1: \documentclass{article}
2:
3: \usepackage{epsfig}
4:
5: \begin{document}
6:
7: \title{Mapping dynamical systems onto complex networks}
8:
9: \author{Ernesto P. Borges$^a$, %\thanks{\tt ernesto@ufba.br}
10: Daniel O. Cajueiro$^b$ %\thanks{\tt danoc@pos.ucb.br}
11: and Roberto F. S. Andrade$^c$\thanks{\tt randrade@ufba.br}
12: \\ $ $ \\
13: {\normalsize $^a$Escola Polit\'ecnica, Universidade Federal da Bahia,} \\
14: {\normalsize R. Aristides Novis, 2, 40210-630 Salvador--BA, Brazil}
15: \\
16: {\normalsize $^b$Departamento de Economia, Universidade Cat\'{o}lica de Bras\'{i}lia} \\
17: {\normalsize 70790-160 Bras\'{i}lia, Brazil}
18: \\
19: {\normalsize $^c$Instituto de F\'{i}sica, Universidade Federal da Bahia,} \\
20: {\normalsize 40210-340 Salvador, Brazil}
21: }
22:
23: \date{}
24:
25: \maketitle
26:
27: \begin{abstract}
28: A procedure to characterize chaotic dynamical systems with concepts of
29: complex networks is pursued, in which a dynamical system is mapped onto a
30: network. The nodes represent the regions of space visited by the system,
31: while edges represent the transitions between these regions. Parameters
32: used to quantify the properties of complex networks, including those related
33: to higher order neighborhoods, are used in the analysis. The methodology is
34: tested for the logistic map, focusing the onset of chaos and chaotic regimes.
35: It is found that the corresponding networks show distinct features, which are
36: associated to the particular type of dynamics that have generated them.
37: \end{abstract}
38: %
39: \begin{quote}
40: \small
41: 89.75.Fb: Structures and organization in complex systems \\
42: 89.75.Hc: Networks and genealogical trees \\
43: 02.10.Ox: Combinatorics; graph theory
44: \end{quote}
45:
46: \section{Introduction}
47: \label{sec:intro}
48:
49: Chaotic dynamical systems are characterized by several measures that
50: quantify how irregular (despite deterministic) the trajectories are. The
51: set of Lyapunov exponents~\cite{eckrue85} provides a measure of the
52: dependence on the initial conditions of their trajectories, while
53: information theory may be used to characterize a dynamical system in terms
54: of the production of entropy. Actually, a dynamical system with a chaotic
55: behavior is regarded as a realization of Shannon's concept of an ergodic
56: information source~\cite{gal68}, as the Kolmogorov-Sinai entropy is
57: equated~\cite{pes77} to the sum of the positive Lyapunov exponents.
58: Further measures to describe a chaotic system include fractal dimensions
59: and singularity spectra~\cite{henpro83,mensre89}. This formalism applies
60: only where the system has, at least, one positive Lyapunov exponent.
61: %
62: However, several authors have studied situations in which
63: the largest Lyapunov exponent
64: vanishes, but the distance between
65: neighbouring trajectories increases in time according to a power law
66: \cite{tsa97,rob04,rob05,rob06}. These situations are usually found at the
67: onset of chaos, where an infinitesimal change of a control parameter
68: drives the system into either a regular or a chaotic regime. These
69: investigations uncovered some of its features
70: related to the sensitivity to the initial conditions ~\cite{tsa97},
71: entropy production per unit time ~\cite{lat00}, multifractal geometry of
72: the attractor ~\cite{lyrtsa98}, relaxation to the system attractor
73: ~\cite{bor02}
74: and multifractal dynamics at the onset of chaos ~\cite{mayoral05}.
75:
76: Recently, the investigation of complex networks has set up a new framework
77: for the analysis of systems with a large number of degrees of freedom.
78: Within it, one has access to the properties of the topological structure
79: underlying the mutual interactions among the system constituents. This
80: approach has been applied to a large variety of actual systems, ranging
81: from social interactions, biological nature, information and electrical
82: distribution \cite{albbar02,lat06}.
83:
84: In this work, we define a mapping of dynamical systems onto a network.
85: This way, the network properties can be used to display new features for
86: the characterization of the system's trajectory. The network nodes
87: correspond to coarse-grained regions (cells) of the phase space visited by
88: the trajectory. Two nodes $r$ and $s$ are linked if, during the time
89: evolution, the trajectory jumps from cell $r$ to cell $s$. Although this
90: naturally offers a construction procedure for a directed network, we
91: consider here undirected networks. In this approach, we are able to find
92: novel geometric and topological properties of the phase space through the
93: measures that have been recently developed to characterize complex
94: networks. The dynamical system, defined on the time domain, is mapped into
95: a node domain, which represents regions of the phase space visited by the
96: trajectory. As it is based on the division of phase space into boxes, this
97: process follows similar construction procedures considered in the
98: evaluation of the fractal dimension of attractors and the Kolmogorov-Sinai
99: entropy.
100:
101: It worths mentioning that some previous works have put together
102: ideas of dynamical systems and complex networks. As we shall see,
103: our approach is rather distinct, as the previous contributions consider
104: synchronization under the assumption of certain regularity in the
105: connection topology~\cite{pik01,atajos04}.
106: The use of a network to represent the phase space evolution of discrete time
107: dynamical systems has been suggested in \cite{stephan}.
108:
109: We restrict ourselves to the analysis of the quadratic map
110: \cite{collet80} described by
111: \begin{equation}\label{eq:logistic}
112: x_{t+1}=1-a x_{t}^{2}\;\;(t=0,1,2,3\cdots)
113: \end{equation}
114: where $x_{t}\in [-1,1]$ and $a\in [0,2]$. Eq.~(\ref{eq:logistic}) leads to a
115: variety of distinct dynamical situations, the properties of which are
116: expected to be manifested in the networks they originate. In particular,
117: we explore those regimes at the onset of chaos that proceed through a
118: bifurcation cascade as well as those close to an intermittency transition
119: and also the full developed chaos.
120:
121:
122: \section{Network characterization}
123: \label{sec:network}
124:
125: An indirected network $R$ is defined only by the number $N$ of nodes and
126: links $L$, represented by an assembly of unordered pairs $S_R{(r,s)}$,
127: $r,s\leq N$, indicating which pairs of nodes are directly connected. This
128: information provides a full description of the network, leading to the
129: computation of the average number of links per node $\langle k\rangle$,
130: the average clustering coefficient $C$, the mean minimal distance among
131: the nodes $\langle d\rangle$, the diameter $D$, the probability
132: distribution of nodes with $k$ links $p(k)$. Other measures, like the
133: assortativity degree \cite{new02} and the distribution of individual node
134: clustering coefficients $C(k)$ with respect to its degree $k$ have also
135: been introduced, but they are not explored here.
136:
137: $R$ can be described by its adjacency matrix $M(R)$. This is not the most
138: concise representation of a network, but it opens the possibility to the
139: evaluation of its spectral properties and, as recently indicated, the
140: higher order neighborhoods $R_{\ell}$, $\ell=1,2,..., D$ \cite{and06}.
141: This is done in a straightforward way, by means of the set of matrices
142: $\{M_{\ell}\}$, so defined that $(M_{\ell})_{r,s}=1$ only if the shortest
143: distance along the network between the nodes $r$ and $s$ is $\ell$.
144: Otherwise, $(M_{\ell})_{r,s}=0$. Although the whole information on the
145: network is entailed in $S_R{(r,s)}$ or in $M(R)$, each $M_{\ell}$
146: condenses information on $R$ that is extracted from $M(R)$ within the
147: quoted framework. This formalism is also consistent with the recently
148: proposed procedure to evaluate the fractal dimension of the network
149: $d_{F,R}$ \cite{son05}, as it naturally leads to the set $\{N_{\ell}\}$
150: required for this evaluation. Here, each $N_{\ell}$ counts the number of
151: pairs of nodes which are $\ell$ steps apart.
152:
153: Within this framework, we consider that each node is the only zeroth order
154: neighbor of itself, and define $M_0=I$, where $I$ indicates the identity
155: matrix. Also, we assume that $M_1=M$. Since the matrix elements of $M(R)$
156: take only the values 0 and 1, the other matrices $M_{\ell}$ of the set are
157: recursively evaluated with the help of Boolean operations \cite{and06}. We
158: shall make further use of a matrix that condenses all information in
159: $\{M_{\ell}(R)\}$. As just discussed, given any two nodes $r$ and $s$, it
160: is clear that $(M_\ell)_{rs}=1$ for just one value of $\ell$. So, if we
161: define a matrix
162: \begin{equation}\label{eq:Matrix}
163: \widehat{M}=\sum_{j=0}^{\ell_{max}}j M_j,
164: \end{equation}
165: it directly informs how many steps apart any two nodes in the network are.
166: Also, it is possible to use the information in $\widehat{M}$ to
167: graphically display the structure of a network with the help of color or
168: grey scale plots.
169:
170: It should also be mentioned that this framework opens the door for a finer
171: characterization of the network, if we consider each $R_\ell$ as an
172: independent network. Thus, several of the properties quoted above used to
173: characterize $R$ can be also evaluated for the evaluation of $R_{\ell}$.
174: This is explored in the next section specifically for the the degree
175: distribution and clustering coefficient $\langle k\rangle_\ell$ and
176: $C(\ell)$.
177:
178:
179: \section{An algorithm to map a dynamical system into a network}
180: \label{sec:algorithm}
181:
182: To map a dynamical system into a network $R$, we use the framework of the
183: algorithm introduced in~\cite{blo90}, to efficiently evaluate the
184: generalized fractal dimensions of fractal structures by the box counting
185: method.
186:
187: Let us consider a dynamical system with $m$ variables. Without loss of
188: generality, one can consider a set of points $\mathcal{Z}\subset \Re^m$
189: consisting of the vectors $z(i)$, $i=1,\cdots,T$, $T \gg 1$, which
190: represent the coordinates of the dynamical system. The components of these
191: vectors, $z_\delta(i)$, $\delta=1,\cdots,m$, are assumed to belong to the
192: interval $[0,1)$. We define a graining in phase space by dividing each
193: phase space axis into $W$ equally sized disjoint intervals, so that the
194: whole phase is spanned by a set of $W^m$ boxes. This represents also the
195: maximal possible number of nodes in a network, as {\em e.g.}, in the case
196: of an ergodic system. Of course, the choice of $W$ defines the graining,
197: and the size of the region represented by a node. In the next section we
198: investigate the effect of $W$ on the obtained networks.
199:
200: Based on ~\cite{blo90}, each point $z(i)$ of the trajectory is mapped into
201: a node of $R$ according to
202: \begin{equation}\label{eq2}
203: n(i)=\sum_{\delta=1}^{m} W^{\delta-1}\mathrm{floor}(W
204: z_\delta(i))\mathrm{,}
205: \end{equation}
206: where $\mathrm{floor}(x)$ is a function that evaluates the largest integer
207: less than $x$.
208: Actually, this is just a simple way to divide the region
209: $[0,1)^m$ in equal parts. Thus, the nodes of the so constructed network
210: represent a box in the coarse grained phase space of the system. After the
211: mapping is complete, the boxes that were not visited by the trajectory are
212: eliminated from the network, as they constitute nodes with zero degree
213: ($k=0$), which do not provide any useful information on the dynamical
214: system. The edges are built as described in the following procedure. Let
215: $z(i)$ and $z(i+1)$ be two consecutive points of the dynamical system.
216: Suppose that these points were previously mapped into the nodes $n(r)$ and
217: $n(s)$, where $0<r,s\le W^m$. Thus, one assumes that there is an edge
218: connecting $n(r)$ and $n(s)$. Here, it is considered that there is only
219: one edge linking $n(r)$ to $n(s)$ if $r\ne s$, {\em i.e.} self-links are
220: not computed.
221:
222: This procedure can lead to directed and weighted networks; however, in
223: this work, we focus on the most simple situation of undirected and
224: un-weighted networks, once our main purpose here is to address the
225: problem, and show that we can extract useful information from it.
226:
227:
228: \section{Results}
229: \label{sec:results}
230:
231: We concentrate our investigation on three distinct regions of $a$,
232: $a=a_c=1.40115518909...$, $a\in[1.749,1.75)$ and $a=2$, which correspond,
233: respectively, to the first period doubling transition, the region close to
234: the tangent bifurcation to the period-three window, and the full developed
235: chaotic state. Representative networks to the distinct chaotic attractors
236: are generated for distinct values of the graining $W$. We only consider
237: trajectories that start on the attractor, in order to avoid spurious nodes
238: (visited only once) that depend on initial conditions. For a fixed value
239: $W$, the network grows as the trajectory evolves in the the phase space
240: with respect to the number of iteration steps $t$. There is no a priori
241: criterion to decide the time $t_F$ after which the network is complete. In
242: this work we have followed how $N$ and $L$ increase with $t$, for a given
243: $W$. We define $t_F$ as the smallest value of $t$ for which
244: $N(t_F)=N(2t_F)$ and $L(t_F)=L(2t_F)$.
245:
246:
247: First let us discuss the effect of $W$ on $N$ and $L$. For the purpose of
248: presenting a full neighborhood analysis of the networks, we have selected
249: here values of $W$ that lead to the maximal number $\approx$ 10000 nodes
250: in the network. Such choice for $W$ clearly depends on $a$.
251: Indeed, due to the strategy adopted for the construction of the networks,
252: $N$ grows with $W$ according to a power law mediated by the fractal
253: dimension of the attractor $d_{F,A}$. This is shown in
254: Fig.~\ref{fig:netparameters}a, where we draw points for $a=a_c$,
255: $a_c+10^{-3}$, $1.749999$ and 2. For $a_c$ we measure the slope
256: $0.54\dots$, which agrees with the known value of $d_{F,A}$ of the period
257: doubling attractor. For all other cases, the slope is 1 within 2\%
258: accuracy, even for $a=a_c+10^{-3}$, which lies already in the chaotic
259: regime. This is in accordance with the fact that $d_{F,A}$ changes in a
260: discontinuous way at $a=a_c$. When $a$ corresponds to a periodic solution,
261: the network becomes finite, so that $N$ and $L$ do not depend on $W$,
262: provided this parameter is large enough.
263:
264: Although we are primarily interested in the properties of the complete
265: network, we can also follow the dependency of $N$ and $L$ with respect to
266: $t$, for a fixed value of $W$. Assuming $N\sim L^z$, this defines a
267: dynamical exponent $z$ in the early stages of the network evolution. The
268: analyzed data indicate that $z\simeq 1$ for all values of $a$.
269: Nevertheless, we find that, in the immediate chaotic neighborhood of
270: $a=1.75$,
271: the laminar phases in the intermittent regime traps the trajectory
272: for long intervals, demanding large time of integration
273: to complete the network.
274: \begin{figure}[tbp]
275: \begin{center}
276: \epsfig{figure=n-w.eps,width=0.95\columnwidth,keepaspectratio,clip=}
277: \epsfig{figure=d-w.eps,width=0.95\columnwidth,keepaspectratio,clip=}
278: \end{center}
279: \caption{(a)~Dependence of $N$ with respect to $W$ for
280: $a=a_c$ (circles), $a=a_c+10^{-3}$ (diamonds), $a=1.749999$ (down
281: triangles), and $a=2$ (up triangles).
282: The the same convention is used in all other figures.
283: The distinct slopes indicate the values of $d_{F,A}$.
284: (b)~Power law dependence of $\langle d\rangle$ (hollow symbols) and $D$
285: (solid symbols) with respect to $N$ for $a=a_c$ and $a=a_c+10^{-3}$. The
286: inset shows logarithmic dependence among the same quantities when
287: $a=1.749999$ and $a=2$.}
288: \label{fig:netparameters}
289: \end{figure}
290:
291: Now we discuss the network properties, following the methodology and
292: parameters indicated before. When appropriate, we extend our discussion to
293: properties of higher order neighborhoods in the network. We find distinct
294: network structures when we consider the chaotic regime or the onset of
295: chaos. Regarding the mean minimal distance $\langle d\rangle$ and diameter
296: $D$, we find that, for the chaotic regime, they grow with respect to $W$
297: (and $N$), in a logarithmic way, similarly to small-world networks
298: \cite{watts98}. If we write $\langle d\rangle=\alpha \log_{10} W$, we find
299: $\alpha\simeq2.4$ and 3.7, respectively, for $a=2$ and
300: $a\in[1.749,1.749999]$, as illustrated in the inset of
301: Fig.~\ref{fig:netparameters}b. As $D$ only assumes integer values, it is
302: possible to expect a similar behavior only in an approximate way, {\em
303: e.g.}, equally sized steps in $D\times \log_{10} W$ plots. Thus, assuming
304: $D=\beta \log_{10} W$, we find $\beta\simeq4$ for $a=2$ with very good
305: accuracy. In the interval $[1.749,1.749999]$, we noticed the presence of
306: fluctuations in the size of the steps, which increases when we approach
307: the threshold $a=1.75$. The results for $a=a_c$ have a completely distinct
308: behavior: $\langle d\rangle$ and $D$ increase as power
309: laws with respect to $W$, as illustrated in the main panel of
310: Fig.~\ref{fig:netparameters}b. For $a=a_c+10^{-3}$, the same dependence
311: prevails. The exponents obtained for $\langle d\rangle$ and $D$ are,
312: respectively, $0.67$ and $0.73$ for $a_c$, and $0.35$ and $0.38$ for
313: $a=a_c+10^{-3}$. $\langle d \rangle$ and $D$ increase smoothly for
314: $a=a_c+10^{-3}$ but at $a=a_c$, the growth of the two quantities present
315: discontinuities and steps. Results in Fig.~\ref{fig:netparameters}b indicate
316: that, unlike the properties of the attractor, reflected by the rash change
317: in the value of $d_{F,A}$, the network properties change slowly when the
318: chaotic regime is reached.
319:
320: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
321: % p(k)
322: We have evaluated the distribution $p(k)$ vs. $k$ in all different
323: regimes.
324: We can see in Fig.~\ref{fig:degree} that, for $a=a_c$, $k$ does not reach
325: large values ($k_{max}\simeq 30$), so that it is not possible to identify
326: a power law decay in this range.
327: For $a=2$ (and also $a=1.749999$) nodes with larger values of $k$ can be found,
328: but $p(k)$ does not follow either a power law.
329:
330:
331: \begin{figure}[tbp]
332: \begin{center}
333: \epsfig{figure=pk.eps,width=0.95\columnwidth,keepaspectratio,clip=}
334: \end{center}
335: \caption{
336: Degree distribution of nodes at $a_c$ (circles)
337: and in the $a=2$ chaotic regime, (up triangles).
338: Behavior close to tangent bifurcation, $a=1.75 - 10^{-4}$, (not shown)
339: is quite similar to $a=2$.}
340: \label{fig:degree}
341: \end{figure}
342:
343: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
344: % fractal dimension
345: Such distinct behavior is also explicit when we analyze the fractal
346: dimension of the network $d_{F,R}$ \cite{son05}. Fig.~\ref{fig:fractaldim}
347: shows that the
348: $a=a_c$ networks have a well defined scaling behavior, which extends over
349: more than two decades, in a quite precise way. On the other hand, a
350: fractal dimension for the networks in the chaotic regime is not
351: evident. First, the small values of $D$ reduces the region of possible
352: scaling behavior. Then, we clearly observe deviations to the expected
353: power-law regime.
354:
355: \begin{figure}[tbp]
356: \begin{center}
357: \epsfig{figure=fractaldim.eps,width=0.95\columnwidth,keepaspectratio,clip=}
358: \end{center}
359: \caption{Clear power law behavior for $N(\ell) \times \ell$ when $a=a_c$,
360: with $d_{F,R}=1.47$. Finite size effects blur this dependence when
361: $\ell\simeq D$. In the chaotic regime, for $a=2$, $d_{F,R}$ can
362: hardly be evaluated. For $a=a_c+10^{-3}$, the points illustrate a slow
363: transition between the two regimes.}
364: \label{fig:fractaldim}
365: \end{figure}
366:
367:
368: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
369: % Clustering coefficient
370: Regarding the clustering coefficient, we obtain $C\equiv0$ for $a=a_c$,
371: indicating the complete absence of triangles in the network. For
372: $a=1.749999$ and $a=2$, we find that $C$ decays with $N$ according to a
373: power law with exponent $\approx$ 0.95, what is not so close to the value
374: $0.75$ observed for the Albert-Barabasi scale-free network
375: \cite{albbar02}. However, its small values indicate that the network has
376: only a small number of triangles.
377:
378: Other features of the network may be drawn if we consider the clustering
379: coefficient of higher order neighborhoods \cite{and06}. To obtain a
380: clearer picture of this analysis consider, for instance, the regular
381: Cayley tree with $\ell=2$ network for the complete regular three neighbors
382: Cayley tree. The $\ell=2$ network is formed by triangles, much like the
383: Husimi cactuses and, correspondingly, a large value for $C(\ell=2)$. The
384: following odd and even numbered neighborhoods are characterized,
385: respectively, by values of $C(\ell)=0$ and $C(\ell)>0$, whereby the values
386: of $C(\ell)$ for a subset of even neighborhoods decrease in a monotonic
387: way. A similar situation is found for the networks investigated here. In
388: Fig.~\ref{fig:neighborhood}a we summarize the sequence of $C(\ell)$ for the
389: three situations under investigation. We see that, for $a=2$ and
390: $a=1.4999$, the oscillatory behavior lasts only until $\ell=5$ and 10,
391: respectively. Also we note that the odd numbered $C(\ell)$ increases until
392: it reaches values as high as those of the even numbered neighborhoods. On
393: the other hand, the $a=a_c$ network has $C(\ell)=0$ for all odd numbered
394: neighborhood. The inset shows that the $C(\ell$=even number$)$ decays with
395: $\ell$ according to a power law, with exponent $\alpha\simeq 1$.
396:
397: \begin{figure}[tbp]
398: \begin{center}
399: \epsfig{figure=clustering.eps,width=0.95\columnwidth,keepaspectratio,clip=}
400: \epsfig{figure=averagedegree.eps,width=0.95\columnwidth,keepaspectratio,clip=}
401: \end{center}
402: \caption{(a)~Behavior of $C(\ell)$ with respect to $\ell$ for chaotic regime
403: and at the onset of chaos.
404: The inset shows a peculiar power law behavior $C(\ell)$
405: at the onset of chaos.
406: (b)~Behavior of $\langle k\rangle_{\ell} \times
407: \ell$.
408: The onset of chaos exhibits a slow increasing
409: value of $\langle k\rangle_{\ell}$ over a large interval of $\ell$,
410: interrupted by finite size effects already present in
411: Fig.~ \protect\ref{fig:fractaldim}.
412: In opposition to this picture, chaotic regime shows a sharp increase
413: of $\langle k\rangle_{\ell}$.}
414: \label{fig:neighborhood}
415: \end{figure}
416:
417: We have also analyzed the average degree $\langle k\rangle_{\ell}$ as a
418: function of the neighborhood $\ell$. Here again we find that the behavior
419: of the chaotic regime and the onset of chaos have distinct features, as
420: exhibited in the Fig.~\ref{fig:neighborhood}b.
421:
422: Finally, we use the information in $\widehat{M}$ to obtain images,
423: in color (or gray) scale,
424: of the network neighborhood structure.
425: They provide a vivid and easy visualization of their distinct
426: properties of the attractor. For $a=2$, the first order neighborhood is
427: distributed
428: along the parabola described by the r.h.s. of (\ref{eq:logistic}) (see
429: Fig.~\ref{fig:colormatrix}a). It shows how the second and higher order
430: neighborhood evolve according to the higher order iterates of the quadratic map.
431: However, mixing and finite size effects stemming from a finite
432: graining smears the higher order iterates. The situation is distinct
433: for $a_c$, when the attractor is a $d_{F,A}=0.54...$ dust spread out
434: in the $[-1,1]$ interval. Only boxes containing part of the dust remain
435: in the network, so that contiguous numbered nodes are not actually neighbors
436: in phase space. The resulting image (Fig.~\ref{fig:colormatrix}b)
437: displays a fine and intertwined tessitura that reflects a very peculiar
438: behavior of the trajectories at the onset of chaos\cite{rob04,rob05,rob06}.
439:
440: \begin{figure}[tbp]
441: \begin{center}
442: \epsfig{figure=colormatrix-chaos.eps,width=0.75\columnwidth,keepaspectratio,clip=}
443:
444: $ $
445:
446: \epsfig{figure=colormatrix-edge.eps,width=0.75\columnwidth,keepaspectratio,clip=}
447: \end{center}
448: \caption{Color scale plots (gray scale in paper version) of $\widehat{M}$
449: for $a=2$~(a) and $a=a_c$ (b). Scale ranges from black and blue
450: ($\ell=0$ and $\ell=1$) to red ($\ell=D$). Number of levels in (a) is much
451: smaller than in (b). Neighborhood structure changes abruptly in the distinct
452: regimes.}
453: \label{fig:colormatrix}
454: \end{figure}
455:
456:
457: \section{Conclusion}
458: \label{sec:conclusions}
459:
460: In this work we explore the idea of mapping chaotic systems onto complex
461: networks. Networks are constructed according to a well defined
462: methodology, and results for the logistic map indicate how their
463: properties can be associated to those of the attractor in phase space.
464: Trajectories in distinct dynamical regimes are explored in order to show
465: how the major differences in phase space are reflected in the networks.
466: The networks show several features of small-world and scale-free networks,
467: but they do not fully match with those networks generated by the specific
468: algorithms described in \cite{watts98,baralb99}. The analysis of the
469: networks in the neighborhood of $a_c$ reveals that the $N\times W$
470: dependence, measured by $d_{F,A}$, has a sharp transition at the onset of
471: chaos. So, the distinct character of the trajectories in phase space is
472: indeed reflected in the network. However, with exception of $C$, the
473: results for the other indices ($\langle d \rangle,D$ and $p(k)$) change in
474: a much smoother way with respect to changes in the parameter $a$.
475:
476:
477: Acknowledgements: This work was supported by the Brazilian
478: Agencies CNPq and FAPESB.
479:
480: \begin{thebibliography}{00}
481:
482: \bibitem{eckrue85}
483: {Eckmann J. P. \and Ruelle D.},
484: {Rev. Mod. Phys.} \textbf{57}, (1985) 617.
485:
486: \bibitem{gal68}
487: {Gallager R. G.},
488: \textit{Information theory and reliable communication}
489: (Addison-Wesley, Reading, MA 1968).
490:
491: \bibitem{pes77}
492: {Pesin Y. B.},
493: {Rus. Math. Surv.} \textbf{32}, (1977) 55.
494:
495: \bibitem{henpro83}
496: {Hentschel H. G. E. \and Procaccia I.},
497: {Physica D} \textbf{8}, (1983) 435.
498:
499: \bibitem{mensre89}
500: {Meneveau C. \and Sreenivasan K. R.},
501: {Phys. Lett. A} \textbf{137}, (1989) 103.
502:
503: \bibitem{tsa97}
504: {Tsallis C., Plastino A. R. \and Zheng W. M.},
505: {Chaos Solitons and Fractals} \textbf{8}, (1997) 885.
506:
507: \bibitem{rob04}
508: {Baldovin F. \and Robledo A.},
509: {Phys. Rev. E} \textbf{69}, (2004) 045202.
510:
511: \bibitem{rob05}
512: {Robledo A.},
513: {Europhys. News} \textbf{36}, (2005) 214.
514:
515: \bibitem{rob06}
516: {Hern\'andez-Sanda\~na H., Robledo A.},
517: {Physica A} \textbf{370}, (2006) 286.
518:
519: \bibitem{lat00}
520: {Latora V., Baranger M., Rapisarda A. \and Tsallis C.},
521: {Phys. Lett. A} \textbf{273}, (2000) 97.
522:
523: \bibitem{lyrtsa98}
524: {Lyra M. L. \and Tsallis C.},
525: {Phys. Rev. Lett.} \textbf{80}, (1998) 53.
526:
527: \bibitem{bor02}
528: {Borges E. P., Tsallis C., Ananos G. F. J. \and de Oliveira P. M. C.},
529: {Phys. Rev. Lett.} \textbf{89}, (2002) 254103.
530:
531: \bibitem{mayoral05}
532: {Mayoral E. \and Robledo A.},
533: {Phys. Rev. E} \textbf{72}, (2005) 026209.
534:
535: \bibitem{albbar02}
536: {Albert R. \and Barabasi A.-L. },
537: {Rev. Mod. Phys.} \textbf{74}, (2002) 47.
538:
539: \bibitem{lat06}
540: {Boccaletti S., Latora V., Moreno Y., Chavez M. \and Hwang D. U.},
541: {Phys. Rep.} \textbf{424}, (2006) 175.
542:
543: \bibitem{pik01}
544: {Pikovsky A., Rosenblum M. \and Kurths J.},
545: \textit{Synchronization - a universal concept in nonlinear science}
546: (Cambridge University Press, Cambridge 2001).
547:
548: \bibitem{atajos04}
549: {Atay F. M. \and Jost J.},
550: {Phys. Rev. Lett.} \textbf{92}, (2004) 144101.
551:
552: \bibitem{stephan}
553: {Thurner S.},
554: {Europhys. News} \textbf{36}, (2005) 218.
555:
556: \bibitem{collet80}
557: {Collet P. \and Eckman J. P.},
558: \textit{Iterated Maps on the Interval as Dynamical Systems}
559: (Birkhauser, Boston 1980).
560:
561: \bibitem{new02}
562: {Newman M. E. J.},
563: {Phys. Rev. Lett.} \textbf{89}, (2002) 208701.
564:
565: \bibitem{and06}
566: {Andrade R. F. S., Miranda J. G. V. \and Lob\~ao T. P.},
567: {Phys. Rev. E} \textbf{73}, (2006) 046101.
568:
569: \bibitem{son05}
570: {Song C. M., Havlin S. \and Makse H. A.},
571: {Nature}\textbf{433}, (2005) 392.
572:
573: \bibitem{blo90}
574: {Block A., VonBloh W. \and Schellnhuber H. J.},
575: {Phys. Rev. A} \textbf{42}, (1990) 1869.
576:
577: \bibitem{watts98}
578: {Watts D. J. \and Strogatz S. H.},
579: {Nature} \textbf{393}, (1998) 440.
580:
581: \bibitem{baralb99}
582: {Barabasi A.-L. \and Albert R.},
583: {Science} \textbf{286}, (1999) 509.
584:
585: \end{thebibliography}
586:
587: \end{document}
588: