1: %\documentclass[amsmath,amssymb,aps]{revtex4}
2: \documentclass[12pt]{iopart}
3: \usepackage{graphicx}
4: \usepackage{cite}
5: \usepackage{amssymb}
6: \newcommand{\be}{\begin{equation}}
7: \newcommand{\ee}{\end{equation}}
8: \newcommand{\bel}[1]{\begin{equation}\label{#1}}
9: \newcommand{\bea}{\begin{eqnarray}}
10: \newcommand{\eea}{\end{eqnarray}}
11: \newcommand{\ba}{\begin{array}}
12: \newcommand{\ea}{\end{array}}
13: \newcommand{\noin}{\noindent}
14: \newcommand{\bra}[1]{\mbox{$\langle \, {#1}\, |$}}
15: \newcommand{\ket}[1]{\mbox{$| \, {#1}\, \rangle$}}
16: \newcommand{\exval}[1]{\mbox{$\langle \, {#1}\, \rangle$}}
17: \newcommand{\smfrac}[2]{\mbox{\small$\frac{#1}{#2}$}}
18: \newcommand{\eps}{\varepsilon} % epsilon
19: \newcommand{\vph}{\varphi} % rundes phi
20: \newcommand{\vth}{\vartheta} % Deutsch-Delta
21: \newcommand{\D}{{\rm d}} % gerades d fuer Ableitungen
22: \newcommand{\II}{{\rm i}} % gerades i fuer komplexe Einheit
23: \def\eg{{\it e.g.} }
24: \def\ie{{\it i.e.} }
25: \renewcommand{\eta}{n}
26:
27: \begin{document}
28:
29: \title{Zero-range process with open boundaries}
30: \author{E. Levine$^{(a)}$, D. Mukamel$^{(a)}$, and G.M. Sch\"{u}tz$^{(b)}$}
31: \address{$(a)$ Department of Physics of Complex Systems,
32: Weizmann Institute of Science, Rehovot, Israel 76100.\\
33: $(b)$ Institut f\"{u}r Festk\"{o}rperforschung, Forschungszentrum
34: J\"{u}lich, 52425 J\"{u}lich, Germany.}
35:
36: \date{\today}
37:
38: \begin{abstract}
39: We calculate the exact stationary distribution of the
40: one-dimensional zero-range process with open boundaries for
41: arbitrary bulk and boundary hopping rates. When such a
42: distribution exists, the steady state has no correlations between
43: sites and is uniquely characterized by a space-dependent fugacity
44: which is a function of the boundary rates and the hopping
45: asymmetry. For strong boundary drive the system has no stationary
46: distribution. In systems which on a ring geometry allow for a
47: condensation transition, a condensate develops at one or both
48: boundary sites. On all other sites the particle distribution
49: approaches a product measure with the finite critical density
50: $\rho_c$. In systems which do not support condensation on a ring,
51: strong boundary drive leads to a condensate at the boundary.
52: However, in this case the local particle density in the interior
53: exhibits a complex algebraic growth in time. We calculate the bulk
54: and boundary growth exponents as a function of the system
55: parameters.
56: \end{abstract}
57:
58: \maketitle
59:
60: \section{Introduction}
61:
62: The zero-range process (ZRP) has originally been introduced in
63: 1970 by Spitzer \cite{Spit70} as a model system for interacting
64: random walks, where particles on a lattice hop randomly to other
65: neighboring sites. The hopping rates $w_n$ depend only on the
66: number of particles $n$ at the departure site. This model has
67: received renewed attention because of the occurrence of a
68: condensation transition \cite{Evan96,Krug96,Oloa98,Evan00,Jeon00a}
69: analogous to Bose-Einstein condensation and because of its close
70: relationship with exclusion processes \cite{Schu03}. Condensation
71: phenomena are well-known in colloidal and granular systems (see
72: \cite{Shim04} for a recent study making a connection with the
73: zero-range process), but appear also in other contexts, such as
74: socio-economics \cite{Burd01} and biological systems \cite{Froh75}
75: as well as in traffic flow \cite{traffic} and network theory
76: \cite{Bian01,Doro03}. In the mapping of the ZRP to exclusion
77: processes in one space dimension, condensation corresponds to
78: phase separation. The ZRP has served for deriving a quantitative
79: criterion for the existence of non-equilibrium phase separation
80: \cite{Kafr02} in the otherwise not yet well-understood driven
81: diffusive systems with two conservation laws. For the occurrence
82: of condensation the dimensionality of the ZRP does not play a role
83: and we shall consider only the one-dimensional (1d) case.
84:
85: Most studies of the ZRP focus on periodic boundary conditions or
86: the infinite system. Under certain conditions on the rates $w_n$
87: (see below) the grand-canonical stationary distribution is a
88: product measure, i.e. there are no correlations between different
89: sites \cite{Andj82}. In addition, an exact coarse-grained
90: description of the {\it dynamics} is possible in this case in
91: terms of a hydrodynamic equation for the particle density
92: $\rho(x,t)$ \cite{Kipn99,Gros03b}.
93:
94: In zero-range processes for which the hopping rates $w_n$ admit
95: condensation, one finds that above a critical density $\rho_c$ a
96: finite fraction of all particles in the system accumulate at a
97: randomly selected site, whereas all other sites have an average
98: density $\rho_c$ \cite{Evan00,Jeon00b,Gros03a}. The large scale
99: dynamics of condensation has been studied in terms of a coarsening
100: process \cite{Gros03a,Godr03}. The steady-state and the dynamics
101: of open systems which may admit condensation has not been
102: addressed so far.
103:
104: In the present work we consider a ZRP on an open chain with
105: arbitrary hopping rates and boundary parameters. Particles are
106: added and removed through the boundaries. In the interior of the
107: system hopping may either be symmetric or biased in one direction.
108: We calculate the exact steady-state distribution, when it exists,
109: and find it to be a product measure. In order to study
110: condensation phenomena in the open system we analyze in detail a
111: particular but generic ZRP which admits condensation. In this
112: model the hopping rates for large $n$ take the form $w_n=1+b/n$.
113: On a periodic lattice it is known that a condensation takes place
114: at high densities when $b>2$. We find that in an open system and
115: for a weak boundary drive the model evolves to a non-critical
116: steady-state. On the other hand, if the boundary drive is
117: sufficiently strong, the system may develop a condensate on one or
118: both of its boundary sites even for $b<2$. The number of particles
119: in the condensate grows linearly in time due to the strong
120: boundary drive. In this case the interior of the system may either
121: (a) reach a sub-critical steady-state, (b) reach a critical
122: steady-state, or (c) it may evolve such that the local particle
123: density exhibits a complex algebraic growth in time. Which
124: behavior is actually realized depends on the boundary rates, on
125: the parameter $b$, and on the asymmetry in the hopping rates.
126:
127: The paper is organized as follows. In Sec.~\ref{sec:def} we define
128: the ZRP with open boundaries, and give some examples of possible
129: hopping rates $w_n$. The exact steady-state distribution for
130: general boundary parameters and rates $w_n$ is derived in
131: Sec.~\ref{sec:st}. In Sec.~\ref{sec:finite} we consider the
132: behavior of finite systems for totally asymmetric, partially
133: asymmetric and symmetric hopping rates. The long-time temporal
134: behavior of local densities is obtained, and supporting numerical
135: simulations are presented. The long-time behavior of bulk
136: densities in an infinite system is exactly obtained using a
137: hydrodynamical approach in Sec.~\ref{sec:hydro}. Finally, we
138: present our summary and conclusions in Sec.~\ref{sec:summary}.
139:
140: \section{Zero-range processes with open boundaries}
141: \label{sec:def}
142:
143: The ZRP on an open 1d lattice with $L$ sites is defined as
144: follows. Each site $k$ may be occupied by an arbitrary number $n$
145: of particles. In the bulk a particle at site $k$ (say, the topmost
146: of $n$ particles) hops randomly (with exponentially distributed
147: waiting time) with rate $p w_n$ to the right and with rate $qw_n$
148: to the left. Without loss of generality, we take throughout this
149: paper $p \geq q$, so that particles in the bulk are driven to the
150: right. At the boundaries these rules are modified. At site $1$ a
151: particle is injected with rate $\alpha$, hops to the right with
152: rate $pw_n$, and is removed with rate $\gamma w_n$. At site $L$ a
153: particle is injected with rate $\delta$, hops to the left with
154: rate $qw_n$, and is removed with rate $\beta w_n$ (see
155: figure~\ref{fig:toy}).
156:
157: \begin{figure}
158: \centerline{\includegraphics[scale=0.45]{toy.eps}}
159: \caption{Graphic representation of the ZRP model with open
160: boundaries.} \label{fig:toy}
161: \end{figure}
162:
163:
164: Examples for such processes include the trivial case of
165: non-interacting particles ($w_n=n$) and the pure chipping process
166: ($w_n=1$) \cite{chip}. In the mapping to exclusion processes where
167: the particle occupation number becomes the interparticle distance,
168: the chipping model defined on a ring maps onto the simple
169: exclusion process which is integrable and can be solved by the
170: Bethe ansatz, combinatorial methods, recursion relations and
171: matrix product methods \cite{Derr98,Schu00}. A largely unexplored
172: but interesting integrable model with rates $w_n = (1-q^n)/(1-q)$
173: that interpolates between these two cases has been found in
174: \cite{Povo04}. The truncated chipping process with $w_n=1$ for $1
175: \leq n \leq K$ and $w_n = \infty$ for $n > K$ maps onto the
176: drop-push model \cite{Schu96} which is also integrable
177: \cite{Alim98}. For $w_1=w$ and $w_n = 1$ for $n\geq 2$ one obtains
178: the version of the non-integrable KLS model \cite{Katz84} that has
179: been introduced in \cite{Anta00} as a toy model for traffic flow
180: with a nonsymmetric current density relation. A parallel updating
181: version of this model was found to correspond to a broader class
182: of traffic models \cite{Levi04T} as well as an integrable family
183: of ZRP's \cite{Povo04b} which includes the asymmetric avalanche
184: process \cite{Prie01}.
185:
186: In a similar mapping to exclusion processes (but in a continuum
187: limit) the family of models with hopping rate \bel{rodmodel} w_n =
188: \left[1+ \frac{b'}{n}\right]^{-1} \ee describes the diffusion of
189: interacting rods on the real line \cite{Scho04}. The condensation
190: model \cite{Oloa98,Evan00,Kafr02,Jeon00b,Gros03a,Godr03} with
191: \bel{condmodel} w_n = 1+ \frac{b}{n} \ee is a generic model that
192: exhibits the condensation phenomenon described in the introduction
193: for $b>2$. Both models are non-integrable and hence alternative
194: tools must be employed for deriving information about their
195: dynamical behavior. We remind the reader that this model is
196: generic in the sense that it represents the complete family of
197: models with rates of the form $w_n = 1 + b/n + O(n^{-s})$ where
198: $s>1$. Other ZRP's which exhibit condensation are defined by the
199: rates $w_n = 1+b/n^\sigma$ with $0<\sigma<1$ and $b>0$
200: \cite{Kafr02} or by rates approaching zero \cite{Jeon00a}.
201:
202:
203: \section{Stationary distribution}
204: \label{sec:st}
205:
206: A state of the model at time $t$ may be defined through a
207: probability measure $P_\eta$ on the set of all configurations
208: $\eta=(\eta_1,\eta_2, \dots, \eta_L)$, $\eta_k\in\mathbb{N}$. Here
209: $\eta_k$ is the number of particles on site $k$. To calculate the
210: stationary distribution it is convenient to represent the
211: generator $H$ of this process in terms of the quantum Hamiltonian
212: formalism \cite{Schu00} where one assigns a basis vector
213: $|\eta\rangle$ of the vector space $(\mathbb{C}^\infty)^{\otimes
214: L}$ to each configuration and the probability vector is defined by
215: $|P\rangle=\sum_\eta P_\eta |\eta\rangle$. It is normalized such
216: that $\langle s|P\rangle=1$ where $\langle s|=\sum_\eta
217: \langle\eta|$ and $\bra{\eta}\eta'\,\rangle =
218: \delta_{\eta,\eta'}$. The time evolution described above is given
219: by the master equation
220: \begin{equation}
221: \label{master}
222: \frac{d}{dt} |P(t)\rangle = -H |P(t)\rangle
223: \end{equation}
224: through the ``quantum Hamiltonian'' $H$. This operator has off-diagonal
225: matrix elements $H_{\eta,\eta'}$ which are
226: the hopping rates between configurations $\eta,\eta'$
227: and complementary diagonal elements to preserve conservation of probability.
228:
229: Since we have only nearest neighbor exchange processes the
230: Hamiltonian in (\ref{master}) can be written as
231: \begin{equation}
232: \label{H}
233: H = h_1 + h_L +\sum_{k=1}^{L-1} h_{k,k+1},
234: \end{equation}
235: where $h_{k,k+1}$ acts nontrivially only on sites $k$ and $k+1$
236: (corresponding to hopping) while $h_1,h_L$ generates the boundary
237: processes specified above. For the ZRP we define the
238: infinite-dimensional particle creation and annihilation matrices
239: \bel{ladder} a^+ = \left(\begin{array}{ccccc}
240: 0 & 0 & 0 & 0 & \dots \\
241: 1 & 0 & 0 & 0 & \dots \\
242: 0 & 1 & 0 & 0 & \dots \\
243: 0 & 0 & 1 & 0 & \dots \\
244: \dots & \dots & \dots & \dots & \dots \\
245: \end{array}\right), \quad
246: a^- = \left(\begin{array}{ccccc}
247: 0 & w_1 & 0 & 0 & \dots \\
248: 0 & 0 & w_2 & 0 & \dots \\
249: 0 & 0 & 0 & w_3 & \dots \\
250: 0 & 0 & 0 & 0 & \dots \\
251: \dots & \dots & \dots & \dots & \dots \\
252: \end{array}\right)
253: \ee
254: as well as the diagonal matrix $d$ with elements
255: $d_{i,j} = w_i \delta_{i,j}$. With these matrices we have
256: \bel{hoppingpart}
257: - h_{k,k+1} = p (a^-_{k}a^+_{k+1} - d_k) + q (a^+_{k}a^-_{k+1} - d_{k+1})
258: \ee
259: and
260: \bel{boundarypart}
261: - h_1 = \alpha (a^+_1 - 1) + \gamma (a^-_1-d_1), \quad
262: - h_L = \delta (a^+_L - 1) + \beta (a^-_L-d_L).
263: \ee
264: The ``ground state'' of $H$ has eigenvalue 0. The corresponding right
265: eigenvector is the stationary distribution which we wish to calculate.
266:
267: Guided by the grand-canonical stationary distribution of the
268: periodic system we consider the grand-canonical single-site
269: particle distribution where the probability to find $n$ particles
270: on site $k$ is given by
271: \bel{marginal} P^\ast(\eta_k=n) =
272: \frac{z_k^n}{Z_k} \prod_{i=1}^n w_i^{-1}.
273: \ee
274: Here the empty
275: product $n=0$ is defined to be equal to 1 and $Z_k$ is the local
276: analogue of the grand-canonical partition function
277: \bel{Z} Z_k
278: \equiv Z(z_k) = \sum_{n=0}^\infty z_k^n \prod_{i=1}^n w_i^{-1}.
279: \ee
280: The corresponding probability vector $|P^\ast_k)$ with the
281: components $P^\ast(\eta_k=n)$ satisfies
282: \bel{coherent} a^+
283: |P^\ast_k) = z_k^{-1} d |P^\ast_k), \quad a^- |P^\ast_k) = z_k
284: |P^\ast_k). \ee
285: The proof of this property is by straightforward
286: calculation.
287:
288: As an ansatz for calculating the stationary distribution we take
289: the $L$-site product measure with the one-site marginals (\ref{marginal})
290: which is given by the tensor product
291: \bel{product}
292: \ket{P^\ast} = |P^\ast_1) \otimes |P^\ast_2) \otimes \dots \otimes |P^\ast_L)
293: \ee
294: and according to (\ref{coherent}) satisfies
295: \bea
296: - H \ket{P^\ast} & = & \left[\sum_{k=1}^{L-1} (pz_k-qz_{k+1})
297: (z_{k+1}^{-1} d_{k+1} - z_k^{-1} d_k) \right.\nonumber \\
298: & & + \left.(\alpha-\gamma z_1) (z_{1}^{-1} d_{1} - 1) + (\delta -
299: \beta z_L) (z_{L}^{-1} d_{L} - 1)\right]\ket{P^\ast}. \eea The
300: uncorrelated particle distribution (\ref{product}) is stationary
301: if and only if all terms on the right hand side of this equation
302: cancel. It is not difficult to see that this is satisfied for the
303: following stationarity conditions on the fugacities $z_k$ \be p
304: z_k - q z_{k+1} = \alpha - \gamma z_1 = \beta z_L -\delta \equiv
305: c\;. \ee The quantity $c$ is the stationary current.
306:
307:
308:
309: This recursion relation has the unique solution \bel{steadystate}
310: z_k = \frac{\left[(\alpha+\delta)(p-q) - \alpha\beta +
311: \gamma\delta\right] \left(\frac{p}{q}\right)^{k-1} - \gamma\delta
312: + \alpha\beta \left(\frac{p}{q}\right)^{L-1}}{\gamma(p-q-\beta) +
313: \beta(p-q+\gamma) \left(\frac{p}{q}\right)^{L-1}} \;,\ee and the
314: current is given by \bel{current} c = (p-q) \frac{- \gamma\delta +
315: \alpha\beta \left(\frac{p}{q}\right)^{L-1}}{\gamma(p-q-\beta) +
316: \beta(p-q+\gamma) \left(\frac{p}{q}\right)^{L-1}}\;. \ee Thus,
317: given the rates $w_n$ the stationary distribution is unique and
318: completely specified by the hopping asymmetry and the boundary
319: parameters. The stationary density profile follows from the
320: fugacity profile through the standard relation \be \rho_k = z_k
321: \frac{\partial}{\partial z_k} \ln{Z_k}\;, \ee where $Z_k = Z(z_k)$
322: is determined by the bulk hopping rules, as given in (\ref{Z}).
323:
324: We remark that up to corrections exponentially small in the system
325: size $L$, the bulk fugacity is a constant \bel{zeff}
326: z_{\mathrm{eff}} = \frac{\alpha}{p-q + \gamma}\;, \ee
327: that depends only on the
328: {\it left} boundary rates. This is in agreement with the observation
329: \cite{Gros03b} for special boundary parameters and can be
330: explained in terms of the more general theory of boundary-induced
331: phase transitions \cite{Kolo98}.
332: %The absence of these generic phase
333: %transitions which occur e.g. in the asymmetric exclusion process
334: %\cite{Schu93,Derr93}
335: %originates in the absence of a maximum in the current-density relation
336: %for the ZRP.
337:
338: Some special cases of (\ref{steadystate}) deserve mentioning.\\
339: (1) Setting the boundary extraction rates equal to the
340: corresponding bulk jump rates, i.e., $\beta = p, \;\gamma = q$,
341: and defining reservoir fugacities $z_{r,l}$ by $\alpha = p z_l,
342: \;\delta = q z_r$, the expression (\ref{steadystate}) reduces to
343: \be z_k = \frac{(z_l-z_r) \left(\frac{p}{q}\right)^{k-1} + z_r -
344: z_l \left(\frac{p}{q}\right)^{L-1}}{1-
345: \left(\frac{p}{q}\right)^{L-1}}\;. \ee This dynamics has a natural
346: interpretation as coupling the system to boundary reservoirs with
347: fugacities $z_r,z_l$ respectively. This case has been considered
348: in \cite{Gros03b} and earlier in \cite{Andj82,deMa84} for $p=q$.
349: The general solution (\ref{steadystate}) for arbitrary boundary
350: rates appears to be a new result for the ZRP.
351: \\
352: (2) For
353: \be
354: \gamma\delta = \alpha\beta \left(\frac{p}{q}\right)^{L-1}
355: \ee
356: the current vanishes and the system is in thermal equilibrium.
357: For the symmetric case $p=q$ this implies constant fugacities $z_k$.\\
358: (3) For symmetric hopping $p=q=1$ the fugacity profile is
359: generally linear \bel{symmetric} z_k =
360: \frac{\alpha+\delta+\alpha\beta(L-1) - (\alpha\beta-\gamma\delta)
361: (k-1)}{\beta+\gamma+\beta\gamma(L-1)}\;, \ee with a current \be c
362: =
363: \frac{\alpha\beta-\gamma\delta}{\beta+\gamma+\beta\gamma(L-1)}\;.
364: \ee Notice that the linear fugacity profile does not imply a
365: linear density profile except in
366: the very special case of non-interacting particles where $z_k = \rho_k$.\\
367: (4) In the totally asymmetric case, $q=0$, we find \bea
368: z_k & = & \frac{\alpha}{p+\gamma} \equiv z \mbox{ for } k\neq L\;,\\
369: z_L & = & \frac{(\alpha+\delta)p+\gamma\delta}{\beta(p+\gamma)}\;.
370: \eea
371: The current is given by $c=p z$.\\
372:
373: In the considerations above we have tacitly assumed that
374: the local partition function $Z_k$ exists for all $k$. Since for
375: suitable choices of boundary parameters the local fugacities at
376: the boundary may take arbitrary values this assumption implies an
377: infinite radius of convergence of $Z_k = Z(z_k)$ which is not the
378: case in the models described above. This raises the question
379: of the long-time behaviour of the ZRP for strong boundary drive,
380: i.e., rates which drive the boundary fugacities out the radius
381: of convergence of $Z$. This is studied in detail in the following
382: sections for the condensation model (\ref{condmodel}).
383: The rod model (\ref{rodmodel}) can be regarded as a generic ZRP
384: without bulk condensation transition, but finite radius of
385: convergence for $Z$. Where appropriate we compare its behaviour
386: with that of the condensation model.
387:
388:
389: \section{Condensation model --- Steady state and dynamics near the boundary}
390: \label{sec:finite}
391:
392: In this section we consider in some detail the long-time behavior
393: of the condensation model (\ref{condmodel}). For this model the
394: local fugacity $z_k$ has to satisfy $z_k\leq1$ in the steady
395: state. On a ring geometry, the model exhibits a condensation for
396: $b>2$ at high density. We first analyze the totally asymmetric
397: case, where particles can only hop to the right. The cases of
398: partial asymmetry and symmetric hopping are then treated.
399:
400: \subsection{Totally asymmetric hopping}
401:
402: For the totally asymmetric case we take $q=\gamma=\delta=0$. For
403: the normalization of time we set $p=1$. The exact steady-state
404: solution (\ref{steadystate}) yields \bea
405: \label{tas1} z_k & = & \alpha \equiv z \mbox{ for } k\neq L\;,\\
406: \label{tas2} z_L & = & \frac{\alpha}{\beta} \;, \eea and the
407: current is given by $c=z$. Since for this model $z$ has to satisfy
408: $z\leq1$, the steady state (\ref{tas1}, \ref{tas2}) is valid only
409: for $\alpha \leq 1$ and $\beta \geq \alpha$. In this case the
410: single-site steady-state distribution is given, for large $n$, by
411: $P^*(\eta_k=n) \sim z^n/n^b$.
412:
413: We now proceed to discuss the dynamical behavior of the model in
414: the case that stationary state does not exist. For $\alpha>1$ the
415: following picture emerges.
416:
417: \vspace{12pt}\noindent {\bf Site 1:}\\[4mm]
418: On site~1 particles are deposited randomly with rate $\alpha > 1$
419: and are removed by hopping to site 2 with rate $1+b/n_1$. Hence
420: the occupation number performs a simple biased random walk on the
421: set $n_1$ of positive integers with drift $\alpha - 1 - b/n_1$
422: which is positive for any $n_1 > b/(\alpha - 1)$. Such a random
423: walk is non-recurrent and reaches asymptotically the mean velocity
424: $v=\alpha - 1$. Hence the mean particle number $N_1(t) =
425: \exval{n_1(t)}$ on site~1 grows asymptotically linearly
426: \bel{site1} N_1(t) \sim (\alpha - 1) t. \ee
427:
428: \vspace{12pt}\noindent {\bf Boundary sites $k>1$:}\\[4mm]
429: We extend the random walk picture (which is strictly valid for site~1)
430: to site~2. On site~2 particles are injected (by hopping from site
431: 1) with rate $1 + b/n_1$ and are removed with rate $1 + b/n_2$.
432: Since $n_1$ increases in time on average the input rate approaches
433: 1 and the occupation number at site 2 performs a biased random
434: walk with hopping rate 1 to the right and rate $1 + b/n_2$ to the
435: left. Whether this random walk is recurrent depends on $b$. The
436: asymptotic behavior has been analyzed in \cite{Levi04}. We merely
437: quote the result: \bel{site2} N_2(t) \sim \left\{ \ba{ll}
438: t^{1/2} & b < 1 \\
439: t^{1/2}/\ln{t} & b = 1 \\
440: t^{1-b/2} & 1 < b < 2 \\
441: \ln{t} & b = 2 \\
442: \rho^\ast & b > 2 \ea \right. \ee
443: The constant
444: \bel{critdens} \rho^\ast = \frac{1}{b-2} \ee
445: is the critical
446: density of the condensation model \cite{Oloa98}. It is approached
447: with a power law correction $t^{1-b/2}$. By applying this random
448: walk picture to further neighboring sites, and assuming scaling,
449: it has been shown that neighboring boundary sites behave
450: asymptotically in the same fashion \cite{Levi04}. Similar analysis
451: shows that the square-root increase of the particle density occurs
452: also for the model (\ref{rodmodel}) for all values of its interaction
453: parameter $b'$.
454:
455: \begin{figure}
456: \centerline{\includegraphics[height=5.5cm]{site14.eps}}
457: \caption{Temporal evolution of local densities, as obtained from
458: numerical simulations of the totally asymmetric condensation
459: model, with $b=3/2$, and $\alpha=2$. The solid line corresponds to
460: the expected growth law $t^{1/4}$, and the dotted line corresponds
461: to linear growth.} \label{fig:simulation1}
462: \end{figure}
463:
464: To test the validity of the random walk picture to sites beyond
465: $k=2$, we carried out numerical simulations of a totally
466: asymmetric model with $L=5$. We first note that since the hopping
467: is totally asymmetric the time evolution of the system up to site
468: $k$ is independent of what happens at sites to the right of $k$.
469: In particular, the dynamics on all sites $k<L$ is independent of
470: $\beta$. Hence, in order to study boundary layers it is sufficient
471: to simulate very small systems of only a few sites. In
472: figure~\ref{fig:simulation1} we present the long-time behavior of
473: the occupation of sites $k=1 - 4$ for $\alpha=2$ and $b=3/2$. It
474: is readily seen that while site $1$ grows linearly in time, sites
475: $2-4$ grow with the expected power law $t^{1/4}$.
476:
477: \vspace{12pt}\noindent {\bf Bulk sites $k\gg 1$:}\\[4mm]
478: %For $1 < b < 2$ the boundary exponent (\ref{site2})
479: %derived above differs from the
480: %hydrodynamic bulk exponent $\kappa$ (\ref{kappa}).
481: The picture of the simple random walk becomes increasingly
482: inaccurate as one enters into the bulk of the system, since
483: injection events onto a site become increasingly correlated in
484: time. This violates the random walk assumption and makes the
485: previous analysis invalid in this case. The temporal behavior of
486: bulk sites $k \gg 1$ can be treated exactly by using a
487: hydrodynamic approach, which yields the behavior of bulk sites in
488: the long-time limit. This analysis, carried out in
489: Sec.~\ref{sec:hydro}, shows a different dynamical behavior in the
490: bulk. Notice, however, that for any {\it finite} system all ``bulk
491: sites'' have finite distance from the boundary and hence behave
492: asymptotically like the boundary sites.
493:
494: \vspace{12pt}\noindent {\bf Site L:}\\[4mm]
495: Since the motion of particles on site $k$ is independent of the
496: motion on sites to their right we expect the bulk result to be
497: asymptotically valid on all sites up to site $L-1$, i.e., there is
498: no right boundary layer with yet another set of growth exponents.
499: On site $L$ the following picture emerges. There is an asymptotic
500: incoming flux $c=1$ and an exit rate $\beta (1+b/n_L)$. For
501: $\beta=1$ the boundary site behaves like a bulk site and we obtain
502: the bulk growth exponent. For $\beta < 1$ the outgoing flux does
503: not compensate the incoming flux which yields asymptotically
504: linear growth $\rho_L(t) = (1-\beta) t$. For $\beta > 1$ a finite
505: stationary chemical potential $z_L = 1/\beta$ is approached.
506:
507: \begin{figure}
508: \centerline{\includegraphics[height=5.5cm]{site5.eps}}
509: \caption{Temporal evolution of local density at the rightmost
510: site, as obtained from numerical simulations of the totally
511: asymmetric condensation model, with $b=3/2$, $\alpha=2$ and
512: $\beta=1/2, 1, 2$. The solid line corresponds to the expected
513: growth law $t^{1/4}$, and the dotted line corresponds to linear
514: growth.} \label{fig:simulation2}
515: \end{figure}
516:
517:
518: In figure~\ref{fig:simulation2} we present simulation results
519: for the long-time dynamics of
520: the occupation of site $k=L=5$. Depending on the value of $\beta$
521: the occupation number either grows linearly ($\beta<1$), grows
522: with the same power law as the bulk ($\beta=1$), or approaches a
523: finite density ($\beta>1$), as expected from the above discussion.
524:
525: To complete the discussion of the totally asymmetric case
526: we remark that for a subcritical left boundary $\alpha < 1$ but
527: supercritical $\beta < \alpha$ the bulk of the
528: system becomes stationary with fugacity $z_{bulk}=\alpha$. On site
529: $L$ a condensate develops with linearly increasing particle density
530: $\rho_L(t) \sim (\alpha-\beta) t$.
531:
532: \subsection{Partially asymmetric hopping}
533:
534: We now analyze the dynamical behavior of the partially asymmetric
535: model in the case where no stationary state exists. We start by
536: considering the case where the rates at both boundaries are such
537: that the occupation of the two boundary sites increase linearly
538: with time. This takes place for $\alpha-\gamma < p-q <
539: \beta-\delta$. Here sites $k=1$ and $k=L$ act as reservoirs for
540: the rest of the system. The effective rates at which the
541: reservoirs exchange particles with the system are
542: $\alpha_{\mathrm{eff}}=\beta_{\mathrm{eff}}=p$ and
543: $\gamma_{\mathrm{eff}}=\delta_{\mathrm{eff}}=q$. The
544: fugacity~(\ref{steadystate}) at sites $k=2,\ldots, L-1$ thus
545: becomes $z_k=1$. The asymptotic temporal behavior~(\ref{site2})
546: holds also in this case.
547:
548: \begin{figure}
549: \centerline{\includegraphics[scale=0.4]{zkasym.eps}} \caption{The
550: fugacity profile (\ref{steadystate}) for the partially asymmetric
551: case, with $b=3/2$, $p=3/4$ and $q=1/4$. Solid line corresponds to
552: a left condensate ($\alpha=3/4, \beta=1, \gamma=1/4, \delta=1/5$)
553: and dashed line corresponds to a right condensate ($\alpha=1/2,
554: \beta=3/4, \gamma=1/4, \delta=1/4$).} \label{fig:zk}
555: \end{figure}
556:
557:
558: The picture changes qualitatively if only one boundary fugacity
559: does not exist. Suppose first that this happens at site 1. In this
560: case the density on site $1$ will increase linearly with time, as
561: in the previous totally asymmetric case. This happens when
562: $\alpha-\gamma > p-q$ and $\beta-\delta>p-q$. As before, site $1$
563: acts as a reservoir for the rest of the system, with effective
564: rates $\alpha_{\mathrm{eff}}=p,\; \gamma_{\mathrm{eff}}=q$. Sites
565: $k=2,\ldots, L-1$ are thus stationary, with the fugacity given by
566: (\ref{steadystate}). The fugacity at sites away from the right
567: boundary approaches $1$, with a deviation exponentially small in
568: the system size (see figure~\ref{fig:zk}). Hence, starting from an
569: empty initial state one expects algebraic growth of the local
570: density until, after a long crossover time, stationarity is
571: reached.
572:
573: If, on the other hand, the right boundary rates drive site $L$ out
574: of equilibrium, then the density on site $L$ increases linearly in
575: time. This happens when $\alpha-\gamma < p-q$ and $\beta-\delta
576: <p-q$. Site $L$ acts effectively as a boundary reservoir with
577: $\beta_{\mathrm{eff}}=p$ and $\delta_{\mathrm{eff}}=q$. As in the
578: preceding case the system becomes stationary, but with a finite
579: (independent of system size) deviation of the bulk fugacity from
580: the critical value $z=1$. Both in the bulk and at the left
581: boundary the density approaches the finite value dictated by the
582: left boundary fugacity (see figure~\ref{fig:zk}).
583:
584: Finally, for $ \beta-\delta < p-q < \alpha-\gamma$ both boundary
585: sites have finite fugacity, and the system reaches a non-critical
586: steady state, as given by (\ref{steadystate}).
587:
588: \subsection{Symmetric hopping}
589:
590: The analysis of symmetric hopping ($p=q=1$) follows very closely
591: that of the partially asymmetric case, except that here the
592: fugacity profile is linear rather than exponential. In particular,
593: if only one boundary is driven out of equilibrium, a condensate
594: appears at that boundary with a particle density which increases
595: linearly with time. This site, say site $1$, acts as a reservoir
596: with $\alpha_{\mathrm{eff}} = \gamma_{\mathrm{eff}}=1$. The
597: fugacity profile decreases linearly from $1-O(1/L)$ at the left
598: boundary to $\delta/\beta+O(1/L)$ at the right boundary, as given
599: by~(\ref{symmetric}).
600:
601: When both boundary sites are supercritical they act as reservoirs
602: with effective rates $\alpha_{\mathrm{eff}} =
603: \beta_{\mathrm{eff}}=\gamma_{\mathrm{eff}} =
604: \delta_{\mathrm{eff}}=1$, yielding $z_k=1$ throughout the system.
605: This is also the case relevant for studying the fluid-mediated
606: interaction of probe particles in two-species systems
607: \cite{Levi04}. We expect similar boundary growth laws as in the
608: asymmetric case. This is because the incoming flux from the
609: direction of the bulk is compensated by the outgoing flux at the
610: boundary and one has an effective random picture with the same
611: rates as above. However, corrections to scaling are expected to be
612: larger as current correlations are now stronger due to hopping
613: contributions from the bulk. Moreover, as opposed to the
614: asymmetric case, in a semi-infinite system both boundaries have
615: the same growth exponents.
616:
617:
618: \section{Hydrodynamics --- Exact analysis of dynamics in the bulk}
619: \label{sec:hydro}
620:
621:
622: In order to analyze the time evolution of sites far away from the
623: boundaries we consider the hydrodynamic limit of the ZRP model.
624: The coarse-grained time evolution of the density profile starting
625: from a non-stationary initial profile can be determined, by
626: adapting standard arguments \cite{Kipn99,Spoh91}, from the
627: continuum limit of the lattice continuity equation. This equation
628: reads \bel{continuity} \frac{d}{dt}\rho_k = c_{k-1} - c_{k}\;, \ee
629: with the local current \bel{bulkcurrent} c_{k} = p z_{k} - q
630: z_{k+1}\;, \ee and $z$ expressed in terms of $\rho$. Together with
631: an appropriate choice of constant boundary fugacities the solution
632: is uniquely determined \cite{Gros03b}.
633:
634: For the driven system one obtains under Eulerian scaling (lattice
635: constant $a\to 0$, $t \to t/a$, system length fixed) \bel{Euler}
636: \frac{\partial}{\partial t}\rho(x,t) = - (p-q)
637: \frac{\partial}{\partial x} z(x,t) + a
638: (p+q)/2\frac{\partial^2}{\partial x^2} z(x,t) \ee where the
639: infinitesimal viscosity term serves as regularization and selects
640: the physical solution of the otherwise ill-posed initial value
641: problem with fixed boundary fugacities. It takes care of a proper
642: description of discontinuities which may arise in the form of
643: shocks or a boundary discontinuity. Indeed, in the large-time
644: limit the density approaches a constant given by the left boundary
645: fugacity, with a jump discontinuity at the right boundary
646: \cite{Gros03b}. This is in agreement with the exact result derived
647: in the previous subsection. We stress that (\ref{Euler}) provides
648: an exact description of the density evolution under Eulerian
649: scaling. It is not a continuum approximation involving a mean
650: field or other assumption. The simple form of (\ref{Euler})
651: originates in local stationarity. The absence of noise is due to
652: Eulerian scaling, i.e., the effects of noise appear on finer
653: scales. For a recent rigorous discussion of the hydrodynamic limit
654: of stochastic particle systems, see \cite{Kipn99,Saad04}.
655:
656:
657: Consider a semi-infinite system which is initially empty and has
658: constant density at the left boundary. This boundary condition
659: induces a rarefaction wave entering the bulk which can be
660: constructed using the method of characteristics. The speed $v_0$
661: of the wave front is given by the zero-density characteristic of
662: (\ref{Euler}) which is the average speed $v_0=(p-q)w_1$ of a
663: single particle. In light of the results of the previous sections
664: we are particularly interested in the case where the left boundary
665: fugacity is equal to 1. For $b<2$ in the condensation model [and
666: for any $b'$ in the rod model (\ref{rodmodel})] this corresponds
667: to an infinite boundary density. On the other hand, for $b>2$ the
668: corresponding boundary density is $\rho_c$~(\ref{critdens}).
669: Therefore we are searching for a scaling solution in terms of the
670: scaling variable $u=x/(v_0t)$ such that $\rho(u) = 0$ for $u \geq
671: 1$ while $\rho=\rho_c$ or $\rho=\infty$ at the left boundary. On
672: physical grounds the solution has to be continuous as no shock
673: discontinuities can develop for the initial state (empty lattice)
674: under consideration. Under these conditions (\ref{Euler}) can be
675: integrated straightforwardly by setting $a=0$ and one finds the
676: implicit representation \bel{solutionz} \frac{d \, z}{d\,\rho} =
677: uw_1\;. \ee This is the solution within the interval $v_1t \leq x
678: \leq v_0t$. Here \bel{coll} v_z = (p-q) \frac{dz}{d\rho} \ee is
679: the collective velocity of the lattice gas which plays the role of
680: the speed of the characteristics for the hydrodynamic equation
681: (\ref{Euler}). Outside this interval one has $z=0$ for $x\geq
682: v_0t$ and $z=1$ for $0\leq x\leq v_1t$. We now analyze this
683: solution in terms of $\rho$.
684:
685:
686: For the model (\ref{rodmodel}) one has $Z=1/(1-z)^{b'+1}$ which
687: yields the fugacity-density relation $z=\rho/(b'+1+\rho)$. Hence
688: \be \rho(u) = (b'+1)\left(\sqrt{\frac{1}{u}} - 1\right). \ee with
689: $v_0=(p-q)/(b'+1)$ and $v_1=0$ for all $b'$. At any fixed $x$ the
690: bulk density increases algebraically \be \rho_{bulk}(t) \sim
691: t^{1/2} \ee for any $b'>0$.
692:
693: For the condensation model the local partition is the
694: hypergeometric function \be Z = {}_2F_1 (1,1;1+b;z) \ee which does
695: not admit an explicit representation of $z$ as a function of
696: $\rho$. However, as the fugacity is an increasing function in time
697: approaching 1 we may analyze its asymptotic behavior by expanding
698: the hypergeometric function around $z=1$ \cite{Gros03a}.
699:
700: \vspace{12pt}\noindent {\bf $b<2$:}\\[4mm]
701: Here the left boundary fugacity $z=1$ corresponds to infinite left
702: boundary density. Moreover $v_1=0$, therefore the solution
703: (\ref{solutionz}) describes the density profile in the interval $0
704: \leq u \leq 1$. This
705: enables us to consider fixed $x$ and study the long time limit.
706: In order to analyze the small $u$ behavior (i.e.
707: $z$ close to 1 and $\rho$ large) one has to distinguish two
708: domains \cite{Gros03a}. For $b<1$ one has $\rho(z) \propto
709: z/(1-z)$ for large $\rho$ while for $1<b<2$ one finds $\rho(z)
710: \propto z/(1-z)^{2-b}$. As a function of $u$ we make the ansatz
711: $\rho \propto 1/u^\kappa$ for the large $t$ (i.e. small $u$)
712: asymptotics. Inserting this into the differential equation
713: (\ref{Euler}) yields a consistent solution with \be \rho_{bulk}(t)
714: \sim t^\kappa \ee only for \bel{kappa}
715: \kappa = \left\{ \ba{ll} \frac{1}{2} & \mbox{ for } b < 1 \\
716: \frac{2-b}{3-b} & \mbox{ for } 1 < b < 2 \ea \right. . \ee We
717: conclude that the bulk density increases algebraically with the
718: universal diffusive exponent $1/2$ for $b<1$ and with a
719: non-universal $b$-dependent exponent in the range $1 < b < 2$ of
720: the condensation model. At $b=1,2$ there are logarithmic
721: corrections which we do not discuss further.
722:
723: For symmetric hopping one describes the density dynamics under
724: diffusive scaling $t \to t/a^2$ and obtains
725: \bel{diffusion}
726: \frac{\partial}{\partial t}\rho(x,t) =
727: p \frac{\partial^2}{\partial x^2} z(x,t)
728: \ee
729: which needs no further regularization. Repeating the previous
730: analysis one finds the same bulk growth exponents as for the
731: asymmetric hopping model.
732:
733: \vspace{12pt}\noindent {\bf $b>2$:}\\[4mm]
734: In the condensation regime the boundary density corresponding to
735: $z=1$ is $\rho_c$~(\ref{critdens}). Again two different regimes
736: are found from the asymptotic analysis of the hypergeometric
737: function. For $b<3$ one has $v_1=0$ and repeating the same
738: analysis as for $1<b<2$ shows that at fixed $x$ the density
739: approaches $\rho_c$ with a power law correction with the same
740: exponent $(2-b)/(3-b)$ as before. This is analogous to the
741: behavior in the boundary sites analysed in the previous section,
742: but the exponent is different. For $b>3$ one finds for the
743: collective velocity \cite{Gros03a} \be v_1 = (p-q)
744: \frac{(b-3)^2(b-2)^2}{(b-1)^2} > 0. \ee Hence an analysis of the
745: long-time behaviour at fixed $x$ is not meaningful. A domain with
746: constant $\rho=\rho_c$ spreads into the system, with a front speed
747: $v=v_1$. This front is preceded by the rarefaction wave
748: (\ref{solutionz}) for $v_1 t < x < v_0 t$. This behaviour as a
749: function of $b$ is unexpected as usually changes in the
750: rarefaction wave of this type are caused by changing the boundary
751: density rather than an interaction parameter of the driven system.
752:
753: We stress that there are two questions that cannot be answered by
754: the hydrodynamic analysis given above. First we apply Eulerian or
755: diffusive scaling respectively. This leaves generally open what
756: happens in any semi-infinite lattice system at finite lattice
757: distance from the boundaries or in a finite system. Any deviation
758: from the results given above which decays on lattice scale as one
759: approaches the bulk cannot be detected within the hydrodynamic
760: description. Boundary layers, which have been analyzed in the
761: previous section and found to have an interesting microscopic
762: structure, would under scaling at best appear as a structureless
763: boundary discontinuity.
764:
765: Secondly, the radius of convergence of $Z$ is 1 and forcing the
766: boundary fugacities $z_1$ or $z_L$ to be larger than 1 implies a
767: breakdown of the assumption of local stationarity underlying the
768: hydrodynamic description, at least in the vicinity of the
769: boundaries. In particular, the hydrodynamic approach is not
770: applicable for analyzing the condensation regime $b>2$ if the
771: boundary density exceeds the critical density of the bulk.
772: However, this regime may be analyzed by the random-walk approach
773: discussed in the previous section.
774:
775:
776: \section{Conclusions}
777: \label{sec:summary}
778:
779: In this paper we studied the dynamical behavior of the zero-range
780: process with open boundary conditions for arbitrary bulk and
781: boundary rates. It is found that for a weak boundary drive the
782: model reaches a steady state. The exact steady-state distribution
783: is calculated and is shown to be a product measure characterized
784: by site-dependent fugacities. In the case of strong drive the
785: system does not reach a steady-state, and its evolution in time is
786: calculated. To this end we considered the condensation model with
787: an initially empty lattice and studied how the local density
788: evolves in time. As long as the bulk dynamics does not permit
789: condensation ($b<2$) the growth of the local density is algebraic
790: in time with exponents that we determined using a random walk
791: picture for the boundary region, and standard hydrodynamic
792: description in the bulk. From this analysis we are led to the
793: conclusion that in the condensation model with $b>2$ (where
794: condensation in a periodic system sets in above the critical bulk
795: density $\rho_c=1/(b-2)$) only the boundary sites develop into a
796: condensate, with a density increasing linearly in time. All bulk
797: sites become stationary with finite local fugacities determined by
798: the boundary rates. Somewhat surprisingly the driven and the
799: symmetric model have the same bulk and boundary growth exponents.
800: The boundary condensate appears also for $b<2$ when no
801: condensation transition exists in a periodic system. It is a
802: general feature of zero-range processes in which $Z$ has a finite
803: radius of convergence.
804:
805: For $b>1$ we observe a precursor to the condensation transition in
806: the sense that the universal diffusive growth for $b<1$ breaks
807: down. Bulk and boundary growth exponents become different and both
808: decrease with $b$, i.e. they become non-universal. It is
809: interesting to note that $b=1$ plays a special role also for the
810: stationary state on a ring geometry. For $b<1$ the stationary
811: probability to find any given site empty vanishes as the density
812: is increased to infinity \cite{Gros03a}, in agreement with
813: intuition. However, for $b>1$ every site has a finite probability
814: of being empty, even if the particle density is infinite. Applied
815: to present scenario this implies the counterintuitive result that
816: even at very late time, when the average particle density in an
817: open system tends to infinity on {\it each site}, one still
818: expects to find a finite fraction of empty sites at any given
819: moment.
820:
821: In this context it is also instructive to study the mean first
822: passage time (MFPT) of the boundary particle density, i.e. the
823: mean time $\tau$ after which a particle number $N$ has been
824: reached for the first time at a given site, starting from an empty
825: site. Using the exact general MFPT expression \cite{Murt89} for
826: the effective random walk defined above one finds \be \tau =
827: \frac{1}{b-1} \left[ \left(\ba{c} N+b \\ b+1 \ea \right) -
828: \left(\ba{c} N+1 \\ 2 \ea \right) \right] \ee where for
829: non-integer $b$ the factorials are defined by the
830: $\Gamma$-function. For large $N$ this quantity has the asymptotic
831: behavior \be \tau \sim \left\{ \ba{ll}
832: N^2 & (b<1) \\
833: N^2\ln{N} & (b=1) \\
834: N^{1+b} & (b>1) \ea \right. . \ee Again there is a transition
835: at $b=1$, with diffusive behavior for $b<1$ and sub-diffusive
836: exploration of the state space for $b>1$. In the bulk a simple
837: random picture for the on-site density dynamics is not valid and a
838: prediction for the bulk MFPT is not possible. For small finite
839: system size one expects boundary behavior everywhere, but with
840: increasing corrections to scaling as one moves away from the
841: boundary. The precise nature of the crossover from the boundary
842: growth exponent to the hydrodynamic bulk growth exponent remains
843: an open problem.
844:
845: \ack G.M.S. thanks the Weizmann Institute for kind hospitality.
846: The support of the Albert Einstein Minerva Center for Theoretical
847: Physics, Israel Science Foundation, and Deutsche
848: Forschungsgemeinschaft (grant Schu827/4), is gratefully
849: acknowledged.
850:
851: \section*{References}
852: \begin{thebibliography}{99}
853:
854: \bibitem{Spit70} F. Spitzer,
855: %\newblock Interaction of Markov Processes
856: \newblock Adv. Math. {\bf 5} 246 (1970).
857:
858: \bibitem{Evan96} M.R. Evans,
859: %Bose-Einstein condensation in disordered exclusion processes and relation
860: %to traffic flow.
861: Europhys. Lett. {\bf 36}, 13-18 (1996).
862:
863: \bibitem{Krug96} J. Krug and P.A. Ferrari,
864: Phase transitions in driven diffusive systems with random rates.
865: J. Phys. A: Math. Gen. {\bf 29}, L465-L471 (1996).
866:
867: \bibitem{Oloa98} O.J. O'Loan, M.R. Evans, and M.E. Cates,
868: Phys. Rev. E {\bf 58} 1404 (1998).
869:
870: \bibitem{Evan00} M.R. Evans,
871: %\newblock Phase transitions in one-dimensional nonequilibrium systems
872: \newblock Braz. J. Phys. {\bf 30} 42 (2000).
873:
874: \bibitem{Jeon00a} I. Jeon and P. March, Condensation transition
875: for zero-range invariant measures. Can. Math. Soc. Conf. Proc {\bf 26},
876: 233-244 (2000).
877:
878: %\bibitem{Muka00} D. Mukamel
879: %%\newblock Phase transitions in nonequilibrium systems
880: %\newblock in: {\it Soft and Fragile Matter: Nonequilibrium dynamics,
881: %metastability and flow} eds M E Cates and M R Evans
882: %\newblock (Bristol, Institute of Physics Publishing, 2000)
883:
884: \bibitem{Schu03} G.~M.~Sch\"utz,
885: %Critical Phenomena and universal dynamics in one-dimensional
886: %driven diffusive systems with two species of particles.
887: J.~Phys.~A {\bf 36} R339 (2003)
888:
889: \bibitem{Shim04}
890: G.M. Shim, B.Y. Park, J.D. Noh, and H. Lee,
891: %Analytic study of the three-urn model for separation of sand.
892: Phys. Rev. E {\bf 70} 031305 (2004).
893:
894: \bibitem{Burd01}
895: Z. Burda, D. Johnston, J. Jurkiewicz, M. Kami\'nski, M.A. Nowak, G. Papp,
896: and I. Zahed,
897: %Wealth condensation in Pareto macroeconomies.
898: Phys. Rev. E {\bf 65} 026102 (2002).
899:
900: \bibitem{Froh75}
901: H. Fr\"ohlich, Evidence for Bose condensation-like excitation of
902: coherent modes in biological systems.
903: Phys. Lett. A {\bf 51}, 21-22 (1975).
904:
905: \bibitem{traffic}
906: D. Chowdhury, L. Santen, and A. Schadschneider, Phys. Rep. {\bf 329},
907: 199 (2000); D. Helbing, Rev. Mod. Phys. {\bf 73}, 1067 (2001).
908:
909: \bibitem{Bian01}
910: G. Bianconi and A.-L. Barab\'asi,
911: %Bose-Einstein condensation in complex networks
912: Phys. Rev. Lett. {\bf 86}, 5632 (2001).
913:
914: \bibitem{Doro03}
915: S.N. Dorogovtsev, J.F.F Mendes, and A. Samukhin, Nucl. Phys. B {\bf
916: 666}, 396 (2003).
917:
918:
919: \bibitem{Kafr02} Y. Kafri, E. Levine, D. Mukamel, G.M. Sch\"{u}tz, and
920: J. T\"or\"ok, Phys. Rev. Lett. {\bf 89}, 035702 (2002)
921:
922: \bibitem{Andj82}
923: E.D. Andjel, Ann. Probab. {\bf 10}, 525 (1982).
924:
925: \bibitem{Kipn99}
926: C. Kipnis and C. Landim, {\it Scaling limits of interacting
927: particle systems} (Springer, Berlin, 1999).
928:
929: \bibitem{Gros03b}
930: S. Grosskinsky and H. Spohn,
931: Bull. Braz. Math. Soc. New Series 34, 489 (2003).
932:
933: \bibitem{Jeon00b} I. Jeon and P. March,
934: %Size of the largest cluster under zero-range invariant measures.
935: Ann. Probab. {\bf 28}, 1162-1194 (2000).
936:
937: \bibitem{Gros03a}
938: S. Grosskinsky, G.M. Sch\"utz, and H. Spohn,
939: J. Stat. Phys. {\bf 113}, 389 (2003)
940:
941: \bibitem{Godr03}
942: C. Godr\`eche, J. Phys. A: Math. Gen. {\bf 36}, 6313-6328 (2003).
943:
944: \bibitem{chip}
945: S.N. Majumdar, S. Krishnamurthy, and M. Barma, Phys. Rev. Lett.
946: {\bf 81}, 3891 (1998).
947:
948: \bibitem{Derr98}
949: B. Derrida, Phys. Rep. {\bf 301}, 65 (1998)
950:
951: \bibitem{Schu00} G.M. Sch\"utz
952: %{\it Exactly solvable models for many-body systems far from equilibrium},
953: in {\it Phase
954: Transitions and Critical Phenomena} vol 19, ed. C. Domb and J.
955: Lebowitz (London: Academic, 2001)
956:
957:
958: \bibitem{Povo04}
959: A.M. Povolotsky, Phys. Rev. E {\bf 69}, 061109 (2004).
960:
961: \bibitem{Schu96}
962: G.M. Sch\"utz, R. Ramaswamy, and M. Barma,
963: J. Phys. A: Math. Gen. {\bf 29}, 837 (1996).
964:
965: \bibitem{Alim98}
966: M. Alimohammadi, V. Karimipour, and M. Khorrami,
967: Phys. Rev. E {\bf 57}, 6370 (1998).
968:
969: \bibitem{Katz84} S. Katz, J. L. Lebowitz, and H. Spohn,
970: J. Stat. Phys. {\bf 34},
971: 497 (1984).
972:
973: \bibitem{Anta00}
974: T. Antal and G.M. Sch\"utz,
975: Phys. Rev. E {\bf 62}, 83 (2000).
976:
977: \bibitem{Levi04T}
978: E. Levine, G. Ziv, L. Gray and D. Mukamel, Physica A {\bf 340},
979: 636 (2004).
980:
981: \bibitem{Povo04b}
982: A.M. Povolotsky and J.F.F. Mendes, cond-mat/0411558.
983:
984: \bibitem{Prie01}
985: V.B. Priezzhev, E.V. Ivashkevich, A.M. Povolotsky, and C.K. Hu,
986: Phys. Rev. Lett. {\bf 87}, 084301 (2001).
987:
988: \bibitem{Scho04}
989: G. Sch\"onherr, cond-mat/0409618,
990: to appear in J. Phys. A: Math. Gen.
991:
992: \bibitem{Kolo98}
993: A.B. Kolomeisky, G.M. Sch\"{u}tz, E.B. Kolomeisky and
994: J.P. Straley,
995: J. Phys. A: Math. Gen. {\bf 31} 6911 (1998).
996:
997: %\bibitem{Schu93}
998: %G. Sch\"utz and E. Domany, J. Stat.
999: %Phys. {\bf 72}, 277 (1993).
1000:
1001: %\bibitem{Derr93}
1002: %B. Derrida, V. Hakim, M.R. Evans, and V. Pasquier J. Phys. A: Math. Gen.
1003: %{\bf 26}, 1493 (1993).
1004:
1005: \bibitem{deMa84}
1006: A. De Masi and P. Ferrari, J. Stat. Phys. {\bf 36}, 81 (1984).
1007:
1008: \bibitem{Levi04} E. Levine, D. Mukamel, and G. M. Sch\"utz, preprint {\tt cond-mat/0412130}.
1009:
1010: \bibitem{Spoh91}
1011: H. Spohn, {\it Large Scale dynamics of interacting particle systems},
1012: (Springer, Berlin, 1991)
1013:
1014: \bibitem{Saad04}
1015: E. Saada, to appear in Oberwolfach Reports.
1016:
1017: \bibitem{Murt89} K.P.N.~Murthy and K.W.~Kehr,
1018: Phys. Rev. A {\bf 40}, 2082 (1989);
1019: {\bf 41}, 1160 (1990) (Erratum).
1020:
1021:
1022: \end{thebibliography}
1023:
1024: \end{document}
1025: