1: \documentclass[pre,preprint,amsmath,amssymb,floatfix]{revtex4}
2:
3: \usepackage{latexsym,amssymb,amsmath}
4: \usepackage{graphics}
5: \usepackage{graphicx}
6: \usepackage{epsfig}
7: \begin{document}
8:
9: \title{Disorder and non-conservation in a driven diffusive system}
10:
11: \author{M. R. Evans$^1$, T. Hanney$^1$ and Y. Kafri$^2$\\
12: $^1$ School of Physics, University of Edinburgh,
13: Mayfield Road, Edinburgh, EH9 3JZ, United Kingdom\\
14: $^2$ Department
15: of Physics, Harvard University, Cambridge, MA 02138}
16:
17:
18: \begin{abstract}
19: We consider a disordered asymmetric exclusion process in which
20: randomly chosen sites do not conserve particle number. The model
21: is motivated by features of many interacting molecular motors such
22: as RNA polymerases. We solve the steady state exactly in the two
23: limits of infinite and vanishing non-conserving rates. The first
24: limit is used as an approximation to large but finite rates and
25: allows the study of Griffiths singularities in a nonequilibrium
26: steady state despite the absence of any transition in the pure
27: model. The disorder is also shown to induce a stretched
28: exponential decay of system density with stretching exponent
29: $\phi= 2/5$.
30: \end{abstract}
31:
32: \maketitle
33:
34: \section{Introduction} \label{intro}
35:
36: Driven diffusive systems serve as simple models for collective
37: phenomena ranging from traffic flow to molecular motors. Moreover,
38: they provide tractable examples of systems far from thermal
39: equilibrium. Studies of one-dimensional driven diffusive systems have
40: shown that many interesting phenomena, which are typically not
41: observed in one dimensional systems in thermal equilibrium,
42: exist. Prominent examples are boundary induced phase transitions and
43: spontaneous symmetry breaking, for reviews see
44: e.g. \cite{Mukamel,Brazil,Schutz}.
45:
46: Most studies have considered systems in which the dynamics are the
47: same everywhere in the system or systems where the dynamics are
48: modified only at the boundaries. However, when trying to relate these
49: systems to many interacting molecular motors, the effects of
50: non-conservation and disorder (i.e. spatial heterogeneity in the
51: dynamics) can not be ignored in many cases.
52:
53: Indeed there have been some studies on the effects of disorder on
54: driven diffusive systems. For example, the effect of assigning a
55: disordered quenched rate to each particle was studied in
56: \cite{KF,E,BJ,JB,TB,K} on a ring geometry. Exact solutions show that
57: at high enough densities a macroscopic number of particles jam behind
58: the slowest particle in the system. The phase transition between the
59: jammed and non-jammed phase is similar to a Bose-Einstein
60: condensation. Work has also been done on an asymmetric exclusion
61: process on a ring where the quenched hopping rates between neighboring
62: sites are drawn at random \cite{TB,K,KP}. For molecular motors moving
63: along a disordered substrate this seems to be the relevant scenario
64: \cite{Harms,Kafri}. It was argued, based on numerics and mean-field
65: solutions, that at high densities the system phase separates into a
66: region of high density coexisting with a low density region. Finally,
67: the combined effect of random hopping rates and open boundary
68: conditions was considered in \cite{ED,HS}. In \cite{ED} it was argued
69: using numerics that the location of phase transition lines may be
70: sample dependent. In \cite{HS} mean-field arguments and numerics
71: indicate the existence of shifts in phase boundaries which, by analogy
72: with equilibrium systems, are expected to be accompanied by emergent
73: Griffiths regions. A review of the effects of disorder in exclusion
74: models has been given in \cite{S}.
75:
76: In this paper we consider another type of disorder. We study an
77: asymmetric exclusion process (ASEP) where {\em non-conserving} sites
78: are chosen at random along the lattice. At these sites particles may
79: attach and detach with specified rates which may also be drawn at
80: random. Thus there are two components to the disorder. A feature of
81: this disorder is that it allows a detailed account of the way in which
82: Griffiths singularities can arise in nonequilibrium steady states. In
83: equilibrium the mechanism leading to Griffiths singularities is well
84: understood: the disorder, e.g. dilution, breaks the system into pure
85: regions and large pure regions may give rise to the exponentially
86: suppressed Griffiths singularities. In the present the case, in the
87: limit of high attachment and detachment rates, the non-conserving
88: sites break the system into driven conserving domains.
89:
90: Non-conservation of particles in driven systems without disorder has
91: previously been considered in the context of molecular motors. The
92: idea is that molecular motors move in a preferred direction along a
93: filament and are able to attach and detach from the filament. In the
94: works so far all sites are non-conserving
95: \cite{PFF,EJS,LKN,PRWKS}. The motivation for the model we study here
96: comes from the fact that some molecular motors only attach and detach
97: at certain sites.
98:
99: More specifically, we give a very simplified description of
100: many interacting RNA polymerase (RNAp) motors acting on a prokaryotic
101: DNA in vitro. Prokaryotic RNA polymerase can initiate without
102: regulatory proteins. Namely, RNAp left in a solution with DNA can
103: produce RNA even if the specific protein which regulates its action
104: (for example, by enhancing or reducing the initiation rates) is
105: present. They can enter and leave the DNA in order to transcribe RNA
106: molecules at specific sites, referred to as promotor and termination
107: sites respectively. In the language of the lattice model we consider a
108: binding of a RNAp to a promotor corresponds to a particle entering the
109: system. The unbinding at the termination site corresponds to a
110: particle leaving the system. In the absence of regulatory proteins the
111: rate of entering the DNA depends on the details of the promotor
112: sites. In such systems the RNA polymerase do not usually move from one
113: gene to another. In the lattice model this would correspond to
114: particles not moving from one stretch of conserving sites to a
115: neighboring one i.e. the limit in which the detachment rates are large
116: at the non-conserving sites. We comment that in principle RNA
117: polymerase may move in different directions along the DNA when
118: transcribing different messenger RNAs, corresponding to particles
119: moving in different directions along different conserving
120: stretches. However, in the limit when the detachment rate is large
121: this will not influence most of the results described in the paper. Of
122: course, our assumption of randomly distributed lengths of genes (or
123: conserving segments) is not expected to hold. However, the model
124: provides a starting point for analyzing more realistic situations.
125:
126: The paper is organised as follows. In Section \ref{model}, we define
127: the model and discuss two limits which are exactly soluble. In
128: Section \ref{GS} we show that the disorder induces Griffiths like
129: singularities as the rates for entering and leaving the lattice are
130: changed. More significantly, it is shown in Section \ref{dynamics}
131: that the presence of the non-conserving sites leads to anomalous
132: relaxation of the system toward the steady state. Specifically, we
133: argue that decay of measurable quantities decay as a function of time,
134: $t$, as a stretched exponential $\exp(-ct^\phi)$, where $c$ is a
135: non-universal constant and $\phi=2/5$. The results are verified
136: numerically. We conclude in Section \ref{CONC}.
137:
138: \section{Model} \label{model}
139:
140: The model we consider is a disordered generalisation of the ASEP. The
141: pure ASEP is defined on a one-dimensional lattice containing $L$ sites
142: and with periodic boundaries. The lattice is occupied by particles
143: subject to an exclusion interaction, which prohibits multiple
144: occupancy of any site. These particles hop with rate one to the
145: nearest neighbour site to the right, provided it is empty, and so the
146: total particle number $N$ is conserved. We introduce non-conservation
147: into this model by allowing, at certain sites (which we will call
148: `disorder' sites), processes which do not conserve the total particle
149: number $N$. Hence each site $l$ ($l=1,\ldots,L$) in the pure model
150: remains a pure site with probability $p$, or becomes a disorder site
151: with probability $(1-p)$. Now, at the disorder sites, labelled by
152: $j=1,\ldots,P$, particles attach with rate $c_j$ or detach with rate
153: $a_j$. In general, we wish to consider heterogeneous rates for the
154: non-conserving processes.
155:
156: To study the model we first consider limits which can be solved
157: exactly. Later, using numerics, we argue that the results are
158: generic. Exact solubility arises when the steady state densities at
159: the disorder sites are determined solely by the attachment and
160: detachment processes. In these cases, since the system is composed of
161: conserving domains of the chain in contact with disorder sites at the
162: boundaries of each domain, the steady state can be written as a
163: product of boundary driven ASEPs in which the densities of the
164: boundary reservoirs are given by the disorder site densities $\rho_j$.
165:
166: Before turning to the disordered ASEP under consideration we recap
167: some facts, which will be useful later, about the boundary driven
168: ASEP. For the boundary driven ASEP, in which particles are injected at
169: the left-hand boundary site with rate $\alpha$ (provided it is vacant)
170: and removed from the right-hand boundary site with rate $\beta$, exact
171: steady state weights for particle configurations can be obtained using
172: a matrix product ansatz \cite{DEHP}. In this ansatz, particle
173: configurations are represented as a product of matrices $X_1 \cdots
174: X_L$ where $X_l = D$ ($E$) if site $l$ is occupied (vacant). The
175: steady state weight of a configuration is given by $\langle \alpha
176: \vert X_1 \cdots X_L \vert \beta \rangle$ provided the matrices $D$
177: and $E$ and the vectors $\langle \alpha \vert$ and $\vert \beta
178: \rangle$ satisfy the relations
179: \begin{equation}
180: DE=D+E \equiv C\;, \quad \alpha \langle \alpha \vert E =
181: \langle \alpha \vert \;\quad {\rm and} \;\quad \beta D \vert \beta \rangle =
182: \vert \beta \rangle\;.
183: \end{equation}
184: From these relations exact expressions for the normalisation $\langle
185: \alpha \vert C^L \vert \beta \rangle$ can be derived which show that
186: the model undergoes a second order phase transition: when both
187: $\alpha$ and $\beta \geq 1/2$ the system is in a maximum current
188: phase, otherwise it is in one of two low current phases. The phase
189: transition between the low current phases is first order. The rates
190: $\alpha$ and $1-\beta$ represent the densities of particles in
191: reservoirs connected to the boundary sites.
192:
193: Next, we use the known results for the boundary driven ASEP to
194: study the disordered case. As stated above, there are limits where
195: the steady state weight of the disordered model factorizes into a
196: product over boundary driven ASEP weights. The two
197: exactly soluble limits are:
198:
199: \begin{itemize}
200: \item $c_j$, $a_j \to \infty$, with $c_j/a_j$ fixed.
201:
202: In this limit, each disorder site $j$ acquires a density $\rho_j$ determined
203: solely by $c_j$ and $a_j$ which
204: obeys the equation of motion,
205: \begin{equation}
206: \frac{\partial \rho_j}{\partial t} = c_j (1-\rho_j) + a_j \rho_j\;.
207: \end{equation}
208: Therefore in the steady state,
209: \begin{equation}
210: \rho_j = \frac{c_j}{c_j+a_j}\;.
211: \label{rho}
212: \end{equation}
213: If we define $n_j$ to be the number of sites between disorder
214: sites $j$ and $j+1$ (i.e. the length of the $j$-th conserving domain),
215: then the normalisation $Z_L(\{n_j\})$ (which is the sum over the
216: steady state weights of all particle configurations on sites excluding
217: the disorder sites), for a given configuration of the disorder sites
218: $\{n_j\}=n_1,\ldots,n_P$, factorises into a product over
219: normalisations for the boundary driven ASEP:
220: \begin{equation} \label{infty}
221: Z_L(\{n_j\}) = \prod_{j=1}^P \langle \rho_j \vert C^{n_j} \vert 1-\rho_{j+1}
222: \rangle
223: \end{equation}
224: and where $\rho_{P+1} = \rho_1$.
225: This is the relevant limit for the model molecular motors discussed
226: in the introduction.
227:
228:
229:
230: \item $c_j$, $a_j \to 0$, with $c_j/a_j$ fixed.
231:
232: In this limit the time between each attachment/detachment event
233: tends to infinity. Therefore, after each event the system reaches
234: a homogeneous steady state of the pure ASEP with periodic
235: boundaries. Thus the system density $\rho = N/L$, satisfies the
236: equation of motion
237: \begin{equation}
238: \frac{\partial \rho}{\partial t} = \left[ \sum_{j=1}^P c_j \right]
239: (1-\rho) +\left[ \sum_{j=1}^P a_j \right] \rho\;.
240: \end{equation}
241: Therefore in the steady state the system density is given by
242: \begin{equation}
243: \rho = \frac{\sum_{j=1}^P c_j}{\sum_{j=1}^P (c_j+a_j)}\;.
244: \end{equation}
245: Because the steady state is homogeneous, all sites, including
246: disorder sites, have the same steady state density $\rho$.
247: Moreover, the steady factorises and there are no correlations
248: between sites. Therefore, one can still write the normalisation
249: in a form similar to (\ref{infty}):
250: \begin{equation} \label{zero}
251: Z_L(\{n_j\}) = \prod_{j=1}^P \langle \rho \vert C^{n_j} \vert 1-\rho
252: \rangle\;.
253: \end{equation}
254: This is because in this case $D$ and $E$ are given by
255: the scalars
256: $1/(1-\rho)$ and $1/\rho$.
257: \end{itemize}
258:
259: In the following we will consider the first limit, $c_j$, $a_j \to
260: \infty$. We use the factorised form (\ref{infty}) with $\rho_j$ given
261: by (\ref{rho}) as an approximation for the case where $c_j$, $a_j$ are
262: large but finite which is relevant for the model of molecular motors.
263: This approximation has a mean-field character, in the sense that
264: correlations are factorised about the disorder sites, however all
265: correlations within conserving domains are retained.
266:
267: \section{Griffiths singularities} \label{GS}
268:
269: We can exploit known properties of the normalisation of the boundary
270: driven ASEP to demonstrate the existence of Griffiths-type
271: singularities in the disordered ASEP \cite{G,M}. As an illustrative
272: example, we consider binary disorder at disorder sites, such that
273: \begin{displaymath}
274: \rho_j = \frac{c_j}{a_j+c_j} = \left\{ \begin{array}{ll} u & \textrm{
275: with probability $q$,} \\ v & \textrm{with probability $1-q$.}
276: \end{array} \right.
277: \end{displaymath}
278: This is the simplest choice of disorder for which Griffiths
279: singularities occur. Generalizations to more complicated
280: situations are straightforward.
281:
282: Using (\ref{infty}) the steady state normalization satisfies
283: \begin{equation}
284: {\rm ln}Z_L = \sum_{j=1}^P W_{n_j} (\rho_j, 1\!-\!\rho_{j+1}) \;,
285: \label{lnZ}
286: \end{equation}
287: where $W_n (\rho_j, 1-\rho_{j+1}) = {\rm ln} \langle \rho_j \vert C^{n_j}
288: \vert 1-\rho_{j+1} \rangle$. In order to perform the disorder average, we
289: write (\ref{lnZ}) in the form
290: \begin{eqnarray}
291: {\rm ln}Z_L = \sum_{n=0}^\infty & \left[ \nu_{u,1-v}(n) \, W_n
292: (u,1\!-\!v) + \nu_{u,1-u}(n) \, W_n (u,1\!-\!u) \right. \nonumber \\
293: & \left. + \nu_{v,1-v}(n) \, W_n (v,1\!-\!v) + \nu_{v,1-u}(n) \, W_n
294: (v,1\!-\!u) \right]
295: \end{eqnarray}
296: where $\nu_{\alpha,\beta}(n)$ is the number of conserving domains of
297: size $n$ bounded by disorder sites at densities $\alpha$ and $1-\beta$.
298:
299: We can average over the configurations of the $\nu_{\alpha,\beta}(n)$
300: by calculating the expectation values $\langle
301: \nu_{\alpha,\beta}(n) \rangle$ in the thermodynamic limit (the angled
302: brackets denote a disorder average). This is
303: achieved by observing that $\lim_{L \to \infty}
304: L^{-1}\langle \nu_{\alpha,\beta}(n) \rangle$ is just the probability that
305: a site is part of an $n$-site conserving domain bounded by disorder
306: sites with densities $\alpha$ and $1-\beta$, hence
307: \begin{eqnarray} \label{w}
308: \lim_{L \to \infty} L^{-1} \langle {\rm ln} Z_L \rangle = (1\!-\!p)^2
309: \sum_{n=0}^\infty p^n &\left\{ q^2
310: W_n(u,1\!-\!u) + q(1\!-\!q)
311: [W_n(u,1\!-\!v) \right.\nonumber \\ &\left. + W_n(v,1\!-\!u)] + (1\!-\!q)^2
312: W_n(v,1\!-\!v) \right\}
313: \end{eqnarray}
314:
315: The form of equation (\ref{w}) is typical of systems which exhibit
316: Griffiths singularities. In equilibrium, these singularities are
317: usually inferred from the properties of the Yang-Lee zeros of the
318: partition function --- we can use the known properties of the Yang-Lee
319: zeros of the the analogous quantity, the normalisation \cite{BE}, to
320: show how Griffiths singularities arise in the disordered
321: nonequilibrium model: For fixed $u \geq 1/2$ say, in the complex
322: $v$-plane and for $n$ arbitrarily large, the zeros of $\langle u \vert
323: C^n \vert 1-v \rangle$ accumulate arbitrarily close to the point
324: $v=1/2$ on the real axis. Therefore there exists a singularity in
325: $W_n(u,1-v)$ arbitrarily close to the point $v=1/2$ which is
326: exponentially suppressed (by a factor $p^n$). Thus, such a
327: Griffiths-type singularity follows whenever $u$ and $v$ are such that
328: at least one of the $W_n(\alpha,\beta)$ in (\ref{w}) lies on the phase
329: boundary of the ASEP i.e. whenever $u$ and/or $v=1/2$.
330:
331: One can go further and consider disorder in the $c_j$'s and $a_j$'s
332: explicitly. For instance, if both $c_j$ and $a_j$ are drawn from
333: binary distributions, then the densities at the disorder sites can
334: assume one of four possible values, each with a different probability
335: in general. However Griffiths singularities still arise whenever any
336: one of these values for the density is 1/2, as before.
337:
338: It is also straightforward to use standard arguments from the study of
339: dilute systems to show that the correlation length remains finite at
340: the Griffiths singularity, as is the case in equilibrium systems. Note
341: that Griffiths singularities emerge in this non-equilibrium system
342: despite the absence of a transition in the pure model. This is in
343: sharp contrast to equilibrium systems.
344:
345:
346: \section{Dynamics: stretched exponential decay of the density to its
347: steady state value} \label{dynamics}
348:
349: In the pure boundary driven ASEP, whenever the system is in or at the
350: boundary of the maximum current phase, the system density decays with
351: time to its stationary value as an exponential with a decay constant
352: that depends on system size $L$ as $L^{z}$, where $z=3/2$ is the
353: dynamic exponent\cite{BW,NAS}. In the low current phase when the
354: boundary injection and ejection rates are equal, a shock exists in the
355: steady state, and the dynamic exponent $z=2$. Otherwise, in the low
356: current phases the relaxation time is finite \cite{BW} and does not
357: depend on $L$
358: %(except from certain non-generic initial conditions such
359: %as the empty lattice).
360: Hence, in the disordered model, whenever
361: contributions to the normalisation (\ref{w}) are in the maximum
362: current phase, the decay of the system density will be determined by
363: the relaxation of these conserving domains.
364:
365: For the following analysis, it is sufficient to consider disorder only
366: in the location of the disorder sites: we consider homogeneous
367: attachment and detachment rates i.e. $c_j=c$ and $a_j=a$. In the case
368: where $c=a$, (\ref{rho}) gives $\rho = 1/2$ at which point conserving
369: domains are on the boundary of the maximum current phase; otherwise
370: the conserving domains are in the low current phases. Thus only when
371: $c=a$ do we expect the decay constant associated with a conserving
372: domain to depend on its size\cite{NAS}.
373:
374: Therefore for $c=a$, in a conserving domain of length $n$, we assume
375: that the particle density $\rho_n(t)$ decays to its steady state value
376: as
377: \begin{equation}
378: \delta\rho_n(t) \equiv \rho_n(t) - 1/2 \sim e^{-\Delta_n t},
379: \end{equation}
380: where $\Delta_n = \Delta_0 n^{-3/2}$.
381: In the disordered case, we need to sum
382: over configurations of the disorder sites. This is achieved in the
383: same way as in the previous section so, in the thermodynamic limit, the
384: decay of the system density $\rho(t)$ becomes
385: \begin{equation}
386: \delta\rho(t) \sim \sum_{n=0}^{\infty} p^n e^{-\Delta_n t}\;,
387: \end{equation}
388: where we retain only the $n$-dependence, as is sufficient to determine
389: the dominant contribution to the form of the relaxation. If we
390: convert the sum into an integral and consider late times so that the
391: integral can be evaluated at the saddle point, we obtain
392: \begin{equation} \label{decay}
393: \delta\rho(t) \sim {\rm exp} (-c t^\phi),
394: \end{equation}
395: where $c$ is a constant and $\phi = (1+z)^{-1} = 2/5$.
396:
397: Equation (\ref{decay}) predicts the decay of the density up to some
398: prefactor, power-law in $t$, with an exponent peculiar to the decay of
399: the density. The stretching exponent
400: $\phi$ should be universal however, in the sense that other
401: correlation functions
402: e.g. the current, should reach their
403: stationary values with the same stretched exponential decay.
404: This result should be valid for more general types of disorder
405: whenever one has conserving domains in the maximal current phase.
406:
407: In figures \ref{fig:decay10} and \ref{fig:decay1} we show the results
408: of simulations. The simulations were run on systems of 10000 sites
409: with periodic boundary conditions and averaged over 1000 histories of
410: the dynamics, starting from an empty lattice and with the same
411: realisation of disorder. The decay of the averaged system density
412: $\delta \rho(t)$, is shown in Figure \ref{fig:decay10} for $a_j = c_j
413: = 10$, and in Figure \ref{fig:decay1} for $a_j = c_j = 1$. In both
414: cases the straight line $t^{2/5}$ is given for reference. Figure
415: \ref{fig:decay10} shows very good agreement with the predicted
416: stretched exponential decay, and even in figure \ref{fig:decay1},
417: where $c_j$ and $a_j$ are not large, the agreement is still quite
418: good. Thus it appears that our result for the stretching exponent
419: holds for finite rates, although its derivation is only exact in the
420: limit of infinite attachment and detachment rates.
421: \begin{figure}[!ht]
422: \begin{minipage}[c]{0.48 \linewidth}
423: %\vspace{-10mm}
424: \includegraphics[width=0.95\textwidth]{dnR10.eps}
425: \caption{Log-log plot of the decay of the density
426: with time for $a_j = c_j = 10$; the initial condition was an empty lattice.}
427: \label{fig:decay10}
428: \end{minipage}
429: \hfill
430: \begin{minipage}[c]{0.48 \linewidth}
431: \includegraphics[width=0.95\textwidth]{dnR1.eps}
432: \caption{Log-log plot of the decay of the density
433: with time for $a_j = c_j = 1$; the initial condition was an empty lattice.}
434: \label{fig:decay1}
435: \end{minipage}
436: \end{figure}
437:
438: \section{Conclusion} \label{CONC}
439: In this work we have studied an ASEP with disorder sites where
440: particles are not conserved. This may provide a basis for a more
441: realistic model for interacting molecular motors such as RNAp. We
442: have used an approximation, exact in the limit of infinite
443: non-conserving rates, the underlying assumption behind which is
444: that the system factorises into conserving domains. According to
445: the ratios of the attachment and detachment rates these domains
446: assume different phases of the ASEP with open boundaries.
447:
448: Within this approximation we can explicitly identify Griffiths
449: singularities. These arise when there are large conserving domains, on
450: the boundary of the maximal current phase. An interesting feature is
451: the prediction of a Griffiths singularity despite the absence of a
452: transition in the pure system.
453:
454: More generally one might ask under what conditions do Griffiths
455: singularities arise in nonequilibrium steady states. In equilibrium
456: systems Griffiths singularities are understood in terms of Yang-Lee
457: zeros of the partition function. In nonequilibrium systems one does
458: not have an energy function, nevertheless one can often identify a
459: quantity that plays the role of a partition function, for example the
460: normalisation (\ref{infty}), and recently there has been progress in
461: understanding the zeros of such quantities \cite{BE}.
462:
463: When there is a spectrum of maximal-current conserving-domain sizes,
464: we have demonstrated that correlation functions undergo a stretched
465: exponential decay with a stretching exponent predicted to be
466: $\phi=2/5$. Moreover simulations suggest this result holds
467: for a wide range of attachment and detachment rates.
468: A related stretched exponential decay has already been
469: observed for the decay of autocorrelations in a bond diluted symmetric
470: exclusion process on a ring \cite{GS} (in this case $z=2$ so
471: $\phi=1/3$).
472:
473: It would be instructive to develop further the approximation that the
474: steady state factorises about the disorder sites. As we saw in
475: Section \ref{model} this approximation is exact in two limits, and we gave
476: expressions for the densities at the disorder sites. It would be
477: interesting to develop a scheme that interpolates between these two
478: limits. Also of interest would be a better understanding of the
479: correlations between the conserving domains which may exist away from
480: the two exact limits and their effect on Griffiths singularities.
481:
482: {\bf Acknowledgments} The authors thank the Max Planck Institute for
483: Complex Systems, Dresden, where this work was initiated, for
484: hospitality. Work by Y.K was supported by the National Science
485: Foundation through grant DMR-0229243 and grant DMR-0231631 and the
486: Harvard Materials Research Laboratory via Grant DMR-0213805. MRE and
487: TH were supported by EPSRC programme grant GR/S10377/01.
488:
489: \begin{thebibliography}{00}
490:
491: \bibitem{Mukamel} D. Mukamel, in: {\it Soft and Fragile Matter:
492: Nonequilibrium dynamics, metastability and flow}, eds. M.E. Cates and
493: M.R. Evans (Bristol, Institute op Physics Publishing) (2000)
494:
495: \bibitem{Brazil}
496: M.R. Evans, Braz. J. Phys. {\bf 30} 42 (2000)
497:
498: \bibitem{Schutz} G.M. Sch{\" u}tz, in: {\it Phase Transitions and
499: Critical Phenomena} Vol. 19, eds. C. Domb and J. Lebowitz (London,
500: Academic Press) (2001)
501:
502: \bibitem{KF}
503: J. Krug and P.A. Ferrari, J. Phys. A {\bf 29} L465 (1996)
504:
505: \bibitem{E}
506: M.R. Evans, Europhys. Lett. {\bf 36} 13 (1996)
507:
508: \bibitem{BJ}
509: M. Barma and K. Jain, Pramana-J. Phys. {\bf 58} 409 (2002)
510:
511: \bibitem{JB}
512: K. Jain and M. Barma, Phys. Rev. Lett. {\bf 91} 135701 (2003)
513:
514: \bibitem{TB}
515: G. Tripathy and M. Barma, Phys. Rev. E {\bf 58} 1911 (1998)
516:
517: \bibitem{K}
518: J. Krug, Braz. J. Phys. {\bf 30} 97 (2000)
519:
520: \bibitem{KP}
521: K.M. Kolwankar and A. Punnoose, Phys. Rev. E {\bf 61} 2453 (2000)
522:
523: \bibitem{Harms}
524: T. Harms and R. Lipowsky, Phys. Rev. Lett. {\bf 79} 2895 (1997)
525:
526: \bibitem{Kafri}
527: Y. Kafri, D.K. Lubensky and D.R. Nelson, Biophys. J. in press.
528:
529: \bibitem{ED}
530: C. Enaud and B. Derrida, cond-mat/0402450
531:
532: \bibitem{HS}
533: R.J. Harris and R.B. Stinchcombe, cond-mat/0403062
534:
535: \bibitem{S}
536: R.B. Stinchcombe, J. Phys.: Condens. Matter {\bf 14} 1473 (2002)
537:
538: \bibitem{PFF}
539: A. Parmeggiani, T. Franosch and E. Frey, Phys. Rev. Lett. {\bf 90}
540: 086601 (2003)
541:
542: \bibitem{EJS}
543: M.R. Evans, R. Juh\'asz and L. Santen, Phys. Rev. E {\bf 68} 026117 (2003)
544:
545: \bibitem{LKN}
546: R. Lipowsky, S. Klumpp and T.M. Nieuwenhuizen, Phys. Rev. Lett. {\bf 87}
547: 108101 (2001); S. Klumpp and R. Lipowsky, J. Stat. Phys. {\bf 113} 233 (2003)
548:
549: \bibitem{PRWKS}
550: V. Popkov, A. Rakos, R. D. Willmann, A. B. Kolomeisky, G. M. Sch{\" u}tz,
551: Phys. Rev. E {\bf 67} 066117 (2003)
552:
553: \bibitem{DEHP}
554: B. Derrida, M.R. Evans, V. Hakim and V. Pasquier,
555: J.~Phys.~A {\bf 26} 1493 (1993)
556:
557: \bibitem{G}
558: R.B. Griffiths, Phys. Rev. Lett. {\bf 23} 17 (1969)
559:
560: \bibitem{M}
561: M. Wortis, Phys. Rev. B {\bf 10} 4665 (1974)
562:
563: \bibitem{BE}
564: R.A. Blythe and M.R. Evans, Braz. J. Phys. {\bf 33} 464
565: (2003); Phys. Rev. Lett. {\bf 89} 080601 (2002)
566:
567: \bibitem{BW}
568: U. Bilstein and B. Wehefritz, J. Phys. A {\bf 30} 4925 (1997)
569:
570: \bibitem{NAS}
571: Z. Nagy, C. Appert and L. Santen, J. Stat. Phys. {\bf 109} 623 (2002)
572:
573: \bibitem{GS}
574: M.D. Grynberg and R.B. Stinchcombe, Phys. Rev. E {\bf 61} 324 (2000)
575:
576: \end{thebibliography}
577:
578: \end{document}
579: