1: %\documentstyle[preprint,aps]{revtex}

2: \documentstyle[prl,aps]{revtex}

3: \twocolumn

4: %\documentstyle[aps]{revtex}

5: \begin{document}

6: \input{epsf}

7: \draft

8: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname@twocolumnfalse\endcsname

9: \title{Normal scaling in globally conserved

10: interface-controlled coarsening of fractal clusters}

11: \author{Avner Peleg $^{1}$, Massimo Conti $^{2}$

12: and Baruch Meerson $^{1}$}

13: \address{$^{1}$ Racah Institute of Physics, Hebrew

14: University   of  Jerusalem, Jerusalem 91904, Israel}

15: \address{$^{2}$ Dipartimento di Matematica e Fisica,

16: Universit\'{a} di Camerino, and Istituto Nazionale di Fisica della

17: Materia, 62032, Camerino, Italy}

18: \maketitle

19: \begin{abstract}

20: Globally conserved

21: interface-controlled 

22: coarsening of fractal clusters exhibits dynamic

23: scale invariance and normal scaling. This is demonstrated 

24: by a numerical solution of

25: the

26: Ginzburg-Landau equation with a global conservation law. 

27: The sharp-interface limit of this equation is

28: volume preserving motion by mean curvature. The scaled form of the 

29: correlation 

30: function 

31: has a power-law tail accommodating the

32: fractal initial condition. The

33: coarsening length exhibits normal scaling with time. 

34: Finally, shrinking of

35: the fractal clusters with time is observed. The difference between global and 

36: local conservation is discussed.

37: \end{abstract}

38: \pacs{PACS numbers: 64.60.Ak, 61.43.Hv, 05.45.Df, 05.70.Fh}

39: \vskip1pc]

40: \narrowtext

41: %\pagebreak

42: Dynamics of growth of order from disorder in coarsening

43: systems with long-range

44: correlations

45: is an intriguing problem which appears in 

46: phase ordering \cite{Bray1} and in many other applications. 

47: Systems with long-range correlations are often characterizable by

48: fractal geometry \cite{Fractals,Vicsek}. Therefore, a lot of

49: attention has been devoted recently

50: to a variety of ``fractal coarsening" 

51: problems 

52: \cite{Toyoki,Jullien1,Irisawa1,Irisawa2,Jullien2,Meerson,Conti1,Conti2,Peleg,Crooks,Kalinin,Streitenberger}. 

53: A typical (though not the only)

54: setting for fractal coarsening

55: is the following. At an earlier stage of the dynamics

56: a fractal cluster (FC)

57: develops due to an

58: instability of growth of the ``minority 

59: phase" \cite{Vicsek,Meakin}. When the mass (or heat)

60: source is depleted, coarsening by

61: surface tension

62: becomes dominant. 

63: The main question concerns possible scaling relations and

64: universality classes in this type of

65: coarsening.

66: 

67: A major 

68: simplifying assumption 

69: in the analysis of a coarsening process is 

70: dynamic scale invariance (DSI) 

71: (see review \cite{Bray1}). 

72: The DSI hypothesis was first applied to fractal

73: coarsening by Toyoki and Honda \cite{Toyoki}

74: who considered systems with non-conserved order parameter. 

75: Implications  

76: of (mass) conservation in fractal coarsening

77: were considered more recently \cite{Jullien1,Meerson}. Most 

78: remarkable of 

79: them is predicted shrinking of the FCs in the process of 

80: coarsening. However, there has been no

81: convincing evidence (neither in experiment, nor in simulations)

82: in favor of DSI in conserved fractal coarsening. Moreover,

83: anomalous scaling and breakdown of DSI were observed

84: in recent simulations 

85: of

86: locally conserved edge-diffusion- \cite{Jullien2}

87: and bulk-diffusion-controlled \cite{Conti1,Conti2}

88: fractal coarsening.

89: This communication reports our finding 

90: that

91: DSI and normal scaling hold in the process of {\it interface-controlled}

92: fractal coarsening with a {\it globally} conserved 

93: order parameter. This system is apparently the first realistic

94: conserved fractal coarsening

95: system where this simplifying and beautiful concept is found to work.

96: 

97: Globally conserved interface-controlled coarsening

98: is accessible in experiment. Consider

99: the sublimation/deposition dynamics of a

100: solid and its vapor in a small closed vessel

101: kept at a (constant) low temperature \cite{airplane}. 

102: As the acoustic time in the gas phase is short compared

103: to the coarsening time, the gas pressure

104: (and, consequently, density) remains uniform in space, 

105: changing only in time. This character of mass 

106: transport

107: makes the coarsening dynamics conserved

108: globally rather than locally, which leads to a 

109: different kinetics.

110: Another example

111: appears in the context of attachment/detachment-controlled 

112: nanoscale fluctuations at solid surfaces 

113: \cite{Williams,Zinke-Allmang}. There is also a strong 

114: recent evidence in favor of interface-controlled transport

115: during the cluster coarsening in driven rapid granular 

116: flows \cite{Aranson1}. 

117: 

118: Here is an outline of the rest of the paper.

119: We shall work with

120: the Ginzburg-Landau equation with

121: a global conservation law. The corresponding

122: sharp interface

123: theory is reducible, at large times, to

124: volume preserving motion by mean curvature.

125: Assuming DSI, one can then

126: predict

127: scaling behavior

128: of the 

129: correlation function, 

130: shrinking of FCs with time and normal scaling

131: of the coarsening length $l(t)$. Our extensive numerical

132: simulations of the coarsening of two-dimensional (2d) 

133: diffusion-limited aggregates (DLAs)

134: support all these

135: predictions. We shall

136: conclude by pointing out the main difference between

137: global and local conservation.

138: 

139: We adopt a simple Landau

140: free energy functional:

141: \begin{equation}

142: F[u]=\int\left[(1/2) ({\bf{\nabla}}u)^2+V(u)+Hu\right]\,d^{d}{\bf r}

143: \,,

144: \label{1}

145: \end{equation}

146: where $V(u)=(1/4) (1-u^{2})^{2}$ is a double-well potential, 

147: $u({\bf r},t)$ is the order parameter 

148: and fluctuations are neglected. 

149: The effective ``magnetic field" $H=H(t)$ 

150: changes in time

151: so as to impose a global conservation law: $<u>=\mbox{const}$,

152: where $<...>$ denotes a spatial average:

153: \begin{equation}

154: <...>= L^{-d}\,\int(...)\, d^{d}{\bf r}

155: \,,

156: \label{2}

157: \end{equation}

158: $L$ is the system size and the integration is

159: over the whole system. The dynamics is described by a simple

160: gradient descent:

161: \begin{equation}

162: \frac{\partial u}{\partial t}=-\frac{\delta F}{\delta u}= \nabla

163: ^{2}u+u-u^{3}-H(t)

164: \,.

165: \label{3}

166: \end{equation}

167: It follows from Eq. (\ref{3}) and the conservation law that

168: $H(t)=<u-u^{3}>$ (either no-flux, or periodic boundary conditions

169: are assumed), so Eq. (\ref{3}) is a nonlocal

170: reaction-diffusion equation \cite{Schimansky,Rubinstein,Mikhailov,MS}.

171: In the context of phase ordering it can be called

172: the Ginzburg-Landau equation with a global conservation law.

173: To make a theoretical progress, one should

174: work in the sharp-interface limit \cite{MS}

175: valid at late times, when the system already consists of large domains 

176: of ``phase 1"

177: and ``phase 2" divided by a sharp interface.

178: At this stage $H(t)$ is both 

179: small, $H(t) \ll 1$, and slowly varying in time. 

180: The phase 

181: field in the phases $1$ and $2$ is uniform and rapidly adjusts to the 

182: current value of $H(t)$, so $u=-1-H(t)/2$

183: and $1-H(t)/2$, respectively. For brevity, we will consider

184: 2d-case. The normal velocity of the

185: interface is

186: \begin{equation}

187: v_{n}(s,t)=\frac{3}{\sqrt{2}}H(t)-\kappa(s,t)

188: \,,

189: \label{8}

190: \end{equation}

191: where $s$ is the coordinate along the interface and $\kappa$ is the

192: local curvature of the interface. The positive sign of $v_n$

193: corresponds to the interface moving toward phase $2$, while

194: $\kappa$ is positive when the interface is

195: convex towards phase $2$.

196: 

197: An equation for $H(t)$ follows from the global

198: conservation law which takes the form

199: \begin{equation}

200: \frac{4 A(t)}{L^{2}}-H(t)=\mbox{const}

201: \,,

202: \label{11}

203: \end{equation}

204: where 

205: $A(t)=\int_{u({\bf r},t)>0}d^{2}{\bf r}$

206: is the cluster area.

207: 

208: Eqs. (\ref{8}) and (\ref{11}) make a closed set and provide

209: a general sharp-interface formulation to the Ginzburg-Landau

210: equation with a global conservation law. Often

211: they

212: can be 

213: simplified further. Compute

214: the area loss rate of the cluster:

215: \begin{equation}

216: \dot{A}(t) = \oint v_{n}(s,t)ds = \Lambda(t) \left[

217: \frac{3}{\sqrt{2}}H(t)-\overline{\kappa(s,t)}\right]

218: \,, \label{12}

219: \end{equation}

220: where $\overline{\kappa(s,t)}$ is the curvature averaged over the whole

221: interface:

222: \begin{equation}

223: \overline{\kappa(s,t)}=\frac{1}{\Lambda(t)}\oint \kappa(s,t)ds

224:  \,,

225: \label{13}

226: \end{equation}

227: and $\Lambda(t)$ is the cluster perimeter. If the cluster area is conserved, 

228: then $H(t) = (\sqrt{2}/3)\overline{\kappa(s,t)}$ which yields

229: \begin{equation}

230: v_{n}(s,t)=\overline{\kappa(s,t)}-\kappa(s,t)

231:  \,.

232: \label{14}

233: \end{equation}

234: This is area-preserving motion by curvature (in 2d), or

235: volume-preserving motion by mean curvature (in 3d)

236: \cite{Rubinstein,MS,Gage}. 

237: Dynamics (\ref{14}) shortens the interface

238: length (in 2d), or area (in 3d) 

239: \cite{Rubinstein,Gage}. This nonlocal

240: coarsening model

241: is simpler than the better known

242: ``Laplacian coarsening model" (derivable 

243: from

244: the Cahn-Hilliard equation \cite{Bray1,Pego}) which describes

245: the late-time asymptotics of the

246: locally-conserved bulk-diffusion-controlled coarsening.

247: 

248: Assuming DSI and using Eq.

249: (\ref{14}), one obtains 

250: normal scaling for 

251: the characteristic coarsening length:

252: $l(t)\sim t^{1/2}$. Therefore,

253: global conservation does not change the scaling law.

254: The same result (again, when assuming DSI)

255: follows from dynamic renormalization group 

256: arguments applied to Eq. (\ref{3}) (with a

257: Gaussian white noise term) \cite{Bray2}. 

258: For short-range correlations 

259: this result was supported by

260: particle simulations of critical \cite{simulations} and off-critical

261: \cite{Majumdar} quench, and by

262: a numerical solution of Eq. (\ref{3}) for both critical, and off-critical

263: quench \cite{CMPS}.

264:  

265: Let us return to fractal coarsening. The initial conditions 

266: represent FCs that are

267: characterizable by their fractal dimension $D$ on an

268: interval of scales between the lower cutoff $\tilde{l}_{0}$ and

269: the upper cutoff $\tilde{L}_{0}$. The DSI-based coarsening

270: scenario \cite{Toyoki,Jullien1,Meerson} assumes that the fractal

271: dimension of the cluster remains constant on a

272: shrinking interval of distances between the lower cutoff

273: $\tilde{l}(t)$ (which has the same dynamic scaling as the 

274: coarsening length), and the upper cutoff $\tilde{L}(t)$. Now, the

275: perimeter $\Lambda$ and area $A$ of the FC can be estimated

276: as \cite {Fractals}

277: \begin{equation}

278: \Lambda \sim l(\tilde{L}/l)^{D} \;\;\; \mbox{and} \;\;\; A \sim

279: l^{2}(\tilde{L}/l)^{D}

280:  \,,

281: \label{15}

282: \end{equation}

283: respectively. Then area conservation yields $L\sim

284: l^{(D-2)/D} \sim t^{-(2-D)/2D}$ which implies that the FC

285: shrinks with time \cite{Jullien1,Meerson}. This follows

286: $\Lambda(t)

287: \sim l^{-1}(t) \sim t^{-1/2}$. One can also predict the

288: asymptotic shape of the equal-time pair 

289: correlation function at large times: $C(r,t) \rightarrow g[r/l(t)]$.

290: At

291: distances $r\ll l(t)$ from a typical reference point inside the

292: cluster the correlation function should obey the Porod law:

293: $g(\xi)= 1-k \xi$

294: with a constant $k$ of order unity. At 

295: $l(t) \ll r \ll \tilde{L}(t)$

296: $g (\xi) \sim \xi^{D-2}$ (see Ref. \onlinecite{Meerson}), a

297: power-law tail with the same exponent as in $C(r,t=0)$.

298: 

299: In order to check these predictions, we solved

300: Eq. (\ref{3}) numerically on a domain

301: $2048 \times 2048$ with no-flux boundary conditions. 

302: The accuracy of the numerical scheme was monitored by

303: checking the (approximate) conservation law (\ref{11}) which 

304: was found to hold

305: with an accuracy better than $0.2${\%}

306: for $t>3$.

307: 

308: We used $10$ different

309: DLA clusters \cite {Witten} as the initial conditions.

310: These clusters (like the one shown

311: in Fig. 1, upper left) had a radius of order $10^3$.

312: To prevent fragmentation at an early stage of coarsening, 

313: the clusters were reinforced by an addition of peripheral

314: sites, similar to Ref. \onlinecite{Irisawa1}. The 

315: average fractal dimension of the

316: initial clusters, determined from the averaged pair correlation

317: function, was $1.75$.

318: 

319: Introducing the density $\rho({\bf

320: r},t)=(1/2) [u({\bf r},t)+1]$, we identified 

321: the cluster as the locus where 

322: $\rho({\bf r},t)\geq 1/2$. Typical snapshots of the

323: coarsening process are shown in Fig. 1. One can see that

324: larger features of the FC grow at the expense of smaller ones. 

325: At late times the cluster shrinks, in

326: agreement with the prediction of DSI \cite{shrinking}. A similar

327: shrinking is evident in the pictures obtained in Monte Carlo simulations

328: of area-preserving

329: interface-controlled coarsening \cite{Irisawa2}, although the 

330: authors of Ref. \onlinecite{Irisawa2} did not comment on it.

331: 

332: To characterize the coarsening process, several quantities were

333: sampled and averaged over the $10$ initial conditions:

334: \begin{enumerate}

335: \item The cluster area.

336: \item The (circularly averaged) correlation function,

337: normalized at r=0:

338: \begin{equation}

339: C(r,t)=\frac{<\rho({\bf r'}+{\bf r},t)\rho({\bf

340: r'},t)>}{<\rho^{2}({\bf r'},t)>} \,. \label{31}

341: \end{equation}

342: \item The coarsening length scale $l(t)$, computed

343: from the equation $C(l,t)=1/2$. 

344: $l(t)$ can be

345: interpreted as the typical width of the cluster branches.

346: \item The cluster perimeter $\Lambda (t)$ computed

347: by a standard 

348: algorithm \cite {Parker}.

349: \end{enumerate}

350: \begin{figure}[h]

351: %\vspace{0.5cm}

352: \hspace{-1.0cm}

353: \rightline{ \epsfxsize = 6.5cm \epsffile{f1.eps}}

354: \vspace{0.1in}

355: \caption{

356: Evolution of a DLA cluster undergoing an

357: interface-controlled coarsening in a

358: globally conserved system. The upper row

359: corresponds to $t=0$ (left) and $12.6$ (right), the lower row to

360: $t=126.4$ (left) and $1856.6$ (right).

361: \label{fig. 1}}

362: \end{figure}

363: 

364: The cluster area was found to be constant with an accuracy

365: better than $0.5{\%}$ for $t>10$, and better than $0.15{\%}$ for

366: $t>100$. Hence, area preserving motion by curvature, Eq. (\ref{14}),

367: provides an accurate description to this regime. Figure 2 shows that,

368: at late times ($t>100$),

369: $C(r,t)$ approaches a scaled form. The scaled function

370: has a long-range power-law tail with an exponent $D-2$ 

371: (the same as in the initial condition), see the inset

372: of Fig. 2. Noticeable is the absence of any

373: additional dynamic length scales, in a striking 

374: contrast to the locally conserved

375: fractal coarsening \cite{Conti2}. The dynamics of

376: $l(t)$ is shown in Fig. 3.

377: The same figure shows a pure $t^{1/2}$ power-law line

378: (serving as a

379: reference for the expected late-time

380: behavior) and a corrected power-law fit

381: $l(t)=l_{0}+bt^{\alpha}$ with $\alpha =0.49$, $b=1.2$

382: and $l_{0}=5.0$. 

383: 

384: \begin{figure}[h]

385: \hspace{-2.6cm}

386: \rightline{ \epsfxsize = 6.0cm \epsffile{f2.eps}}

387: \vspace{0.1in} 

388: \caption{Scaling form of the

389: correlation function $C(r,t)$ for time moments

390: $t=400.0$, 587.0, 1264.8 and

391: 1856.6. The inset shows the same data on a log-log

392: plot. The solid line, serving as a reference, represents a

393: power-law with an exponent $D - 2 = - 0.25$.

394: \label{fig. 2}}

395: \end{figure}

396: 

397: \begin{figure}[h]

398: \hspace{-2.6cm}

399: \rightline{ \epsfxsize = 6.0cm \epsffile{f3.eps}}

400: \vspace{0.1in} 

401: \caption{The coarsening

402: length $l(t)$ versus time (circles). The solid line is a

403: corrected power-law fit: $l(t)=l_{0}+bt^{\alpha}$ with

404: $\alpha=0.49$, $l_{0}=5.0$ and $b=1.2$. The dotted line represents a pure

405: $t^{1/2}$ power law. The inset shows the time-dependent

406: effective dynamic exponent $\alpha_{0}(t)$ versus

407: $1/l(t)$. The solid line is a linear fit. 

408: \label{fig. 3}}

409: \end{figure}

410: 

411: The inset of Fig. 3 illustrates a different method \cite{Huse} 

412: of determining the dynamic exponent. One defines

413: a (time-dependent) effective exponent:

414: $\alpha_{0}(t)=d \ln l(t)/d \ln t$. Under the normal scaling 

415: assumption, one can determine the

416: ``true" dynamic exponent by plotting $\alpha_{0}(t)$ versus 

417: $l^{-1}(t)$ and extrapolating it to $t \to \infty$, that is 

418: to $l^{-1}(t) \to 0$. The numerical value of $\alpha_{0}(t)$ is computed

419: from

420: $\alpha_{0}(t)=\log_{10}\left[l(10t)/l(t)\right]$.

421: This procedure yields $\alpha = 0.50$. Therefore, $l(t)$ exhibits normal

422: scaling.

423: 

424: The same procedures were used for an analysis of the dynamic 

425: behavior of the cluster perimeter

426: $\Lambda (t)$. We found that $\Lambda^{-1}(t)$ exhibits the same normal

427: scaling: $\Lambda^{-1}(t)\sim t^{1/2}$. Irisawa {\it et al.}

428: \cite{Irisawa2}

429: reported an exponent $0.38$ for the inverse perimeter.

430: Their graph shows, however, that the 

431: effective

432: exponent increases at late times. We believe that a 

433: careful analysis of their data

434: would also lead to an exponent of $1/2$.

435: 

436: 

437: Thus, all predictions following from the DSI hypothesis: the normal

438: scaling of $l(t)$, shrinking of the FC and scaling behavior

439: of correlation function (including its

440: power-law tail), are confirmed by 

441: numerical simulations. We therefore

442: conclude that globally-conserved interface-controlled fractal

443: coarsening

444: exhibits DSI and normal scaling. This

445: behavior stands in contrast to the breakdown of

446: scale invariance observed in diffusion-controlled coarsening of

447: FCs \cite {Conti1,Conti2}, where the order

448: parameter is conserved {\it locally}. The mechanism of scaling violations

449: in locally-conserved systems is not entirely

450: clear at present, therefore

451: a comparison between the two types of systems can be instructive.

452: We relate the difference in scaling behavior 

453: to an important (and simple) difference in the character of transport. 

454: Global

455: transport, characteristic for interface-controlled

456: systems, is uninhibited by Laplacian screening effects typical for 

457: locally conserved systems.

458: Therefore, large-scale dynamics is always present in globally-conserved

459: systems, but is suppressed in locally-conserved ones.

460: This difference is observed already in a very

461: simpler setting of an area-preserving coarsening (shrinking) 

462: of long slender

463: bars. In the locally-conserved case the

464: bar acquires a dumbbell shape, while its initial width remains

465: (almost) constant and

466: represents a relevant length scale until late times

467: \cite{Conti2}. On the contrary, in the globally

468: conserved case the shrinking bar 

469: has a finger-like shape, and its dimensions are changing on

470: the same time scale \cite {Meerson2}.

471: 

472: We are grateful to Azi Lipshtat for

473: help. This work was supported in part by a grant from the Israel

474: Science Foundation, administered by the Israel Academy of Sciences and

475: Humanities.

476: \begin{references}

477: \bibitem{Bray1} A.J. Bray, Adv. Phys. {\bf 43}, 357 (1994).

478: \bibitem{Fractals} B.B. Mandelbrot, {\it The Fractal Geometry of

479: Nature} (Freeman, San Francisco, 1982); J. Feder, {\it Fractals}

480: (Plenum, New York, 1988).

481: \bibitem{Vicsek} T. Vicsek, {\it Fractal Growth Phenomena}

482: (World Scientific, Singapore, 1992).

483: \bibitem{Toyoki}H. Toyoki and K. Honda, Phys. Lett. {\bf 111A},

484: 367 (1985).

485: \bibitem{Jullien1} R. Semp\'{e}r\'{e}, D. Bourret, T. Woignier, J.

486: Phalippou, and R. Jullien, Phys. Rev. Lett. {\bf 71}, 3307 (1993).

487: \bibitem{Irisawa1} T. Irisawa, M. Uwaha, and Y. Saito, Europhys.

488: Lett. {\bf 30}, 139 (1995).

489: \bibitem{Irisawa2} T. Irisawa, M. Uwaha, and Y. Saito, Fractals {\bf 4}

490: 251 (1996).

491: \bibitem{Jullien2} N. Olivi-Tran, R. Thouy and R. Jullien, J.

492: Phys. I {\bf 6}, 557 (1996).

493: \bibitem{Meerson}B. Meerson and P.V. Sasorov,

494: cond-mat/9708036.

495: \bibitem{Conti1} M. Conti, B. Meerson, and P.V. Sasorov, Phys. Rev.

496: Lett. {\bf 80} 4693 (1998).

497: \bibitem{Conti2} M. Conti, B. Meerson, and P.V. Sasorov,

498: cond-mat/9912426.

499: \bibitem{Peleg} A. Peleg and B. Meerson, Phys. Rev. E {\bf 59}, 1238 (1999);

500: {\bf 62} (2000).

501: \bibitem{Crooks} G.E. Crooks, B. Ostrovsky, and Y. Bar-Yam, Phys.

502: Rev. E {\bf 60}, 4559 (1999).

503: \bibitem{Kalinin} S.V. Kalinin {\it et al.}, Phys. Rev. E {\bf 61}, 

504: 1189 (2000).

505: \bibitem{Streitenberger} P. Streitenberger, in {\it Paradigms of Complexity.

506: Fractals and Structures in Sciences}, edited by M.N. Novak (World

507: Scientific, Singapore, 2000), p. 135.

508: \bibitem{Meakin} P. Meakin, {\it Fractals, Scaling and Growth Far

509: from Equilibrium} (Cambridge University Press, Cambridge, 1997).

510: \bibitem{airplane} One can observe an uncontrolled version of

511: this phenomenon while 

512: looking at slowly changing ice patterns on

513: the double window of an airplane flying at a high altitude. 

514: \bibitem{Williams} Z. Toroczkai and E. Williams,

515: Phys. Today {\bf 52} (12), 24 (1999).

516: \bibitem{Zinke-Allmang} M. Zinke-Allmang, L.C. Feldman, and

517: M.H. Grabow, Surface Sci. Reports {\bf 16}, 277 (1992).

518: \bibitem{Aranson1} I. S. Aranson {\it et al.}, Phys. Rev. Lett. {\bf 84}, 

519: 3306 (2000). 

520: \bibitem{Schimansky} L. Schimansky-Geier, Ch. Z\"{u}licke,

521: and E. Sch\"{o}ll, Z. Phys. B {\bf 84}, 433 (1991).

522: \bibitem{Rubinstein} J. Rubinstein and P. Sternberg, IMA J. Appl.

523: Math. {\bf 48}, 249 (1992).

524: \bibitem{Mikhailov}A.S. Mikhailov, {\it Foundations of Synergetics

525: I. Distributed Active Systems} (Springer-Verlag, Berlin, 1993).

526: \bibitem{MS} B. Meerson and P.V. Sasorov, Phys. Rev. E {\bf 53}, 3491

527: (1996).

528: \bibitem{Gage} M. Gage, Contemporary Math. {\bf 51}, 51 (1986).

529: \bibitem{Pego} R.L. Pego, Proc. R. Soc. Lond. A {\bf 422}, 261 (1989).

530: \bibitem{Bray2} A.J. Bray, Phys. Rev. Lett. {\bf 66}, 2048 (1991),

531: and references therein.

532: \bibitem{simulations} J.F. Annett and J.R. Banavar, Phys. Rev. Lett. 

533: {\bf 68}, 2941 (1992); L.L. Moseley, P.W. Gibbs, and N. Jan, J. 

534: Stat. Phys. {\bf 67}, 813 ( 1992); A.D. Rutenberg, Phys. Rev. 

535: E {\bf 54}, 972 (1996).

536: \bibitem{Majumdar} C. Sire and S.N. Majumdar, Phys. Rev. E {\bf 52}, 244

537: (1995).

538: \bibitem{CMPS} M. Conti, B. Meerson, A. Peleg, and P.V. Sasorov,

539: in preparation.

540: \bibitem{Witten} T.A. Witten, Jr. and L.M. Sander, Phys. Rev.

541: Lett. {\bf 47}, 1400 (1981).

542: \bibitem{shrinking} The predicted ``shrinking exponent" 

543: $(D-2)/(2 D) \simeq - 0.07 $ is too small to be measured accurately. 

544: \bibitem{Parker} J.R. Parker, {\it Practical Computer Vision Using C}

545: (Wiley, New York, 1993), p. 51.

546: \bibitem{Huse} D.A. Huse, Phys. Rev. B, {\bf 34}, 7845 (1986).

547: \bibitem{Meerson2} A. Peleg, M. Conti,

548: B. Meerson, and A.J. Vilenkin, in preparation.

549: \end{references}

550: \end{document}

551: