1: \documentclass[nobm]{epl}
2:
3: \title{Small-world behavior in a system of mobile elements}
4: \shorttitle{Dynamic Small-Worlds}
5: \author{Susanna C. Manrubia\inst{1} \and Jordi Delgado\inst{2}
6: \and Bartolo Luque\inst{3}}
7: \shortauthor{S.C. Manrubia \etal}
8: \institute{
9: \inst{1} Max Planck Institute for Colloids and Interfaces, Theory
10: Division.\\ 14424 Potsdam, Germany \\
11: \inst{2} Departament de Llenguatges i Sistemes
12: Inform\`atics, Universitat Polit\`ecnica de \\ Catalunya. Campus Nord,
13: M\`odul C6, 08034 Barcelona, Spain \\
14: \inst{3} Centro de Astrobiolog\'{\i}a (CAB), INTA. Carretera de Ajalvir
15: Km. 4, 28850. \\ Torrej\'on de Ardoz, Madrid, Spain}
16:
17: \pacs{87.23.Ge}{Dynamics of social systems}
18: \pacs{05.10.-a}{Computational methods in statistical physics and nonlinear
19: dynamics}
20: \pacs{05.40.-a}{Fluctuation phenomena, random processes, noise, and
21: Brownian motion}
22:
23: \begin{document}
24: %\verb|nobm|
25: \maketitle
26:
27: \begin{abstract}
28: We analyze the propagation of activity in a system of mobile automata.
29: A number $\rho L^d$ of elements move
30: as random walkers on a lattice of dimension $d$, while with a small
31: probability $p$ they can jump to any empty site in the system. We show that
32: this system behaves as a Dynamic Small-World (DSW) and present analytic
33: and numerical results for several quantities. Our analysis shows that the
34: persistence time $T^*$ (equivalent to the persistence size $L^*$ of
35: small-world networks) scales as $T^* \sim (\rho p)^{-\tau}$, with $\tau =
36: 1/(d+1)$.
37: \end{abstract}
38:
39: The network of collaboration of actors in movies, of world airports, or of
40: neural connections in the worm {\it Caenorhabditis
41: elegans}, can be represented in the
42: form of a graph. Persons, airports, and neurons stand for the
43: nodes of the network, and a bond is drawn when two actors are cast
44: together in a film, when a non-stop flight exists between two airports, or
45: when a neural connection is present, respectively \cite{Watts,Classes}.
46: The relational graphs resulting from these apparently disparate systems have
47: two interesting properties that qualify them as {\it small-world} graphs:
48: They are highly clustered at the local scale (similar to what happens in
49: regular graphs), and the average number of bonds $\bar \ell$ that separate
50: two randomly chosen nodes in a graph of size $L$ scales as $\bar{\ell} \sim
51: \log L$ (a characteristic property of random graphs).
52:
53: Small-world (SW) graphs efficiently interpolate between regular and random
54: graphs thanks to a small number $p L^d$ of {\it shortcuts} (long-range
55: connections, $p \ll 1$) which are
56: superimposed on a regular lattice formed by $L^d$ sites. Much attention has
57: been devoted recently to the topological properties of such graphs, which
58: determine the characteristic path length $\bar \ell$ and its scaling
59: properties \cite{Top,pers}, and to the effect that a small-world-like
60: connectivity might have on the properties of dynamical
61: systems~\cite{Dyn,Dyn2}.
62: Up to now, the attention has been only focused on this relevant type of
63: networks, SW. Nonetheless, the restriction of
64: activity propagating {\it on a quenched, undirected graph}, where elements
65: are fixed in their positions and information propagates in a symmetric way
66: (from element $A$ to element $B$ as well as {\it vice versa}), is a strong
67: one, and perhaps not always the best representation of a world formed by
68: mobile elements where connections (relational bonds) appear and disappear
69: in the course of time. In this sense, recent approaches have tried to model
70: the formation and evolution of the topology in SW and other networks throught
71: slow
72: changes in their bonds \cite{Bara,Doro}. Our focus here is not in this large
73: time-scale, but on the shorter time-scale in which the movement of the
74: elements in the physical space is essential in defining contacts between
75: nodes. As an example, consider the case of propagation of a
76: tropical disease. Such events usually start in a ``far-away'' place (in the
77: Euclidean sense), from which the carrier of the disease flies away.
78: He or she arrives at a new destination and transmits the disease, and the
79: process repeats with the new carriers. This is a clear example of a directed,
80: dynamical graph, with moving elements which create and remove connections
81: in a partly stochastic fashion. Indeed, in all cases where the carrier is not
82: taking active part in the transmission of the signal, random encounters might
83: play a main role when compared to the fixed network of acquaintances.
84:
85: With the aim of analyzing the effect of movement in the transmission of a
86: signal, we introduce a simple model for the formation of a Dynamic
87: Small-World (DSW), and study the propagation of activity among mobile elements
88: in the system. We define quantities analogous to the characteristic path
89: length $\bar \ell$ and to the clustering coefficient $C$ in order to quantify
90: the similarities with SW. Our analytical and numerical calculations show that
91: DSW are quantitatively different from (quenched) SW.
92:
93: The model is defined as follows. Consider a lattice formed by $L^d$ sites
94: embedded in a $d-$dimensional space, where $\rho L^d$ automata can move.
95: The density of automata $\rho$ is a parameter of the model. A time step
96: consists of one move of all the active elements in the system plus the
97: synchronous transmission of activity to new inactive neighbours. The precise
98: implementation of these rules is as follows: i) Choose one of the active
99: elements according to a random sequential updating (i.e. each active element
100: will be updated once and only once in an uncorrelated fashion with respect
101: to the previous move). With probability $p$, the automaton
102: selects an empty site in the system and jumps there. With the complementary
103: probability $1-p$, it looks for a free site in its neighbourhood and moves to
104: it. The neighbourhood consists of the $2d$ adjacent sites. ii) Once all the
105: active automata have moved, activity is simultaneously transmitted by each
106: active element to any other inactive element in its neighbourhood.
107: As initial condition an automaton is selected at random and activated at
108: $t=0$, the rest of the system remains inactive. The active element starts
109: moving according to rule i). When an inactive element is met at a
110: neighbouring site, activity is transmitted according to ii). We use periodic
111: boundary conditions in our simulations.
112:
113: The probability $p$ of jumping to a far-away
114: site is the equivalent of the density of shortcuts in SW. In that
115: case, the activity would propagate to one new bond per time step, and one could
116: define a velocity of propagation $v_{\rm SW}=1$\cite{Top}. In the DSW case,
117: the time required to transmit the activity to the chosen element
118: depends on the density of automata $\rho$, so we expect $v_{\rm DSW} =
119: v(\rho) \le 1$.
120:
121: We start by defining two quantities analogous to the characteristic path length
122: $\bar \ell (p)$ and the clustering coefficient $C(p)$ of SW
123: (for details, see Ref. \cite{Watts}).
124: In our case, the quantity equivalent to $\bar \ell$
125: is the characteristic time $T (\rho, p)$ for the activity starting at
126: the seed to reach a randomly chosen element in the system, averaged
127: over many independent realizations of the
128: process. Because $v_{\rm SW}=1$ in SW, $\bar \ell$ and $T$ coincide.
129: We will call the analogous of the clustering coefficient {\it neighbouring
130: coefficient}, and define it in the following way. Consider the set
131: of particles located in the neighbourhood of element $i$ at time $t$,
132: ${\cal{N}}_t(i)$. At the next time step, this set will be in principle
133: different, since all active elements have tried to move to new sites. We
134: define the neighbouring coefficient for element $i$ at time $t$ as $K_t(i;
135: \rho, p)= (2d)^{-1} {\cal{U}}_t(i)$, where
136: ${\cal{U}}_t(i) \equiv \textrm{card} \{{\cal{N}}_t(i) \cap
137: {\cal{N}}_{t-1}(i) \}$. The average neighbouring
138: coefficient for a system with parameters $p$ and $\rho$
139: is $K (\rho, p) = \langle (\rho L^d)^{-1} \sum_i K_t(i; \rho, p) \rangle$,
140: where the brackets $\langle . \rangle$ represent a temporal average.
141: Our numerical simulations in $d=1, \; 2$ reveal
142: that the small-world behavior of our system is well described through these
143: two quantities (see Fig. \ref{fig1}). Indeed, for a wide range of $p$ values,
144: the local configuration is maintained (the value of $K$ is close to unity),
145: while the typical number of time steps required to reach a randomly chosen
146: automaton from the initial source is small and scales as $\log L$. This
147: behavior is characteristic of SW systems.
148:
149: \begin{figure}[tbp]
150: \twofigures[width=6.6cm,height=6cm]{fig1.eps}{fig2.eps}
151: \caption{Normalized characteristic time $T(\rho, p) / T(\rho, 0)$ and
152: normalized neighbouring coefficient $K(\rho, p) / K(\rho, 0)$ as a function
153: of $p$ for two different values of the density, as shown in the legend, in a
154: one-dimensional system of size $L=5000$ (averages over $10^3$ independent
155: realizations have been performed).}
156: \label{fig1}
157: \caption{Schematic representation of the propagation of activity in a Dynamic
158: Small-World. The initial condition corresponds to a single active particle
159: which moves as a random walk. The spatio-temporal triangle signaling the
160: spreading of activity has a half-angle $\alpha = \arctan v$. A second
161: triangle starts the first time that an automaton jumps out of $A_1$, at a time
162: $t_1=(v \rho p)^{-1/2}$. When $A_k=A_1$, the $k$th jump takes
163: place.}
164: \label{fig2}
165: \end{figure}
166:
167: Our system admits a geometrical description suitable for calculating
168: several quantities analytically. We will focus our interest on the
169: growth of the volume $V(t)$ occupied by $a(t)$ active elements
170: as a function of time $t$. There is a simple
171: relation between the active volume and the number of active particles:
172: $\rho V(t) = a(t)$. In the following, we restrict our calculations
173: to the case $d=1$. Some generalizations to arbitrary $d$ are
174: presented at the end of the letter.
175:
176: At $t_0=0$, only one element is active.
177: This element moves as a random walker during about $t \sim \rho^{-2}$ time
178: steps (the average distance between elements is of order $1/\rho$),
179: until it meets a second element at a neighbouring site, and
180: activity is transmitted. This kind of spreading of activity persists for
181: $t_1$ time steps,
182: where $t_1$ signals the jump of the first active element
183: to a domain disconnected (in terms of activity) from the previous one.
184: For $p^{-1/2} \gg \rho^{-2}$ (contact through
185: random walker moves is much more frequent than contact through long-distance
186: jumps, see below), we can use a continuous approximation to the problem,
187: as shown schematically
188: in Fig. \ref{fig2}. The vertical axis represents physical distance in a
189: one-dimensional space, and the horizontal axis represents time. At time
190: $t$, active elements cover a volume $V(t)$ equal to the sum of the lengths
191: obtained from the intersection of a
192: vertical line drawn at $t$ with the triangles. For the particular case
193: $p=0$ the activity propagates at a constant velocity $v(\rho)$ and, in
194: the geometrical representation, only one triangle is present.
195:
196: We can easily calculate the average time $t_1$
197: when the first jump (or shortcut) takes place. To this end, we make the
198: annealed approximation $t \propto \langle \sqrt{n(t)} \rangle
199: \sim \sqrt{\langle n(t) \rangle}$ (where $n(t)$ is the total number
200: of attempted jumps up to time $t$), and obtain
201:
202: \begin{equation}\label{t1}
203: t_1 = \sqrt{1 \over v \rho p} \; .
204: \end{equation}
205: According to the picture in Fig. \ref{fig2} we can also calculate the times
206: for the appearance of the next jumps,
207:
208: \begin{equation}\label{t2}
209: t_2 = {1 + \sqrt 3 \over 2} t_1 \;\;\;\;\;\;\;\;\;\;\;\;\;
210: t_3 = {1 \over 2} \left[ 1 + {\sqrt 3 \over 3} + {2 \sqrt{6} \over 3}
211: \right] t_1 \; ,
212: \end{equation}
213: and so on.
214: Note that jumps are not equidistant in time, and that
215: they happen more and more frequently for longer times: $(t_1-t_0) > (t_2 - t_1)
216: > \dots > (t_{k+1}-t_k) > \dots$, and these differences go to zero for $k \to
217: \infty$.
218: In general, we can write an expression for the time by which $k$ jumps
219: have taken place,
220:
221: \begin{equation}\label{kk}
222: t_1^2 k = \sum_{i=0}^{k-1} (t_k - t_i)^2 \; .
223: \end{equation}
224: If we consider the new variable $\tau_k = (t_k - t_{k-1})/t_1$ the difference
225: between consecutive times satisfies
226:
227: \begin{equation}\label{ittau}
228: k \tau_k^2 + 2 \tau_k \; \sum_{i=1}^{k-1} i \tau_i =1 \; .
229: \end{equation}
230: We look for a solution of the form
231: %\begin{equation}
232: $\tau_k = (a + \epsilon_{k-1} - \epsilon_k)/k$.
233: %\end{equation}
234: Direct substitution into Eq. (\ref{ittau}) yields $a = {1 / \sqrt{2}}, \;
235: \epsilon_0 = {a/2}$, and $\epsilon_k \sim {b'/ k}$, where $b'$ is
236: obtained by imposing a final condition for $k \to \infty$.
237: In the original discrete variables, we obtain now
238:
239: \begin{equation}\label{tk}
240: t_k = t_1 \sum_{i=1}^k \tau_i \simeq t_1 \left( {1 \over \sqrt{2}}
241: \sum_{i=1}^k {1 \over i} + b - {c \over k^2} \right) \; .
242: \end{equation}
243: The coefficients of this expression are numerically obtained by fitting the
244: iterative solution of Eq. (\ref{ittau}) to the expression above. This returns
245: $b=0.3124 \dots , \; c= 0.0337 \dots$.
246:
247: \begin{figure}[t]
248: \twofigures[width=6.6cm]{fig3b.eps}{newcollapse.eps}
249: \caption{Volume $V(t)$ covered by active particles.
250: Main plot: The continuous line
251: corresponds to $2vt$ for $t<t_1$ and to our solution
252: Eq. (\ref{Solvol}) for $t>t_1$. These simulations were made
253: for systems of size $L=10^3$ (circles), $10^4$ (crosses), and $10^5$ (squares)
254: with $\rho = 0.8$ and
255: $p=10^{-5}$. Inset: In the limit $t \gg t_1$ and $L \to \infty$, $V(t)$ grows
256: exponentially at a rate $\sqrt{2 v \rho p}$. This result (continuous line,
257: with velocity numerically computed at $p=0$), is compared with numerical
258: interpolations for systems of size $L=10^5$ (circles), with $\rho=0.4$ and
259: $p=10^{-4}$ in the exponential regime.}
260: \label{fig3}
261: \caption{Data collapse for the characteristic time $T$ as a function of
262: the rescaled system parameters. The simulations were performed in dimension
263: $d=1$ and averaged over 100 independent realizations. Other parameters
264: are as shown in the legend. There are corrections to scaling for $p
265: \to 1$, where the assumption $p^{-1/2} \gg \rho^{-2}$ breaks down.}
266: \label{fig4}
267: \end{figure}
268:
269: For time $t<t_1$, the volume grows linearly, $V(t<t_1) \simeq 2 v t$. We can
270: also define the volume at the $k$th jump (for $t_k \ge t_1$) as
271:
272: \begin{equation}\label{vol}
273: V(k) = 2 v \sum_{i=0}^{k-1} (t_k - t_i) = 2 v \left( k t_k
274: -\sum_{i=1}^{k-1} t_i \right) \; .
275: \end{equation}
276: It follows from our previous calculations that
277:
278: \begin{equation}
279: {V(k) \over 2 v t_1} = \sum_{i=1}^k i \tau_i = \sum_{i=1}^k
280: (a + \epsilon_{k-1} - \epsilon_k) = a k + \epsilon_0 - \epsilon_k \; .
281: \end{equation}
282: Inverting Eq. (\ref{tk}) and substituting above, we get the asymptotic
283: expression
284:
285: \begin{equation}\label{Solvol}
286: V(t) = \sqrt{2} v t_1 \exp \left\{ \sqrt{2}
287: \left[ {t \over t_1} - \left( {{\bf C} \over \sqrt{2}} + b \right) \right]
288: \right\} + O (t^{-1})\; ,
289: \end{equation}
290: where {\bf C}$=0.577215 \dots$ is Euler's constant.
291: Our numerical and analytical results are compared in Fig. \ref{fig3}.
292:
293: In these calculations, we have assumed that the time by which
294: $k$ jumps have taken place is a deterministic variable defined by $t_k
295: \sim \sqrt{\langle n(t_k) \rangle}$, thus discarding fluctuations in $n(t)$.
296: If we calculate the fluctuations in the jumping process, it turns out that
297: their
298: characteristic size is proportional to the approximate value $t_k$. The
299: annealed approximation is quite good for $\rho$ close to unity and $p \to 0$,
300: but the effect of the fluctuations is usually visible in the fact that
301: the apparent velocity at $t<t_1$ is larger than $v(\rho)$ when $p > 0$
302: (this explains the small discrepancy seen in Fig. \ref{fig3} for the largest
303: size). The bending at finite $V(t)$ observed in Fig. \ref{fig3} is a finite
304: size effect: $V(t)$ saturates at the system size. The exponential regime is
305: nonetheless well captured by our approach: In an infinite system, activity
306: would propagate at a rate $\sqrt{2}/t_1 = \sqrt{2 v \rho p}$, as shown in the
307: inset of Fig. \ref{fig3}.
308:
309: In the light of the above results, we have studied the scaling properties of
310: the characteristic time $T(\rho, p)$ with the system size $L$, the
311: jumping probability $p$, and the density of automata $\rho$, in a way
312: similar to what has been done for SW \cite{NW}. First, we need to estimate
313: the dependence of a persistence time $T^* \equiv t_1$ \cite{pers} on the
314: parameters of the system.
315: Despite having a whole spectrum of characteristic times, as given by
316: (\ref{tk}), all of the $t_k$ scale in the same
317: form as $t_1$. Hence, we conjecture that $T^* \propto (v \rho p)^{-1/2}$
318: in $d=1$. Another relevant quantity with dimensions of time is $L/v(\rho)$.
319: We propose the following scaling for $T$,
320:
321: \begin{equation}
322: T \propto {L \over v(\rho)} f \left( L \left({\rho p \over v(\rho)}
323: \right)^{1/2}
324: \right) \; ,
325: \label{scal}
326: \end{equation}
327: where $f(z)$ is a function of the dimensionless parameter $z=L/L^*$, and
328: we have defined the persistence size to be $L^* = v T^*$. We show in Fig.
329: \ref{fig4} the resulting data collapse for several $T(\rho,p)-$ curves,
330: where $T(\rho,p)$ is the (minimum) time required for the activation of
331: all particles in the system. Our numerical results support the scaling
332: ansatz (\ref{scal}).
333:
334: This result can be extended to dimensions $d>1$. Our previous
335: calculations rely on the determination of a persistence time $T^*$,
336: which can be estimated in any dimension $d$ from a geometrical
337: representation of the system.
338: In general, we can talk of activity propagating in a $d-$dimensional lattice,
339: at a velocity
340: $v(\rho)$, from a source placed at the center of a hyper-sphere of radius
341: $r(t) = v(\rho) t$. The first
342: jump in this $d-$dimensional space will take place when the volume of the
343: $(d+1)-$dimensional cone (where time has been included as an
344: additional dimension) contains $1/p$ automata.
345: That is to say, the persistence time is defined by
346:
347: \begin{equation}
348: 1 =
349: {2 \pi^{d/2} \over \Gamma(d/2)} v^d(\rho) \rho p \int_0^{T_d^*} {t^d} dt \; ,
350: \end{equation}
351: which yields
352:
353: \begin{equation}
354: T_d^* = \left[ {(d+1) \Gamma(d/2) \over 2 \pi^{d/2}} {1 \over v^d(\rho)
355: \rho p} \right]^{1/(1+d)} \; .
356: \end{equation}
357:
358: It has been recently argued \cite{pers} that the presence of a finite
359: persistence size
360: in SW (or a finite persistence time in our case) is a finite size effect
361: implying that the transition that the system undergoes at $p \to 0$ is first
362: order. In both SW and DSW, $L^* \sim p^{-\tau}$, with an exponent
363: $\tau_{\rm SW}=1/d$ in the quenched case, while we have just shown that in
364: the dynamical system $\tau_{\rm DSW}=1/(d+1)$. We also expect a first-order
365: critical transition for DSW at $p \to 0$. It is not difficult to
366: understand where the difference in $\tau$ comes from. In DSW, effective jumps,
367: that is, shortcuts able to carry the activity to an inactive domain, can
368: originate in the bulk of the system, and not only at the boundaries of each
369: active area, as is the case in SW \cite{NW}. The introduction of time
370: as an additional ``effective'' dimension
371: naturally leads to the calculated change in the scaling, and establishes
372: a symmetry between time and space for DSW.
373:
374: The efficiency of a system with shortcuts can be measured in
375: terms of the average time that a signal takes to propagate, $T^*$. For
376: fixed parameters $L$, $p$, and $\rho$, $T^*_{SW} \simeq p^{-1}$, while
377: $T^*_{DSW} \simeq p^{-1/2}$. Since $p<1$, $T^*_{DSW} < T^*_{SW}$.
378: In other words, if we are going to design an algorithm to make a signal
379: propagate, it would be better to invest effort in dynamical elements able
380: to follow arbitrary paths than to construct fixed channels between fixed
381: elements.
382:
383: Dynamic Small-Worlds might be an adequate description for a class of
384: systems where the physical movement of the elements is the key ingredient
385: for the transmission of a signal, and where
386: a small-world-like behavior may be present. Another example of this class of
387: systems may be some species
388: of ants, where two different types of individuals, characterized by
389: two different speeds of movement, are known to coexist \cite{Ants}. This
390: moves our model to the field of collective computation, and suggests that a
391: small-world behavior might be relevant for the overall performance of such
392: systems. Some authors have studied epidemics in models where
393: elements are able to move on a lattice \cite{Boccara}. It would be interesting
394: to analyze the dynamics of such a system on a DSW. According to the results
395: presented here, they should differ quantitatively from current studies of
396: epidemics on SW networks \cite{Dyn}.
397: The case $p=0$ of our model closely corresponds to an active process of the
398: type $A+B \to 2B$ with strong inter-particle interactions. The
399: spreading of the active front depends on the interplay between the
400: movement of the active particle at the boundary and the position of
401: the closest inactive particle in a non-trivial way. This problem is
402: similar to that of a diffusing prey hunted by $N$ predators \cite{Redner}.
403:
404: Our model admits a number of generalizations.
405: Recent studies have stressed the non-random structure of real networks.
406: The scale-free topology of the World Wide Web (WWW), for instance, arises
407: from a dynamical
408: multiplicative process through which a small fraction of nodes accumulates
409: most of the links \cite{Bara}.
410: In the DSW here presented all sites have equal probability of generating or
411: receiving the next shortcut, resulting in a Poisson distribution of
412: shortcuts per site. This rule can be easily modified if our sites are
413: provided with memory, such that the
414: probability of being selected by an automaton as a landing site depends on
415: the number of automata which already visited it. This would generate
416: a skewed distribution of shortcuts, with a small set of frequently visited
417: sites. Hubs in the WWW would have their equivalent in `cities' (strongly
418: prefered sites) in the
419: physical space. Actually, the number of potential shortcuts of a real city
420: (its population) distributes similarly to the number of connections per node
421: in the WWW, and results from a similar growth process \cite{Zan}. The presence
422: of highly connected nodes in the WWW directly affects the tolerance of the
423: network against errors \cite{AlCo}. It would be
424: interesting to explore if the modification of the landing rule has also further
425: consequences in the performance of DSW, for instance in the detection of
426: the minimal path between two points by an automaton furnished with
427: local information \cite{Klein}. Further analysis of the DSW model here
428: introduced will be the subject of forthcoming works.
429:
430: \acknowledgments
431: The authors acknowledge interesting discussions with U. Bastolla,
432: P. Grassberger, J. Shillcock, N. Shnerb, and D.H. Zanette.
433: JD was supported by the UPC grant PR98-11.
434:
435: \begin{thebibliography}{99}
436:
437: \bibitem{Watts} \Name{Watts D. J.} \Book{Small Worlds. The dynamics of
438: networks between order and randomness} \Publ{Princeton studies in complexity,
439: Princeton University Press} \Year{1999}; \Name{Watts D. J. \and Strogatz
440: S. H.} \REVIEW{Nature}{393}{1998}{440}.
441:
442: \bibitem{Classes} \Name{Nunes Amaral L. A., Scala A., Barth\'el\'emy M., \and
443: Stanley H. E.} \REVIEW{Proc. Natl. Acad. Sci. USA}{97}{2000}{11149}.
444:
445: \bibitem{Top} \Name{Newman M. E. J. \and Watts D. J.}
446: \REVIEW{Phys. Lett. A}{263}{1999}{341};
447: \Name{Barrat A. \and Weigt M.}
448: \REVIEW{Eur. Phys. J. B}{13}{2000}{457};
449: \Name{Newman M. E. J., Moore C., \and Watts D. J.}
450: \REVIEW{Phys. Rev. Lett.}{84}{2000}{3201}.
451:
452: \bibitem{pers} \Name{Argollo de Menezes M., Moukarzel C.F., \and Penna T. J.
453: P.} \REVIEW{Europhys. Lett.}{50}{2000}{574}.
454:
455: \bibitem{Dyn} \Name{Moukarzel C. F.} \REVIEW{Phys. Rev. E}{60}{1999}{R6263}.
456:
457: \bibitem{Dyn2} \Name{La\-go-Fer\-n\'an\-dez L. F., Huer\-ta R., Cor\-ba\-cho
458: F., \and Si\-g\"uen\-za J. A.} \REVIEW{Phys. Rev. Lett.}{84}{2000}{2758};
459: \Name{Moore C. \and Newman M. E. J.} \REVIEW{Phys. Rev. E}{61}{2000}{5678}.
460:
461: \bibitem{Bara} \Name{Barab\'asi A.-L. \and Albert R.}
462: \REVIEW{Science}{286}{1999}{509}.
463:
464: \bibitem{Doro} \Name{Dorogovtsev S. N. \and Mendes
465: J. F. F.} \REVIEW{Phys. Rev. E}{62}{2000}{1842}; \Name{Bornholdt S. \and
466: Rohlf T.} \REVIEW{Phys. Rev. Lett.}{84}{2000}{6114}.
467:
468: \bibitem{NW} \Name{Newman M. E. J. \and Watts D. J.} \REVIEW{Phys. Rev.
469: E}{60}{1999}{7332}.
470:
471: \bibitem{Ants} \Name{H\"olldobler B. \and Wilson E. O.} \Book{The Ants}
472: \Publ{Berlin: Springer-Verlag} \Year{1990}.
473:
474: \bibitem{Boccara} \Name{Boccara N., Cheong K., \and Oram M.} \REVIEW{J.
475: Phys. A: Math. Gen.}{27}{1994}{1585}.
476:
477: \bibitem{Redner} \Name{Krapivsky P. L. \and Redner S.} \REVIEW{J. Phys. A:
478: Math. Gen.}{29}{1996}{5347}.
479:
480: \bibitem{Zan} \Name{Zanette D. H. \and Manrubia S. C.} \REVIEW{Phys. Rev.
481: Lett.}{79}{1997}{523}.
482:
483: \bibitem{AlCo} \Name{Albert R., Jeong H., \and Barab\'asi A.-L.}
484: \REVIEW{Nature}{406}{2000}{378}; \Name{Cohen R., Erez K., ben-Avraham D.,
485: \and Havlin S.} \REVIEW{Phys. Rev. Lett.}{85}{2000}{4626}.
486:
487: \bibitem{Klein} \Name{Kleinberg J. M.} \REVIEW{Nature}{406}{2000}{845}.
488:
489: \end{thebibliography}
490:
491: \end{document}
492:
493:
494:
495:
496:
497:
498:
499:
500:
501:
502:
503:
504:
505:
506: