1: \documentclass[amssymb]{article}
2:
3: \usepackage{graphicx}
4:
5: %\addtolength {\textwidth} {99pt} \addtolength {\hoffset} {-49pt}
6:
7: \begin{document}
8:
9: \begin{center}
10:
11: {\bf \Large Discrete Laplacian Growth: Linear Stability vs Fractal Formation}\\
12:
13: \vspace{5mm}
14:
15: {\Large Igor Loutsenko, Oksana Yermolayeva}\\
16:
17: \vspace{3mm}
18:
19: ICTP, Trieste, 34014, Italy\\
20:
21: \vspace{5mm}
22:
23: e-mail: loutsenk@maths.ox.ac.uk, oyermola@ictp.it
24:
25: \vspace{5mm}
26:
27: Abstract
28:
29: \end{center}
30:
31: \begin{quote}
32:
33: We introduce stochastic Discrete Laplacian Growth and consider its
34: deterministic continuous version. These are reminiscent respectively
35: to well-known Diffusion Limited Aggregation and Hele-Shaw free
36: boundary problem for the interface propagation. We study correlation
37: between stability of deterministic free-boundary problem and
38: macroscopic fractal growth in the corresponding discrete problem. It turns out
39: that fractal growth in the discrete problem is not influenced by stability
40: of its deterministic version. Using this
41: fact one can easily provide a qualitative analytic description of
42: the Discrete Laplacian Growth.
43:
44: \end{quote}
45:
46: \begin{section}{Introduction}
47:
48: The Laplacian Growth or zero surface tension Hele-Shaw free-boundary
49: problem ( see e.g. \cite{B}, \cite{VE}) describes time evolution of a
50: domain $\Omega=\Omega(t)$ in the complex plane with $z=x+iy$, when the boundary
51: $\partial\Omega$ of the domain is driven by the gradient of a scalar
52: field $\phi=\phi(x,y,t)$ (see Fig. \ref{continuous}). The field is
53: harmonic, except for several singular points ("sources" and "sinks"), in
54: the exterior or the interior of $\Omega$. The corresponding
55: Hele-Shaw problems are referred as exterior and interior
56: respectively. The field $\phi$ vanishes at the domain boundary
57: ("interface") $\phi(\partial\Omega)=0$. Well possedness and
58: stability of the problem depends strongly on types of singularities.
59:
60: For instance, the exterior "sink"-driven Hele-Shaw problem is a
61: problem of expanding a simply-connected domain $\Omega$ whose
62: boundary is driven by the field harmonic in exterior of $\Omega$. It is
63: ill-posed and linearly unstable almost for any initial conditions.
64: On the contrary, the exterior problem for contracting $\Omega$ (which is
65: a "source"-driven time reversal of the above problem) is linearly
66: stable and well posed.
67:
68: A discrete, stochastic version of the exterior Hele-Shaw problem
69: (often called Diffusion Limited Aggregation or DLA,
70: \cite{GL},\cite{H},\cite{WS}) describes formation of a cluster of
71: particles on two dimensional (e.g. square) lattice. No more than one
72: particle can occupy a lattice cite. The particles stick one by one
73: to the cluster. The probability for a particle to occupy a given
74: (next to the cluster) cite is proportional to the value of a lattice
75: harmonic field at that cite. The field has logarithmic "sink"-type
76: singularity at infinity. It vanishes on the cluster and is updated
77: after each cluster increment.
78:
79: The cluster in such a discrete problem is a fractal and one may
80: think that instability and ill-possedness of the deterministic
81: continuous version of such a discrete model (i.e. exterior Hele-Shaw
82: free-boundary flow for expanding domain $\Omega$) is a manifestation
83: of this fact. In other words, the fractal interface of the discrete
84: problem may not be approximated by an analytic boundary curve of
85: its deterministic counterpart.
86:
87: In this article we consider a "mix" of interior and exterior
88: problems for simply-connected bounded domain, when the interface is
89: driven by the scalar field $\phi$ which is harmonic almost everywhere
90: except for the interface $\partial\Omega$ itself, where $\phi$ vanishes, and
91: two logarithmic singular points. One of these points is placed at
92: $z=0$ and other at $z=\infty$. In such systems the growth depends on
93: a parameter $\lambda$ which is a measure of the "mix" between the exterior
94: unstable ($\lambda=0$) and the interior stable ($\lambda=1$) processes,
95: with $\lambda=1/2$ being a "neutral stability" growth.
96:
97: One may expect that the fractal formations are not visible on
98: macroscopic scale for discrete version of such models when
99: $\lambda>1/2$ and when the interface is linearly stable. This turns
100: out to be true only for extreme stability point $\lambda=1$, while the
101: macroscopic fractal formations are present for any $\lambda<1$ in
102: the discrete model.
103:
104: Note that the Discrete Laplacian Growth (DLG) introduced here differs from
105: the Diffusion Limited Aggregation (DLA) by simultaneous consideration of both
106: exterior and interior processes on the lattice defined in a similar way.
107: In the case of DLA one considers a lattice
108: cluster and a complement to it, while in DLG the lattice is divided
109: into interior domain, discrete boundary and exterior domain.
110: Although the law of the cluster growth in pure exterior limit
111: ($\lambda=0$) of DLG is locally different from that of DLA,
112: both models have the same continuous version and belong essentially
113: to the same class.
114:
115: %this difference vanishes in the continuous limit and both models
116: %belong to essentially the same class.
117:
118: In the next section we describe, in details, the continuous version of
119: the growth processes under consideration and perform its linear
120: stability analysis.
121:
122: \end{section}
123:
124: \begin{section}{Continuous Model}
125:
126: \begin{figure}
127: \centering
128: \includegraphics{fig.ps}
129: \caption{Problem setting (left) and the domain area increment along
130: the boundary segment $a<l<b$ during the time interval $(t,t+dt)$
131: (right)} \label{continuous}
132: \end{figure}
133:
134: Let, for simplicity, $\Omega$ be a simply connected, bounded domain
135: in the $z=x+iy$ plane with the point $z=0$ inside the domain. We
136: denote by $\phi_+$ the field $\phi(x,y,t)$ in the interior
137: and by $\phi_-$ in the exterior of the domain respectively (see Fig.
138: \ref{continuous}). Field $\phi_\pm$ is harmonic in the
139: interior/exterior of $\Omega$ except for points $z=0$ and $z=\infty$
140: \begin{equation}
141: \Delta \phi_\pm=0, \quad
142: z\not\in\left\{0,\infty,\partial\Omega\right\}. \label{laplacian}
143: \end{equation}
144: It is continuous across the boundary and vanishes on it
145: \begin{equation}
146: \phi_\pm(z\in\partial\Omega)=0 \label{boundary}
147: \end{equation}
148: The field has logarithmic singularities at $z=0$ and $z=\infty$
149: $$
150: \phi_+\to \frac{A_+}{2\pi}\log|z|, \quad z\to 0, \quad
151: $$
152: $$
153: \phi_-\to\frac{A_-}{2\pi}\log|z|, \quad z\to \infty
154: $$
155: where $A_\pm$ are constants.
156:
157: Consider the situation when the boundary dynamics is governed by
158: gradients of $\phi_\pm$
159: $$
160: v_n=\alpha_+\frac{\partial\phi_+}{\partial
161: n}+\alpha_-\frac{\partial\phi_-}{\partial n},
162: $$
163: where $\alpha_\pm$ are constants, $v_n$ denotes the normal velocity
164: of the boundary
165: $$
166: v_n={\rm Re}(n d\bar z/dt), \quad \bar z=x-iy, \quad
167: z\in\partial\Omega,
168: $$
169: $n=n_x+in_y, |n|=1$ stands for the exterior normal to
170: $\partial\Omega$ and $\partial/\partial n$ denotes the normal
171: derivative at the boundary.
172:
173: Rescaling the harmonic fields $\phi_\pm$ and the time variable $t$
174: $$
175: \phi_+\to-\frac{A_+}{\lambda}\phi_+, \quad
176: \phi_-\to\frac{A_-}{1-\lambda}\phi_-, \quad t\to
177: \frac{\lambda}{\alpha_+A_+}t, \quad
178: $$
179: where
180: $$
181: \lambda=\frac{A_+\alpha_+}{A_-\alpha_-+A_+\alpha_+},
182: $$
183: we rewrite the above dynamic law for the boundary velocity as
184: \begin{equation}
185: v_n=\frac{\partial\phi_-}{\partial n}-\frac{\partial\phi_+}{\partial
186: n} \label{velocity},
187: \end{equation}
188: and field asymptotic now depend on a "stability" parameter $\lambda$
189: as
190: \begin{equation}
191: \phi_+\to \frac{-\lambda}{2\pi}\log|z|, \quad z\to 0, \quad
192: \phi_-\to\frac{1-\lambda}{2\pi}\log|z|, \quad z\to \infty.
193: \label{sources}
194: \end{equation}
195: %respectively.
196:
197: Equations
198: (\ref{laplacian},\ref{boundary},\ref{velocity},\ref{sources})
199: together with the initial condition $\Omega(t=0)$ constitute a
200: free-boundary, initial value problem for $\Omega(t)$.
201:
202: Note, that the case $\lambda=0$ corresponds to the exterior Hele-Shaw
203: problem (with $\phi_+=0$) while $\lambda=1$ refers to the interior Hele-Shaw
204: problem (with $\phi_-=0$, respectively).
205:
206: We are interested in discrete stochastic processes of lattice
207: cluster growth, corresponding to the deterministic problem
208: (\ref{laplacian}-\ref{sources}).
209:
210: Consider first the case
211: of cluster growing from a single particle at origin that
212: corresponds to the circular solution of deterministic model. It is easy
213: to see that such solution is an expanding circle of radius
214: $r(t)=\sqrt{t/\pi}$ and $\phi_\pm=\phi^{(0)}_\pm $, where
215: $$
216: \phi_+^{(0)}=\frac{-\lambda}{2\pi}(\log|z|-\frac{1}{2}\log\frac{t}{\pi}),
217: \quad |z|<r(t), \quad
218: $$
219: $$
220: \phi^{(0)}_-=\frac{1-\lambda}{2\pi}(\log|z|-\frac{1}{2}\log\frac{t}{\pi}),
221: \quad |z|>r(t)
222: $$
223: Let us now study linear stability of this solution, considering
224: small perturbations of the circle $|z|=r(t)$ by the $k$-periodic
225: function
226: $$
227: |z(\theta,t)|=r(t)+\epsilon r^{(k)}(t)\sin(k\theta)
228: $$
229: where $\theta=(0,2\pi)$ is a polar angle on the $z$-plane
230: ($z=|z|\exp(i\theta)$) and $k$ is a positive integer. The fields
231: $\phi_\pm$ satisfying conditions (\ref{laplacian}), (\ref{sources})
232: are of the form
233: $$
234: \phi_\pm=\phi_\pm^{(0)}+\epsilon\phi^{(k)}(t)|z/r(t)|^{\pm
235: k}\sin(k\theta).
236: $$
237: Substituting this and previous equations to (\ref{boundary}) and
238: (\ref{velocity}) in the first order of $\epsilon$-series we get
239: \begin{equation}
240: \frac{dr^{(k)}}{dt}=\frac{1-2\lambda}{2t}kr^{(k)} \label{analysis}
241: \end{equation}
242: Since $k>0$, the perturbations grow in time when $\lambda<1/2$ and
243: vanish in time when $\lambda>1/2$, with $\lambda=1/2$ being the "neutral
244: stability" point.
245:
246: In this article we consider continuous systems labelled by stability
247: parameter $\lambda$ in the range $0\le\lambda\le
248: 1$. The $\lambda=0$ case (exterior Hele-Shaw problem
249: for expanding interior domain) is unstable and
250: ill-posed problem, whose discrete version manifests fractal growth.
251: On the other extreme of the interval, at $\lambda=1$, we have stable and
252: well-posed interior expansion problem.
253:
254: Now it is natural to ask the question: Whether the stability of interior
255: continuous problems (i.e. those at $\lambda>1/2$) may depress the macroscopic (i.e. visible
256: in continuous limit) fractal growth of boundaries
257: of corresponding discrete models?
258:
259: To study this question, we are to introduce discrete analogue of the above problem
260: (\ref{laplacian}-\ref{sources}).
261:
262: \end{section}
263:
264: \begin{section}{Discrete Model}
265:
266: One may think of the free boundary problem
267: (\ref{laplacian}-\ref{sources}) as that of dynamics of an ideal
268: conducting contour $\partial\Omega$ in two-dimensional electric
269: field. The field is created by a Coulomb charge of value $-\lambda$
270: at $z=0$ and unit charge distributed with linear density
271: $\sigma(l,t)$ along the contour. The contour is "an ideal
272: conductor", and this means that the potential (at fixed $t$) is
273: constant along $\partial\Omega$. Here $l$ stands for natural
274: parameter along the contour $\left\{\partial\Omega: z=\zeta(l,t),
275: |d\zeta(l)|=dl\right\}$ and $\oint_{\partial\Omega}\sigma(l,t)dl=1$.
276: The harmonic field $\phi(x,y,t)$ is (modulo coordinate independent
277: function of $t$) the electrostatic potential created by the above
278: charges
279: $$
280: \phi=\frac{-\lambda}{2\pi}\log|z|+\frac{1}{2\pi}\oint_{\partial\Omega}\sigma(l,t)\log|z-\zeta(l,t)|dl
281: $$
282: The gradient of the potential is the electric vector field, that has
283: jump of magnitude $\sigma(l)$ across the boundary. Therefore from
284: (\ref{velocity}) it follows that the normal velocity of the contour
285: equals its linear density
286: $$
287: v_n(l,t)=\sigma(l,t)
288: $$
289: This and the previous equations reduce the free-boundary problem on the
290: plane to a dynamical problem on the contour only. It is easy to see that
291: the density $\sigma$ is non-negative, and the interior domain
292: $\Omega$ expands in time.
293:
294: The domain area increment $dP_{(a,b)}$ along the interface segment
295: $a<l<b$ during the time $dt$ (see Fig. \ref{continuous})
296: \begin{equation}
297: \frac{dP_{(a,b)}}{d
298: t}=\int_{a}^{b}v_n(l)dl=\int_{a}^{b}\sigma(l)dl=q_{(a,b)}
299: \label{probability}
300: \end{equation}
301: is proportional to the electric charge $q_{(a,b)}$ concentrated on
302: this segment. This fact suggests to consider the boundary charge as
303: the cluster increment probability in discrete analogue of the above
304: model.
305:
306: \begin{figure}[htp]
307: \includegraphics{fig1.ps}
308: \caption{An example of the cluster growth. Black squares
309: ("particles") belong to the cluster $\Omega$, white squares with
310: gray sides form the cluster boundary $\partial\Omega$. Black dots
311: denote charged (boundary) cites and gray dots are uncharged lattice
312: cites.}\label{explaincluster}
313: \end{figure}
314:
315: Let us now take a square lattice, paint elementary square cells in
316: white color and label centers of squares by a pair of integers
317: $(m,n)$ that are discrete coordinates along the $x$ and $y$
318: directions. Consider the following stochastic process: At the first
319: step, the $(0,0)$ square is painted in black. At the next step we
320: paint in black one of four cites that are next-neighbors of this
321: smallest cluster e.t.c. (see Figure \ref{explaincluster}). This
322: procedure is continued by coloring, at each step, any randomly
323: chosen white next-neighbors of the cluster (i.e. adding randomly a
324: "particle" to the cluster) with the probability described below.
325:
326: We use the same notation $\Omega$ for the cluster as we used for the
327: continuous domain. The cluster boundary, which consists of all white
328: next-neighbors of the cluster, is denoted, by analogy with
329: continuous case, by $\partial\Omega$.
330:
331: At each step of the process we can define a lattice function
332: $\phi_{m,n}$ which satisfies the difference equation
333: \begin{equation}
334: \phi_{n-1,m}+\phi_{n+1,m}+\phi_{n,m-1}+\phi_{n,m+1}-4\phi_{n,m}=-\lambda\delta_{n,0}\delta_{m,0},
335: \quad (n,m)\not\in\partial\Omega \label{lattice_laplacian}
336: \end{equation}
337: everywhere except for the boundary, where it vanishes
338: \begin{equation}
339: \phi_{n,m}=0, \quad (n,m)\in\partial\Omega. \label{lattice_boundary}
340: \end{equation}
341: Its asymptotics is like follows
342: \begin{equation}
343: \phi_{n,m}\to\frac{1-\lambda}{4\pi}\log(m^2+n^2), \quad
344: m^2+n^2\to\infty \label{lattice_sources}
345: \end{equation}
346: The left hand side of (\ref{lattice_laplacian}) is a lattice Laplace
347: operator, and $\delta$ on the right hand side stands for Kronecker
348: delta symbol. Equations
349: (\ref{lattice_laplacian}-\ref{lattice_sources}) are lattice analogues
350: of (\ref{laplacian}, \ref{boundary}, \ref{sources}).
351:
352: As in the continuous case, the potential $\phi_{n,m}$ can be
353: expressed in terms of charges, that are now placed at the boundary
354: cites (see Fig. \ref{explaincluster}).
355:
356: Introducing the lattice Coulomb potential (or the Green function)
357: $G_{n,m}$, a such that
358: $$
359: G_{n-1,m}+G_{n+1,m}+G_{n,m-1}+G_{n,m+1}-4G_{n,m}=\delta_{n,0}\delta_{m,0},
360: \quad -\infty<n,m<\infty
361: $$
362: and
363: $$
364: G_{n,m}\to\frac{1}{4\pi}\log(m^2+n^2), \quad m^2+n^2\to\infty
365: $$
366: one expresses $\phi_{n,m}$ in terms of $q_{i,j},
367: (i,j)\in\partial\Omega$ as
368: $$
369: \phi_{n,m}=\sum_{(i,j)\in\partial\Omega}q_{i,j}G_{n-i,m-j}-\lambda
370: G_{n,m}+C
371: $$
372: where $C$ is a constant.
373:
374: It's easy to see that the boundary charges are nonnegative $q_{i,j}\ge
375: 0$. From (\ref{lattice_laplacian}), (\ref{lattice_sources}) and by the
376: definition of the Green function it follows that
377: \begin{equation}
378: \sum_{(i,j)\in\partial\Omega}q_{i,j}=1. \label{sum}
379: \end{equation}
380: By direct analogy with continuous case (c.f. (\ref{probability})) we
381: now consider $q_{i,j}$ as a probability of the $(i,j)$ square of the
382: boundary $\partial\Omega$ to join the cluster $\Omega$.
383:
384: Therefore, at each step of the growth process we have to solve the
385: linear algebraic system consisting of equation (\ref{sum}) and
386: \begin{equation}
387: \sum_{(i',j')\in\partial\Omega}q_{i',j'}G_{i-i',j-j'}-\lambda
388: G_{i,j}+C=0, \quad (i,j) \in \partial\Omega \label{potential}
389: \end{equation}
390: for unknowns $q_{i,j}, (i,j)\in\partial\Omega$ and $C$, then repaint
391: one of the boundary squares in the black color with the probability
392: $q_{i,j}$.
393:
394: Note, that the lattice function for boundary cites satisfies the following
395: equation
396: $$
397: \phi_{i-1,j}+\phi_{i+1,j}+\phi_{j-1,i}+\phi_{j+1,i}=q_{i,j}, \quad
398: (i,j)\in\partial\Omega
399: $$
400: and the cluster increment rule can be also reformulated in terms of
401: the potential $\phi$ as follows: Probability of the $(i,j)$ boundary cite to
402: join the cluster equals to the sum of potentials on the next-neighbors of
403: that cite.
404:
405: \end{section}
406:
407: \begin{section}{Fractal and Continuous Growth}
408:
409: \begin{figure}[htp]
410: \centering
411: \includegraphics{fig2.ps}
412: \caption{Clusters (left) and their boundaries (right) for the pure
413: exterior ($\lambda=1$) problem (top) and the pure interior problem
414: $\lambda=0$ (bottom). To visualize the time dynamics, we prescribe
415: colors to cluster particles. The color ranges from green to blue,
416: depending on time (or step) at which the particle joined the cluster
417: (green for the initial step, blue for the final
418: step)}\label{extremes}
419: \end{figure}
420:
421: \begin{figure}[htp]
422: \centering
423: \includegraphics{fig3_1.ps}
424: \includegraphics{fig3_2.ps}
425: \caption{Examples of clusters growing from a single particle for
426: $\lambda=0.25$ (top) and $\lambda=0.65$ (bottom)}\label{lambdas}
427: \end{figure}
428:
429: Since in the case of pure exterior $\lambda=0$ problem the model
430: under consideration differs from other discretizations of the
431: Laplacian Growth by local details only, we have to expect the
432: qualitative behavior to be similar to that of Diffusion Limited
433: Aggregation. Indeed, for $\lambda=0$ one observes the pure fractal
434: growth (see Figure \ref{extremes}). Numerical calculations give the
435: following estimate for Hausdorff dimension of fractal boundary of a
436: tree-like cluster (see Appendix for details of numerical
437: calculations)
438: \begin{equation}
439: d=1.64\pm0.02 \label{dimension}
440: \end{equation}
441: Note that in the $\lambda=0$ case the dimension of the boundary
442: coincides with that of the cluster.
443:
444: In the case of the pure interior problem $\lambda=1$, the Discrete
445: Laplacian Growth shows dynamics, close to that of its deterministic
446: continuous counterpart: The cluster boundary is not a fractal, and
447: tends to solution of the continuous problem as its size increases
448: (see Figure \ref{extremes}). Such a behavior was expected due to
449: stability of the interior continuous problem for expanding cluster.
450:
451: Since the continuous problem (\ref{laplacian}-\ref{sources}) is
452: stable for $\lambda>1/2$ (c.f. (\ref{analysis})) one would expect
453: similar (close to deterministic) behavior of Discrete Laplacian
454: Growth for such $\lambda$. This turns out not to be so: Instead, one
455: observes separation of the growth in two fractal components, one of which is
456: quasi-regular and the other is macroscopic, i.e. visible in continuous
457: limit, constituting a finite fraction of the cluster. In the range
458: $0<\lambda<1$, a big (grown from a single particle) cluster consists
459: of a quasi-circular center and branches going out of it (see
460: Figure \ref{lambdas}). The Hausdorff dimension of such cluster
461: boundary equals the same $d$ as in (\ref{dimension}).
462:
463: One can interpret such a behavior in the following way: The discrete
464: growth can be viewed as a competition between the averaged process tending
465: to a continuous limit, and probabilistic fluctuations. Such
466: fluctuations drive the system beyond the linear stability region when
467: $\lambda<1$.
468:
469: \end{section}
470:
471: \begin{section}{Qualitative Analysis}
472:
473: One might try to explain the above fractal properties of growth in
474: the stability region $1/2<\lambda<1$ by the fact that the cluster
475: evolution considered so far starts rather with a single particle than
476: with quasi-continuous
477: macroscopic initial conditions.
478: And indeed, to describe properly the continuous limit we have to
479: start with a cluster of dimension 2, containing $N\to\infty$
480: particles. It is also necessary that, after rescaling the lattice
481: spacing (or cluster linear size) by the factor $N^{-1/2}$, the
482: initial boundary $\partial\Omega(t=0)$ tends to an analytic curve
483: as $N\to\infty$. The time, which is the step number in the discrete
484: model, is rescaled by the factor $1/N$.
485:
486: Such growth process remains, nevertheless, a
487: superposition of smooth (analytic) and macroscopic fractal
488: components in the limit $N\to\infty$: The quasi-analytic part of the
489: boundary $\partial\Omega(t)$ evolves as in the pure interior problem,
490: while the quasi-analytic growth is superposed with the fractal growth of
491: the pure exterior problem. In other words, the "mixed" interior-exterior
492: problem separates into interior and exterior parts in zero-lattice
493: spacing limit.
494:
495: It is easy to find the rate of growth of quasi-analytic to fractal
496: components provided the separation always takes place for
497: $0<\lambda<1$: The tree-like fractal part mainly grows on the tips
498: of the "branches", since the exterior field $\phi_-$ is screened out
499: in fjords between the branches, while the interior field $\phi_+$ is
500: screened out inside the branches. Therefore, in zero-lattice spacing
501: limit, the interior (quasi-analytic) part of the boundary evolves
502: like there were only one logarithmic singularity
503: $$
504: \phi_+\to\frac{- \lambda}{2\pi}\log|z|, \quad z\to 0, \quad {\rm and}
505: \quad \phi_-=0,
506: $$
507: while the fractal part of the curve grows as if
508: $$
509: \phi_-\to\frac{1-\lambda}{2\pi}\log|z|, \quad z\to\infty, \quad {\rm
510: and} \quad \phi_+=0.
511: $$
512: The ratio of growth between the fractal and quasi-analytic parts (in
513: number of particles per unit of time) is
514: $$
515: (1-\lambda)/\lambda.
516: $$
517:
518: It is also straightforward to show that the separation always takes
519: place for $0<\lambda<1$. Let's consider the case when $\lambda$ is
520: close to 1, since the separation of the fractal component here
521: implies the separation for a smaller positive $\lambda$. Suppose
522: again that initial cluster has a big size $N\to\infty$ and its
523: rescaled boundary is close to an analytic curve, so the number of
524: the boundary cites is of order $\sqrt{N}$.
525:
526: Consider now a perturbation (fluctuation) of the boundary. Suppose
527: that the fluctuation linear size (in lattice spacings) is $l$. For
528: $\lambda$ close to $1$, the charge of the particles on the
529: fluctuation tip (i.e. most distant from the quasi-analytic part of
530: the boundary cites of the fluctuation) is of order
531: $(1-\lambda)l^{1-D/2}/\sqrt{N}$, where $1\le D\le 2$. For instance,
532: $D=1$ for one-dimensional, "crack"-like fluctuations of length $l$,
533: while $D=2$ for a "bump"-like 2-dimensional fluctuations of diameter
534: $l$. Charge of particles on the quasi-analytic part of the boundary
535: is of order $1/\sqrt{N}$. A fluctuation tend to grow (rather than
536: collapse) when the probability for a particle to join the cluster at
537: the tip of fluctuation exceeds the probability for the
538: quasi-analytic part of the boundary. Since the probability is
539: proportional to the charge, from the above discussion it follows
540: that fluctuations tend to grow when their linear size $l$ exceeds
541: some critical value which is proportional to
542: $(1-\lambda)^{-\frac{2}{2-D}}$. This value is asymptotically
543: independent of cluster size $N$.
544:
545: Since the critical size of fluctuation is $N$-independent, the
546: possible number of critical fluctuations along the boundary
547: is proportional to the
548: boundary length $\sqrt{N}$. It then follows that starting from the
549: close to analytic boundary, a big cluster develops critical
550: fluctuations in the number of steps $f(\lambda)\sqrt{N}$, where
551: $f(\lambda)$ is some function of $\lambda$ only. This number of
552: steps corresponds to the time $t=f(\lambda)/\sqrt{N}$ in the scaling
553: limit.
554:
555: Therefore, in the $N\to\infty$ scaling limit, the cluster develops
556: critical fluctuations in zero time and the separation of growth into
557: the fractal and continuous components takes place immediately if
558: $0<\lambda<1$.
559:
560: \end{section}
561:
562: \begin{section}{Conclusions}
563:
564: In the present work we have introduced the Discrete Laplacian Growth
565: (DLG) model and studied correlation between the stability of its
566: continuous version and the fractal growth of boundary. The
567: stability, turns out, does not guarantee an absence of the fractal
568: growth. The growth process separates into superposition of
569: "continuous" and fractal component. In other words, in the
570: $N\to\infty$ scaling limit, the discrete growth does not tend to
571: solution of continuous model even if the later is stable and
572: well-posed.
573:
574: There remain a few questions to address:
575:
576: 1) Since, in $N\to\infty$ limit, DLG separates in two independent
577: processes, one may think about implications of such a separation in
578: the continuous model. For instance, whether this separation can be traced
579: (in some, possibly "hidden", form) in continuous model or it is just
580: a consequence of the discretization?
581:
582: 2) It is also known that both, exterior and interior problems possess
583: an infinite number of conserved quantities (harmonic moments) in the
584: scaling $N\to\infty$ limit \cite{R,VE}. Does this imply (in view of the
585: separation) existence of an infinite number of conserved quantities
586: in "mixed" models?
587:
588: 3) In this article we presented numerical estimate (\ref{dimension})
589: for the fractal dimension of DLG. Although DLG differs from the
590: Diffusion Limited Aggregation (DLA) only locally, their fractal
591: dimensions do not coincide. This result is not unexpected, since
592: dimension of DLA is sensitive even to the lattice type (see e.g.
593: \cite{H}) and can not be considered as a general characteristics
594: of a class of models. It would
595: be interesting to understand which quantities are universal, i.e.
596: independent of local details of a discrete scheme.
597:
598:
599: \end{section}
600:
601: \begin{section}{Appendix}
602:
603: Numerical simulations of the Discrete Laplacian Growth have been
604: performed by updating the discrete boundary (presented by a set of pairs
605: of integer coordinates
606: $\left(i(1),j(1)\right)$,$\left(i(2),j(2)\right)$, $\dots,
607: \left(i(K),j(K)\right)$, where $K$ is the current boundary length)
608: with the probability given by solutions of linear system
609: (\ref{sum},\ref{potential}). In this method an explicit solution of
610: Laplace equation at each step is not needed, since the coefficients of
611: Eq. (\ref{potential}) are defined by the coordinates of
612: the boundary cites only.
613:
614: The Green function (or lattice Coulomb potential) $G_{n,m}$ can be
615: computed in several ways. To speed up the calculations, we have used
616: the following approximation for the lattice Green function
617: $$
618: G_{m,n}=\frac{1}{4\pi}\log(m^2+n^2)-\frac{m^4+n^4-6m^2n^2}{24\pi(m^2+n^2)^3}
619: $$
620: when the point $(n,m)$ is in the exterior of the square $-L\le n\le
621: L, -L\le m\le L$, and numerical solution for $G_{n,m}$ in the
622: interior of this square (with the function value at the square
623: boundary given by the above approximation). In this approach, the
624: Green function is represented numerically as (computed once)
625: $2L+1\times 2L+1$ table extended by the above equation. This
626: approximation is very precise: for example, for $L=100$ the maximum
627: deviation from the exact lattice Green function is of order
628: $10^{-10}$.
629:
630: We split solution of the system (\ref{sum},\ref{potential}) in two
631: parts:
632: $$
633: \sum_{k'=1}^KG_{i(k)-i(k'),j(k)-j(k')}q^{\rm ex}_{k'}=1, \quad
634: \sum_{k'=1}^KG_{i(k)-i(k'),j(k)-j(k')}q^{\rm in}_{k'}=G_{i(k),j(k)}
635: $$
636: $$
637: q_k=-C q^{\rm ex}_k+\lambda q_k^{\rm in}, \quad C=-\frac{1-\lambda
638: \sum_{k=1}^K q_k^{\rm in}}{\sum_{k=1}^K q_k^{\rm ex}},
639: $$
640: where $q_k$ stands for $q_{i(k),j(k)}$. The $K\times K$ matrix
641: $G_{i(k)-i(k'),j(k)-j(k')}$ is symmetric and positive definite, so
642: we made use of the rapidly converging conjugate gradient method \cite{KA} for
643: the numerical solution.
644:
645: \end{section}
646:
647: \vspace{5mm}
648:
649: {\bf \Large Acknowledgement}\\
650:
651:
652: The authors would like to acknowledge useful information and help
653: received from Professor V.Kravtsov.
654:
655:
656: \vspace{5mm}
657:
658: \begin{thebibliography} {Bibliography}
659:
660: \bibitem{B} D. Bensimon, L. Kadanoff, S. Liang, B. Shraiman, C. Tang, Viscous flows in
661: two dimensions, Rev. Mod. Phys. 58 (1986) 977.
662:
663: \bibitem{GL} J.Gollub, J. Langer, Pattern formation in non equilibrium
664: physics, Rev. Mod. Phys. 71 (1999) 396-404
665:
666: \bibitem{H} T. Halsey, Diffusion limited aggregation: a model for pattern
667: formation, Physcis Today 53 (2000) 36
668:
669: \bibitem{KA} Kendell A. Atkinson (1988), An introduction to numerical analysis (2nd ed.), Section 8.9, John Wiley and Sons. ISBN 0-471-50023-2
670:
671: \bibitem{R} S. Richardson, Hele Shaw flows with a free boundary produced
672: by the injection of fluid into a narrow channel. J.Fluid Mech., 56,
673: part 4, (1972), pp.609-618
674:
675: \bibitem{VE} A.N. Varchenko and P.I. Etingof, Why the boundary of a round drop becomes a curve of order four, American Mathematical Society, University Lecture Series, 3, (1994)
676:
677: \bibitem{WS} T. Witten and L. Sander, Diffusion limited aggregation: a kinetic
678: critical phenomenon, Phys. Rev. Lett. 47 (1981) 1400
679:
680:
681:
682:
683:
684:
685: \end{thebibliography}
686:
687: \end{document}
688: