1: \documentclass{epl}
2: \newcommand{\be}{\begin{equation}}
3: \newcommand{\bea}{\begin{eqnarray}}
4: \renewcommand{\d}{{\rm d}}
5: \newcommand{\ee}{\end{equation}}
6: \newcommand{\eea}{\end{eqnarray}}
7: \newcommand{\mean}[1]{\left\langle#1\right\rangle}
8: \renewcommand{\skew}{_{{\rm skew}}}
9: \newcommand{\eff}{_{{\rm eff}}}
10: \newcommand{\vd}{{\bf d}}
11: \renewcommand{\c}{{\bf c}}
12: \newcommand{\e}{{\bf e}}
13: \renewcommand{\k}{{\bf k}}
14: \newcommand{\x}{{\bf x}}
15: \newcommand{\y}{{\bf y}}
16: \newcommand{\0}{{\bf 0}}
17: \newcommand{\V}{{\bf V}}
18: \newcommand{\p}{\partial}
19: \newcommand{\dt}{\delta t}
20: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}
21: \newcommand{\frah}[2]{#1/#2}
22:
23: \title{Anomalous aging phenomena caused by drift velocities}
24: \shorttitle{Anomalous aging phenomena}
25: \author{J.M.~Luck\inst{1}\footnote{E-mail: luck@spht.saclay.cea.fr}
26: \and Anita Mehta\inst{2}\footnote{E-mail: anita@boson.bose.res.in}}
27: \institute{
28: \inst{1}Service de Physique Th\'eorique,
29: CEA Saclay, 91191 Gif-sur-Yvette cedex, France\\
30: \inst{2}ICTP, Strada Costiera 11, 34100 Trieste, Italy,
31: and S N Bose National Centre for Basic Sciences, Block JD Sector 3,
32: Salt Lake, Calcutta 700098, India}
33: \pacs{05.70.Ln}{Nonequilibrium thermodynamics, irreversible processes}
34: %\pacs{61.43.Fs}{Glasses}
35: \begin{document}
36: \maketitle
37: \begin{abstract}
38: We demonstrate via several examples that a uniform drift velocity
39: gives rise to anomalous aging,
40: characterized by a specific form for the two-time correlation functions,
41: in a variety of statistical-mechanical systems far from equilibrium.
42: Our first example concerns the oscillatory phase
43: observed recently in a model of competitive learning.
44: Further examples, where the proposed theory is exact,
45: include the voter model and the Ohta-Jasnow-Kawasaki theory for domain growth
46: in any dimension, and a theory for the smoothing of sandpile surfaces.
47: \end{abstract}
48:
49: \def\fun{
50: \begin{figure}
51: \vskip 7cm{\hskip 1.4cm}
52: \special{psfile=f1.ps hoffset=0 voffset=0 hsize=300 vsize=210}
53: \caption{Logarithmic plot of efficiency correlation function $C(t,0)$
54: against $t$.
55: Full line: least-square fit of data for $t\ge3$ (slope $=-0.34$).}
56: \label{fun}
57: \end{figure}}
58:
59: \def\fde{
60: \begin{figure}
61: \vskip 7cm{\hskip 1.4cm}
62: \special{psfile=f2.ps hoffset=0 voffset=0 hsize=300 vsize=210}
63: \caption{Logarithmic plot of efficiency correlation function $C(t,s)$
64: against $\tau=t-s$, for several values of the waiting time~$s$.}
65: \label{fde}
66: \end{figure}}
67:
68: \def\ftr{
69: \begin{figure}
70: \vskip 7cm{\hskip 1.4cm}
71: \special{psfile=f3.ps hoffset=0 voffset=0 hsize=300 vsize=210}
72: \caption{Logarithmic plot of the data of Figure~\ref{fde},
73: against $\tau^2/(\tau+0.55\,s)$.
74: Full line: least-square fit of data with abscissa $>2$ (slope $=-0.33$).}
75: \label{ftr}
76: \end{figure}}
77:
78: Aging is a characteristic feature of systems
79: such as glasses and spin glasses in their low-temperature phase.
80: More generally, aging is a consequence of slow dynamics,
81: which can be observed in a broad range of systems
82: far from equilibrium~\cite{angell}.
83: Aging phenomena are commonly described in terms of the two-time
84: correlation and response functions of the relevant variables,
85: which often exhibit scaling behavior in the two-time plane~\cite{aging}.
86:
87: In this Letter we demonstrate that this aging scenario is strongly affected
88: by the presence of a uniform drift velocity,
89: either intrinsic to the microscopic dynamics,
90: or imposed by external conditions.
91: On general grounds, a uniform velocity field is expected
92: to be relevant for long-time dynamics.
93: Consider for simplicity a diffusive relaxational dynamics,
94: characterized by a diffusion constant $D$.
95: To a non-zero velocity $V$ naturally corresponds a finite relaxation
96: time,~$D/V^2$, beyond which the dynamical evolution
97: can be qualitatively different from that with~$V=0$.
98:
99: Non-equilibrium phenomena in the presence of velocity fields,
100: either uniform (drift) or non-uniform (shear),
101: and more generally in a range of situations with a preferred direction,
102: have been described in recent works.
103: For instance, it has been shown that a shear flow
104: induces a strong anisotropy in coarsening dynamics~\cite{bc}.
105: The influence of drift on phase-ordering dyna\-mics
106: has also been investigated in Ref.~\cite{brag}.
107: The peculiar scaling properties of the flocking model of Ref.~\cite{ole}
108: are yet another consequence of directionality in non-equilibrium systems.
109:
110: Here we show that a variety of models
111: exhibit a novel type of non-equilibrium dynamics
112: in the presence of a uniform drift velocity $V$,
113: characterized by the specific scaling form~(\ref{th})
114: for the two-time correlation function.
115: We refer to this phenomenon as anomalous aging.
116: Uniform drift velocities cannot here be transformed away naively,
117: by invoking Galilean invariance.
118:
119: We first investigate in detail
120: the oscillatory phase described recently in the context of a model
121: of competitive learning~\cite{pre}.
122: The anomalous aging behavior of this specific model
123: will be explained at a quantitative level
124: by the combined effects of drift and diffusion.
125: Anomalous aging will then be shown to be a common feature
126: of a range of non-equilibrium models in the presence of a drift,
127: including the voter model, the OJK theory for domain growth,
128: and the smoothing of sandpile surfaces~\cite{mln,coupled}.
129:
130: We begin with a brief description of our model
131: of competitive learning~\cite{pre}.
132: Two types of individuals, considered as quick and slow learners,
133: are respectively classified by their efficiencies
134: (coded as Ising variables $\eta_\x=\pm1$),
135: and occupy the sites of a regular lattice.
136: The updating dynamics of the efficiencies involves two steps.
137: First, a majority rule favors the conversion of every individual
138: to the dominant type among its neighbors.
139: In the next step, there is a majority rule
140: involving the outcomes of individuals in some stochastic game,
141: favoring the conversion of every individual to the
142: type which has had the most favorable outcomes among its neighbors
143: at the last iteration.
144: Our model has two variants, the `interfacial' and the `cooperative',
145: which differ in the precise formulation of the second step.
146:
147: We consider the situation where both types
148: of individuals, while distinguishable, are equally slow learners,
149: so that no type wins {\it a priori}.
150: We refer to this case as coexistence, in analogy with ferromagnetic models.
151: Several kinds of ordered and disordered phases,
152: i.e., types of non-equilibrium steady states,
153: have been found in the phase diagram of both variants of the model~\cite{pre};
154: most phase transitions were found to belong either to the Ising
155: or to the voter universality classes.
156: However, an unusual type of coarsening was found in the
157: cooperative version of the model,
158: where the order parameter underwent more or less regular oscillations in time,
159: hence the name {\it oscillatory phase}.
160:
161: As the model of competitive learning is entirely defined by dynamical rules,
162: without reference to a Hamiltonian
163: (and therefore without the concept of detailed balance),
164: details of the dynamics can influence its steady states.
165: It turns out that
166: the updating scheme, including the order in which sites are updated
167: (sequential or parallel, ordered or random), {\em does}
168: affect the phase diagram.
169: From now on, we focus on the cooperative model at coexistence
170: on the square lattice,
171: characterized by a single parameter, $0\le p\le1$,
172: which represents an individual's probability of obtaining a successful outcome.
173: In Ref.~\cite{pre} ordered sequential dynamics were used,
174: and an oscillatory phase was found for $p<p_o$ $(p_o\approx0.136)$.
175: On the other hand,
176: parallel dynamics leads to a ferromagnetically ordered phase
177: for $p<p_c$ $(p_c\approx0.259)$,
178: while random sequential dynamics leads to a disordered phase
179: with very weak correlations between efficiencies.
180:
181: The special feature of ordered sequential dynamics
182: that is responsible for the oscillatory phase
183: turns out to be its {\it directionality}.
184: Consider a square sample with sites $\x=(m,n)$ with $m,n=1,\dots,L$.
185: For ordered sequential dynamics, neglecting boundary effects,
186: the next site updated after $\x_t=(m,n)$ is
187: $\x_{t+\dt}=(m,n+1)$, with $\dt=1/L^2$.
188: We define the directionality of this updating scheme as
189: $\vd=\mean{\x_{t+\dt}-\x_t}$.
190: We have $\vd=(0,1)$ for ordered sequential dynamics,
191: while $\vd=\0$ for parallel dynamics or random sequential dynamics.
192: Other updating schemes have been considered,
193: including alternating sequential dynamics:
194: the sample is first swept in a regular way,
195: such as e.g.~$\x_{t+\dt}=(m,n+1)$ if $\x_t=(m,n)$,
196: and then in a different regular way,
197: such as $\x_{t+\dt}=(m,n-1)$, or $\x_{t+\dt}=(m+1,n)$.
198: We have $\vd=\0$ in the first example,
199: and $\vd=\frac12(1,1)\ne\0$ in the second.
200: The model exhibits an oscillatory phase
201: at low~$p$ if, and only if, the updating scheme has
202: a net directionality $\vd\ne\0$.
203:
204: In the following, we restrict the analysis to the prototypical example
205: of ordered sequential dynamics.
206: Spin (efficiency) patterns are advected with a constant drift velocity $\V$,
207: due to the directionality $\vd$ of the dynamics.
208: The direction of $\V$ is given by the arrow of $\vd$,
209: while its strength $V$ depends e.g. on the parameter $p$.
210: If periodic boundary conditions are used, spin clusters wrap around the sample,
211: so that the magnetization oscillates more or less regularly,
212: with period $T_1=2L/V$, which defines a time scale associated with
213: the drift velocity.
214: This phenomenon was depicted in Figure~11 of Ref.~\cite{pre}.
215: A correlation analysis of a long magnetization record
216: allows for the accurate determination of $V$.
217: We find that $V$ decreases roughly linearly with $p$,
218: from $V\approx1.14$ at $p=0$ to $V\approx0.97$ at $p=p_o\approx0.136$.
219: The drift velocity
220: thus remains of order unity throughout the oscillatory phase,
221: including the critical point $p_o$,
222: where it does not seem to exhibit any kind of critical behavior.
223:
224: Next, a coarsening phenomenon takes place
225: if the system starts from a random initial state
226: with uncorrelated efficiencies.
227: The commonly accepted picture of phase-ordering dynamics
228: with no conservation law~\cite{bray} involves diffusive behavior,
229: hence the emergence of a second time scale, $T_2=L^2/D$,
230: with the diffusion coefficient $D$ being a microscopic constant.
231: The competition between these two time scales will turn out to be central
232: to anomalous aging [see paragraph below Eq.~(\ref{taueff})].
233:
234: Along the lines of investigations of aging properties in glassy systems,
235: we next measured the two-time correlation function
236: of the efficiency of a given individual,
237: $C(t,s)=\mean{\eta_\x(s)\eta_\x(t)}$,
238: with $0\le s$ (waiting time) $\le t=s+\tau$ (observation time).
239: We have chosen to work at $p=0$ for definiteness.
240: For each value of the waiting time $s$,
241: data were averaged over more than 1,000 samples
242: of size $300\times 300$.
243: The particular case $s=0$ corresponds to the correlation
244: of the efficiencies at time $t$ with their random initial values.
245: Figures~\ref{fun} and~\ref{fde} respectively show plots
246: of $C(t,0)$ and $C(t,s)$ for $s>0$.
247: Figure~\ref{fun} demonstrates that $C(t,0)$ decays exponentially,
248: just as in any equilibrium process,
249: with a microscopic relaxation time $\tau_0\approx1/0.34\approx2.9$,
250: while Figure~\ref{fde} clearly shows non-stationarity or aging effects:
251: correlations decay more and more slowly when the waiting time is increased.
252:
253: \fun
254:
255: \fde
256:
257: In order to elucidate these observations,
258: and especially to understand better the role of the drift velocity $V$
259: on the non-equilibrium dynamics of the model,
260: we propose the following phenomenological linear theory
261: meant to incorporate the combined effects of drift and diffusion,
262: in a natural fashion and at a minimal level of sophistication.
263: The efficiencies $\eta_\x$ are replaced by a continuous field $\phi(\x,t)$.
264: The equal-time correlation function $C(\x,t)=\mean{\phi(\x,t)\phi(\0,t)}$
265: is assumed to obey the diffusion equation
266: \be
267: \frah{\p C(\x,t)}{\p t}=D_1\nabla^2C(\x,t),
268: \label{eq1}
269: \ee
270: as the drift velocity is expected to have no effect on one-time observables.
271: A disordered initial state corresponds to
272: $C(\x,0)=\delta^{(d)}(\x)$ at $t=0$, in dimension $d$.
273: The two-time correlation function
274: $C(\x,t,s)=\mean{\phi(\x,t)\phi(\0,s)}$ is assumed
275: to be affected by both the drift and the diffusion mechanisms:
276: \be
277: \frah{\p C(\x,t,s)}{\p t}=D_2\nabla^2C(\x,t,s)-\V\cdot\nabla C(\x,t,s)
278: \label{eq2}
279: \ee
280: $(t\ge s)$, with initial condition $C(\x,s,s)=C(\x,s)$ at $t=s$.
281: The diffusion coefficients $D_1$ and $D_2$ are possibly different.
282: Convoluting the Green's functions of both equations,
283: and neglecting pre-exponential factors, we obtain
284: \be
285: C(t=s+\tau,s)\approx\exp\left(-\frac{V^2\tau^2}{4(D_1s+D_2\tau)}\right).
286: \label{th}
287: \ee
288:
289: This analytical result is in excellent agreement with our results above.
290: First, Eq.~(\ref{th}) predicts that $C(t,0)$ decays exponentially,
291: with a relaxation time $\tau_0=4D_2/V^2$, as seen in
292: Figure~\ref{fun}, where
293: the values of $V$ and $\tau_0$ yield $D_2\approx0.94$.
294: Second, in order to check the specific form of Eq.~(\ref{th}),
295: the data for $\ln C(t,s)$ have been plotted
296: as a function of $\tau^2/(\tau+\alpha s)$.
297: A convincing data collapse and linear behavior are observed
298: (Figure~\ref{ftr}) for the optimal value $\alpha\approx0.55$,
299: yielding $D_1=\alpha D_2\approx0.52$.
300: Moreover, Eq.~(\ref{th}) predicts that the slopes
301: of Figures~\ref{fun} and~\ref{ftr} are equal;
302: the measured slopes of the least-square fits, $-0.33$ and $-0.34$,
303: coincide within errors, in agreement with the above.
304:
305: \ftr
306:
307: Our result~(\ref{th}) is the centerpiece of {\it anomalous aging}.
308: The correlation function therein is clearly aging,
309: in the general sense that it is not stationary,
310: and decays more and more slowly as the waiting time $s$ is increased.
311: The expression~(\ref{th}) does not, however,
312: exhibit any simple scaling form as a function of the time ratio $\tau/s$,
313: since it involves the intrinsic time scale $\tau_0=4D_2/V^2$,
314: emblematic of the competition between drift and diffusion,
315: as expected from the discussion in the beginning of this Letter.
316:
317: For $\tau\ll s$,
318: the correlation function in Eq.~(\ref{th}) has a Gaussian scaling form
319: \be
320: C(t,s)\approx\exp\big(-\tau^2/(\tau_1s)\big)\qquad(\tau=t-s\ll s),
321: \label{bulk}
322: \ee
323: with $\tau_1=4D_1/V^2=\tau_0D_1/D_2$.
324: The correlation is characterized by an effective relaxation time
325: \be
326: \tau\eff(s)\approx(\tau_1s)^{1/2},
327: \label{taueff}
328: \ee
329: which diverges, albeit anomalously slowly, with the waiting time $s$.
330:
331: Eq.~(\ref{taueff}) can be recast as $(V\tau\eff)^2\approx4D_1s$,
332: to be put in perspective with the similar quadratic relationship
333: $(VT_1)^2=4DT_2=4L^2$ which holds between both time scales introduced above.
334: The stationary or time-translationally-invariant (TTI) regime
335: usually observed for $\tau\ll s$,
336: where the correlation function only depends on $\tau$,
337: is absent in the present case, having already
338: died away before our continuum description applied.
339: The TTI regime is also too small to be visible on Figure~\ref{fde}.
340:
341: For $\tau\gg s$, the result~(\ref{th}) only depends on $\tau$,
342: and falls off exponentially, as
343: \be
344: C(t,s)\approx\exp(-\tau/\tau_0)\qquad(\tau=t-s\gg s),
345: \label{tail}
346: \ee
347: just as $C(t,0)$.
348: This paradoxical emergence of a translationally-invariant regime
349: for {\it long} time separations is one of the reasons
350: we refer to this aging as {\em anomalous}.
351: The crossover between both kinds of behavior~(\ref{bulk}) and~(\ref{tail})
352: takes place for $\tau\sim s$,
353: where the correlation is already exponentially small,
354: of order $\exp(-s/\tau_0)$.
355:
356: In agreement with general expectations on aging phenomena~\cite{aging},
357: the result~(\ref{th}) can be recast as $C(t,s)\approx F[h(t)/h(s)]$,
358: with $h(t)=\exp\left(2(t/\tau_1)^{1/2}\right)$
359: and $F(x)=\exp\left(-(\ln x)^2\right)$.
360: Anomalous aging thus appears as a special case of sub-aging,
361: using the terminology of Ref.~\cite{aging},
362: as the effective age function $h(t)$ grows much more rapidly
363: than real time $t$.
364: The effective relaxation time~(\ref{taueff}) is recovered as
365: $\tau\eff(t)\sim h(t)/(\d h(t)/\d t)$.
366: A similar behavior, with the same age function $h(t)$,
367: has been found~\cite{let} in the spherical model with random
368: asymmetric couplings at zero temperature.
369: The ferromagnetic spherical model with a conserved order parameter~\cite{bepjb}
370: provides another interesting instance of anomalous aging,
371: involving a scaling function similar to the above $F(x)$.
372:
373: Since Eq.~(\ref{th}) is a simple consequence
374: of the combined effects of drift and diffusion,
375: it might be expected to apply to a variety of situations.
376: Several examples are discussed below,
377: where anomalous aging applies exactly in the regime of long times.
378:
379: First, our linear theory is exact for the voter model.
380: The dynamics of this model~\cite{L,vm,vp}
381: amounts to saying that a given site, say $\x$,
382: takes the opinion of any of its neighbors, say $\y=\x+\e$,
383: with the lattice vector $\e$ being chosen at random with some weight $w(\e)$.
384: If all vectors $\e$ have equal weights
385: ($w(\e)=1/z$ for all $\e$, with $z$ the coordination number),
386: the usual voter model is recovered,
387: with its well-known diffusive dynamics~\cite{vm,vp}.
388: In the general (asymmetric) case where some neighbors
389: are preferred over others,
390: patterns of opinions are advected with a velocity $\V=-\sum_\e\e\,w(\e)$.
391: We have shown that the equal-time and two-time correlation functions
392: obey Eqs.~(\ref{eq1}) and~(\ref{eq2}) in the continuum limit, with the identity
393: \be
394: D_1=2D_2,
395: \label{iden}
396: \ee
397: so that anomalous aging applies exactly to the voter model.
398:
399: Second, the Ohta-Jasnow-Kawasaki (OJK) theory~\cite{ojk},
400: a linear approximation commonly used in the study
401: of phase-ordering dynamics~\cite{bray},
402: also yields, in the presence of drift, the anomalous aging result~(\ref{th}).
403: The main difference between the OJK theory and our linear theory
404: is that linear equations are assumed to be obeyed
405: by the field $\phi(\x,t)$ itself in the OJK theory,
406: and by its correlations in our approach.
407: We have checked that, despite this difference, anomalous aging
408: (Eq.~(\ref{th})) applies exactly to the OJK theory with drift,
409: again with the simultaneous validity of Eq.~(\ref{iden}).
410:
411: Our last example relates to the smoothing of sandpile surfaces~\cite{coupled}
412: represented in earlier work by noisy coupled
413: equations for the fields $h(\x,t)$ and $\rho(\x,t)$.
414: Here $h(\x,t)$ represents the local height fluctuations of the sandpile surface
415: ( caused by clusters of `stuck' grains) defined with respect to a mean surface,
416: while $\rho(\x,t)$
417: represents the local density of mobile grains rolling down the
418: clusters~\cite{mln}.
419: The simplest example of these coupled equations is fully linear:
420: \be
421: \left\{
422: \begin{array}{l}
423: \frah{\p h}{\p t}=D_h\nabla^2h+\c\cdot\nabla h+\eta(\x,t),\\
424: \frah{\p\rho}{\p t}=D_\rho\nabla^2\rho-\c\cdot\nabla h,
425: \end{array}
426: \right.%}
427: \label{hr}
428: \ee
429: where $\eta(\x,t)$ is Gaussian white noise, such that
430: $\mean{\eta(\x,t)\eta(\x',t')}=\Delta^2\delta^{(d)}(\x-\x')\delta(t-t')$.
431: The equation for $h(\x,t)$ is nothing but the Edwards-Wilkinson (EW)
432: equation~\cite{EW} with a flow (drift) term added;
433: in the context of sandpiles, this represents the effect of tilt~\cite{mln}.
434:
435: In Ref.~\cite{coupled} it was found that
436: while $S_h(\k,t)$, a one-time quantity,
437: shows only the smoothing exponents of the EW equation
438: (i.e., the flow term can indeed be Galilean transformed away in this case),
439: $S_h(\k,\omega)$, a quantity which involves an integration over many times,
440: shows a clear crossover from EW exponents to a {\it smoothing}
441: state with $\alpha=\beta=0$, when the drift velocity is large enough.
442: Evaluating the two-time correlation function of the $h$-field, we find
443: \be
444: C_h(t,s)
445: =\mean{h(\0,t)h(\0,s)}
446: =\Delta^2\int_0^s\d u\,\big(4\pi D_h(t+s-2u)\big)^{-d/2}
447: \exp\left(-\frac{c^2(t-s)^2}{4D_h(t+s-2u)}\right).
448: \ee
449: In the scaling regime, where both $s$ and $t-s=\tau$ are large
450: compared to the intrinsic time scale $\tau_0=4D_h/c^2$,
451: the integral is dominated by microscopic times $u\sim\tau_0$.
452: We thus recover the anomalous aging result~(\ref{th}),
453: with $D_1=2D_h$, $D_2=D_h$, so that Eq.~(\ref{iden}) again holds.
454:
455: A corollary of this result is that anomalous aging
456: (or, equivalently, the anomalous smoothing referred to
457: in Ref.~\cite{coupled}) will manifest itself at experimentally
458: observable times only for large~$c$ -- this, in the sandpile context,
459: would correspond to unstable packings
460: of grains susceptible to rolling under tilt~\cite{mln}.
461: It would be interesting to see if measurements on such a surface indicate
462: that it is composed of large domains characterized by
463: EW roughening exponents stacked end on end.
464: Physically, this is one possible realization
465: of the anomalous smoothing scenario -- {\em inter-domain} roughness
466: measurements would here show no significant variation in roughness,
467: while {\em intra-domain} measurements would show EW roughening exponents.
468:
469: To sum up, we have demonstrated that a uniform drift velocity
470: is relevant for the long-time dynamics of systems far from equilibrium.
471: We have shown that the prediction~(\ref{th})
472: of a simple linear theory, referred to as anomalous aging,
473: provides an accurate or even exact description of the non-equilibrium
474: dynamics of a wide variety of systems, including the oscillatory phase
475: observed recently in a model of competitive learning, the voter model,
476: the Ohta-Jasnow-Kawasaki theory for domain growth,
477: and a theory for the smoothing of sandpile surfaces.
478:
479: \acknowledgments
480:
481: It is a pleasure for us to thank L. Berthier, L. Cugliandolo, S. Franz,
482: and M. Laessig for fruitful discussions.
483:
484: \begin{thebibliography}{99}
485:
486: \bibitem{angell} C.A. Angell, Science {\bf 267}, 1924 (1995).
487:
488: \bibitem{aging} For reviews, see: E. Vincent, J. Hammann, M. Ocio,
489: J.P. Bouchaud, and L.F. Cugliandolo,
490: in {\it Complex Behavior of Glassy Systems},
491: Springer Lecture Notes in Physics {\bf 492}, 184 (1997),
492: preprint cond-mat/9607224;
493: J.P. Bouchaud, L.F. Cugliandolo, J. Kurchan, and M. M\'ezard,
494: in {\it Spin Glasses and Random Fields},
495: Directions in Condensed Matter Physics, vol.~{\bf 12},
496: ed. A.P. Young (World Scientific, Singapore, 1998),
497: preprint cond-mat/9702070.
498:
499: \bibitem{bc} A.J. Bray and A. Cavagna, J. Phys. A {\bf 33}, L305 (2000);
500: A. Cavagna, A.J. Bray, and R.D.M. Travasso, Phys. Rev. E {\bf 62}, 4702 (2000).
501:
502: \bibitem{brag} J. Bragard, P.L. Ramazza, F.T. Arecchi, S. Boccaletti,
503: and L. Kramer, Phys. Rev. E {\bf 61}, R6045 (2000).
504:
505: \bibitem{ole} O.J. O'Loan and M.R. Evans, J. Phys. A {\bf 32}, L99 (1999).
506:
507: \bibitem{pre} Anita Mehta and J.M. Luck, Phys. Rev. E {\bf 60}, 5218 (1999).
508:
509: \bibitem{mln} Anita Mehta, J.M. Luck, and R.J. Needs, Phys. Rev. E {\bf 53},
510: 92 (1996).
511:
512: \bibitem{coupled} P. Biswas, A. Majumdar, Anita Mehta, and J.K. Bhattacharjee,
513: Physica {\bf A 248}, 379 (1997).
514:
515: \bibitem{bray} A.J. Bray, Adv. Phys. {\bf 43}, 357 (1994).
516:
517: \bibitem{let} L.F. Cugliandolo, J. Kurchan, P. Le Doussal, and L. Peliti,
518: Phys. Rev. Lett. {\bf 78}, 350 (1997).
519:
520: \bibitem{bepjb} L. Berthier, Eur. Phys. J. B {\bf 17}, 689 (2000).
521:
522: \bibitem{L} T.M. Liggett, {\it Interacting Particle Systems} (Springer,
523: New-York, 1985).
524:
525: \bibitem{vm} J.T. Cox and D. Griffeath, Ann. Probab. {\bf 11}, 876 (1983);
526: Contemp. Math. {\bf 41}, 55 (1985);
527: M. Bramson, J.T. Cox, and D. Griffeath, Probab. Th. Rel. Fields {\bf
528: 77}, 613 (1988); J.T. Cox, Ann. Probab. {\bf 16}, 1559 (1988).
529:
530: \bibitem{vp} M. Scheucher and H. Spohn, J. Stat. Phys. {\bf 53}, 279 (1988);
531: P.L. Krapivsky, Phys. Rev. A {\bf 45}, 1067 (1992).
532:
533: \bibitem{ojk} T. Ohta, D. Jasnow, and K. Kawasaki, Phys. Rev. Lett. {\bf 49},
534: 1223 (1982).
535:
536: \bibitem{EW} S.F. Edwards and D.R. Wilkinson, Proc. Roy. Soc. London
537: {\bf A 381}, 17 (1982).
538:
539: \end{thebibliography}
540: \end{document}
541: