1: \documentclass[aps,twocolumn,groupedaddress,showkeys,showpacs]{revtex4}
2: \usepackage[dvips]{graphicx}
3: \bibliographystyle{apsrev}
4:
5: \begin{document}
6:
7: \title{Analytic study of the three-urn model for separation of sand}
8:
9: \author{G.\ M.\ Shim}
10: \author{B.\ Y.\ Park}
11: \author{J.\ D.\ Noh}
12: \author{Hoyun Lee}
13: \affiliation{Department of Physics, Chungnam National University,
14: Daejeon 305-764, Korea}
15:
16: \date{\today}
17:
18: \begin{abstract}
19: We present an analytic study of the three-urn model for separation of sand.
20: We solve analytically the master equation and the first-passage problem.
21: We find that the stationary probability distribution obeys the detailed balance
22: and is governed by the {\it free energy}.
23: We find that the characteristic lifetime of a cluster diverges algebraically
24: with exponent $1/3$ at the limit of stability.
25: \end{abstract}
26:
27: \pacs{45.70.--n,68.35.Rh}
28: %pacs:granular systems, phase transitions and critical phenomena
29: \keywords{granular, urn model, master equation, critical phenomena, symmetry breaking.}
30:
31: \maketitle
32:
33: %%%%%%%%%%%%%%%%%%%%%%%% section 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
34: \section{INTRODUCTION}
35:
36: A granular system consisting of macroscopic particles exhibits extremely
37: rich phenomena, which has been studied both experimentally and
38: theoretically~\cite{Kadanoff}. One of such interesting phenomena is the
39: spatial separation of shaken sand.
40: In the experiment by Schlichting and Nordmeier~\cite{Schlichting96},
41: granular particles are prepared in a box which is mounted on a shaker and
42: separated into two equal parts by a wall. There is a slit on a wall through
43: which particles can move from one compartment to the other.
44: Under a certain shaking condition, the granular particles simultaneously
45: separate into dense and dilute regions, which will not occur for gaseous
46: particles, due to the dissipative nature of the macroscopic particle
47: collision.
48:
49: The emergence of symmetry breaking in shaken sand was first explained by
50: Eggers using a hydrodynamic approach~\cite{Eggers99}, and later by
51: Lipowski and Droz using an urn model which is supposed to
52: capture the essence of the experimental system~\cite{Lipowski02a}.
53: In the urn model, $N$ granular particles are distributed into $L$
54: urns. And each particle can jump from one urn to another with the
55: probability controlled by a parameter, called an effective temperature,
56: which depends explicitly on the density of particles in each urn.
57: The dissipative nature of the particle collision is incorporated into the
58: model with the density-dependent effective temperature.
59:
60: The urn model is simple enough to allow extensive numerical
61: calculations~\cite{Lipowski02a,Coppex02}
62: as well as analytical studies~\cite{Shim03,Bena03}.
63: The two-urn~($L=2$) model shows a rich structure with
64: symmetric, mixed, and asymmetric phases separated with continuous
65: and discontinuous transitions as well as the tricritical point.
66: In the symmetric phase particles are distributed equally in each urn,
67: while the symmetry is broken in the asymmetric phase. In the mixed phase,
68: both the symmetric and asymmetric states are stable.
69: It was also found that the characteristic time which it takes to reach
70: the symmetric state from an asymmetric state is given by
71: the same free energy function which governs the stationary probability
72: distribution~\cite{Shim03}.
73: Coppex et al.\ also numerically investigated the three-urn model to
74: find the absence of continuous transition~\cite{Coppex02}.
75: They also obtained that the characteristic time at the limit of stability
76: diverges algebraically as the number of particles increases~(see also
77: Ref.~\cite{vanderMeer}).
78:
79: In this paper, we present the results of an analytic study to the
80: master equation and the characteristic time scale in the three-urn model.
81: We solve the master equation in the thermodynamic limit to find that the
82: solution shows a nature of deformed wave.
83: We also obtain the stationary probability distribution in a finite system,
84: which interestingly obeys the detailed balance. Combining the knowledge of
85: the characteristic scales of those distribution and the mean-field flux
86: equation, we obtain the scaling property of the characteristic time scale
87: at the limit of stability.
88:
89: The paper is organized as follows.
90: In Sec.~\ref{sec:II} we briefly review the model and its master equation.
91: In Sec.~\ref{sec:III} we present the analytic solution of the master
92: equation in the thermodynamic limit and the analytic expression of
93: the stationary probability distribution.
94: In Sec.~\ref{sec:IV} we investigate the scaling law of the
95: characteristic time scale.
96: Section \ref{sec:V} is devoted to conclusions and discussions.
97:
98: %%%%%%%%%%%%%%%%%%%%%%%% sec 2 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
99: \section{MODEL AND ITS MASTER EQUATION} \label{sec:II}
100:
101: The model introduced by Coppex et al.\ \cite{Coppex02} is
102: defined as follows. $N$ particles are distributed between three urns,
103: and the number of particles in each urn is denoted as $N_1$, $N_2$,
104: and $N_3=N-N_1-N_2$, respectively.
105: At each time of updates one of the $N$ particles is randomly chosen.
106: Let $n$ be a fraction of the total number of particles in the urn
107: which the selected particle belongs to.
108: With probability $\exp(-\frac{1}{T(n)})$ the selected particle moves
109: to a randomly chosen neighboring urn. $T(n) = T_0+\Delta(1-n)$ is the
110: effective temperature of an urn with particles $nN$ that measures the
111: strength of fluctuations in the urn.
112: For more detailed description of the model,
113: we refer readers to Ref.~\cite{Lipowski02a}.
114:
115: It is easy to derive the master equation for the probability
116: distribution $p(N_1,N_2,N_3,t)$ that there are $N_i$ particles in
117: urn $i$ at time $t$
118: \begin{eqnarray}
119: && p(N_1,N_2,N_3,t+1) =
120: \nonumber \\
121: &&\hspace{1em} F\bigl(\frac{N_1+1}{N}\bigr) \frac12\biggl(
122: p(N_1+1,N_2-1,N_3,t)
123: \nonumber \\
124: &&\hspace{5em} +p(N_1+1,N_2,N_3-1,t) \biggr)
125: \nonumber\\
126: &&\hspace{1em} + F\bigl(\frac{N_2+1}{N}\bigr) \frac12\biggl(
127: p(N_1-1,N_2+1,N_3,t)
128: \nonumber \\
129: &&\hspace{5em} +p(N_1,N_2+1,N_3-1,t) \biggr)
130: \nonumber\\
131: &&\hspace{1em} + F\bigl(\frac{N_3+1}{N}\bigr) \frac12\biggl(
132: p(N_1-1,N_2,N_3+1,t)
133: \nonumber \\
134: &&\hspace{5em} +p(N_1,N_2-1,N_3+1,t) \biggr)
135: \nonumber\\
136: &&\hspace{1em} + \Bigl[ 1-\sum_{i=1}^3 F\bigl(\frac{N_i}{N}\bigr)
137: \Bigr]\ p(N_1,N_2,N_3,t)
138: \,,\label{eq:master}\end{eqnarray}
139: where $F(n) = n \exp(-\frac{1}{T(n)})$ measures the flux of
140: particles leaving the given urn.
141: Here we introduced for convenience the notation $p(N_1,N_2,N_3,t)=0$
142: if $N_1+N_2+N_3 \ne N$ or any $N_i$'s is either less than 0 or greater
143: than $N$.
144:
145: Let's denote the occupancies of the urns by $n_i = N_i/N$.
146: The time evolution of the averaged particle occupancies
147: is governed by the equations
148: \begin{equation}\label{eq:averagedoccupancy}
149: \langle n_i\rangle_{t+1} = \langle n_i\rangle_t
150: +\frac1{N} \langle {\cal F}_i(\vec{n}) \rangle_t
151: \,,\end{equation}
152: where $\langle \cdots \rangle_t =
153: \sum_{N_1,N_2,N_3} (\cdots) p(N_1,N_2,N_3,t)$,
154: and ${\cal F}_i(\vec{n}) = \frac12\sum_{k=1}^3F(n_k)-\frac32F(n_i)$.
155: The unit of time may be chosen so that there is a single update per
156: a particle on average. Scaling the time by $N$, expanding
157: Eq.~(\ref{eq:averagedoccupancy}) with respect to $\frac{1}{N}$,
158: and using the mean-field approximation in evaluating the average, we get
159: \begin{equation}\label{eq:differentialoccupancy}
160: \frac{d}{dt}n_i(t) = {\cal F}_i(\vec{n}(t))
161: \,.\end{equation}
162: where $n_i(t) = \langle n_i \rangle_t$.
163:
164: Detailed analysis on the existence of the stable stationary solutions of
165: Eq.~(\ref{eq:differentialoccupancy}) was done by Coppex et al. \cite{Coppex02}.
166: We here display their phase diagram in Fig.~\ref{fig:phasediagram} to make
167: our paper as self-contained as possible.
168: The stable symmetric solution ($n_1=n_2=n_3$) exists in region I, III, and
169: IV while the stable asymmetric solution ($n_1>n_2=n_3$) exists in region II,
170: III and IV. Note that
171: Eq.~(\ref{eq:differentialoccupancy}) is obtained with mean-field
172: approximation that $\langle {\cal F}(\vec{n})\rangle = {\cal F}(\langle
173: \vec{n}\rangle)$. However, as will be shown later, the relation becomes
174: exact for the delta-peaked initial probability distribution, so is the phase diagram.
175:
176:
177: \begin{figure}[tbp]%--------------------------------------------%
178: \includegraphics*[width=\columnwidth]{fig1.eps}
179: \caption{\label{fig:phasediagram} Phase diagram of the three-urn model
180: \cite{Coppex02}.
181: The symmetric solution vanishes continuously on the solid line
182: while the asymmetric one disappears discontinuously on the
183: dotted line. The transition of the behavior of the stationary
184: probability distribution is denoted by the dashed line. For
185: the inset, see Sec.~\ref{sec:IV}.}
186: \end{figure}%--------------------------------------------------%
187:
188:
189: %%%%%%%%%%%%%%%%%%%%%%%% section 3 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
190: \section{THE SOLUTION OF THE MASTER EQUATION} \label{sec:III}
191:
192: We are mainly interested in investigating the properties of the
193: infinite system.
194: Consider the thermodynamic limit $N \rightarrow \infty$ with
195: the occupancies of the urns $n_i$ being fixed.
196: Representing the probability distribution by $n_i$ instead of $N_i$,
197: scaling the time by $N$,
198: expanding Eq.~(\ref{eq:master}) with respect to $\frac{1}{N}$,
199: and keeping the terms up to the first order,
200: we arrive at the partial differential equation
201: \begin{equation}\label{eq:continuousmaster}
202: \frac{\partial}{\partial t}p(\vec{n},t)
203: + \frac{\partial}{\partial n_1}\bigl[
204: {\cal F}_1(\vec{n})p(\vec{n},t) \bigr]
205: + \frac{\partial}{\partial n_2}\bigl[
206: {\cal F}_2(\vec{n})p(\vec{n},t) \bigr]
207: = 0
208: \,.\end{equation}
209: Here we would like to remind that $n_3 = 1-n_1-n_2$ so that the independent
210: variables are $n_1, n_2$.
211:
212: Note that Eq.~(\ref{eq:continuousmaster}) may be interpreted as the continuity
213: equation for the probability with velocity $({\cal F}_1,{\cal F}_2)$.
214: Since the velocity is already given as a function of $\vec{n}$, the
215: solution to Eq.~(\ref{eq:continuousmaster}) can be formally found as
216: follows. Time evolution of a point located at $\vec{r}=(x,y,1-x-y)$ at $t=0$
217: is determined by the Eq.~(\ref{eq:differentialoccupancy}). We denote it by
218: $\vec{R}(t;\vec{r})$.
219: We show its typical trajectories in Fig.~\ref{fig:flow}.
220: In the symmetric phase (region I in Fig.~\ref{fig:phasediagram}) there is
221: only the stable fixed point corresponding to the symmetric state so that
222: every trajectory flows to that point. This is shown in Fig.~\ref{fig:flow}(a).
223: Figure \ref{fig:flow}(b) shows a typical behavior for the asymmetric
224: phase (region II in Fig.~\ref{fig:phasediagram}). In this case, there are
225: three stable fixed points corresponding to asymmetric states and one
226: unstable fixed points corresponding to the symmetric state as well as
227: three saddle points. The trajectories are separated by the separatrices
228: denoted by dashed lines.
229: Finally Fig.~\ref{fig:flow}(c) shows a
230: typical behavior for the mixed phase (region III and IV in
231: Fig.~\ref{fig:phasediagram}). There are one stable symmetric fixed point
232: and three asymmetric fixed points as well as three saddle points.
233: The trajectories flowing toward the stable fixed points are separated
234: by the separatrices.
235: \begin{figure}[tbp]%--------------------------------------------%
236: \includegraphics*[width=7cm]{fig2a.eps}
237: \includegraphics*[width=7cm]{fig2b.eps}
238: \includegraphics*[width=7cm]{fig2c.eps}
239: \caption{\label{fig:flow} Typical trajectories of
240: Eq.~(\ref{eq:differentialoccupancy}).
241: We plotted the trajectories in the diagonal plane of a unit cube
242: so that each variable $n_i$ are treated in the same way.
243: (a) $\Delta=0.5, T_0=0.5$ (symmetric phase)
244: (b) $\Delta=0.25, T_0=0.05$ (asymmetric phase)
245: (c) $\Delta=1.0, T_0=0.1$ (mixed phase).
246: The stable, the unstable, and the saddle point are represented with a filled
247: circle, a filled square, and a cross, respectively.}
248: \end{figure}%--------------------------------------------------%
249:
250: Since the continuity equation implies that the probability at point
251: $\vec{r}$ evolves according to $\vec{R}(t;\vec{r})$,
252: we obtain a formal solution of Eq.~(\ref{eq:continuousmaster})
253: \begin{equation}\label{eq:formalsolution}
254: p(\vec{n},t) = \int dxdy\, p(\vec{r},0)
255: \delta^{(2)}\biggl(\vec{n}-\vec{R}(t;\vec{r})\biggr)
256: \,.\end{equation}
257: It implies that the initial probability distribution in the basin of
258: attraction associated with a stable fixed point will eventually
259: accumulate at that point. As a consequence, in the long time limit
260: $t\rightarrow\infty$ the probability distribution becomes a sum of delta
261: peaks at the stable fixed points denoted by $\vec{n}_i$;
262: \begin{equation}\label{eq:probdistinfty}
263: p(\vec{n},\infty) = \sum_i p_i \delta^{(2)}(\vec{n}-\vec{n}_i)
264: \,,\end{equation}
265: where $p_i$ are the sum of the initial probabilities in the basin of
266: attraction associated with $\vec{n}_i$.
267:
268: Now we will consider another limit in the master equation
269: (\ref{eq:master}), namely take the long time limit $t \rightarrow
270: \infty$ before we take the limit $N \rightarrow \infty$.
271: That is, we are looking for the stationary probability distribution for a
272: finite system. Though this limit may not properly reflect the properties
273: of the infinite system, we expect it will reveal interesting properties
274: of the system concerning the characteristic times (See \cite{Shim03} in
275: two-urn case.).
276:
277: Let's first take the long time limit of $t \rightarrow \infty$ in
278: Eq.~(\ref{eq:master}). In this limit we may drop off the time dependence in
279: the probability distribution, which now takes the form
280: \begin{eqnarray}
281: \left[ F(\frac{N_1+1}{N}) \frac{p(N_1+1,N_2-1,N_3)}{p(N_1,N_2,N_3)}
282: - F(\frac{N_2}{N}) \right] &+& \nonumber \\
283: \left[ F(\frac{N_2+1}{N}) \frac{p(N_1,N_2+1,N_3-1)}{p(N_1,N_2,N_3)}
284: - F(\frac{N_3}{N}) \right] &+& \nonumber \\
285: \left[ F(\frac{N_3+1}{N}) \frac{p(N_1-1,N_2,N_3+1)}{p(N_1,N_2,N_3)}
286: - F(\frac{N_1}{N}) \right] &+& \nonumber \\
287: \left[ F(\frac{N_1+1}{N}) \frac{p(N_1+1,N_2,N_3-1)}{p(N_1,N_2,N_3)}
288: - F(\frac{N_3}{N}) \right] &+& \nonumber \\
289: \left[ F(\frac{N_3+1}{N}) \frac{p(N_1,N_2-1,N_3+1)}{p(N_1,N_2,N_3)}
290: - F(\frac{N_2}{N}) \right] &+& \nonumber \\
291: \left[ F(\frac{N_2+1}{N}) \frac{p(N_1-1,N_2+1,N_3)}{p(N_1,N_2,N_3)}
292: - F(\frac{N_1}{N}) \right] &=& 0
293: \!.\label{eq:stationaryprob}\end{eqnarray}
294:
295: In contrast to the corresponding equation (13) in the two-urn model
296: \cite{Shim03}, it does not show a simple tridiagonal structure.
297: However, we can show that the stationary probability distribution
298: determined by Eq.~(\ref{eq:stationaryprob}) obeys the {\it detailed balance}
299: \begin{equation}\label{eq:detailedbalance}
300: F\bigl(\frac{N_i+1}{N}\bigr)p(N_1',N_2',N_3')
301: =
302: F\bigl(\frac{N_j}{N}\bigr)p(N_1,N_2,N_3)
303: \,,\end{equation}
304: where $\{i,j,k\}=\{1,2,3\}$ and $N_i'=N_i+1, N_j'=N_j-1,N_k'=N_k$.
305: We also would like to note that this kind of relation is obeyed in
306: two-urn model.
307: Equation (\ref{eq:stationaryprob}) essentially tells us that probability
308: distribution with $N_j$ is given by that with $N_j-1$, which is in turn
309: given by that with $N_j-2$, and so on. Repeatedly using
310: Eq.~(\ref{eq:stationaryprob}) with $i$ as 1 or 2 and $i=3$, we obtain
311: \begin{equation}\label{eq:stationarysolution}
312: p(N_1,N_2,N_3) = \frac{
313: \prod_{k=1}^{N_1+N_2}F\bigl(\frac{N-k+1}{N}\bigr)
314: }{
315: \prod_{i=1}^{N_1}F\bigl(\frac{i}{N}\bigr)
316: \prod_{j=1}^{N_2}F\bigl(\frac{j}{N}\bigr)
317: }p(0,0,N)
318: \,,\end{equation}
319: where $p(0,0,N)$ appears as an overall factor to normalize the
320: probabilities so that we get
321: $$
322: p(0,0,N) = \biggl[1+\sum_{N_1=0}^N\sum_{N_2=0}^{N-N_1} \frac{
323: \prod_{k=1}^{N_1+N_2}F\bigl(\frac{N-k+1}{N}\bigr)
324: }{
325: \prod_{i=1}^{N_1}F\bigl(\frac{i}{N}\bigr)
326: \prod_{j=1}^{N_2}F\bigl(\frac{j}{N}\bigr) }
327: \biggr]^{-1} \,.
328: $$
329:
330: Now let's take the limit $N \rightarrow \infty$.
331: With $\frac{N_1}{N} = n_1$, $\frac{N_2}{N}=n_2$,
332: and scaling the probability distribution by
333: $N^2$, the stationary probability distribution for large $N$ now becomes
334: \begin{equation}\label{eq:statcontinprob}
335: p(\vec{n}) \approx \frac{ e^{NG(\vec{n})} }{
336: \int_0^1 dx \int_0^{1-x} dy \, e^{NG(\vec{r})} }
337: \,,\end{equation}
338: where
339: \begin{eqnarray}
340: G(\vec{n}) &=& \int_0^{n_1+n_2} dt \, \bigl[ \ln F(1-t) \bigr]
341: \nonumber \\ \label{eq:nfe}
342: && - \int_0^{n_1} dt \, \bigl[ \ln F(t) \bigr]
343: - \int_0^{n_2} dt \, \bigl[ \ln F(t) \bigr]
344: \,.\end{eqnarray}
345: We will call $G(\vec{n})$ the (negative) free energy function.
346:
347: In the limit $N \rightarrow \infty$, the main contribution to the
348: stationary probability distribution comes only from the maximum of
349: $G(\cdot)$, and it becomes the sum of delta peaks.
350: The maximum of $G(\vec{n})$ occurs when
351: \begin{equation}
352: \frac{\partial}{\partial n_1}G(\vec{n})
353: =\frac{\partial}{\partial n_2}G(\vec{n})
354: =0
355: \end{equation}
356: or
357: \begin{equation}
358: F(n_1) = F(n_2) = F(n_3)
359: \,.\end{equation}
360: This condition is equivalent to the stationary condition of the flux
361: equation (\ref{eq:differentialoccupancy}). Note that in region II only
362: the three stable asymmetric solutions having the same maximum
363: exist, while in region I only the symmetric solution is stable.
364: Therefore the stationary probability distribution has the triple peaks
365: in region II and only the central peak in region I.
366: In region III and IV both the
367: symmetric and the asymmetric solutions are stable so that the maximum
368: of $G(\vec{n})$ should be determined by comparing its values at the
369: stable fixed points. The crossover of the maximum point occurs when
370: both values coincide. This implies that the transition between
371: the triple peaks and the central single peak in the probability
372: distribution is determined by the condition
373: $\Delta G = G(\vec{n}_a)-G(1/3,1/3,1/3) = 0$, where $\vec{n}_a$ is one of
374: the stable asymmetric fixed points.
375: This condition yields a line that separates two regions III and IV.
376:
377:
378:
379: %%%%%%%%%%%%%%%%%%%%%%%% section 4 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
380: \section{CHARACTERISTIC TIME Scale}\label{sec:IV}
381: One of the interesting phenomena in sands separated by compartments is
382: a sudden collapse of a granular cluster~\cite{vanderMeer,Coppex02}.
383: It is observed experimentally that a granular cluster with majority
384: of sand grains in a single compartment remains stable for a long time
385: until it collapses abruptly and diffuses over all
386: compartments~\cite{vanderMeer}. The urn model proposed in
387: Ref.~\cite{Coppex02} displays the same phenomenon: In region I
388: a granular cluster is stable up to a time scale $\tau$, after which sand
389: particles are distributed uniformly over all boxes. Approaching the phase
390: boundary I-IV, the characteristic time scale diverges. At the phase
391: boundary, it is found numerically that $\tau$
392: scales as $\tau \sim N^z$ with $z\simeq 0.32$~\cite{Coppex02}.
393: In this section, the scaling law for the characteristic time $\tau$
394: at and near the phase boundary is derived analytically.
395:
396: The master equation in Eq.~(\ref{eq:continuousmaster}) in the large $N$
397: limit does not contain a diffusive term. It implies that a delta-peaked
398: probability distribution remains delta-peaked during time evolution, which
399: enabled us to write down the formal solution in
400: Eq.~(\ref{eq:formalsolution}). The dispersion-less nature also guarantees
401: that the mean field approximation in Eq.~(\ref{eq:differentialoccupancy})
402: for the occupancy $n_i$ becomes exact as long as the initial values of
403: $n_i$'s are prescribed. Hence we can study the dynamics of the granular
404: cluster using Eq.~(\ref{eq:differentialoccupancy}) with the initial
405: condition $n_1=n_2=0$ and $n_3=1$.
406:
407: With $n_1=n_2=n$ and $n_3 = 1-2n$, one can rewrite
408: Eq.~(\ref{eq:differentialoccupancy}) as
409: \begin{equation}
410: \dot{n} = V(n) \equiv \frac{1}{2}\left\{F(1-2n)-F(n)\right\}
411: \end{equation}
412: with the initial condition $n(t=0)=0$.
413: The cluster dynamics is then determined from the property of the flow
414: function $V(n)$. In the inset of Fig.~\ref{fig:phasediagram},
415: we show the plots of the
416: flow function at different values of $T_0$ with $\Delta=0.3$ fixed. At
417: $\Delta=0.3$, the critical point separating the region I and VI is given by
418: $T_0 = T_{0c} = 0.1698\cdots$. For $T_0>T_{0c}$, $n=1/3$ is the unique
419: stationary state solution; $n$ grows from zero to 1/3 to reach the symmetric
420: state with $n_1=n_2=n_3=1/3$. Note that there exists a local minimum in
421: $V(n)$ at $0<n_0<1/3$ where the flow velocity is minimum.
422: Hence the value of $n$ remains at the intermediate value $n\simeq n_0$
423: for a long time, and then converges to $n=1/3$ quickly.
424: It turns out that the sudden collapse of a granular
425: cluster~\cite{Coppex02,vanderMeer} is due to the local minimum in $V(n)$.
426: The characteristic time scale or the life time $\tau$ is given by
427: \begin{equation}\label{eq:lifetime}
428: \tau = \int_0^{n_0} \frac{dn}{V(n)} \ .
429: \end{equation}
430: As $T_0\rightarrow T_{0c}^+$, the minimum flow velocity $V(n_0)$ decreases
431: and hence $\tau$ increases.
432: The asymmetric solution with $n_1=n_2<n_3$ emerges at $T=T_{0c}$ when
433: $V(n_{0})=0$ and $\tau$ diverges.
434:
435:
436: The scaling law for the characteristic time scale is determined from the
437: analytic property of $V(n)$ near $T_0\simeq T_{0c}$ and $n\simeq n_0$.
438: One can expand as $V(n) \simeq a (n-n_0)^2 + b (T_0 - T_{0c})$,
439: where $a$ and $b$ are constants of order one and all higher order terms in
440: $(T_0-T_{0c})$ are irrelevant. Inserting it into Eq.~(\ref{eq:lifetime}),
441: one obtains that the characteristic time scale diverges as $\tau\sim
442: (T_0-T_{0c})^{-1/2}$~(see also Ref.~\cite{vanderMeer}).
443: At $T_0 = T_{0c}$, the granular system approaches the asymmetric state
444: algebraically in time as $|n-n_0| \sim t^{-1}$ whose life time $\tau$
445: diverges. That is to say, the asymmetric state is stable in the
446: $N\rightarrow \infty$ limit.
447:
448: However, for finite $N$, the system could evolve into the symmetric state
449: at $T_0 = T_{0c}$ with the help of statistical fluctuations in $n$.
450: Indeed, the probability distribution $p$ has a finite dispersion for finite
451: $N$, even though it becomes a sum of delta peaks for infinite $N$.
452: In order to estimate the order of magnitude of fluctuations in $n$, we
453: investigate the stationary probability distribution in
454: Eq.~(\ref{eq:statcontinprob}) at $T_0 = T_{0c}$ in detail. Near the
455: asymmetric state with $n_1=n_2=n$ and $n_3 = 1-2n$, the (negative) free
456: energy function in Eq.~(\ref{eq:nfe}) is written as $G(n) = \int_0^{2n}dt
457: \ln F(1-t) - 2 \int_0^n dt \ln F(t)$. Now we expand it near $n=n_0$ to yield
458: $G(n) = G(n_0) + G'(n_0)(n-n_0) + \frac{1}{2}G''(n_0) (n-n_0)^2 +
459: \frac{1}{3} G'''(n_0) (n-n_0)^3 + \cdots$. After straightforward
460: calculations, one can show that
461: \begin{eqnarray}
462: G'(n_0) &=& 2\ln (F(1-2n_0)/F(n_0)) \\
463: G''(n_0) &=& \frac{2}{F(n_0)}V'(n_0) \\
464: G'''(n_0) &=& \frac{2}{F(n_0)}V''(n_0) \ .
465: \end{eqnarray}
466: Note that $G'(n_0)=G''(n_0)=0$ since $V(n_0) =
467: V'(n_0)=0$. Therefore, the (negative) free energy function can be
468: approximated as $G(n) \simeq c (n-n_0)^3$ with a constant $c$, and hence the
469: probability distribution as $p_s (n) \simeq e^{cN(n-n_0)^3}$ up to a
470: normalization constant.
471:
472: The stationary probability distribution suggests that the magnitude of
473: the fluctuation in $n$ near $n_0$ is of order $\delta n \sim N^{-1/3}$.
474: With the help of the fluctuation, the granular system could get away from
475: the asymmetric state when $|n(t) - n_0| \sim \delta n$, and flow into the
476: symmetric state. Therefore the characteristic time scale is given by
477: $\tau \simeq \int_0^{n_0-\delta n} dt / V(t)$, which results in $\tau \sim N^z$
478: with the dynamic exponent
479: \begin{equation}\label{eq:dynamicexponent}
480: z = 1/3 \ .
481: \end{equation}
482: Our analytic result confirms the numerical result $z \simeq 0.32$ reported
483: in Ref.~\cite{Coppex02}.
484:
485:
486: %{\large\it To be done by Dr. Noh.}
487:
488:
489:
490: %%%%%%%%%%%%%%%%%%%%%%%% section 6 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
491: \section{CONCLUSIONS}\label{sec:V}
492:
493: We analytically investigate the three-urn model introduced by Coppex
494: et al.\ \cite{Coppex02}
495: We formally solve the master equation of the
496: model in the thermodynamic limit and find how the probability
497: distribution evolves.
498: In the long time limit, the probability distribution becomes
499: delta peaks only at the stable fixed points.
500: The strength of a delta peak is equal to the sum of initial
501: probabilities in the basin of attraction associated with that fixed
502: point.
503:
504: We solve exactly the stationary probability distribution where we
505: take the long time limit before we take thermodynamic limit.
506: We find the distribution obeys the detailed balance.
507: Regardless of the initial probability distribution it shows triple
508: peaks or a single central peak depending on the parameters of the
509: system. The final formula of the stationary probability distribution
510: resembles that of the equilibrium systems, where the transition from
511: the triple peaks to the single peak is determined by the condition
512: that {\it free energies} of two phases become equal.
513:
514: We also obtain the exact scaling law for the characteristic time scale
515: $\tau$ which it takes to reach the symmetric state from an asymmetric
516: state near and at the phase boundary I-VI. In the symmetric phase~(region
517: I), the granular cluster is unstable and $\tau$ is finite. It grows as one
518: approach the phase boundary as $\tau \sim (T_0 - T_{0c})^{-1/2}$.
519: At $T_0 = T_{0c}$, the characteristic time diverges algebraically as $\tau
520: \sim N^z$ with the dynamic exponent $z=1/3$.
521:
522:
523: %%%%%%%%%%%%%%%%%%%%%%%% references %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
524: \begin{thebibliography}{99}
525: \bibitem{Kadanoff} L. P. Kadanoff, Rev. Mod. Phys. {\bf 71}, 435 (1999).
526: \bibitem{Schlichting96} H.\ J.\ Schlichting and V.\ Nordmeier,
527: Math.\ Naturwiss.\ Unterr.\ {\bf 49}, 323 (1996).
528: \bibitem{Eggers99} J.\ Eggers,
529: Phys.\ Rev.\ Lett.\ {\bf 83}, 5322 (1999).
530: \bibitem{Lipowski02a} A.\ Lipowski and M.\ Droz,
531: Phys.\ Rev.\ E {\bf 65}, 031307 (2002).
532: \bibitem{Coppex02} F.\ Coppex, M.\ Droz, and A.\ Lipowski,
533: Phys.\ Rev.\ E {\bf 66}, 011305 (2002).
534: \bibitem{Shim03} G.\ M.\ Shim, B.\ Y.\ Park, and H.\ Lee,
535: Phys.\ Rev.\ E {\bf 67} 011301 (2003).
536: \bibitem{Bena03} I. Bena, F. Coppex, M. Droz, and A. Lipowski,
537: Phys. Rev. Lett. {\bf 91}, 160602 (2003).
538: \bibitem{vanderMeer} D. van der Meer, K. van der Weele, and D. Lohse,
539: Phys. Rev. Lett. {\bf 88} 174302 (2002).
540: %\bibitem{Ehrenfest90} P.\ Ehrenfest and T.\ Ehrenfest,
541: % {\it The Conceptual Foundations of the
542: % Statistical Approach in Mechanics}
543: % (Dover, New York, 1990);
544: % M.\ Kac and J.\ Logan,
545: % in {\it Fluctuation Phenomena},
546: % edited by E.\ W.\ Montroll and J.\ L.\ Lebowitz
547: % (North-Holland, Amsterdam, 1987).
548: %\bibitem{Lipowski02b} A.\ Lipowski and M.\ Droz, cond-mat/0201472.
549: \end{thebibliography}
550:
551: \end{document}
552:
553: