1: \documentstyle[pre,aps,preprint,epsfig]{revtex}
2: \renewcommand{\thesubsection}{\arabic{subsection}}
3: \renewcommand{\theenumi}{\roman{enumi}}
4: \renewcommand{\labelenumi}{(\theenumi)}
5: \tightenlines
6: \begin{document}
7:
8:
9: \title{THE VISCOUS SLOWING DOWN OF SUPERCOOLED LIQUIDS
10: AND THE GLASS TRANSITION: PHENOMENOLOGY, CONCEPTS, AND MODELS}
11: \author{G. TARJUS}\address{
12: Laboratoire de Physique Th{\'e}orique des Liquides, Universit{\'e} Pierre et
13: Marie Curie, 4 Place Jussieu, 75252 Paris cedex 05, France}
14: \author{D. KIVELSON}
15: \address{Department of Chemistry and Biochemistry,
16: University of California, Los Angeles, CA 90095, USA}
17: \maketitle
18: \begin{abstract}
19: The viscous slowing down of supercooled liquids that leads to glass
20: formation can be considered as a classical, and is assuredly a
21: thoroughly studied, example of a "jamming process". In this review, we
22: stress the distinctive features characterizing the phenomenon. We also
23: discuss the main theoretical approaches, with an emphasis on the
24: concepts (free volume, dynamic freezing and mode-coupling
25: approximations, configurational entropy and energy landscape,
26: frustration) that could be useful in other areas of physics where
27: jamming processes are encountered.
28: \end{abstract}
29: \vspace{0.5cm}
30:
31: To appear in:
32: {\it Jamming and Rheology: Constrained Dynamics on Microscopic and Macroscopic
33: Scales}\\
34: {\it S. Edwards, A. liu, and S. Nagel Eds.}
35:
36:
37: \section{INTRODUCTION}
38: When cooling a liquid, usually under isobaric P=1atm conditions, one
39: can often bypass crystallization, thereby obtaining a supercooled
40: liquid that is metastable relative to the crystal. When the
41: temperature is further lowered, the viscosity of the liquid, as well
42: as the relaxation times associated with the primary ($\alpha$) relaxation
43: of all kinds of structural, dielectric, macro- and micro-scopic
44: observables, increase rapidly, until a temperature is reached at which
45: the liquid can no longer flow and equilibrate in the time scale of the
46: experiment. The system effectively appears as a rigid amorphous
47: material and is then called a glass. Glass formation thus results from
48: the strong viscous slowing down of a liquid with decreasing
49: temperature\footnote{Although not as widely used, there are other ways
50: of generating glassy structures, such as vapor deposition, in situ
51: polymerization or chemical reactions.\cite{R3}} , a slowing down that
52: can be considered as a classical example of a ``jamming process''. A
53: characteristic of this process that is unanimously recognized, a
54: unanimity rare in this otherwise quite open and controversial field,
55: is that it is a dynamic effect. The so-called ``glass transition'' is
56: not a {\it bona fide} thermodynamic phase transition, but represents a
57: crossover below which a liquid falls out of equilibrium on the
58: experimental time scale. The transition temperature, $T_g$ depends on
59: this time scale, set either by the observation time (corresponding,
60: for instance, to a relaxation time of $10^2$ or $10^3$ sec or a
61: viscosity of $10^{13}$ Poise) and/or by the cooling rate (in a typical
62: differential scanning calorimetry measurement, $10$ K per minute).
63: The dependence on cooling rate is, however, weak, a difference of few
64: K in $T_g$ for an order-of-magnitude change of the rate; this is so
65: because on further lowering of the temperature the viscosity and
66: $\alpha$-relaxation times rapidly become enormous and out of reach of any
67: experimental technique.
68:
69: In the following, we shall focus on the jamming process occurring in
70: the supercooled liquid state.\cite{R1,R2} Both the
71: crystal\footnote{Note that for several liquids, such as
72: m-fluoroaniline and dibutylphtalate at atmospheric pressure and
73: atactic polymers, crystallization has never been observed.}, which is
74: the stable phase below the melting point $T_m$ but can be ignored in
75: discussing the glass transition, and the glassy state itself will be
76: excluded from the present discussion. By appropriately eliminating
77: the crystal (experimentally as well as theoretically), metastable
78: supercooled liquids can be treated by equilibrium thermodynamics,
79: statistical mechanics, and conventional linear-response formalisms.
80: Glasses on the other hand, although mechanically stable, are
81: out-of-equilibrium states; especially near $T_g$, they display
82: nonlinear responses and relaxations known as aging or annealing, and
83: their properties depend on their history of preparation. \cite{R4,R5}
84: These phenomena will not be considered in this review.
85:
86: \section{SALIENT PHENOMENOLOGY }
87: The distinctive feature of glass-forming liquids is the dramatic,
88: continuous increase of viscosity and $\alpha$-relaxation times with
89: decreasing temperature. This sort of jamming is observed in a large
90: variety of substances: covalently bonded systems like $SiO_2$,
91: hydrogen-bonded liquids, ionic mixtures, polymers, colloidal
92: suspensions, molecular van der Waals liquids, etc. The emphasis will
93: be placed on those liquids (the vast majority) that do not form 2- or
94: 3-dimensional networks of strong bonds because they show the most
95: striking behavior when passing from the high-temperature liquid phase
96: to the deeply supercooled and very viscous regime.
97:
98: \subsection{Strong, super-Arrhenius T-dependence of
99: viscosity and $\alpha$-relaxation times } The viscosity $\eta$ and
100: $\alpha$-relaxation times can change by $15$ orders of magnitude for a
101: mere decrease of temperature by a factor two\footnote{The (shear)
102: viscosity $\eta$ can be related to a time characteristic of
103: $\alpha$-relaxation, the average shear stress relaxation time $\tau_s$, by
104: $\eta=G_\infty \tau_s$, where $G_\infty$ is the infinite-frequency shear
105: modulus; $G_\infty$ is typically of the order of
106: $10^{10}-10^{11}$ erg.cm$^{-3}$, so that a viscosity of $10^{13}$
107: Poise roughly
108: corresponds to a time of $10^2$ or $10^3$ sec.}. Such a dramatic
109: variation is conveniently represented on a logarithmic plot of $\eta$ or
110: $\tau_\alpha$ versus $1/T,$ i.e., an Arrhenius plot: see Fig. 1a. A
111: system like $GeO_2$, an example of a network-forming system, is
112: characterized by an almost linear variation, which indicates an
113: Arrhenius temperature dependence. For all other liquids on the figure
114: there is a marked upward curvature, which represents a
115: faster-than-Arrhenius, or super-Arrhenius, temperature dependence and
116: is often described by an empirical Vogel-Fulcher-Tammann (VFT) formula
117: (also called Williams-Landel-Ferry formula in the context of polymers
118: studies\cite{R6}),
119: \begin{equation}
120: \label{eq:1}
121: \tau_\alpha =\tau_0\exp\left(D\frac{T}{T-T_0}\right),
122: \end{equation}
123: where $\tau_0$, $D$ and $T_0<T_g$ are adjustable parameters. On the
124: basis of such Arrhenius plots, with the temperature scaled by $T_g$,
125: Angell proposed the now standard classification of glass-forming
126: liquids into {\it strong} (Arrhenius-like) and {\it fragile}
127: (super-Arrhenius) systems;\cite{R7} in Eq.~(\ref{eq:1}), the smaller
128: the value of $ D$, the more fragile the liquid. There are, of course,
129: alternative fitting formulas that have been used, some which do not
130: imply a singularity at a nonzero temperature as does the expression in
131: Eq.~(\ref{eq:1}).\cite{R8}
132:
133: A different way of representing the phenomenon is to plot the
134: effective activation free energy for $\alpha$-relaxation, $E(T)$, obtained
135: from
136:
137: \begin{equation}
138: \label{eq:2}
139: \tau_\alpha =\tau_{\alpha,\infty}\exp\left(\frac{E(T)}{k_BT}\right),
140: \end{equation}
141: where $k_B$ is the Boltzmann constant and $\tau_{\alpha,\infty}$ is a high-$T$
142: relaxation time, or from a similar equation for the viscosity. This is
143: illustrated in Fig. 1b, where the temperature has been scaled for each
144: liquid to a temperature T* above which the dependence is roughly
145: Arrhenius-like. Although the determination of $T^*$ is subject to
146: some uncertainty,\cite{R8,R9} the procedure emphasizes the crossover
147: from Arrhenius-like to super-Arrhenius behavior that is typical of and
148: quite distinct in most supercooled liquids. The appreciable size of
149: the effective activation free energies $E(T)$, namely, $40 k_BT_g$ at
150: the glass transition, is indicative of thermally activated
151: dynamics. Such a large effective activation free energy for weakly
152: bonded fragile molecular liquids such as orthoterphenyl is an
153: intriguing feature of the phenomenology. Another peculiar property of
154: $E(T)$ for fragile systems is that it increases significantly between
155: $T^*$ and $T_g$ (a factor $3$, i.e., a factor of $5$ or $6$ in units
156: of the thermal energy $k_BT$, for weakly bonded fragile liquids).
157: Such a variation is not commonly encountered. For instance, in the
158: field of critical phenomena, the slowing down of dynamics that occurs
159: when approaching the critical point is usually characterized by a
160: power law growth of the relaxation time; in terms of effective
161: activation free energy, this corresponds to a logarithmic growth and
162: it is slower than the variation described by the VFT formula,
163: Eq.~(\ref{eq:1}). Unusually strong slowing down, with exponentially
164: growing times similar to Eq.~(\ref{eq:1}), is found in some disordered
165: systems like the random field Ising model and it is known as
166: ``activated dynamic scaling''.\cite{R10}
167:
168: \subsection{Nonexponential relaxations}
169: In an ``ordinary'' liquid above the melting point, relaxation functions
170: are usually well described, after some transient time, by a simple
171: exponential decay. Deviations are observed, but they are neither
172: systematic nor very marked. The situation changes at lower
173: temperatures, and the $\alpha$-relaxation is no longer characterized by an
174: exponential decay. A better representation is provided by a
175: ``stretched exponential'' (or Kohlrausch-Williams-Watts function),
176:
177: \begin{equation}
178: \label{eq:3}
179: f_\alpha(t)\propto \exp\left[-\left(\frac{t}{\tau_\alpha}\right)^\beta \right],
180: \end{equation}
181: where $\beta$ is the stretching parameter; the smaller $\beta$ the more
182: ``stretched'' the relaxation. Although not unambiguously established,
183: the degree of departure from exponential behavior, or stretching,
184: appears to increase (i.e., $\beta$ decreases) with decreasing
185: temperature.
186:
187: Alternatively, in frequency space, the spectrum of the imaginary part
188: of the susceptibility, which is characterized by a peak at a frequency
189: $\omega_\alpha \propto 1/ \tau_\alpha $, tends to be broader (when plotted as a function
190: of $log(\omega)$ ) than the simple Lorentzian or Debye spectrum that is
191: just the Fourier transform of the time-derivative of an exponential
192: relaxation function (see Fig. 2). Fitting formulas related to
193: Eq.~(\ref{eq:3}), formulas like the Cole-Davidson function for
194: frequency-dependent susceptibilities, $(1-i(\omega/\omega_\alpha ))^{-\beta'}$, are used
195: to fit the spectroscopic data, but similar trends are observed: the
196: $\alpha$ peak, as observed for instance in the imaginary part of the
197: dielectric susceptibility as a function of $log(\omega)$, broadens as the
198: temperature is lowered towards $T_g$,\cite{R11} which indicates
199: increasing departure from Debye/exponential behavior. Except for
200: network-forming systems, the stretching of the $\alpha$ relaxation is
201: significant ($\beta$ is typically between $0.3$ and $0.6$ for fragile
202: liquids at $T_g$). However, and this point may not have been given
203: enough attention, the stretching, or broadening in frequency space, is
204: relatively small when compared to the extremely rapid variation with
205: temperature of the $\alpha$-relaxation time itself. This is to be
206: contrasted for instance with the activated critical slowing discussed
207: above. There, the power law growth of the activation free energy when
208: the temperature is decreased toward the critical point comes with a
209: more striking stretching of the relaxation function that occurs on a
210: logarithmic scale: in this case, in place of a stretched exponential
211: behavior as in Eq.~(\ref{eq:3}), $\ln(f(t))$ goes as some power of
212: $(1/\ln(t))$.\cite{R12}
213: \subsection{No marked changes in structural quantities}
214:
215: It is tempting to associate the huge increase in $\alpha$-relaxation times
216: and viscosity with the growth of a {\it structural} correlation
217: length. However, no such growth has been detected so far in
218: supercooled liquids. Quite to the contrary, the variation of
219: structure in liquids and glasses, as measured in neutron and X-ray
220: diffraction experiments, appear rather bland\cite{R13,R14} (see Fig.
221: 3). The ordinary, high-temperature liquid has only short-range order
222: whose signature in the static structure factor $S(Q)$ is a broad peak
223: (or a split peak for some molecular systems as illustrated in Fig. 3)
224: at a wave vector $Q$ that roughly corresponds in real space to some
225: typical mean distance between neighboring molecules. As the
226: temperature is lowered and the supercooled regime is entered, there
227: are small, continuous variations of the structure factor that mostly
228: reflect the change in density (typically, a $5\%$ change between $T_m$
229: and $T_g$) and, possibly, some adjustments in the local arrangements
230: of the molecules. There is no sign, however, of a significantly
231: growing correlation length, nor of the appearance of a super-molecular
232: length.
233:
234: In network-forming and
235: $H$-bonded systems, an additional ``pre-peak'' is sometimes detectable at
236: wave vectors somewhat lower than that of the main peak, but it is
237: attributed to specific effects induced by the strongly directional
238: intermolecular bonds and not to a length scale that would correlate
239: with the viscous slowing down.\cite{R15}
240:
241: In
242: contrast to this lack of structural signature for the existence of an
243: increasing super-molecular correlation length with decreasing
244: temperature, there is significant evidence, as discussed below, that
245: corresponding ``dynamical'' correlation length do exist.
246:
247: \subsection{Rapid entropy decrease and Kauzmann paradox }
248:
249:
250: The absence of marked changes in the structure, {\it at least at the
251: level of two-particle density correlations}, or of a strong increase
252: in any directly measured static susceptibility is a puzzling feature
253: of the jamming process associated with glass formation. The only
254: static quantity that shows behavior that might be relevant is the
255: entropy. Below the melting point, $T_m$, the heat capacity $C_p(T)$
256: of a supercooled liquid is larger than that of the corresponding
257: crystal. (At $T_g$, the $C_p$ of the liquid drops to a value that is
258: characteristic of the glass and is close to the $C_p$ of the crystal,
259: but this is a consequence of the system no longer being properly
260: equilibrated.) As a result of this ``excess'' heat capacity, the entropy
261: difference between the liquid and the crystal decreases with
262: temperature, typically by a factor of $3$ between $T_m$ and $T_g$ for
263: fragile liquids. The effect is illustrated in Fig. 4 and leads to the
264: famous Kauzmann paradox:\cite{R16} if the entropy difference is
265: extrapolated to temperatures below $T_g$, its extrapolated value
266: vanishes at some nonzero temperature $T_K$, which results in the
267: unpleasant feature that the entropy of the liquid becomes equal to
268: that of the crystal (even more unpleasant: if the extrapolation is
269: carried to still lower temperatures, the entropy of the liquid becomes
270: negative, which violates the third law of thermodynamics). The paradox
271: is that this {\it extrapolated }entropy crisis is avoided for a purely
272: dynamic reason, the intervention of the glass transition: what would
273: occur if one were able to keep the supercooled liquid equilibrated
274: down to temperatures below $T_g$? There are certainly many ways to
275: answer the question. The paradox could be resolved by the existence
276: between $T_g$ and $T_K$ of an intrinsic limit of metastability of the
277: liquid\cite{R16} or of a second-order phase transition\footnote{Note
278: that a low-$T$ first-order transition does not resolve the paradox
279: because it can be supercooled.} (a speculation that gains additional
280: credibility with the observation that the VFT temperature $T_0$ at
281: which the extrapolated viscosity and $\alpha$-relaxation times diverge
282: (see Eq.~(\ref{eq:1})) is often found close to
283: $T_K$\cite{R17,R18}). Even more simply, one might find that the
284: extrapolation breaks down above $T_K$ and that the entropy-difference
285: curve levels off and goes smoothly to zero at zero $K$, in much the
286: same way as it does in the Debye theory of crystals. These are of
287: course all speculations, but it remains that the rapid decrease of the
288: entropy of the supercooled liquid relative to that of the crystal
289: represents an intriguing aspect of the phenomenology of fragile
290: glass-formers.
291:
292: \subsection{Two-step relaxation and secondary processes}
293: As we stressed before, the salient features related to glass formation
294: concern the long-time (low-frequency) primary or $\alpha$ relaxation. As
295: the $\alpha$-relaxation time increases with decreasing temperature, so too
296: does the window between this time and typical microscopic, picosecond
297: or sub-picosecond times. When the relaxation function is plotted
298: against the logarithm of the time, one then observes what is sometimes
299: called a ``two-step relaxation''. An illustration is given in Fig. 5
300: by the dynamic structure factor of the fragile ionic glass-former
301: $Ca_{0.4}K_{0.6} (NO_3)_{1.4}$ obtained by neutron
302: techniques.\cite{R19} At high temperature, the relaxation function is
303: essentially a one-step process. However, as the liquid becomes more
304: viscous, the relaxation proceeds in two steps separated by a plateau.
305: Although the terminology is far from being universally accepted, the
306: approach to the plateau from the short-time side is often referred to
307: as $\beta$ or fast-$\beta$ relaxation. If one is to fit the long-time part
308: by a stretched exponential (Eq.~(\ref{eq:3})), there is a large range
309: of ``mesoscopic'' times that is not adequately described and that widens
310: as the temperature is lowered. Power law functions of time are often
311: used to reproduce the relaxation function in this mesoscopic range.
312:
313: This two-step relaxation feature is common to all fragile liquids. In
314: addition, there may also appear additional secondary processes,
315: detected first by Johari and Goldstein in dielectric spectroscopy.
316: \cite{R20} Such secondary processes, whose presence and strength
317: strongly vary from one liquid to another, have characteristic
318: frequencies that are intermediate between those of the $\alpha$ and
319: fast-$\beta$ relaxations. They are denoted Johari-Goldstein-$\beta$,
320: slow-$\beta$, or simply $\beta$ processes. \cite{R2,R21} To make the
321: description more complete, one should also mention the so-called
322: ``boson peak'' that may be present on the high-frequency side
323: ($\sim 10^2-10^3Ghz$) of light and neutron scattering (or absorption)
324: spectra.\cite{R2,R22} Here we do not discuss either the slow-$\beta$
325: processes or the boson
326: peak.
327:
328: \section{ A SELECTION OF QUESTIONS }
329: After this brief review of the salient aspects of the phenomenology of
330: supercooled liquids as they get glassy, we discuss in more detail a
331: number of questions, whose answers give justification or put
332: constraints on the theoretical picture one can build to explain the
333: viscous slowing down.
334: \subsection{How universal is the behavior of glass-forming
335: liquids ?}
336:
337: Universality is a key concept in physics and it has proven to be
338: central in the field of critical phenomena. By the standards of
339: critical phenomena studies, the observed behavior of glass-forming
340: liquids is not universal, the main reason being that no singularity is
341: detected experimentally (or approached asymptotically close), as
342: stressed above. However, if one is willing to take a broader view of
343: the notion of universality, one can find considerable generality or
344: ``universality'' in the properties, those mostly associated with
345: long-time and low-frequency phenomena, that characterize the approach
346: to the glass transition. For instance, the super-Arrhenius
347: $T$-dependence of the viscosity and $\alpha$-relaxation times and the
348: nonexponential character of the relaxation function are observed for
349: virtually all glass-formers, be they polymeric, $H$-bonded, ionic, van
350: der Waals, etc., with the exception of a minority of strong
351: network-forming systems; and, for a given liquid, these properties are
352: found by a large variety of experimental techniques, such as
353: dielectric relaxation, light and neutron scattering, NMR, viscosity
354: measurements, specific heat spectroscopy, volume and enthalpy
355: relaxation, optical probe methods.
356:
357: The presence of an underlying "universality" is supported by the fact
358: that experimental data covering a wide range of temperatures and a
359: great diversity of substances can be collapsed onto master curves with
360: only a small number of species-dependent adjustable parameters. A
361: good example is provided by the scaling plot of the
362: frequency-dependent dielectric susceptibility proposed by Nagel and
363: coworkers.\cite{R11} As shown in Fig. 6, the data taken over a $13$
364: decade range of frequencies for many different liquids can be placed
365: with good accuracy onto a single curve after scaling with only three
366: parameters associated with the $\alpha$-peak position, width and
367: intensity. The master-curve for the temperature dependence of the
368: effective activation free energy for viscosity and $\alpha$ relaxation put
369: forward by Kivelson {\it et al.} \cite{R8} is another example. It is
370: nonetheless fair to say that the fits resulting from these various
371: scaling procedures are far from perfect, which leaves room for debate
372: and conflicting interpretations. One can also ask the question whether
373: the universality holds only up to the implied high-frequency cut-off
374: of the susceptibility scaling curve of Nagel {\it et al.}
375: (i.e., whether one should be focusing on slow behavior) or
376: whether it extends higher in light of the fact that similarities have
377: also been observed in the high-frequency susceptibilities.
378: \cite{R23}
379: \subsection{ Is the $\alpha$-relaxation homogeneous or heterogeneous?}
380: The observation stressed above that the $\alpha$-relaxation is
381: nonexponential in the supercooled liquid range and its representation
382: by a stretched exponential as in Eq.~(\ref{eq:2}) can be formally
383: interpreted in terms of a superposition of exponentially decaying
384: functions with a distribution of relaxation times; but, this {\it per
385: se} does not guarantee that the dynamics be ``heterogeneous'', in the
386: sense that relaxation of the molecules differs from one environment to
387: another with the environment life time being longer than the
388: relaxation time. An alternative explanation can be offered within a
389: ``homogeneous'' picture in which relaxation of the molecules everywhere
390: in the liquid is intrinsically nonexponential.
391:
392: In recent years, there has been mounting evidence that
393: heterogeneities, sufficiently long-lived to be relevant to the
394: $\alpha$-relaxation and to be at least partly responsible for its
395: nonexponential feature, do exist in supercooled liquids.\cite{R24} The
396: heterogeneous character of the slow dynamics has been demonstrated in
397: several experiments: multi-dimensional NMR, \cite{R25} photobleaching
398: probe rotation measurements, \cite{R26} nonresonant dielectric hole
399: burning. \cite{R27} These techniques involve the selection of a
400: sub-ensemble of molecules in the sample that is characterized by a
401: fairly narrow distribution of relaxation times (and in general a
402: relaxation slower than average) and the further monitoring of the
403: gradual return to the equilibrium situation. Additional evidence of
404: the spatially heterogeneous nature of the dynamics in fragile
405: supercooled liquids is provided by the so-called breakdown of the
406: Stokes-Einstein relation between the translational diffusion constant
407: and the viscosity, and the concomitant ``decoupling'' between rotational
408: and translational time scales:\cite{R28,R29} see Fig.7.
409:
410: The size of the heterogeneities is not directly observable in the
411: above mentioned experiments, but various estimates, obtained, e. g.,
412: from optical studies of the rotational relaxation of probes of varying
413: size, \cite{R30} NMR measurements, \cite{R31} the study of excess
414: light scattering, \cite{R32} and the influence of a well-defined
415: $3$-dimensional confinement \cite{R33} lead to a typical length of
416: several nanometers in different fragile liquids near $T_g$. One should
417: recall that these signatures are all dynamical and that no signature
418: at such a length scale has been detected so far in small-angle neutron
419: and X-ray diffraction data. If the heterogeneous character of the
420: $\alpha$-relaxation appear reasonably well established, at least for
421: deeply supercooled fragile liquids, several points concerning the
422: lifetime, the size, and the nature of the heterogeneities need still
423: be settled.
424:
425: \subsection{Is density or temperature the dominant control variable?}
426: The phenomenon of viscous
427: slowing down and glass formation as it is studied most of the time
428: (and described in the preceding sections) takes place under isobaric
429: $P=1atm$ conditions. As a consequence, when the temperature is lowered,
430: there is also an increase of the density of the liquid. This increase
431: is small (a typical variation of $5\%$ between $T_m$
432: and $T_g$), but it could still have a
433: major influence on the dynamics. Actually, there are theoretical
434: models of jamming, such as those based on free volume concepts and
435: hard sphere systems, that attribute the spectacular increase of
436: viscosity and $\alpha$-relaxation
437: times of fragile glass-formers (almost) entirely to the density
438: changes. It is thus important to evaluate the role of density and
439: temperature in driving the jamming process that leads to the glass
440: transition {\it at 1 atm}.
441:
442: Basic models and theories are usually formulated in terms of either
443: density or temperature as control variable, but experiments are
444: carried out with pressure and temperature as external control
445: variables. The data must be converted, when enough experimental
446: results are available, in order to analyze the influence of density at
447: constant temperature and that of temperature at constant density, for
448: a range of density and temperature that is characteristic of the
449: phenomenon {\it at 1 atm}. Extant analyses\cite{R34} are far from
450: exhaustive. However, as illustrated in Fig. 8, the characteristic
451: super-Arrhenius T-dependence of the viscosity, $\eta$, and
452: $\alpha$-relaxation times, $\tau_\alpha$, appears predominantly due to the
453: variation of temperature and not to that of density. This conclusion
454: is confirmed by a comparative study of the contributions induced by
455: density variations (at constant temperature) and by temperature
456: variations (at constant density) to the rate of change of $\eta$ and
457: $\tau_\alpha$ at constant (low) pressure in the viscous liquid regime of
458: several molecular and polymeric glass-formers. \cite{R34}
459:
460: How general is the above result? Temperature
461: appears to be the dominant variable controlling the viscous
462: super-Arrhenius slowing down of supercooled liquids at low pressure,
463: but this may not be the case at much higher pressure (although not
464: much data are presently available to confirm this point), and it is
465: most likely not true for describing the concentration-driven
466: congestion of dynamics in colloidal suspensions. In the absence of a
467: ``super-universal'' picture of the jamming associated with
468: glass formation, we shall restrict ourselves, as we have implicitly
469: done above, to the consideration of supercooled liquids at 1 atm.
470:
471: \subsection{What are the relevant characteristic temperatures?}
472:
473: There is no unbiased way of presenting the phenomenology of
474: glass-forming liquids. Choices must be made about the emphasis put on
475: the different aspects, about the best graphic representations, and
476: about the way in which one analyzes experimental data. To make sense
477: out of the wealth of observations and measurements, it is natural to
478: look for characteristic temperatures about which to organize and scale
479: the data. However, since no singularity is directly detected, the
480: selection of one or several relevant temperatures is far from
481: straightforward. The temperatures that can be easily determined
482: experimentally are the boiling point, $T_b$, the melting point, $T_m$,
483: and the glass transition temperature(s), $T_g$. Unfortunately, the
484: former two are generally considered as irrelevant to the jamming
485: phenomenon, and the latter has an operational rather than a
486: fundamental nature (see the introduction).
487:
488: Several other candidates have been suggested, that can be split into
489: two groups. First, there are ``extrapolation temperatures'' {\it below}
490: $T_g$, i.e., temperatures dynamically inaccessible to supercooled
491: liquids, at which extrapolated behavior diverges or becomes singular.
492: This is the case of the VFT temperature $T_0$ (see Eq. 1) and the
493: Kauzmann temperature $T_K$ (see section II-4 and Fig. 4). In the
494: second group are ``crossover temperatures'' {\it above} $T_g$ at which a
495: new phenomenon seems to appear (decoupling of rotations and
496: translations, emergence of a secondary $\beta$ process, etc.), a
497: crossover of behavior or a change of
498: $\alpha$-relaxation mechanism seems to take place (passage from
499: Arrhenius to super-Arrhenius $T$-dependence, arrest of the relaxation
500: mechanisms described by the mode-coupling theory, putative emergence
501: of activated barrier crossing processes, etc.). A variety of such
502: temperatures for the fragile glass-former $OTP$ are shown in Fig. 9:
503: the location of the different characteristic temperatures is
504: illustrated on an Arrhenius plot of the viscosity. It is interesting
505: to note that most putative crossover behaviors occur in the region of
506: strong curvature where $\eta \sim 1-10^2$ Poise or $\tau_\alpha \sim 10^{-10}-10^{-8}$
507: sec, while, on the other hand, the temperatures obtained by
508: extrapolation of data to low T lie fairly close to each other, some 40
509: K below $T_g$.
510:
511: \subsection{What can be learned from computer simulations?}
512:
513: Computer simulation studies, \cite{R39} in particular those based on
514: Molecular Dynamics algorithms, have proven extremely valuable in
515: investigating the structure and dynamics of simple, ordinary liquids.
516: Their contribution to the understanding of the glass transition is
517: unfortunately limited, the main reason being the restricted range of
518: lengths and times that are accessible to Molecular Dynamics
519: simulations: typical simulations on atomistic models consider
520: $10^3-10^4$ atoms and can follow relaxations for less than $10^{-8}$
521: sec (when expressing the elementary time step in terms of parameters
522: characteristic of simple liquids). As a result, the viscous, deeply
523: supercooled regime of real glass-forming liquids, where strong
524: super-Arrhenius behavior, heterogeneous dynamics, and other
525: significant features associated with the jamming process develop, is
526: out of reach, as is the laboratory glass transition that occurs on a
527: time scale of $10^2$ or $10^3$ sec. Simple liquid models do form
528: ``glasses'' on the observation, i.e., simulation, time with many of
529: the attributes of the laboratory glass transition: abrupt change in
530: the thermodynamic coefficients, dependence on the cooling rate, aging
531: effects, etc\footnote{ However, these models are effectively
532: high-temperature structures, since they correspond to liquid
533: configurations that are kinetically arrested at a temperature at which
534: the primary relaxation time is of the order only of nanoseconds. Even
535: with systems specially designed to avoid crystallization, such as
536: binary Lennard-Jones mixtures, the cooling rates to prepare glasses
537: ($10^8$ to $10^9$ K/sec) are orders of magnitude higher than those
538: commonly used in experiments.}. However, these supercooled simulation
539: liquids are not truly fragile (the $T$-dependence of the
540: $\alpha$-relaxation time shows only small departure from Arrhenius
541: behavior) and are not deeply supercooled so that one{\'{}}s ability to
542: extract insights into the deeply supercooled fragile liquids is
543: questionable.
544:
545: Computer simulations can be useful in studying the moderately
546: supercooled liquid region, where one can observe the onset of viscous
547: slowing down (see for instance Fig. 9). The major interest of such
548: studies is that static and dynamic quantities that are not
549: experimentally accessible can be investigated, such as multi-body
550: (beyond two-particle) correlations involving a variety of variables
551: and microscopic mechanisms for transport and relaxation. They also
552: allow for testing in detail theoretical predictions made in the
553: relevant window of times and lengths (e. g., those of the
554: mode-coupling theory) and for analyzing properties associated with
555: configurational or phase space (see below).
556:
557: In addition to the much studied one- or two-component systems of
558: spheres with spherically symmetric interaction potentials,\cite{R39}
559: the models investigated in computer simulations can be divided into
560: two main groups: on one hand, more realistic microscopic models for
561: molecular glass-formers that attempt to describe species-specific
562: effects;\cite{R40} on the other hand, more schematic systems,
563: coarse-grained representations, \cite{R41} lower-dimensional systems
564: \cite{R42} or toy-models,\cite{R43}
565: that bear less detailed resemblance with real glass-formers, but can
566: be studied on much longer time scales and with bigger system sizes.
567:
568:
569:
570: \section{THEORETICAL APPROACHES.}
571: There is a large number of theories, models, or simply empirical
572: formulae that attempt to reproduce pieces of the phenomenology of
573: supercooled liquids. There are fewer approaches, however, that
574: address the question of why and how the viscous slowing down leading
575: to the glass transition, with its salient characteristics described in
576: the preceding sections, occurs in liquids as they are cooled. In the
577: following, we shall briefly review the main theoretical approaches,
578: with an emphasis on the concepts and methods that may prove useful in
579: other areas of physics where some sort of jamming process is also
580: encountered\footnote{Models addressing more specific questions, such
581: as the ``coupling model'' \cite{R44} or the ``continuous time random
582: walk'', approach\cite{R45} are discussed in the reviews cited in
583: \cite{R1}.}. More specifically, we shall discuss phenomenological
584: models based on free volume and configurational entropy, the
585: description of a purely dynamic arrest resulting from mode-coupling
586: approximations, ideas relying on the consideration of the topographic
587: properties of the configurational space (energy and free-energy
588: landscapes) or on the analogy with generalized spin glass models, and
589: approaches centered on the concept of frustration.
590:
591: \subsection{Free volume }
592: Free-volume models rest on the assumption that molecular transport in
593: viscous fluids occurs only when voids having a volume large enough to
594: accommodate a molecule form by the redistribution of some ``free
595: volume'', where this latter is loosely defined as some surplus volume
596: that is not taken up by the molecules. In the standard presentation
597: by Cohen and Turnbull,\cite{R46} a molecule in a dense fluid is mostly
598: confined to a cage formed by its nearest neighbors. The local free
599: volume, $v_f$, is roughly that part of a cage space
600: which exceeds that taken by a molecule. It is assumed that between
601: two events contributing to molecular transport, a reshuffling of free
602: volume among the cages occurs at no cost of energy. Assuming also
603: that the local free volumes are statistically uncorrelated, one
604: derives a probability distribution, $P(v_f)$, which is exponential,
605: \begin{equation}
606: \label{eq:4}
607: P(v_f)\propto \exp\left(-\gamma \frac{v_f}{\overline{v_f}}\right),
608: \end{equation}
609: where $\overline{v_f}$ is the average free volume per molecule and $\gamma$ is
610: a constant of order $1$. Since the limiting mechanism for the
611: diffusion of a molecule is the occurrence of a void, i.e., a local
612: free volume $v_f$ larger than some critical value, $v_0$, that is
613: approximately equal to the molecular volume, the diffusion constant
614: $D$ is given by the probability of finding a free volume equal to
615: $v_0$; this leads to an expression for $D$, and by extension for the
616: viscosity $\eta$,
617:
618: \begin{equation}
619: \label{eq:5}
620: \eta\propto 1/D\propto \exp\left(\gamma \frac{v_0}{\overline{v_f}}\right),
621: \end{equation}
622: which is similar to the formula first proposed by Doolittle.\cite{R47}
623:
624: In the Cohen-Turnbull formulation, the average free volume per
625: molecule is given by $\overline{v_f}=v-v_0$, where $v=1/ \rho$ is the average
626: total volume per molecule. The free-volume concept, in zeroth order,
627: relies on a hard-sphere picture in which thermal activation plays no
628: role. For application to real liquids, temperature enters through the
629: fact that molecules, or molecular segments in the case of polymers,
630: are not truly ``hard'' and that, consequently, the constant-pressure
631: volume is temperature-dependent,
632: \begin{equation}
633: \label{eq:6}
634: \overline{v_f(T)}\propto \alpha_P(T-T_0),
635: \end{equation}
636: where $\alpha_P$ is the coefficient of isobaric expansivity and $T_0$ is
637: the temperature at which all free volume is consumed, i.e., $v=v_0$.
638: Inserting the above equation in the Doolittle formula,
639: Eq.~(\ref{eq:4}), gives the VFT expression, Eq.~(\ref{eq:1}). An
640: unanswered, but fundamental question associated with Eq.~(\ref{eq:6})
641: is why the free volume should be consumed at a nonzero temperature,
642: $T_0$? An extended version of the free-volume approach has been
643: developed by Cohen and Grest, in which the cages or ``cells'' are
644: divided into two groups, liquid-like and solid-like, and concepts from
645: percolation theory are included to describe the dependence upon the
646: fraction of liquid-like cells. \cite{R48} (See also the model for
647: molecular diffusivity in fluids of long rod molecules by Edwards and
648: Vilgis.\cite{R49})
649:
650: The main criticisms of the free volume models are
651: (i) that the concept of free volume is ill-defined, which results in a
652: variety of interpretations and difficulty in finding a proper
653: operational procedure even for simple model systems, and (ii) that the
654: pressure dependence of the viscosity (and $\alpha$-relaxation times) is not adequately
655: reproduced. This latter feature has been emphasized in many
656: studies, \cite{R17,R34,R50} and it is a
657: consequence of the observation made above (see II-3) that the viscous
658: slowing down of glass-forming liquids at 1 atm and more generally at
659: low pressure is primarily controlled by temperature and not by density
660: or volume. Glass formation in supercooled liquids does not
661: predominantly results from the drainage of free volume, but rather
662: from thermally activated processes.
663:
664: \subsection{Mode-coupling approximations}
665: The theory of glass-forming liquids that has had the highest
666: visibility for more than a decade is the mode coupling
667: theory. \cite{R51} It predicts a dynamic arrest of the
668: liquid structural relaxation without any significant change in the
669: static properties. All structural quantities are assumed to behave
670: smoothly and jamming results from a nonlinear feedback mechanism that
671: affects the relaxation of the density fluctuations. Formally, the
672: theory involves an analysis of a set of nonlinear integro-differential
673: equations describing the evolution of pair correlation functions of
674: wave-vector- and time-dependent fluctuations that characterize the
675: liquid. These equations have the form of generalized Langevin
676: equations, and they can be derived by using the Zwanzig-Mori
677: projection-operator formalism. The equation for the quantity of prime
678: interest in the theory, the (normalized) correlation function of the
679: density fluctuations,
680:
681: \begin{equation}
682: \label{eq:7}
683: \phi_Q(t)= \frac{<\rho_Q(t)\rho_Q^*(0)>}{<|\rho_Q(0)|^2>},
684: \end{equation}
685: where $\rho_Q(t)=\sum_j\exp(i{\bf Q}{\bf r}_j)$
686: and ${\bf r}_j$
687: denotes the position of the $j$th particle, can be written as
688: \begin{equation}
689: \label{eq:8}
690: \frac{d^2}{dt^2}\phi_Q(t)+\Omega^2_Q\phi_Q(t)+\int_0^tdt'm_Q(t-t') \frac{d}{dt'}\phi_Q(t')=0,
691: \end{equation}
692: where $\Omega_Q$ is a microscopic frequency obtainable from the static
693: structure factor, $S(Q)\propto<~|\rho_Q(0)|^2> $, and $m_Q(t)$ is the
694: time-dependent memory function that is formally related to the
695: correlation function of a Q-dependent random force. The above equation
696: being exact, the crux of the mode-coupling approach consists in
697: formulating an approximate expression for $m_Q(t)$. The mode-coupling
698: scheme has been implemented for liquids both in the frame of the
699: kinetic theory of fluids \cite{R51} and that of the fluctuating
700: nonlinear hydrodynamics. \cite{R52} It essentially boils down to
701: approximating the memory function $m_Q(t)$ as the sum of a bare
702: contribution coming from the fast relaxing variables and a
703: mode-coupling contribution coming from the slowly decaying bilinear
704: density modes,
705: \begin{equation}
706: \label{eq:9}
707: m_Q(t)=\gamma\delta(t)+\sum_{{\bf Q'}}V_{\bf QQ'}\phi_{\bf Q'}(t)\phi_{|{\bf Q-Q'}|}(t),
708: \end{equation}
709: where the vertices $V_{\bf QQ'}$ can be expressed in terms of the static
710: structure factor. The self-consistent solution of the resulting set of
711: nonlinear equations predicts a slowing of the relaxation that is
712: attributed, within a purely homogeneous picture (see II-2), to a cage
713: effect and to the feedback mechanism above mentioned. This solution
714: exhibits a dynamic arrest at a critical point, $T_c$, which represents
715: a transition from an ergodic to a nonergodic state with no concomitant
716: singularity in the thermodynamics and structure of the system. The
717: main achievements of the mode-coupling approach are the predicted
718: anomalous increase in relaxation time and the appearance of a two-step
719: relaxation process with decreasing temperature, as indeed observed in
720: real fragile glass-formers (compare Fig. 10 to Fig. 7) and in
721: molecular dynamics simulations. \cite{R39} Early on, however, it was
722: realized, both from empirical fits to experimental data and from
723: comparison to simulation data on model systems, that the dynamic
724: arrest at $T_c$ did not describe the observed glass transition at
725: $T_g$ nor the transition to an ``ideal glass'' at a temperature below
726: $T_g$. This is illustrated in Fig. 11. Thus, the $T_c$ was
727: interpreted as a temperature above $T_g$. The singularity at $T_c$ is
728: avoided because of the breakdown of the simple mode-coupling
729: approximation, Eq. 9, and the $T_c$ of what is called the ``idealized''
730: mode-coupling theory is taken as a crossover below which additional
731: relaxation mechanisms, such as activated processes, presumably take
732: over. Unfortunately, beyond some empirical introduction,
733: \cite{R51,R52} activated processes are not theoretically described by
734: mode-coupling approaches, and so the theory of $\alpha$ relaxation has not
735: been extended to temperatures below $T_c$. To draw once again a
736: parallel with critical phenomena (where a singularity occurs at $T_c$
737: in the structure and the thermodynamics of the system), mode-coupling
738: approximations, as formulated for instance by Kawasaki, are known to
739: describe quite well the standard critical slowing down, but not the
740: activated dynamic scaling such as that observed in the random field
741: Ising model (see section I-1). This failure is related to the
742: underlying nature of the approximation that corresponds to a one-loop
743: self-consistent resummation scheme in a perturbative treatment
744: \cite{R52,R54} (see also below in III-4 the parallel with spin glass models).
745:
746: Mode-coupling approaches can thus describe at best the dynamics of
747: moderately supercooled liquids\footnote{It is possible that they are
748: also applicable to the fast-$\beta$ relaxations even below
749: $T_c$.\cite{R51}} (see Fig. 11). Because of the many detailed
750: predictions it makes in this regime, the mode-coupling theory has
751: stimulated the use and the development of experimental techniques,
752: such as neutron and depolarized light scattering, and molecular
753: dynamics simulations that are able to probe the early stage of the
754: viscous slowing down; but, the very fact that the predicted dynamic
755: singularity is not observed makes it difficult to reach any clear-cut
756: conclusion about the quantitative adequacy of the theory, and this has
757: led to much debate in recent years. \cite{R55}
758:
759: \subsection{ Configurational entropy and (free) energy landscape }
760: The existence of a crossover temperature in the moderately supercooled
761: liquid region where $\alpha$-relaxation times are of the order of
762: $10^{-9}$ sec (hence in the same region as the $T_c$ predicted by the
763: mode-coupling theory) was advocated 30 years ago by
764: Goldstein.\cite{R56} Goldstein argued that below this crossover flow
765: is dominated by potential energy barriers that are high compared to
766: thermal energies and slow relaxation occurs as a result of thermally
767: activated processes taking the system from one minimum of the
768: potential energy hypersurface to another. The idea that molecular
769: transport in viscous liquids approaching the glass transition could be
770: best described by invoking motion of the representative state point of
771: the system on the potential energy hypersurface had also been
772: suggested by Gibbs. \cite{R57} In his view, the slowing down of
773: relaxations with decreasing temperature is related to a decrease of
774: the number of available minima and to the increasing difficulty for
775: the system to find such minima. The viscous slowing down would thus
776: result from the decrease of some ``configurational entropy'' that is a
777: measure of the number of accessible minima. These two concepts,
778: potential energy hypersurface, also denoted ``energy landscape'', and
779: ``configurational entropy'', have gained a renewed interest in recent
780: years, boosted by the analogy with the situation encountered in
781: several generalized spin glass models (see below).
782:
783: The Adam-Gibbs approach \cite{R58}
784: represents a phenomenological attempt to
785: relate the $\alpha$-relaxation time of a glass-forming liquid
786: to the ``configurational entropy''. In the picture,
787: $\alpha$-relaxation takes place by increasingly
788: cooperative rearrangements of groups of molecules. Any such group,
789: called a ``cooperatively rearranging region'', is assumed to relax
790: independently of the others. It is a kind of long-lived
791: heterogeneity. Molecular motion is activated and the effective
792: activated free energy is equal to the typical energy barrier per
793: molecule, which is taken as independent of temperature, times the
794: number of molecules that are necessary to form a cooperatively
795: rearranging region whose size permits a transition from one
796: configuration to a new one independently of the environment. This
797: latter number goes as the inverse of the configurational entropy per
798: molecule, $S_c(T)/N$, where $N$ is the total number
799: of molecules in the sample. Since $S_c$
800: decreases with decreasing temperature, the reasoning leads
801: to an effective activation free energy that grows with decreasing
802: temperature, i. e., to a super-Arrhenius behavior,
803: \begin{equation}
804: \label{eq:10}
805: \tau_\alpha=\tau_0\exp\left(\frac{C}{TS_c(T)}\right),
806: \end{equation}
807: where $C$ is proportional to $N$ times the typical energy barrier per
808: molecule. If the configurational entropy vanishes at a nonzero
809: temperature, an assumption somewhat analogous to that in
810: Eq.~(\ref{eq:6}) for the free volume model, but one that is inherent
811: for instance in the Gibbs-di Marzio approximate mean field treatment
812: of a lattice model of linear polymeric chains, \cite{R59} then the
813: $\alpha$-relaxation times diverge at this same nonzero temperature. In
814: particular if the configurational entropy is identified as the entropy
815: difference between the supercooled liquid and the
816: crystal\footnote{This phenomenological choice for the entropy of
817: configuration has been criticized by Goldstein who showed for several
818: glass-formers that only half of the entropy difference between the
819: liquid and the crystal comes from strictly ``configurational'' sources;
820: the remainder comes mostly from changes in vibrational anharmonicity
821: or differences in the number of molecular groups able to engage in
822: local motions. \cite{R60}}, the Adam-Gibbs theory allows one to
823: correlate the extrapolated divergence of the $\alpha$-relaxation times
824: with the Kauzmann paradox (see I-4): the Kauzmann temperature $T_K$
825: would then signal a singularity both in the dynamics and in the
826: thermodynamics of a supercooled liquid\footnote{A recent careful, but
827: conjectural analysis of dielectric relaxation data suggests that these
828: data are consistent with the existence of a critical point, both
829: structural and dynamical, at the approximate $T_0$ specified by the
830: VTF expression in Eq.~(\ref{eq:1}).\cite{R61}}. Note also that by
831: using a hyperbolic temperature dependence to fit the experimental data
832: on the heat capacity difference between the liquid and the crystal,
833: $\Delta C_P(T)=K/T$, and using
834: this formula to extrapolate the configurational entropy down to the
835: Kauzmann temperature, one converts Eq.~(\ref{eq:10}) to a VFT formula,
836: \begin{equation}
837: \label{eq:11}
838: \tau_\alpha=\tau_0\exp\left(\frac{CT_K}{K(T-T_K)}\right),
839: \end{equation}
840: with the VFT temperature $T_0$ equal to the Kauzmann temperature
841: $T_K$. When comparing to experimental data, the
842: configurational-entropy based expressions provide a good description
843: at least over a restricted temperature range, but the resulting
844: estimates for the critical number of molecules composing a
845: cooperatively rearranging region is often found to be unphysically
846: small (only a few molecules at $T_g$).
847: \cite{R1}
848:
849: Building upon the early suggestion made by Goldstein, \cite{R62} and
850: others proposed that the apparent passage with decreasing temperature
851: from flow dynamics described by a mode-coupling approach to activated
852: dynamics such as pictured by the configurational-entropy theory of
853: Adam and Gibbs could be rationalized by considering the physics of
854: exploration of the energy landscape: see Fig. 12. The energy
855: landscape is the potential energy in configurational space. It can be
856: envisaged as an incredibly complex, multi-dimensional ($3N$ dimensions
857: for a system of $N$ particles) set of hills, valleys, basins,
858: saddle-points, and passage-ways around the hills. At constant volume
859: and constant number of particles, this landscape is independent of
860: temperature. However, the fraction of space that is statistically
861: accessible to the representative state point of the system decreases
862: with decreasing temperature, and the system becomes constrained to
863: deeper and deeper wells. (Recall that below the melting point the
864: deepest energy minima corresponding to the crystalline part of
865: configurational space must be excluded when studying the supercooled
866: liquid.) At low enough temperature, when the representative point of
867: the supercooled liquid is mostly found in fairly deep and narrow
868: wells, it seems reasonable to define a ``configurational entropy'' that
869: is proportional to the logarithm of the number of minima that are
870: accessible at a given temperature. The liquid configurations
871: corresponding to these accessible minima have been called ``inherent
872: structures'' and Stillinger and coworkers have devised a
873: gradient-descent mapping procedure to find the inherent structures and
874: study their properties in computer simulations. \cite{R63}
875: Interestingly, Stillinger has also shown, with fairly general
876: arguments, that if one is to use the above defined notion of
877: configurational entropy, an ``ideal glass transition'' of the type
878: commonly associated with the Kauzmann paradox, i. e., one
879: characterized by the vanishing of the configurational entropy at a
880: nonzero temperature, cannot occur for systems of limited molecular
881: weight and short-range interactions. \cite{R64}
882:
883: It may be more fruitful to investigate in place of the potential
884: energy landscape a free-energy landscape. Such a landscape can only be
885: defined if one is able to construct a free-energy functional by a
886: suitable coarse-graining procedure, as can be done for instance in the
887: case of mean-field spin glass models (see below). A free-energy
888: landscape is temperature-dependent, and it is important to note that
889: the ``configurational entropy'', also called ``complexity'',\cite{R65}
890: that one can define from the logarithm of the number of
891: accessible free-energy minima differs from the ``configurational
892: entropy'' computed from the potential energy landscape. In particular,
893: the behavior of the complexity is not restricted by the Stillinger
894: arguments given above.
895:
896: The ``landscape paradigm'' is very appealing in rationalizing many
897: observations on liquids and glasses and, more generally, in
898: establishing a framework to describe qualitatively slow dynamics in
899: complex systems that span a wide range of scientific fields.\cite{R66}
900: It has been used to motivate, in addition to the Adam-Gibbs theory and
901: other phenomenological approaches like the soft-potential model,
902: \cite{R67} simple stochastic models of transport based on master
903: equations. \cite{R68} Nevertheless, it has not so far offered a
904: way for elucidating the {\it physical} mechanism that is responsible
905: for the distinctive features of the viscous slowing down of
906: supercooled liquids.
907:
908: \subsection{Analogy with generalized spin glass models}
909: If one takes seriously the observation that the extrapolated
910: temperature dependence of both the viscosity (and $\alpha$-relaxation
911: times) and the ``configurational'' entropy (taken as the difference of
912: entropy between the liquid and the crystal) become divergent or
913: singular at essentially the same temperature $T_0\simeq T_K$(see Fig. 9),
914: one is naturally led to postulate the existence at this temperature of
915: an underlying thermodynamic transition, usually referred to as the
916: ``ideal glass transition''. Looking for analogies with phase
917: transitions in spin glasses is then appealing. However, the kind of
918: dynamic activated scaling that would be required to describe the
919: slowing down of relaxations when approaching the ideal glass
920: transition (see I-1) is not found in the most studied Ising spin
921: glasses. \cite{R69} Kirkpatrick, Thirumalai, and Wolynes argued that
922: generalized spin glass models, such as Potts glasses and random {\it
923: p}-spin systems, would be better candidates. \cite{R70,R71} The
924: random {\it p}-spin model, for instance, is defined by the following
925: hamiltonian:
926: \begin{equation}
927: \label{eq:12}
928: H=\sum_{i_1<i_2<\ldots<i_p}J_{i_1i_2\ldots i_p}\sigma_{i_1}\sigma_{i_2}\ldots\sigma_{i_p},
929: \end{equation}
930: where the $\sigma_{i}$'s are Ising variables and the couplings
931: $J_{i_1i_2\ldots i_p}$'s are quenched independent random variables that
932: can take positive and negative values according to a given probability
933: distribution. The behavior of these systems, {\it at least when
934: solved in the mean-field limit where the interactions between spins
935: have infinite range}, bears many similarities with the theoretical
936: description of glass-forming liquids outlined above. Indeed,
937: mean-field Potts glasses (with a number of states strictly larger than
938: 4) and mean-field p-spin models (with $p\geq 3$) have essentially the
939: following characteristics:
940: \begin{enumerate}
941: \item
942: At
943: high temperature, the system is in a fully disordered (paramagnetic)
944: state. At a temperature $T_D$, there appears an
945: exponentially large number of metastable ``glassy'' states whose overall
946: contribution to the partition function is equal to that of the
947: paramagnetic minimum. The free energy and all other static equilibrium
948: quantities are fully regular at $T_D$, but the dynamics have a
949: singularity of the exact same type as that found in the mode-coupling
950: theory of liquids. At $T_D$, the system is trapped in
951: one of the metastable free-energy minima, and ergocity is broken.
952:
953: \item Below $T_D$ , a peculiar situation occurs. The
954: partition function has contributions from, both, the paramagnetic
955: state and the exponentially large number of ``glassy'' (free-energy)
956: minima, the logarithm of which defines the configurational entropy or
957: complexity. This latter decreases as the temperature is lowered.
958: \item
959: At a nonzero temperature $T_s<T_D$,
960: the configurational entropy vanishes. The system undergoes a
961: {\it bona fide }thermodynamic transition to a spin-glass phase. The
962: transition has been termed ``random first order''
963: \cite{R70,R71} because it is second-order in the usual thermodynamic sense
964: (with, e. g., no latent heat), but shows a discontinuous jump in the
965: order parameter. (Technically, within the replica formalism, it
966: corresponds to a one-step replica symmetry breaking with a
967: discontinuous jump of the Edwards-Anderson order
968: parameter.\cite{R69,R70,R71})
969: \end{enumerate}
970:
971:
972: These mean-field systems are the simplest, analytically tractable
973: models found so far that display a high-temperature mode-coupling
974: dynamic singularity, a nontrivial free-energy landscape, and a
975: low-temperature ideal (spin) glass transition with an ``entropy
976: crisis''\footnote{They are also aging phenomena, as discussed in Ref
977: \cite{R5}.}. Analyzing them sheds light on the mode-coupling
978: approximation, whose validity for fluid systems is otherwise hard to
979: assess. The mode-coupling approximation becomes exact in the
980: mean-field limit, because the barriers separating the metastable
981: minima diverge (in the thermodynamic limit) at and below $T_D$ as a
982: result of the assumed infinite range of the interactions. One expects
983: that in a finite-range model, provided the same type of free-energy
984: landscape is still encountered, barriers are large but finite and
985: ergodicity is restored by thermally activated processes. Accordingly,
986: the dynamic transition is smeared out, and the activated relaxation
987: mechanisms that take over must be described in a nonperturbative way,
988: as suggested for instance by Kirkpatrick, Thirumalai, and Wolynes
989: \cite{R71} in their dynamic scaling approach based on entropic
990: droplets.
991:
992: An advantage of an analogy between glass-forming liquids and
993: generalized spin glasses\footnote{See also the frustrated percolation
994: model\cite{R72}.} is that the powerful tools that have been developed
995: in the theory of spin-glass models to characterize the order parameter
996: and the properties associated with the existence of a large number of
997: metastable glassy states, among which the replica formalism,
998: \cite{R69} can be used {\it mutatis mutandis }to{\it }study liquids and
999: glasses. \cite{R73} However, to make the analogy really successful,
1000: one must still find a short-range model (even more convincing would be
1001: a model without quenched disorder) that actually displays activated
1002: dynamic scaling and a random first-order transition and make progress
1003: in describing the slow relaxation.
1004:
1005:
1006: \subsection{Intrinsic frustration without randomness }
1007: Spin glasses, and related systems like orientational glasses, vortex
1008: glasses and vulcanized matter, \cite{R74} owe their fascinating
1009: behavior to two main ingredients: {\it randomness}, namely the
1010: presence of an externally imposed quenched disorder, and {\it
1011: frustration}, which expresses the impossibility of simultaneously
1012: minimizing all the interaction terms in the energy function of the
1013: system. Liquids and glasses (sometimes called "structural glasses" to
1014: stress the difference with spin glasses) have no quenched randomness,
1015: but frustration has been suggested as a key feature to explain the
1016: phenomena associated with glass formation. \cite{R75,R76,R77,R78,R79}
1017: Frustration in this context is attributed to a competition between a
1018: short-range tendency for the extension of a locally preferred order
1019: and global constraints that preclude the periodic tiling of the whole
1020: space with the local structure.
1021:
1022: The best studied example of such an intrinsic frustration concerns
1023: single-component systems of spherical particles interacting with
1024: simple pair potentials. What is usually called "geometric" or
1025: "topological" frustration can be more easily understood by comparing
1026: the situations encountered in $2$ and $3$ dimensions \cite{R76} (see
1027: Fig. 13). In $2$ dimensions, the arrangement of disks that is
1028: locally preferred, in the sense that it maximizes the density and
1029: minimizes the energy, is a hexagon of 6 disks around a central one,
1030: and this hexagonal structure can be extended to the whole space to
1031: form a triangular lattice. In $3$ dimensions, as was shown long ago
1032: by Frank, \cite{R80} the locally preferred cluster of spheres is an
1033: icosahedron; however, the $5$-fold rotational symmetry characteristic
1034: of icosahedral order is incompatible with translational symmetry, and
1035: formation of a periodic icosahedral crystal is forbidden. Geometric
1036: or topological frustration is thus absent in the $2$-dimensional case
1037: but present in the $3$-dimensional case. A consequence of this, for
1038: instance, is that crystallization is continuous, or weakly
1039: first-order, in $2$ dimensions (with some subtleties related to
1040: ordering in $2$ dimensions \cite{R81}) whereas it is strongly
1041: first-order in $3$ dimensions and accompanied by the breaking of the
1042: local icosahedral structure to make the face-centered-cubic or
1043: hexagonal-close-packed order that allows to tile space periodically.
1044: (In contrast, aligned cubes in 3 dimensions have no frustration and
1045: undergo a continuous freezing transition to a crystalline state.
1046: \cite{R82}). The geometric frustration that affects spheres in
1047: $3$-dimensional Euclidean space can be relieved in curved space with a
1048: specially tuned curvature; the creation of topological defects
1049: (disclination lines) can then be viewed as the result of forcing the
1050: ideal icosahedral ordering into "flat" space. This picture of sphere
1051: packing disrupted by frustration has been further developed in models
1052: for simple atomic systems and metallic glasses, \cite{R75,R76} and the
1053: slowing down of relaxations has been tentatively attributed to the
1054: topological constraints that hinder the kinetics of the entangled
1055: defect lines; \cite{R76} however, the treatment remains only
1056: qualitative and incomplete.
1057:
1058: A significant difficulty in applying the concept of geometric or
1059: topological frustration to supercooled liquids is that real fragile
1060: glass-formers are in general either mixtures or single-component
1061: systems of nonspherical molecules with a variety of shapes, all of
1062: which obscures the detailed mechanisms and constraints that are
1063: responsible for the frustration. Attempts have been made to get
1064: around this problem by proposing a more coarse-grained description of
1065: frustration\footnote{In addition, several ``toy models''possessing
1066: frustration, but no quenched disorder have been studied by computer
1067: simulation: see for instance Ref.\cite{R83}.}. \cite{R77,R78,R79}
1068:
1069: In Stillinger's "tear and repair" mechanism for relaxation and shear
1070: flow \cite{R78} and in the more recently introduced
1071: "frustration-limited domain theory",\cite{R79} frustration is
1072: described as the source of a strain free energy that opposes the
1073: spatial extension of the locally preferred structure and grows
1074: super-extensively with system size. It results in the breaking up of
1075: the liquid into domains, whose size and growth with decreasing
1076: temperature are limited by frustration, the weaker the frustration the
1077: larger the domains. The super-Arrhenius temperature dependence of the
1078: viscosity and $\alpha$-relaxation times and the heterogeneous nature of
1079: the dynamics are attributed to these domains (see also
1080: Ref.\cite{R77}). Progress has been made along these lines, by making
1081: use of a scaling approach based on the concept of avoided critical
1082: behavior\footnote{This approach differs from both those based upon
1083: spin-glass analogies and those in which the slow kinetics are
1084: attributed to frustration-induced entangled defect lines in that these
1085: others scale about a low-temperature characteristic point signifying
1086: ultimate slowing down, whereas in the frustration-limited domain
1087: theory the scaling is carried out about a high-temperature
1088: characteristic point signifying the initiation of anomalously slow
1089: dynamics. For the same reason, it also differs from the domain (or
1090: cluster) picture that has been proposed on the basis of an analogy
1091: between a supercooled liquid approaching the glass transition and a
1092: mean-field model with purely repulsive interactions near its
1093: spinodal.\cite{R85}}. \cite{R79,R84} However, the putative order
1094: variable characterizing the locally preferred structure of the liquid
1095: has not yet been properly identified, and, as in the case of the
1096: generalized spin-glass models discussed above, one must still give
1097: convincing evidence that the $3$-dimensional statistical-mechanical
1098: frustrated models that have been suggested as minimal theoretical
1099: descriptions do show the expected activated dynamics.
1100:
1101: \section{CONCLUSION}
1102: The viscous slowing down of supercooled liquids that leads to glass
1103: formation can be considered as a classical and thoroughly studied
1104: example of a "jamming process". In this review, we have stressed the
1105: distinctive features characterizing the phenomenon: strong,
1106: super-Arrhenius temperature dependence of the viscosity and the
1107: $\alpha$-relaxation times, nonexponential and heterogeneous character of
1108: the $\alpha$ relaxation, absence of marked changes in the structural
1109: (static) quantities, rapid decrease of the liquid entropy relative to
1110: that of the crystal, appearance of a sequence of steps (or regimes) in
1111: the relaxation functions. These features are common to most
1112: glass-forming liquids (with the exception of systems forming $2$- and
1113: $3$-dimensional networks of strong intermolecular bonds). We have
1114: also discussed the main theoretical approaches that have been proposed
1115: to describe the origin and the nature of the viscous slowing down and
1116: of the glass transition. We have emphasized the concepts, such as free
1117: volume, dynamic freezing and mode-coupling approximations,
1118: configurational entropy and (free) energy landscape, and frustration,
1119: that could be useful in other areas of physics where jamming processes
1120: are encountered.
1121: \begin{references}
1122:
1123: \bibitem{R1} Other aspects of supercooled liquids and glasses are
1124: presented in previous reviews:J. J{\"a}ckle, Rep. Prog. Phys. {\bf 49},
1125: 171 (1986),C. A. Angell, J. Non-Cryst. Solids {\bf 131-133}, 13
1126: (1991), M. D. Ediger, C. A. Angell, and S. R. Nagel,
1127: J. Phys. Chem. {\bf 100}, 13200 (1996).
1128:
1129: \bibitem{R2} More specific and detailed presentations can be found
1130: for instance in the {\it Proceedings of the International Meetings on
1131: ``Relaxations in Complex Systems''}: (i) K. L. Ngai and G. B. Wright
1132: Eds., Office of Naval Research, Arlington (1985); (ii) K. L. Ngai and
1133: G. B. Wright Eds.,{\bf }J. Non-Cryst. Solids {\bf 131-133} (1991);
1134: (iii) K. L. Ngai, E. Riande, and G. B. Wright Eds.,{\bf }J.
1135: Non-Cryst. Solids {\bf 172-174} (1994); (iv) K. L. Ngai Ed.,{\bf
1136: }J. Non-Cryst. Solids {\bf 235-238} (1998).
1137:
1138: \bibitem{R3} C. A. Angell, Science {\bf 267}, 1924 (1995).
1139: \bibitem{R4} L. C. E. Struik, {\it Physical Aging in Amorphous Polymers
1140: and Other Materials} (Elsevier, Amsterdam, 1978). G. W. Scherer,
1141: {\it Relaxation in Glasses and Composites } (John Wiley, New York,
1142: 1986). I. M. Hodge, J. Non-Cryst. Solids {\bf 169}, 211 (1994).
1143: \bibitem{R5} J. P. Bouchaud, L. Cugliandolo, J. Kurchan, and
1144: M. Mezard, in {\it Spin Glasses and Random Fields}, A. P. Young
1145: Ed. (World Scientific, Singapore, 1998), pg. 161. J. Kurchan, this
1146: volume.
1147: \bibitem{R6} J. D. Ferry, {\it Viscoelastic Properties of Polymers
1148: } (John Wiley, New York, 3$^{rd}$ Edition, 1980).
1149: \bibitem{R7} C. A. Angell, in {\it Relaxation in Complex Systems
1150: }, K. L. Ngai and G. B. Wright Eds.{\bf }(Office of Naval Research,
1151: Arlington, 1985), pg. 3.
1152: \bibitem{R8} D. Kivelson, G. Tarjus, X-L Zhao, and S. A. Kivelson,
1153: Phys. Rev. E {\bf 53}, 751 (1996).
1154:
1155: \bibitem{R9} F. Stickel, E. W. Fisher, and R. Richert,
1156: J. Chem. Phys. {\bf 102}, 6251 (1995); {\it ibid} {\bf 104}, 2043
1157: (1996).
1158:
1159: \bibitem{R10} D. S. Fisher, G. M. Grinstein,
1160: and A. Khurana, Physics Today Dec. 1988, 56 (1988).
1161:
1162: \bibitem{R11} P.K. Dixon, L. Wu, S.R. Nagel, B.D. Williams and
1163: J.P. Carini, Phys. Rev. Lett. {\bf 65}, 1108 (1990). L. Wu,
1164: P. K. Dixon, S. R. Nagel, B. D. Williams, and J. P. Carini,
1165: J. Non-Cryst. Solids {\bf 131-133}, 32 (1991). S. R. Nagel, in
1166: {\it "Phase Transitions and Relaxation in Systems with Competing Energy
1167: Scales"}, T. Riste and D. Sherrington Eds. (Kluwer Academic,
1168: Netherlands, 1993), pg. 259.
1169:
1170: \bibitem{R12} J. Villain, J. Phys. (Paris) {\bf 46}, 1843 (1985);
1171: D. S. Fisher, Phys. Rev. Lett. {\bf 56}, 416 (1986). A. T. Ogielski
1172: and D. A. Huse, Phys. Rev. Lett. {\bf 56}, 1298 (1986).
1173:
1174: \bibitem{R13} A. T{\"u}lle, H. Schober, J. Wuttke,
1175: and F. Fujara, Phys. Rev. E. {\bf 56}, 809 (1997).
1176:
1177: \bibitem{R14} See also: B. Frick, D. Richter, and Cl. Richter,
1178: Europhys. Lett. {\bf 9}, 557 (1989). E. Kartini, M. F. Collins,
1179: B. Collier, F. Mezei, and E. C. Swensson, Phys. Rev. B {\bf 54}, 6292
1180: (1996). R. Leheny, N. Menon, S. R. Nagel, K. Volin, D. L. Price, and
1181: P. Thiyagarjan, J. Chem. Phys. {\bf 105}, 7783 (1996).
1182:
1183: \bibitem{R15} S. R. Elliot, Nature {\bf 354}, 445 (1991); J. Phys. Condens. Matter {\bf 4}, 7661 (1992). D. Morineau, C. Alba-Simionesco, M.-C. Bellissent-Funel, M.-F. Lauthie, Europhys. Lett. {\bf 43}, 195 (1998).
1184:
1185: \bibitem{R16} W. Kauzmann, Chem. Rev. {\bf 43}, 219 (1948).
1186:
1187: \bibitem{R17} C. A. Angell, J. Res. Natl. Inst. Stand. Technol. {\bf 102}, 171 (1997).
1188: \bibitem{R18}{\bf }R. Richert and C. A. Angell, J. Chem. Phys. {\bf 108}, 9016 (1998).
1189: \bibitem{R19} W. Knaak, F. Mezei, and B. Farago, Europhys. Lett. {\bf 7}, 529 (1988).
1190: \bibitem{R20} G. P. Johari and M. Goldstein, J. Chem. Phys. {\bf 53},
1191: 2372 (1970).
1192:
1193: \bibitem{R21} E. R{\"o}ssler, Phys. Rev. Lett. {\bf 69},
1194: 1620 (1992). E. R{\"o}ssler, U. Warschewske, P. Eiermann, A. P. Sokolov, D. Quitmann, J. Non-Cryst. Solids, {\bf 172-174}, 113 (1994). E. Bartsch, F. Fujara, B. Geil, M. Kiebel, W. Petry, W. Schnauss, H. Sillescu, and J. Wuttke, Physica A {\bf 201}, 223 (1993).
1195:
1196: \bibitem{R22} E. R{\"o}ssler, V. N. Novikov, and A.P. Sokolov, Phase Transitions {\bf 63}, 201 (1997), and references therein.
1197:
1198: \bibitem{R23} U. Schneider, P. Lukenheimer, R. Brand, and A. Loidl, J. Non-Cryst. Solids {\bf 235-237}, 173 (1998).
1199:
1200: \bibitem{R24} See for instance the review by H. Sillescu, J. Non-Cryst. Solids, in press (1999).
1201:
1202: \bibitem{R25} K. Schmidt-Rohr and H.W. Spiess, Phys. Rev. Lett.
1203: {\bf 66}, 3020 (1991). A. Heuer, M. Wilhelm, H. Zimmermann, and
1204: H. W. Spiess,
1205: Phys. Rev. Lett.{\bf 75}, 2851 (1995). R. B{\"u}hmer, G. Hinze,
1206: G. Diezemann,
1207: B. Geil, and H. Sillescu, Europhys. Lett. {\bf 36}, 55 (1996).
1208: R. B{\"u}hmer, G. Diezemann, G. Hinze, and H. Sillescu,
1209: J. Chem. Phys. {\bf 108}, 890 (1998);
1210:
1211: \bibitem{R26} M. T. Cicerone and M. D. Ediger, J. Chem. Phys. {\bf
1212: 103}, 5684 (1995); C.-Y. Wang and M. D. Ediger, J. Phys. Chem. B , in
1213: press (1999).
1214:
1215: \bibitem{R27} B. Schiener, R. B{\"u}hmer, A Loidl, and R. V. Chamberlin,
1216: Science{\bf 274}, 752 (1996); B. Schiener, R. V. Chamberlin,
1217: G. Diezemann, and R. B{\"u}hmer, J. Chem. Phys.{\bf 107}, 7746 (1997).
1218: \bibitem{R28} F. Fujara, B. Geil, H. Sillescu and G. Fleischer,
1219: Z. Phys. B-Condensed Matter {\bf 88}, 195 (1992). I. Chang,
1220: F. Fujara, G. Heuberger, T. Mangel, and H. Sillescu,
1221: J. Non-Cryst. Solids {\bf 172-174}, 248 (1994).
1222:
1223: \bibitem{R29} M. T. Cicerone and M. D. Ediger, J. Chem. Phys. {\bf
1224: 104}, 7210 (1996). F. Blackburn, C.-Y. Wang, and M. D. Ediger,
1225: J. Phys. Chem. {\bf 100}, 18249 (1996).
1226: \bibitem{R30} M. T. Cicerone, F. R. Blackburn,
1227: and M. D. Ediger, J. Chem. Phys. {\bf 102}, 471 (1995). M. T. Cicerone
1228: and M. D. Ediger, J. Non-Cryst. Solids {\bf 235-238}, (1998).
1229: \bibitem{R31} U. Tracht, M. Wilhelm,
1230: A. Heuer, H. Feng, K. Schmidt-Rohr, and H.W. Spiess,
1231: Phys. Rev. Lett., 2727 (1998).
1232: \bibitem{R32} C. T. Moynihan and J. Schroeder,
1233: J. Non-Cryst. Solids {\bf 160}, 52 (1993).
1234: \bibitem{R33} G. Barut, P. Pissis, R. Pelster,
1235: and G. Nimtz, Phys. Rev. Lett.{\bf 80}, 3543 (1998).
1236: \bibitem{R34} M. L. Ferrer, C. Lawrence,
1237: B. G. Demirjian, D. Kivelson, C. Alba-Simionesco, and G. Tarjus,
1238: J. Chem. Phys. {\bf 109}, 8010 (1998), and references therein.
1239: \bibitem{R35} W. T. Laughlin and D. R. Uhlmann, J. Phys. Chem. {\bf 76}, 2317
1240: (1972). M. Cukierman, J. W. Lane, and and D. R. Uhlmann,
1241: J. Chem. Phys. {\bf 59}, 3639 (1973). R. G. Greet and D. Turnbull,
1242: J. Chem. Phys. {\bf 46}, 1243 (1967).
1243: \bibitem{R36} C. Hansen,
1244: F. Stickel, T. Berger, R. Richert, and E. W. Fischer,
1245: J. Chem. Phys. {\bf 107}, 1086 (1997)
1246: \bibitem{R37} R. R{\"o}ssler and
1247: A. P. Sokolov, Chem. Geol. {\bf 128}, 143 (1996); E. R{\"o}ssler,
1248: K.-U. Hess, and V. N. Novikov, J. Non-Cryst. Solids {\bf 223}, 207
1249: (1998).
1250: \bibitem{R38} W. Petry, E. Bartsch, F. Fujara, M. Kiebel, H. Sillescu, and B. Farago, Z. Phys.B {\bf 83}, 175 (1991).
1251:
1252: \bibitem{R39} See for instance the various reviews: J. L. Barrat and
1253: M. L. Klein, Annu. Rev. Phys. Chem. {\bf 42}, 23 (1991). J. P. Hansen,
1254: Physica A {\bf 201}, 138 (1993). W. Kob, J. Phys. Condens. Matter, in
1255: press (1999), as well as the special issue: {\it "Glasses
1256: and the glass transition. Challenges in Materials Theory and
1257: Simulation"},{\it } S. Glotzer Ed. (Computational Materials Science
1258: {\bf 4}, 1995).
1259:
1260: \bibitem{R40} P. Sindzingre and M. L. Klein, J. Chem. Phys. {\bf 96},
1261: 4681 (1992). P. H. Poole, F. Sciortino, U. Essmann, and H. E. Stanley,
1262: Nature {\bf 360}, 324 (1992). L. J. Lewis and G. Wahnstrom,
1263: Phys. Rev. E {\bf 50}, 3865 (1994). M. Wilson and P. A. Madden,
1264: Phys. Rev. Lett. {\bf 72}, 3033 (1994). J. Horbach, W. Kob, and
1265: K. Binder, Phil. Mag. B {\bf 77}, 297 (1998).
1266:
1267: \bibitem{R41}{\bf }K. Binder, J. Baschnagel, W. Paul, H. P. Wittmann,
1268: and W. Wolfgardt, Computational Materials Science {\bf 4}, 309 (1995),
1269: and references therein.
1270:
1271: \bibitem{R42}{\bf }T. Muranaka and Y. Hiwatari, Phys. Rev. E {\bf 51},
1272: R2735 (1995); D. N. Perera and P. Harrowell, Phys. Rev. E {\bf 52},
1273: 1694 (1995); R. Yamamoto and A. Onuki, J. Phys. Soc. Jpn. {\bf 66},
1274: 2545 (1997). A. I. Mel’cuk, R. A. Ramos, H. Gould, W. Klein, and
1275: R. D. Mountain, Phys. Rev. Lett. {\bf 75}, 2522 (1995).
1276:
1277: \bibitem{R43}{\bf }G. H. Fredrickson and H. C. Andersen,
1278: Phys. Rev. Lett. {\bf 53}, 1244 (1984). T. A. Weber and F.
1279: H. Stillinger, Phys. Rev. B {\bf 36}, 7043 (1986). F. Ritort,
1280: Phys. Rev. Lett. {\bf 75}, 1190 (1995). M. Foley and P. Harrowell, J.
1281: Chem. Phys. {\bf 98}, 5069 (1993). See also J. J{\"a}ckle,
1282: Prog. Theor. Phys. Suppl. {\bf 126}, 53 (1997).
1283:
1284: \bibitem{R44} K. L. Ngai and
1285: R. W. Rendell, in {\it Supercooled Liquids: Advances and Novel
1286: Applications}, J. Fourkas, D. Kivelson, U. Mohanty, and K. A. Nelson
1287: Eds. (ACS Symposium Series 676, Washington, 1997), pg. 45.
1288: \bibitem{R45} J. T. Bendler and M. F. Schesinger,
1289: in {\it ``Relaxations in Complex Systems''} K. L. Ngai and G. B.
1290: Wright Eds. (Office of Naval Research, Arlington, 1985), pg. 261.
1291: \bibitem{R46}
1292: M. H. Cohen and D. J. Turnbull, J. Chem. Phys. {\bf 31}, 1164 (1959);
1293: D. Turnbull and M. H. Cohen, {\it ibid} {\bf 34}, 120 (1961), {\bf
1294: 52}, 3038 (1970).
1295: \bibitem{R47} A. K. Doolittle, J. Appl. Phys. {\bf 22}, 1471 (1951).
1296: \bibitem{R48} G. S. Grest and M. H. Cohen,
1297: Adv. Chem. Phys. {\bf 48}, 455 (1981).
1298: \bibitem{R49} S. F. Edwards and Th. Vilgis, Phys. Scripta T {\bf 13},
1299: 7 (1986).
1300:
1301: \bibitem{R50} M. Goldstein, J. Phys. Chem. {\bf 77}, 667
1302: (1973). J. P. Johari and E. Whalley, Faraday Symp. Chem. Soc. {\bf 6},
1303: 23 (1973). S. A. Brawer, {\it Relaxation in Viscous Liquids and
1304: Glasses} (American Ceramic Society, Colombus, 1985). D. M. Colucci,
1305: G. B. McKenna, J. J. Filliben, A. Lee, D. B. Curliss, K. B. Bowman,
1306: and J. D. Russel, J. Polym. Sci. B: Polym. Phys. {\bf 351}, 1561
1307: (1997).
1308:
1309: \bibitem{R51} W. G{\"o}tze, in {\it "Liquids, Freezing, and the
1310: Glass Transition"}, J. P. Hansen, D. Levesque, and J. Zinn-Justin
1311: Eds. (North Holland, Amsterdam, 1991), pg. 287. W. G{\"o}tze and
1312: L. Sj{\"o}gren, Rep. Prog. Phys. {\bf 55}, 241 (1992). W. G{\"o}tze, J. Phys.:
1313: Cond. Mat. {\bf 11}, A1 (1999).
1314: \bibitem{R52} S. P. Das and G. F. Mazenko, Phys. Rev. A {\bf 34},
1315: 2265 (1986). B. Kim and G.F. Mazenko, Adv. Chem. Phys. {\bf 78}, 129 (1980).
1316: \bibitem{R53} K. Kawasaki,
1317: in {\it Phase Transitions and Critical Phenomena}, C. Domb and
1318: M. S. Green Eds; (Academic Press, New York, 1976), vol. 5a.
1319:
1320: \bibitem{R54} J. P. Bouchaud, L. Cugliandolo, J. Kurchan,
1321: and M. Mezard, Physica A {\bf 226}, 243 (1996). C. Z.-W. Liu and
1322: I. Oppenheim, Physica A {\bf 235}, 369 (1997).
1323: \bibitem{R55} For a sample of viewpoints and contributions, see:
1324: G. Li, W. M. Du, X.K. Chen, H. Z. Cummins, and N. J. Tao, Phys. Rev. A
1325: {\bf 45}, 3867 (1992); G. Li, W. M. Du, A. Sakai, and H. Z. Cummins,
1326: Phys. Rev. A {\bf 46}, 3343 (1992); H. Z. Cummins, W. M. Du,M. Fuchs,
1327: W. G{\"o}tze, S. Hildebrand, G. Li, and N. J. Tao, Phys. Rev. E {\bf 47},
1328: 4223 (1993). X. C. Zeng, D. Kivelson, and G. Tarjus, Phys. Rev. E
1329: {\bf 50}, 1711 (1994); P. K. Dixon, N. Menon, and S. R. Nagel,
1330: Phys. Rev. E {\bf 50}, 1717 (1994). A. P. Sokolov, W. Steffen, and
1331: E. R{\"o}ssler, Phys. Rev. E {\bf 52}, 5105 (1995). F. Mezei and
1332: M. Russina, J. Phys.: Cond. Mat. {\bf 11}, A341 (1999). J. Gapinski,
1333: W. Steffen, A. Patkowski, A. P. Sokolov, A. Kisliuk, U. Buchenau, M.
1334: Russina, F. Mezei, and A. Schober, preprint (1998).
1335:
1336: \bibitem{R56} M. Goldstein, J. Chem. Phys. {\bf 51}, 3728 (1969).
1337: \bibitem{R57} J. H. Gibbs, in {\it Modern Aspects of the Vitreous State}, J. D. Mackenzie Ed. (Butterworths, London, 1960), vol. 2, pg. 152.
1338: \bibitem{R58} G. Adam and J. H. Gibbs, J. Chem. Phys. {\bf 43}, 139 (1965).
1339:
1340: \bibitem{R59} J. H. Gibbs and E. A. Di Marzio,
1341: J. Chem. Phys. {\bf 28}, 373 (1958).
1342: \bibitem{R60} M. Goldstein, J. Chem. Phys. {\bf 64}, 4767 (1976).
1343: \bibitem{R61} N. Menon and S. R. Nagel, Phys. Rev. Lett. {\bf 74}, 1230 (1995).
1344: \bibitem{R62} C. A. Angell, J. Phys. Chem. Solids {\bf 49}, 863 (1988).
1345: \bibitem{R63} F. H. Stillinger and T. A. Weber, Science {\bf 225}, 983 (1983);
1346: F. H. Stillinger, {\it ibid} {\bf 267}, 1935 (1995); S. Sastry, P.
1347: G. Debenedetti, and F. H. Stillinger, Nature {\bf 393}, 554
1348: (1998). See also R. J. Speedy and P. G. Debenedetti, Mol. Phys. {\bf
1349: 88}, 1293 (1996). T. Keyes, J. Chem. Phys. {\bf 101}, 5081 (1992).
1350:
1351: \bibitem{R64} F. H. Stillinger, J. Chem. Phys. {\bf 88}, 7818 (1988).
1352: \bibitem{R65} R. G. Palmer, Adv. Phys. {\bf 31}, 669 (1982).
1353:
1354: \bibitem{R66} C. A. Angell, Nature {\bf 393}, 521 (1998).
1355: \bibitem{R67} U. Z{\"u}rcher and T. Keyes, in {\it Supercooled
1356: Liquids: Advances and Novel Applications}, J. Fourkas et al. Eds. (ACS
1357: Symposium Series 676, Washington, 1997), pg. 82.
1358:
1359: \bibitem{R68} S. A. Brawer, J. Chem. Phys. {\bf 81}, 954 (1984);
1360: H. B{\"a}ssler, Phys. Rev. Lett. {\bf 58}, 767 (1987); J. C. Dyre,
1361: Phys. Rev. Lett. {\bf 58}, 792 (1987); U. Mohanty, I. Oppenheim, and
1362: C. H. Taubes, Science {\bf 266}, 425 (1994); G. Diezemann,
1363: J. Chem. Phys. {\bf 107} (1997).
1364:
1365: \bibitem{R69} See for instance: K. Binder and A. P. Young,
1366: Rev. Mod. Phys. {\bf 58}, 801 (1986);
1367:
1368: {\it Spin Glass Theory and Beyond}, M. Mezard, G. Parisi,
1369: and M. A. Virasoro Eds. (World Scientific, Singapore, 1987);
1370: {\it Spin Glasses and Random Fields}, A. P. Young
1371: Ed. (World Scientific, Singapore, 1998).
1372:
1373: \bibitem{R70} T. R. Kirkpatrick and P. G. Wolynes, Phys. Rev. A{\bf
1374: 35}, 3072 (1987); Phys. Rev. B{\bf 36}, 8552 (1987); T. R. Kirkpatrick
1375: and D. Thirumalai, Phys. Rev. Lett. {\bf 58}, 2091 (1987); J. Phys. A
1376: {\bf 22}, L149 (1989).
1377: \bibitem{R71} T. R. Kirkpatrick, D. Thirumalai, and P. G. Wolynes,
1378: Phys. Rev. A {\bf 40}, 1045 (1989).
1379: \bibitem{R72} M. Nicodemi and A. Coniglio,
1380: J. Phys. A {\bf 30}, L187 (1997); A. Coniglio, A. de Candia, and
1381: M. Nicodemi, J. Phys.: Cond. Mat. {\bf 11}, A167 (1999).
1382: \bibitem{R73} G. Parisi,
1383: in {\it Supercooled Liquids: Advances and Novel Applications}, J.
1384: Fourkas, D. Kivelson, U. Mohanty, and K. A. Nelson Eds. (ACS Symposium
1385: Series 676, Washington, 1997), pg. 110. R. Monasson, Phys. Rev.
1386: Lett. {\bf 75}, 2847 (1995). S. Franz and G. Parisi, J. Physique
1387: (Paris) I {\bf 5}, 1401 (1995). M. Mezard and G. Parisi, J. Phys. A:
1388: Math. Gen. {\bf 29}, 6515 (1996); cond-mat/9812180 (1998).
1389: \bibitem{R74} See the articles collected in
1390: {\it Spin Glasses and Random
1391: Fields}, A. P. Young Ed. (World Scientific, Singapore, 1998).
1392:
1393: \bibitem{R75}
1394: M. Kleman and J. F. Sadoc, J. Phys. Lett. (Paris) {\bf 40}, L569
1395: (1979). J. P. Sethna, Phys. Rev. Lett. {\bf 51}, 2198
1396: (1983). S. Sachdev and D. R. Nelson, Phys. Rev. Lett. {\bf 53}, 1947
1397: (1984); Phys. Rev. B {\bf 32}, 1480 (1985). S. Sachdev, Phys. Rev. B
1398: {\bf 33}, 6395 (1986).
1399:
1400: \bibitem{R76} D. R. Nelson and F. Spaepen, Solid State Phys. {\bf 42},
1401: 1 (1989).
1402: \bibitem{R77} J. P. Sethna, J. D. Shore, and M. Huang,
1403: Phys. Rev. B {\bf 44}, 4943 (1991).
1404: \bibitem{R78} F. H. Stillinger,
1405: J. Chem. Phys. {\bf 89}, 6461 (1988); F.H. Stillinger and
1406: J. A. Hodgdon, Phys. Rev. E {\bf 50}, 2064 (1994).
1407:
1408: \bibitem{R79} D. Kivelson,
1409: S. A. Kivelson, X.-L. Zhao, Z. Nussinov, and G. Tarjus, Physica A {\bf
1410: 219}, 27 (1995); G. Tarjus and D. Kivelson, J. Chem. Phys. {\bf 103},
1411: 3071 (1995); D. Kivelson and G. Tarjus, Phil. Mag. B{\bf 77}, 245
1412: (1998).
1413: \bibitem{R80} F.C. Frank, Proc. Royal Soc. London {\bf 215A}, 43
1414: (1952).
1415: \bibitem{R81} K. J. Standburg, Rev. Mod. Phys. {\bf 60},161
1416: (1988).
1417: \bibitem{R82} E. A. Jagla, Phys. Rev. E {\bf 58}, 4701 (1998).
1418: \bibitem{R83} L. Gu
1419: and B. Chakraborty, Mat. Res. Symp. {\bf 455}, 229 (1997). B. Kim and
1420: S. J. Lee, Phys. Rev. Lett. {\bf 79}, 3709 (1997). S. J. Lee and
1421: B. Kim, cond-mat/9901077.
1422: \bibitem{R84}
1423: L. Chayes, V. J. Emery, S.A. Kivelson,
1424: Z. Nussinov, and G. Tarjus, Physica A {\bf 225}, 129
1425: (1996).
1426: \bibitem{R85}
1427: W. Klein, H. Gould, R. A. Ramos, I. Clejan, and
1428: A. I. Mel{\'{}}cuk, Physica A {\bf 205}, 738 (1994).
1429:
1430: \end{references}
1431:
1432: \begin{figure}
1433:
1434: \caption[99]{
1435: Super-Arrhenius $T$-dependence of the viscosity $\eta$ and
1436: $\alpha$-relaxation times $\tau_\alpha $ in glass-forming liquids. { \bf a)}
1437: Logarithm (base 10) of $\eta$ and $\tau_\alpha $ versus reduced inverse
1438: temperature $T_g/T$ for several liquids. For $GeO_2$, a system forming
1439: a network of strong intermolecular bonds, the variation is almost
1440: linear, whereas the other liquids (glycerol, m-toluidine, and
1441: ortho-terphenyl) are characterized, below some temperature $T^*$, by a
1442: strong departure from linear dependence: the behavior is nearly
1443: Arrhenius in the former case and super-Arrhenius in the latter. (Data
1444: taken from references cited in Ref.\cite{R8} and from
1445: C. Alba-Simionesco, private communication.) { \bf b) } Effective
1446: activation free energy $E(T)$, obtained from data shown in {\bf a)},
1447: as a function of inverse temperature. Both $E(T)$ and $T$ are divided
1448: by the crossover temperature $T^*$ shown in { \bf a) }.}
1449:
1450: \end{figure}
1451:
1452: \begin{figure}
1453: \caption[99]{
1454: Imaginary part of the dielectric susceptibility $\chi''$ of liquid
1455: m-toluidine versus $ log_{10}(\omega)$ for several
1456: temperatures close to $T_g$ ($T_g=183.5 K$). The inset
1457: shows that the $\alpha$ peak is broader than a Debye spectrum
1458: that would correspond to a purely exponential relaxation in
1459: time. (Data from C. Tschirwitz, E. R{\"o}ssler, and
1460: C. Alba-Simionesco, private communication.)}
1461: \end{figure}
1462:
1463: \begin{figure}
1464: \caption[99]{Static
1465: structure factor $S(Q)$ of liquid (deuterated) ortho-terphenyl at
1466: several temperatures from just below melting ($T_m=329 K$) to
1467: just above the glass transition ($T_g=243 K$). The inset shows
1468: the weak variation with temperature of $S(Q)$ for three values of $Q$
1469: indicated by the symbols above the $Q$-axis. (From Ref.\cite{R13}.)}
1470: \end{figure}
1471: \begin{figure}
1472: \caption[99]{
1473: Kauzmann's representation of the ``entropy paradox'': entropy difference
1474: $\Delta S$ between the liquid and the crystal (normalized by its value $\Delta
1475: S_m$ at the melting point) versus reduced temperature $T/T_m$. The
1476: break in the slopes of the full lines signals the glass transition at
1477: $T_g$. The dashed lines indicate an extrapolation of the entropy
1478: difference curves below $T_g$. Except for the strong, network-forming
1479: liquid $B_2O_3$, the extrapolated entropy difference vanishes at a
1480: nonzero temperature $T_K$. (Data from Refs.\cite{R16},\cite{R18}, and
1481: from H. Fujimori and C. Alba-Simionesco, private communication.)}
1482: \end{figure}
1483: \begin{figure}
1484: \caption[99]{Time dependence of the (normalized)
1485: dynamic structure factor $S(Q,t)/S(Q)$ of liquid
1486: $Ca_{0.4}K_{0.6}(NO_3)_{1.4}$ (for
1487: $Q\simeq 1.9{\AA}^{-1} $) at various
1488: temperatures. The continuous lines are obtained by Fourier
1489: transforming neutron time-of-flight data and the symbols represent
1490: neutron spin-echo results. When $T$ decreases, two steps separated by a
1491: plateau appear in the relaxation. (From Ref.\cite{R19}.)}
1492: \end{figure}
1493: \begin{figure}
1494: \caption[99]{
1495: Scaling plot for the imaginary part $\chi''$ of the dielectric
1496: susceptibility of several glass-forming liquids (glycerol, salol,
1497: propylene glycol, dibutyl-phtalate, $\alpha$-phenyl-o-cresol,
1498: ortho-terphenyl, ortho-phenylphenol). Experimental data similar to
1499: those shown in Fig. 2, but for 13 decades of frequency, are collapsed
1500: for all temperatures and all liquids onto the master-curve
1501: $w^{-1}log_{10}(\chi''\omega_\alpha/ \Delta\chi\omega)$ vs $w^{-1}(1+w^{-1})log_{10}(\omega/
1502: \omega_\alpha)$; $\omega_\alpha$ is the $\alpha$-peak position, $w$ is a shape factor that
1503: characterizes the deviation from Debye behavior, and $\Delta\chi$ is the
1504: static susceptibility. (From Ref.\cite{R11}.) }
1505: \end{figure}
1506: \begin{figure}
1507: \caption[99]{``Decoupling'' between rotational and
1508: translational time scales: logarithm (base 10) of the rotational
1509: ($D_r$) and translational ($D_t$) diffusion coefficients for
1510: ortho-terphenyl as a function of temperature and viscosity. $D_r$
1511: (diamonds) follows the viscosity at all temperatures whereas $D_t$
1512: (squares, triangles, and dots) departs from viscosity (and from $D_r$)
1513: below $T\simeq 290 K$. (From Ref.\cite{R28}).}
1514: \end{figure}
1515: \begin{figure}
1516: \caption[99]{
1517: Relative influence of temperature ($T$) and density ( $\rho$) on the
1518: viscous slowing down of liquid triphenyl phosphite. { \bf a) }
1519: $log_{10}(\eta)$ versus $\rho$ at constant $T$ for several isotherms.
1520: Also shown are the isobaric data at $P=1atm$ (under atmospheric
1521: pressure, the glass transition takes place at $T_g\simeq 200K$ and
1522: $\rho_g\simeq1.275g/cm^3$). {\bf b)} $log_{10}(\eta)$ versus $T$ at constant
1523: $\rho$ for several isochores. Note that the change of viscosity is much
1524: smaller with density (at constant $T$) than it is with temperature (at
1525: constant $\rho$) for the range of temperature and density characteristic
1526: of the liquid and supercooled liquid phases at atmospheric pressure.
1527: (From Ref.\cite{R34}.) }
1528: \end{figure}
1529: \begin{figure}
1530: \caption[99]{Illustration of the various
1531: choices of characteristic temperature for describing the viscous
1532: slowing down of liquid ortho-terphenyl. $log_{10}(\eta)$ is plotted
1533: versus inverse temperature. ``Extrapolation'' temperatures: $T_0(VFT$,
1534: see Eq. 1)$=200-202 K$ \cite{R18,R35}, $T_K$(Kauzmann, see I-4)=$204
1535: K$ \cite{R18}. ``Crossover'' temperatures: $T^*$(see Fig. 1 and
1536: III-5)=$350 K$\cite{R8}, $T_c$(MCT, see III-2)=$276-290 K$
1537: \cite{R38}, $T_A=455 K$ \cite{R36}, $T_B=290 K$ \cite{R37}, $T_X=289
1538: K$\cite{R37}. Also shown are the experimentally measured boiling
1539: ($T_b=610 K$), melting ($T_m=311 K$), and glass transition ($T_g=246
1540: K$) temperatures. (Viscosity data from Ref.
1541: \cite{R38}.) }
1542: \end{figure}
1543:
1544: \begin{figure}
1545: \caption[99]{Mode-coupling scenario of kinetic freezing and appearance
1546: of a $2$-step relaxation: time dependence of the (normalized)
1547: density-density correlation function for a schematic mode-coupling
1548: model. The curves from $A$ to $G$ correspond to the approach toward
1549: the dynamic singularity from the ergodic state. The other curves
1550: correspond to the nonergodic state. (From Ref.\cite{R51}.)}
1551: \end{figure}
1552: \begin{figure}
1553: \caption[99]{Breakdown of
1554: the ``idealized'' mode-coupling theory illustrated on $log_{10}(\eta)$
1555: versus $1/T$ for liquid ortho-terphenyl
1556: (see caption to Fig. 9). The continuous line is the mode-coupling fit
1557: to the experimental data. The predictions break down below a point at
1558: which the viscosity is of the order of 10 Poise; the dynamic
1559: singularity is not observed, and $T_c$ is interpreted as a
1560: crossover temperature.}
1561: \end{figure}
1562: \begin{figure}
1563: \caption[99]{Illustration of the putative relation
1564: between the rapid increase of $\alpha$-relaxation time ({\bf a}), the decrease
1565: of the entropy difference between the liquid and the crystal ({\bf
1566: b}), and the characteristic energy level on the (schematic) potential
1567: energy landscape ({\bf c}) for a fragile glass-former at various
1568: temperatures. The ideal glass level corresponds to the Kauzmann
1569: temperature (see I-4) and to an extrapolated divergence of the
1570: relaxation time; Goldstein's crossover (see III-3) takes place
1571: somewhat below point $2$. (From Ref.\cite{R17}.)}
1572: \end{figure}
1573: \begin{figure}
1574: \caption[99]{Illustration of geometric
1575: frustration for spherical particles. {\bf a)} Packing in 2 dimensions:
1576: equilateral triangles are preferred locally and combine to form a
1577: hexagonal local cluster that can tile space to generate a close-packed
1578: triangular lattice. {\bf b)} Packing in $3$ dimensions: tetrahedra are
1579: preferred locally and combine (with slight distortions) to form a
1580: regular icosahedral cluster; however, the $5$-fold symmetry axes of the
1581: icosahedron preclude a simple icosahedral space-filling lattice. (From
1582: Ref. \cite{R76}.) }
1583: \end{figure}
1584: \end{document}
1585: