Frequency synchronization in random oscillator network
Extracted source text ยท arXiv cond-mat/0406580
% $Id: letter.tex,v 1.8 2004/06/02 07:23:03 miya Exp $
\documentclass[pre,preprint]{revtex4}
\usepackage[dvipdfm]{graphics}
\usepackage{subfigure}
\begin{document}
\title{Frequency synchronization in random oscillator network}
\author{Takashi Ichinomiya}
\affiliation{Meme Media laboratory, Hokkaido Univ. Sapporo, Hokkaido, Japan}
\email{miya@aurora.es.hokudai.ac.jp}
\date{12 Nov 2003}
\begin{abstract}
We study the frequency-synchronization of randomly coupled oscillators. By
analyzing the continuum limit, we obtain the sufficient condition for the
mean-field type synchronization. We especially find that the critical
coupling constant $K$ becomes 0 in the random scale free network,
$P(k)\propto k^{-\gamma}$, if $2 < \gamma \le 3$. Numerical simulations
in finite networks are consistent with these analysis.
\end{abstract}
\pacs{89.75.Hc, 05.45.Xt}
\maketitle
\section{Introduction}
Recently, it has become clear that the complex network plays an important
role in many natural and artificial systems, such as neural network,
metabolic systems, power supply system, Internet, and so on
\cite{Albert,Dorogovtsev}.
In particular, we have recognized that many networks have scale-free topology;
the distribution of the degree obeys the power law,
$P(k)\sim k^{-\gamma}$.
The study of the scale-free network now attracts the interests of many
researchers in mathematics, physics, engineering and biology.
The dynamics in the network systems is one of the important themes of
the investigation of complex network.
In this paper, we study the synchronization of the random network of
oscillators. The phase synchronization in complex network has been
studied by several authors\cite{Timme, Wang}, while
the frequency synchronization has not been studied so much.
One of the important work on this problem was made by
Watts\cite{Watts}. He suggested from numerical simulation that
mean-field type synchronization
occurs in small-world network such as Watts-Strogatz model. His study was
followed by the work of Hong {\it et al.}, in which phase-diagram and
critical exponent are numerically studied in detail\cite{Hong}.
These works showed that mean-field type synchronization, that
Kuramoto observed in globally-coupled oscillators\cite{Kuramoto}, appears
also in the small-world network. However, such a study in scale-free
network has not been performed yet.
In this paper, we analytically study the frequency synchronization
in the random network of oscillators. By analyzing the
continuum limit of this model, we obtain the sufficient condition for
the synchronization.
Our result shows that in the scale-free random network the threshold for
synchronization is absent if $2 < \gamma \le 3$. We also carry out the
numerical simulations and the results are consistent with this analysis.
This paper is constructed as follows: the next section describes the
model of oscillator network and derive the continuum
limit equation. Section \ref{analysis} is devoted to derive the
sufficient condition for the synchronization from the continuum limit
equation. We show that the order parameter is different from the one
used in previous works, and especially we conclude that the threshold
for the synchronization disappears in the random
scale-free network. These
results are consistent with the results of the numerical simulations,
which are described in
section \ref{simulation}. In the last section we make a summary of this
paper and discuss the relation to the other properties of the scale-free
network.
\section{Oscillator network model and its continuum limit}
First we describe the model we study in this paper. We study the network
with N-nodes. At each nodes, there exists a oscillator and the phase of the
oscillator $\theta_i$ is developed as
\begin{equation}
\frac{\partial \theta_i}{\partial t}=\omega_i + K\sum_j a_{i,j}\sin(\theta_j-\theta_i).\label{problem}
\end{equation}
where $K$ is the coupling constant, $a_{i,j}$ is 1 if the nodes
$i$ and $j$ are connected, and 0 otherwise. $\omega_i$ is a random
number, whose distribution is given by the function $N(\omega)$.
For the analytic study, it is convenient to use the continuum limit
equation. We define $P(k)$ as the
distribution of nodes with degree $k$, and $\rho(k,\omega; t,\theta)$
the density of oscillators with phase $\theta$ at time $t$, for given
$\omega$ and $k$.
We assume that $\rho(k,\omega; t,\theta)$ is normalized as
\begin{equation}
\int_0^{2 \pi} \rho(k,\omega; t, \theta) d \theta = 1.
\end{equation}
For simplicity, we assume $N(\omega)=N(-\omega)$. Under this assumption,
we suppose that the collective oscillation corresponds to the
stable solution,
$\frac{d \rho}{dt} = 0$, in this model.
Now we construct the continuum limit equation for the network of
oscillators.
The evolution of $\rho$ is determined by the continuity equation $\partial \rho /\partial t = -\partial (\rho v)/\partial \theta$, where $v$ is defined by
the continuum limit of the r.h.s of eq.(\ref{problem}). Because one randomly
selected edge connects to the
node of degree $k$, frequency $\omega$, phase $\theta$ with the
probability $kP(k)N(\omega)\rho(k,\omega;t,\theta)/\int dk k P(k)$,
$\rho(k,\omega; t, \theta)$
obeys the equation
\begin{equation}
\frac{\partial \rho(k,\omega; t,\theta)}{\partial t} = -\frac{\partial}
{\partial \theta}\left[\rho(k,\omega;t,\theta)\left(\omega+ \frac{Kk\int
d\omega^{\prime}\int dk^{\prime}\int d \theta^{\prime}N(\omega^{\prime})
P(k^{\prime})k^{\prime}
\rho(k^\prime,\omega^{\prime};t,\theta^{\prime})\sin(\theta-\theta^{\prime})}{\int dk^{\prime} P(k^{\prime})k^{\prime}}\right)\right].\label{master}
\end{equation}
In the next section, we study the mean-field solution of this equation.
\section{Mean-field analysis of random oscillator network\label{analysis}}
In this section, we study the sufficient condition for the
synchronization using eq.(\ref{master}).
First we introduce order parameter $(r,\psi)$ as
\begin{equation}
re^{i \psi} = \int d\omega \int dk \int d\theta N(\omega)P(k)k \rho(k,\omega;t,\theta) e^{i \theta}/\int dk P(k)k.\label{mf}
\end{equation}
This order parameter is different from the one which is used in previous work
in small-world model\cite{Watts,Hong}. In the previous works,
$\sum_i e^{i \theta_i}/N$ is used for the mean-field, while
our order parameter corresponds to $\sum_i k_i e^{i \theta_i}/\sum_i k_i$,
where $k_i$ is the degree of the node $i$.
However, from eq.(\ref{master}) it seems natural to use eq.(\ref{mf}) as the
mean-field value the in random network. Here we note that $0\leq r \leq 1$.
Inserting eq.(\ref{mf}) into eq.(\ref{master}), we get
\begin{equation}
\frac{\partial\rho(k,\omega; t,\theta)}{\partial t} =-\frac{\partial}{\partial \theta}(\rho(k,\omega;t,\theta)(\omega + K k r \sin(\psi-\theta))).\label{mf-eq}
\end{equation}
The time-independent solution of $\rho$ is then
\begin{equation}
\frac{\partial}{\partial \theta}(\rho(k,\omega;t,\theta)(\omega + K k r \sin(\psi-\theta))) = 0.
\end{equation}
Without a loss of generality, we can assume $\psi=0$.
Since we want to seek the solution which corresponds to the Kuramoto's solution
in globally coupled oscillators, we assume the solution of this equation as
\begin{equation}
\rho(k,\omega;\theta)=\left\{
\begin{array}{ll}
\delta(\theta - \arcsin(\frac{\omega}{Kkr})) & \mbox{ if } \frac{|\omega|}{Kkr} \le 1 \\
\frac{C(k,\omega)}{|\omega-Kkr\sin\theta|} & \mbox{ otherwise,}
\end{array}
\right.\label{density}
\end{equation}
where $C(k,\omega)$ is the normalization factor. Here we note that $\rho$
depends on both $K$ and $k$. This equation means that $Kk$
corresponds to the coupling between mean-field and the oscillator.
Inserting eq.(\ref{density}) into eq.(\ref{mf}), we get the equation for $r$,
\begin{equation}
r = \int d\omega\int dk\int d\theta N(\omega)kP(k)\rho(k,\omega;\theta)e^{i \theta}/\int dk k P(k).
\end{equation}
To calculate this integral, first we divide the integral over $\omega$.
\begin{eqnarray}
\int d\omega\int dk\int d\theta
N(\omega)kP(k)\rho(k,\omega;\theta)e^{i \theta} &=& \int dk \int d
\theta
\left(\int_{-Kkr}^{Kkr}d\omega+\int_{-\infty}^{-Kkr}d\omega+\int_{Kkr}^{\infty}
d\omega\right)\nonumber\\
&\times& N(\omega)kP(k)\rho(k,\omega;\theta)e^{i\theta} \label{mf-eq-2}
\end{eqnarray}
The contribution from the integral at $\omega< -Kkr$ and
$\omega > Kkr$ is 0 if $N(\omega)= N(-\omega)$, because
\begin{equation}
(\int_{-\infty}^{-Kkr} d \omega+\int_{Kkr}^{\infty}d \omega )
N(\omega)\rho(k,\omega;\theta)e^{i\theta}=\int_{Kkr}^{\infty}N(\omega)e^{i\theta}C(k,\omega)(\frac{1}{\omega-Kkr\sin\theta}+\frac{1}{\omega+Kkr\sin\theta}).\label{for-0}
\end{equation}
The integral of the r.h.s of eq.(\ref{for-0}) over $\theta$ equals to 0.
Therefore eq.(\ref{mf-eq-2}) is equivalent to
\begin{equation}
r = \int dk\int_{-Kkr}^{Kkr} N(\omega)k P(k)\exp(i \arcsin(\frac{\omega}{Kkr}))/\int dk k P(k).
\end{equation}
If we assume $\arcsin(\frac{\omega}{Kkr})$ is between $[-\pi/2,\pi/2]$, we get
\begin{eqnarray}
r \int dk k P(k) &=& \int dk \int_{-Kkr}^{Kkr} d\omega N(\omega)k P(k) \sqrt{1-\left(\frac{\omega}{Kkr}\right)^2}\nonumber\\
&=& \int dk \int_{-1}^{1} d\omega^{\prime}k P(k)N(Kkr\omega^{\prime})\sqrt{1-\omega^{\prime 2}}\times Kkr\nonumber\\
&=& Kr \int dk k^2 P(k)\int _{-1}^{1}d\omega^{\prime}N(Kkr\omega^{\prime})\sqrt{1-\omega^{\prime 2}}.\label{finaleq}
\end{eqnarray}
If $r \neq 0$, we get
\begin{equation}
\int dk k P(k) = K \int dk k^2 P(k)\int _{-1}^{1}d\omega^{\prime}N(Kkr\omega^{\prime})\sqrt{1-\omega^{\prime 2}}
\label{final2}
\end{equation}
The l.h.s of this equation is independent of $r$ and we define the
r.h.s of this equation as $f(r)$.
At $r=1$, $f(r)$ is not larger than $\int d k k P(k) $,
because
\begin{eqnarray}
\int dk k^2 P(k)\int_{-1}^{1} d\omega^{\prime} N(Kkr)\sqrt{1-\omega^{\prime 2}}
&\le& \int dk k^2 P(k)\int_{-1}^{1} d\omega^{\prime} N(Kkr\omega^{\prime})\nonumber\\
&\le& \int dk k^2 P(k) \frac{1}{Kkr}\int_{-\infty}^{\infty} d\omega^{\prime\prime}N(\omega^{\prime\prime})\nonumber\\
&=& \frac{\int dk kP(k)}{Kr},
\end{eqnarray}
here we use the relation $\int_{-\infty}^{\infty} d\omega N(\omega)=1$.
Therefore
the sufficient condition that eq.(\ref{finaleq}) have solution at
$0 < r \le 1 $
is that $f(r) > \int dk k P(k)$ at $r=0$,
\begin{equation}
\frac{KN(0)\pi\int dk k^2 P(k)}{2 \int dk k P(k)} > 1.\label{critical}
\end{equation}
This is the sufficient condition for the synchronization in random
network of oscillators.
The most impressive point of this equation is that in the random scale-free
network, $P(k)\propto k^{-\gamma}$,
this condition is satisfied for any $K > 0$ if $2 < \gamma\le 3$, because
$\int dk k^2 P(k)/\int dk k P(k)$ diverges. Therefore we have no
threshold for the synchronization in the random scale-free
network. This seems similar to the absence of the threshold in
susceptible-infected-susceptible(SIS) model\cite{SIS0}. We will discuss
on this similarity in later section.
In this section, we derive the sufficient condition for the
synchronization in random network of oscillators, using the continuum
limit equation. In the next section, we show that the analysis above is
in good agreement with the results of the numerical simulations.
\section{Numerical simulation of synchronization\label{simulation}}
In this section, we show the result of the numerical simulations of the
random network of oscillators. In the all simulations,
we take $N(\omega)$ as $N(\omega)=0.5$ if $-1.0< \omega< 1.0$, and 0
otherwise.
First we show the result on the 1000-node
Erd\H{o}s-R\'enyi random network model.
We choose the probability of coupling $p=0.005$, which gives
$\int dk kP(k)=5.0$ and $\int dk k^2 P(k) = 29.7$
on average.
In this case, estimated critical $K$ is $K_c =0.214$. Each
simulation is carried out 100 times.
In Fig.\ref{omega-vel-randomnet} we plot the relation between $\omega_i$ and
$\frac{d \theta_i}{dt}$ after a long time ($t=200$) when $K=0.15$ and $0.30$.
In the case of $K=0.15$, $d\theta/dt$ seems to depend on $\omega$ linearly.
On the other hand at $K=0.30$ many oscillators seem to be synchronized at
$d\theta/dt = 0$. This figure strongly suggests that the synchronization occurs
between $K=0.15$ and 0.30.
\begin{figure*}[t]
\mbox{
% \subfigure[$K=0.15$]{\resizebox{.4\textwidth}{!}{\includegraphics{k0.15wv-random.eps}}}
% \subfigure[$K=0.30$]{\resizebox{.4\textwidth}{!}{\includegraphics{k0.30wv-random.eps}}
\subfigure[$K=0.15$]{\resizebox{.4\textwidth}{!}{\includegraphics{figure1a.ps}}}
\subfigure[$K=0.30$]{\resizebox{.4\textwidth}{!}{\includegraphics{figure1b.ps}}
}
}
\caption{$(\omega,d\theta/dt)$ distribution of oscillators in random
network, for K=0.15 and K= 0.30.\label{omega-vel-randomnet}}
\end{figure*}
We plot the relation between $\omega$
and $\theta$ for $K=0.15$ and $K=0.30$ in
Fig. \ref{omega-theta-randomnet}. We find a clear difference
between these two cases. In the case of $K=0.30$, the distribution of
$(\omega_i,\theta_i)$ is apparently non-uniform, while at $K=0.15$ we
cannot find any structure. In the case of $K=0.30$, $\theta$ seems to depend
linearly on $\omega$. However, from the previous analysis we suggest that
$\theta$ depends on both $\omega$ and $k$. To clarify
the degree dependence, we plot $(\omega,\theta)$ for the nodes with
the degree $k=$ 3, 5 and 7 at $K=0.30$ in Fig. \ref{omega-theta-randomnet-k}
. We also plot $\arcsin(\omega/Kkr)$ in these figures. The average of
$r$ is
0.623 in our simulation. From these figures, we find that
distribution of $(\omega,\theta)$ seems to be concentrated around a single
line. The concentration line qualitatively coincides with
$\theta=\arcsin(\omega/Kkr)$. This result suggests that our mean-field
defined by eq.(\ref{mf}) is the correct one.
\begin{figure*}[t]
\mbox{
% \subfigure[$K=0.15$]{\resizebox{.4\textwidth}{!}{\includegraphics{k0.15-random.eps}}}
% \subfigure[$K=0.30$]{\resizebox{.4\textwidth}{!}{\includegraphics{k0.30-random.eps}}}
\subfigure[$K=0.15$]{\resizebox{.4\textwidth}{!}{\includegraphics{figure2a.ps}}}
\subfigure[$K=0.30$]{\resizebox{.4\textwidth}{!}{\includegraphics{figure2b.ps}} }
}
\caption{$(\omega,\theta)$ distribution of oscillators in random network, K=0.15 and K= 0.30.\label{omega-theta-randomnet}}
\end{figure*}
\begin{figure*}[t]
\mbox{
% \subfigure[$k=3$]{\resizebox{.3\textwidth}{!}{\includegraphics{k0.30-random-k3.eps}}}
% \subfigure[$k=5$]{\resizebox{.3\textwidth}{!}{\includegraphics{k0.30-random-k5.eps}}}
% \subfigure[$k=7$]{\resizebox{.3\textwidth}{!}{\includegraphics{k0.30-random-k7.eps}}}
\subfigure[$k=3$]{\resizebox{.3\textwidth}{!}{\includegraphics{figure3a.ps}}}
\subfigure[$k=5$]{\resizebox{.3\textwidth}{!}{\includegraphics{figure3b.ps}}}
\subfigure[$k=7$]{\resizebox{.3\textwidth}{!}{\includegraphics{figure3c.ps}}}
}
\caption{$(\omega,\theta)$ distribution of oscillators with degree 3, 5 and 7
in random network, K= 0.30.\label{omega-theta-randomnet-k}}
\end{figure*}
To estimate the critical coupling $K_c$, we plot $K$ dependence of the
average of the order parameter $r_{\mbox{av.}}$ in
Fig.\ref{K-r}. $r_{\mbox{av.}}$ is less than 0.1
and shows weak dependence on $K$ at $K<0.2$. This
non-zero value of $r_{\mbox{av.}}$ is due to the finite-size effect.
On the other hand, at $K>0.2$, $r_{\mbox{av.}}$
increases rapidly as the interaction increases. This figure suggest that
$K_c$ is about 0.2, which is in agreement with our
analysis. Therefore we conclude that all numerical results are consistent with
our analysis.
\begin{figure}[t]
% \resizebox{.4\textwidth}{!}{\includegraphics{mfvalue.eps}}
\resizebox{.4\textwidth}{!}{\includegraphics{figure4.ps}}
\caption{Interaction dependence of mean-field parameter $r_{\mbox{av.}}$.\label{K-r}}
\end{figure}
From these simulation, we find that our mean-field theory is applicable to
Erd\H{o}s-R\'{e}nyi model. However, the most impressive suggestion of our
analysis is the absence of the threshold in the random scale-free
network. In the following, we show the result of simulation in random
scale-free network with $\gamma=2.5$.
In Fig. {\ref{scalefree-r}}, we show the relation between order
parameter $r$ and coupling constant $K$ for
N=500, 1000, 2000 and 4000. At N=500 , $r_{\mbox{av.}}$ rapidly increase above
$K\sim 0.16$, which is
qualitatively consistent with the $K_c\sim 0.175$ estimated from
eq.(\ref{finaleq}). As the network size increases, $r_{\mbox{av.}}$ at small
coupling
decreases, which suggests that the finite $r_{\mbox{av.}}$ at small coupling
is the finite-size
effect. The order parameter begin to increases rapidly above $K_c$. We
note that by increasing the system size,
the increase of order parameter begins at smaller coupling. This
means that the critical coupling $K_c$ decreases as the system
size increases. We also show $K_c$ estimated from eq.(\ref{finaleq}) in
this figure. The estimated $K_c$ qualitatively coincide with the
coupling constant at which the order parameter increases rapidly. We
conclude that our analysis and the results of the numerical simulation
show a good agreement also in the random scale-free network. These results
suggest that in the infinite size scale-free network the critical
coupling constant $K_c$ becomes zero, just same as the continuum limit
equation.
To compare the results of the numerical simulation and the analysis
more precisely, we need a more accurate estimation of $K_c$ from the
numerical simulation. In the case of the globally coupled networks and
Watts-Strogatz model, $K_c$ is numerically obtained as the point at which
$N^{0.25} r_{\mbox{av.}}$ becomes independent of the size of the
network\cite{Hong}. In their analysis, there exists an assumption that
$K_c$ does not depend on the size of the network. On the other hand,
our analysis and simulation show that $K_c$ depends clearly on the size of
the network through the average of the square of the degrees. Therefore
it is impossible to obtain accurate $K_c$ from finite-size analysis.
The exact estimation of $K_c$ is a difficult task.
However, we find that $K_c$ derived from eq.(\ref{finaleq}) seems to have
a strong relation to the phase transition. We rescale $K$
by $K_c$ which is obtained from eq.(\ref{finaleq}), and plot the
relation between $N^{0.25} r_{\mbox{av.}} $ and $K/K_c$ in
Fig.\ref{scalefree-finite-size-scaling}. In the case of
$N=1000, 2000$ and $4000$,
the well-defined crossing point exists at $K/K_c=1$. In
the case of $N=500$, $N^{0.25}r_{\mbox{av.}}$ at $K/K_c=1$ is a little
larger than in the other cases. However, this difference is small, and it
seems that $K/K_c=1.0$ is the crossing point at large $N$. This result
is similar to the results of the finite-size scaling in the
globally coupled networks and Watts-Strogatz model. In these models,
there exists a crossing point at $K=K_c$.
On the other hand, our analysis is not a precise determination of the
critical coupling strength. To avoid the size dependence of $K_c$, we
rescale $K$ and we have no guarantee that such a rescaling is
valid for scale-free network model. However, our result strongly
suggests that $K_c$ obtained from the numerical simulation coincides with
the result of the analytic solution.
To conclude this section, we carried out the simulations on
Erd\H{o}s-R\'{e}nyi model and the random scale-free network. All the
results of these simulations show a qualitative agreement with the
analysis in the previous section.
\begin{figure}[t]
\resizebox{.4\textwidth}{!}{\includegraphics{figure5.ps}}
\caption{Interaction dependence of the mean-field parameter in the random
scale-free network for N=500,1000,2000 and 4000. The arrow shows $K_c$ estimated from
eq.(\ref{finaleq}). Simulations for each parameters are carried out at
least 50 realization of the networks.\label{scalefree-r}}
\end{figure}
\begin{figure}[t]
\resizebox{.4\textwidth}{!}{\includegraphics{figure6.ps}}
\caption{The relation between $N^{0.25} r_{\mbox{av.}}$ and $K/K_c$, where
$K_c$ is the value estimated from eq.(\ref{finaleq}).\label{scalefree-finite-size-scaling}}.
\end{figure}
\section{Summary and Discussion}
In this paper, we study the frequency synchronization of the random oscillator
network.
By analyzing the continuum limit equation, we find that mean-field type
synchronization occurs in
random network model. We obtain the sufficient condition for the
synchronization. Especially we find that
the threshold for the synchronization is absent in scale-free random
network if $2 < \gamma \le 3$.
The results of numerical simulations are in good agreement with this
analysis.
One of the most astonishing result in the dynamics of scale-free network is
the absence of epidemic threshold in SIS model.
Our result seems to be similar to the result in SIS model,
however, there exists a large difference between them.
In SIS model the absence of epidemic threshold is the result of the
divergence of
$k_{\mbox{nn}}$, the mean degree of the nearest neighbor
nodes\cite{SIS}. On the other
hand,
in our model the absence of threshold originates from the degree-dependence
of
the coupling between order parameter and oscillators.
The coupling between the oscillators and the mean-field
is proportional to the degree of the nodes, as shown in eq.(\ref{mf-eq}), and
the contribution to the order parameter from the oscillator is also
proportional to the degree of the node, shown in eq.(\ref{mf}).
These degree-dependence result in the $k^{2}$-dependence of
eq.(\ref{critical}), which leads to the absence of the threshold in random
scale-free network.
Therefore
there is a large difference between the absence of threshold in SIS model
and the synchronization, though these are apparently
similar results.
To clarify this difference, we will need to study the
synchronization in the other network models. As Egu\'{\i}luz
and Klemm has shown, scale-free network with large clustering
coefficient has
epidemic threshold in SIS model\cite{Eguiluz}, due to the smallness of
$k_{\mbox{nn}}$. The different behavior of the threshold may appear in our
oscillator network, because the absence of threshold is not caused by the
divergence of $k_{\mbox{nn}}$, but by degree dependence of
mean field-oscillator coupling. The study of the synchronization in
other scale-free network models is a future problem.
\begin{acknowledgements}
We acknowledge to Y. Nishiura, M. Iima and T. Yanagita for fruitful comments.
\end{acknowledgements}
\begin{thebibliography}{99}
\bibitem{Albert}R. Albert and A-L. Barab\'{a}shi, Rev. Mod. Phys. {\bf 74} 47(2002).
\bibitem{Dorogovtsev}S. N. Dorogovtsev and J. F. F. Mendes, {\it Evolution of Networks} (Oxford Press, 2003).
\bibitem{Timme}M. Timme , F. Wolf and T. Geisel, cond-mat/0306512.
\bibitem{Wang}X. F. Wang, and G. Chen, IEEE trans. on Circuits and systems-I,{bf 49} 1057(2002).
\bibitem {Watts} D. J. Watts, {\it Small Worlds}(Princeton, 1999).
\bibitem{Hong}H. Hong, M. Y. Choi and B. J. Kim, Phys. Rev. E {\bf 65} 026139.
\bibitem{Kuramoto} Y. Kuramoto, {\it Chemical Oscillations, Waves, and Turbulence}, (Springer, Berlin, 1984).
\bibitem{SIS0} R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. {\bf 86} 3200(2001).
\bibitem{SIS} M. Bog\~{u}n\'{a}, R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. {\bf 90} 028701(2003).
\bibitem{Eguiluz} V. M. Egu\'{\i}luz and K. Klemm, Phys. Rev. Lett. {\bf 89} 108701.
\end{thebibliography}
\end{document}