1: %--------------------- KPZ Mean field approximation ---------------------------
2:
3: \documentclass[11pt,towside,floatfix]{article}
4: \usepackage{amsmath,amssymb,epsf,graphicx,cite,bbm}
5: \parskip 2mm
6: \begin{document}
7:
8: \newcommand{\mathbold}[1]{\mbox{{\boldmath $#1$}}}
9: \newcommand{\beq}{\begin{equation}}
10: \newcommand{\eeq}{\end{equation}}
11: \newcommand{\beqn}{\begin{eqnarray}}
12: \newcommand{\eeqn}{\end{eqnarray}}
13: \newcommand{\bearr}{\begin{array}}
14: \newcommand{\enarr}{\end{array}}
15: \newcommand{\derp}[2]{\frac{\partial{#1}}{\partial{#2}}}
16: \newcommand{\toref}[1]{\mbox{(\ref{#1})}}
17: \newcommand{\ket}[1]{|#1\rangle}
18: \newcommand{\bra}[1]{\langle#1|}
19: \newcommand{\bracket}[2]{\langle#1|#2\rangle}
20: \newcommand{\comm}[2]{\left[#1,#2\right]}
21:
22: \newcommand{\anticomm}[2]{\left\{#1,#2\right\}}
23: \newcommand{\dcomm}[3]{\left[#1,#2,#3\right]}
24: \newcommand{\danticomm}[3]{\left\{#1,#2,#3\right\}}
25: \newcommand{\expval}[3]{\langle#1|#2|#3\rangle}
26: \newcommand{\redexpval}[3]{\langle#1||#2||#3\rangle}
27: \newcommand{\media}[1]{\langle#1\rangle}
28: \newcommand{\eps}{\varepsilon}
29:
30: \title{Mean field theory for skewed height profiles\\ in KPZ growth processes}
31: \maketitle
32:
33: \begin{center}
34: Francesco Ginelli and Haye Hinrichsen
35:
36: \vglue 3mm
37: Institut f{\"u}r Theoretische Physik und Astrophysik, \\
38: Universit{\"a}t W{\"u}rzburg, D-97074 W{\"u}rzburg, Germany
39:
40: \end{center}
41:
42: \begin{abstract}
43: We propose a mean field theory for interfaces growing according to the
44: Kardar-Parisi-Zhang (KPZ) equation in 1+1 dimensions. The mean field equations
45: are formulated in terms of densities at different heights, taking surface
46: tension and the influence of the nonlinear term in the KPZ equation into account.
47: Although spatial correlations are neglected, the mean field equations still reflect
48: the spatial dimensionality of the system. In the special case of Edwards-Wilkinson growth,
49: our mean field theory correctly reproduces all features. In the presence of a
50: nonlinear term one observes a crossover to a KPZ-like behavior with the correct
51: dynamical exponent $z=3/2$. In particular we compute the skewed interface profile
52: during roughening, and we study the influence of a co-moving
53: reflecting wall, which has been discussed recently in the context of nonequilibrium
54: wetting and synchronization transitions. Also here the mean field approximation
55: reproduces all qualitative features of the full KPZ equation, although with different
56: values of the surface exponents.
57: \end{abstract}
58:
59:
60:
61:
62: %--------------------------------------------------------------------------
63: % Macros:
64: %--------------------------------------------------------------------------
65:
66: \def\xvec{\vec{x}} % Vector x
67: \def\nupar{{\nu_\parallel}} % temporal exponent
68: \def\nuperp{{\nu_\perp}} % spatial exponent
69: %\def\onethird{{\text{$\frac13$}}} % spatial exponent
70:
71: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
72: \section{Introduction}
73: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
74:
75: Since many years the physical properties of deposition-evaporation processes on a planar
76: surface have been studied theoretically by analyzing appropriate
77: stochastic growth models that capture the essential features of the
78: experimental realm~\cite{Dietrich}. In most of these models the configuration of
79: the growing surface is described by a height variable $h(\xvec,t)$ that yields the height
80: of the interface between deposited layer and gas phase above point $\xvec$ of
81: the substrate at time $t$. Starting with a certain initial configuration, the
82: interface then evolves according to certain stochastic rules.
83:
84: Depending on the specific dynamic rules for deposition and evaporation and their
85: symmetries, the temporal evolution of the interface may be described on a
86: coarse-grained scale by a stochastic differential equation, one of the
87: simplest and more general one being the celebrated Kardar-Parisi-Zhang (KPZ)
88: equation~\cite{KPZ}
89: %
90: %
91: \begin{equation}
92: \label{KPZeq}
93: \frac{\partial}{\partial t}h(\xvec,t) \;=\; v_0 + D \nabla^2 h(\xvec,t)
94: +\frac{\lambda}{2} [\nabla h(\xvec,t)]^2 + \xi(\xvec,t)\,.
95: \end{equation}
96: %
97: %
98: Here $v_0$ is the average growth velocity which can be set to zero in a
99: co-moving frame, the Laplacian accounts for surface tension of the interface,
100: and $\xi(\xvec,t)$ is an uncorrelated white Gaussian noise generated by the stochastic
101: nature of deposition and evaporation. Moreover, the nonlinear term $(\nabla h)^2$
102: it's the simplest one which breaks the invariance under reflections $h \to -h$.
103:
104: As many models for interface growth, the KPZ equation exhibits dynamic scaling,
105: i.e., starting with a flat configuration the interface width
106: $w(t)=\sqrt{\langle h^2 \rangle - \langle h \rangle^2}$ (where $\langle \cdot \rangle$
107: denotes average over space and ensemble realizations) first increases as a
108: power law $w(t)\sim t^\gamma$ until it saturates in a finite system of linear
109: size $L$ at a stationary value $w_{\text{stat}} \sim L^\alpha$. The crossover from a roughening to a
110: stationary state is described by the well-known Family-Vicsek scaling form~\cite{FVScaling}
111: %
112: %
113: \begin{equation}
114: \label{VWScaling}
115: w(t) \;\sim\; t^\gamma \,g(t/L^z)\,,
116: \end{equation}
117: %
118: %
119: where $g$ is a universal scaling function and $z=\alpha/\gamma$ is the dynamical exponent.
120:
121: The width is actually related to the second moment of the height distribution
122: profile $P_L(h,t)$, which is defined as the normalized probability to find the
123: interface at a randomly chosen lattice site at height $h$. Clearly,
124: the height distribution contains much more information about the interface
125: morphology than the width alone.
126: As for the width, dynamic scaling implies a scaling form for the height
127: distribution which in a co-moving frame may be written
128: as\footnote{The scaling functions $g$ and $f$ are related by
129: $g^2(u)=\int_0^\infty f(u,v)v^2\,dv-[\int_0^\infty f(v,u)v\,dv]^2$.}
130: %
131: %
132: \begin{equation}
133: \label{ScalingForm}
134: P_L(h,t) \;=\; t^{-\gamma} \, f(h/t^\gamma,t/L^z) \,.
135: \end{equation}
136: %
137: %
138: Obviously, both the critical exponents $\alpha$ and $\gamma$ and the shape of height
139: profile during roughening or after saturation reflect the symmetries of the
140: growth process under consideration. The simple case of invariance under
141: the reflection $h \to -h$ can be studied by imposing $\lambda = 0$ in
142: Eq. \toref{KPZeq}. In this case the linear Edwards-Wilkinson (EW) equation
143: \cite{EW} is recovered, and the critical growth exponents in 1+1 dimensions take the
144: values $\alpha=1/2$ and $\gamma=1/4$. Moreover, the height profile
145: of 1+1-dimensional EW processes is known to be a simple Gaussian distribution
146: both in the dynamically roughening phase as well as in the stationary state.
147:
148: In more realistic growth models, where nearest neighbour interactions play a
149: role in the dynamics of the growing interface, reflection symmetry is broken
150: and the nonlinear KPZ term has to be taken into account.
151: In 1+1 dimensions such a term is known to be {\it relevant} in the renormalization group
152: sense. Therefore, even when the
153: reflection symmetry is weakly violated (i.e., if $\lambda$ is small),
154: the scaling behavior of an infinite system will eventually cross over from EW to
155: KPZ scaling, the latter being characterized by the
156: exponents $\alpha=1/2$, $\gamma=1/3$, and $z=\alpha/\gamma=3/2$.
157:
158: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
159: \begin{figure}
160: \centerline{\includegraphics[width=105mm]{profdemo.eps}}
161: \caption{\label{figprofile} \small
162: Skewed interface profile. Left panel: Numerically determined profile $f(z):=
163: f(z,0)$ of a roughening KPZ interface in 1+1 dimensions for $\lambda<0$ (solid
164: line) compared to a Gaussian distribution of the same width (dashed
165: line). Right panel: Effective exponents $\eta_{eff}(z)$ (see text)
166: in comparison with the numerical estimates $\eta_+=1.6(2)$ and
167: $\eta_-=2.4(2)$ reported in Ref.~\cite{KBM91a} (representing the error bars as
168: dashed boxes) and the theoretical predictions for the PNG model
169: \cite{Prahofer2000} (marked by full horizontal lines).
170: }
171: \end{figure}
172: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
173:
174: With a non-symmetric term being present there is no longer any reason for the height
175: distribution $P_L(h,t)$ to be symmetric with respect to $h$. Although in 1+1
176: dimensions a KPZ interface of a {\it finite} system after saturation still happens to be
177: symmetric and Gaussian (see e.g.~\cite{Barabasi}), the profile of a {\em roughening}
178: KPZ interface before saturation is indeed skewed~\cite{Zhang}, reflecting the asymmetry of
179: the nonlinear term. In what follows we therefore restrict ourselves
180: to the roughening process before saturation, i.e., $t \ll L^z$, regarding a virtually
181: infinite system at finite times. Formally speaking, this can be achieved by
182: taking the thermodynamic limit $L \to \infty$ {\it before} the
183: time--asymptotic limit $t \to \infty$ is carried out.
184: In particular, we are interested in the scaling function
185: $f(z) := f(z,0)$, which renders the rescaled shape of the skewed profile after
186: sufficiently long time (see left panel of Fig.~\ref{figprofile}).
187:
188: The function $f(z)$ is known to be universal, i.e.,
189: the asymptotic shape of the skewed profile is
190: fully determined by the underlying KPZ field theory
191: and does not depend on the microscopic details of the model.
192: It has been suggested that the finite--time height distribution,
193: especially the form of its tails, is
194: approximately given by a stretched exponential
195: %
196: %
197: \beq
198: P_L(h,t) \propto \exp \left[-\mu\left(|h-\langle h
199: \rangle|/t^{\gamma}\right)^{\eta_{\pm}}\right] \quad t \ll L^z \,,
200: \label{SkewP}
201: \eeq
202: %
203: %
204: meaning that $f(z) \sim \exp(-\mu|z|^{\eta_\pm})$. Here $\mu$ is a
205: metric factor while the exponents
206: $\eta_{\pm}$ refer to the two different tails of the distribution with
207: $\pm \lambda (h-\langle h \rangle)> 0$.
208: Because of the skewness both exponents are expected to be different.
209: An argument based on a replica scaling
210: analysis of directed polymers~\cite{Zhang2}, whose free energy fluctuations
211: corresponds to the height fluctuations of a KPZ interface,
212: suggests the value $\eta_+ = 3/2$.
213:
214: As a breakthrough, Pr{\"a}hofer and Spohn have shown recently~\cite{Prahofer2000}
215: that the finite--time rescaled height
216: profile of the polynuclear growth model (PNG) \cite{PNG}, a model which is believed
217: to belong to the KPZ universality class, equals the Gaussian orthogonal ensemble
218: (GOE) Tracy-Widom distribution. This immediatly leads to
219: %
220: %
221: \beq
222: \eta_+ = 3/2 \,, \quad \eta_-=3.
223: \eeq
224: %
225: %
226: Numerical simulations reported in literature
227: concerning both directed polymers and KPZ lattice models~\cite{KBM91a}
228: give $\eta_+ = 1.6(2)$ and $\eta_- = 2.4(2)$, the latter value being not in
229: agreement with theoretical predictions. However, since these results
230: were obtained more than a decade ago the numerical precision was limited.
231: Performing similar simulations using the so-called single
232: step model~\cite{SSM} (see Sec. 2) we measured the effective exponent
233: $
234: \eta_{\rm eff}(z) = \frac{z}{\ln f(z)} \, \frac{d}{dz} \ln f(z)
235: $
236: which according to Eq.~(\ref{SkewP}) should converge to $\eta_\pm$ as $|z| \to \infty$.
237: As it can be seen in the right panel of Fig.~\ref{figprofile}, our numerical results
238: are in agreement with theoretical predictions for the PNG model while being incompatible with the previous estimate for $\eta_-$ of Ref.~\cite{KBM91a}.
239: Clearly, the center of the scaling function (i.e. small values of $z$) is
240: described only approximately by Eq.~(\ref{SkewP}).
241:
242: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
243: \begin{figure}
244: \centerline{\includegraphics[width=80mm]{walldemo.eps}}
245: \caption{\label{figwall} \small
246: Snapshots of a roughening interface in a reference frame where the asymptotic
247: velocity is zero. The figure shows a free interface with $\lambda{<}0$ (upper panel)
248: compared to interfaces confined by a co-moving lower wall in the two cases $\lambda{<}0$
249: (middle panel) and $\lambda{>}0$ (lower panel).
250: Simulations have been performed using the single step model (see Sec. 2) and
251: snapshots have been taken after 2048 time steps. }
252: \end{figure}
253: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
254:
255: Looking at a snapshot of a roughening interface in 1+1 dimensions, it is almost
256: impossible to recognize the influence of the KPZ nonlinearity by naked eye.
257: Its influence, however, is much more pronounced in the presence of a hard-core wall.
258: The wall is fixed in a frame where the asymptotic velocity of the interface vanishes
259: and interacts with the interface solely by preventing excursions to negative heights.
260: As can be seen in Fig.~\ref{figwall}, in presence of such a wall one can easily
261: appreciate the dramatic difference emerging when the sign of $\lambda$ is
262: changed\footnote{Alternatively one may compare a lower and an upper wall while
263: keeping the sign of $\lambda$ fixed.}.
264: Surprisingly, for $\lambda<0$ the interface touches the wall only occasionally,
265: while a high density of contact points is observed for $\lambda>0$.
266:
267: The properties of a KPZ interface close to a reflecting wall has been studied recently
268: in the context of nonequilibrium wetting~\cite{Wetting}, where the interface
269: describes a wetting layer on a planar substrate. Upon varying the average
270: growth velocity $v_0$ the interface undergoes a {\it depinning} transition between a
271: pinned phase, in which portions of the interface remains attached to the wall, to a
272: depinned phase, where the interfaces detaches entirely and starts moving upwards.
273: At the critical point, where the asymptotic interface velocity is zero,
274: a second order phase transition takes place and various scaling laws can be singled out.
275: In addition, the case $\lambda <0$ describes the critical properties of most synchronization
276: transitions in spatially extended chaotic systems \cite{Livi, Pikov}.
277:
278: Previous numerical simulations suggested that the temporal
279: decay of the density $\rho_0(t)$ of contact points, where the interface touches the wall,
280: obeys the power law ~\cite{MN1,MN2}
281: %
282: %
283: \begin{equation}
284: \label{theta}
285: \rho_0(t) \sim t^{-\theta},\qquad \theta \approx
286: \left\{
287: \begin{array}{cc}
288: 1.1(1) & \text{ if } \lambda<0 \\
289: 3/4 & \text{ if } \lambda=0 \\
290: 0.22(2) & \text{ if } \lambda>0
291: \end{array}
292: \right.
293: \quad,
294: \end{equation}
295: %
296: %
297: where the exponent $3/4$ can be obtained from a transfer matrix calculation
298: \cite{Wetting}. Moreover, a hyperscaling relation observed in simulations
299: starting with a single pinned site~\cite{Droz}
300: suggests the rational value $\theta = 7/6$.
301: Obviously, the different values of the exponents reflect the asymmetry of the
302: nonlinear term with respect to reflections $h \to -h$. Moreover, the pronounced numerical
303: variation of $\theta$ by a factor of 5 explains why the snapshots
304: in Fig.~\ref{figwall} are so strikingly different.
305:
306: Interestingly, the profile of a
307: roughening KPZ interface next to a wall cannot be described
308: in terms of an appropriate generalized GOE Tracy-Widom distribution because of
309: emerging nonlinear term. More generally, the critical behavior of such a bounded growth process in
310: 1+1 dimensions is not easily accessible by analytical
311: means. For example, renormalization group techniques fail either
312: due to the presence of a strong--coupling fixed point unaccessible
313: by perturbative approaches in the case $\lambda<0$~\cite{MNRG} or
314: due to essential singularities arising for $\lambda>0$.
315: Therefore, the primary aim of the present paper is to
316: discuss this case within a suitable mean field approximation.
317: The mean field theory to be constructed should incorporate the asymmetry caused by
318: the nonlinear term and should render a skewed height distribution with similar
319: properties as in the full model. Although the mean field theory ignores space,
320: it should not resemble a naive infinite-dimensional limit (where a KPZ interface
321: is always smooth), instead it should reflect to some extent the dimensionality of
322: space in the thermodynamic limit $L \to \infty$. Moreover, the desired theory
323: should be as simple as possible and exactly solvable. In the following
324: sections we propose and solve a mean field theory which meets these requirements.
325:
326: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
327: \section{Mean field equations}
328: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
329:
330: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
331: \begin{figure}
332: \centerline{\includegraphics[width=125mm]{ssm.eps}}
333: \caption{\label{figssm} \small
334: Single step model in 1+1 dimension. Left panel: The simulation usually starts with
335: a flat interface, realized as a horizontal sawtooth pattern.
336: On selecting a site with a local minimum
337: a diamond (rhombus) is deposited with probability~$p$, flipping up the interface
338: by two units. Similarly, if the selected site happens to be a local maximum, a diamond
339: may evaporate with probability $1-p$, flipping the interface downward by two units.
340: Right panel: For $p>\frac12$ the interface roughens and propagates upwards.
341: }
342: \end{figure}
343: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
344:
345: The mean field equations proposed here are inspired by a particular model,
346: the so-called single step model (SSM)~\cite{SSM}, which is probably the simplest
347: and most compelling lattice model for KPZ-type interface growth.
348:
349: In the single step model the growing interface is represented by a set of
350: integer heights $n_i \in \mathbbm{N}$ residing at the sites $i=1\ldots L$ of a
351: one-dimensional lattice of length $L$ with periodic boundary conditions,
352: obeying the restriction
353: %
354: %
355: \begin{equation}
356: \label{restriction}
357: n_{i+1}-n_i=\pm 1\,.
358: \end{equation}
359: %
360: %
361: The interface evolves in time according random sequential updates as follows:
362: At each sub--time step $dt=1/L$ a site $i$ is chosen at random.
363: If the interface has a local minimum at site $i$ (i.e., $n_i < n_{i\pm 1}$)
364: the height $n_i$ is increased by $2$ with probability $p\in[0,1]$. This
365: update can be pictured as depositing a diamond (see Fig.~\ref{figssm}),
366: transforming a local minimum into a local maximum. Similarly, if the selected site
367: happens to be a local maximum ($n_i > n_{i\pm 1}$) the height $n_i$ is decreased
368: by two units with probability $1-p$.
369:
370: For $p=1/2$ the propagation velocity of the interface is zero and
371: the evolution rules satisfy detailed balance, as described by the
372: EW equation. For $p \neq 1/2$, however, the propagation velocity
373: is nonzero, depending on the roughness and the average slope of the interface.
374: In this case the SSM exhibits KPZ growth with
375: $\lambda$ being equal to $\frac12-p$.
376:
377: By identifying upward segments $n_{i+1}-n_i=1$ with particles and
378: downward segments $n_{i+1}-n_i=-1$ with vacancies,
379: the single step model can be mapped exactly onto a partially
380: asymmetric exclusion process (ASEP)~\cite{ASEP} of diffusing particles with density $1/2$.
381: Since the ASEP is known to evolve towards an uncorrelated product state with a
382: current $j=p/2-1/4$, it is immediately clear that the propagation velocity of
383: the interface tends to
384: %
385: %
386: \begin{equation}
387: v_\infty = \lim_{t \to \infty} v(t) = p-\frac12
388: \end{equation}
389: %
390: %
391: as $t \to \infty$.\footnote{Initially the velocity is higher, and the
392: excess velocity $|v(t)-v_\infty|$ decays as $t^{-1/3}$.}
393: Moreover, the mapping to the ASEP allows one to solve the
394: model via Bethe ansatz\cite{Gwa92}, making it possible to derive the
395: KPZ dynamical exponent $z=3/2$ and various other quantities exactly.
396: Other rigorous results concerning shape fluctuations in the ASEP can also be
397: found in Ref. \cite{Johansson}.
398:
399: In order to formulate a mean field theory for the single step model,
400: let $N_u(n,t)$ and $N_d(n,t)$ be the probabilities of finding an upward or downward
401: segment with their lower edge rooted at height level $n$. Let us first consider a deposition
402: process, in which a local minimum at level $n$ is flipped into a local maximum
403: at level $n+2$.
404: Having selected a random site, the probability to find such a local minimum
405: at a given height can
406: be approximated as follows. Clearly, the probability of finding a downward
407: segment on the left side terminating at height $n$ is $N_d(n,t)/L$, where $L$
408: is the system size. With this probability, knowing that the height of the selected site
409: is $n$, the adjacent segment to the right can only go up or down so that
410: the \textit{conditional} probability to find an upward segment is
411: given by $N_u(n,t) /(N_u(n,t)+N_d(n-1,t))$. Ignoring possible correlations
412: the total probability of finding a local minimum at
413: height $n$ is the product of these two expressions.
414: The deposition process, taking place with probability $p$, therefore leads to a loss
415: of probability at level $n$
416: %
417: %
418: \begin{equation}
419: \label{depositionevent}
420: \begin{split}
421: N_u(n,t) \to N_u(n,t+dt) &= N_u(n,t) - \frac{p}{L} \, \frac{N_d(n,t)N_u(n,t)}{(N_u(n,t)+N_d(n-1,t))}\\
422: N_d(n,t) \to N_d(n,t+dt) &= N_d(n,t) - \frac{p}{L} \, \frac{N_d(n,t)N_u(n,t)}{(N_u(n,t)+N_d(n-1,t))}
423: \end{split}
424: \end{equation}
425: %
426: %
427: and a corresponding gain at level $n+1$. Similar expressions can be derived
428: for the evaporation process. Obviously, this approximation accounts
429: for the restriction~(\ref{restriction}) and the one-dimensional structure
430: of the model but disregards possible nearest-neighbor correlations.
431:
432: The structure of Eq.~(\ref{depositionevent}) suggests that the probabilities $N_u(n,t)$ and $N_d(n,t)$ evolve exactly in the same way. In fact, it is easy to see that
433: in a system with periodic boundary conditions the numbers of
434: upward and downward segments are exactly equal. Assuming the same to hold
435: in an infinite system we have
436: %
437: %
438: \begin{equation}
439: N_u(n,t)=N_d(n,t)
440: \end{equation}
441: %
442: %
443: for every $n$ and $t$. Thus, introducing a combined probability density
444: %
445: %
446: \begin{equation}
447: P(n,t) = \frac{N_d(n,t)+N_u(n,t)}{2L}
448: \end{equation}
449: %
450: %
451: the loss at level $n$ due to a deposition event in Eq.~(\ref{depositionevent}) can be recast as
452: %
453: %
454: \begin{equation}
455: P(n,t) \to P(n,t+dt)=P(n,t)-p\,\frac{P_n(t)^2}{P_n(t)+P_{n-1}(t)} \,.
456: \label{mf2}
457: \end{equation}
458: %
459: %
460: Collecting all loss and gain contributions due to deposition and evaporation
461: one arrives at the following set of mean field equations
462: %
463: %
464: \begin{eqnarray}
465: \frac{\partial}{\partial t} P_n(t) \;=\;
466: p\left[ \frac{P_{n-1}(t)^2}{P_{n-1}(t)+P_{n-2}(t)} -
467: \frac{P_n(t)^2}{P_n(t)+P_{n-1}(t)}\right] \nonumber \\
468: +(1-p)\left[ \frac{P_{n+1}(t)^2}{P_{n+1}(t)+P_{n+2}(t)} -
469: \frac{P_n(t)^2}{P_{n}(t)+P_{n+1}(t)}\right]
470: \label{MF}
471: \end{eqnarray}
472: %
473: %
474: which serve as a starting point for all further calculations throughout this paper.
475: Notice that the form of the denominators appearing on the r.h.s of Eq.~\toref{MF}
476: is a consequence of the restriction~(\ref{restriction}), and that some care
477: has to be taken when one of them vanishes. Since the numerators are quadratic
478: we assume that each of these terms is zero {\it whenever}
479: their denominator vanishes.
480:
481: Introducing a probability current flowing between neighboring levels
482: %
483: %
484: \begin{equation}
485: \label{current}
486: J_{n,n+1}(t) = p\,\frac{P_n(t)^2}{P_n(t)+P_{n-1}(t)} \,-\,
487: (1-p)\frac{P_{n+1}(t)^2}{P_{n+1}(t)+P_{n+2}(t)}
488: \end{equation}
489: %
490: %
491: these equations can be also written as
492: %
493: %
494: \begin{equation}
495: \label{MFJ}
496: \frac{\partial}{\partial t} P_n(t) \;=\;
497: J_{n-1,n}(t) - J_{n,n+1}(t) \,.
498: \end{equation}
499: %
500: %
501: Obviously, they conserve probability $\sum_{k=-\infty}^{+\infty} P_k(t)$
502: so that the integrated probability distribution
503: %
504: %
505: \begin{equation}
506: Q_n(t) := \sum_{k=-\infty}^{n} P_k(t)
507: \end{equation}
508: %
509: %
510: satisfies the simple evolution equation
511: %
512: %
513: \begin{equation}
514: \frac{\partial}{\partial t} Q_n(t) \;=\; -J_{n,n+1}(t) \,.
515: \end{equation}
516: %
517: %
518: By construction these mean field equations reflect both the one-dimensional
519: structure as well as the restriction but they ignore spatial correlations between
520: the segments. The full model does exhibit such correlations, but it evolves towards
521: a trivial state without correlations (corresponding to a simple
522: product state in the ASEP). Although this trivial state is never reached
523: in an infinite system, it may explain why the mean field equations proposed
524: here reproduce so many of the observed phenomena faithfully, some of them
525: even exactly, as will be shown in the following sections.
526:
527: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
528: \section{Exact solution of the mean field equations}
529: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
530:
531: Let us first consider the case of a free interface, where the height index $n$
532: runs over all integers from $-\infty$ to $+\infty$.
533:
534: As the KPZ equation is invariant under appropriate rescaling of
535: space, time and height \cite{Barabasi}, we can carry out the continuum limit by
536: introducing a new height variable $h=n\Delta$, where $\Delta$ is the new
537: height unit of the rescaled system.
538: In order to investigate the asymptotic properties of the roughening processes let us assume that
539: $P_n(t)$ varies only slowly with $n$
540: and expand the r.h.s. of Eq.~(\ref{MF})
541: as a Taylor series around $h$. Keeping contributions up to fourth order
542: in $\Delta$ we obtain the partial differential equation
543: %
544: %
545: \begin{eqnarray}
546: \label{PDMF}
547: \frac{\partial}{\partial t} P(h,t)&=&
548: \frac{\Delta(1-2p)}{2} P'(h,t) + \\
549: &&\frac{\Delta^3(2p-1)}{24}\left[\frac{3P'(h,t)^3}{P(h,t)^2}-
550: \frac{6P'(h,t)P''(h,t)}{P(h,t)}+
551: 4P'''(h,t)\right] +
552: \nonumber\\
553: &&\frac{\Delta^4}{8}\,\,\biggl[\frac{2P'(h,t)^4}{P(h,t)^3}-
554: \frac{5P'(h,t)^2P''(h,t)}{P(h,t)^2}+ \nonumber\\
555: && \hspace{9mm}
556: \frac{2P''(h,t)^2}{P(h,t)}+\frac{2P'(h,t)P'''(h,t)}{P(h,t)}-P''''(h,t)\biggr] \nonumber\,,
557: \end{eqnarray}
558: %
559: %
560: where the prime stands for a partial derivative with respect to $h$. Obviously
561: the leading term
562: of order $\Delta$ on the r.h.s. generates a uniform propagation of the
563: probability distribution, hence, the average height
564: $\langle h(t) \rangle$ of the interface will asymptotically grow with the
565: linear velocity
566: %
567: %
568: \beq
569: \label{vel}
570: v=\Delta\left(p-\frac12\right)
571: \eeq
572: %
573: %
574: plus some sublinear correction terms.
575: Assuming ordinary Family-Vicsek scaling~\cite{FVScaling},
576: it is therefore near at hand to test the validity of the scaling form
577: %
578: %
579: \begin{equation}
580: \label{ansatz}
581: P(h,t) \;=\; t^{-\gamma} \, f\Bigl( \frac{h-vt}{t^{\gamma}}\Bigr)
582: \end{equation}
583: %
584: %
585: which -- by definition -- conserves the integrated probability
586: $\int_{-\infty}^{+\infty} \text{d}h P(h,t)$.
587: Notice that the normalization of the height probability distribution
588: implies the scaling function $f(z)$ to be normalized as well.
589: In what follows we solve Eq.~\toref{PDMF} both in the equilibrium case
590: $p=1/2$ and the non--equilibrium case $p\neq 1/2$ confirming that
591: our results do not depend on $\Delta$. In particular, we will show that
592: higher order terms appearing in the expansion of Eq. \toref{MF}
593: turns to be irrelevant, vanishing in the asymptotic limit $t \to \infty$.
594: The correct asymptotic behavior of Eq. \toref{MF} will be therefore recovered by setting
595: $\Delta =1$.
596:
597: %--------------------------------------------------------------------------------
598: \subsection{Equilibrium roughening of a free interface}
599: %--------------------------------------------------------------------------------
600: %
601: %
602: We start analyzing the special case $p=1/2$, where the dynamic processes of
603: the full model are known to exhibit detailed balance. In this case the velocity
604: $v$ is zero and the first-order and third-order contributions
605: on the r.h.s. of Eq.~\toref{PDMF} vanish.
606: Inserting the Ansatz~\toref{ansatz} into Eq.~(\ref{PDMF}) we find that,
607: up to fourth order, the partial differential equation reduces to a non-trivial
608: ordinary differential equation for the scaling function
609: if and only if $\gamma_{\rm eq}=1/4$ (the subscript denoting the equilibrium case).
610: This is exactly the value predicted by the EW theory for equilibrium roughening.
611: Moreover one easily notices that by fixing $\gamma_{\rm eq}=1/4$
612: higher order terms $O(\Delta^5)$ occurring in the Taylor expansion
613: of Eq.~\toref{MF} are irrelevant in the
614: asymptotic limit $t \to \infty$. The differential equation therefore reads
615: %
616: %
617: \begin{eqnarray}
618: \label{EW1}
619: &&\frac{1}{f_{\rm eq}(z)^3}\,\biggl[f_{\rm eq}(z)^4 -\frac{5}{2}\Delta^4 f_{\rm eq}(z)f_{\rm eq}'(z)^2f_{\rm eq}''(z)+\\
620: &&\hspace{15mm}\Delta^4 f_{\rm eq}(z)^2 \Bigl(f_{\rm eq}''(z)^2 +f_{\rm eq}'(z)f_{\rm eq}'''(z)\Bigr) +
621: \nonumber \\
622: &&\hspace{15mm}f_{\rm eq}(z)^3\Bigl(zf_{\rm eq}'(z)-\frac{\Delta^4}{2}f_{\rm eq}''''(z)\Bigr) +
623: \Delta^4 f_{\rm eq}'(z)^4 \biggr] = 0\,,\nonumber
624: \end{eqnarray}
625: %
626: %
627: where $z=h\,t^{-\gamma_{\rm eq}}$ denotes the scaling variable. Integrating both sides we obtain
628: %
629: %
630: \begin{equation}
631: \label{EW2}
632: zf_{\rm eq}(z) - \frac{\Delta^4}{2} \left[\frac{f'_{\rm eq}(z)^3} {f_{\rm eq}(z)^2}
633: - 2 \frac{f'_{\rm eq}(z) f''_{\rm eq}(z)}{f_{\rm eq}(z)}
634: + f'''_{\rm eq}(z)\right] = 0\,.
635: \end{equation}
636: %
637: %
638: This equation admits the two simple solutions
639: %
640: %
641: \beq
642: f^{\rm free}_{\rm eq}(z) =
643: \frac{1}{\Delta 2^{1/4} \sqrt{\pi}} \exp \left(-\frac{z^2}{\Delta^2 \sqrt{2}} \right)
644: \label{EW2free}
645: \eeq
646: %
647: %
648: and
649: %
650: %
651: \beq
652: f^{\rm bound}_{\rm eq}(z) =
653: \frac{2^{1/4}}{\Delta^3 \sqrt{\pi}} \,z^2\,\exp \left(-\frac{z^2}{\Delta^2 \sqrt{2}} \right)
654: \label{EW2bound}
655: \eeq
656: %
657: %
658: which have been normalized over the real line.
659: The first solution $f^{\rm free}_{\rm eq}$ is a simple Gaussian and
660: represents the physical solution for a free interface
661: starting with a flat initial condition $h(x,t)=0$.
662: The second solution $f^{\rm bound}_{\rm eq}$ is characterized by two different
663: maxima over the real line and is therefore dismissed as unphysical
664: in the free case. However, as we will see in Sect. 4, this solution becomes physically
665: meaningful in the presence of a hard-core wall at zero height.
666: %In addition, there exists a third solution which behaves as $f(z) \sim z^3$ as $z \to 0$
667: %and is therefore unphysical in the free case (due to the restriction
668: %$f(z)\geq 0 \quad \forall z$).
669:
670: To summarize we note that the mean field equation for $p=1/2$ does indeed capture
671: the features of one-dimensional EW roughening
672: in the thermodynamic limit $L\to\infty$.
673:
674:
675: %--------------------------------------------------------------------------------
676: \subsection{Nonequilibrium roughening of a free interface}
677: %--------------------------------------------------------------------------------
678:
679: We now turn to the nonequilibrium case $p\neq1/2$. Inserting again the
680: scaling form~\toref{ansatz} and the expression for the velocity~\toref{vel}
681: into the partial differential equation~\toref{PDMF}, we find that,
682: up to {\em third} order, the partial differential equation reduces to a
683: non-trivial ordinary differential equation for the scaling function
684: (i.e., without explicit occurance of $t$)
685: if and only if $\gamma=1/3$. The differential equation then reads
686: %
687: %
688: \begin{eqnarray}
689: \label{NonlinDE}
690: &&\frac{1}{f(z)^2} \Bigl[8 f(z)^3 + 3 k f'(z)^3 - 6 k f(z)f'(z)f''(z) + \\
691: && \hspace{12mm} 4 f(z)^2 \bigl(2z\,f'(z) + k f'''(z)\bigr)\Bigr] \;=\; 0\,,\nonumber
692: \end{eqnarray}
693: %
694: %
695: where
696: $
697: z = (h-vt)\,t^{-\gamma}
698: $
699: denotes the scaling variable in the comoving reference frame and
700: %
701: %
702: \begin{equation}
703: k=(2p-1)\Delta^3\neq 0\,.
704: \label{metricf}
705: \end{equation}
706: %
707: %
708: As in the equilibrium case the postulate of Family-Vicsek scaling
709: applied to the mean field equation already determines the roughening exponent.
710: Remarkably the value $\gamma=1/3$ coincides exactly with the known value
711: of a KPZ process in 1+1 dimensions.
712:
713: As can be verified easily, upon fixing $\gamma=1/3$,
714: the fourth order terms (and all higher order terms)
715: of the Taylor expansion turn out to be irrelevant in the asymptotic limit
716: $t \to \infty$ and thus generate only short time corrections to the scaling function.
717: It is also worth commenting that asymptotic EW scaling can only be seen in the
718: symmetric case $p=1/2$. For any small deviation from this value the third-order
719: terms do not vanish, leading eventually to a crossover to KPZ scaling in the
720: limit $t \to \infty$. Therefore, the mean field equations
721: nicely reproduce the character of the KPZ nonlinearity as a relevant perturbation.
722:
723: Assuming that $f(z) \neq 0$ and integrating both sides
724: of Eq.~(\ref{NonlinDE}) one obtains a simplified equation
725: %of the form
726: %
727: %
728: %\begin{equation}
729: %8 z f(z) - \frac{3 k f'(z)^2}{f(z)} + 4kf''(z)\;=\;0
730: %\end{equation}
731: %
732: %
733: which, by substituting $f(z)=u(z)^4$, can be further reduced to a simple Airy differential equation
734: %
735: %
736: \begin{equation}
737: zu(z)+2ku''(z)\;=\;0
738: \label{Airy}
739: \end{equation}
740: %
741: %
742: with the general solution
743: %
744: %
745: \begin{equation}
746: \label{generalsolution}
747: u(z) = \left\{
748: \begin{array}{cc}
749: c_1 \text{Ai}\Bigl(\frac{-z}{(2k)^{1/3}}\Bigr) +
750: c_2 \text{Bi}\Bigl(\frac{-z}{(2k)^{1/3}}\Bigr)
751: &
752: \quad \text{ for } k \geq 0 \\[3mm]
753: c_1 \text{Ai}\Bigl(\frac{z}{(-2k)^{1/3}}\Bigr) +
754: c_2 \text{Bi}\Bigl(\frac{z}{(-2k)^{1/3}}\Bigr)
755: &\quad \mbox{ for } k < 0
756: \end{array}
757: \right.
758: \end{equation}
759: %
760: %
761: where $\text{Ai}(z)$ and $\text{Bi}(z)$ are Airy functions (see for
762: instance \cite{Abramowitz}).
763: For given $|k|$ the two solutions differ only by a reflection $z\to -z$ so
764: that for the rest of this section we can
765: restrict ourselves to the case $k>0$ (which corresponds to a negative
766: nonlinear term, i.e. $\lambda < 0$).
767:
768: The two integration constants have to be chosen such that $f(z)$ is properly normalized
769: and the appropriate boundary conditions are satisfied. For a free interface,
770: the scaling function $f(z)$
771: has to vanish for $z\to \pm \infty$ in such a way that all of its moments are
772: finite, hence $c_2=0$. Surprisingly, the remaining solution
773: {\em oscillates} for $z \to \infty$ and does not vanish fast enough to yield
774: finite moments.
775: We conclude that the physically meaningful
776: solution extends from $z=-\infty$ to the first root of the Airy function $z_0$
777: (where $f(z_0)=f'(z_0)=f''(z_0)=0$) and
778: vanishes elsewhere. The solution for the free interface therefore reads
779: %
780: %
781: \begin{equation}
782: \label{freesolution}
783: f(z) \;=\;
784: \left\{
785: \begin{array}{ll}
786: \frac{1}{\mathcal N}\,\left[\text{Ai}\Bigl(\frac{-z}{(2k)^{1/3}}\Bigr) \right]^4
787: &\quad \text{ for } -\infty < z < z_0 \\[3mm]
788: 0 & \quad \text{ for } \,\, z_0 \leq z < +\infty
789: \end{array}
790: \right.
791: \end{equation}
792: %
793: %
794: where $z_0 \simeq 2.94583 \, k^{1/3}$ and ${\mathcal N} \simeq 0.127153 \,
795: k^{1/3}$. As it can
796: be seen, the parameter $k$, describing the strength of the KPZ nonlinearity,
797: appears here as a simple metric factor in the scaling function. Notice that
798: by Eq. \toref{metricf} the height unit $\Delta$ has been absorbed in $k$.
799:
800: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
801: \begin{figure}
802: \centerline{\includegraphics[width=75mm]{profile.eps}}
803: \caption{\label{figfreeprofile} \small
804: Rescaled skewed profile obtained within the mean field approximation
805: (solid line) compared to the numerically determined KPZ profile
806: using the single step model with $p=1$.
807: Differences between the two profiles can be highlighted by
808: their mean value $a_1$. The mean value controls the KPZ excess velocity and
809: it is known to scale as $\lambda^{1/3}$.
810: While the mean field profile is characterized by $a_1 \simeq 1.13184$,
811: direct numerical estimate renders $a_{1,SSM} \simeq 0.60$.
812: The vertical axis is plotted in a logarithmic scale.
813: }
814: \end{figure}
815: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
816:
817: Figure~\ref{figfreeprofile} shows the solution~(\ref{freesolution}) in comparison with
818: the numerically determined profile of a freely roughening KPZ interface. Although the
819: two curves are different due to the approximative character of the mean field theory,
820: they share essential qualitative features. To quantify those it is instructive to
821: compute the skewness
822: %
823: %
824: \begin{equation}
825: S = \frac{c_3}{c_2^{3/2}} = \frac{a_3-3a_2a_1+2a_1^3}{(a_2-a_1^2)^{3/2}} \quad,
826: \end{equation}
827: %
828: %
829: where $c_n$ is the $n^{\rm th}$ moment of the height probability distribution and
830: $a_n=\int_{-\infty}^{z_0} dz\,z^n f(z)$ denotes the $n^{\rm th}$ central
831: moment of the rescaled
832: profile $f(z)$. The skewness is expected to be
833: universal in modulo for all KPZ growth processes, with
834: its sign being equal
835: to the sign of the nonlinear term.
836: Known numerical estimates \cite{KrMHH92} give the value
837: $|S_{\rm KPZ} | = 0.28 \pm 0.04$,
838: which is in good agreement with the theoretically computed skewness for the
839: PNG model\footnote{The PNG model is characterized by a positive nonlinear term.}
840: \cite{Prahofer2000},
841: %
842: %
843: \beq
844: S_{PNG} \simeq 0.2935 \ .
845: \eeq
846: %
847: %
848: Mean field theory, on the other hand, renders the value
849: %
850: %
851: \begin{equation}
852: S_{\rm MF} \simeq \pm 0.465970 \,,
853: \end{equation}
854: %
855: %
856: where the positive (negative) sign correspond to the case $k<0$ ($k>0$).
857: Although this value is different from direct numerical estimates,
858: it has the correct sign and the same order of magnitude, showing that the mean field
859: theory captures qualitatively the influence of the nonlinear KPZ term.
860:
861: Surprisingly, the mean field theory predicts that the interface profile is
862: asymptotically bounded for $z>0$ at a {\it finite} value $z_0$. This means that
863: within mean field the advancing front of the distribution exhibits a sharp
864: cutoff rather than a stretched exponential tail. However, on the opposite side,
865: where $z$ is negative, the profile does indeed decay as a stretched exponential:
866: %
867: %
868: \begin{equation}
869: f(z) \sim |z|^{-1/4} \, e^{-\frac{2}{3\sqrt{2k}} |z|^{3/2}}
870: \qquad (z \to -\infty)
871: \end{equation}
872: %
873: %
874: This result suggests that $\eta_+=3/2$, which coincides with the theoretical
875: value predicted in the context of directed polymers and of the PNG model.
876: On the other hand, $\eta_-$ does not exist within the
877: mean field approximation, which therefore fails to correctly describe large
878: negative (w.r.t. the sign of the KPZ nonlinearity) height fluctuations.
879:
880: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
881: \section{Roughening in the vicinity of a wall}
882: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
883: %
884: %
885: We now modify the single step model and the associated
886: mean field equations in order to incorporate the presence of
887: a hard core wall. Our aim is to determine
888: the surface exponent $\theta$ introduced in
889: Eq.~(\ref{theta}) within the mean field approximation.
890: In terms of the continuous height variable $h$,
891: the density of pinned sites can be defined as the integral
892: of the height probability distribution between the hard core wall and some
893: arbitrary small height level $h_0$, i.e.,
894: %
895: %
896: \beq
897: \rho(t) = \int_{vt }^{h_0+ vt } P(h,t)\,dh
898: = \int_{0}^{h_0 t^{-\gamma}}
899: f(z) \,dz \quad,
900: \label{pippo}
901: \eeq
902: %
903: %
904: so that the surface exponent is completely determined by the behavior of the
905: scaling function $f(z)$ for $z\ll 1$.
906:
907: In order to formulate the appropriate boundary condition in the
908: continuum limit, one has to resort to the discrete formulation of the problem.
909: Following the approach outlined in Ref. \cite{Ginelli03},
910: the wall is initially located at zero height
911: and moves {\it discontinuously}
912: with the average velocity $v=p-\frac12$, its actual height being given by
913: $n_0(t)=\lfloor vt \rfloor$ (where $\lfloor \cdot \rfloor$
914: indicates the integer part).
915: The interface is restricted to evolve above
916: the wall, i.e.
917: %
918: %
919: \begin{equation}
920: P_n(t)=0 \qquad (n \leq n_0(t))
921: \end{equation}
922: %
923: %
924: so that the mean field equations \toref{MF} have to be modified accordingly
925: close to the wall. In particular there is no probability
926: current between level $n_0$ and $n_0+1$, so that
927: %
928: %
929: \begin{equation}
930: \label{MFJbnd}
931: \frac{\partial}{\partial t} P_{n_0+1}(t) \;=\; - J_{n_0+1,n_0+2}(t) \quad,
932: \end{equation}
933: %
934: %
935: while Eq. \toref{MFJ} still holds for $n > n_0+1$.
936: Depending on $p$ one has to distinguish three different cases.
937: If $v>0$ the wall advances by one unit in time intervals $\Delta t = 1/v$, flipping
938: up all local minima at level $n_0+1$ by two units. This means that $P_{n_0+1}$ is
939: increased by $P_{n_0}$ while $P_{n_0}$ is set zero. If $v<0$ the wall retracts
940: by one unit in time intervals $\Delta t = |1/v|$, allowing height level $n_0$, which
941: was previously set to zero, to become nonzero during the subsequent evolution.
942: Finally, for $v=0$ the wall does not move, i.e. $n_0=0$ for all times $t$.
943: The moving wall makes it difficult to specify the correct boundary conditions,
944: so out of equilibrium we will derive them in the special cases $p=1$ and
945: $p=0$, where the KPZ nonlinearity is maximal. Our reasoning, which once more
946: relies on a series expansion in the proximities of the wall, shows that
947: both a pushing ($p=1$) and a retracting ($p=\frac12$ wall impose a Dirichlet
948: boundary condition for the scaling function. Surprisingly, it turns out that
949: a retracting wall ($p=0$) does not fix any boundary condition for the scaling
950: function, which is free to assume any finite value at wall level, thus
951: justifying the high density of pinned sites which is numerically observed
952: in the case $\lambda>0$ (see Fig. \ref{figwall}).
953: General scaling arguments suggest that results obtained for $p=1$ ($p=0$) hold for any
954: $p>1/2$ ($p<1/2$).
955:
956:
957: %--------------------------------------------------------------------------------
958: \subsection{Depinning in the equilibrium case $p=1/2$}
959: \label{EWdepin}
960: %--------------------------------------------------------------------------------
961:
962: In the equilibrium case we have $v=0$ so that the wall does not move.
963: As shown in the Appendix, a wall at zero height
964: imposes a Dirichlet boundary condition $f(0)=0$.
965: Obviously, the only solution satisfying this boundary condition is Eq.~\toref{EW2bound}
966: %
967: %
968: \beq
969: f^{\rm bound}_{\rm eq}(z) =
970: \frac{2^{5/4}}{\Delta^3 \sqrt{\pi}} \,z^2\,\exp \left(-\frac{z^2}{\Delta^2 \sqrt{2}} \right)
971: \label{EW2boundnormalized}
972: \eeq
973: %
974: %
975: which has been normalized here over the positive real axis. With this solution
976: we can immediately read off the surface exponent from Eq. \toref{pippo},
977: %
978: %
979: \begin{equation}
980: \theta^{\rm MF}_{\rm EW} = 3/4.
981: \end{equation}
982: %
983: %
984: We note that this value coincides exactly with the known exponent for
985: Edwards Wilkinson growth next to a wall.
986: %
987: %--------------------------------------------------------------------------------
988: \subsection{Depinning in the non-equilibrium case $p=1$}
989: \label{KPZdepin}
990: %--------------------------------------------------------------------------------
991: %
992: %
993: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
994: \begin{figure}
995: \centerline{\includegraphics[width=120mm]{withwall.eps}}
996: \caption{\label{figwithwall} \small
997: Rescaled mean field profiles in the presence of a hard core wall compared
998: to the corresponding profiles in the SSM for $p=1$ (left panel)
999: and $p=0$ (right panel).
1000: The vertical axis have been plotted in a logarithmic scale.
1001: }
1002: \end{figure}
1003: %++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1004: %
1005: For $p>1/2$ the wall advances discontinuously which makes it more difficult to
1006: specify the boundary conditions. As shown in the Appendix, in this case
1007: the co-moving wall again leads to a Dirichlet
1008: boundary condition $f(0)=0$ for the scaling function. According to
1009: Eq.~(\ref{generalsolution}) the corresponding solution then reads
1010: %
1011: %
1012: \begin{equation}
1013: \label{uppersolution}
1014: f(z) \;=\;
1015: \left\{
1016: \begin{array}{ll}
1017: \frac{1}{\mathcal N}\,\left[\text{Ai}\Bigl(\frac{-z}{(2k)^{1/3}}\Bigr) -
1018: \frac{1}{\sqrt{3}} \text{Bi}\Bigl(\frac{-z}{(2k)^{1/3}}\Bigr)\right]^4
1019: &\quad \text{ for } 0 \leq z < z_1 \\[3mm]
1020: 0 & \quad \text{ for } \,\, z_1 \leq z < +\infty
1021: \end{array}
1022: \right.
1023: \end{equation}
1024: %
1025: %
1026: where $z_1 \simeq 3.32426 \, k^{1/3}$ is the first positive root of
1027: the scaling function $f(z)$ (where also its first three derivatives vanish)
1028: and ${\mathcal N} \simeq 0.133454 \, k^{1/3}$.
1029: As in the free case the profile exhibits a sharp cutoff (see Fig.~\ref{figwithwall}),
1030: although at a different value of $z$. Since $f(z) \sim z^4$ for $z \to 0$,
1031: the surface exponent is given by
1032: %
1033: %
1034: \begin{equation}
1035: \theta^{\rm MF}_{\rm p=1} = 4/3,
1036: \end{equation}
1037: %
1038: %
1039: This values has to be compared with the numerical estimate
1040: $\theta^{\rm KPZ}_{\lambda{<}0} = 1.1(1)$ in Eq.~(\ref{theta}).
1041:
1042: %--------------------------------------------------------------------------------
1043: \subsection{Depinning in the non-equilibrium case $p=0$}
1044: %--------------------------------------------------------------------------------
1045: %
1046: For $p<1/2$ the wall moves discontinuously backward.
1047: As shown in the Appendix, this situation is special in so far as the retracting
1048: wall does not specify any boundary condition on the scaling function, allowing $f(0)$
1049: to be nonzero. In fact, according to Eq.~(\ref{generalsolution}) the only
1050: normalizable solution is given by
1051: %
1052: %
1053: \begin{equation}
1054: \label{lowersolution}
1055: f(z) \;=\;
1056: \frac{1}{\mathcal N}\,\left[\text{Ai}\Bigl(\frac{z}{(2k)^{1/3}}\Bigr)\right]^4
1057: \end{equation}
1058: %
1059: %
1060: with the normalization constant ${\mathcal N} \simeq 0.00584355 \, k^{1/3}$.
1061: Since $f(0)>0$ the surface exponent is given by
1062: %
1063: %
1064: \begin{equation}
1065: \theta^{\rm MF}_{\rm p=0} = 1/3,
1066: \end{equation}
1067: %
1068: %
1069: This values has to be compared with the numerical estimate
1070: $\theta^{\rm KPZ}_{\lambda{>}0} = 0.22(2)$ in Eq.~(\ref{theta}).
1071:
1072: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1073: \section{Conclusions}
1074: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1075: In this paper we presented a mean field theory for 1+1 dimensional nonlinear
1076: growth processes evolving according to a KPZ equation.
1077: It is worth stressing that our approach does not neglect
1078: all types of fluctuations, as the term mean field usually suggests.
1079: Instead our equations retain information about the one dimensional
1080: spatial structure and the height restriction $h_i - h_{i+1} = \pm 1$.
1081: Therefore, our mean field theory is not expected to hold
1082: exactly above some upper critical dimension, rather it serves as an
1083: approximation for the one dimensional case. Although we neglect
1084: spatial correlation between local slopes of
1085: the roughening interface, the theory has a very predictive power.
1086: Its success can be ascribed
1087: to a fluctuation dissipation theorem which is known to hold only for the 1+1
1088: dimensional case \cite{Barabasi}.
1089: Moreover, the mean field equations presented herein have been derived
1090: in the thermodynamic limit $L \to \infty $, and thus they are suited for
1091: probing finite time behavior, i.e. $t \ll L^z$.
1092:
1093: Our approach is therefore successful in predicting a power law decay for the
1094: density of interfacial sites pinned to the substrate, thus supporting previous
1095: numerical studies of the non--equilibrium case. Although the mean field
1096: surface exponents $\theta$ differ from the numerically estimated values,
1097: our theory correctly reproduces the dramatic difference between the surface
1098: behavior in the presence of a negative or positive nonlinear KPZ term.
1099:
1100: For what concerns the finite--time bulk properties of a KPZ interface, our
1101: simple mean field theory correctly reproduces the exact value for the
1102: roughening exponent $\gamma=1/3$ and the skewed nature of the finite time
1103: height probability distribution. We found that one of the two tails
1104: of such a distribution decays as a stretched exponential with exponent
1105: $\eta_+ = 3/2$, thus confirming previous results obtained in the context
1106: of directed polymers and for the PNG model.
1107: While our theory successfully predicts large {\it positive} (w.r.t. to the sign of
1108: the KPZ nonlinearity) height fluctuations, it fails in describing
1109: large {\it negative} ones, exhibiting a sharp cut--off for the negative tail.
1110: It is therefore interesting to note that
1111: a simple scaling argument \cite{PNG} proposed in the context of the PNG model
1112: directly relates $\eta_+$ to the roughening exponent, i.e. $\eta_+ =
1113: 1/(1-\gamma)$, while no corresponding argument can be worked out for
1114: $\eta_-$. This is due to the fact that height fluctuations
1115: with the same sign w.r.t. the KPZ nonlinear term manifest as ``bumps'' (or
1116: ``holes'') which grows laterally, while height fluctuations with the opposite
1117: sign manifest as ``holes'' (or''bumps'') which shrink laterally.
1118:
1119: It is also worth noticing that the mean field theory reproduces correctly
1120: almost all bulk nonlinear critical properties at finite times,
1121: while it gives only approximate results for the surface exponent
1122: $\theta$. This is an indication that the substrate introduces
1123: spatial correlations between local slopes at low height levels.
1124: It is our belief that a detailed study of the SSM with a hard substrate
1125: may eventually lead to the exact analytical knowledge of critical
1126: depinning properties. It would also be interesting to find out whether the
1127: methods introduced in Ref.~\cite{Prahofer2000}
1128: can be applied to the problem of a KPZ interface with a wall.
1129:
1130: Finally, equilibrium results are correctly reproduced as a marginal case, and
1131: small out of equilibrium corrections eventually leads to full KPZ behavior.
1132:
1133:
1134: \noindent
1135: {\bf Acknowledgements}\\
1136: We would like to thank A. Politi for stimulating discussions concerning the
1137: simplest formulation of a mean field equation for a KPZ interface.
1138:
1139: %
1140: %
1141: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1142: \appendix
1143: \section{Boundary conditions in the presence of a hard core wall}
1144: \label{app}
1145: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1146: %
1147: In order to solve the mean field equation \toref{PDMF} when a hard core wall
1148: is imposed, it is necessary to go back to the discrete formulation of
1149: the problem and to consider the
1150: proper evolution equation for the height probability distribution
1151: \toref{MFJbnd} close to the wall.
1152: In the following we analyze the special cases $v=0$ and $v = \pm \frac12$ in
1153: order to derive the corresponding boundary conditions for the scaling function $f(z)$.
1154:
1155:
1156: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1157: \subsection{$p=\frac12, v=0$}
1158: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1159:
1160: In this case the wall does not move and $n_0=0$ for all times $t$.
1161: According to Eq.~(\ref{MFJbnd}) the density of contact points $P_1(t)$ then evolves as
1162: %
1163: %
1164: \begin{equation}
1165: \label{MFJ1}
1166: \frac{\partial}{\partial t} P_1(t) \;=\; - J_{1,2}(t) \;,
1167: \end{equation}
1168: %
1169: %
1170: where
1171: \beq
1172: J_{1,2}(t) = \frac{P_1(t)}{2}-\frac{P_2(t)^2}{2(P_2(t)+P_3(t))}=0.
1173: \eeq
1174: In the limit $t \to \infty$ this equation implies a Dirichlet boundary condition $f(0)=0$.
1175: To see this let us assume that $f(0)\neq0$ with $|f'(0)|<\infty$. Assuming EW scaling
1176: the first three probabilities would be given by
1177: $P_1(t) \simeq P_2(t) \simeq P_3(t) \simeq t^{-1/4} f(0)$ (where we expanded
1178: the scaling function $f$ around $z=0$ keeping only the leading term),
1179: giving rise to a current $J_{1,2}(t) \simeq \frac14 t^{-1/4} f(0)$.
1180: Since the l.h.s. scales as $t^{-5/4}$, Eq.~(\ref{MFJ1}) cannot hold unless $f(0)=0$.
1181:
1182: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1183: \subsection{$p=1, v=\frac12$}
1184: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1185:
1186: For $p=1$ one has $\Delta t = 2$, i.e.
1187: the wall advances by one unit after every second time step.
1188: To find the appropriate boundary condition for $f(z)$ (now assuming KPZ
1189: scaling with $\gamma=1/3$), let us first consider
1190: the continuous temporal evolution between two advancements when the wall is
1191: fixed at some height ${n_0}\lfloor -t_0/2 \rfloor$.
1192: As in the previous case the probabilities
1193: $P_{n}(t)$ with $n \leq n_0$ vanish. According to Eqs. \toref{current}, \toref{MFJ} and
1194: \toref{MFJbnd}, the height probability distribution
1195: at the first two levels above the wall obeys to the differential equations
1196: %
1197: %
1198: \begin{equation}
1199: \label{LowerDGL}
1200: \begin{split}
1201: &\frac{\partial}{\partial t} P_{n_0+1}(t) \;=\; - P_{n_0+1}(t) \\
1202: &\frac{\partial}{\partial t} P_{n_0+2}(t) \;=\; + P_{n_0+1}(t) -
1203: \frac{P_{n_0+2}^2} {P_{n_0+1}+P_{n_0+2}}\,.
1204: \end{split}
1205: \end{equation}
1206: %
1207: %
1208: Let us again suppose that $f(0) \neq 0$ and $|f'(0)| < \infty$, i.e.,
1209: just after the advancement of the wall at time $t_0$ we assume that
1210: to leading order $P_{n_0+1}(t_0) \simeq P_{n_0+2}(t_0) \simeq t_0^{-1/3} f(0) =: c\,(t_0)$.
1211: Iterating the differential equations~(\ref{LowerDGL}) over two
1212: time steps, and by assuming $c(t) \simeq c(t_0)$ for $t_0\leq t \leq t_0+2$,
1213: one obtains $P_{n_0+1}(t_0+2)\simeq c\,(t_0) e^{-2}$.
1214: On the other hand, by numerically solving the differential equation for level $n_0+2$
1215: \beq
1216: P_{n_0+2}(t) \;\simeq\; + c\,(t_0) e^{-(t-t_0)} -
1217: \frac{P_{n_0+2}^2} {c\,(t_0) e^{-(t-t_0)}+P_{n_0+2}}\,
1218: \eeq
1219: one gets
1220: $P_{n_0+2}(t_0+2) \approx 0.596 \, c(t_0)$. At time $t_0+2$ the wall
1221: advances by one unit, i.e. all local minima at height $n_0$ are flipped upwards,
1222: meaning that $P_{n_0+1}(t_0+2)$ is first added to $P_{n_0+2}(t_0+2)$ and then
1223: set to zero. Just after advancement $P_{n_0+2}(t_0+2)\approx 0.732\, c(t_0)$
1224: which is in contradiction the assumption unless $f(0)=0$. Hence for $p=1$
1225: the wall imposes again a Dirichlet boundary condition.
1226:
1227: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1228: \subsection{$p=0, v=-\frac12$}
1229: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1230: %
1231: For $p<0$ the wall retracts by one unit after every second time step.
1232: Let us first consider the continuous temporal evolution between two moves when the wall is
1233: fixed at height ${n_0}=\lfloor -t_0/2 \rfloor$. As usual the probabilities
1234: $P_n(t)$ with $n \leq n_0$ vanish and the he first two levels,
1235: where the height probability distribution is nonzero, evolve according
1236: to the differential equations
1237: %
1238: %
1239: \begin{equation}
1240: \label{UpperDGL}
1241: \begin{split}
1242: &\frac{\partial}{\partial t} P_{n_0+1}(t) \;=\; \frac{P_{n_0+2}^2}{P_{n_0+2}+P_{n_0+3}} \\
1243: &\frac{\partial}{\partial t} P_{n_0+2}(t) \;=\; -\frac{P_{n_0+2}^2}{P_{n_0+2}+P_{n_0+3}}
1244: +\frac{P_{n_0+3}^2} {P_{n_0+3}+P_{n_0+4}}
1245: \end{split}
1246: \end{equation}
1247: %
1248: %
1249: Just after retraction level $n_0+1$ can be visited by the interface for the
1250: first time, hence $P_{n_0+1}(t_0)$ is initially zero and becomes nonzero as time evolves.
1251: Assuming that $f(0) \neq 0$ and $|f'(0)| < \infty$, to leading order
1252: the other probabilities have the
1253: initial values $P_{n_0+2}(t_0) \simeq P_{n_0+3}(t_0)\simeq t_0^{-1/3} f(0) =: c\,(t_0)$.
1254: Iterating the differential equations~(\ref{LowerDGL}) over two
1255: time steps one obtains $P_{n_0+1}(t_0+2)\simeq c\,(t_0)$ while the higher levels remain unchanged.
1256: Thus the almost constant probability distribution in the vicinity the wall is simply
1257: 'extended' to the new level that becomes available by retraction of the wall,
1258: meaning that a nonzero value $f(0) \neq 0$ and $|f'(0)| < \infty$ is consistent
1259: with the equations~(\ref{UpperDGL}). Loosely speaking the wall retracts so quickly
1260: that it does not impose a specific boundary condition, allowing $f(0)$ to take
1261: any positive value. The equations are built in such a way that this value is
1262: simply copied from the following height level.
1263:
1264: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1265: \begin{thebibliography}{99}
1266: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1267:
1268: \bibitem{Dietrich}
1269: S. Dietrich, in {\it Phase Transition and Critical Phenomena}, edited by
1270: C. Domb and J.L Lebowitz (Academic Press, London, Orlando, 1988), Vol 12, p. 1.
1271:
1272: \bibitem{KPZ}
1273: M. Kardar, G. Parisi and Y--C Zhang, Phys. Rev. Lett.
1274: {\bf 56} 889 (1986).
1275:
1276: \bibitem{FVScaling}
1277: F. Family and T. Vicsek, J. Phys. A {\bf 18}, L75 (1985).
1278:
1279: \bibitem{EW}
1280: S.F. Edwards and D.R. Wilkinson, Proc. R. Soc. London A {\bf 381}, 17 (1982).
1281:
1282: \bibitem{Barabasi}
1283: A.-L. Barabasi and H.E. Stanley, {\it Fractal Concepts in Surface Growth}
1284: (Cambridge University Press, 1995).
1285:
1286: \bibitem{Zhang}
1287: T. Halpin-Healy and Y.-C. Zhang, Phys Rep. {\bf 254}, 215 (1995).
1288:
1289: \bibitem{Zhang2}
1290: Y.-C. Zhang, Europhys. Lett. {\bf 9}, 113 (1989).
1291:
1292: \bibitem{Prahofer2000}
1293: M. Pr{\"a}hofer and H. Spohn,
1294: Phys. Rev. Lett. {\bf 84}, 4882-4885 (2000);
1295: Physica A {\bf 279}, 342 (2000);
1296: J. Stat. Phys. {\bf 108} 1071 (2002);
1297: J. Stat. Phys. {\bf 115} 255 (2004).
1298:
1299: \bibitem{PNG}
1300: E. Ben-Naim, A.R. Bishop, I. Daruka and P.L. Krapivsky, J. Phys. A {\bf 31},
1301: 5001 (1998).
1302:
1303: \bibitem{KBM91a}
1304: J.M. Kim, A.J. Bray and M.A. Moore, Phys. Rev. A {\bf 44}, 2345 (1991).
1305:
1306: \bibitem{SSM}
1307: M. Plischke, Z. R\`acz and D. Liu, Phys Rev B {\bf 35}, 3485, (1987).
1308:
1309: \bibitem{Wetting}
1310: H. Hinrichsen, R. Livi, D. Mukamel, and A. Politi,
1311: Phys. Rev. Lett. {\bf 79}, 2710 (1997).
1312:
1313: \bibitem{Livi}
1314: L. Baroni, R. Livi and A. Torcini, Phys Rev. E {\bf 63}, 036226 (2001).
1315:
1316: \bibitem{Pikov}
1317: V. Ahlers and A. Pikovsky, Phys. Rev. Lett. {\bf 88}, 254101 (2002).
1318:
1319: \bibitem{MN1}
1320: Y. Tu, G. Grinstein and M.A Mun\~oz, Phys. Rev Lett. {\bf 78}, 274 (1997).
1321:
1322: \bibitem{MN2}
1323: M.A Mun\~oz, F. de los Santos and A. Achahbar, Brazilian J. of Physics {\bf
1324: 33}, 443 (2003)
1325:
1326: \bibitem{Droz}
1327: M. Droz and A. Lipowski, Phys. Rev. E {\bf 67}, 056204 (2003).
1328:
1329: \bibitem{MNRG}
1330: G. Grinstein, M.A Mun\~oz, and Y. Tu, Phys. Rev Lett. {\bf 76}, 4376 (1996).
1331:
1332: \bibitem{ASEP}
1333: T. M. Ligget, {\it Interacting Particle Systems}, (Springer-Verlag, New York 1985).
1334:
1335: \bibitem{Gwa92}
1336: L.-H. Gwa and H. Spohn, Phys. Rev. Lett. {\bf 68}, 725 (1992).
1337:
1338: \bibitem{Johansson}
1339: K. Johansson, Comm. Math. Phys {\bf 209}, 437 (2000).
1340:
1341: \bibitem{Abramowitz}
1342: M. Abramowitz and I. Stegun, {\it Handbook of Mathematical Functions} 10th
1343: printing (New York: Dover, 1972).
1344:
1345: \bibitem{KrMHH92}
1346: J. Krug, P. Meakin and T. Halpin-Healy, Phys. Rev. A {\bf 45}, 638 (1992).
1347:
1348: \bibitem{Ginelli03}
1349: F. Ginelli, V. Ahlers, R. Livi, D. Mukamel, A. Pikovsky,
1350: A. Politi, and A. Torcini,
1351: Phys. Rev. E {\bf 68}, 065102(R) (2003).
1352:
1353: %\bibitem{Sneppen92}
1354: %K. Sneppen, J. Krug, M. H. Jensen, J. Jayaprakash, and T. Bohr,
1355: %Phys. Rev. E {\bf 46}, R7351 (1992).
1356:
1357: \end{thebibliography}
1358: \end{document}
1359: