cond-mat0206205/cbd.tex
1: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
3: %\renewcommand{\rmdefault}{ptm}
4: \usepackage{graphicx,epsf}
5: \begin{document}
6: 
7: 
8: \title{An analytic model for a cooperative ballistic deposition in one dimension}
9: \author{M. Kamrul Hassan$^{\dagger,\ddagger}$, Niels Wessel$^\dagger$, and J\"urgen Kurths$^\dagger$}
10: \affiliation{
11: $^\dagger$University of Potsdam, Department of Physics, Postfach 601553, D-14415 Potsdam, Germany \\
12: $^\ddagger$ University of Dhaka, Department of Physics, Theoretical Physics Division, Dhaka 1000, Bangladesh
13: }
14: %\email{khassan@agnld.uni-potsdam.de}
15: \date{\today}
16: 
17: 
18: 
19: 
20: \begin{abstract}%
21: 
22: We formulate a model for a cooperative ballistic deposition (CBD) process whereby the incoming particles are 
23: correlated with the ones already adsorbed via attractive force. The strength of the correlation is controlled by a 
24: tunable parameter $a$ that interpolates the classical car parking problem at $a=0$, the ballistic deposition at $a=1$ and
25: the CBD model at $a>1$. The effects of the correlation in the CBD model are as follows. 
26: The jamming coverage $q(a)$ increases with the strength of attraction $a$ due to an ever increasing tendency of cluster 
27: formation. The system almost reaches the closest packing structure as $a\rightarrow\infty$ but never forms a 
28: percolating cluster which is typical to 1D system.  
29: In the large $a$ regime, the mean cluster size $k$ increases as $a^{1/2}$. Furthermore, 
30: the asymptotic approach towards the closest packing is purely algebraic both with $a$ as $q(\infty)-q(a) \sim a^{-1/2}$
31: and with $k$ as $q(\infty)-q(k) \sim k^{-1}$ where $q(\infty)\simeq 1$.
32: \end{abstract}
33: 
34: \pacs{05.20.Dd,02.50.-r,68.43.-h}
35: 
36:  \maketitle
37: 
38: \section{Introduction}
39: 
40: The kinetics of a monolayer growth by the deposition of macromolecules and colloidal particles onto solid substrates 
41: has been the subject of extensive research for the past years (see \cite{kn.evans,kn.tarjus,kn.schaaf} for extensive
42: review). The reason is well justified because its importance 
43: and significance cover many seemingly unrelated topic in physics, chemistry, biology and other branches of 
44: science and technology. From a theoretical point of view, the random sequential adsorption (RSA) of a 
45: monodisperse particle is one of the simplest model that can describe deposition phenomena \cite{kn.renyi}. 
46: In this process, particles are deposited randomly, one at each time step, with the strict restriction that overlapping is forbidden. This can be described by the 
47: following algorithm. (i) At each time step, a random position is chosen from the whole substrate and is assigned to the 
48: center of the particle picked for deposition. (ii) If the incoming particle collides with a previously adsorbed one, 
49: the trial attempt
50: is rejected; otherwise it is adsorbed irreversibly. (iii) In either case, the time is increased by one unit and the steps
51: (i) - (ii) are repeated until the system reaches a state when particles can no longer be adsorbed.
52: 
53: One of the virtue of the RSA model is that like many statistical physics problems, it is exactly solvable in one dimension 
54: in both its continuum and lattice version for some specific cases. The one dimensional continuum version of the model is 
55: popularly known as the {\it random car parking} (RCP) problem and has attracted much attention.
56: Despite the inherent simplicity in the RSA model, it still captures essential generic features of the process and 
57: has proved to describe successfully the behavior of many experimental systems, namely, the adsorption of proteins, 
58: latex and colloidal particles \cite{kn.feder,kn.onoda,kn.mann}. Nevertheless, there have been continuous research 
59: efforts to include various important physical features to make it more realistic and thus covering a wider range of 
60: real life situations \cite{kn.micha,kn.pagonabarraga,kn.rodgers}.
61: Along this road, a good deal of progress has already been achieved and yet we are far away from a complete theory.
62: In recent years it has received extra momentum and the number of papers published in the recent years is a clear 
63: testimony to it \cite{kn.schaaf,kn.hassan,kn.bonnier,kn.pastor,kn.lee,kn.penrose}. 
64: 
65: 
66: 
67: The strongest criticism of the RSA model is its outright rejection of particles that fall upon an already adsorbed one. 
68: The most successful model overcoming this criticism is known as the ballistic deposition (BD) 
69: process \cite{kn.talbot,kn.jullien}. The BD model is best explained in terms of the deposition of disks of 
70: diameter $m$ on a line. In this case, whenever an incoming disk overlaps with an already adsorbed one, it is allowed 
71: to roll over the latter disk following the path of the steepest descent. In doing so, the disk can either touch the 
72: adsorbing plane (global minimum) or it may 
73: find itself trapped in the local minimum formed by two or more connected disks.
74: In the former case, the disk is irreversibly attached with the one it rolled over leaving no gaps
75: in between, while in the latter case the trial attempt is rejected.
76: Both in the simple RSA and the BD model, only a short range hard-core repulsion via the excluded volume 
77: effects is taken into account. All forms of  
78: long-range interactions between the particles in the adsorbed and adsorbing phases are ignored.
79: There are some fragmented attempts though, 
80: to include some specific forms of interactions such as the electrostatic, dipolar and the hydrodynamic interactions
81: \cite{kn.electrostatic,kn.dipolar,kn.hydrodynamic}. 
82:  
83: 
84: 
85: In this article, we consider a model that includes the 
86: attractive force between the particles in the adsorbing and the adsorbed phase mimicking the long range interaction. 
87: To study the underlying mechanisms 
88: in such  complex phenomena, like the deposition processes, it is of great advantage to have 
89: a flexible model which allows explicit analytical solutions with different possible mechanisms.
90: In order to increase the flexibility of the model,
91: we further generalise it by introducing a parameter $a$ that can tune the strength of 
92: the attractive forces. This would certainly facilitate the study of the general effect of long-range interactions in
93: the whole process. We can recover the simple RSA results by setting $a=0$, the BD results by setting $a=1$ and while  
94: $a>1$, we show that the model describes the cooperative ballistic deposition (CBD). Obviously, the strength of correlation
95: is determined by the strength of the attractive force. This results in an increased overlapping with the already adsorbed 
96: particles. However, once a particle collides with an already adsorbed one, it follows the rule of the simple BD. 
97: Thus, as the strength of interaction increases, we expect an increasing rate of successful adsorptions.
98: The analytical part of the model we study in this work is similar to the one previously studied by Viot {\it et al} 
99: \cite{kn.viot}. However, we go far beyond that not only by offering a completely new model but also by giving
100: an in-depth analysis and an exact physical explanation as well as extracting more insightful results from
101: several aspects. First, we give
102: a general explanation of the model and a means to convert it into a simple and well known BD model based on a philosophical
103: approach. Our ideas are backed up and well supported by direct numerical simulations which 
104: enables to understand the physical nature of the system described by each term of the rate equations we used for 
105: analytical solutions. Second, we present more extensive results showing the 
106: asymptotic approach of the coverage towards the jamming limit in terms of the various parameters involved in the processes.
107:    
108: 
109:  
110: %\begin{eqnarray}  
111: %\label{eq:gap_1}
112: %{{\partial c(x,t)}\over{\partial t}} & = &- (x-\bar\sigma)c(x,t) +  q\a \sigma \Big[c(x+\sigma,t)-c(x,t)\Big] 
113: %\nonumber \\ && + p\beta m\sigma\Big [c(x+m\sigma,t)-c(x,t)\Big] \nonumber \\ &&
114: % 2q\int_{x+\sigma}^\infty c(y,t) dy + 2p\int_{x+m\sigma}^\infty c(y,t)dy
115: %\end{eqnarray}
116: 
117: \section{cbd model}
118:  
119: 
120: We consider a system which consists of a reservoir of particles with diameter $m$ lying in the immediate vicinity of 
121: a substrate which is an infinitely extended line. 
122: The adsorbing particles may be in the gas or in the fluid phase and arrive in the adsorbing plane through
123: Brownian motion. As soon as a particle comes into contact with a gap large enough to accommodate it, it 
124: is then adsorbed immediately and irreversibly. In addition, the incoming particle that touches an already adsorbed 
125: one, is allowed to follow the BD rules to form a monolayer. The simplicity of the 1D problem lies in the following 
126: situations. Every successful deposition of a particle on a given gap divides it into smaller gaps having the same 
127: geometry as the parent gap. It is this {\it shielding property}, found only in 
128: 1D, which we shall use to gain further insight into the problem and tackle it analytically. For simplicity's sake, we assume 
129: that the daughter gaps are uncorrelated, irrespective of the island size separating the gaps from their neighbours, 
130: so that we can treat each gap as an independent entity. We further assume that each roll-over motion is
131: completed prior to the next trial attempt for deposition. 
132: 
133: \begin{figure}[!htb]
134: \centerline{\includegraphics[width=8.0cm]%
135: {vir_ballis.eps}}
136:  \caption{Schematic illustration of the model in different situations. 
137: The shaded regions between $AB$ and $CD$ in (a) have local minimums and play no role except for kinetic reasons. 
138: We therefore eliminate all such shaded regions so that we have a system where all the gaps are separated 
139: from their neighbours by only one disk as shown in (b). The dotted lines in (b) represent the probability 
140: distribution in different regimes of a given gap.
141: The steepest descent path in (c) is artificially increased by using the
142: idea of a virtual disk at the expense of lowering the height of PD so that every point of $x+(2a-1)m$ is now
143: equally likely to be chosen by the trial attempt.}
144: \label{fig:1}
145: \end{figure}
146: 
147: At this point, it is useful to discuss the classical RSA and the simple BD model before introducing the
148: cooperative BD model. In the classical RSA, an incoming particle is 
149: adsorbed successfully in a gap of size $x$, if the center of the incoming particle arrives in any place but $m/2$ 
150: away from either edges of the gap. This means that only $(x-m)$ of a given gap $x$ is accessible for adsorption, 
151: which we have illustrated in Fig. 1(a). In the BD model, on the other hand, those particles that fall on an already 
152: adsorbed one may reach the substrate via the roll over motion. The deposition via rolling is successful if 
153: the center of the incoming particle falls within a distance of $m/2$ on either sides of both edges. 
154: It is then adsorbed on the respective edge creating a 
155: new gap of size $(x-m)$. That is, for a given gap $x$, the total position accessible to a new arrival is $(x+m)$ which is
156: shown in Fig. 1(a). Note that any particle dropping in the shaded regimes $AB$ or $CD$ 
157: are considered to be trapped due to the local minimum and will never reach the global minimum or the adsorbing substrate. 
158: Thus, if we are not interested in the kinetic aspect of the process, we can safely delete the shaded regimes as if they
159: did not exist and assume that the neighbouring gaps are separated by only one disk as shown in Fig. 1(b).    
160: We can thus define each gap as an independent isolated interval bordered on either end by a semidisk so that if 
161: we had connected the two remote ends it would then form a ring with one particle at the joint.
162: 
163: 
164: We are now in a position to introduce the long range attractive force among the incoming particles and the 
165: particles in the 
166: adsorbed phase. The question is how to incorporate it? First, we need to understand the effects of such an
167: attractive force. The most significant one is that each adsorbed particle 
168: will tend to attract the incoming particle towards it. This immediately breaks the random nature of the process as the 
169: incoming particles are more likely to land on an already adsorbed particle than on the gap. That is, as far as the
170: selection of the trial position is concerned, we have two distinct probability distribution (PD)
171: regimes as indicated in Fig 1(b). First, the {\it force free regime} where the attractive force is strictly zero. 
172: Second, the {\it force field regime} where the 
173: adsorbed particle exerts force and hence can influence the selection of the trial position. However, for analytical 
174: simplicity, we assume a square well potential around the center of each adsorbed particle of width $2m$. That is,
175: the strength of the attractive force is equal anywhere within the force field regime. This is 
176: indicated by the dotted lines in Fig. 1(b) representing the PD of the position chosen by the trial attempt. 
177: The flat PD implies that all the points within that regime have the same {\it a priori probability}.
178: 
179: \begin{figure}[!htb]
180: \centerline{\includegraphics[width=8.0cm]%
181: {vir_bal_dep.eps}}
182:  \caption{Schematic illustration of the rules for the cooperative ballistic deposition model. 
183: The incoming disk is adsorbed directly as shown in (i) if it is dropped  within $(x-m)$. The deposition via the rolling 
184: mechanism is depicted in (ii). Here, any disk that falls on the steepest descent path of the virtual disk, OP, 
185: is assumed to be dropping on its corresponding equivalent point of the real particle QP from where it can 
186: successfully reach the global minimum.}
187:  \label{fig:2}
188: \end{figure}
189: 
190: It is noteworthy to mention that the simple BD model 
191: refers to the case where  both regimes have the same flat PD height and hence the whole substrate represents
192: the zero force field or zero attractive force. Now, as soon as we switch on the attractive force, the height of the 
193: PD around each adsorbed particle will increase to a degree depending on the strength of the attractive force. However, we can lower the PD height
194: by increasing the width keeping the total area unchanged. In this way, we can make the whole system having the 
195: same PD height and treat it like the simple BD model where an incoming particle can land anywhere in the substrate
196: with equal probability including the disks that are already occupied as represented in Fig. 1(c). 
197: That is, we can envision it as follows: 
198: Prior to selecting a position for an incoming particle, we replace the size of the depositing particles used for
199: defining the gaps by a virtual disk of diameter $R>m$, without altering the gap size. In this way, we increase the 
200: cross-section of collision for an incoming particle in order to count the additional collision probability due to
201: the attractive force. Once the position for the next trial deposition is chosen, 
202: we can immediately return to the system with disks having diameter $m$ and  proceed according to the simple BD rules
203: as depicted in Fig. 2.
204: In doing so, we artificially increase the probability with which an incoming particle may collide with an already 
205: adsorbed disk thus enhancing the probability of adsorption via the rolling mechanism and mimicking the effect of 
206: the attractive force. One can thus expect an enhanced adsorption probability near the 
207: two extreme ends of each gap as the virtual disk size increases and, in the limit $R\longrightarrow \infty$, we can only
208: expect the adsorption via the rolling mechanism except in the very early stage where the virtual diameter and the gap size
209: may be of the same order in size. 
210: %The present model has the further very intuitive interpretation.
211: % We can envision such process as the effect of long-range attractive interactions 
212: %that eventually tend to attract the incoming particles towards the  already adsorbed  particles. 
213: %The strength of attraction in this model is measured in terms of the virtual disk size $R$. 
214: 
215: \section{Analytical solution}
216: 
217: To address the problem described above analytically, 
218: we adopt the well studied rate equation approach of the gap size distribution function or concentration 
219: $c(x,t)$. Let us define $c(x,t)dx$ as the 
220: number of gaps at time $t$ in the size range between $x$ and $x+dx$. 
221: The kinetics of adsorption of the monodisperse particles can then be described by the following set of rate equations:
222: \begin{eqnarray}
223: {{\partial c(x,t)}\over{\partial t}}& = &-(x-m)c(x,t)+2 \int_{x+m}^\infty c(y,t)dy \nonumber \\ && +
224: 2am \Big \{c(x+m,t)-c(x,t)\Big\}
225: \end{eqnarray}
226: for $x\geq m$ and
227: \begin{equation}
228: {{\partial c(x,t)}\over{\partial t}} = 2 \int_{x+m}^\infty c(y,t)dy +2am c(x+m,t)
229: \end{equation}
230: for $x<m$. The above rate equations are mean-field in nature as the fluctuations and correlations are ignored. 
231: The rate equation approach is 
232: based on the assumption that the creation and annihilation of gaps are independent of the size 
233: of the neighboring gaps. The first two terms of Eq. (1) and the first term of Eq. (2) are the same as that of the simple 
234: RSA process and thus describe the creation and destruction of a gap of size $x$ due to the direct 
235: adsorption of size $m$ on size $y\geq x+m$ or on size $x$, respectively. The remaining terms in both equations also 
236: describe the creation and destruction of gaps but due to the rolling motion over the steepest descent path 
237: to travel the maximum linear path $am$, where $a$ is a dimensionless constant number that we can tune. 
238: The factor `2' in the integral terms accounts for
239: the fact that any of the two new gaps created upon a direct deposition on the gap size  $y\geq x+m$ can be of
240: size $x$; whereas the same factor in the remaining terms takes into account that a gap of size $x$ can be 
241: created or destroyed from either end by adsorption. In order to understand the role of $a$, 
242: it is convenient to rewrite Eq. (1) as
243: \begin{eqnarray}
244: {{\partial c(x,t)}\over{\partial t}}& = &-\{x+(2a-1)m\}c(x,t)+2 \int_{x+m}^\infty c(y,t)dy \nonumber \\ && +
245: 2amc(x+m,t).
246: \end{eqnarray}
247: 
248: 
249: The term $\{x+(2a-1)m\}$ in the above equation is the key to understand the role of $a$. Note that by setting
250: $a=0$ we recover the classical RSA case where $(x-m)$ of a given gap $x$ is accessible for adsorption, which is
251: consistent with our discussion in the previous section. Similarly, $a=1$ describes the simple BD model where the 
252: total positions accessible to a new arrival is $(x+m)$ which is again consistent (see Fig. 1). 
253: In general, $\{x+(2a-1)m\}$ means that a given gap of size $x$ is bounded by a semidisk
254: of diameter $R=(2a-1)m$ while the adsorbing particles are of size $m$. That is, any particle that falls within a distance
255: $am$ from either end of  $\{x+(2a-1)m\}$ effectively will collide with the virtual disk. Every point of the steepest 
256: descent of the virtual path $OP$ in Fig. 2 has its corresponding equivalent point on the real path $QP$. Therefore, an 
257: incoming particle falling on the virtual path is assumed as if it were falling on the exact equivalent position 
258: of the real path and vice versa.   
259:    
260: 
261: To solve Eq. (1) we seek a trial solution of the following form  
262: \begin{equation}
263: c(x,t)=A(t)e^{-(x-m)t},
264: \end{equation}
265:  where $A(t)$ is still an undetermined quantity fixed by the initial condition. Let us assume a monodisperse initial
266: condition $c(x,0)=\delta(x-L)/L$ so that we have
267: \begin{equation}
268:  \lim_{L\longrightarrow \infty}\int_0^L c(x,0)dx =0, \hspace{0.3cm} \lim_{t \longrightarrow 0}\int_0^\infty xc(x,t)dx=1.
269: \end{equation}
270: Substituting the trial solution into Eq. (1), 
271: we obtain the following differential equation for $A(t)$
272: \begin{equation}
273: {{d\ln A(t)}\over{dt}}={{2e^{-m t}}\over{t}}+2ae^{-m t}.  
274: \end{equation}
275: Solving it, satisfying the initial conditions, we get
276: \begin{equation}
277: A(t)=t^2F(a, m t)
278: \end{equation}
279: where the auxiliary function $F(a,m t)$ is defined as
280: \begin{equation}
281: F(a, m t)= e^{-2\int_0^{m t}{{1-e^{-u}}\over{u}}du+2a(1-m t -e^{-m t})}.
282: \end{equation}
283: To obtain $c(x,t)$ for $x<m$, we substitute the solution of Eq. (1) into Eq. (2) and then upon a direct integration 
284: we get
285: \begin{equation}
286: c(x,t)=\int_0^t u(2+ am u)F(a,m u)e^{-xu}du.
287: \end{equation}
288: The solutions $c(x,t)$ can provide a complete analytical description of the process including its kinetic aspect.
289: All we need now is to find
290: useful ways of using these solutions for computing various physical quantities of interest such as the jamming coverage, 
291: the mean number density, the mean cluster size, etc. 
292: 
293: \section{Numerical Simulations}
294: 
295: \begin{figure}[!htb]
296: \centerline{\includegraphics[width=8.0cm]%
297: {vir_sim_1.eps}}
298:  \caption{Schematic illustration of the recursive simulation scheme.}
299:  \label{fig:3}
300: \end{figure}
301: To test the physical description of our CBD model, we have simulated it on a computer. One obvious constraint of the 
302: simulation is of course the finite size effect. However, for a sufficiently large substrate in comparison to the 
303: depositing particles, the finite-size effect can be made sufficiently small.  
304: To simplify the simulation, we use the approximation mentioned in the previous section that the gaps are 
305: uncorrelated and can be treated independently. This allows us to treat the problem in a recursive way if we focus on 
306: the coverage only and ignore the kinetics of the process. In brief,  the description of the recursive scheme of the 
307: simulation is as follows. Let us assume that the initial gap of unit interval $[0,1]$ bordered on both ends by semidisks 
308: of radius $(2a-1)m/2$ as shown in Fig. 3. We then generate a random 
309: number $n$ from the interval $[-(2a-1)m/2,1+(2a-1)m/2]$ to assign to the center of the incoming particle. The incoming 
310: particle is then adsorbed directly creating two new smaller intervals if $n$ lies in the 
311: interval $[m/2,1-m/2]$. Otherwise, the disk is adsorbed on one of the edges of $[0,1]$ creating only one new gap 
312: of interval $[m,1]$ or of $[0,1-m]$ depending on if is $n \in [-(2a-1)m/2,m/2]$ or $n\in [1-m/2,1+(2a-1)m/2]$ respectively.
313: % of size $[a+m/2,1]$ or $[0,1-(a+m/2)]$ depending
314: %on the random number if it is within $[0,a]$ or $[1-a,0]$ respectively. 
315: In the case when the disk is adsorbed in one of the two edges, we increase the 
316: counter which provides the information on the mean cluster size of the system. We then continue the process assuming
317: each new gap is again bordered by virtual semidisk and treat them in the same fashion as for the first step until we 
318: have no more gaps of size $\geq m$. At this point we add all the 
319: gaps of size $< m$ and finally, using this, we can immediately calculate the jamming coverage and the mean cluster size.
320: We have performed the simulation with substrate size $\sim 10^6m-10^8m$ within the interval $[0,1]$ and found an 
321: excellent match with the corresponding analytical results up to a several digits. We also noticed that increasing the 
322: substrate size by decreasing the $m$ value only contributes to a higher order precision as expected. 
323: 
324: %We have a gap of size $x$ which is defined as an interval having half of a disk of diameter $m$ at its two remote ends.
325: %However, for the CBD model before tossing for the random number we replace the disks that define the given gap $x$ by 
326: %the virtual disks of diameter $(2a-1)m$ keeping the gap size intact (see Fig. 1). 
327: %The first step can be described as follows. We find 
328: %a random position from the length $x+(2a-1)m$ to assign to the center of the incoming article. The incoming particle is 
329: %then adsorbed directly creating two new smaller gaps if the chosen position lies within  $(x-m)$ as shown in Fig. 1.
330: %Otherwise, the disk is adsorbed via rolling creating only one new gap $(x-m)$. In the latter case, we increase the 
331: %counter which can provides us the information on mean cluster size of the system. We continue the process treating each 
332: %new gap in the same fashion as we described for
333: %the first step until we have no more gaps of size $x\geq m$. At this point we add all the 
334: %gaps of size $\leq m$ and finally using this we can immediately calculate the jamming coverage.
335: 
336: 
337: \section{Results and analysis}
338: 
339: 
340: The fraction of the line covered by the adsorbing particles or 
341: the coverage $\theta(a,t)$ at different instants of time can be defined as
342: \begin{equation}
343: \theta(a,t)=1-\int_0^\infty xc(x,t)dx,
344: \end{equation}
345: while the number density is defined by the following relation
346: \begin{equation}
347: N(a,t)=\int_0^\infty c(x,t)dx.
348: \end{equation}
349: However, we find it more convenient to handle their rate equation rather than their definition itself. 
350: The kinetic equation for the coverage is  
351: \begin{equation}
352: \label{cov}
353: {{d\theta(a,t)}\over{dt}}=m\int_m^\infty\{x+m(2 a-1)\}c(x,t)dx =\Phi(a,t).
354: \end{equation}
355: Here, the quantity $\Phi(a,t)$ is the fraction of the substrate accessible to a new particle at a given time $t$.
356: The kinetic equation for the number density on the other hand is
357: \begin{equation}
358: \label{num}
359: {{dN(a,t)}\over{dt}}=\int_m^\infty (x-m)c(x,t)dx.
360: \end{equation}
361: The above two equations can be combined together to obtain
362: \begin{equation}
363: {{d\theta(a,t)}\over{dt}}=m{{dN(t)}\over{dt}}+2am\int_m^\infty c(x,t)dx.
364: \end{equation}
365: Note that in the simple RSA, one gap corresponds to one particle 
366: and thus we have $\theta(0,t)=m N(0,t)$ reflecting the fact that the average particle size is the same as the size of
367: the adsorbing particles. However, in the present case, for $a>0$, the second term 
368: of the above equation describes the cluster formation. That is, the first term on the right hand side of Eq. (13) 
369: takes into account the direct deposition while the effect of the rolling mechanism is described by the second term.
370: Using the solution for the appropriate boundary into Eq. (\ref{cov}) yields
371: \begin{equation}
372: \theta(a,t)=\int_0^{mt}F(a,u)(1+2au)du.
373: \end{equation}
374: This can provide all the information about kinetics aspect of the process, namely how the coverage evolves in time. 
375: %The fraction of the substrate accessible to a new arrival is $\Phi(a,0)=1$ and $\Phi(a,\infty)=1$ and at intermediate
376: %time $t$ it can be obtained from the follwoing relation,
377: %\begin{equation}
378: %\Phi(a,t)= F(a,mt)(1+2amt).
379: %\end{equation}
380: \begin{figure}[!htb]
381: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
382: {sim_1.eps}}
383:  \caption{Jamming coverage as a function of $a$ that measures the strength of attraction: $q(a)$ vs $a$.}
384:   \label{fig:4}
385: \end{figure}
386: \begin{figure}[!htb]
387: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
388: {nfig_1a.eps}}
389: \caption{The linear fits of $\ln(q(\infty)-q(a))$ vs $\ln(a)$ having slope $-1/2$ in the large $a$ regime reveals 
390: that the convergence of the jamming coverage towards the closest packing is power-law in nature.}
391:   \label{fig:5}
392: \end{figure}
393: %\begin{figure}[!htb]
394: %\centerline{\includegraphics[width=80mm,height=60mm,angle=0]%
395: %{cov_vs_a.eps}}
396: % \caption{Dependence of coverage on the distance the precursor species can travel: $q$ vs $a$.}
397: %  \label{fig:2}
398: %\end{figure}
399: %\begin{figure}[!htb]
400: %\centerline{\includegraphics[width=8.0cm]%
401: %{logcov_vs_loga.eps}}
402: % \caption{The convergence of $q(a)$ towards its asymptotic value: 
403: %$Y=\ln[0.999998-q(a)]$ vs $X=\ln[a]$.}
404: %  \label{fig:3}
405: %\end{figure}
406: One of the characteristics of the deposition process is that the system reaches a state of dead-lock 
407: in a finite time when particles can no longer be adsorbed. This is typically known as the jamming limit 
408: and the exact critical time to reach such a state should depend on $a$.
409: However, till to-date there do not exist any theoretical means to pin down the exact critical time for reaching the 
410: jamming limit. Nevertheless, we can safely calculate the coverage in the jamming limit as 
411: \begin{equation}
412: q(a)=\lim_{t\longrightarrow \infty} \theta(a,t).
413: \end{equation}
414: The jamming coverage has been of special interest in the study of the deposition phenomena as it can 
415: uniquely characterize the structure of the resulting monolayer.  From the exact expression for
416: the coverage, Eq. (15), it is of particular interest to know the approach of the coverage $\theta(a,t)$ to the 
417: corresponding jamming limit $q(a)$. We find that beyond the transient behavior, the system reaches its
418: corresponding asymptotic coverage namely the jamming limit exponentially, with a decay constant $2a$, 
419: multiplied by an algebraic prefactor $t^{-1}$
420: \begin{equation}
421: q(a)-\theta(a,t) \sim t^{-1}e^{-2at}, 
422: \end{equation}
423: which was also reported in \cite{kn.viot}. 
424: Obviously, the classical RSA ($a=0$), we recover the power-law behavior which is also known as 
425: Feder's law \cite{kn.feder1}. Here for $a>0$, the exponential approach towards the jamming limit reflects the fact that 
426: the increasing number of particles that land on an already adsorbed particles are successfully accommodated via 
427: rolling. Another interesting point to check is how the jamming limit varies as we increase the 
428: strength of interaction $a$. In other words, we want to see how the 
429: jamming limit changes as we increase the degree of correlation between the particles in the adsorbed and adsorbing phases.
430: Fig. 4 shows  a sharp rise in the jamming coverage at low $a$ and a slow rise towards 
431: the closest packing in the large $a$ regime. In an attempt to quantify the slow regime we plot 
432: $\ln(q(\infty)-q(a))$ against $\ln(a)$ in Fig. 5 and find that the jamming coverage converges towards the closest packing
433: obeying a power law 
434: \begin{equation}
435: q(\infty)-q(a) \sim a^{-1/2},
436: \end{equation}
437: where $q(\infty)\approx 1$. It is important to note here that the system never reaches a complete closest packing ($q=1$) 
438: even for $a\longrightarrow \infty$. This is due to the fact that the substrate size too is of the same order as that 
439: of the size of the virtual disk, hence there is always a non-zero probability for a direct deposition at least
440: in the early stage. We attempted to check it in the computer choosing both the virtual diameter and the initial 
441: substrate size to be of the same order and large enough to minimize the finite size effect. This is exactly the case 
442: described by the analytical model as we let $a\longrightarrow \infty$. In doing so we never find 
443: a cluster covering the whole substrate . We checked it over and over again by increasing the substrate size and
444: the virtual disk size up to $\sim 10^10m$. Nevertheless, neither analytical solution nor the simulation could give us 
445: an exact estimate for $q(\infty)$.
446: %\begin{figure}[!htb]
447: %\centerline{\includegraphics[width=8.5cm]%
448: %{assymp_coverage.eps}}
449: % \caption{The convergence of $q(a)$ towards its asymptotic value: 
450: %$Y=\ln[0.99998-q(a)]$ vs $X=\ln[a]$.}
451: %  \label{fig:2}
452: %\end{figure}
453: \begin{figure}[!htb]
454: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
455: {N_vs_q_1.eps}}
456:  \caption{Time dependence of the number density as a function of the coverage: The curves
457: from top to bottom represents $N(a,t)$ vs $\theta(a,t)$ for $a=0,1,5$ and $10$.}
458:   \label{fig:6}
459: \end{figure}
460: 
461: We now intend to obtain an exact expression for the number density by substituting the solution $c(x,t)$ for $x>m$
462: into Eq. (\ref{num}) which yields
463: \begin{equation}
464: N(a,t)={{\int_0^{m t} F(a,u)du}\over{m}}.
465: \end{equation}
466: %\begin{figure}[!htb]
467: %\centerline{\includegraphics[width=75mm,height=50mm,angle=0]%
468: %{a_k.eps}}
469: %\caption{Mean cluster size $k$ against $a$.}
470: %  \label{fig:6}
471: %\end{figure}
472: \begin{figure}[!htb]
473: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
474: {lna_lnk.eps}}
475: \caption{The linear fits of the plot of $\ln(k)$ vs $\ln(a)$ in the large $a$ regime with slope $1/2$ shows
476: that the mean cluster size increases as $\sim a^{1/2}$.}
477:   \label{fig:7}
478: \end{figure}
479: %\begin{figure}[!htb]
480: %\centerline{\includegraphics[width=8.0cm]%
481: %{logacs_vs_loga.eps}}
482: %\caption{Mean cluster size $s$ vs $a$ in the jamming limit for generalized ballistic deposition with $m=1$.}
483: %  \label{fig:6}
484: %\end{figure}
485: The above relation for the number density immediately implies that it depends on the size of the adsorbing particles. 
486: Here it is noteworthy to mention that the mean number density for the classical RSA ($a=0$) is simply 
487: the coverage divided by the size of the adsorbing particles. In this case, the mean number density increases 
488: linearly with the coverage as time proceeds having a slope $m$ during the process (see Fig. 6). 
489: However, as soon as a particle that fell on a previously adsorbed one can 
490: roll over the latter, the linear relation between $N(a,t)$ and $\theta(a,t)$ is immediately ceased and it is
491: replaced by a non-linear relation (see Fig. 6). Therefore, the mere knowledge of one of 
492: the two is not sufficient to obtain the other. This is due to the 
493: fact that the mean cluster size is different from the size of the adsorbing particle as the system keeps producing 
494: connected clusters of different sizes depending on the value of $a$. Fig. 3 shows that for $a>0$ the number density
495: grows linearly at a very low coverage. This is due to the fact that, at an initial stage the incoming particles hardly 
496: encounter any pre-occupied species and therefore there exists almost no cluster. However, as the substrate gets crowded,
497:  it is evident from Fig. 6 that the mean number density increases 
498: in a nonlinear fashion and the strength of non-linearity increases with increasing $a$. Therefore, to obtain the coverage 
499: (the number density) from the number density (coverage) we need to know the mean cluster size. 
500: %\begin{figure}[!htb]
501: %\centerline{\includegraphics[width=8.5cm]%
502: %{cluster_size_t.eps}}
503: % \caption{Mean cluster size $<l(t)>$ vs $t$ for generalized ballistic deposition with $m=1$.}
504: %  \label{fig:2}
505: %\end{figure}
506: The expression for the coverage $\theta(a,t)$ and  the mean number density $N(a,t)$  at different 
507: instants of time can give us an estimate of how the mean cluster size defined as
508: \begin{equation}
509: s(t)={{\theta(a,t)}\over{N(a,t)}}
510: \end{equation}
511: grows in time and with the strength of $a$. We find that for a given size of the adsorbing species, 
512: the mean cluster size in the jamming limit is 
513: \begin{equation}
514: \lim_{t \longrightarrow \infty}{{\theta(a,t)}\over{N(a,t)}}=s={{\int_0^{\infty}F(a,u)(1+2au)du}\over
515: {\int_0^{\infty} F(a,u)du}}m.
516: \end{equation}
517: %The above relation shows that for a given $m$, the mean cluster size increases monotonously and remains as a constant
518: %quantity.
519: 
520: Obviously, like the mean number density,  the mean cluster size should depend on the 
521: size of the adsorbing particle. However, the ratio between the two $k=s/m$ remains constant in the jamming limit and
522: therefore it is useful to call it the universal mean cluster size. It is note worthy to mention here that for the simple 
523: RSA we get $k=1$ and hence the mean cluster size is the same as that of the size of the 
524: adsorbing species. However, for $a>0$ we find that the universal mean cluster $k>1$ and $k$ increases  
525: monotonously with increasing $a$. Furthermore, Fig. 7 reveals that in the large $a$ regime the universal mean 
526: cluster size $k$ increases as
527: \begin{equation}
528: k\sim a^{1/2}.
529: \end{equation} 
530: The probability of adsorption of particles without overlapping with any preadsorbed particle decreases with
531: time and for all $a$; however the strength of decrease gets sharper and sharper as $a$ increases. 
532: For $a \longrightarrow \infty$, only at a very
533: initial stage some particles may be adsorbed by direct deposition without overlapping. Thus, the system 
534: never reaches a state of closest packing but of almost closest packing. This is well supported by our numerical
535: simulation.     
536: The jamming coverage thus increases with increasing $k$ (see Fig. 8). We find that like 
537: the $q(a)$ vs $a$, the approach of the jamming limit towards almost closest packing against $k$ also follows
538: a power law form but with a different exponent. As shown in Fig. 9, the plot of $q(k(a\longrightarrow \infty))-q(k)$
539: versus $k$ in the logarithmic scale along both axis is well fitted by a straight-line with slope $1$ and hence 
540: \begin{equation}
541: q(k(a\longrightarrow \infty))-q(k) \sim k^{-1},
542: \end{equation}
543: where $q(k(a\longrightarrow \infty)) \approx 1$.
544: \begin{figure}[!htb]
545: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
546: {sim_2.eps}}
547: \caption{The jamming coverage as a function of the mean cluster size $k$.}
548:   \label{fig:8}
549: \end{figure}
550: \begin{figure}[!htb]
551: \centerline{\includegraphics[width=80mm,height=50mm,angle=0]%
552: {nfig2a.eps}}
553: \caption{The linear fits of $\ln(q(\infty)-q(a))$ vs $\ln(k)$ in the large $k$ regime shows that the jamming 
554: coverage increases in a power-law fashion $\sim k^{-1}$.}
555:   \label{fig:9}
556: \end{figure}
557: 
558: %\begin{figure}[!htb]
559: %\centerline{\includegraphics[width=8.0cm]%
560: %{cov_vs_acs.eps}}
561: % \caption{Dependence of the jamming coverage on the mean cluster size in the same limit: $q(k)$ vs $k$.}
562: %  \label{fig:7}
563: %\end{figure}
564: %\begin{figure}[!htb]
565: %\centerline{\includegraphics[width=8.0cm]%
566: %{logcov_vs_logacs.eps}}
567: % \caption{Dependence of the jamming coverage on the mean cluster and its asymptotic behavior: 
568: %$Y=\ln[0.999998-q(k)]$ vs $X=\ln[k]$.}
569: %  \label{fig:8}
570: %\end{figure}
571: 
572: 
573: %\begin{figure}[!htb]
574: %\centerline{\includegraphics[width=8.5cm]%
575: %{mean_size.eps}}
576: % \caption{Mean cluster size $<l>$ vs $a$ in the jamming limit for generalized ballistic deposition with $m=1$.}
577: %  \label{fig:1}
578: %\end{figure}
579: 
580: 
581: \section{Discussion}
582: 
583: We first discuss various possible interpretations for different regimes constituted by different values of $a$
584: in our CBD model.
585: It is interesting to note that when $a=1$, the size of the virtual and the real disk coincides and hence every point along
586: the substrate has the equal probability of occupation by the incoming particle. In this case, 
587: particles that fall on an already adsorbed one may travel up to a linear distance of its own size $m$ 
588: via the rolling motion following the steepest descent path. The particle can either be trapped in the local
589: minimum or it can reach the global minimum. In the former case, it is rejected while in the latter case it
590: is adsorbed irreversibly touching the one it just rolled over, provided there is at least a gap to accommodate
591: it. The situation for $0<a<1$ is also interesting as it describes a mixed process composed of the simple RSA and the 
592: BD process. In this case, whenever an incoming disk encounters 
593: another disk, it is then allowed to roll over the latter disk with probability $a$.
594: The rolling mechanism over the previously adsorbed disk can be assumed to be due to the 
595: gravitational pull towards the adsorbing surface. The ballistic deposition model ($0<a\leq 1$) can 
596: thus describe the situation where the transport of the adsorbing species is dominated by gravitational effects. However,
597: the model with $1<a<\infty$ can describe the situation where the gravitational effects as well as the attractive 
598: interaction between the elements of the adsorbed phase and the incoming particles play the dominant role.
599:  
600: 
601: It is noteworthy to mention that Viot {\it et al} also generalised the  ballistic deposition model 
602: and gave the following definition  \cite{kn.viot}.
603: The disks of unit diameter are dropped uniformly and sequentially one at each time step. The disks can either reach 
604: the adsorbing plane or fall on an already adsorbed disk or on a cluster of disks. In the former case, the trial attempt 
605: is retained with a probability $q=(1-p)$. In the latter case, on the other hand, the trial disk follows the path of the 
606: steepest descent over the disk it encountered. The disk is then adsorbed with probability $p$ 
607: provided the particle can reach the global minimum by the roll over motion; otherwise it is trapped in an elevated 
608: position and it is rejected. In \cite{kn.viot} the parameter $a$ is defined as $a=p/q$. 
609: %For example, if we choose
610: %the case $a=2$ it would mean that the disk that could reach the surface without overlapping is adsorbed with probability
611: %$q=1/3$ and the one that reaches the surface via rolling is adsorbed with probability $p=2/3$. 
612: Therefore, the direct
613: adsorption and the deposition via the rolling mechanism  will both cause an increase in the rejection rate of the trial 
614: attempts for deposition. As a result, the approach to the jamming coverage in time should be slower as $a$ increases. 
615: However, the analytical solutions reveal an opposite behaviour, see Eq. (17). 
616: 
617: Note that it is the rejection criterion 
618: that determines how fast or how slow the coverage should evolve and finally reach the jamming limit. We would like to 
619: point out here that the apparent ambiguity arises because one cannot define  $a=p/q$ due to the following reasons.
620: We can rewrite Eqs. (1) and (2) for  $a=p/(1-p)$ upon multiplying both equations by $(1-p)$. 
621: %\begin{eqnarray}
622: %{{\partial c(x,t)}\over{\partial t}}& = &(1-p)\Big \{-(x-m)c(x,t)+2\int_{x+m}^\infty c(y,t)dy\Big \} \nonumber \\ && +
623: %2pm \Big \{c(x+m,t)-c(x,t)\Big\}
624: %\end{eqnarray}
625: %for $x\geq m$ and
626: %\begin{equation}
627: %{{\partial c(x,t)}\over{\partial t}} = 2 (1-p)\int_{x+m}^\infty c(y,t)dy +2pm c(x+m,t)
628: %\end{equation}
629: %for $x<m$.
630: If we then set $p=1$ in these equations we find that the remaining terms alone are incapable of describing any 
631: meaningful physical process. The point to emphasize here is that the roll over mechanism comes after the trial attempt for
632: deposition is made. The trial attempt to deposit the particle
633: is the primary event at each time step of the process which may then be followed by the roll over motion upon collision.
634: Therefore, the two events, the direct adsorption and the deposition via rolling are not mutually exclusive. 
635: 
636: 
637: The model we have presented in this article is solved exactly 
638: by means of a kinetic equation approach and supported by the numerical simulations based on the recursive algorithm. 
639: The exact match between the analytical solution and the numerical simulation helped not only to  confirm the validity 
640: of the mean-field approximation  but also to shed a deeper insight into the nature of the problem.
641: The basic principle of the model is the same as that of the simple ballistic deposition process. We have 
642: extended the simple BD by adding a certain degree of correlation 
643: between the adsorbing particles and those already adsorbed. To increase the flexibility of the model, we allowed a  
644: parameter $a$ that can tune the strength of the correlation which is induced by the attractive force. 
645: Instead of using the attractive force directly, we have shown a way of transforming it into a virtual situation, 
646: which is then just the simple BD model. 
647:   
648: The most significant consequence of the presence of the attractive force is that it results in an increased packing 
649: fraction and the mean clsuter size due to the formation of the higher order connected clusters. 
650: Moreover, the jamming coverage increases with the increasing degree of strength of the attractive force.
651: Similar results has also been recently reported by Pastor-Satorras and Rubi \cite{kn.rubi}, who 
652: studied a model of correlated sequential adsorption by numerical simulation both in one and two dimensions. 
653: However, unlike the square well type PD studied here, they used a Gaussian and exponential type PD around the center 
654: of each adsorbed particle. Nevertheless, despite the apparent differences in the detailed nature of the 
655: forces or in the PD,
656: the qualitative behaviour seems remarkably identical to what we have found in this article. 
657: Pastor-Satorras and Rubi too observed the similar trend of the increase of jamming
658: coverage as well as the mean cluster size. In addition, they too reported the approach towards the closest packing 
659: in the limit where the correlation is maximal. This reveals that the qualitative effect of the attractive force is 
660: insensitive to the detailed nature of the force. However, in addition to solving the model analytically  
661: we were able to quantify the effect of the attractive force. To this end, we have shown that in the strong force 
662: regime the convergence towards the closest packing against the strength of the 
663: attractive force follows a power-law relation $q(\infty)-q(a)\sim t^{-1/2}$. Perhpas, the emergence of such a power-law 
664: behaviour implies a universal nature of the phenomena including the exponent in the sense that it is independent of 
665: the detailed nature of the attractive force. However, at this point it is just a conjecture and we intend to investigate 
666: it in our future work.
667: 
668: \section{Conclusion}
669: 
670: In this article, we have presented an extension of the simple BD model by incoporating an attractive force betweent the
671: elements of the adsorbed phase and the incoming particle. 
672: The most significant consequence of the presence of the attractive force is the increase in packing fraction or the
673: jamming coverage of the resulting monolayer as we increase the strength of the attractive force $a$. 
674: This is manifested through the increase in the mean cluster size $k$ and in the strong force regimes it increases
675: as  $\sim a^{1/2}$. We have shown that the system exhibits a power-law approach of the jamming coverage towards the 
676: closest packing both in $a$ and $k$ but with different exponents $\sim a^{-1/2}$ and $\sim k^{-1}$ respectively, 
677: except for the weak field. It is important to note that although the jamming coverage increases 
678: with the degree of correlation {\it vis-a-vis} the mean cluster size, we can never create one single connected cluster 
679: spanning the whole substrate and giving the coverage $q=1$ which is indeed typical to 1D problem. 
680: Nevertheless, it indicates the potential structural phase transition in heigher dimensions and it has been indeed 
681: observed in \cite{kn.rubi}.
682:  
683: Finally, as our model produces connected clusters of different sizes, it may well explain some aspects
684: of RSA of polydisperse mixture in some sense. However, the mixture of particles has the following restrictions. 
685: Namely, the mixture contains a strict lower and upper cut-off and all the particles of 
686: intermediate size are of the integral multiple of the smallest particle in the mixture. 
687: 
688: \acknowledgements
689: 
690: This work was supported by the Alexander von Humboldt-Foundation (M.\ K. \ H.) and
691: the Deutsche Volkswagen-Stiftung (N.\ W.).   
692: 
693: 
694: 
695: 
696: %\begin{figure}[!htb]
697: %\centerline{\includegraphics[width=8.5cm]%
698: %{assymp_coverage.eps}}
699: % \caption{Asymptotic approach of mean cluster size. The top curve represents 
700: %$Y=\log[t(\theta(\infty)-\theta(t))]$ vs $t$
701: %and the bottom curve represents $Y=\log[\theta(\infty)-\theta(t)]$ vs $t$ with $m=1$.}
702: %  \label{fig:3}
703: %\end{figure}
704: 
705: \begin{thebibliography}{99}  
706: 
707: \bibitem{kn.evans} J. W. Evans, Rev. Mod. Phys. {\bf 65}, 1281 (1993).
708: \bibitem{kn.tarjus} J. Talbot, G. Tarjus, P. R. Van Tassel and P. Viot, Colloid Surface A {\bf 165}, 287 (2000).
709: \bibitem{kn.schaaf} P. Schaaf, J.-C. Voegel, and B. Senger, J. Phys. Chem. B {\bf 104}, 2204 (2000).
710: \bibitem{kn.renyi} A. Renyi, Publ. Math. Inst. Hung. Acad. Sci {\bf 3}, 109 (1958);
711: J. J. Gonzalez, P. C. Hemmer and J. S. Hoye, Chem. Phys. {\bf 3}, 288 (1974).
712: \bibitem{kn.feder} J. Feder and I. Giaever, J. Colloid Interface Sci. {\bf 78}, 144 (1980); A. Yokoyama, 
713: K. R. Srinivasan and H. S. Fogler, J. Colloid Interface Sci. {\bf 126}, 141 (1988).
714: \bibitem{kn.onoda} G. Y. Onoda and E. G. Liniger, Phys. Rev. A {\bf 33},715 (1986).
715: \bibitem{kn.mann} P. Wjtaszczyk, E. K. Mann, B. Senger, J.-C. Voegel and P. Schaaf, J. Chem. Phys. {\bf 103}, 8285 (1995).
716: 
717: \bibitem{kn.micha} M. Semmler, J. Ricka and M. Borkovec,  Colloid Surface A {\bf 165}, 79 (2000)
718: \bibitem{kn.pagonabarraga} I. Pagonabarraga, P. Wjtaszczyk, J. M. Rubi, B. Senger, J.-C. Voegel and, P. Schaaf,
719: J. Chem. Phys. {\bf 105}, 99 (1993).
720: \bibitem{kn.rodgers} G. J. Rodgers  and Z. Tavassoli, Phys. Lett. A {\bf 246}, 252 (1998);
721:   G. J. Rodgers and J. A. N. Filipe, J. Phys. A: Math. Gen. {\bf 30}, 3449 (1997).
722: \bibitem{kn.hassan} M. K. Hassan,  J. Schmidt, B. Blasius, and J. Kurths, Phys. Rev E {\bf 65}, 045103(R) (2002);
723:  M. K. Hassan and J. Kurths, J. Phys. A: Math. Gen. {\bf 34}, 7517 (2001).
724: \bibitem{kn.bonnier}  B. Bonnier, Phys. Rev. E {\bf 64}, 066111 (2001).
725: \bibitem{kn.pastor}  R. Pastor-Satorras and J. M. Rubi, Phys. Rev. E {\bf 64 }, 016103 (2001).
726: \bibitem{kn.lee} For an issue exclusively dedicated to the subject see:
727: Colloid Surface A {\bf 165} (2000).
728: \bibitem{kn.penrose} For recent rigorous mathematical articles see: 
729: M. D. Penrose and  J. E. Yukich, Ann. Appl Probab. {\bf 12}, 272 (2002); M. D. Penrose, 
730:  J. Stat Phys. {\bf 105}, 561 (2001); M. D. Penrose, Commun. Math. Phys. {\bf 218}, 153 (2001).    
731: \bibitem{kn.talbot} J. Talbot and S. M. Ricci, Phys. Rev. Lett {\bf 68}, 958 (1992).
732: \bibitem{kn.jullien} R. Jullien and P. Meakin, J. Phys. A: Math. Gen. {\bf 25}, L189 (1992).
733: \bibitem{kn.electrostatic} M. Semmler, E. K. Mann, J. Ricka, and M. Borkovec, Langmuir {\bf 14}, 5127 (1998).
734: \bibitem{kn.dipolar} R. Pastor-Satorras and J. M. Rubi, Phys. Rev. Lett. {\bf 80}, 5373 (1998).
735: \bibitem{kn.hydrodynamic} P. Wojtaszczyk and J. B. Avalos, Phys. Rev. Lett. {\bf 80}, 754 (1998).
736: \bibitem{kn.viot} P. Viot, G. Tarjus, and J. Talbot,  Phys. Rev E {\bf 48}, 480 (1993).
737: \bibitem{kn.feder1} J. Feder, J. Theor. Biol. {\bf 87}, 237 (1980); R. H. Swendsen, Phys. Rev A {\bf 24}, 504 (1981); 
738: Y. Pomeau, J. Phys. A: Math. Gen. {\bf 13}, L193 (1980).
739: \bibitem{kn.rubi} R. Pastor-Satorras and J. M. Rubi, Phys. Rev. E {\bf 64}, 016103 (2001).
740: %\bibitem{kn.privman} {\it Nonequilibrium statistical mechanics in one dimension} ed. V. Privman (Cambridge University 
741: %Press, 1997).  
742: 
743: 
744: 
745: 
746: \end{thebibliography} 
747: 
748: 
749: 
750: \end{document}
751: 
752: 
753: