1: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}

2: 

3: \usepackage{graphicx}% Include figure files

4: \usepackage{dcolumn}% Align table columns on decimal point

5: \usepackage{bm}% bold math

6: 

7: \begin{document}

8: \title{Diffusion-limited-aggregation on a directed small world network}

9: \author{Jie Ren$^{1}$}

10: \author{Wen-Xu Wang$^{2}$}

11: \author{Gang Yan$^{3}$}

12: \author{Bing-Hong Wang$^{2}$}

13: \email{bhwang@ustc.edu.cn}

14: \affiliation{$^{1}$Department of Physics,

15: \\$^{2}$Department of Modern Physics,

16: \\$^{3}$Department of Electronic Science and Technology, \\University of

17: Science and Technology of China, Hefei, 230026, PR China }

18: 

19: \date{\today}

20: 

21: \begin{abstract}

22: For real world systems, nonuniform medium is ubiquitous. Therefore,

23: we investigate the diffusion-limited-aggregation process on a two

24: dimensional directed small-world network instead of regular lattice.

25: The network structure is established by rewiring connections on the

26: two dimensional directed lattice. Those rewired edges are controlled

27: by two parameters $\theta$ and $m$, which characterize the spatial

28: length and the density of the long-range connections, respectively.

29: Simulations show that there exists a maximum value of the fractal

30: dimension when $\theta$ equals zero. Interestingly, we find that the

31: symmetry of the aggregation pattern is broken when rewired

32: connections are long enough, which may be an explanation for the

33: formation of asymmetrical fractal in nature. Then, we perform

34: multifractal analysis on the patterns further.

35: \end{abstract}

36: 

37: \maketitle

38: \section{INTRODUCTION}

39: Nonequilibrium growth models leading naturally to self-organized

40: fractal patterns, the structure of which strongly depends on the

41: dynamics of the growth process, are of continuing interest due to

42: their relevance for many important fields\cite{Meakin}.

43: Diffusion-limited -aggregation (DLA)\cite{Witten} is probably the

44: most remarkable growth model for pattern formation. This model

45: generates complex and mysterious fractal

46: structures\cite{Vicsek,Stanley}, which seem to be generated as well

47: in natural systems in which growth is controlled by diffusive,

48: including dielectric breakdown\cite{dielectric}, electrochemical

49: deposition\cite{electrochemical}, colloid aggregation\cite{Kolb},

50: film growth\cite{Elam}, viscous fingering\cite{viscous}, Laplacian

51: flow\cite{Laplacian} etc.

52: 

53: In the original DLA model\cite{Witten}, particles released at a

54: point distant from the cluster execute random walks until they find

55: a nearest neighbor site of the cluster and irreversibly stick at

56: this site. However, for many physical processes, the medium is

57: nonuniform so that the probability of jumping from site $i$ to site

58: $j$ is usually not equal to that from $j$ to $i$. Moreover, except

59: nearestneighboring jumps, there also exist some nonlocal jumps

60: through which the particle can move to a distant site at a step. One

61: example is the diffusion of adatoms on metal surfaces, in which the

62: long jumps play a dominating role\cite{Long1,Long2}. A bunch of

63: defects or impurities in the substrate may also play the part of the

64: long-jump path in the case of weak absorbate-substrate

65: interaction\cite{Long3}, which is important for the thin-film

66: growth, heterogeneous catalysis, and oxidation. Hence, the

67: traditional undirected regular lattice will miss important

68: information of medium and it is unconformity to characterize the

69: actual DLA process. This thus calls for the use of network structure

70: with directed long-range connections.

71: 

72: On the other hand, complex networks have recently attracted an

73: increasing interest among physicists\cite{Review1,Review2,Review3}.

74: In particular, small-world (SW) networks, introduced by Watts and

75: Strogatz\cite{WS}, have been extensively studied because they

76: constitute an interesting attempt to translate the complex topology

77: of social, biological, and physical networks into a simple model.

78: Two dimensional (2D) SW networks result from randomly rewiring a

79: fraction of links of a 2D regular lattice. Several dynamical models

80: have recently been studied in order to understand the effect of SW

81: topology on classical systems such as the Ising model\cite{Ising},

82: the spread of epidemics\cite{Epidemic}, aggregation\cite{Huang},

83: random walks\cite{RW}, etc. Such models are expected to capture the

84: essential features of the complicated processes taking place on real

85: networks.

86: 

87: In this paper, we investigate the DLA process on a 2D directed SW

88: network, in which the directed links correspond to the directed

89: irreversibly jumps and the node is regarded as the lattice point of

90: the real space, respectively.

91: 

92: \section{THE MODEL}

93: In order to construct the directed SW network, we start from a 2D

94: square lattice of size $L\times L$ consisting of sites linked to

95: their four nearest neighbors by both outgoing and incoming links, as

96: shown in Fig. 1. Then, we reconnect nearest-neighbor outgoing links

97: to a different site chosen at random with the probability

98: \begin{equation}

99: p(r)\sim e^{-\theta r},

100: \end{equation}

101: where $r$ is the lattice distance between the two selected sites and

102: $\theta(\theta\in[-1,1])$ is the clustering exponent which

103: characterizes the localized degree of the long-range links in the

104: lattice. The formula corresponds to the diffusivity represented by

105: Arrhenius relation as usual in surface

106: science\cite{Long1,Long2,Long3}. The reconstructing process is

107: repeated until $m$, the number of long-range rewiring connections,

108: reaches a desired value. Note that by performing this procedure

109: every site will have exactly four fixed outgoing links and a random

110: number of incoming links. When the clustering exponent $\theta$=0,

111: we have a uniform distribution over long-range connections, and the

112: present model reduces to the basic 2D directed SW

113: network\cite{directed}. As $\theta\rightarrow1$

114: ($\theta\rightarrow1$ denotes $\theta$ tends to 1), the long-range

115: links of a site become more and more local in its vicinity. In

116: reverse, as $\theta\rightarrow-1$, the long-range rewiring outgoing

117: links are in favor of the farther sites. Thus, the clustering

118: exponent $\theta$ serves as a structural parameter controlling the

119: spatial length of the long-range connections.

120: 

121: \begin{figure}

122: \scalebox{0.4}[0.4]{\includegraphics{SW.eps}}

123: \caption{\label{fig:epsart} Sketch map of a directed 2D SW lattice.

124: Dotted lines represent rewired connections. Arrows indicate the

125: direction of the corresponding connection.}

126: \end{figure}

127: 

128: Based on the directed 2D SW network as constructed above, we have

129: performed extensive numerical simulations for the DLA with size of

130: the reconstructed lattice $L$=1000 with number of particles

131: $N$=10000. Staring from an immobile seed at the center of the

132: lattice, a particle is released at a random position where is depart

133: from the outer radius of the growing pattern. Then the particle

134: jumps along the direction from the current site to one of its linked

135: sites which are not occupied by the growing pattern, with equal

136: probability step by step. At last, the particle irreversibly sticks

137: at the nearest neighbor site of the growing pattern in terms of the

138: physical distance and the pattern will grow gradually. To reduce the

139: effect of fluctuation, the calculated result is taken average over

140: 10 different network realizations and 10 independent runs for each

141: network configuration for each set of parameters $(\theta, m)$.

142: 

143: \section{SIMULATION RESULT AND DISCUSSION}

144: \begin{figure}

145: \scalebox{0.5}[0.5]{\includegraphics{pattern.eps}} \caption{The

146: simulation patterns as a function of the clustering exponent for the

147: number of long-range connections $m$=1000, 10000 and 100000

148: respectively. Each color represents 1000 particles in order.}

149: \end{figure}

150: 

151: Fig 2 illustrates the patterns of DLA which exhibit rich behaviors

152: for different parameters $\theta$ and $m$. For each $\theta$, it can

153: be seen that with the increase of $m$ the patterns of DLA become

154: thicker and denser, however, which is not obvious for large

155: $\theta$, approximately 1. For each $m$, the pattern is nearly the

156: most dense when $\theta$=0. While, it gets thin and sparse when

157: $\theta$ departs from 0 tending to 1 or $-1$. However, it is

158: astonishing that the symmetry of the aggregation pattern is markedly

159: broken while $\theta<0$, which is more obvious as $\theta$ tends to

160: $-1$. To quantify the patterns of DLA, we calculate the fractal

161: dimensions $D_0$ of the DLA by box-counting

162: method\cite{Vicsek,Stanley}, which are shown in Fig. 3. It is clear

163: that there exists a maximum value of $D_0$ when $\theta$ equals 0

164: for each $m$. It can be seen that the more $\theta$ departs from 0,

165: the more $D_0$ decreases. Moreover, it is found that $D_0$ decreases

166: more fast when $\theta\rightarrow1$ than $\theta\rightarrow-1$.

167: 

168: \begin{figure}

169: \scalebox{0.85}[0.85]{\includegraphics{D0.eps}}

170: \caption{\label{fig:epsart} The fractal dimension $D_0$ of patterns

171: as a function of the clustering exponent $\theta$ for several $m$.}

172: \end{figure}

173: 

174: It is well-known that the special randomly branching, open structure

175: of a DLA pattern results from the effects of screening, which is

176: manifested through the fact that the tips of most advanced branches

177: capture the incoming diffusing particles most effectively. In the

178: present work, due to the long-range connections, particles can jump

179: directly to distant sites, including the traditionally completely

180: screened deep fjord. The nonlocal connections effectively weaken the

181: screening effect so that the pattern of aggregate becomes compact

182: and the fractal dimension $D_0$ increases with $m$ increasing. On

183: the other hand, the clustering exponent $\theta$, which restricts

184: the spatial length of the long-range connection, affects the

185: morphology of the aggregate and the fractal dimension $D_0$. As

186: $\theta\rightarrow1$, the long-range links are restricted more and

187: more local in its vicinity so that the capacity of weakening the

188: screening effect vanishes gradually. Finally, $D_0$ does not vary

189: and the morphology of the aggregate seems like the original DLA

190: pattern, as shown in Fig. 2 and Fig. 3. When $\theta$=0, the spatial

191: lengths of the long-range connections are entire random, neither too

192: distant nor too local and they have a uniform distribution. Then,

193: due to intensive weakening for the screening effect, the random

194: particle has the chance to appear on arbitrary sites on the

195: underlying network so that the pattern becomes thick and compact,

196: corresponding to increase of $D_0$. However, as

197: $\theta\rightarrow-1$, the long-range links tend to the sites as

198: distant as possible and the irreversible jumps along directed links

199: break the symmetry of dynamics. Thus, small fluctuations are

200: enhanced, and this instability together with the randomness inherent

201: in the model leads to a complex asymmetrical behavior.

202: 

203: However, the fractal dimension $D_0$ is a rough description because

204: the pattern becomes asymmetric while $\theta\rightarrow-1$. So, we

205: have performed the multifractal analyse\cite{multi1,multi2} here to

206: see more details. It should be noted that our measurements concern

207: the pattern itself other than its harmonic measure\cite{multi}.

208: 

209: Further characterization of the DLA patterns can be achieved by

210: determining the generalized fractal dimensions $D_q$. Cover the

211: pattern with a grid of square boxes of size $\varepsilon$ and define

212: $P_i(\varepsilon)$ to be the relative portion of the pattern in cell

213: $i$, and define $N(\varepsilon)$ to be the total number of boxes of

214: size $\varepsilon$ needed to cover the whole pattern. The relative

215: portion $P_i(\varepsilon)$ can be described as multifractal as:

216: \begin{equation}

217: P_i(\varepsilon)\sim\varepsilon^{\alpha},

218: \end{equation}

219: \begin{equation}

220: N_\alpha(\varepsilon)\sim \varepsilon^{-f(\alpha)},

221: \end{equation}

222: where $\alpha$ is the singularity, $N_\alpha(\varepsilon)$ the

223: number of small squares of relative size $\varepsilon$ with the same

224: singularity, and $f(\alpha)$ is the fractal dimension.

225: 

226: \begin{figure}

227: \scalebox{0.85}[0.85]{\includegraphics{multi1.eps}}

228: \caption{\label{fig:epsart} The multifractal spectra $f(\alpha)$ of

229: the distribution of the patterns, with various number of long-range

230: connections, $m$=1000, 10000, 100000, respectively.}

231: \end{figure}

232: 

233: To describe quantitatively the subtle geometrical feature of the

234: pattern, the fractal dimension $f(\alpha)$ can be obtain from the

235: partition function $\chi_q$:

236: \begin{equation}

237: \chi_q=\sum_{i}^{N(\varepsilon)}P_i^q(\varepsilon),

238: \end{equation}

239: and its power law of $\varepsilon$,

240: \begin{equation}

241: \tau_q=\lim_{\varepsilon\rightarrow0}\frac{\ln\chi_q}{\ln\varepsilon},

242: \end{equation}

243: where $q$ is the moment order and $\tau_q$ the index of the pow law.

244: The generalized fractal dimension is defined as:

245: \begin{equation}

246: D_q=\frac{\tau_q}{q-1},

247: \end{equation}

248: Then, $(\alpha, f(\alpha))$ can be obtained from $(q, D_q)$ by

249: performing the Legendre transformation:

250: \begin{equation}

251: \alpha=\frac{d}{dq}[(q-1)D_q],

252: \end{equation}

253: \begin{equation}

254: f(\alpha)=\alpha q-(q-1)D_q,

255: \end{equation}

256: In our calculation, the moment order $q$ is taken for -30 to 30.

257: 

258: 

259: \begin{figure}

260: \scalebox{0.85}[0.85]{\includegraphics{multi2.eps}}

261: \caption{\label{fig:epsart} The multifractal spectra with several

262: value of the cluster exponent $\theta$, -1, -0.01, 0, 0.01, 1, for

263: $m$=10000.}

264: \end{figure}

265: 

266: We have calculated the multifractal spectra $f(\alpha)$ of the

267: distribution of the patterns, with various number of long-range

268: connections, $m$=1000, 10000, 100000, respectively, for a original

269: directed 2D SW lattice, $\theta$=0. Figure 4 shows the result. It

270: can be seen that the curve becomes higher and the range of

271: singularity $\alpha$ becomes wider with increasing the number of the

272: long-range connections. In Table I, the multifractal parameters of

273: the distribution are listed. The multifractal spectrum can be used

274: to provide more information about the subtle geometrical difference,

275: because of the $\alpha_{max}$ and $\alpha_{min}$ connecting with the

276: smallest probability and the largest probability of the spatial

277: distribution [show by Eq .(2)]. The result (Table I) show that

278: $\alpha_{max}$ and $\alpha_{min}$ both increase with increasing $m$,

279: while $\Delta\alpha$=$\alpha_{max}-\alpha_{min}$ also increases,

280: indicating that the pattern becomes less irregular, less nonuniform,

281: and more dense. Moreover, Fig. 5 illustrates that the multifractal

282: curves with several value of the cluster exponent $\theta$, -1,

283: -0.01, 0, 0.01, 1, for a directed 2D SW lattice, $m$=10000. More

284: quantitative details can be seen in Table II. It illustrates that

285: the range of $\alpha$ is the broadest and the curve is the maximal

286: when $\theta$=0, suggesting the pattern is the most dense, compact

287: and regular, which corresponds to Fig. 2 and Fig. 3 showed above.

288: 

289: \begin{center}

290:  {TABLE I: Some multifractal parameters of Figure 4.}

291:   \begin{tabular}{cccc}\hline\hline

292:    $m$   & 1000 & 10000 & 100000 \\ \hline

293:    $\alpha_{min}$   & 1.598 & 1.726 & 1.786 \\

294:    $\alpha_{max}$   & 3.332 & 3.591 & 4.034 \\

295:    $\Delta\alpha=\alpha_{max}-\alpha_{min}$   & 1.734 & 1.865 & 2.248 \\

296:    $f(\alpha_{min})$   & 0.684 & 0.752 & 1.251 \\

297:    $f(\alpha_{max})$   & 0.303 & 0.180 & 0.147 \\

298:    $\Delta f=f(\alpha_{min})-f(\alpha_{max})$   & 0.381 & 0.572 & 1.104 \\

299:    \hline\hline

300:  \end{tabular}

301: \end{center}

302: 

303: 

304: 

305: 

306: \begin{center}

307:  {TABLE II: Some multifractal parameters of Figure 5.}

308:   \begin{tabular}{cccccc}\hline\hline

309:    $\theta$   & -1 & -0.01 & 0 & 0.01 & 1 \\ \hline

310:    $\alpha_{min}$   & 1.532 & 1.630 & 1.726 & 1.648 & 1.392 \\

311:    $\alpha_{max}$   & 2.969 & 3.307 & 3.591 & 3.109 & 2.727 \\

312:    $\Delta\alpha=\alpha_{max}-\alpha_{min}$   & 1.437 & 1.677 & 1.865 & 1.461 & 1.335 \\

313:    $f(\alpha_{min})$   & 0.352 & 0.652 & 0.752 & 0.676 & 0.347 \\

314:    $f(\alpha_{max})$   & 0.438 & 0.326 & 0.180 & 0.635 & 0.470 \\

315:    $\Delta f=f(\alpha_{min})-f(\alpha_{max})$   & -0.086 & 0.326 & 0.572 & 0.041 & -0.123 \\ \hline\hline

316:  \end{tabular}

317: \end{center}

318: 

319: \section{CONCLUSION}

320: 

321: In summarize, we have investigated the DLA process on a nonuniform

322: medium which is characterized by a directed 2D SW lattice with two

323: introduced parameters $(\theta, m)$ which govern the style of the

324: pattern. It is found that as $m$ increases, the aggregation pattern

325: become thicker and denser, which indicates the fractal dimension

326: increases. We also figure out that there exists a maximum value of

327: $D_0$ in the case of $\theta=0$ for any value of $m$, which implies

328: the densest aggregation pattern corresponds to the cases of entire

329: randomly length of long-range connections, neither too long nor too

330: local. Interestingly, we find that the symmetry of the aggregation

331: pattern is broken when rewired connections are long enough. The

332: directed long-range links contribute to the formation of

333: asymmetrical patterns. The random walk of the particles along the

334: directed links is irreversible so that the principle of detailed

335: balance is broken. Hence, the asymmetry of the dynamics finally

336: results in the asymmetry patterns. To give detailed description of

337: the asymmetrical pattern, we have performed multifractal analysis on

338: the patterns. The subtle geometrical difference among these patterns

339: for different parameter value can be provided by the multifractal

340: parameters. Although the model we have proposed is very simple, the

341: simulation results demonstrate that it can capture most of the

342: general features of asymmetrical growth processes. Other than the

343: traditional asymmetrical factor such as gravity, magnetic field,

344: electric field, etc, the asymmetrical factor of our model is the

345: directed link, which causes the break of dynamics symmetry inherent.

346: It may be an new explanation for the formation of asymmetrical

347: fractal behavior in nature.

348: 

349: \begin{thebibliography}{10}

350: \bibitem{Meakin} P. Meakin, {\sl Fractal, Scaling and Growth far from Equilibrium} (Cambridge

351: University Press, Cambridge, U.K.,1998).

352: 

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

354: \textbf{47}, 1400 (1981).

355: 

356: \bibitem{Vicsek} T. Vicsek, {\sl Fractal Growth Phenomena} (World Scientific,

357: (Singapore, 1992).

358: 

359: \bibitem{Stanley} A.-L. Barabasi and H. E. Stanley, {\sl Fractal Concepts on Surface

360: Growth} (Cambridge University Press, Cambridge, UK, 1995).

361: 

362: \bibitem{dielectric} L. Niemeyer, L. Pietronero, and H. J. Wiesmann,

363: Phys. Rev. Lett. \textbf{52}, 1033 (1984).

364: 

365: \bibitem{electrochemical} R. M. Brady and R. C. Ball, Nature

366: (London) \textbf{309}, 225 (1984); M. Matsushita, M. Sano, Y.

367: Hayakawa, H. Honjo and Y. Sawada, Phys. Rev. Lett. \textbf{53}, 286

368: (1984).

369: 

370: \bibitem{Kolb} M. Kolb, R. Botet, and R. Jullien, Phys. Rev. Lett.

371: \textbf{51}, 1123 (1983).

372: 

373: \bibitem{Elam} W. T. Elam, S. A. Wolf, J. Sprague, D. U. Gubser, D. Van Vechten, G. L.

374: Barz, and P. Meakin, Phys. Rev. Lett. \textbf{54}, 701 (1985).

375: 

376: \bibitem{viscous} K. J. M{\aa}l{\o}y, J. Feder, and T. J{\o}ssang,

377: Phys. Rev. Lett. \textbf{55}, 2688 (1985).

378: 

379: \bibitem{Laplacian} L. Paterson, Phys. Rev. Lett. \textbf{52},1621

380: (1984).

381: 

382: \bibitem{Long1} D. C. Senft and G. Ehrlich, Phys. Rev. Lett.

383: \textbf{74}, 274 (1995).

384: 

385: \bibitem{Long2} T. R. Linderoth, S. Horch, E. L{\ae}gsgaard, I. Stensgaard, and

386: F. Besenbacher, Phys. Rev. Lett. \textbf{78}, 4978 (1997).

387: 

388: \bibitem{Long3} M. Schunack, T. R. Linderoth, F. Rosei, E. L{\ae}gsgaard, I. Stensgaard, and F.

389: Besenbacher, Phys. Rev. Lett. \textbf{88}, 156102 (2002).

390: 

391: \bibitem{Review1} R. Albert and A.-L. Barab\'asi, Rev. Mod. Phys.

392: \textbf{74}, 47 (2002).

393: 

394: \bibitem{Review2} S. N. Dorogovtsev and J. F. F. Mendes, Adv. Phys. \textbf{51}, 1079 (2002).

395: 

396: \bibitem{Review3} M. E. J. Newman, SIAM Review \textbf{45}, 167

397: (2003).

398: 

399: \bibitem{WS} D. J. Watts and S. H. Strogatz, Nature (London) \textbf{393}, 440

400: (1998); S. H. Strogatz, \emph{ibid}. \textbf{410}, 268 (2001).

401: 

402: \bibitem{Ising} A. Barrat and M. Weigt, Eur. Phys. J. B \textbf{13}, 547 (2000).

403: 

404: \bibitem{Epidemic} R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. \textbf{86}, 3200

405: (2001).

406: 

407: \bibitem{Huang} S.-Y. Huang, X.-W. Zou, Z.-G. Shao, Z.-J. Tan, and

408: Z.-Z. Jin, Phys. Rev. E \textbf{69}, 067104 (2004).

409: 

410: \bibitem{RW} J. D. Noh, and H. Rieger, Phys. Rev. Lett.

411: \textbf{92}, 118701 (2004).

412: 

413: \bibitem{directed} A. D. S\'{a}nchez, J. M. L\'{o}pez, and M. A.

414: Rodr\'{\i}guez, Phys. Rev. Lett. \textbf{88}, 048701 (2002).

415: 

416: \bibitem{multi1} H. G. E. Hentschel and I. Procaccia, Physica D \textbf{8}, 435 (1983).

417: 

418: \bibitem{multi2} T. C. Halsey, M. H. Jensen, L. P. Kadanoff, I. Procaccia, and B. I.

419: Shraiman, Phys. Rev. A \textbf{33}, 1141 (1986).

420: 

421: \bibitem{multi} O. Praud and H. L. Swinney, Phys. Rev. E

422: \textbf{72}, 011406 (2005).

423: 

424: \end{thebibliography}

425: 

426: \end{document}

427: