1: \documentstyle[aps,prl,epsfig]{revtex}
2: \draft
3: \begin{document}
4:
5: \title{Self-organized and driven phase synchronization in coupled map
6: scale free networks}
7: \author {Sarika Jalan\footnote{e-mail: sarika@prl.ernet.in} and
8: R. E. Amritkar\footnote{e-mail: amritkar@prl.ernet.in}}
9: \address{Physical Research Laboratory, Navrangpura, Ahmedabad 380 009, India.}
10: \maketitle
11:
12: \begin{abstract}
13: As a model of evolving networks, we study coupled logistic maps on scale
14: free networks. For small coupling strengths nodes show turbulent behavior
15: but form phase synchronized clusters as coupling increases. We identify two
16: different ways of cluster formation. For small coupling strengths we get
17: {\it self-organized clusters} which have mostly intra-cluster couplings and
18: for large coupling strengths there is a crossover and reorganization to
19: {\it driven clusters} which have mostly inter-cluster couplings. In the
20: novel driven synchronization the nodes of one cluster are driven by those
21: of the others.
22: \end{abstract}
23:
24: Recently, there is considerable interest in complex systems described by
25: networks or graphs with complex topology \cite{Strogatz}.
26: One significant recent discovery in the field of complex networks is
27: the observation that a number of naturally occurring large and complex networks
28: are scale free, i.e. the probability that a node
29: in the network is connected to $k$ other nodes decays as a power
30: law. These networks are found in many diverse systems such as
31: the nervous systems \cite{koch},
32: social groups \cite{social},
33: world wide web \cite{www}, metabolic networks \cite{metabolic},
34: food webs \cite{food} and citation network \cite{citation}.
35: Barabasi and Albert \cite{scalefree} have given a simple growth model
36: based on preferential attachments for the scale free networks.
37:
38: Most networks in the real world consist of dynamical
39: elements interacting with each other. Thus in order to understand properties of such
40: networks, we study a coupled map model of scale free networks.
41: Over the past decade, much work on coupled maps has used
42: regular coupling schemes. They show rich phenomenology arising when
43: opposing tendencies compete; the nonlinear dynamics of the maps
44: which in the chaotic regime tends to separate the orbits of different elements,
45: and the couplings that tend to synchronize them.
46: Coupled map lattices with nearest neighbor
47: or short range interactions show interesting spatio-temporal patterns,
48: and intermittent behaviour \cite{CML}. Globally coupled
49: maps (GCM) where each node is connected with all other nodes, show
50: interesting synchronized behaviour \cite{GCM}.
51: Random networks with large number of connections also show
52: synchronized behaviour for large coupling strengths.
53: \cite{random-net}.
54:
55: In this letter, we study the dynamics of coupled maps on scale free
56: networks. In particular, we investigate the clustering and synchronization
57: properties of such dynamically evolving networks. We find that as the
58: network evolves, it splits into
59: several phase synchronized clusters. Phase synchronization is obviously
60: because of the couplings between the nodes of the network and may be
61: achieved in two different ways. (i) The nodes of a cluster can be
62: synchronized because of intra-cluster couplings. We refer to
63: this as the {\it self-organized synchronization}.
64: (ii) Alternately, the nodes of
65: a cluster can be synchronized because of inter-cluster couplings. We
66: refer to this as the {\it driven
67: synchronization}. We find examples of both these types of phase
68: synchronized clusters in scale free networks with a crossover and
69: reorganization of nodes between
70: the two types as the coupling strength is varied. For small couplings
71: synchronization is of the self-organized type while for large
72: couplings it is of the driven type. We will discuss several examples
73: of such clusters in natural systems afterwards.
74:
75: Consider a network of $N$ nodes that are coupled with each
76: other through connections of the scale free type. Let each node of the
77: network be assigned a dynamical variable $x^i, i=1,2,\ldots,N$. The
78: evolution of the dynamical variables is given by
79: \begin{equation}
80: x^{i}_{t + 1} = (1 - \epsilon) f( x^{i}_t ) + \frac{\epsilon}{\sum_j C_{i j}}
81: \sum C_{ij} g( x^{j}_t )
82: \label{coupleddyn}
83: \end{equation}
84: where $x^{i}_n$ is the dynamical variable of the i-th node $(1 \le i \le N)$ at
85: the $t-th$ time step, C is the coupling matrix with elements $C_{ij}$
86: taking values $1$ or $0$ depending upon whether i and j are
87: connected or not. Note that C is a symmetric matrix with diagonal
88: elements zero. The function $f(x)$ defines the local nonlinear map and
89: the function $g(x)$ defines the nature of coupling between the nodes.
90: In this paper, we define the local dynamics by the logistic map,
91: $f(x) = \mu x (1 - x)$. The coupling function is taken as the
92: identity mapping, $g(x)=x$.
93:
94: The scale free network of $N$ nodes is generated by using the model of
95: Barabasi et.al. \cite{model}.
96: Starting with a small number, $m_0$, of nodes, at every time
97: step a new node with $m \le m_0$ connections is added. The probability
98: $\pi(k_i)$ that a
99: connection starting from this new node is connected
100: to a node $i$ depends on the degree
101: $k_i$ of node $i$ (preferential attachment) and is given by
102: $\pi(k_i) = (k_i + 1)/(\sum_j (k_j + 1))$.
103: After $\tau$ time steps the
104: model leads to a random network with $N = \tau + m_0$ nodes and $m \tau$
105: connections. This model leads to a scale free network, i.e. the
106: probability $P(k)$ that a node has a degree $k$ decays as a power law,
107: $P(k) \sim k^\lambda$ where $\lambda$ is a constant.
108: For the type of
109: probability law $\pi(k)$ that we have used $\lambda=3$. Other forms
110: for the probability $\pi(k)$ are possible which give different
111: values of $\lambda$. However, the results reported in this letter do
112: not depend on the exact form of $\pi(k)$ except that it should lead to
113: a scale free network.
114:
115: Synchronization of coupled dynamical systems may be defined in various
116: ways. The perfect synchronization corresponds to
117: the dynamical variables for different nodes having identical
118: values. The phase synchronization
119: corresponds to the dynamical variables for different nodes having
120: values with some definite relations \cite{phase}. In scale free
121: networks, we find that when the local dynamics of the
122: nodes (i.e. function $f(x)$) is in the chaotic zone, perfect
123: synchronization is obtained only for
124: clusters with small number of nodes. However,
125: interesting results are obtained when we study phase synchronized
126: behaviour. For our study we define the phase synchronization
127: as follows \cite{syn}.
128: Let $n_i$ and $n_j$ denote
129: the number of times the dynamical variables $x^i_t$ and $x^j_t$,
130: $t=1,2,\ldots,T$ for the nodes $i$ and $j$ show local
131: minima during the time interval $T$. Let $n_{ij}$
132: denote the number of times these local minima match
133: with each other. We define the phase distance
134: between the nodes $i$ and $j$ as $d_{ij}=1-2n_{ij}/(n_i+n_j)$. Clearly, $d_{ij}=0$
135: when all minima of variables $x^i$ and $x^j$ match with each other
136: and $d_{ij}=1$ when none of the minima match.
137: We say that nodes $i$ and $j$ are phase synchronized if $d_{ij}=0$.
138: Also, a cluster of nodes is phase synchronized if all pairs of nodes
139: belonging to that cluster are phase synchronized.
140:
141: \begin{figure}
142:
143: \begin{center}
144: \epsfxsize 8cm \epsfysize 8cm
145: \epsfbox[150 375 495 718]{fig1.ps}
146: \caption{Phase diagram showing turbulent, phase synchronized and
147: coherent regions in the two parameter space of $\mu$ and
148: $\epsilon$. Different regions are I. Turbulent region, II. Mixed
149: region, III. Partially ordered region, IV. Ordered quasiperiodic
150: region, V. Ordered periodic region, VI. Coherent region. Calculations
151: are for $N=50, m=1, T=100$. Region boundaries are determined based on changes
152: in the behaviour of the largest Lyapunov exponent (see Fig.~(2)) and
153: observing the asymptotic behaviour using several
154: initial conditions. The dashed lines indicate uncertainties in
155: determining the boundaries. The inset shows the phase diagram for the entire
156: range of parameter $\mu$ i.e. from 0 to 4.}
157: \end{center}
158: \label{phase}
159:
160: \end{figure}
161:
162: We now present numerical results of our model. We generate scale free
163: networks using the algorithm defined above and then study coupled
164: dynamics of variables associated with nodes of the network. Starting
165: from random initial conditions the dynamics of Eq.~(\ref{coupleddyn}),
166: after an initial transient, leads to interesting phase synchronized
167: behaviour. Fig.~(1) shows the phase diagram in the two parameter space
168: defined by $\mu$ and $\epsilon$ for $m=1, N=50, T=100$. To understand the
169: phase diagram let us first consider different states which are
170: obtained by the coupled dynamics. \\
171: (a) Coherent state: Nodes form a single synchronized cluster. \\
172: (b) Ordered state: Nodes form two or more clusters. \\
173: (c) Partially ordered states: Nodes form a few clusters with some
174: nodes not attached to any clusters. \\
175: (d) Turbulent state: All nodes behave chaotically with no cluster
176: formation. \\
177:
178: \begin{figure}
179: \begin{center}
180: \epsfxsize 8cm \epsfysize 8cm
181: \epsfbox[146 373 490 712]{fig2.ps}
182: \caption{Largest Lyapunov exponent, $\lambda$, is plotted as a function of
183: $\epsilon$ for $\mu = 4.0$. Different regions are labeled as in Fig.~(1).}
184: \label{lya}
185: \end{center}
186: \end{figure}
187:
188: For $\mu<3$, we get a stable coherent region (region VI) with all nodes having the
189: fixed point value. To understand the remaining phase diagram, consider
190: the line $\mu=4$. Fig.~(2) shows the largest Lyapunov exponent
191: $\lambda$ as a function of the coupling strength $\epsilon$ for
192: $\mu=4$. We can identify five different regions
193: as $\epsilon$ increases from 0 to 1; namely turbulent region, mixed
194: region, partially ordered region, ordered quasiperiodic region and
195: ordered periodic region as shown by regions I to V in Figs.~(1) and~(2).
196: For small values of $\epsilon$, we observe a turbulent
197: behaviour with all nodes evolving chaotically and there is no phase
198: synchronization. As $\epsilon$ increases further we get into a mixed
199: region (region II) which shows variety of phase synchronized behaviour, namely
200: ordered chaotic, ordered quasiperiodic,
201: ordered periodic and partially ordered, depending on the initial
202: conditions. Next region (region III) shows partially ordered chaotic behaviour. Here,
203: the number of clusters as well as the number of nodes in the clusters depend on the
204: initial conditions and also they change with time. There are
205: several isolated nodes not belonging to any cluster. Many of these
206: nodes are of the floating type which keep on switching intermittently
207: between independent evolution and phase synchronized evolution
208: attached to some cluster.
209: Last two regions are ordered quasiperiodic
210: and ordered periodic regions. In these regions, the network always splits into
211: two clusters. The two clusters are perfectly anti-phase synchronized
212: with each other, i.e. when the nodes belonging to one cluster show
213: minima those belonging to the other cluster show maxima.
214:
215: We now investigate the nature of phase ordering in different regions
216: of the phase diagram. We first concentrate on the middle of
217: regions II and V where we observe ordered
218: periodic behaviour. In both cases the largest Lyapunov
219: exponent is negative. In Fig.~(3a) and~(3b) we show the coupling matrix
220: $C$ (solid circles) and the nodes belonging to the two clusters
221: in regions II and V respectively. In Fig.~(3a) we observe that
222: there are no inter-cluster couplings between the nodes of the two clusters except
223: one coupling i.e. all the couplings except one are of the
224: intra-cluster type. The phase synchronization in this case is clearly of the
225: {\it self-organized} type. Exactly opposite behaviour is observed for the
226: region V (Fig.~(3b)). Here, we find that all the couplings are of inter-cluster
227: type with no intra-cluster couplings. This is clearly the phenomena of {\it driven
228: synchronization} where nodes of one cluster are driven into a phase
229: synchronized state due to the couplings with nodes of the other
230: cluster. The phenomena of driven synchronization in this region is a very robust one
231: in the sense that it is obtained for almost all initial conditions, the
232: transient time is very small, the nodes belonging to the two clusters
233: are uniquely determined and we get a stable solution.
234:
235: We observe that for small values of $\epsilon$ the self organized
236: behaviour dominates while for large $\epsilon$ driven behaviour dominates.
237: As the coupling parameter $\epsilon$ increases from zero and we enter
238: region II, we observe phase synchronized clusters of the self organized
239: type. Region III acts as a
240: crossover region from the self-organized to the driven behaviour. Here, the
241: clusters are of mixed type. The number of inter-cluster couplings
242: is approximately same as the number of intra-cluster couplings.
243: In this region there is a competition between the self-organized and driven
244: behaviour. This appears to be the reason for the formation of several
245: clusters and floating nodes as well as the sensitivity of these to the
246: initial conditions. As $\epsilon$ increases, we get into region IV where the driven
247: synchronization dominates and most of the connections between the nodes
248: are of the inter-cluster type. This driven
249: synchronization is further stabilized in region V with two
250: perfectly anti-phase synchronized driven clusters.
251:
252:
253: \begin{figure}
254: \begin{center}
255: \epsfxsize 8cm \epsfysize 8cm
256: \epsfbox[145 352 485 714]{fig3a.ps}
257: \epsfxsize 8cm \epsfysize 8cm
258: \epsfbox[145 352 485 714]{fig3b.ps}
259: \caption{Figures (a) and (b) show formation of phase synchronized
260: clusters for $\mu=4$ and $\epsilon=0.16$ and $0.8$ respectively. For
261: clarity of display, the
262: nodes are renumbered in each case such that nodes belonging to the
263: same cluster have sequential numbers. The nodes of clusters are shown
264: by open circles in squares. The elements of the coupling
265: matrix are shown by solid circles.}
266: \end{center}
267: \label{cluster}
268: \end{figure}
269:
270:
271: Geometrically, the organization of the network into connections of both
272: self-organized and driven types is always possible for $m=1$. For
273: $m=1$, our growth algorithm generates a tree type structure. A tree can
274: always be broken into two or more disjoint clusters with only
275: intra-cluster couplings by breaking one
276: or more connections. Clearly, this splitting is not unique. A tree can
277: also be divided into two clusters by putting connected nodes into different
278: clusters. This division is unique and leads to two clusters with only
279: inter-cluster couplings.
280:
281: For $m >1$ the dynamics of Eq.~(\ref{coupleddyn}) leads to a similar
282: phase diagram as in Fig~(1) with region II dominated by
283: self-organized synchronization and regions IV and V dominated by driven
284: synchronization. Though perfect inter- and
285: intra-cluster couplings between
286: the nodes as displayed in Figs.~(3a) and~(3b) are no longer
287: observed, clustering in region II is such that most of the couplings are
288: of intra-cluster type while for regions IV and V they
289: are of the inter-cluster type. As $m$ increases the
290: regions I and II are mostly
291: unaffected, but regions IV and V shrink while region III grows in size.
292:
293: The phenomena of self-organized and driven behaviour persists for the
294: largest size network that we have studied ($N=1000$). The mixed region
295: showing self-organized behaviour is mostly unaffected while the
296: ordered regions showing driven behaviour show a small shrinking in size.
297:
298: We have generated the networks using the preferential attachment law
299: for $\pi(k_i)$ leading to a power law distribution of the
300: degree of nodes. If instead we use a non-preferential law of
301: attachment i.e. the probability of connecting to any node does not
302: depend on the degree of that node but is a constant, then the
303: distribution of degree of nodes shows
304: an exponential dependence rather than a power law. In this case we
305: do not observe any phase synchronized behaviour. Thus the scale free
306: nature appears to be important for observing the phase synchronized
307: behaviour.
308:
309: There are several examples of self-organized and driven behaviour in
310: naturally occurring systems. Self-organized behaviour is more common
311: and is easily observed. Examples are social, ethnic and religious
312: groups, political groups, cartel of
313: industries and countries, herds of animals and flocks of birds,
314: different dynamic transitions such as self-organized criticality etc. The
315: driven behaviour is not so common. An interesting example is the
316: behaviour of fans during a match between traditional rivals.
317: Before the match the fans may act as individuals
318: (turbulent behavior) or form self-organized clusters such as a
319: single cluster of fans of the game or several clusters of fans of
320: different star players. During the match there can be
321: a crossover to a driven behaviour. When the
322: match reaches a feverish pitch, i.e. the strength of the interaction
323: increases, the fans are likely to form two phase synchronized groups.
324: The response of the two groups will be
325: anti-phase synchronized with each other. Another example of crossover
326: to a driven behaviour is the conflict in Bosnia where a society organized
327: into villages and towns was split into ethnic groups.
328:
329: As discussed in the introduction several naturally occurring networks
330: show scale free behaviour and it is likely to be a generic behaviour
331: of several naturally growing networks. We expect to observe both the self-organized
332: and driven synchronization behaviour reported in this paper in such
333: systems. Network properties of many examples of self-organized and
334: driven behaviour discussed above
335: are not known and it is possible that some of these
336: examples may have scale free type of networks.
337:
338: To conclude we have found interesting self-organized and driven phase
339: synchronization behaviour in coupled maps on scale free
340: networks. Self-organized synchronization is
341: characterized by dominant intra-cluster couplings and is found when
342: strength of the couplings is small as compared to the local dynamics.
343: As the coupling strength increases there is a crossover from
344: the self-organized to the driven behaviour which also involves
345: reorganization of nodes into different clusters.
346: The driven behaviour is characterized by inter-cluster couplings
347: and is found when strength of the couplings is large as compared to
348: the local dynamics.
349:
350: One of us (REA) thanks Professor H. Kanz for useful
351: discussions and Max-Plank Institute for
352: the Physics of Complex Systems, Dresden for hospitality.
353:
354:
355: \begin{thebibliography}{99}
356:
357: \bibitem{Strogatz} S. H. Strogatz, Nature, {\bf 410}, 268 (2001).
358:
359: \bibitem{koch} C. Koch and G. Laurent, Science {\bf 284}, 96 (1999).
360: \bibitem{social} S. Wasserman and K. Faust, Social Network Analysis,
361: Cambridge Univ. Press, Cambridge 1994.
362: \bibitem{www} R. Albert, H. Jeong, A. -L. Barabasi, Nature, {\bf 401},
363: 130 (1999).
364: \bibitem{metabolic} H. Jeong, B. Tomber, R. Albert, Z. N. Oltvai,
365: A. -L. Barabasi, Nature, {\bf 407}, 651 (2000).
366: \bibitem{food} Richard J. Williams, Nea D. Martinez, Nature,
367: {\bf 404}, 180 (2000).
368: \bibitem{citation} S. Render, Euro. Phys. J. B. {\bf 4}, 131 (1998).
369: \bibitem{scalefree} A. -L. Barabasi, R. Albert, Science, {\bf 286}, 509 (1999).
370:
371: \bibitem{model} A. -L. Barabasi, R. Albert, H. Jeong, Physica A, {\bf 281},
372: 69 (2000).
373:
374: \bibitem{CML} K. Kaneko, Physica D, {\bf 34}, 1 (1989).
375: \bibitem{GCM} K. Kaneko, Physica D, {\bf 41}, 137 ( 1990 ).
376:
377: \bibitem{random-net} C. Susanna, Manrubia and A. S. Mikhailov,
378: PRE, {\bf 60}, 1579 (1999).
379: Oct 2001.
380:
381: \bibitem{phase} M. G. Rosenblum, A. S. Pikovsky, and J.
382: Kurth, PRL, {\bf 76}, 1804 (1996); W. Wang, Z. Liu, Bambi Hu, Phys. rev. Lett.
383: {\bf 84}, 2610 (2000).
384:
385: \bibitem{syn} F. S. de San Roman, S. Boccaletti, D. Maza and H. Mancini,
386: Phys. Rev. Lett. {\bf 81}, 3639 (1998).
387:
388: \end{thebibliography}
389:
390: \end{document}
391: