1: % ****** Start of file apssamp.tex ******
2: %
3: % This file is part of the APS files in the REVTeX 4 distribution.
4: % Version 4.0 of REVTeX, August 2001
5: %
6: % Copyright (c) 2001 The American Physical Society.
7: %
8: % See the REVTeX 4 README file for restrictions and more information.
9: %
10: % TeX'ing this file requires that you have AMS-LaTeX 2.0 installed
11: % as well as the rest of the prerequisites for REVTeX 4.0
12: %
13: % See the REVTeX 4 README file
14: % It also requires running BibTeX. The commands are as follows:
15: %
16: % 1) latex apssamp.tex
17: % 2) bibtex apssamp
18: % 3) latex apssamp.tex
19: % 4) latex apssamp.tex
20: %
21: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
22: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
23:
24: % Some other (several out of many) possibilities
25: %\documentclass[preprint,aps]{revtex4}
26: %\documentclass[preprint,aps,draft]{revtex4}
27: %\documentclass[prb]{revtex4}% Physical Review B
28:
29: \usepackage{graphicx}% Include figure files
30: \usepackage{dcolumn}% Align table columns on decimal point
31: \usepackage{bm}% bold math
32:
33: %\nofiles
34:
35: \begin{document}
36:
37: %\preprint{Ao-Zheng-2005-XXX}
38:
39: \title{Partial Synchronization on Complex Networks}% Force line breaks with \\
40: \author{Bin Ao and Zhigang Zheng\footnote{Corresponding author: zgzheng@bnu.edu.cn.}}
41: \address{
42: Department of Physics and the Beijing-Hong-Kong-Singapore Joint
43: Center for Nonlinear and Complex Systems (Beijing), Beijing Normal
44: University, Beijing 100875, CHINA }
45: %\date{\today}
46:
47:
48: \begin{abstract}
49: Network topology plays an important role in governing the collective
50: dynamics. Partial synchronization (PaS) on regular networks with a few
51: non-local links is explored. Different PaS patterns out of the
52: symmetry breaking are observed for different ways of non-local
53: couplings. The criterion for the emergence of PaS is studied. The
54: emergence of PaS is related to the loss of degeneration in
55: Lyapunov exponent spectrum. Theoretical and numerical analysis
56: indicate that non-local coupling may drastically change the
57: dynamical feature of the network, emphasizing the important
58: topological dependence of collective dynamics on complex networks.
59: \end{abstract}
60:
61: \pacs{05.45.Xt, 87.10.+e.}% PACS, the Physics and Astronomy
62: % Classification Scheme.
63: %\keywords{Suggested keywords}%Use showkeys class option if keyword
64: %display desired
65: \maketitle
66:
67: Synchronization, as a universal cooperative behavior and a
68: fundamental mechanism in nature, has been extensively studied in
69: relating to numerous phenomena in physics, chemistry, and biology
70: \cite{1,2,3,4}. In recent years, there has been a growing interest
71: in the synchronization of spatiotemporal systems, especially in
72: synchronous dynamics on networks. Synchronization on typical
73: complex networks, e.g., on small-world networks \cite{5},
74: scale-free networks \cite{ab,jalan} or weighted
75: network\cite{hwang} have been investigated recently\cite{6}. In
76: spite of these efforts, a number of fundamental questions still
77: remain open. For example, most of previous works have been focused
78: on the onset of global synchronization, while much less was
79: explored for partial synchronization prior to the global case.
80: However, there exist rich synchronous dynamics on complex
81: networks, and many of these behaviors are closely related to
82: partially synchronous motions. Moreover, the mechanism for
83: synchronization on complex networks is still not clear. A good
84: understanding of this issue should be relevant to many collective
85: behaviors in spatiotemporal systems, especially in complex
86: networks.
87:
88: An important feature of complex networks is the existence of
89: non-local links. On the one hand, dynamics on general complex
90: networks (e.g., small-world or scale-free types) are usually
91: difficult to study due to the sophisticated network topology; On
92: the other hand, simple networks, especially regular networks,
93: cannot be ideal models in describing real networks. Therefore, one
94: should find some kind of network that is relatively simple but
95: complex enough to reflect properties of typical complex networks.
96: It is intuitive and feasible to consider a regular network plus
97: only a few non-local links, which forms the simplest "complex"
98: network. This can be regarded as a good bridge between regular
99: networks and general complex networks. In this Letter, we address
100: the issue of synchronization on rings of coupled oscillators with
101: a few shortcut links. We reveal important partial synchronization
102: (PaS) phenomena induced by symmetry breaking in the presence of
103: non-local couplings. Various synchronous patterns are observed for
104: different types of non-local links. This reveals the significant
105: role of network topology in governing the global dynamics. We
106: analyze in detail the conditions for the emergence of PaS and give
107: the criterion of the topological dependence of PaS.
108:
109: We apply certain $m$-dimensional dynamics to nodes in a given network:
110: $\dot{\vec{x}}=\vec{f}(\vec{x}),\vec{x}=(x_1,x_2,\cdots,x_m)$. The
111: network is formed by using the linear coupling as edges. The
112: dynamical equations of the network can be written as
113: \begin{eqnarray}
114: \dot{\vec{X}}=\vec{F}(\vec{X})+\varepsilon\Gamma\otimes C \vec{X},
115: \label{e1}
116: \end{eqnarray}
117: where $\vec{X}=(\vec{x}^1,\vec{x}^2,\cdots,\vec{x}^N)$,
118: $\vec{F}=(\vec{f}^1,\vec{f}^2,\cdots,\vec{f}^N)$, and $N$ is the
119: number of nodes. $\varepsilon$ denotes the coupling strength, and
120: $\Gamma:\mathbb{R}^m \rightarrow \mathbb{R}^m$ characterizes the
121: coupling scheme among the nodes. $C=M-D$, where $M$ is the
122: adjacency matrix of the network (the element $M_{ij}$ denotes the
123: number of the edges that link node $i$ and $j$). $D$ is a diagonal
124: matrix, and $D_{ii}=\sum_{j=1}^{N}M_{ij}$. Therefore
125: $\sum_{j=1}^{N}C_{ij}=0$, and its largest eigenvalue $\lambda_1^C$
126: is 0 (we adopt $\lambda_1\geq\lambda_2\geq\cdots\geq\lambda_N$
127: throughout this Letter).
128:
129: \begin{figure}
130: \includegraphics[width=3.5in,height=1.7in]{figure1.eps}
131: \caption{\label{f1}Different Topologies of regular networks with
132: non-local links. (a): One shortcut leads to five synchronized
133: clusters(10 nodes); (b): One shortcut leads to six clusters (10
134: nodes).(c): One shortcut leads to five clusters (9 nodes). (d):
135: Two parallel shortcuts; (e)Two crossed shortcuts; (f)The
136: lambda-type network.}
137: \end{figure}
138:
139: We first briefly describe the synchronous phenomena we observed in
140: the above networks. Above all, one can define the average distance
141: between two oscillators $i$ and $j$ by $d_{ij}=\lim_{T\to
142: \infty}(T^{-1}\int\|\vec{x_i}-\vec{x_j}\|dt)$, and say they are in
143: a synchronous state in the sense that $d_{ij}\to0$. Denote the
144: number of non-local links by $N_{s}$. When $N_{s}=1$, the
145: synchronous patterns are shown in Figs.~\ref{f1}(a)-(c), where the
146: nodes with the same symbol will synchronize to each other, forming
147: several synchronous clusters. We call the formation of these
148: synchronous clusters "partial synchronization" (PaS). Several
149: interesting phenomena can be observed: (1) PaS emerges not only
150: for oscillators directly linked by the non-local couplings, but
151: also for those without shortcuts; (2) Synchronous clusters are
152: formed by satisfying the mirror symmetry of lattices, i.e., if
153: $(i,j)$ is a pair linked by a shortcut, then $(i+k,j-k)$ are all
154: synchronous pairs. More interestingly, if there are two non-local
155: links, the emergence of PaS also need not only satisfy the mirror
156: symmetry of network, but also depend on the linking form. For
157: example, PaS can always be achieved on networks with "parallel"
158: links [the network with two shortcuts $(i,j)$ and $(i+k,j-k)$], as
159: shown in Fig.~\ref{f1}(d). However, for networks with "cross"
160: shortcuts [two shortcuts $(i,j)$ and $(i+k,j+k)$] shown in
161: Fig.~\ref{f1}(e) and networks with "lambda"-type shortcuts [two
162: shortcuts $(i,i+k)$ and $(j,j+k)$] shown in Fig.~\ref{f1}(f), PaS
163: emerges only for some choices of $k$. If PaS emerges on "cross"
164: and "lambda" networks, synchronization may not emerge between
165: pairs that are directly linked.
166:
167: The above PaS behavior implies the important role of network
168: topology in governing the collective behaviors. It is important to
169: understand the mechanism of the above interesting PaS phenomena.
170: For example, why does PaS appear on networks with mirror symmetry
171: of sites? What is the manifestation in the dynamical exponents at
172: the onset of PaS? Moreover, can we find a simple criterion in
173: exactly determining whether PaS can appear in a given network? It
174: is our central task to answer these important questions.
175:
176: We start with theoretical analysis on the stability of complete
177: synchronization \cite{7}. For the synchronous state
178: $\vec{x}^1(t)=\vec{x}^2(t)=\cdots=\vec{x}^N(t)=\vec{s}(t)$, by
179: linearizing Eq.~(\ref{e1}) near the synchronous manifold
180: $\vec{s(t)}$, i.e., $\vec{x}^i=\vec{s}+\vec{u}^i
181: (i=1,2,\cdots,N)$, one may get
182: %\begin{eqnarray}
183: $\dot{\vec{U}}=[\vec{I}_N\otimes\emph{D}\vec{f}+\varepsilon\Gamma\otimes
184: C ]\vec{U}, \label{e2}$
185: %\end{eqnarray}
186: where $\vec{U}=(\vec{u}^1,\vec{u}^2,\cdots,\vec{u}^N)$ and
187: $\emph{D}\vec{f}$ is the Jacobian function of $\vec{f}$ near the
188: synchronous state $\vec{s}(t)$. Noticing that the first term of
189: the new equation is block diagonal with $\emph{m}\times \emph{m}$
190: blocks, one can diagonalize $C$. Thus it becomes
191: %\begin{eqnarray}
192: $\dot{\vec{v}}_k=[\emph{D}\vec{f}+\varepsilon\lambda_k\Gamma
193: ]\vec{v}_k,(k=1,2,\cdots,N), \label{e3}$
194: %\end{eqnarray}
195: where $\lambda_k$ are eigenvalues of $C$. Then Eq.~(\ref{e1})
196: becomes $\emph{N}$ independent blocks . To judge whether CS can
197: emerge on a network, one needs to find the regime where the
198: largest Lyapunov exponent of the generic equation
199: $\dot{\vec{v}}=[\emph{D}\vec{f}+\alpha\Gamma ]\vec{v}$ is negative
200: on the $\alpha$ plane ($\alpha$ is complex). If all
201: $\varepsilon\lambda_k (k=1,2,\cdots,N)$ are in the
202: negative(stable) regime (except $\lambda_1=0$), then the CS state
203: will be stable. Therefore, the essence of this method is to divide
204: the system into two subsystems which dominate the dynamics on the
205: synchronous manifold(the equation with $\lambda_1=0$) and that on
206: the transversal manifold(the equation with
207: {$\lambda_i$}$(i=2,3,\cdots,N)$), and then the emergence of the CS
208: is determined by the stability of the transversal system.
209:
210: For the case of PaS, one also needs the system to be divided into
211: two parts, and the stability of one of them (the transverse
212: system) determine the emergence of the PaS. On the one hand, the
213: symmetry of a network can be defined by the invariance under a
214: permutation transformation. Taking the case in Fig.~\ref{f1}(a) as
215: an example, its symmetry requires that $F_N C F_N^{-1}=C,$ where
216: $F_{N}$ is a counteridentity matrix that satisfies
217: $F_{i,N-i+1}=1$, for $i=1,2,\cdots,N$ and $F_{ij}=0$ for all
218: $i+j\neq N/2+1$, thus the network has a mirror symmetry. It is
219: easy to verify that PaS is a solution of the differential
220: equation, then each node in a cluster should possess the same
221: dynamics \cite{8}. Therefore the mirror symmetry is needed for the
222: emergence of PaS. On the other hand, one can introduce new
223: variables $\vec{W}=\emph{S}\vec{X}$, where $\vec{W}=(\vec{w}^1,
224: \vec{w}^2, \cdots, \vec{w}^{N}),$
225: \begin{eqnarray}
226: S =\left( \begin{array}{ccc}
227: \emph{I}_{N/2} & F_{N/2} \\
228: \emph{I}_{N/2} & -F_{N/2}
229: \end{array}\right).
230: \label{e4}
231: \end{eqnarray}
232: Here $\emph{I}_{N/2}$ is a $\emph{N}$/2-order identity matrix. For
233: a network with a mirror symmetry, the above transformation can
234: divide the system into two subsystems. Although this requires the
235: adjacent matrix $C$ to be persymmetric (i.e., $C$ is symmetric
236: about its anti-diagonal), the condition can always be satisfied by
237: relabelling the node for a network with a mirror symmetry.
238: Furthermore, if a network satisfies the mirror symmetry and its
239: label obeys the rule in Fig.~(\ref{f1}), the adjacent matrix $C$
240: can be block diagonalized by the transformation (\ref{e4}):
241: $$\emph{G}=\emph{SCS}^{-1}
242: =\left(\begin{array}{cc}
243: A & 0 \\
244: 0 & B
245: \end{array}\right).$$
246: This transformation only re-distributes the eigenvalue spectrum of
247: the matrix $C$ into two blocks $A$ and $B$. Since
248: $\sum_{j=1}^{N/2}A_{ij}=0$, the largest eigenvalue of $A$ is 0.
249:
250: By using the above transformation and considering that we only
251: care about the dynamics near the synchronous manifold, the system
252: can be separated into two parts, and they can be linearized near
253: their synchronous manifold respectively:
254: \begin{eqnarray}
255: \delta
256: \dot{\vec{W}}_{im}=[D\vec{F}(\vec{W}_{im})+\varepsilon\Gamma
257: A]\delta \vec{W}_{im}, \label{e5}
258: \end{eqnarray}
259: \begin{eqnarray}
260: \delta\dot{\vec{W}}_{tv}=[D\vec{F}(\vec{W}_{tv})+\varepsilon\Gamma
261: B] \delta\vec{W}_{tv}, \label{e6}
262: \end{eqnarray}
263: where $\vec{W}_{im}=(\vec{w}^1, \vec{w}^2, \cdots, \vec{w}^{N/2})$
264: and $\vec{W}_{tv}=(\vec{w}^{N/2+1}, \vec{w}^{N/2+2}, \cdots,
265: \vec{w}^{N}).$ For the PaS case discussed above, we have
266: $\vec{W}_{im}=\vec{X}_1+\vec{X}_2,\vec{W}_{tv}=\vec{X}_1-\vec{X}_2$,
267: where $\vec{X}_1=(\vec{x}^1,\vec{x}^2,\cdots,\vec{x}^{N/2})$, and
268: $\vec{X}_2=(\vec{x}^{N/2+1},\vec{x}^{N/2+2},\cdots,\vec{x}^{N})$.
269: Here, $\vec{W}_{im}$ denotes the dynamics on synchronous manifold
270: and $\vec{W}_{tv}$ denotes the dynamics on transversal manifold.
271:
272: Back to Eqs.(\ref{e5}) and (\ref{e6}). If one increase the
273: coupling from 0, synchronization will be achieved for (3) and (4)
274: when $\varepsilon\lambda^{im}_2$ ($\lambda^{im}_2$ is the second
275: largest eigenvalue of $A$) and $\varepsilon\lambda^{tv}_1$
276: ($\lambda^{tv}_1$ is the largest eigenvalue of $B$) respectively
277: fall in the stable regime of the $\alpha$ plane. The
278: synchronization (the stability of synchronous state) of
279: Eq.(\ref{e6}) suggest the emergence of PaS. If
280: $\lambda_2^{im}<\lambda_1^{tv}$, the synchronization of
281: Eq.(\ref{e5}) will be achieved at a smaller coupling than that of
282: Eq.(\ref{e6}), thus a global synchronization state has been
283: achieved when Eq.(\ref{e6}) is stable. Therefore the emergence of
284: PaS requires
285: \begin{eqnarray}
286: \lambda_2^{im}>\lambda_1^{tv}. \label{e7}
287: \end{eqnarray}
288:
289: Now let us test the validity of this criterion and numerically
290: explore the PaS phenomena. We take the Lorenz oscillator as the
291: node of the network: $dx/dt=\sigma(y-x)$, $dy/dt=rx-y-xz$,
292: $dz/dt=xy-bz$. Here $\sigma=10, r=27, b=8/3.$ We adopt the
293: coupling scheme by the 3$\times$3 matrix $\Gamma$ as
294: $\Gamma_{2,1}=1$ and $\Gamma_{i,j}=0$ for $i\neq2,j\neq1$. Let us
295: first try to find all possible networks that satisfy the mirror
296: symmetry. Taking the network with two non-local links for instance
297: (the case of $N_c=1$ can always topologically lead to PaS). For
298: convenience we always take node $i=1$ as one of the ends. The
299: second and the third ends are labelled as $j$ and $k$ ($k>j$),
300: respectively. Due to the mirror symmetry, the label of the fourth
301: end is $n_4=k+j-1$. There are three possible links that hold the
302: mirror symmetry , as shown in Figs.~\ref{f1}(d),(e) and (f).
303: Noticing the translational symmetry of the ring network, we have
304: $2\leqslant j\leqslant N/2$ and $j\leqslant k\leqslant N/2$. Thus
305: by varying $j$ and $k$, all symmetric networks with two non-local
306: couplings can be found [Fig.~\ref{f2}(a)].
307:
308: \begin{figure}
309: \includegraphics[width=3in,height=2.1in]{figure2.eps}
310: \caption{\label{f2}(a): The label rule for $N_s=2$; (b)-(d): The
311: comparison between $\varepsilon_C-\varepsilon_{PaS}$ and
312: $\lambda_2^{im}-\lambda_1^{tv}$, N=20.
313: $\lambda_2^{im}-\lambda_1^{tv}$ are amplified by a proper times;
314: (b): Parallel network, j=3; (c): Cross network, k=9; (d): Lambda
315: network, j=3; (e): Topology graph of the network for $N_s=3$; (f):
316: The same as (b)-(d) for $N_s=3$.}
317: \end{figure}
318:
319: \begin{figure}
320: \includegraphics[width=3.5in,height=1.3in]{figure3.eps}
321: \caption{\label{f3} The PaS regimes (labelled by sparse lines).
322: (a): cross network; (b): lambda network.}
323: \end{figure}
324:
325: We use $\varepsilon_{PaS}$ and $\varepsilon_C$ to represent the
326: critical coupling of PaS and CS, respectively. Figs.\ref{f2}(b)
327: and (c) give a comparison between differences $\Delta \lambda
328: =\lambda_2^{im}-\lambda_1^{tv}$ and $\Delta \varepsilon
329: =\varepsilon_C-\varepsilon_{PaS}$ for different networks. It is
330: instructive to find that $\Delta \lambda$ passes zero when $\Delta
331: \varepsilon$ becomes zero. $\Delta \varepsilon \rightarrow 0$
332: means the coincidence between CS and PaS, implying that PaS
333: doesn't appear. This good correspondence verifies our criterion in
334: (\ref{e7}). Our investigations reveal that one-shortcut networks
335: and "parallel" networks can always achieve PaS because both
336: $\Delta \lambda$ and $\Delta \varepsilon$ are positive for all
337: $k$, as shown in Fig.\ref{f3}(b) for the "parallel" case. For
338: "cross" and "lambda" networks, PaS occurs only for some
339: configurations, as shown in Figs.\ref{f3}(c) and (d). To
340: illustrate these results, we plot the PaS critical line $\Delta
341: \lambda=0$ in the ($2j/N,2k/N$) plane. Fig.\ref{f3}(a) plots the
342: PaS regime for "cross" networks, and Fig.\ref{f3}(b) gives the
343: phase diagram of "lambda" networks. Different lines correspond to
344: different nodes number. These lines coincide with each other
345: pretty well, indicating that PaS depends on the topology of
346: networks and is independent of numbers of nodes.
347:
348: The above criterion for PaS can also be applied to networks with
349: more than two non-local links, while the discussions become more
350: complicated. However, the above phenomena and criterion still
351: remain valid. Fig.\ref{f2}(f) shows one case of three shortcuts.
352: There are 20 nodes in a network, and two non-local links are fixed
353: to be (11,3) and (11,19), and the third link is (11+$m$,11-$m$),
354: where $m=1,2,\cdots,9$, see Fig.\ref{f2}(e). PaS occurs only for
355: the $m=4, 5$ cases, as shown in Fig.\ref{f2}(f).
356:
357: Let's observe the manifestation of the Lyapunov exponent (LE)
358: spectrum (LES) of the network at the onset of PaS. Fig.\ref{f4}(a)
359: gives the first four largest LE's for a ring of 6 nodes without
360: shortcuts. One can find that the third and the fourth largest LE's
361: pass zero at the same coupling ($\varepsilon=\varepsilon_C$),
362: indicating the degeneration of LE's at the onset of CS for ring
363: network. Fig.\ref{f4}(b) is the four largest LE's of a ring
364: network with one non-local link (1,4). Obviously the presence of
365: non-local links leads to a breaking of degeneration of LES. The
366: fourth largest LE passes zero at a very small coupling. It also
367: can be found that $d_{1,4},d_{2,3},d_{5,6}$ become 0 at the same
368: coupling, as shown in Fig.\ref{f4}(c), but
369: $d_{1,2},d_{3,4},d_{1,6},d_{4,5}\neq 0$. Therefore, PaS emerges if
370: and only if there only are two positive LE's. This conclusion was
371: tested to be true for different node numbers and different
372: shortcut numbers and link ways.
373:
374: These results can be interpreted by analyzing eigenvalues of the
375: adjacent matrix. The adjacent matrix of a ring network is a
376: circulant matrix, its eigenvalues are conjugated besides the
377: largest and the smallest ones. So when the coupling increases,
378: $\varepsilon\lambda_2$ and $\varepsilon\lambda_3$ will fall into
379: the stable regime at same critical coupling, above which the third
380: and fourth LE's corresponding to $\lambda_2$ and $\lambda_3$
381: become negative (there are always at least one positive LE and one
382: zero LE for a chaotic system). When non-local couplings are
383: applied, the degeneration in the eigenvalues and LES is lost.
384: Since the emergence of PaS requires
385: $\lambda_2^{im}>\lambda_1^{tv}$, when $\varepsilon\lambda_1^{tv}$
386: is in stable regime, there are only two
387: eigenvalues($\lambda_1^{im}=0$ and $\lambda_2^{im}$) outside of
388: the stable regime. That is, there are only two positive LE's. One
389: can also calculate the conditional LE spectrum of the synchronous
390: manifold (Eq.(\ref{e5})) and its corresponding transversal
391: manifold (Eq.(\ref{e6})). It can be found from Fig.\ref{f4}(d)
392: that there are only two positive conditional LE's of synchronous
393: state(solid line) when the largest conditional LE of the
394: transversal manifold (dotted line) passes zero.
395:
396: \begin{figure}
397: \includegraphics[width=3in,height=2in]{figure4.eps}
398: \caption{\label{f4} (a): The first four largest LEs of a ring
399: network; (b): The first four largest LE's of a ring network with
400: one shortcut (1,4); (c): average trajectory distances
401: $d_{1,4},d_{2,3},d_{5,6}$ (solid lines), and
402: $d_{1,2},d_{3,4},d_{1,6},d_{4,5}$ (dashed lines) varying with the
403: coupling, where the network is the same as in (b); (d): The three
404: largest LE's of the synchronous manifold (solid lines) and the
405: largest LE of the transversal manifold.}
406: \end{figure}
407:
408: In conclusion, in this letter we explored synchronization
409: behaviors of complex networks with a few non-local links. The
410: criterion for the emergence of PaS is theoretically given and
411: verified in numerical simulations. The manifestation of Lyapunov
412: exponent spectrum is discussed for PaS. Our results and criterion
413: proposed in this Letter are general, and they are independent of
414: the dynamics on sites and coupling forms. we tested different
415: dynamics on the nodes, such as Lorenz oscillators, Rossler
416: systems, and nonlinear maps (e.g., the logistic map and circle
417: maps). All these numerical experiments support our criterions and
418: conclusions. These "simple" complex networks are not only simple
419: but also typical, and it can be a useful tool to study more
420: complex networks. The synchronization behavior we studied in this
421: letter reveal the importance of network topology. Different types
422: of shortcuts may drastically change the topology and furthermore
423: the dynamics of a network. Our work should illuminate a good step
424: in understanding the synchronization processes on general complex
425: networks from a dynamical point of view.
426:
427: \begin{acknowledgments}
428: We thank Prof. Gang Hu for constructive comments and suggestions,
429: and we are also grateful to Drs. H.Chen, Z.Cao, N.Zheng and X.Feng
430: for very useful discussions and help.% This work is supported in
431: %part by the key NNSF of China, the FANEDD, the TRAPOYT in Higher
432: %Education Institutions of MOE, and the Foundation of Doctoral
433: %Training.
434: \end{acknowledgments}
435:
436: \begin{thebibliography}{99}
437: \bibitem{1}L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. 64,
438: 821(1990); A. Pikovsky, M. Rosenblum and J. Kurths,
439: \textit{Synchronization: A universal concept in nonlinear
440: sciences} (Cambridge University Press, Cambridge, 2001); S.
441: Boccaletti, C. Grebogi, Y. C. Lai, H. Mancini, and D.Maza, Phys.
442: Rep. 329, 103 (2000). S. Boccaletti, J. Kurths, G. Osipov, D. L.
443: Valladares, C. S. Zhou, Phys. Rep. 366, 1 (2002).
444: \bibitem{2}M. G. Rosenblum, A. S. Pikovsky, and J. Kurths Phys. Rev. Lett. 76, 1804
445: (1996).
446: \bibitem{3}Z. Zheng, G. Hu, and B. Hu Phys. Rev. Lett. 81, 5318
447: (1998). U. Ernst \textit{et al.}, \textit{ibid} 74, 1570 (1995).
448: G. Hu \textit{et al.}, \textit{ibid} 85, 3377 (2000).
449: \bibitem{4}N. Parekh, S. Parthasarathy, and S. Sinha Phys. Rev. Lett. 81, 1401
450: (1998).
451: \bibitem{5}D. J. Watts and S. H. Strogatz, Nature (London) 393, 440 (1998).
452: \bibitem{ab}S.H. Strogatz, Nature 410, 268 (2001); R. Albert and A.L. Barabasi, Rev. Mod. Phys. 74, 47 (2002);
453: M.E.J. Newman, SIAM Rev. 45, 167(2003).
454: \bibitem{jalan}S. Jalan and R. E. Amritkar Phys. Rev. Lett. 90 014101 (2003).
455: \bibitem{hwang}D.-U. Hwang, M. Chavez \textit{et al.},
456: Phys. Rev. Lett. 94, 138701 (2005). M. Chavez, D.-U. Hwang
457: \textit{et al.}, Phys. Rev. Lett. 94, 218701 (2005).
458: \bibitem{6}M. E. J. Newman \textit{et al.}, Phys. Rev. Lett. 84, 3201
459: (2000); L. F. Lago-Fernandez \textit{et al.}, \textit{ibid} 84,
460: 2758 (2000); M. Barahona and L. M. Pecora, \textit{ibid} 89,
461: 054101 (2002); G. W. Wei \textit{et al.}, \textit{ibid} 89,
462: 284103(2002); T. Nishikawa \textit{et al.}, \textit{ibid} 91,
463: 014101(2003); F. M. Atay \textit{et al.}, \textit{ibid} 92,
464: 144101(2004); Y. Jiang \textit{et al.}, Phys. Rev. E 68,
465: 065201(R)(2003); J. G. Restrepo \textit{et al.}, Phys. Rev. E 69,
466: 066215(2004); Y. Moreno and A. F. Pacheco, Europhys. Lett. 68,
467: 603(2004).
468: \bibitem{siam}I. Stewart, M Golubitsky and M. Pivato, SIAM J.
469: Applied Dynamical Systems 2(4) 609 (2003)
470: \bibitem{7}J. Yang \textit{et al.}, Phys. Rev. Lett. 80, 496
471: (1998); L. M. Pecora and T. L. Carroll, \textit{ibid} 80, 2109
472: (1998).
473: \bibitem{8}Y. Zhang \textit{et al.}, Phys. Rev. E 63,026211(2001).
474: \end{thebibliography}
475:
476:
477: \end{document}
478: %
479: % ****** End of file PaS.tex ******
480: