1: \documentclass[aps,prl,twocolumn,showpacs,floatfix]{revtex4}

2: \usepackage{graphics}

3: \usepackage{epsfig}

4: \usepackage{subfigure}

5: \usepackage{amsmath,amsfonts,amssymb,graphicx}

6: 

7: \begin{document}

8: \title{Synchronization reveals correlation between oscillators on networks}

9: 

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

11: \author{Huijie Yang$^{2}$}

12: %\author{Yi-Cheng Zhang$^{1}$}

13: \affiliation{$^{1}$Department of Physics, University of Fribourg,

14: CH-1700, Fribourg, Switzerland\\

15: $^{2}$ Department of Physics and Center for Computational Science

16: and Engineering, National University of Singapore, Singapore 117542}

17: 

18: \date{today}

19: 

20: \begin{abstract}

21: The understanding of synchronization ranging from natural to social

22: systems has driven the interests of scientists from different

23: disciplines. Here, we have investigated the synchronization dynamics

24: of the Kuramoto dynamics departing from the fully synchronized

25: regime. We have got the analytic expression of the dynamical

26: correlation between pairs of oscillators that reveals the relation

27: between the network dynamics and the underlying topology. Moreover,

28: it also reveals the internal structure of networks that can be used

29: as a new algorithm to detect community structures. Further, we have

30: proposed a new measure about the synchronization in complex networks

31: and scrutinize it in small-world and scale-free networks. Our

32: results indicate that the more heterogeneous and ``smaller" the

33: network is, the more closely it would be synchronized by the

34: collective dynamics.

35: \end{abstract}

36: 

37: \pacs{05.45.Xt, 89.75.Fb}

38: 

39: \maketitle

40: 

41: The last decade has witnessed the rapid explosion of interest in

42: complex networks, which are found in many fields as diverse as

43: biology, technology and social organizations

44: \cite{review1,review2,review3,review4}. The relation between

45: structure and function becomes a key area in the study of complex

46: networks. In the study of networked-dynamics, the emergent

47: synchronization of interacting oscillators generally has occupied a

48: privileged position because of its rich applications in variety of

49: areas ranging from Neuroscience to Sociology

50: \cite{sync1,sync2,sync3,sync4,sync5,sync6,sync7,sync8,sync9,sync10,sync11,sync12,sync13}.

51: In neural systems, the brain dynamics is characterized by

52: synchronization phenomena for a given topology of synapses. And in

53: metabolic networks, hundreds of interconnected biochemical reactions

54: are responsible for the biomass and energy fluxes which are adjusted

55: to optimize the robustness of synchronized behavior. Other example

56: is the synchronized or coordinated behavior of communication

57: patterns in social organizations. The study of synchronization

58: provides us with insights into the key issue: how the collective

59: dynamics couple the relationships between the systematic function

60: and the underlying topology.

61: 

62: One of the most successful attempts to understand synchronization

63: phenomena is due to the Kuramoto model (KM) \cite{KM,KM1}, which is

64: rich enough for many different contexts, including superconducting

65: currents in Josephson junction arrays, emerging coherence in

66: populations of chemical oscillators, and the accuracy of central

67: circadian pacemakers in insects and vertebrates, \emph{etc.}. This

68: model describes a population of $N$ coupled phase oscillators which

69: evolves in time according to the following dynamics

70: \begin{equation}

71: \frac{d\theta_i}{dt} = \omega_i + \sigma

72: \sum_{j}A_{ij}\sin(\theta_{j}-\theta_{i}) + \xi_{i}(t).\label{eq:KM}

73: \end{equation}

74: Here, $\theta_i$ represents the phase of the $i$th oscillator,

75: $\omega_i$ the intrinsic frequency, $\sigma$ the coupling constant,

76: $A_{ij}$ the effective coupling between the oscillators, and

77: $\xi_{i}(t)$ is the white noise due to the complicated environment,

78: with expectation and variance

79: \begin{eqnarray}

80: \langle \xi_i(t)\rangle &=& 0, \nonumber \\

81: \langle \xi_i(t) \xi_j(t') \rangle &=& 2 \delta_{ij} \delta(t-t').

82: \nonumber

83: \end{eqnarray}

84: There are many attractors that each oscillator rotates at its

85: natural frequency when incoherent. As the coupling strength

86: increases, some units become resonant. At sufficiently strong

87: coupling, when coherent, there is only one attractor of the

88: dynamics, and all oscillators rotate at the same frequency,

89: $\overline\omega_i$, the average of $\omega_i$. We call it fully

90: synchronized.

91: 

92: The synchronization problem is solved mainly in the mean-field

93: approach which, unfortunately, is not usually fulfilled in real

94: systems. The nontrivial pattern of connectivity in complex networks

95: brings the non mean-field properties and incorporates many new

96: questions to the research of synchronization. Because of the elegant

97: work of Pecora and Carroll \cite{MSF}, the Master Stability Function

98: (MSF) formalism allows us to analysis the dynamical

99: synchronizability in terms of purely topological porperties,

100: independent of details of unit dynamics. However, the MSF requires

101: several constraints and has limited applications that it only deals

102: with the stability of the exact synchronized states, \emph{i.e.}, it

103: is used to study infinitesimal deviations from the synchronization

104: manifold of chaotic oscillators. However, many realistic dynamics

105: are only close to or even far from the full synchronization.

106: 

107: In this work we study the KM dynamics departing from the fully

108: synchronized state. We get the analytic expression of the dynamical

109: correlation between any two oscillators which reveals the relation

110: between the network dynamics and the underlying topology. It could

111: be used as a new algorithm to detect community structures. Moreover,

112: we propose a new measure about the synchronization in complex

113: networks and scrutinize it in Small-World and Scale-Free networks.

114: 

115: \begin{figure}

116: \scalebox{0.75}[0.9]{\includegraphics{net.eps}}

117: \scalebox{0.85}[0.9]{\includegraphics{cor.eps}}

118: \caption{\label{fig:SFI} Top row: Illustration of the Santa Fe

119: Institute collaboration network. Bottom left: The analytic

120: correlation matrix resulting from Eq. \ref{eq:cor}. Bottom right:

121: The simulated results of the correlation between pairs of

122: oscillators from the Kuramoto model. The colors are a gradation

123: between white (strong correlation) and black (weak correlation).}

124: \end{figure}

125: 

126: When the KM dynamics is close to the attractor of synchronization,

127: the phase differences are small and then the sine coupling function

128: can be approximated linearly, so that we can analyze the linearized

129: dynamics of the Kuramoto model in terms of the Laplacian matrix,

130: \begin{eqnarray}

131: \frac{d\theta_i}{dt} &=& - \sigma \sum_{j}a_{ij}(\theta_{i}-\theta_{j}) + \xi_{i}(t) \nonumber \\

132: &=& - \sigma

133: \sum_{j}L_{ij}\theta_{j} + \xi_{i}(t), \label{eq:Laplace}

134: \end{eqnarray}

135: where

136: $L_{ij}=\delta_{ij}\sum_{m}a_{im}-a_{ij}=\delta_{ij}k_i-a_{ij}$.

137: $a_{ij}$ stands for the adjacency matrix ($1$ if nodes $i$ and $j$

138: are connected and $0$ otherwise), $\delta_{ij}$ is the Kronecker

139: delta and the degree $k_i$ is defined as the number of nodes to

140: which the node $i$ is connected. Thus, we can rewrite the equation

141: in terms of the normal modes,

142: \begin{equation}

143: \frac{d\vartheta_{\alpha}}{dt} = - \sigma

144: \lambda_{\alpha}\vartheta_{\alpha} + \zeta_{\alpha}(t),

145: \end{equation}

146: where $\vartheta_{\alpha} = \sum_{j}\psi_{\alpha j}\theta_{j}$,

147: $\zeta_{\alpha} = \sum_{j}\psi_{\alpha j}\xi_{j}$, and $\psi_{\alpha

148: j}$ denotes the $\alpha$th normalized eigenvector of the Laplacian,

149: $\lambda_{\alpha}$ is the corresponding eigenvalue for

150: $\alpha=0,...,N-1$. Considering $\xi_{i}(t)$ is delta correlated, we

151: can find easily that $\zeta_{\alpha}(t)$ is also delta correlated:

152: $\langle\zeta_{\alpha}(t)\zeta_{\beta}(t')\rangle=2\delta_{\alpha\beta}\delta(t-t')$.

153: 

154: Without loss of generalization, we set $\sigma=1$,

155: $\zeta_{\alpha}(0)=0$. Using stochastic calculus methods

156: \cite{method}, we get

157: \begin{equation}

158: \vartheta_{\alpha}(t) = \int_{0}^t e^{- \lambda_{\alpha} (t-t')}

159: \zeta_{\alpha}(t') dt',

160: \end{equation}

161: so that

162: \begin{eqnarray}

163: \langle \vartheta_{\alpha}(t)^2 \rangle &=& \int_{0}^t \int_{0}^t

164: e^{- \lambda_{\alpha} (t-t')} e^{- \lambda_{\alpha} (t-t'')}\langle \zeta_{\alpha}(t') \zeta_{\alpha}(t'') \rangle dt' dt'' \nonumber\\

165: &=& \int_{0}^t 2 e^{-2 \lambda_{\alpha} (t-t')} dt' \nonumber\\

166: &=& \frac{1}{ \lambda_{\alpha}}(1-e^{-2 \lambda_{\alpha}t})

167: \end{eqnarray}

168: Thus, the time scales of the dynamical relaxation of the eigenmodes

169: are inversely proportional to the corresponding eigenvalues. And for

170: large $t\gg 1/2\lambda_{\alpha}$, we have

171: \begin{equation}

172: \langle \vartheta_{\alpha}(t)^2 \rangle=1/\lambda_{\alpha}.

173: \end{equation}

174: Transforming it back to the basis of $\theta_{i}$ by substituting

175: $\theta_{i}=\sum_{\alpha}\psi_{\alpha i}\vartheta_{\alpha}$, we find

176: the steady-state correlation function:

177: \begin{eqnarray}

178: \langle (\theta_i-\overline \theta) (\theta_j-\overline

179: \theta)\rangle &=& \sum_{\alpha=1}^{N-1} \psi_{\alpha i}

180: \psi_{\alpha j}\langle \vartheta_{\alpha}^2 \rangle \nonumber\\

181: &=& \sum_{\alpha=1}^{N-1} \frac{1}{\lambda_{\alpha}} \psi_{\alpha i}

182: \psi_{\alpha j}, \label{eq:cor}

183: \end{eqnarray}

184: where $\overline \theta = \frac{1}{N}\sum_i \theta_i$ and

185: $\langle...\rangle$ stands for the average over initial random

186: phases. This analytic expression gives the dynamical correlation

187: between any pair of oscillators just in terms of the properties of

188: the underlying topology. Actually, the right term of Eq.

189: \ref{eq:cor} gives the pseudo-inverse of the Laplacian matrix, which

190: acts like a Green Function. It could be used to detect the community

191: structure in various complex networks.

192: 

193: One real world example with known community structure is the

194: collaboration network of the Santa Fe Institute from ref.

195: \cite{Girvan}. The top row of Fig. 1 illustrates the largest

196: component of the collaboration graph, which consists of $118$

197: scientists denoted by nodes. The edge is drawn between a pair of

198: scientists if they coauthored one or more papers. Obviously, the

199: scientists of different interests group in different communities.

200: 

201: In the bottom row of Fig. 1, we illustrate the dynamical correlation

202: between any pairs of nodes on which the KM oscillators is placed.

203: The left one is the analytic results from the relation Eq.

204: \ref{eq:cor} and the right is the results of numerical simulations

205: from Eq. \ref{eq:KM}. They accord with each other very well. In the

206: two figures we can identify the distinct dynamics-based communities

207: wherein nodes become strong correlated in groups, coherently with

208: their topological structure. Moreover, smaller subcommunities can be

209: found clearly in these dynamics-based communities which corresponds

210: with the sketch map of the collaboration network. Although some

211: nodes in the same dynamics-based community are not directly

212: connected, they are topologically equivalent and belong to the same

213: structural level. Therefore, they have strong correlation on

214: dynamics. The dynamics-based groups indicates the different

215: functional modules. When applied to the specific internal structure,

216: which might be the fingerprint of different functional groups for

217: social or biological networks, the dynamical correlation may help us

218: understand better the interplay between systemic structure and

219: function.

220: 

221: \begin{figure}

222: \scalebox{0.8}[0.80]{\includegraphics{k_lamda.eps}}

223: \caption{\label{fig:epsart} (color online). Illustration of the same

224: hierarchical structure in the eigenvalues spectra and degree series.

225: It implies a progressive cascade of the synchronization pattern from

226: the central core to the nodes with smaller degrees. The network data

227: is used as the same as in Fig. \ref{fig:SFI}.}

228: \end{figure}

229: 

230: %=============Also, we can get

231: %\begin{eqnarray}

232: %\langle (\theta_i-\theta_j)^2 \rangle &=& \sum_{\alpha=1}^{N-1}

233: %(\psi_{\alpha i} - \psi_{\alpha j})^2 \langle \vartheta_{\alpha}^2\rangle \nonumber\\

234: %&=& \sum_{\alpha=1}^{N-1} \frac{1}{\lambda_{\alpha}} (\psi_{\alpha

235: %i} - \psi_{\alpha j})^2.

236: %\end{eqnarray}

237: %===========================(This part is not necessary, may be moved to the reference?)

238: 

239: In another way, we can use the mean-field approximation to rewrite

240: the Eq. \ref{eq:Laplace} as:

241: \begin{eqnarray}

242: \frac{d\theta_i}{dt} &=& - \sum_{j}A_{ij}(\theta_{i} - \theta_{j}) +

243: \xi_{i}(t)  \nonumber\\

244: &=& - k_{i}(\theta_{i} - \overline \theta) + \xi_{i}(t).

245: \end{eqnarray}

246: Using the same stochastic calculus method, we get

247: \begin{eqnarray}

248: \langle (\theta_i-\overline \theta)^2\rangle = \frac{1}{k_i}(1-e^{-2

249: k_i t}),

250: \end{eqnarray}

251: which indicates that the time scales of the dynamical relaxation of

252: the observed phases are inversely proportional to the degrees of the

253: corresponding oscillators. The different connectivities in the

254: topology give rise to the corresponding order of the eigenvalues,

255: which are illustrated in Fig. 2. The hubs with the large eigenvalues

256: induce the fast relaxations, \emph{i.e.}, the evolutional pattern of

257: synchronization starts first at the hubs and then pervades almost

258: the whole network in a hierarchical structure across the nodes with

259: smaller degrees. And for large $t\gg1/2k_i$, we have got $\langle

260: (\theta_i-\overline \theta_i)^2\rangle = 1/k_i$, which indicates

261: that hubs are synchronized more closely by the collective dynamics.

262: 

263: For a given realization of networks, we use $R$ to denote the

264: average variance of phase variables which characterizes the

265: collective synchronized ability of the underlying network:

266: \begin{eqnarray}

267: R = \frac{1}{N} \sum^{N}_{i}\langle

268: (\theta_i-\overline\theta)^2\rangle ,

269: \end{eqnarray}

270: The smaller the value of $R$, the smaller the fluctuation of network

271: is, \emph{i.e.}, the more synchronized . Substituting the Eq.

272: \ref{eq:cor} and considering the orthonormality of $\psi_{\alpha

273: i}$, we have

274: \begin{eqnarray}

275: R = \frac{1}{N} \sum_{\alpha=1}^{N-1} \frac{1}{\lambda_{\alpha}}.

276: \label{eq:sync}

277: \end{eqnarray}

278: Therefore, independent of details of unit dynamics, we can

279: characterize the dynamical synchronizability of networked systems

280: just in terms of the spectra of the Laplacian matrix, \emph{i.e.},

281: the purely topological porperties. It is the same as the spirit of

282: the MSF. However, in contrast to the MSF, the new measure Eq.

283: \ref{eq:sync} considers the spectrum in its entirety, not only the

284: extremal eigenvalues. What this new measure emphasizes is on the

285: fluctuation of the process of synchronization, rather than on

286: rigorous bounds for the threshold of desynchronization. In the

287: following, we use this new quantity about synchronization, $R$ to

288: measure the synchronizable ability in small-world (SW) and

289: scale-free (SF) networks, respectively.

290: 

291: \begin{figure}

292: \scalebox{0.8}[0.85]{\includegraphics{SW_SF.eps}}

293: \caption{\label{fig:swsf} (a) The synchronizability of SW networks

294: of size $N=1000$, $\langle k \rangle=8$. The average network

295: distance $D$ and the new measure $R$ decreases with the rewiring

296: probability $p$ in the same trend, while the standard deviation $S$

297: of the degree distribution increases. (b) The synchronizability of

298: SF networks of size $N=1000$, $k_0=4$. The values of $D$ and $R$

299: increase with the scaling exponent $\gamma$, while the values of $S$

300: are decreased. Since less $R$ means more synchronizable, all the

301: results in SW and SF networks indicate that the network would be

302: more synchronizable as it becomes ``smaller" and more heterogeneous.

303: All plots are averaged over 100 realizations.}

304: \end{figure}

305: 

306: We first consider the Watts-Strogatz model of the SW networks

307: \cite{WS}, which is constructed by a rewiring process on a regular

308: ring graph. The probability of rewiring connection is controlled by

309: a parameter $p$, by tuning which the obtained network possesses both

310: short average distance and high heterogeneity that nodes no longer

311: have the same degrees, instead they follow a Poisson distribution.

312: We define $S$ as the standard deviation of the degree distribution

313: and $D$ as the average network distance, averaged over all pairs of

314: nodes. The upper layer of Fig. \ref{fig:swsf}(a) shows the

315: dependence of $S$ and $D$ on the rewiring probability $p$ in the SW

316: network. As $p$ increases, the variance $S$ increases, which implies

317: that the degree distribution becomes more broad and heterogeneous.

318: And as expect, the network distance $D$ decreases as the rewiring

319: probability $p$ (or the heterogeneity of the degree distribution)

320: increases. In the lower layer of Fig. \ref{fig:swsf}(a), $R$ is

321: observed to decrease, which implies enhancement of the

322: synchronizable ability.

323: 

324: Secondly, we implement the same measures on the SF network. The

325: model \cite{SF} is generated by randomly connecting nodes, forcing

326: each nodes $i$ to have connectivity $k_i\geq k_0$ which follows the

327: probability distribution $P(k)\sim k^{-\gamma}$. Note that

328: decreasing $\gamma$ increases $S$ by inducing the longer tail in the

329: connectivity distribution and smaller $\gamma$ gives rise to shorter

330: $D$, as exhibited in the upper layer of Fig. \ref{fig:swsf}(b). And

331: in its lower layer, $R$ is observed to increase as $\gamma$ is

332: increased, which indicates less synchronized when less

333: heterogeneous.

334: % and differs from the results in (T. Nishikawa, A.E.Motter, Y-C Lai, and F.C. Hoppensteadt PRL \textbf{91}, 014101 (2003)).

335: 

336: The results shown in Fig. \ref{fig:swsf} imply that the

337: synchronizable ability on the SW and SF networks is improved as the

338: heterogeneity of the degree distribution is increased or as the

339: average network distance is decreased. In other words, as the

340: network becomes more heterogeneous and ``smaller", with $S$ gets

341: larger and $D$ gets shorter, it would becomes synchronized more

342: closely by the collective dynamics.

343: 

344: In summary, we have investigated the synchronization dynamics of the

345: KM departing from the fully synchronized state. We have got the

346: analytic expression of the dynamical correlation between pairs of

347: oscillators that reveals the relation between the network dynamics

348: and the underlying topology. Moreover, it also reveal the internal

349: structure of networks that can be used as a new algorithm to detect

350: community structures. Further, we have proposed a new measure about

351: the synchronization in complex networks and scrutinize it in SW and

352: SF networks. Our results indicate that the more heterogeneous and

353: ``smaller" the network is, the more closely it would be synchronized

354: by the collective dynamics. We hope our study can provide new

355: insights into the understanding of the role of synchronization

356: between the network structure and function. We also expect it can

357: provide new tools to detect community structure and to analyze the

358: ubiquitous synchronization phenomenon.

359: 

360: 

361: 

362: \begin{thebibliography}{10}

363: \bibitem{review1} S. Strogatz, Nature (London) \textbf{410}, 268 (2001).

364: \bibitem{review2} R. Albert and A.-L. Barabasi, Rev. Mod. Phys. \textbf{74}, 47 (2002).

365: \bibitem{review3} M. E. J. Newman, SIAM Rev. \textbf{45}, 167 (2003).

366: \bibitem{review4} S. Boccaletti, V. Latora, Y. Moreno, M. Chavez,

367: and D.-U. Hwang, Phys. Rep. \textbf{424}, 175 (2006).

368: 

369: \bibitem{sync1}  S. H. Strogatz, {\sl Sync: The Emerging Science of

370: Spontaneous Order} (Hyperion, New York, 2003).

371: \bibitem{sync2} M. Barahona and L. M. Pecora, Phys. Rev. Lett. \textbf{89}, 054101 (2002).

372: \bibitem{sync3} T. Nishikawa, A. E. Motter, Y.-C. Lai, and F. C. Hoppensteadt, Phys. Rev. Lett. \textbf{91}, 014101 (2003).

373: \bibitem{sync4} H. Hong, B. J. Kim, M. Y. Choi, and H. Park, Phys. Rev. E \textbf{69}, 067105 (2004).

374: \bibitem{sync5} Y. Moreno and A. F. Pacheco, Europhys. Lett. \textbf{68}, 603 (2004).

375: \bibitem{sync6} A. E. Motter, C. Zhou, and J. Kurths, Phys. Rev. E \textbf{71}, 016116 (2005).

376: \bibitem{sync7} E. Oh, K. Rho, H. Hong, and B. Kahng, Phys. Rev. E \textbf{72}, 047101 (2005).

377: \bibitem{sync8} L. Donetti, P. I. Hurtado, and M. A. Mu\~{n}oz, Phys. Rev. Lett. \textbf{95}, 188701 (2005).

378: \bibitem{sync9} M. Chavez, D.-U. Hwang, A. Amann, H. G. E. Hentschel, and S. Boccaletti, Phys. Rev. Lett. \textbf{94}, 218701 (2005).

379: \bibitem{sync10} J. G. Restrepo, E. Ott, and B. R. Hunt, Chaos \textbf{16}, 015107 (2006).

380: \bibitem{sync11} C. Zhou and J. Kurths, Chaos \textbf{16},015104 (2006)

381: \bibitem{sync12} A. Arenas, A. Diaz-Guilera, and C. J. Perez-Vicente, Phys. Rev. Lett. \textbf{96}, 114102 (2006).

382: \bibitem{sync13} J. G\'{o}mez-Garde\~{n}es, Y. Moreno, and Alex Arenas, Phys. Rev. Lett. \textbf{98}, 034101 (2007).

383: 

384: \bibitem{KM} Y. Kuramoto, {\sl Chemical Oscillations, Waves and Turbulence}

385: (Springer-Verlag, Berlin, 1984).

386: 

387: \bibitem{KM1} J. A. Acebron, L. L. Bonilla, C. J. Perez Vicente, F.

388: Ritort, and R. Spigler, Rev. Mod. Phys. \textbf{77}, 137 (2005).

389: 

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

391: 

392: \bibitem{method} N. G. Van Kampen, {\sl Stochastic Processes in Physics and

393: Chemistry} (North-Holland, Amsterdam, 1981).

394: 

395: \bibitem{Girvan} M. Girvan and M. E. J. Newman PNAS \textbf{99}, 7821 (2002).

396: 

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

398: 

399: \bibitem{SF} M. E. J. Newman, S. H. Strogatz, and D. J. Watts, Phys. Rev. E \textbf{64}, 026118 (2001).

400: 

401: \end{thebibliography}

402: \end{document}

403: