1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \catcode`\@=11
4: \topmargin 0pt
5: \oddsidemargin 0pt
6: \headheight 0pt
7: \headsep 0pt
8: \textheight 9in
9: \textwidth 6.25in
10: \marginparwidth 0.875in
11: \def\numberbysection{\@addtoreset{equation}{section}
12: \def\theequation{\thesection.\arabic{equation}}}
13: \numberbysection
14: \def\baselinestretch{1.05}
15: \newcommand{\beq}{\begin{equation}}
16: \newcommand{\beqa}{\begin{eqnarray}}
17: \newcommand{\eeq}{\end{equation}}
18: \newcommand{\eeqa}{\end{eqnarray}}
19: \renewcommand{\a}{\alpha}
20: \newcommand{\abs}[1]{\vert#1\vert}
21: \renewcommand{\d}{{\rm d}}
22: \newcommand{\de}{_{(2)}}
23: \newcommand\dis[1]{\displaystyle#1}
24: \newcommand\ds[1]{\frad{\d#1}{\d s}}
25: \newcommand\dt[1]{\frad{\d#1}{\d t}}
26: \newcommand{\e}{{\rm e}}
27: \newcommand{\evec}{{\bf e}}
28: \newcommand{\eps}{\varepsilon}
29: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}
30: \newcommand{\g}{\gamma}
31: \newcommand{\gbar}{\overline{g}}
32: \newcommand{\infy}{_{(\infty)}}
33: \newcommand{\infymax}{_{(\infty){\rm max}}}
34: \newcommand{\li}{_{\rm lim}}
35: \renewcommand{\max}{_{\rm max}}
36: \newcommand{\m}{{\bf m}}
37: \newcommand{\mean}[1]{\langle#1\rangle}
38: \newcommand{\meansur}[1]{\langle\!\langle#1\rangle\!\rangle}
39: \newcommand{\n}{{\bf n}}
40: \newcommand{\pl}{$\bullet$}
41: \newcommand{\s}{\sigma}
42: \newcommand{\un}{_{(1)}}
43: \newcommand{\vi}{$\circ$}
44: \renewcommand{\L}{\Lambda}
45: \newcommand{\N}{{\cal N}}
46: \newcommand{\0}{{\bf 0}}
47: \begin{document}
48: \centerline{\Large\bf A deterministic model of competitive cluster growth:}
49: \vspace{.3cm}
50: \centerline{\Large\bf glassy dynamics, metastability and pattern formation}
51: \vspace{1cm}
52:
53: \centerline{\large J.M.~Luck$^{a,}$\footnote{luck@spht.saclay.cea.fr}
54: and Anita Mehta$^{b,}$\footnote{anita@boson.bose.res.in}}
55: \vspace{1cm}
56:
57: \noindent $^a$Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},
58: CEA Saclay, 91191 Gif-sur-Yvette cedex, France
59:
60: \noindent $^b$S.N.~Bose National Centre for Basic Sciences, Block JD,
61: Sector 3, Salt Lake, Calcutta 700098, India
62: \vspace{1cm}
63:
64: \begin{abstract}
65: We investigate a model of interacting clusters which compete for growth.
66: For a finite assembly of coupled clusters, the largest one always wins,
67: so that all but this one die out in a finite time.
68: This scenario of `survival of the biggest'
69: still holds in the mean-field limit, where the model exhibits
70: glassy dynamics, with two well separated time scales,
71: corresponding to individual and collective behaviour.
72: The survival probability of a cluster
73: eventually falls off according to the universal law $(\ln t)^{-1/2}$.
74: Beyond mean field, the dynamics exhibits both aging and metastability,
75: with a finite fraction of the clusters
76: surviving forever and forming a non-trivial spatial pattern.
77: \end{abstract}
78: \vfill
79: %\noindent To be submitted for publication to the European Physical Journal B
80:
81: \noindent P.A.C.S.: 05.45.--a, 47.54.+r, 89.75.--k, 64.60.My.
82:
83: \newpage
84: \setcounter{footnote}{0}
85: \section{Introduction}
86:
87: Non-equilibrium dynamics can lead to counter-intuitive situations.
88: One is used, for example, to the premise of equilibration:
89: in a system with unequally distributed masses,
90: the effect of most `physical' interactions
91: would be to bring the system to an equilibrium state
92: where masses are distributed equally.
93: This equilibration principle is known to fail in some physical instances,
94: mostly in the presence of long-range forces,
95: the prototypical example being gravitational forces.
96: In fact, as is well known, the effect of gravitation is to amplify forever
97: the contrasts in mass distribution throughout the Universe~\cite{pee}.
98:
99: In this work, we interest ourselves in an extreme case of disequilibration.
100: The model investigated below
101: deals with immobile interacting clusters which compete for growth.
102: Although it arose in an astrophysical context,
103: that of mass accretion by black holes coupled by the radiation field
104: in a brane world~\cite{archan,I},
105: its emergent features are relevant to a far wider range of problems.
106: The present model is strongly out-of-equilibrium
107: and it obeys no mass conservation law.
108: A variety of transient behaviour is therefore possible:
109: for example, two interacting clusters can both decay,
110: or both grow before one of them dies out.
111: At late stages, the model follows the {\it survival of the biggest} scenario,
112: an example of Darwinism in a physical system.
113: In the mean-field geometry,
114: the largest cluster generically wins out over all the rest;
115: in finite dimensions, one has the possibility that infinitely many clusters
116: survive and grow forever, provided each of them is isolated,
117: in a sense that will be clearer later on.
118: This is actually quite meaningful
119: in the original astrophysical context, since it could, with appropriate
120: modifications, be adopted to model the scenario of primordial black holes
121: evolving to a size such that they survive in the present era.
122:
123: The present model, defined in Section~\ref{model},
124: deals with an assembly of pointlike, immobile clusters,
125: which are entirely characterised by their masses.
126: Cluster masses evolve according to coupled deterministic,
127: non-linear first-order equations.
128: We address a medley of situations ranging from finite
129: assemblies of clusters, to the thermodynamic limit, examined
130: both in the mean-field geometry and on a lattice
131: with nearest-neighbour interactions.
132: Section~\ref{one} describes the dynamical behaviour
133: of a single isolated cluster:
134: a large enough cluster, whose initial mass exceeds some threshold,
135: grows forever, whereas a smaller one evaporates
136: and disappears in a finite time.
137: Section~\ref{two} concerns our findings on two interacting clusters,
138: and more generally finitely many coupled clusters:
139: the generic scenario is then the survival of the biggest, so that
140: the largest cluster wins out over all the rest.
141: The mean-field regime of a large collection of weakly coupled clusters
142: is investigated in Section~\ref{mft}.
143: The system exhibits aging and glassy dynamics,
144: involving two well-separated time scales.
145: The cluster survival probability decays according to
146: the universal law $(\ln t)^{-1/2}$.
147: In Section~\ref{latt}, we examine the model with nearest-neighbour interactions
148: in finite dimension: the dynamics now exhibits both aging and metastability.
149: The finite fraction of survivors, i.e., clusters which survive and grow forever,
150: builds a non-trivial spatial pattern.
151: In the Discussion (Section~\ref{discussion}),
152: we put our results in perspective with other growth models.
153:
154: \section{The model}
155: \label{model}
156:
157: The model investigated in this work
158: is a direct generalisation of that derived in~\cite{archan,I}.
159: Consider $n$ pointlike, immobile clusters,
160: which are entirely characterised by their time-dependent masses
161: $m_i(t)$ for $i=1,\dots,n$.
162: The cluster masses evolve according to the following
163: coupled deterministic, first-order equations:
164: \beq
165: \dt{m_i}=\left(\frac{\a}{t}-\frac{1}{t^{1/2}}\sum_jg_{ij}\dt{m_j}\right)m_i
166: -\frac{1}{m_i}.
167: \label{dtm}
168: \eeq
169:
170: These dynamical equations were originally written to model the
171: kinetics of black hole growth in a radiation fluid~\cite{archan,I}.
172: In that context, they only hold after some microscopic initial time $t_0$.
173: The positive (gain) term in the right-hand side of~(\ref{dtm}) represents
174: mass accretion by the black hole from the surrounding fluid.
175: The accretion rate in the large parenthesis is the sum of the free rate
176: for an isolated black hole, proportional to the parameter~$\a>1/2$,
177: and of the rate induced by all the other black holes via the surrounding fluid.
178: The coupling $g_{ij}$ between black holes $i$ and $j$
179: is proportional to the inverse square distance $d_{ij}^2(t_0)$
180: between them at the initial time $t_0$.
181: The negative (loss) term in the right-hand side of~(\ref{dtm}) represents
182: evaporation due to Hawking radiation.
183:
184: In this paper, the dynamical equations~(\ref{dtm}),
185: or equivalently~(\ref{dsx}), are now seen as representing
186: competitive cluster growth.
187: The functional form of the original equations
188: derived in~\cite{I} is kept unchanged for definiteness.
189: The symmetric matrix of couplings $g_{ij}$ reflects the underlying geometry.
190: Succeeding sections will deal with a raft of scenarios
191: of ever-increasing complexity, ranging from the dynamics of two coupled
192: clusters to that of infinitely many.
193:
194: It turns out to be convenient to switch from physical time $t$ to reduced
195: (logarithmic) time
196: \beq
197: s=\ln\frac{t}{t_0},
198: \eeq
199: so that the initial time $t_0$ is mapped onto the origin $s=0$.
200: Furthermore, we introduce for convenience the reduced masses and square masses:
201: \beq
202: x_i=\frac{m_i}{t^{1/2}},\qquad y_i=x_i^2=\frac{m_i^2}{t}.
203: \eeq
204: The dynamical equations~(\ref{dtm}) then become the following
205: {\it autonomous} equations
206: \beq
207: \ds{x_i}\equiv x'_i
208: =\left(\frac{2\a-1}{2}-\sum_jg_{ij}\left(\frac{x_j}{2}+x'_j\right)\right)x_i
209: -\frac{1}{x_i}
210: \label{dsx}
211: \eeq
212: for the reduced masses $x_i(s)$,
213: which exhibit no explicit dependence on the reduced time~$s$~\cite{I}.
214: Throughout the following, accents will denote differentiation with respect
215: to the reduced time $s$.
216: It will also be assumed that the couplings~$g_{ij}$
217: are small enough, so that
218: \beq
219: {\rm det}\,(\delta_{ij}+g_{ij}\,x_i)_{i,j=1,\dots,n}>0.
220: \label{cd}
221: \eeq
222: When this inequality holds,
223: the time derivatives $x'_i$ can be solved explicitly from the
224: implicit dynamical equations~(\ref{dsx}), so that the dynamics is regular.
225: Whenever the regularity condition~(\ref{cd}) holds at $s=0$,
226: it turns out to be preserved by the dynamics.
227:
228: \section{One isolated cluster}
229: \label{one}
230:
231: The simplest situation is that of a single isolated cluster of mass $m(t)$.
232: The dynamical equation~(\ref{dtm}) reads
233: \beq
234: \dt{m}=\frac{\a m}{t}-\frac{1}{m}.
235: \eeq
236: The dynamical equation~(\ref{dsx}) simplify to the following ones
237: for the reduced mass $x(s)$ and square mass $y(s)$:
238: \beqa
239: &&x'=\frac{2\a-1}{2}\,x-\frac{1}{x},\\
240: &&y'=(2\a-1)y-2.\label{yoneds}
241: \eeqa
242: Equation~(\ref{yoneds}) is the easier to solve.
243: It yields at once
244: \beq
245: y(s)=y_\star+(y_0-y_\star)\e^{(2\a-1)s},
246: \label{yone}
247: \eeq
248: where $m_0=m(t_0)$ and $y_0=m_0^2/t_0$ are the initial values
249: of $m(t)$ and $y(t)$, respectively, whereas
250: \beq
251: y_\star=\frac{2}{2\a-1}
252: \label{ystar}
253: \eeq
254: is the unstable fixed point of~(\ref{yoneds}).
255:
256: Returning to physical variables,~(\ref{yone}) reads
257: \beq
258: m(t)^2=y_\star t+(m_0^2-y_\star t_0)\left(\frac{t}{t_0}\right)^{2\a}.
259: \label{m1}
260: \eeq
261: This implies that for all $\a>1/2$, we have two kinds of behaviour:
262:
263: \begin{itemize}
264:
265: \item
266: Large clusters, whose initial mass is such that $y_0>y_\star$,
267: i.e., $m_0$ is larger than the mass threshold
268: \beq
269: m_\star=(y_\star t_0)^{1/2}=\left(\frac{2t_0}{2\a-1}\right)^{1/2},
270: \label{thre}
271: \eeq
272: are {\it survivors:} they survive and keep on growing forever.
273: Equation~(\ref{m1}) reads alternatively
274: \beq
275: m(t)^2=m_\star^2\,\frac{t}{t_0}
276: +(m_0^2-m_\star^2)\left(\frac{t}{t_0}\right)^{2\a}.
277: \eeq
278: The second term is the leading one at late times, for all $\a>1/2$.
279:
280: \item
281: Small clusters, whose initial mass is below the threshold: $y_0<y_\star$,
282: i.e., $m_0<m_\star$, evaporate and die out in a finite reduced time,
283: \beq
284: s(y_0)=\frac{1}{2\a-1}\ln\frac{y_\star}{y_\star-y_0},
285: \label{sdis}
286: \eeq
287: which diverges logarithmically
288: as the mass threshold $m_\star$ is approached from below.
289: The corresponding physical time,
290: \beq
291: t(y_0)=t_0\left(\frac{y_\star}{y_\star-y_0}\right)^{1/(2\a-1)},
292: \eeq
293: diverges as a power law.
294:
295: \end{itemize}
296:
297: Even in this simple case of independent clusters, we get an indication
298: of a Darwinian scenario (where size, i.e., mass replaces fitness):
299: the biggest clusters survive, while the smaller ones die out.
300:
301: Consider now a very large assembly of isolated, i.e., non-interacting, clusters,
302: characterised by the (continuous) probability distribution function $P(y_0)$
303: of their initial square masses $y_0$.
304: One of the quantities of most interest is the survival probability $S(s)$,
305: defined as the fraction of the clusters which have survived
306: up to reduced time $s$.
307: Surviving clusters are exactly those whose initial square mass
308: obeys $y_0>Y(s)$, where the time-dependent threshold $Y(s)$
309: is the inverse of $s(y_0)$ introduced in~(\ref{sdis}):
310: \beq
311: Y(s)=\left(1-\e^{-(2\a-1)s}\right)y_\star.
312: \label{bigy}
313: \eeq
314: The survival probability at time $s$ reads therefore
315: \beq
316: S(s)=\int_{Y(s)}^\infty P(y_0)\,\d y_0.
317: \label{ss}
318: \eeq
319:
320: The limit survival probability\footnote{The subscript $(1)$ recalls
321: that this result holds for isolated clusters.}
322: $S\un$ is defined as the fraction of survivors,
323: i.e., clusters which survive and grow forever.
324: These are the clusters whose initial mass is above the threshold
325: $y_\star$ introduced in~(\ref{thre}).
326: We thus obtain
327: \beq
328: S\un=\lim_{s\to\infty}S(s)=\int_{\dis{y_\star}}^\infty P(y_0)\,\d y_0.
329: \label{s1}
330: \eeq
331:
332: For simplicity, we shall often consider in the following
333: an exponential distribution of initial square masses:
334: \beq
335: P(y_0)=\mu\,\e^{-\mu y_0}.
336: \label{initexpo}
337: \eeq
338: The result~(\ref{s1}) then reads
339: \beq
340: S\un=\e^{-\mu y^\star}.
341: \label{sunexpo}
342: \eeq
343:
344: \section{Two interacting clusters}
345: \label{two}
346:
347: Until now, we have considered only independent clusters:
348: these survive or not, depending on their initial masses.
349: We now turn to the more interesting situation of interacting clusters.
350: The form of the interactions was derived in earlier work~\cite{I}
351: and is such that clusters could `feed on' each other:
352: thus, smaller clusters disappear faster
353: as if they were swallowed by the larger ones.
354: In this section, we explore the details of the simplest possible case,
355: that of two interacting clusters with masses $m_1(t)$ and $m_2(t)$,
356: and interaction strength $g=g_{12}=g_{21}$.
357:
358: We look successively at the special case of equal masses (Section~4.1)
359: and at the generic case of unequal masses (Section~4.2).
360: Only in the first case are the two masses able to survive forever,
361: growing more slowly than if they had been alone.
362: In the second case, the bigger cluster swallows the smaller one, generically.
363:
364: The dynamical equations~(\ref{dsx}) read
365: \beq
366: \matrix{
367: x'_1=\left(\frad{2\a-1}{2}-g\left(\frad{x_2}{2}+x'_2\right)\right)x_1
368: -\frad{1}{x_1},\cr\cr
369: x'_2=\left(\frad{2\a-1}{2}-g\left(\frad{x_1}{2}+x'_1\right)\right)x_2
370: -\frad{1}{x_2}.}
371: \eeq
372: Solving these equations for the time derivatives, we obtain
373: \beq
374: \matrix{
375: x'_1=\frad{(2\a-1)x_1^2x_2-2x_2+2g(1-\a x_2^2)x_1^2+g^2x_1^3x_2^2}
376: {2x_1x_2(1-g^2x_1x_2)},\cr\cr
377: x'_2=\frad{(2\a-1)x_1x_2^2-2x_1+2g(1-\a x_1^2)x_2^2+g^2x_1^2x_2^3}
378: {2x_1x_2(1-g^2x_1x_2)}.}
379: \label{dsx2}
380: \eeq
381: In the generic situation of two unequal masses, the above dynamical equations
382: already illustrate the full complexity of the problem.
383: The regularity condition~(\ref{cd}), which reads
384: \beq
385: 1-g^2x_1x_2>0
386: \label{cd2}
387: \eeq
388: in the case of two clusters, is needed for the denominators not to vanish.
389:
390: \subsection{Equal masses}
391: \label{twoeq}
392:
393: The main results of this section can be explained physically
394: in the following way:
395: since the interactions cause each mass to `feed on' the other,
396: overly strong interactions will lead to a strongly depletive effect on both,
397: as a result of which neither survives.
398: On the other hand, a weakly interacting pair of equal mass clusters can,
399: provided their masses are above a threshold, survive in gentle symbiosis;
400: both depletion and accretion keep occurring at comparable rates,
401: and the pair survive forever.
402:
403: Consider now two clusters whose masses are equal at the initial time~$t_0$.
404: This symmetry is clearly preserved by the dynamics.
405: Let $x(s)$ be the common value of their reduced mass.
406: Equations~(\ref{dsx2}) simplify to
407: \beq
408: x'=\frac{(2\a-1)x^2-2-gx^3}{2x(1+gx)}.
409: \label{dsx1}
410: \eeq
411:
412: The fixed points of the above dynamical equation, given by
413: \beq
414: (2\a-1)x^2-2-gx^3=0,
415: \label{fp}
416: \eeq
417: dictate the qualitative features of the dynamics.
418: There is a critical value of the coupling,
419: \beq
420: g_c=\left(\frac{2(2\a-1)^3}{27}\right)^{1/2},
421: \label{gc}
422: \eeq
423: which separates two kinds of behaviour:
424:
425: \begin{itemize}
426:
427: \item For a large enough coupling ($g>g_c$),~(\ref{fp}) has no real
428: positive root, leading to the absence of a physical fixed point.
429: The reduced mass $x(s)$ of both clusters decreases monotonically
430: until they simultaneously die out in a finite reduced time.
431:
432: \item For a small enough coupling ($g<g_c$),~(\ref{fp}) has three real roots,
433: two of which are positive and correspond to physical fixed points:
434: \beq
435: y_\star^{1/2}<x\un\hbox{ (unstable) }<(3y_\star)^{1/2}<x\de\hbox{ (stable)}.
436: \eeq
437: Small clusters, such that $x_0<x\un$,
438: are attracted by $x=0$, so that both disappear in a finite time.
439: Large clusters, such that $x_0>x\un$, are attracted by $x\de$:
440: those pairs of clusters are survivors, and their common mass grows as
441: \beq
442: m(t)\approx x\de t^{1/2}.
443: \eeq
444: Note that this growth rate is slower than that of an isolated
445: cluster [see~(\ref{m1})].
446:
447: \end{itemize}
448:
449: In other words, small masses, no matter what the coupling strength,
450: die out in a finite time.
451: For large reduced masses, the role of the coupling strength matters.
452: For~$g<g_c$, those larger than $x\un$ survive forever,
453: and grow more slowly than if they had been isolated;
454: for $g>g_c$, all die eventually.
455:
456: These results match those of Section~\ref{one}
457: in the limit of a vanishingly small coupling.
458: The unstable fixed point has a finite limit $x\un\to y_\star^{1/2}$,
459: whereas the stable one diverges as $x\de\approx(2\a-1)/g$.
460: We recall from Section~\ref{one} that $y_\star$ is the threshold
461: above which independent clusters survive forever.
462: The unstable fixed point $x\un$,
463: which likewise separates dying from surviving clusters,
464: thus reduces to $y_\star$ in the $g\to0$ limit, as it should.
465: The main new element with respect to the non-interacting limit is the existence
466: of the stable fixed point $x\de$.
467: However, its effect is to attract all masses above $x\un$ to itself,
468: so that here, too, the effective behaviour is unchanged
469: with respect to the noninteracting case; masses above $x\un$ survive forever.
470:
471: Finally, it is worth coming back to the regularity condition~(\ref{cd2}).
472: In the present situation of two equal masses, this condition reads
473: \beq
474: x<x\li,\qquad x\li=\frac{1}{g}.
475: \label{cd1}
476: \eeq
477: The limiting value $x\li$ plays no special role in the dynamics
478: of two clusters with equal masses.
479: It will however play an important part
480: in the transient dynamical behaviour of two unequal masses,
481: to be studied in Section~4.2.
482: It is therefore worth investigating the fate of $x\li$.
483: Equation~(\ref{dsx1}) implies
484: \beq
485: x'\vert_{x=x\li}=\frac{\a-1-g^2}{2g}.
486: \label{lido}
487: \eeq
488: This expression singles out the following value of the coupling strength:
489: \beq
490: g\li=(\a-1)^{1/2}\qquad(\a>1).
491: \label{gli}
492: \eeq
493: For $g<g\li$, the right-hand side of~(\ref{lido}) is positive,
494: so that $x\li$ flows toward larger values of $x$.
495: Conversely, for $g>g\li$, the right-hand side of~(\ref{lido}) is negative,
496: so that~$x\li$ flows toward smaller values of $x$.
497: We present in Figure~\ref{figa}, for further reference,
498: the phase diagram of the two-cluster problem in the $\alpha$--$g$ plane.
499: Four different phases can be defined (see caption),
500: according to the number of real positive fixed points,
501: and to the number of those obeying the condition~(\ref{cd1}).
502: The phase boundaries are determined by equations~(\ref{gc}) and~(\ref{gli}).
503:
504: \begin{figure}[htb]
505: \begin{center}
506: \includegraphics[angle=90,width=.6\linewidth]{figa.eps}
507: \caption{\small
508: Phase diagram of the two-cluster problem in the $\alpha$--$g$ plane.
509: Full line: critical coupling $g=g_c(\a)$ of~(\ref{gc}).
510: Dashed line: $g=g\li(\a)$ of~(\ref{gli}).
511: Phase~I: no real positive fixed point.
512: Phase~II: two fixed points, both obeying~(\ref{cd1}): $x\un<x\de<x\li$.
513: Phase~III: two fixed points,
514: only $x\un$ obeys~(\ref{cd1}): $x\un<x\li<x\de$.
515: Phase~IV: two fixed points, none obeying~(\ref{cd1}): $x\li<x\un<x\de$.
516: Full symbol: quadruple point ($\a=5/4$, $g=1/2$): $x\un=x\de=x\li=2$.}
517: \label{figa}
518: \end{center}
519: \end{figure}
520:
521: \subsection{Unequal masses}
522: \label{twoun}
523:
524: In the general case where the clusters have unequal masses,
525: the role of the interactions is inherently disequilibrating:
526: mass differences, however small initially, get rapidly amplified, leading
527: to the generic scenario of the survival of the biggest.
528: We shall investigate successively three stages in the dynamics
529: of two clusters with slightly unequal masses.
530:
531: \subsubsection*{Early stage: linear stability analysis}
532:
533: Consider first the early stage of the dynamics
534: for two clusters with a small mass difference.
535: Setting
536: \beq
537: x_1(s)=x(s)+\eps(s),\qquad x_2(s)=x(s)-\eps(s),
538: \eeq
539: to first order in the difference $\eps(s)$,
540: the mean reduced mass $x(s)$ obeys~(\ref{dsx1}),
541: while the difference itself obeys the linear equation
542: \beq
543: \eps'(s)=\L(x(s))\,\eps(s),
544: \eeq
545: where the instantaneous Lyapunov exponent $\L(x)$ reads
546: \beq
547: \L(x)=\L_0(x)+\frac{g(2+gx+\a gx^3)}{x(1-g^2x^2)},\qquad
548: \L_0(x)=\frac{2\a-1}{2}+\frac{1}{x^2}.
549: \eeq
550: The full Lyapunov exponent $\L(x)$
551: of two interacting clusters is therefore larger than
552: the Lyapunov exponent $\L_0(x)$ in the absence of coupling,
553: which is in turn larger than the constant $(2\a-1)/2$.
554: As underlined above, interactions thus enhance disequilibration.
555: In any case, irrespective of the mean initial mass and of the coupling,
556: any small initial mass difference diverges
557: exponentially in the early stages of the dynamics.
558: In particular, the fixed point $x\de$ of Section~\ref{twoeq},
559: which is stable against a symmetric perturbation
560: of the form $\delta x_1=\delta x_2$,
561: is always linearly unstable against an asymmetric perturbation of the form
562: $\delta x_1=-\delta x_2=\eps$.
563:
564: \subsubsection*{Intermediate stage: transient behaviour in the various phases}
565:
566: Later stages of the dynamics cannot be described in closed form,
567: because of the non-linearity of~(\ref{dsx2}).
568: The detailed transient time dependence of both masses depends on the location
569: of the parameters $\a$ and $g$ in the phase diagram of Figure~\ref{figa},
570: especially when the initial mass difference is small.
571:
572: These features are illustrated in Figure~\ref{figb},
573: showing the shape of typical trajectories in the $g\,x_1$--$g\,x_2$ plane.
574: The dashed line shows the limit of the regularity condition~(\ref{cd2}),
575: so that allowed pairs of reduced masses are below that line.
576: Each full line shows a trajectory starting
577: with a small mass difference $\eps=\pm10^{-2}$.
578: The mass difference then grows monotonically,
579: until the trajectory hits either of the co-ordinate axes
580: in some finite time~$s_1$, when the lighter mass disappears.
581: Each panel corresponds to a typical choice of $\a$ and $g$
582: in each of the four phases.
583: Phases~I and~IV are very similar:
584: the lighter mass always decreases monotonically,
585: whereas the larger one decreases in a first stage.
586: If the larger mass is large enough,
587: it may then start increasing before the lighter dies out.
588: Phases~II and~III are also similar:
589: both masses may increase in a first stage
590: if their difference is small enough.
591:
592: \begin{figure}[htb]
593: \begin{center}
594: \includegraphics[angle=90,width=.35\linewidth]{figb1.eps}
595: {\hskip 1mm}
596: \includegraphics[angle=90,width=.35\linewidth]{figb2.eps}
597: \vskip 2mm
598:
599: \includegraphics[angle=90,width=.35\linewidth]{figb3.eps}
600: {\hskip 1mm}
601: \includegraphics[angle=90,width=.35\linewidth]{figb4.eps}
602: \caption{\small
603: Typical trajectories in the $g\,x_1$--$g\,x_2$ plane,
604: for one choice of the parameters $\a$ and $g$
605: in each of the four phases shown in Figure~\ref{figa}.
606: Full lines: a few trajectories with an initial mass difference
607: $\eps=\pm10^{-2}$.
608: Dashed line: limit of the regularity condition~(\ref{cd2}),
609: i.e., $gx_1\cdot gx_2=1$.
610: Full symbols: fixed points.}
611: \label{figb}
612: \end{center}
613: \end{figure}
614:
615: \subsubsection*{Late stage: survival of the biggest}
616:
617: The values of the parameters $\a$ and $g$ become asymptotically
618: irrelevant in the late stages of the dynamical evolution.
619: Indeed, as illustrated above,
620: any trajectory eventually hits either co-ordinate axis.
621: There is therefore one single generic scenario
622: for two clusters with unequal initial masses,
623: i.e., that of survival of the biggest:
624:
625: \begin{itemize}
626:
627: \item The smaller one dies out in a finite time $s_1$.
628:
629: \item The larger one then evolves according to the results of Section~\ref{one}.
630: Depending on the value of its mass at reduced time $s_1$, it may
631: either also disappear in a finite time (for $y(s_1)<y_\star$),
632: or survive and grow forever (for $y(s_1)>y_\star$).
633:
634: \end{itemize}
635:
636: The above results still hold qualitatively
637: for any finite number $n\ge2$ of fully interacting clusters
638: (all couplings are non-zero).
639: In the generic situation of unequal masses,
640: the scenario of survival of the biggest applies:
641: the $n-1$ smaller clusters die out one after the other,
642: while only the largest one may become a survivor.
643:
644: \section{Mean-field limit}
645: \label{mft}
646:
647: Having explored the behaviour of finitely many interacting clusters,
648: we now turn our attention to collective behaviour
649: in the thermodynamical limit of an infinite assembly of interacting clusters.
650: In the present section, we focus on the mean-field regime of
651: long-range interactions.
652: The main feature of mean-field dynamics is again
653: that all the clusters eventually die out, except the largest one.
654: The clusters which survive up to time $s$ are those whose initial reduced
655: square mass exceeds some time-dependent mass threshold $Y(s)$,
656: to be determined below.
657: The general case is necessarily somewhat formal (Section~5.1).
658: If, however, one considers the regime of weak interactions,
659: the formalism simplifies considerably (Section~5.2).
660:
661: \subsection{General formalism}
662:
663: We consider the mean-field limit of a large assembly of clusters ($n\gg1$),
664: assuming that all the couplings $g_{ij}$ have the same value $g$.
665: We perform the usual rescaling of the interaction strength
666: in mean-field models:
667: \beq
668: g=\frac{\gbar}{n}.
669: \eeq
670:
671: The problem simplifies drastically in the thermodynamic limit,
672: defined as usual as the $n\to\infty$ limit at fixed $\gbar$.
673: In this limit, the coupling strength of any cluster to its whole environment,
674: measured by $\gbar$, remains of order unity,
675: whereas the strength of the coupling between any two different clusters,
676: measured by $g$, falls off as $1/n$.
677:
678: In this thermodynamic limit,~(\ref{dsx}) implies the following
679: dynamical equation
680: \beq
681: y'(s)=\g(s)y(s)-2
682: \label{mfds}
683: \eeq
684: for the reduced square mass $y(s)$ of any of the clusters.
685: We have introduced the notation
686: \beq
687: \g(s)=2\a-1-\gbar\left(M(s)+2M'(s)\right)
688: \label{gam}
689: \eeq
690: for the effective growth rate of the square mass $y(s)$, where
691: \beq
692: M(s)=\mean{x}_s
693: =\mean{y^{1/2}}_s=\lim_{n\to\infty}\frac{1}{n}\sum_i y_i(s)^{1/2}
694: \label{stat1st}
695: \eeq
696: is the mean reduced mass of the clusters at reduced time $s$.
697:
698: Despite its apparent simplicity,~(\ref{mfds}) is non-trivial,
699: because of its self-consistency:
700: its right-hand-side indeed involves $\g(s)$, and therefore $M(s)$,
701: and therefore the solution of~(\ref{mfds}) itself.
702: This self-consistent problem can be solved formally as follows.
703: First, we have on differentiating~(\ref{stat1st})
704: \beq
705: M'(s)=\frac{\g(s)}{2}\,M(s)-N(s),
706: \eeq
707: with
708: \beq
709: N(s)=\mean{y^{-1/2}}_s=\lim_{n\to\infty}\frac{1}{n}
710: \sum_{i:y_i(s)>0}y_i(s)^{-1/2},
711: \eeq
712: where only non-zero values of $y_i$, corresponding to clusters $i$
713: which survive at time $s$, are involved in the sum.
714: The effective rate $\g(s)$ can therefore be solved from~(\ref{gam}):
715: \beq
716: \g(s)=\frac{2\a-1+\gbar\,(2N(s)-M(s))}{1+\gbar M(s)}.
717: \label{gam2}
718: \eeq
719: On the other hand, a formal integration of~(\ref{mfds}) yields
720: \beq
721: y(s)=G(s)\,(y_0-Y(s)),
722: \label{y0ys}
723: \eeq
724: with
725: \beq
726: G(s)=\exp\left(\int_0^s\g(u)\,\d u\right),
727: \qquad Y(s)=2\int_0^s\frac{\d u}{G(u)},
728: \eeq
729: so that
730: \beq
731: G(s)=\frac{2}{Y'(s)},\qquad\g(s)=-\frac{Y''(s)}{Y'(s)}.
732: \label{y2}
733: \eeq
734:
735: These steps lead to the following picture of the mean-field dynamics.
736: For a given initial value $y_0=y(0)$ of the reduced square mass,
737: the solution~(\ref{y0ys}) holds only as long as $y(s)$ is positive,
738: or, equivalently, $y_0>Y(s)$.
739: Hence the smaller clusters, with initial square masses $y_0<Y(s)$,
740: have already disappeared at reduced time $s$,
741: while larger ones, such that $y_0>Y(s)$,
742: have their square masses shifted and dilated from $y_0$ to $y(s)$,
743: according to~(\ref{y0ys}).
744:
745: This has strong echoes of the case of many non-interacting clusters.
746: Recall that there existed a mass threshold, also called $Y(s)$
747: in~(\ref{bigy}), below which all particles had died at time $s$,
748: and above which they survived.
749: The quantity~$Y(s)$ in~(\ref{y0ys}) above
750: generalises this threshold in the presence of a mean-field coupling.
751: Now, clusters below this threshold die as before, while the mass
752: of a cluster grows, as a result of interactions, from $y_0$ to $y(s)$.
753: In the absence of coupling ($\gbar=0$), the present~$Y(s)$ reduces to that
754: obtained in~(\ref{bigy}).
755:
756: The above formalism allows us, provided the continuous probability
757: distribution of initial square masses $P(y_0)$ is known,
758: to express all the quantities of interest in terms of
759: a single (so far unknown) dynamical quantity, the threshold $Y(s)$.
760: In terms of this, $M(s)$ and $N(s)$ read
761: \beqa
762: &&M(s)=G(s)^{1/2}
763: \int_{Y(s)}^\infty(y_0-Y(s))^{1/2}P(y_0)\,\d y_0,\hfill\label{momm}\\
764: &&N(s)=G(s)^{-1/2}
765: \int_{Y(s)}^\infty(y_0-Y(s))^{-1/2}P(y_0)\,\d y_0.\hfill
766: \eeqa
767:
768: Equation~(\ref{gam2}) then provides a self-consistent
769: equation for the unknown quantity~$Y(s)$,
770: as it only involves $Y(s)$ itself and its first and second derivatives.
771: Of course, the resulting non-linear integro-differential equation
772: cannot be solved in closed form in general.
773:
774: To recapitulate, the program for the mean-field solution
775: of the dynamics of $n\gg1$ clusters is the following:
776: Equation~(\ref{y0ys}) gives the growth law of any cluster mass
777: in terms of the time-dependent mass threshold $Y(s)$.
778: Only clusters above this threshold survive at reduced time $s$,
779: as the rest have died.
780: The calculation of $Y(s)$ can be done self-consistently,
781: at least in principle, from~(\ref{gam2}).
782: Quantities of interest, such as the mean mass of surviving clusters at time $s$,
783: can then be calculated via~(\ref{momm}) and similar expressions.
784:
785: The definition of the survival probability $S(s)$ is
786: unaltered by the presence of interactions,
787: and is still given by equation~(\ref{ss}), i.e.,
788: \beq
789: S(s)=\int_{Y(s)}^\infty P(y_0)\,\d y_0.
790: \label{ss2}
791: \eeq
792: The mean reduced mass of the surviving clusters reads
793: \beq
794: \meansur{x}_s=\meansur{y^{1/2}}_s=\frac{M(s)}{S(s)},
795: \label{xmean}
796: \eeq
797: where $\meansur{\cdots}_s$ denotes the normalised mean
798: over the clusters surviving at reduced time~$s$.
799: In general, average quantities computed over all
800: clusters, dead and alive, must be renormalised by
801: $S(s)$ in order to get an appropriately normalised
802: average over surviving clusters at any time $s$.
803:
804: \subsection{Weak-coupling regime}
805:
806: The results of Section 5.1 hold for arbitrary values of $\gbar$,
807: and their formal nature admits of no further simplification.
808: However, a much greater transparency is achieved
809: in the regime where the rescaled coupling $\gbar$ is small.
810:
811: In this regime, the dynamics consists of two successive stages.
812: In Stage~I, the clusters behave as if they were isolated,
813: i.e., their masses evolve according to the results of Section~\ref{one}.
814: This fast stage of the dynamics therefore corresponds to individual behaviour.
815: The only surviving clusters after Stage~I are
816: those whose initial masses exceed the threshold~(\ref{thre}).
817: The effect of interactions sets in at Stage~II.
818: This slow stage of the dynamics corresponds to collective behaviour.
819: All but the largest cluster eventually die out during this stage.
820:
821: The weakly interacting mean-field regime of our model of interacting
822: clusters therefore exhibits characteristic features
823: of glassy systems~\cite{glassyrefs}.
824: These aging phenomena originate in the presence
825: of two well-separated time scales of fast and slow dynamics,
826: with a ratio of respective time scales growing as $1/\gbar^2$.
827: Another striking feature of our model is the universality
828: of the main asymptotic results in Stage~II dynamics:
829: the survival probability falls off generically as $(\ln t)^{-1/2}$,
830: whereas the mean square mass of survivors grows as~$t\ln t$.
831: As is well known~\cite{glassyrefs},
832: such logarithmic behaviour is another telltale sign of glassy dynamics.
833: We describe all of this below in more detail.
834:
835: \subsubsection*{Stage~I: Fast individual dynamics}
836:
837: In this first stage, interactions are essentially irrelevant,
838: and the dynamics is fast.
839: The mass of each cluster evolves according to Section~\ref{one},
840: independently of all the others, as if it were isolated.
841: The survival probability decays from $S(0)=1$
842: to the plateau value $S\un$ of~(\ref{s1}),
843: whereas the time-dependent threshold $Y(s)$ of~(\ref{bigy})
844: increases from $Y(0)=0$ to~$y_\star$ of~(\ref{ystar}).
845:
846: \subsubsection*{Stage~II: Slow collective dynamics}
847:
848: In this second stage, the interactions are responsible
849: for a slow collective dynamics in the weak-coupling regime.
850:
851: The evolution throughout Stage~II can be described as follows.
852: Starting from the assumption (to be checked later on)
853: that the dynamics is slow, we have $\g(s)\ll1$ and $M'(s)\ll M(s)$.
854: Equation~(\ref{gam}) therefore simplifies to
855: \beq
856: M(s)\approx\frac{2\a-1}{\gbar}.
857: \label{mslow}
858: \eeq
859:
860: Now, inserting~(\ref{y2}) and~(\ref{mslow}) in~(\ref{momm}),
861: we obtain the following closed differential equation
862: for the unknown $Y(s)$:
863: \beq
864: Y'(s)\approx\frac{2\,\gbar^2}{(2\a-1)^2}\,R(Y)^2,
865: \label{yslow}
866: \eeq
867: where the function
868: \beq
869: R(Y)=\int_Y^\infty(y_0-Y)^{1/2}P(y_0)\,\d y_0
870: \eeq
871: is entirely determined by the initial mass distribution.
872:
873: The behaviour of the threshold $Y(s)$ throughout Stage~II
874: is obtained by integrating~(\ref{yslow}),
875: with an initial value equal to the plateau value $y_\star$ of~(\ref{ystar}):
876: \beq
877: \int_{y_\star}^Y\frac{\d y}{R(y)^2}\approx\frac{2\,\gbar^2\,s}{(2\a-1)^2}.
878: \label{ws}
879: \eeq
880: Equation~(\ref{ws}) contains the key to the dynamical behaviour in Stage~II.
881:
882: First, the characteristic time of the collective dynamics,
883: \beq
884: s_c\sim\frac{(2\a-1)^2}{\gbar^2},
885: \label{scmf}
886: \eeq
887: becomes arbitrarily large in the weak-coupling regime ($\gbar\to0$).
888: The assumption of slow dynamics is thus fully justified.
889:
890: Then, the analysis of the long-time dynamics goes as follows.
891: Note that the left-hand side of~(\ref{ws}) diverges as $s\to\infty$.
892: The only way this can occur is if $R(Y)$ falls off to zero for long times,
893: i.e., if the threshold $Y(s)$ goes to the maximum possible
894: initial mass $y\max$, i.e., more formally,
895: the upper bound of the continuous distribution $P(y_0)$.
896: The survival probability~$S(s)$ then also falls off to zero for long times.
897: We can conclude that the whole population of clusters
898: which survived Stage~I eventually disappears during
899: Stage~II of the mean-field dynamics.
900: For a finite mean-field system consisting of $n$ coupled clusters,
901: at most one of them will survive forever,
902: as already mentioned at the end of Section 4.2.
903:
904: Having established the general pattern,
905: we now specialise to specific distributions of initial masses,
906: in order to obtain quantitative predictions.
907: We first consider an exponential distribution of initial masses.
908: We will find that the results
909: obtainable from it can be generalised to a raft of other distributions.
910:
911: With the exponential distribution~(\ref{initexpo}),
912: equations~(\ref{ss2}),~(\ref{ws}) yield
913: \beq
914: \e^{2\mu Y(s)}=\frac{1}{S(s)^2}
915: \approx\e^{2\mu y_\star}+\frac{\pi\,\gbar^2\,s}{(2\a-1)^2}.
916: \eeq
917: In the late times of Stage~II, the survival probability therefore decays as
918: \beq
919: S(s)\approx\frac{2\a-1}{\gbar}\,(Cs)^{-1/2},
920: \label{slate}
921: \eeq
922: with
923: \beq
924: C=\pi
925: \label{cpi}
926: \eeq
927: for the chosen exponential distribution,
928: irrespectively of $\a$, $\mu$, and~$\gbar$, provided the latter is small enough.
929:
930: Equations~(\ref{xmean}),~(\ref{mslow}) then lead to
931: \beq
932: \meansur{x}_s\approx(Cs)^{1/2}.
933: \eeq
934:
935: For a final presentation of the above results, we return to physical variables.
936: In terms of these, the survival probability falls off as
937: \beq
938: S(t)\approx\frac{2\a-1}{\gbar}\,\left(C\,\ln\frac{t}{t_0}\right)^{-1/2},
939: \label{stlate}
940: \eeq
941: whereas the mean mass of the surviving clusters grows as
942: \beq
943: \meansur{m}_t\approx\left(C\,t\,\ln\frac{t}{t_0}\right)^{1/2}.
944: \label{mtlate}
945: \eeq
946:
947: The universality inherent in the scaling results~(\ref{slate})--(\ref{mtlate})
948: is unusual, because it includes the prefactor $C$,
949: which is itself independent of the details of the initial distribution $P(y_0)$
950: of square masses.
951: It can indeed be checked explicitly that $C$ only depends
952: on the {\it tail exponent}
953: of this distribution in the vicinity of its upper bound~$y\max$:
954:
955: \begin{itemize}
956:
957: \item In the bounded case ($y\max$ finite),
958: assuming that the distribution has a power-law behaviour
959: $P(y_0)\approx A(y\max-y_0)^{a-1}$ for $y_0\to y\max$,
960: with a tail exponent $a>0$, we obtain
961: \beq
962: C=\pi a\left(\frac{\Gamma(a+1)}{\Gamma\!\left(a+\frac32\right)}\right)^2.
963: \label{ca}
964: \eeq
965:
966: \item In the unbounded case ($y\max=\infty$),
967: assuming that the distribution has a power-law behaviour
968: $P(y_0)\approx B\,y_0^{-b-1}$ for $y_0\to\infty$,
969: with a tail exponent $b>1/2$, we obtain
970: \beq
971: C=\pi b\left(\frac{\Gamma\!\left(b-\frac12\right)}{\Gamma(b)}\right)^2.
972: \label{cb}
973: \eeq
974:
975: \end{itemize}
976:
977: It is worth noticing that both results~(\ref{ca}) and~(\ref{cb})
978: smoothly converge to the particular value $C=\pi$ of~(\ref{cpi}),
979: as the tail exponents $a$ and $b$ get large:
980: \beq
981: C=\pi\left(1-\frac{3}{4a}+\cdots\right)=\pi\left(1+\frac{3}{4b}+\cdots\right).
982: \eeq
983:
984: \section{Finite-dimensional lattices}
985: \label{latt}
986:
987: The mean-field regime described above can be thought of
988: as an infinite-dimensional limit of our model.
989: Now, in order to include the effects of fluctuations,
990: we consider a finite-dimensional lattice model.
991: In addition to the emergence of two well-separated time scales,
992: the model now displays {\it metastability}:
993: the system gets finally trapped forever in a non-trivial attractor,
994: where every surviving cluster is isolated.
995:
996: More specifically, clusters sit at the vertices $\n$ of a regular lattice.
997: Every pair of nearest neighbours interacts with a uniform coupling strength $g$.
998: Numerical simulations have been performed on hypercubic lattices:
999: the chain ($D=1$), the square lattice ($D=2$), and the cubic lattice ($D=3$).
1000: The coordination number of these lattices is $z=2D$.
1001: Throughout this section,
1002: initial masses are given by the exponential distribution~(\ref{initexpo}).
1003: Unless otherwise stated, we set $\a=1$, $g=10^{-4}$,
1004: and $\mu$ such that~(\ref{sunexpo}) yields $S\un=0.9$.
1005:
1006: The main focus will be the limit of weak coupling ($g\ll1$).
1007: We therefore simplify the dynamical equations~(\ref{dsx}),
1008: by keeping terms up to first order in $g$.
1009: The resulting explicit equations
1010: \beq
1011: x'_\n=\left(\frac{2\a-1}{2}+g\sum_{\m}\left(\frac{1}{x_\m}-\a x_\m\right)
1012: \right)x_\n-\frac{1}{x_\n},
1013: \label{dsg}
1014: \eeq
1015: where $\m$ runs over the $z$ nearest neighbours of site $\n$,
1016: are solved numerically by means of a standard first-order scheme.
1017:
1018: \subsection{Two-step dynamics}
1019:
1020: In the weak-coupling regime, the dynamics generated by~(\ref{dsg})
1021: again consists of two successive well-separated stages.
1022: As before, fast individual dynamics are exhibited in Stage~I,
1023: while Stage~II is the arena for slow collective dynamics.
1024: The effects of going beyond mean field are only palpable
1025: in the latter stage, since interactions are irrelevant in Stage~I.
1026:
1027: \subsubsection*{Stage~I: Fast individual dynamics}
1028:
1029: There is little new here with respect to the mean-field regime.
1030: The mass of each cluster evolves as if it were isolated, as before.
1031: The survival probability $S(s)$ decays rather fast
1032: from $S(0)=1$ to its plateau value $S\un$ of~(\ref{s1}).
1033:
1034: \subsubsection*{Stage~II: Slow collective dynamics}
1035:
1036: The slow collective dynamics throughout Stage~II
1037: is now very different from the mean-field regime.
1038: The survival probability $S(s)$ indeed decays
1039: from its plateau value $S\un$ to a non-trivial limiting value $S\infy$,
1040: because of metastability, as will be shown below.
1041:
1042: The collective dynamics throughout Stage~II
1043: is very slow in the weak-coupling regime.
1044: Consider now~(\ref{dsg}) for two neighbouring clusters $\n$ and $\m$
1045: which have both survived Stage~I.
1046: The contribution of cluster $\m$ to the large parenthesis in the
1047: right-hand side of~(\ref{dsg}) is proportional to~$\a gx_\m$.
1048: In the absence of coupling,
1049: we have $x_\m\sim\e^{(2\a-1)s/2}$, by virtue of~(\ref{yone}).
1050: The characteristic time scale of Stage~II
1051: is reached when the product~$gx_\m$ becomes of order unity.
1052: It reads therefore
1053: \beq
1054: s_c\approx\frac{2}{2\a-1}\,\ln\frac{1}{g},
1055: \label{sc}
1056: \eeq
1057: i.e.,
1058: \beq
1059: t_c\sim t_0\,g^{-2/(2\a-1)}.
1060: \eeq
1061: The separation of time scales between the fast individual
1062: and the slow collective dynamics
1063: is therefore again parametrically large in the weak-coupling limit,
1064: although the divergence of the collective time scale is much less pronounced
1065: than in the mean-field limit~[see~(\ref{scmf})].
1066: The glassiness of the dynamics, with its manifest two-step relaxation,
1067: is illustrated in Figure~\ref{figc}.
1068: This figure
1069: shows a plot of the decay of the survival probability $S(s)$ in one dimension.
1070: The dynamical equations~(\ref{dsg})
1071: have been integrated numerically for a chain of $10^6$ clusters,
1072: until every surviving cluster is isolated (see below).
1073: Both stages of the dynamics appear clearly on the plot,
1074: as well as the expected plateau value $S\un=0.8$,
1075: and the occurrence of a non-trivial limit survival probability
1076: $S\infy\approx0.4134$.
1077: Each curve corresponds to an interaction strength $g$
1078: a decade apart from its neighbour.
1079: It is accordingly shifted by~$2\,\ln 10$ (thick bar),
1080: in accord with the estimate~(\ref{sc}).
1081:
1082: \begin{figure}[htb]
1083: \begin{center}
1084: \includegraphics[angle=90,width=.6\linewidth]{figc.eps}
1085: \caption{\small
1086: Plot of the survival probability $S(s)$ on the chain with $S\un=0.8$.
1087: Left to right:
1088: Full line: $g=10^{-3}$.
1089: Dashed line: $g=10^{-4}$.
1090: Long-dashed line: $g=10^{-5}$.
1091: Dash-dotted line: $g=10^{-6}$.
1092: The thick bar has length $2\,\ln 10=4.605$~(see text).}
1093: \label{figc}
1094: \end{center}
1095: \end{figure}
1096:
1097: At the end of Stage~II of the dynamics, i.e.,
1098: in practice after a very long time,
1099: the system is left in a non-trivial {\it attractor},
1100: which consists in a pattern where each cluster is {\it isolated}:
1101: all its first neighbours are dead, and is therefore a {\it survivor}:
1102: it survives and keeps growing forever.
1103: In the following, we shall call these attractors {\it metastable states},
1104: in analogy with a variety of statistical-mechanical systems,
1105: where metastable states have been identified
1106: under various names in different contexts:
1107: valleys~\cite{ks}, pure states~\cite{tap,ktw},
1108: inherent structures~\cite{sw}, quasi-states~\cite{fv}.
1109: The common feature of metastable states in all these situations
1110: is that their number~$\N$
1111: generically grows exponentially with the system size (number of sites) $N$:
1112: \beq
1113: \N\sim\exp(N\Sigma).
1114: \eeq
1115: The quantity $\Sigma$ is usually referred to as the configurational entropy,
1116: or complexity.
1117:
1118: The rest of this section is devoted to various characteristics
1119: of these attractors, such as their density
1120: (equal to the limit survival probability), spatial patterns,
1121: spatial correlations, and mass distribution of survivors.
1122:
1123: \subsection{Limit survival probability}
1124:
1125: The limit survival probability $S\infy$,
1126: already emphasised in Figure~\ref{figc},
1127: is just the density of a typical attractor,
1128: i.e., the fraction of the initial clusters which survive forever
1129: and take part in the attractor.
1130: The limit survival probability obeys the inequalities
1131: \beq
1132: S\infy\le S\un,\qquad S\infy\le1/2.
1133: \eeq
1134: The first inequality expresses that clusters can only disappear:
1135: the difference $1-S\un$ (resp.~$S\un-S\infy$) is the fraction of clusters
1136: which die out during Stage~I (resp.~Stage~II).
1137: The second inequality is a consequence of the fact that
1138: each surviving cluster is isolated.
1139: The densest configuration of lattice sites
1140: obeying this condition consists in occupying all the sites
1141: of either of the two sublattices, whose density is exactly $1/2$.
1142: This value $1/2$ of the highest density holds for the large family
1143: of so-called {\it bipartite} lattices,
1144: which includes hypercubic lattices (chain, square lattice, cubic lattice, ...).
1145: It is, however, not universal,
1146: and would e.g.~be only $1/3$ for the triangular lattice.
1147:
1148: In a given class of initial mass distributions,
1149: the limit survival probability $S\infy$
1150: is a monotonically increasing function of the plateau value $S\un$,
1151: starting from $S\infy=0$ for $S\un=0$,
1152: and going to a non-trivial maximum value $S\infymax<1/2$
1153: in the $S\un\to1$ limit.
1154:
1155: In the regime where $S\un$ is small,
1156: it can be shown that $S\infy$ is also small,
1157: and that it depends on $S\un$ alone.
1158: To do so, let us introduce the concept of supercluster.
1159: In analogy with a percolation cluster,
1160: a supercluster is defined as a set of $k\ge1$ connected clusters
1161: which have survived Stage~I,
1162: and such that all their neighbours have disappeared during Stage~I.
1163: The fate of superclusters depends on their size $k$ as follows.
1164:
1165: \begin{itemize}
1166:
1167: \item[$\star$] $k=1$:
1168: If a supercluster consists of a single isolated cluster,
1169: it evolves in Stage~II according to the dynamics of Section~\ref{one}:
1170: it is a survivor, because its reduced square mass exceeds
1171: the threshold $y^\star$ of~(\ref{ystar}).
1172: For independent of initial masses,
1173: a supercluster with $k=1$ occurs with density $p_1=S\un(1-S\un)^{2D}$.
1174:
1175: \item[$\star$] $k=2$:
1176: If a supercluster consists of a pair of neighbouring clusters
1177: (represented as~\pl\pl)
1178: both clusters evolve according to the dynamics of Section~\ref{twoun}:
1179: the smaller dies out, while the larger is a survivor.
1180: We are thus left with~{\pl\vi} or~{\vi\pl} in the late stages of the dynamics.
1181: Such an event takes place with density $p_2=S\un^2(1-S\un)^{2(2D-1)}$.
1182:
1183: \item[$\star$] $k\ge3$:
1184: If three or more surviving clusters form a supercluster,
1185: they may a priori have more than one possible fate.
1186: Consider for instance a linear supercluster of three clusters (\pl\pl\pl).
1187: If the middle one disappears first (\pl\vi\pl),
1188: the two end ones are isolated, and both will be survivors.
1189: If one of the end ones disappears first (e.g.~\pl\pl\vi),
1190: the other two form an interacting pair,
1191: and only the larger of those two will survive forever (e.g.~\pl\vi\vi).
1192: The pattern of the survivors,
1193: and even their number, therefore cannot be predicted a priori.
1194:
1195: \end{itemize}
1196:
1197: The above enumeration implies $S\infy=p_1+p_2/2+\cdots$,
1198: where the dots stand for the unknown contribution of superclusters
1199: with $k\ge3$.
1200: As $p_1\sim S\un$, $p_2\sim S\un^2$, and so on, we are left with the expansion
1201: \beq
1202: S\infy=S\un-D\,S\un^2+\cdots
1203: \label{expan}
1204: \eeq
1205: The dependence of $S\infy$ on details of the initial mass distribution
1206: at fixed $S\un$ therefore only appears at order $S\un^3$.
1207:
1208: In the converse limit $S\un\to1$, the limit survival probability
1209: reaches a non-trivial maximum value $S\infymax<1/2$,
1210: which depends very weakly on the mass distribution.
1211: For instance, in one dimension one has
1212: $S\infymax\approx0.441$ for an exponential distribution
1213: and $S\infymax\approx0.446$ for a uniform distribution.
1214: Figure~\ref{figd} shows a plot of the limit survival probability $S\infy$
1215: (fraction of clusters that survive both fast and slow dynamics)
1216: against the plateau survival probability $S\un$
1217: (fraction of clusters that survive the fast dynamics of Stage~I),
1218: for an exponential mass distribution in one, two, and three dimensions.
1219:
1220: \begin{figure}[htb]
1221: \begin{center}
1222: \includegraphics[angle=90,width=.6\linewidth]{figd.eps}
1223: \caption{\small
1224: Plot of the limit survival probability $S\infy$
1225: against the plateau survival probability~$S\un$.
1226: Top to bottom: one dimension (circles), two dimensions (squares),
1227: three dimensions (triangles).
1228: Full lines (hardly visible through symbols): rational fits
1229: based on [2/2] Pad\'e approximants
1230: incorporating both terms of the expansion~(\ref{expan}).}
1231: \label{figd}
1232: \end{center}
1233: \end{figure}
1234:
1235: \subsection{Spatial patterns of attractors}
1236:
1237: We have seen already that the system generically ends up
1238: trapped in an attractor, or metastable state, where each cluster is isolated,
1239: i.e., surrounded by empty sites, and therefore grows forever.
1240: The spatial patterns generated by survivors
1241: are therefore absolutely fixed once they are created by the dynamics.
1242:
1243: We now turn to a descriptive investigation of these patterns.
1244: One of our main rationales for this exploration of spatial patterns
1245: derives from the cosmological model which was at the origin of the present
1246: model~\cite{I}.
1247: In that context, it was of interest to obtain both the mass distribution
1248: of black holes and their spatial pattern.
1249:
1250: For ease of visualisation, we consider a square lattice.
1251: In the limit of highest density ($S\infy=1/2$),
1252: there are only two possible `ground-state' configurations of the system:
1253: one where the first sublattice is full of survivors, while the second is empty,
1254: and vice-versa.
1255: In this limit,
1256: patterns of surviving clusters are therefore perfect checkerboards.
1257:
1258: In order to describe the patterns of attractors,
1259: we are led to introduce at every site~$\n$,
1260: with integer co-ordinates $(n_1,n_2,\dots,n_D)$,
1261: both the {\it survival index}
1262: \beq
1263: \s_\n=\left\{\matrix{
1264: 1\quad\hbox{if the cluster at site~$\n$ is a survivor,}\hfill\cr
1265: 0\quad\hbox{else,}\hfill}\right.%}
1266: \eeq
1267: and the {\it checkerboard index}
1268: \beq
1269: \phi_\n=(-1)^{\s_\n+n_1+\cdots+n_D}.
1270: \eeq
1271: The survival index depicts very simply the pattern of surviving clusters
1272: surrounded by empty sites.
1273: The checkerboard index, on the other hand, represents,
1274: for each site, the local choice of one of the two symmetry-related
1275: `ground states', i.e., of one of the two sublattices.
1276: This is easiest to understand using a one-dimensional example:
1277: the two ground states are $+-+-+\cdots$ or $-+-+-\cdots$
1278: All the $\phi_n$ are equal to $-1$ in the first pattern,
1279: and equal to $+1$ in the second pattern.
1280: The checkerboard index $\phi_\n$ thus classifies
1281: each site according to the particular ground state selected locally
1282: at this site.
1283:
1284: If the initial masses are large enough, so that the plateau
1285: survival probability $S\un$ after Stage~I is close to unity,
1286: the limit survival probability $S\infy$ is not far
1287: from its `ideal' highest value of~$1/2$.
1288: In this regime, attractors clearly exhibit a local checkerboard structure,
1289: as well as frozen-in defects with respect to a perfect checkerboard.
1290: The random structure of defects is entirely inherited
1291: from the random distribution of initial masses,
1292: because the dynamics is deterministic.
1293:
1294: Figure~\ref{fige} shows a map of the survival index
1295: and of the checkerboard index
1296: for the same attractor of a $150^2$ sample of the square lattice.
1297: This attractor has a density $S\infy\approx0.371$.
1298: For greater clarity, we also zoom into a part of size $40^2$,
1299: in order to show better the correspondence between
1300: patterns of the survival and checkerboard indices.
1301: The frozen-in defects cause what appear to be little rivulets of voids
1302: which surround patches of perfect checkerboard.
1303: These islands of checkerboard are represented by black
1304: or white patches in the lower right-hand figure, depending on their parity,
1305: which is clearly visible from a comparison of the two figures.
1306:
1307: \begin{figure}[htb]
1308: \begin{center}
1309: \includegraphics[angle=0,width=.35\linewidth]{fige1.eps}
1310: {\hskip 1mm}
1311: \includegraphics[angle=0,width=.35\linewidth]{fige2.eps}
1312: \vskip 3mm
1313:
1314: \includegraphics[angle=0,width=.35\linewidth]{figei1.eps}
1315: {\hskip 1mm}
1316: \includegraphics[angle=0,width=.35\linewidth]{figei2.eps}
1317: \caption{\small Two complementary representations
1318: of a typical pattern of surviving clusters on the square lattice,
1319: with $S\un=0.9$, so that $S\infy\approx0.371$.
1320: Left: Map of the survival index.
1321: Black (resp.~white) squares represent $\s_\n=1$ (resp.~$\s_\n=0$),
1322: i.e., surviving (resp.~dead) sites.
1323: Right: Map of the checkerboard index.
1324: Black (resp.~white) squares represent $\phi_\n=+1$ (resp.~$\phi_\n=-1$).
1325: Top panels show a $150^2$ sample.
1326: Bottom panels show an enlargement of a $40^2$ region
1327: near the centre of the sample.}
1328: \label{fige}
1329: \end{center}
1330: \end{figure}
1331:
1332: \subsection{Spatial correlations}
1333:
1334: The main consequence of short-range interactions
1335: is the generation of correlations between clusters.
1336: In our model, a study of such correlations is especially meaningful
1337: in the long-time limit, when all the clusters which are still present
1338: are in fact survivors.
1339: In this regime, neighbouring sites are fully anticorrelated,
1340: because each survivor is surrounded by voids.
1341: However, at least close to the limit $S\infy=1/2$,
1342: the next-nearest neighbours of a surviving cluster
1343: are expected to contain another survivor with high probability.
1344: Also, most survivors at late times should be quite massive,
1345: as they have both survived Stage~I and then Stage~II.
1346: We may thus expect that survival and mass correlations
1347: exhibit a rather similar dependence on the distance.
1348: These expectations are borne out by the following detailed study of correlation
1349: functions.
1350:
1351: Let us introduce the two-point correlation functions
1352: of the survival index and of the reduced mass at separation $\n$:
1353: \beq
1354: C_\s(\n)=\frac{\mean{\s_\0\s_\n}}{S\infy},\qquad
1355: C_x(\n)=\frac{\mean{x_\0x_\n}}{\mean{x^2}},
1356: \eeq
1357: where $\mean{\dots}$ stands for a normalised ensemble average
1358: at a very late stage of the dynamics.
1359: These correlations are normalised so as to have
1360: $C_\s(\0)=C_x(\0)=1$, whereas $C_\s(\evec)=C_x(\evec)=0$,
1361: where $\evec$ is any unit vector of the lattice.
1362: Their disconnected parts read
1363: \beq
1364: C_\s(\infty)=\lim_{\abs{\n}\to\infty}C_\s(\n)=S\infy,\qquad
1365: C_x(\infty)=\lim_{\abs{\n}\to\infty}C_x(\n)=\frac{\mean{x}^2}{\mean{x^2}}.
1366: \label{disco}
1367: \eeq
1368:
1369: It is worth noticing that the correlations of the checkerboard index
1370: are not independent of those of the survival index.
1371: We have indeed:
1372: \beq
1373: \mean{\phi_\0\phi_\n}=(-1)^{n_1+\cdots+n_D}(4S\infy C_\s(\n)-4S\infy+1).
1374: \eeq
1375: The existence of such an identity is quite natural,
1376: although it may seem surprising at first sight.
1377: Indeed the checkerboard index is only a different bookkeeping method
1378: for the same data on the positions of the survivors.
1379:
1380: \begin{figure}[htb]
1381: \begin{center}
1382: \includegraphics[angle=90,width=.6\linewidth]{figf.eps}
1383: \caption{\small Plot of the correlation functions
1384: against the distance $n$ along the chain.
1385: Empty symbols: correlation $C_\s(n)$ of the survival index.
1386: Full symbols: correlation $C_x(n)$ of the reduced mass.}
1387: \label{figf}
1388: \end{center}
1389: \end{figure}
1390:
1391: Figure~\ref{figf} shows a plot of the correlation functions
1392: $C_\s(n)$ and $C_x(n)$ against distance, in one dimension.
1393: Both correlation functions exhibit a fast oscillatory convergence
1394: toward their disconnected parts.
1395: Similar features are observed for the on-axis correlations
1396: $C_\s(n\evec)$ and $C_x(n\evec)$ on the square and cubic lattices.
1397: Figure~\ref{figgh} shows
1398: a plot of the logarithm of the absolute on-axis connected correlations
1399: $C_\s^c(n\evec)=C_\s(n\evec)-C_\s(\infty)$
1400: and $C_x^c(n\evec)=C_x(n\evec)-C_x(\infty)$, against distance,
1401: in one, two, and three dimensions.
1402: The connected correlations are observed to fall off very fast to zero,
1403: so fast that it is hard to fit the precise form of their asymptotic decay.
1404: Neither a conventional exponential fall-off
1405: nor a more exotic super-exponential behaviour
1406: can be ruled out from the available data.
1407: A more accurate investigation of this point
1408: will form the subject of future investigations.
1409: It is worth recalling that,
1410: in the context of zero-temperature dynamics of Ising spin chains,
1411: the super-exponential fall-off of correlations in metastable states
1412: has been emphasised as a signature of the generation
1413: of a non-trivial measure on the space of attractors~\cite{smedt},
1414: i.e., loosely speaking,
1415: of the violation of Edwards' flatness hypothesis~\cite{edwards}.
1416:
1417: \begin{figure}[htb]
1418: \begin{center}
1419: \includegraphics[angle=90,width=.48\linewidth]{figg.eps}
1420: \includegraphics[angle=90,width=.48\linewidth]{figh.eps}
1421: \caption{\small Logarithmic plot of the absolute
1422: connected on-axis correlation functions, against distance $n$.
1423: Left: survival correlation.
1424: Right: mass correlation.
1425: Top to bottom: one dimension (circles), two dimensions (squares),
1426: three dimensions (triangles).}
1427: \label{figgh}
1428: \end{center}
1429: \end{figure}
1430:
1431: \subsection{Mass distribution of survivors}
1432:
1433: To conclude this section, we examine the mass distribution of the survivors.
1434: In the late stages of the dynamics, when every cluster is isolated,
1435: its reduced mass grows according to Section~3, i.e., $x\sim\e^{(2\a-1)s/2}$.
1436: The mean reduced mass $\meansur{x}$ of the surviving clusters
1437: therefore exactly follows the same growth law.
1438: Hence it is natural to measure cluster masses with respect to their mean,
1439: and to introduce the ratios
1440: \beq
1441: X_\n=\frac{x_\n}{\meansur{x}}=\frac{m_\n}{\meansur{m}}.
1442: \eeq
1443: All the rescaled variables $X_\n$ are independent of time
1444: in the very late stages of the dynamics.
1445: They are therefore expected to have a well-defined
1446: limit probability distribution $P\infy(X)$.
1447: The first moment of this distribution is identically $\meansur{X}=1$,
1448: whereas~(\ref{disco}) implies
1449: \beq
1450: \meansur{X^2}
1451: =\frac{\meansur{x^2}}{\meansur{x}^2}
1452: =S\infy\frac{\mean{x^2}}{\mean{x}^2}
1453: =\frac{C_\s(\infty)}{C_x(\infty)}.
1454: \eeq
1455: We have measured the values $\meansur{X^2}\approx1.73$,
1456: $\meansur{X^2}\approx1.93$, and $\meansur{X^2}\approx2.10$,
1457: respectively in one, two, and three dimensions.
1458: Figure~\ref{figi} shows a plot of the whole rescaled distribution
1459: $P\infy(X)$ of the masses of survivors in these three cases.
1460: This distribution is observed to be both rather structureless
1461: and weakly dependent on dimensionality.
1462:
1463: \begin{figure}[htb]
1464: \begin{center}
1465: \includegraphics[angle=90,width=.6\linewidth]{figi.eps}
1466: \caption{\small
1467: Plot of the limit distribution $P\infy(X)$
1468: of the rescaled mass $X$ of surviving clusters.
1469: Circles: one dimension.
1470: Squares: two dimensions.
1471: Triangles: three dimensions.}
1472: \label{figi}
1473: \end{center}
1474: \end{figure}
1475:
1476: The cosmological origins of the present model
1477: were our main motivation for looking at the mass distribution of survivors.
1478: In that context, survivors would be the descendants of primordial black holes,
1479: which would form part of the dark matter in the Universe.
1480: The above results suggest that the population of survivors
1481: is essentially given by a single mass scale,
1482: growing as the mass of a single isolated cluster,
1483: whereas the superimposed cluster-to-cluster fluctuations
1484: are described by the harmless distribution $P\infy(X)$.
1485: This qualitative picture clearly ignores any cosmological details,
1486: but may still be of general interest from the viewpoint
1487: of using statistical-mechanical methods to probe such issues.
1488:
1489: \section{Discussion}
1490: \label{discussion}
1491:
1492: In the above we have presented and investigated in detail
1493: the many facets of a novel and very rich model
1494: at the interface between non-linear dynamics
1495: and non-equilibrium statistical mechanics.
1496: Despite its origins in a rather exotic context,
1497: namely accretion dynamics of black holes in a brane world~\cite{archan,I},
1498: the present model is of potential interest in many other situations,
1499: at least as far as qualitative features are concerned.
1500:
1501: As recalled in the beginning of the Introduction,
1502: the premise of equilibration generally holds in most physical instances.
1503: The rare exceptions to this principle
1504: include classical systems with long-range forces,
1505: with the noticeable example of the large-scale structure of the Universe.
1506: Similar instabilities, where tiny initial differences get amplified forever,
1507: are also met in other sciences,
1508: with one well-known example being the {\it rich-get-richer} principle
1509: in economics~\cite{simon}.
1510: It has been realised more recently that non-equilibrium
1511: statistical-mechanical models may exhibit a similar phenomenon,
1512: where a single microscopic state acquires a large population
1513: by virtue of a condensation transition, even in one dimension.
1514: The appearance of such a condensate can be viewed as a classical
1515: and non-equilibrium analogue of Bose-Einstein condensation~\cite{cond}.
1516: The scenario of {\it survival of the biggest} arising from
1517: our model can be viewed as an extreme example of this instability,
1518: where the condensate ends up containing the entire mass.
1519: We reiterate that the model is non-conserving, in the sense that the final
1520: cluster becomes eventually more massive than all the earlier ones put together.
1521: The physical reason for this is that the interaction term
1522: derives from a radiation field in its cosmological incarnation~\cite{I},
1523: which can be regarded as a mass reservoir.
1524:
1525: The present model also has many other specific features
1526: of interest in each of the geometries considered; we address each one in turn.
1527:
1528: First, for a finite assembly of coupled clusters,
1529: the model provides an interesting example of a deterministic dynamical system
1530: describing the evolution of competing agents.
1531: This class of problems has been studied at length in biophysics,
1532: one particularly well-known example being the Lotka-Volterra system
1533: in population dynamics as described by predator-prey models~\cite{murray}.
1534: The present model exhibits a whole variety of types of trajectories
1535: in the transient regime, encoded in Phases~I to~IV
1536: of the phase diagram shown in Figure~\ref{figa}.
1537: Another distinguishing feature of our model is that only one survivor
1538: remains after a sufficiently long time, leading to the description
1539: of this model as a {\it winner-takes-all} type of model,
1540: despite the lack of a conservation law.
1541:
1542: Next, the thermodynamical mean-field limit of our model
1543: clearly shows many features of glassy behaviour.
1544: The evolution consists of two successive stages:
1545: a fast individual dynamics in Stage~I,
1546: followed by a slow collective dynamics in Stage~II.
1547: In the weak-coupling regime, the characteristic reduced time scale $s_c$
1548: of the slow dynamics diverges as~$1/\gbar^2$ [see~(\ref{scmf})];
1549: the two time scales are well-separated
1550: in a way which is very reminiscent of the $\a$ and $\beta$ relaxations
1551: observed in most glassy systems~\cite{glassyrefs}.
1552: A particularly interesting feature in this context
1553: is the universality of Stage~II asymptotics,
1554: such as~(\ref{stlate}) or~(\ref{mtlate}),
1555: and the more unusual universality of the prefactor $C$,
1556: which only depends on the tail exponent of the initial mass distribution.
1557: A further distinguishing characteristic
1558: is that the time separation between fast and slow dynamics
1559: in our model is simply given in terms of the coupling constant,
1560: and becomes parametrically large in the weak-coupling regime; this
1561: is true both within and beyond mean-field theory
1562: [see respectively~(\ref{scmf}) and~(\ref{sc})].
1563: In most conventional glassy systems, the separation of time scales
1564: is governed by the appearance of a slowly growing length scale~$L(t)$,
1565: associated with some kind of ordering.
1566: By contrast, the glassiness in the present situation has dynamical origins:
1567: our model has features that are similar to driven systems,
1568: where time-scale separation arises either from
1569: a non-zero drift velocity~$V$~\cite{ourvoter}
1570: or a non-zero shear rate~$\gamma$~\cite{brayetal}.
1571: Such time-scale separations become parametrically large with the divergence
1572: (as $V\to0$ or $\gamma\to0$ for those models, and as~$g\to0$ in the present
1573: case) of the slow time scale.
1574:
1575: Beyond mean-field,
1576: e.g.~on finite-dimensional lattices with nearest-neighbour couplings,
1577: the principle of {\it survival of the biggest} only applies locally;
1578: thus isolated clusters of dissimilar sizes are able to survive independently.
1579: At least qualitatively, this recalls
1580: local screening mechanisms in a variety of growth models,
1581: including cluster aggregation in suspensions~\cite{hydrodynamic}.
1582: In our model, however, the screening is extreme, in the sense that
1583: growing clusters, once isolated, are survivors,
1584: i.e., survive and keep on growing forever.
1585: A direct consequence of this is that the model exhibits both aging
1586: and metastability in a way that is qualitatively similar to
1587: what is observed in the mean-field limit, even though the time scale
1588: of the slow stage of the dynamics diverges less rapidly
1589: in the weak-coupling regime.
1590: Furthermore, the aging phase gets interrupted,
1591: as the system gets eventually trapped in a metastable state
1592: where a finite fraction $S\infy$ of the entire lattice is occupied
1593: by isolated clusters which survive forever.
1594: If the clusters are initially large enough,
1595: the density $S\infy$ is only slightly below $1/2$,
1596: so that the spatial pattern of survivors has a local checkerboard structure.
1597: While similar checkerboard patterns have been observed
1598: in coupled map lattices~\cite{K,k},
1599: our attractors are distinguished by their absolute stability:
1600: they are created irreversibly by a deterministic dynamics
1601: from the fluctuations in the initial distribution of their initial masses.
1602: Once created, they then survive forever.
1603: Many questions regarding the statistics of the metastable states
1604: thus obtained remain open.
1605: It would be most interesting to know whether they
1606: are generated with a uniform measure~\`a la Edwards~\cite{edwards}
1607: in an appropriately defined ensemble, or with a highly non-trivial one,
1608: as suggested by recent investigations
1609: of one-dimensional spin models~\cite{smedt,franz}.
1610:
1611: \subsubsection*{Acknowledgements}
1612:
1613: It is a pleasure to thank Archan Majumdar for fruitful discussions,
1614: and Hugues Chat\'e for valuable comments,
1615: and especially for making us aware of References~\cite{K,k}.
1616:
1617: \begin{thebibliography}{99}
1618:
1619: \bibitem{pee}
1620: see e.g. P.J.E. Peebles, {\it The Large-Scale Structure of the Universe}
1621: (Princeton University Press, 1980).
1622:
1623: \bibitem{archan}
1624: A.S. Majumdar, Phys. Rev. Lett. {\bf 90}, 031303 (2003).
1625:
1626: \bibitem{I}
1627: A.S. Majumdar, A. Mehta, and J.M. Luck, preprint astro-ph/0311148.
1628:
1629: \bibitem{glassyrefs}
1630: see e.g. J.L. Barrat, J. Dalibard, M. Feigelman, and J. Kurchan (eds.),
1631: {\it Slow Relaxations and Nonequilibrium Dynamics in Condensed Matter},
1632: Proceedings of Les Houches Summer School, Session LXXVII (Springer, Berlin,
1633: 2003).
1634:
1635: \bibitem{ks}
1636: S. Kirkpatrick and D. Sherrington, Phys. Rev. B {\bf 17}, 4384 (1978).
1637:
1638: \bibitem{tap}
1639: D.J. Thouless, P.W. Anderson, and R.G. Palmer, Phil. Mag. {\bf 35}, 593 (1977).
1640:
1641: \bibitem{ktw}
1642: T.R. Kirkpatrick and P.G. Wolynes, Phys. Rev. A {\bf 35}, 3072 (1987);
1643: T.R. Kirkpatrick and D. Thirumalai, Phys. Rev. B {\bf 36}, 5388 (1987);
1644: T.R. Kirkpatrick and P.G. Wolynes, Phys. Rev. B {\bf 36}, 8552 (1987);
1645: D. Thirumalai and T.R. Kirkpatrick, Phys. Rev. B {\bf 38}, 4881 (1988).
1646:
1647: \bibitem{sw}
1648: F.H. Stillinger and T.A. Weber, Phys. Rev. A {\bf 25}, 978 (1982);
1649: Science {\bf 225}, 983 (1984).
1650:
1651: \bibitem{fv}
1652: S. Franz and M.A. Virasoro, J. Phys. A {\bf 33}, 891 (2000).
1653:
1654: \bibitem{smedt}
1655: G. De Smedt, C. Godr\`eche, and J.M. Luck, Eur. Phys. J. B {\bf 27}, 363 (2002).
1656:
1657: \bibitem{edwards}
1658: S.F. Edwards, in {\it Granular Matter: An Interdisciplinary Approach},
1659: A. Mehta (ed.) (Springer, New York, 1994).
1660:
1661: \bibitem{simon}
1662: see e.g. Y. Ijiri and H.A. Simon, {\it Skew Distributions and the Size
1663: of Business Firms} (North-Holland, New York, 1977).
1664:
1665: \bibitem{cond}
1666: see e.g. D. Mukamel, in {\it Soft and Fragile Matter:
1667: Nonequilibrium Dynamics, Metastability and Flow}, M.E. Cates and M.R. Evans,
1668: eds. (IOP Publishing, Bristol, 2000).
1669:
1670: \bibitem{murray}
1671: see e.g. J.D. Murray, {\it Mathematical Biology} (Springer, Berlin, 1989).
1672:
1673: \bibitem{ourvoter}
1674: A. Mehta and J.M. Luck, Phys. Rev. E {\bf 60}, 5218 (1999);
1675: J.M. Luck and A. Mehta, Europhys. Lett. {\bf 57}, 46 (2002).
1676:
1677: \bibitem{brayetal}
1678: A.J. Bray and A. Cavagna, J. Phys. A {\bf 33}, L305 (2000);
1679: A. Cavagna, A.J. Bray, and R.D.M. Travasso, Phys. Rev. E {\bf 62}, 4702 (2000);
1680: J. Bragard, P.L. Ramazza, F.T. Arecchi, S. Boccaletti, and L. Kramer,
1681: Phys. Rev. E {\bf 61}, R6045 (2000).
1682:
1683: \bibitem{hydrodynamic}
1684: see e.g. R. Jullien and R. Botet, {\it Aggregation and Fractal Aggregates}
1685: (World Scientific, Singapore, 1987);
1686: T. Vicsek, {\it Fractal Growth Phenomena}
1687: (World Scientific, Singapore, 1989);
1688: P. Meakin, {\it Fractals, Scaling, and Growth far from Equilibrium},
1689: Cambridge Nonlinear Science Series, vol.~{\bf 5}
1690: (Cambridge University Press, 1998).
1691:
1692: \bibitem{K}
1693: R. Kapral, Phys. Rev. A {\bf 31}, 3868 (1985).
1694:
1695: \bibitem{k}
1696: K. Kaneko, Physica D {\bf 34}, 1 (1989); {\bf 37}, 60 (1989);
1697: {\bf 41}, 137 (1990).
1698:
1699: \bibitem{franz}
1700: J. Berg, S. Franz, and M. Sellitto, Eur. Phys. J. B {\bf 26}, 349 (2002).
1701:
1702: \end{thebibliography}
1703: \end{document}
1704: