1: \documentclass[11pt]{elsart}

2: \usepackage[dvips]{graphicx}

3: 

4: \newtheorem{property}{Property}

5: \setlength{\parindent}{8pt}

6: \begin{document}

7: \begin{frontmatter}

8: 

9: \title{Relations between Average Distance, Heterogeneity and Network Synchronizability}

10: 

11: \author{Ming Zhao$^{a}$ }

12: \author{Tao Zhou$^{a,b}$ }

13: \ead{zhutou@ustc.edu}

14: \author{Bing-Hong Wang$^{a}$ }

15: \ead{bhwang@ustc.edu.cn}

16: \author{Gang Yan$^{b}$ }

17: \author{Hui-Jie Yang$^{a}$ }

18: \author{and Wen-Jie Bai$^{c}$ }

19: 

20: \address

21: {$^{a}$ Department of Modern Physics and Nonlinear Science Center,

22: University of Science and Technology of China,  Hefei Anhui,

23: 230026, PR China }

24: 

25: \address

26: {$^{b}$ Department of Electronic Science and

27: Technology, University of Science and Technology of China,  Hefei

28: Anhui, 230026, PR China}

29: 

30: \address

31: {$^{c}$ Department of Chemistry, University of Science and

32: Technology of China,  Hefei Anhui, 230026, PR China}

33: 

34: \begin{abstract}

35: By using the random interchanging algorithm, we investigate the

36: relations between average distance, standard deviation of degree

37: distribution and synchronizability of complex networks. We find that

38: both increasing the average distance and magnifying the degree

39: deviation will make the network synchronize harder. Only the

40: combination of short average distance and small standard deviation

41: of degree distribution that ensures strong synchronizability. Some

42: previous studies assert that the maximal betweenness is a right

43: quantity to estimate network synchronizability: the larger the

44: maximal betweenness, the poorer the network synchronizability. Here

45: we address an interesting case, which strongly suggests that the

46: single quantity, maximal betweenness, may not give a comprehensive

47: description of network synchronizability.

48: 

49: \begin{keyword}

50: Synchronizability\sep Complex Networks \sep Average Distance\sep

51: Heterogeneity

52: \PACS 89.75.-k\sep 05.45.Xt

53: \end{keyword}

54: \end{abstract}

55: 

56: \date{}

57: \end{frontmatter}

58: 

59: 

60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

61: \section{Introduction}

62: A variety of systems in nature can be described by complex

63: networks and the most important statistical features of complex

64: networks are the small-world effect and scale-free

65: property\cite{Review1,Review2,Review3,Review4}. Networks that have

66: small average distance as random networks and large clustering

67: coefficient as regular ones are called small-world

68: networks\cite{WS}. And the scale-free property means the degree

69: distribution of networks obeys the power-law form\cite{SFN}. One

70: of the ultimate goals of researches on complex networks is to

71: understand how the structure of complex networks affects the

72: dynamical process taking place on them, such as traffic

73: flow\cite{Traffic1,Traffic2,Traffic3,Traffic4,Traffic5}, epidemic

74: spread\cite{Epidemic1,Epidemic2,Epidemic3,Epidemic4,Epidemic5,Epidemic6},

75: cascading behavior\cite{Cascade1,Cascade2,Cascade3}, and so on.

76: 

77: The large networks of coupled dynamical systems that exhibit

78: synchronized state are subjects of great interest. Previous

79: studies have demonstrated that scale-free and small-world networks

80: are much easier to synchronize than regular

81: lattice\cite{Easier1,Easier2,Easier3,Easier4}. Then what makes

82: complex networks synchronize so easily? It is intuitively believed

83: that shorter average distance predicts better

84: synchronizability\cite{Easier2,Easier3,Zhou2005}. However, it is

85: found that to decrease average distance will make some complex

86: networks synchronize even harder\cite{Betweenness1}. More

87: bewilderingly, a very recent work suggests that on some

88: synchronization systems, the synchronizability is independent of

89: the average distance\cite{Hasegawa}. Some authors also addressed

90: that the homogeneous distribution of degree will lead to better

91: synchronizability. Hong \emph{et al.}\cite{Betweenness2}

92: investigate the relationship between network synchronizability and

93: various topological ingredients, including average distance,

94: heterogeneity, and betweenness of Watts-Strogatz (WS)

95: networks\cite{WS}. They suggest the maximal betweenness a right

96: indicator for synchronizability. This tentative conclusion has

97: been widely accepted now\cite{Cluster,Betweenness3,Oh}. Recently,

98: several researchers examine the effect of clustering coefficient

99: on the synchronization by using Kuramoto model\cite{Cluster} or

100: master stability function\cite{WuXiang} and find that increasing

101: clustering coefficient will hinder the global synchronization. All

102: the four topological ingredients, average distance, heterogeneity

103: (measured by the standard deviation of degree distribution),

104: betweenness and clustering coefficient, may reflect the networks

105: synchronizability to some extent, but which one or ones indicate

106: the network synchronizability simply and exactly?

107: 

108: There is another problem need to be mentioned. When investigating

109: the relations between various topological ingredients and network

110: synchronizability, some parameters, such as rewiring probability

111: $p$ of WS networks\cite{Betweenness2} or power-law exponent

112: $\gamma$ of the degree distribution in scale-free

113: networks\cite{Betweenness1}, are adjusted to modulate other

114: topological ingredients, like average distance or heterogeneity of

115: degree distribution. However, in this process all the topological

116: ingredients keep changing with the adjusting of these parameters.

117: It is impossible to get clear relation between an ingredient and

118: synchronizability when other ingredients are still varying,

119: especially when we do not know the tracks of their motions.

120: %It is not proper to judge an ingredient's impact on synchronizability

121: %when other ingredients are still varying, especially when we do

122: %not know the tracks of their motions.

123: 

124: Here in this paper, we try to discuss the relationship between these

125: ingredients and the synchronization of complex networks precisely.

126: This paper is organized as follows. In section 2, we give a brief

127: review on how to measure the network synchronizability. And then, in

128: section 3, the so-called random interchanging

129: algorithm\cite{Reshuffle1,Reshuffle2,Reshuffle3} is introduced,

130: which allows one to manipulate the clustering coefficient while

131: keeping the network's degree distribution unchanged. The main

132: simulations are shown in section 4, and an interesting case is laid

133: out and discussed in section 5. Finally, in section 6, we sum up

134: this paper and discuss the relevance of our work to the real world.

135: 

136: \begin{figure}

137: \begin{center}

138:  \scalebox{0.5}[0.5]{\includegraphics{graph1}}

139: \caption{The sketch maps of random interchanging algorithm.}

140: \end{center}

141: \end{figure}

142: 

143: \section{Network Synchronizability}

144: In this section, we will introduce a generic model of coupled

145: oscillators on networks and a useful measure\cite{master1}, which

146: is often used to test the stability of the global synchronized

147: states. Consider $N$ identical dynamical systems (oscillators)

148: with the same output function, which are located on the vertices

149: of a network and coupled linearly and symmetrically with

150: neighbors. The coupling fashion ensures the synchronization

151: manifold an invariant manifold, and the dynamics can be locally

152: linearized near the synchronous state. The state of the $i$th

153: oscillator is denoted by $\textbf{x}^i$, we get the set of

154: equations of motion governing the dynamics of the $N$ coupled

155: oscillators:

156: \begin{equation}

157: \dot{\textbf{x}}^i=\textbf{F}(\textbf{x}^i)+\eta\sum_{j=1}^NG_{ij}\textbf{H}(\textbf{x}^j),

158: \end{equation}

159: where $\dot{\textbf{x}}^i=\textbf{F}(\textbf{x}^i)$ governs the

160: dynamics of individual oscillator, $\textbf{H}(\textbf{x}^j)$ is

161: the output function and $\eta$ the coupling strength. The $N\times

162: N$ Laplacian $\textbf{G}$ is given by

163: \begin{equation}

164:     G_{ij}=\left\{

165:     \begin{array}{cc}

166:     k_i   &\mbox{for $i=j$}\\

167:      -1    &\mbox{for $j\in\Lambda_i$}   ,\\

168:      0    &\mbox{otherwise}

169:     \end{array}

170:     \right.

171: \end{equation}

172: where $\Lambda_i$ denotes the neighbor set of node $i$. Because of

173: the positive semidefinite of $\textbf{G}$, all the eigenvalues of

174: it are nonnegative reals and the smallest eigenvalue $\theta_0$ is

175: always zero, for the rows of $\textbf{G}$ have zero sum. If all

176: the nodes are connected, there is only one zero eigenvalue. Thus,

177: the eigenvalues can be ranked as

178: $\theta_0<\theta_1\leq\cdots\leq\theta_{N-1}$. The ratio of the

179: maximum eigenvalue $\theta_{N-1}$ to the smallest nonzero one

180: $\theta_1$ is widely used to measure the synchronizability of the

181: network\cite{master1}, if the eigenratio $R=\theta_{N-1}/\theta_1$

182: satisfies

183: \begin{equation}

184: R<\alpha_2/\alpha_1,

185: \end{equation}

186: we say the network is synchronizable. The right-hand side

187: $\alpha_2/\alpha_1$ of this inequality depends on the dynamics of

188: individual oscillator and the output function (one can see ref.

189: \cite{master2} for details), while the eigenratio $R$ depends only

190: on the Laplacian $\textbf{G}$. $R$ indicates the synchronizability

191: of the network, the smaller it is the better synchronizability and

192: vice versa. In this paper, for universality, we will not address a

193: particular dynamical system, but concentrate on how the network

194: topology affects the eigenratio $R$.

195: 

196: \section{The Random Interchanging Algorithm}

197: To investigate the structural effects on network

198: synchronizability, we use random interchanging algorithm

199: \cite{Reshuffle1,Reshuffle2,Reshuffle3} to adjust clustering

200: coefficient while keeping degree distribution unchanged. The

201: procedure is as follows:

202: 

203: (1) Randomly pick two existing edges $e_1=x_1x_2$ and

204: $e_2=x_3x_4$, such that $x_1\neq x_2\neq x_3\neq x_4$ and there is

205: no edge between $x_1$ and $x_4$ as well as $x_2$ and $x_3$.

206: 

207: (2) Interchange these two edges, that is, connect $x_1$ and $x_4$

208: as well as $x_2$ and $x_3$, and remove the edges $e_1$ and $e_2$.

209: 

210: (3) Ensure the network is still connected and compute whether this

211: interchange increases/decreases the network clustering

212: coefficient. If it does, accept the new configuration, else

213: recover the old one.

214: 

215: (4) Repeat step (1) unless the desired clustering coefficient is

216: achieved.

217: 

218: Since this algorithm only rewires connections and does not change

219: the degree of any node, the degree distribution as well as the

220: degree sequence is fixed. Figure 1 provides a sketch maps of

221: random interchanging algorithm, which may help us understanding

222: the program flow.

223: 

224: \section{Simulations}

225: In the random interchanging process, operations that bring

226: nonlocal couplings will reduce network average distance $L$

227: \cite{Easier3,HuJK}, and at the same time the clustering

228: coefficient $C$ will be reduced\cite{Cluster2}. Figure 2 and 3

229: exhibit the relationship between $L$ and $C$. In figure 2, the

230: original networks are the WS networks with size $N=2000$, average

231: degree $<k>=4$, and standard deviations of degree distributions

232: $\sigma$=0.2, 0.5, 0.6 and 0.87, respectively. In figure 3, the

233: original networks are the  extensional BA

234: networks\cite{ExtBA1,ExtBA2} with $N=2000$, average degree

235: $<k>=12$ and standard deviations of degree distributions

236: $\sigma$=18.43, 19.26, 20.65 and 21.26, respectively. Here, the

237: different standard deviations for WS networks and extensional BA

238: networks are obtained by adjusting the rewiring probability $p$

239: and the power-law exponent $\gamma$, respectively. Clearly, the

240: trends of $L$ and $C$ are qualitatively the same. We have checked

241: that the positive correlation between $L$ and $C$ is not sensitive

242: to the network size, the average degree and the standard deviation

243: of degree distribution. In this paper, we examine the relation

244: between average distance and synchronizability, and the relation

245: between clustering coefficient and synchronizability can be

246: obtained easily.

247: 

248: \begin{figure}

249: \begin{center}

250: \scalebox{1.2}[1.2]{\includegraphics{graph2}} \caption{(Color

251: online) The relationship between average distance $L$ and clustering

252: coefficient $C$ when the original networks are the WS networks. The

253: black squares, red circles, green up-triangles and blue

254: down-triangles represent the cases of $\sigma=$0.20, 0.50, 0.60 and

255: 0.87, respectively. All the data are the average over 20 different

256: realizations.}

257: \end{center}

258: \end{figure}

259: 

260: 

261: The eigenratios for WS and extensional BA model are obtained

262: numerical and their behaviors with the average distance $L$ at

263: different standard deviations of degree distributions are

264: exhibited in Figs. 4 and 5 respectively. From each curve in Fig. 4

265: it can be seen that with the increasing of average distance $L$,

266: the eigenratio $R$ grows, which means shorter average distance

267: predicts better synchronizability. The similar result is obtained

268: for extensional BA network shown in Fig. 5. The present result is

269: consistent with the very recent result\cite{Cluster,WuXiang} that

270: the larger clustering coefficient will inhibit global

271: synchronization in scale-free network.

272: 

273: It can also be seen from Figs. 4 and 5 , at equal average distance

274: $L$ the larger the standard deviation of degree $\sigma$ is, the

275: larger the eigenratio $R$ will be, indicating networks with a

276: homogeneous distribution of connectivity are more synchronizable

277: than heterogeneous ones when average distance $L$ keeps constant.

278: 

279: \begin{figure}

280: \begin{center}

281: \scalebox{1.2}[1.2]{\includegraphics{graph3}} \caption{(Color

282: online) The relationship between average distance $L$ and clustering

283: coefficient $C$ when the original networks are the extensional BA

284: networks. The black squares, red circles, green up-triangles and

285: blue down-triangles represent the cases of $\sigma=$18.43, 19.26 and

286: 21.26, respectively. All the data are the average over 20 different

287: realizations.}

288: \end{center}

289: \end{figure}

290: 

291: Average distance and heterogeneity of connections are topological

292: ingredients for network synchronization. Shortening average

293: distance and making the connections more homogeneous solely will

294: increase network synchronizability, however, only their

295: combination could make the network have strong synchronizability.

296: The average distance of star coupling network is very short,

297: $L\rightarrow2$ as $N\rightarrow\infty$, while the standard

298: deviation of degree is very large, $\sigma\sim\sqrt{N}$, and the

299: eigenration $R\rightarrow N$, suggesting that it is hard to

300: synchronize when the network size is large. In one-dimensional

301: ring lattice, all the nodes have equal degree, thus $\sigma=0$.

302: However, the average distance $L\sim N/4z$ is too large (here $z$

303: denotes the coordination number\cite{Newman1999}), thus it is also

304: very hard to synchronize with the increasing network

305: size\cite{ex1}.

306: 

307: \begin{figure}

308: \begin{center}

309: \scalebox{1.2}[1.2]{\includegraphics{graph4}} \caption{(Color

310: online) The eigenratio $R$ vs average distance $L$ when the original

311: networks are the WS networks. Eigenratio $R$ shows positive

312: correlation with average distance $L$ when standard deviation of

313: degree $\sigma$ is fixed, and at a fixed value of $L$, $R$ will

314: increase with the rising of $\sigma$. For a variety of chaotic

315: oscillators, $\alpha_2/\alpha_1$ ranges from 5 to 100\cite{Easier3},

316: so we only investigate the situations with eigenratios less than

317: about 100. All the data are the average over 20 different

318: realizations.}

319: \end{center}

320: \end{figure}

321: 

322: For many scale-free network models, heterogeneous distribution of

323: connectivity tends to reduce the average network

324: distance\cite{HandL,Betweenness1}, WS small-world network obeys

325: the same law\cite{Betweenness2}. Thus, in this two kind of

326: networks, the increasing of heterogeneity will diminish the

327: average distance. While, for WS small-world network the standard

328: deviation of degree is very small, the network synchronizability

329: is mainly determined by average distance, with $L$'s decreasing,

330: although $\sigma$ increases, $R$ will still be diminished, the

331: network becomes more synchronizable. For some real-life scale-free

332: networks, because of their ultra-small feature\cite{HandL,HandZ}

333: ($L\sim \ln\ln N$ or even shorter) and large degree deviation, $R$

334: strongly depends on heterogeneity, the more heterogeneous is the

335: harder to synchronize. Therefore, shortening average distance is

336: an effective way to enhance small-world network synchronizability

337: and diminishing the heterogeneity of connectivity can make

338: scale-free network synchronize easier.

339: 

340: \begin{figure}

341: \begin{center}

342: \scalebox{1.2}[1.2]{\includegraphics{graph5}} \caption{(Color

343: online) The eigenratio $R$ vs average distance $L$ when the original

344: networks are the extensional BA networks. Eigenratio $R$ shows

345: positive correlation with average distance $L$ when standard

346: deviation of degree $\sigma$ is fixed, and at a fixed value of $L$,

347: $R$ will increase with the rising of $\sigma$. All the data are the

348: average over 20 different realizations.}

349: \end{center}

350: \end{figure}

351: 

352: \section{An Interesting Case: Smaller Maximal Betweenness may not Indicate Better Synchronizability}

353: Some previous studies suggest that the maximal betweenness

354: centrality $B_{max}$ is a suitable indicator for predicting

355: synchronizability on complex

356: networks\cite{Betweenness2,Cluster,Betweenness3}; the larger

357: $B_{max}$ is, the poorer the synchronizability. The betweenness

358: centrality of node $n$ is defined as the probability that a

359: randomly selected shortest path of a randomly picked pair of nodes

360: contains the node $n$\cite{BetweennessDef1,BetweennessDef2}

361: \begin{equation}

362: B_{n}:=\frac{1}{(N-1)(N-2)}\sum_{i\neq j\neq

363: n}\frac{g_{ij}(n)}{g_{ij}},

364: \end{equation}

365: where $g_{ij}$ is the number of shortest paths between nodes $i$

366: and $j$, and $g_{ij}(n)$ is the number of those paths passing

367: through node $n$. From the definition of betweenness centrality,

368: it is easy to get the relationship between the average betweenness

369: centrality $<B>$ and the average distance $L$

370: \begin{equation}

371: <B>=\frac{N(N-1)(L-1)}{N(N-1)(N-2)}=\frac{L-1}{N-2}.

372: \end{equation}

373: Previous studies indicate that there exists strongly positive

374: correlation between degree and betweenness

375: centrality\cite{load1,load2}, that is to say, the node with larger

376: degree will statistically have higher betweenness centrality.

377: Therefore, betweenness centrality is approximately determined by

378: average distance and degree heterogeneity: its average value is

379: determined by the average distance, and its breadth by the

380: heterogeneity of connectivity. Therefore, $B_{max}$ can reflect

381: the influences of both the average distance and degree

382: heterogeneity, which may be the reason why some authors think that

383: $B_{max}$ is a suitable quantity to estimate network

384: synchronizability.

385: 

386: \begin{figure}

387: \begin{center}

388: \scalebox{1.2}[1.2]{\includegraphics{graph6}} \caption{(Color

389: online) The eigenratio $R$ vs maximal betweenness $B_{max}$ when the

390: original network are the WS networks. The eigenratio $R$ is positive

391: correlated with maximal betweenness $B_{max}$ when standard

392: deviation of degree $\sigma$ is fixed. All the data are the average

393: over 20 different realizations.}

394: \end{center}

395: \end{figure}

396: 

397: The insets of figure 6 and 7 respectively show the changes of

398: maximal betweenness with average distance at different standard

399: deviation of degree. For WS network, the increasing of average

400: distance or degree deviation solely always induces the increasing

401: of maximal betweenness. It is consistent with the former analysis.

402: However, for extensional BA network, maximal betweenness is not

403: sensitive to average distance but increases with the increasing of

404: degree deviation clearly.

405: 

406: \begin{figure}

407: \begin{center}

408: \scalebox{1.2}[1.2]{\includegraphics{graph7}} \caption{(Color

409: online) The eigenratio $R$ vs maximal betweenness $B_{max}$ when the

410: original network the extensional BA networks. Clearly, no positive

411: correlation between $R$ and $B_{max}$ can be observed under this

412: case. All the data are the average over 20 different realizations.}

413: \end{center}

414: \end{figure}

415: 

416: By running the random interchanging algorithm, we computing $R$

417: and $B_{max}$ for different configurations, figure 6 and 7 show

418: the correlation between them at fixed $\sigma$ for WS network and

419: extensional BA network, respectively. When the original network is

420: a WS network, there exist strongly positive correlation between

421: $R$ and $B_{max}$ at fixed $\sigma$ (see figure 6), which support

422: the previous conclusion\cite{Betweenness2}. But for different

423: degree deviation, networks shows different $R$ at the same

424: $B_{max}$, although the number of nodes and edges are the same.

425: When the original network is an extensional BA network at fixed

426: $\sigma$, the positive correlation between $R$ and $B_{max}$

427: vanishes. The simulation results show that the single quantity,

428: maximal betweenness $B_{max}$, may not give a comprehensive

429: description of network synchronizability.

430: 

431: \section{Conclusion Remarks}

432: In conclusion, with the help of random interchanging algorithm we

433: show that shorter average distance and homogeneity solely will

434: lead to better synchronizability, but only their combination could

435: make the network easy to synchronize.

436: 

437: Some Numerical studies have been done to check if the maximal

438: betweenness $B_{max}$ is a proper quantity to estimate network

439: synchronizability. The simulation results strongly suggest that

440: the single quantity, $B_{max}$, may not give a comprehensive

441: description of network synchronizability.

442: 

443: It is worthwhile to emphasize that this work is not only of

444: theoretical interest, but also of practical value. The clear

445: picture of topological effects on network synchronizability may

446: provide us a guideline to design algorithm aiming at enhancing or

447: reducing the network synchronizability\cite{zz}.

448: 

449: \subsection*{Acknowledgement}

450: This work is supported by the National Natural Science Foundation of

451: China under Grant No. 10472116, 70471033, 70571074, 10532060,

452: 10547004 and 70271070, the Specialized Research Fund for the

453: Doctoral Program of Higher Education (SRFDP No.20020358009), Special

454: Research Founds for Theoretical Physics Frontier Problems (NSFC

455: Grant No. A0524701), and Specialized Program under President Funding

456: of Chinese Academy of Science.

457: 

458: \begin{thebibliography}{1}

459: \bibitem{Review1} R. Albert and A. -L. Barab\'{a}si, Rev. Mod. Phys. 74 (2002) 47.

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

461: \bibitem{Review3} M. E. J. Newman, SIAM Review 45 (2003) 167.

462: \bibitem{Review4} S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D. -U. Hwang, Phys. Rep. {\bf 424}, 175 (2006).

463: \bibitem{WS} D. J. Watts and S. H. Strogatz, Nature 393 (1998) 440.

464: \bibitem{SFN} A. -L. Barab\'{a}si and R. Albert, Science 286 (1999) 509.

465: \bibitem{Traffic1} B. Tadi\'{c}, S. Thurner, and G. J. Rodgers, Phys. Rev. E 69 (2004) 036102.

466: \bibitem{Traffic2} L. Zhao, Y. -C. Lai, K. Park and N. Ye, Phys. Rev. E 71 (2005) 026125.

467: \bibitem{Traffic3} G. Yan, T. Zhou, B. Hu, Z. -Q. Fu and B. -H. Wang, arXiv: cond-mat/0505366 (Phys. Rev. E In Press).

468: \bibitem{Traffic4} C. -Y. Yin, B. -H. Wang, W. -X. Wang, T. Zhou, and H. -J. Yang, Phys. Lett. A 351 (2006) 220.

469: \bibitem{Traffic5} W. -X. Wang, B. -H. Wang, C. -Y. Yin, Y. -B. Xie, and T. Zhou, Phys. Rev. E {\bf 73}, 026111 (2006).

470: \bibitem{Epidemic1} R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. 86 (2001) 3200.

471: \bibitem{Epidemic2} G. Yan, T. Zhou, J. Wang, Z. -Q. Fu and B. -H. Wang, Chin. Phys. Lett. 22 (2005) 510.

472: \bibitem{Epidemic3} T. Zhou, G. Yan and B. -H. Wang, Phys. Rev. E 71 (2005) 046141.

473: \bibitem{Epidemic4} T. Zhou, Z. -Q. Fu, and B. -H. Wang, arXiv: physics/0508096 (Prog. Natl. Sci. In Press).

474: \bibitem{Epidemic5} Z. -H. Liu, Y. -C. Lai, and N. Ye, Phys. Rev. E 67 (2003) 031911.

475: \bibitem{Epidemic6} Z. -H. Liu, and B. -B. Hu, Europhys. Lett. 72 (2005) 315.

476: \bibitem{Cascade1} A. E. Motter and Y. -C. Lai, Phys. Rev. E 66 (2002) 065102.

477: \bibitem{Cascade2} K. -I. Goh, D. -S. Lee, B. Kahng and D. Kim, Phys. Rev. Lett. 91 (2003) 148701.

478: \bibitem{Cascade3} T. Zhou and B. -H. Wang, Chin. Phys. Lett. 22 (2005) 1072.

479: \bibitem{Easier1} L. F. Lago-Fern\'{a}ndez, R. Huerta, F. Corbacho, and J. A. Sig\"{u}enza, Phys. Rev. Lett. 84 (2000) 2758.

480: \bibitem{Easier2} X. F. Wang and G. Chen, Int. J. Bifurcation Chaos Appl. Sci. Eng. 12 (2002) 187.

481: \bibitem{Easier3} M. Barahona and L. M. Pecora, Phys. Rev. Lett. 89 (2002) 054101.

482: \bibitem{Easier4} P. G. Lind, J. A. C. Gallas, and H. J. Herrmann, Phys. Rev. E 70 (2004) 056207.

483: \bibitem{Zhou2005} T. Zhou, M. Zhao, and B. -H. Wang, Phys. Rev. E {\bf 73}, 037101 (2006).

484: \bibitem{Betweenness1} T. Nishikawa, A. E. Motter, Y. -C. Lai, and F. C. Hoppensteadt, Phys. Rev. Lett. 91 (2003) 014101.

485: \bibitem{Hasegawa} H. Hasegawa, Phys. Rev. E 70 (2004) 066107.

486: \bibitem{Betweenness2} H. Hong, B. J. Kim, M. Y. Choi, and H. Park, Phys. Rev. E 69 (2004) 067105.

487: \bibitem{Cluster} P. N. McGraw and M. Menzinger, Phys. Rev. E 72 (2005) 015101.

488: \bibitem{Betweenness3} D. -S. Lee, Phys. Rev. E 72 (2005) 026208.

489: \bibitem{Oh} E. Oh, K. Rho, H. Hong, and B. Kahng, Phys. Rev. E 72 (2005) 047101.

490: \bibitem{WuXiang} X. Wu, B. -H. Wang, T. Zhou, W. -X. Wang, M. Zhao, and H. -J. Yang, arXiv: con-mat/0511655 (Chin. Phys. Lett. In Press).

491: \bibitem{Reshuffle1} S. Maslov, and K. Sneppen, Science 296 (2002) 910.

492: \bibitem{Reshuffle2} B. J. Kim, Phys. Rev. E 69 (2004) 045101(R).

493: \bibitem{Reshuffle3} B. Wang, H. -W. Tang, T. Zhou, and Z. -L. Xu, arXiv: cond-mat/0512079 (Physica A doi:10.1016/j.physa.2005.12.050).

494: \bibitem{master1} L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. 80 (1998) 2109.

495: \bibitem{master2} L. M. Pecora, and M. Barahona, Chaos and Complexity Letters 1 (2005) 61.

496: \bibitem{HuJK} P. M. Gade, and C. -K. Hu, Phys. Rev. E 62 (2000) 6409.

497: \bibitem{Cluster2} E. Ravasz and A. -L. Barab\'{a}si, Phys. Rev. E 67 (2003) 026112.

498: \bibitem{ExtBA1} S. N. Dorogovtsev, J. F. F. Mendes, and A. N. Samukhin, Phys. Rev. Lett. 85 (2000) 4633.

499: \bibitem{ExtBA2} P. L. Krapivsky, and S. Redner, Phys. Rev. E 63 (2001) 066123.

500: \bibitem{Newman1999} M. E. J. Newman, and D. J. Watts, Phys. Rev. E 60 (1999) 7332.

501: \bibitem{ex1} In one-dimensional ring lattice, the eigenratio $R=1/\sin^{2}(\pi z/N)$, thus it is very hard to synchroniza when $N$ gets large.

502: \bibitem{HandL} R. Cohen and S. Havlin, Phys. Rev. Lett. 90 (2003) 058701.

503: \bibitem{HandZ} T. Zhou, B. -H. Wang, P. -M. Hui, and K. P. Chan, arXiv: cond-mat/0405258 (Physica A doi:10.1016/j.physa.2005.11.011).

504: \bibitem{BetweennessDef1} L. C. Freeman, Sociometry 40 (1977) 35.

505: \bibitem{BetweennessDef2} M. E. J. Newman, Phys. Rev. E 64 (2001) 016132.

506: \bibitem{load1} K. -I. Goh, B. Kahng, and D. Kim, Phys. Rev. Lett. 87 (2001) 278701.

507: \bibitem{load2} M. Barth\'{e}l\'{e}my, Eur. Phys. J. B 38 (2004) 163.

508: \bibitem{zz} M. Zhao, T. Zhou, B. -H. Wang, and W. -X. Wang, Phys. Rev. E 72 (2005) 057102.

509: \end{thebibliography}

510: \end{document}

511: 

512: 

513: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

514: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

515: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

516: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

517: