1: \documentclass[aps,twocolumn]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{dcolumn}% Align table columns on decimal point
4: \usepackage{amssymb}
5:
6: \begin{document}
7:
8: \title{Detecting local synchronization in coupled chaotic systems}
9: \author{L. Pastur, S. Boccaletti and P.L. Ramazza}
10: \affiliation{Istituto Nazionale di Ottica Applicata, Largo E.
11: Fermi 6, 50125 Florence, Italy}
12: \date{\today}
13:
14: \begin{abstract}
15: We introduce a technique to detect and quantify local functional
16: dependencies between coupled chaotic systems. The method estimates
17: the fraction of locally syncronized configurations, in a pair of signals
18: with an arbitrary state of global syncronization.
19: Application to a pair of interacting R\"ossler oscillators shows
20: that our method is capable to quantify the number of dynamical
21: configurations where a local prediction task is possible, also in
22: absence of global synchronization features.
23:
24: \vglue 0.4 truecm
25:
26: PACS: 05.45.Tp,05.45.Xt,05.45.-a,05.45.Ac
27:
28: \end{abstract}
29:
30:
31:
32: \maketitle
33:
34:
35:
36: In the past years much attention has been devoted to characterize
37: coupled chaotic systems exhibiting synchronization regimes
38: \cite{reviews}. In this framework, different synchronization
39: features have been studied, such as, e.g., identical and
40: generalized synchronization \cite{complete,generalized}, phase
41: synchronization \cite{phase}, lag and intermittent lag
42: synchronization \cite{intermittentlag}. Furthermore,
43: synchronization effects have been explored in natural phenomena
44: \cite{nature}, and controlled laboratory experiments
45: \cite{experiment}.
46:
47: In this context, various attempts to provide unifying definitions
48: for encompassing the different synchronization phenomena have been
49: pursuited \cite{previous}. Recently, a formal approach to the
50: problem has been put forward \cite{definition}, in which the
51: unifying property of synchronization is established in the
52: emergence of local functional dependencies between neighborhoods
53: of particular phase space configurations in the projected spaces
54: of the two coupled subsystems. The approach assumes a system {\bf
55: Z} $\in {\Bbb R}^{m_1+m_2}$ divisible into two coupled subsystems,
56: {\bf X} $\in {\Bbb R}^{m_1}$ and {\bf Y} $\in {\Bbb R}^{m_2}$. In
57: this framework, synchronization is equivalent to predictability of
58: one subsystem's values from another, \textit{i.e.} that an event
59: $\tilde y$ in {\bf Y} always occurs when a particular event
60: $\tilde x$ in {\bf X} occurs. However, when searching for evidence
61: of synchronization in data, one seldom has data that fall right on
62: a given $\tilde x$ or on a given $\tilde y$. Rather, the closer
63: $x(t)$ is to $\tilde x$ the closer $y(t)$ is to $\tilde y$. The
64: latter statement is captured rigorously by a local {\em
65: continuous} function; namely, the trajectories of $x(t)$ close to
66: $\tilde x$ are mapped near to $\tilde y$ by a local function that
67: is continuous at the point $(\tilde x, \tilde y)$, and that, near
68: ($\tilde x, \tilde y$) describes well the predictability of
69: subsystem {\bf Y} dynamics from subsystem {\bf X} dynamics. Ref.
70: \cite{definition} gives a general, formal mathematical ground to
71: the above statements, and establishes the sufficient conditions
72: for a system to display global synchronization features,
73: \textit{i.e.} to admit local functional dependencies regardless on
74: the particular choice of the ($\tilde x, \tilde y$) phase space
75: configuration.
76:
77: For a generic pair of coupled chaotic systems, however, it is to
78: be expected that synchronization occurs only at some locations (if
79: any) of the phase space, and not globally. In this case, a
80: continuous functional dependence of $y(t)$ on $x(t)$ will exist
81: only locally around a set of synchronization points $\{ \tilde
82: x_s, \tilde y_s \}$.
83:
84: Implementation of a search for local functional dependencies
85: requires two separate steps: a preliminary one in which the two
86: interacting subsystems {\bf X} and {\bf Y} are properly identified
87: within the original dynamical systems {\bf Z}, and their
88: dimensionalities measured, and a second one in which the local
89: synchronization points ($\tilde x, \tilde y$) are detected. The
90: first problem was solved recently in Ref. \cite{disentangle} by
91: means of a modification of the {\it false nearest neighbors}
92: algorithm \cite{abarbanel}, allowing for a separate measurement of
93: the dimensionalities of weakly coupled systems in the case of
94: emergent synchronization motions.
95:
96: In this paper, we will address the second step of the search by
97: introducing the {\it synchronization points percentage} ({\it
98: SPP}) indicator, and show how one can gather information on local
99: synchronization properties emerging in coupled chaotic systems.
100:
101: We start by assuming to have $N$ data points in {\bf Z} $\in {\Bbb
102: R}^{m_1+m_2}$. By means of a proper subspace reconstruction
103: \cite{disentangle}, we end up with $N$ data points in {\bf X} $\in
104: {\Bbb R}^{m_1}$ and $N$ corresponding images in {\bf Y} $\in {\Bbb
105: R}^{m_2}$. We then pick a specific point $\tilde x \in$ {\bf X}
106: and consider its image $\tilde y \in$ {\bf Y}.
107:
108: The first task consists in identifying proper domains and
109: co-domains for a statistical analysis of the existence of
110: functional dependency. For this purpose, we choose a pair of
111: positive real numbers $(\varepsilon_k, \delta)$ (the index $k$
112: being an integer), and consider the volume $U_{\varepsilon_k}
113: \subset$ {\bf X} ($V_\delta \subset$ {\bf Y}) containing all
114: points whose $m_1$-distance ($m_2$-distance) from $\tilde x$
115: ($\tilde y$) is smaller than $\varepsilon_k$ ($\delta$).
116: Furthermore, we look at all points in {\bf X} falling within
117: $U_{\varepsilon_k}$, and verify the imaging condition, that is we
118: ask ourselves whether or not all images of the points in
119: $U_{\varepsilon_k}$ fall within $V_\delta$. If the answer is no,
120: we choose $\varepsilon_{k+1} < \varepsilon_k$, and repeat the
121: above procedure. If for all $k$ the imaging condition is not
122: satisfied, the task ends with the conclusion that no local
123: functional dependency exists in the vicinity of the chosen
124: configuration $(\tilde x,\tilde y)$. If, instead, for a given
125: $\tilde k$ the imaging condition is verified, the task ends with
126: the identification of a valid pair $(\varepsilon_{\tilde
127: k},\delta)$, over which one has to test for the existence of a
128: continuous functional relationship.
129:
130: Fig.\ref{fig1} helps in understanding the schematic representation
131: of the procedure. In the following we will denote with $U \subset$
132: {\bf X} ($V \subset$ {\bf Y}) the neighborhood
133: $U_{\varepsilon_{\tilde k}}$ ($V_\delta$) surrounding $\tilde x$
134: ($\tilde y$), and assume that $m < N$ points fall within $U$. By
135: construction, the number of points falling within $V$ will be $n
136: \geq m$, reflecting the fact that $V$ might host also images of
137: points not belonging to $U$.
138:
139: \begin{figure}[!h]
140: \includegraphics[width=80mm]{technique.eps}
141: \caption{Schematic representation of the statistical continuity
142: analysis. Upper part shows the reconstructed trajectories in the
143: two subspaces {\bf X} and {\bf Y}, and the location of the points
144: $\tilde x$ and $\tilde y$. The lower part zooms on the $U,V$
145: neighborhoods. For $\varepsilon=\varepsilon_{k+1}$, $V$ contains
146: all images of the $m$ points in $U$ (solid circles), plus images
147: of other points (empty squares) from outside $U$. For
148: $\varepsilon=\varepsilon_{k}$, some points in $U$ (empty circles)
149: have images outside $V$.}\label{fig1}
150: \end{figure}
151:
152: The probability of a single point falling within $V$ is $P(V)
153: \equiv n/N$, and the probability that $m$ points fall within $V$
154: by pure chance is $P_m(V) = P(V)^m = \left( \frac{n}{N}
155: \right)^m$. This latter quantity, for reasonable choices of $n,m$
156: [reasonable pairs $(\varepsilon_{\tilde k},\delta)$], is a very
157: small number. However, one has to fix a {\it confidence level} of
158: comparison, for assessing existence of a local continuous function
159: between the two neighborhoods. This problem was addressed in
160: \cite{PecCarHea95}, where the {\it continuity statistics} method
161: was proposed. This consists in calculating the quantity $b_P$,
162: defined as
163:
164: \begin{eqnarray}
165: \label{uno} b_P = \max_{q=1,...,m} B(q,m;P),
166: \end{eqnarray}
167:
168: where $B(q,m;P)$ is the binomial distribution, giving the
169: probability that $q \leq m$ events out of $m$ attempts are
170: realized for a process of elementary probability $P$.
171:
172: As said above, the presence of a single data within $V$ has
173: probability $P(V)$. The quantity $b_P$ (for $P=P(V)$) represents
174: then the maximum over $q$ of the probability that, given $m$
175: points, $q$ out of them fall into $V$. Hence, a level of
176: confidence for the existence of a continuous function can be
177: estimated in terms of the ratio
178:
179: \begin{eqnarray}
180: \label{due} \Theta = \frac{P_m(V)}{b_P}.
181: \end{eqnarray}
182:
183: If $\Theta \approx 1$ we have no trustable information about the
184: existence of such a functional relationship, insofar as the chance
185: probability of having our $m$ points in $V$ is of the same order
186: of the maximum probability of having events in $V$ out of $m$
187: attempts. On the contrary, if $\Theta \ll 1$, the chance
188: probability of having our $m$ points in $V$ is negligible compared
189: to $b_P$. Thus one concludes that the two sets $U$ and $V$ are the
190: domain and co-domain respectively of a local continuous function
191: mapping states in {\bf X} close to $\tilde x$ to states in {\bf Y}
192: close to $\tilde y$. This answers the practical question of
193: predicting states in {\bf Y} with error $\delta$ from measurements
194: of states in {\bf X} with error $\varepsilon_{\tilde k}$.
195:
196: We have made use of the original formulation of the continuity
197: statistics \cite{PecCarHea95}, that explicitly considers $P=P(V)$
198: in Eq.(\ref{due}). More recently, the same Authors of
199: \cite{PecCarHea95} have proposed an alternative way for measuring
200: the confidence level, by choosing $P=1/2$ in the denominator of
201: Eq. (2), corresponding to an hypothesis of equal probability for
202: an attempt to fall within or outside the selected box
203: \cite{ECCpaper}.
204:
205: Our technique for characterizing synchronization consists then of
206: the three following points: {\it i)} check the imaging of
207: neighborhoods of a given configuration $\tilde x$ into
208: neighborhoods of $\tilde y$; {\it ii)} assess the degree of
209: confidence that such an imaging process comes from the existence
210: of a local continuous function; {\it iii)} repeat points i) and
211: ii) for all $N$ pairs of configurations ($\tilde x, \tilde y$)
212: available in the data set. This procedure allows a classification
213: of the different dynamical states into locally synchronized and
214: non synchronized ones. As a result one can introduce the {\it
215: synchronization points percentage} ({\it SPP}) indicator, as the
216: ratio between the total number $\tilde n$ of locally synchronized
217: configurations and the total number of available points $N$.
218:
219: The proposed method can be applied to any kind of multivariate
220: data set, for the detection of hidden local synchronization
221: properties, that cannot be detected by global indicators, such as
222: correlation functions, Lyapunov exponents, Lyapunov functionals,
223: or any other kind of time (or ensemble) average indicators that
224: unavoidably result in mixing locally synchronized and
225: unsynchronized configurations. As a result, the {\it SPP}
226: indicator furnishes relevant information in all those cases in
227: which synchronization states emerge locally in phase space, to
228: detect predictability properties that are limited to some subset
229: of the dynamics.
230:
231:
232: In order to illustrate the robustness of the method, in the
233: following we will refer to a test case, represented by a pair of
234: non identical bidirectionally coupled chaotic R\"ossler
235: oscillators. Here $m_1=m_2=3$, and the subspaces {\bf X} and {\bf
236: Y} contain state vectors {\bf x}$\equiv (x_1,y_1,z_1)$ and {\bf
237: y}$\equiv (x_2,y_2,z_2)$ whose evolution is ruled by
238:
239:
240: \begin{eqnarray}
241: %
242: \label{Rossler}
243: \dot{x}_{1,2}&=&-\omega
244: _{1,2}y_{1,2}-z_{1,2}+\epsilon ( x_{2,1}-x_{1,2}),\nonumber\\
245: \dot{y}_{1,2}&=&\omega _{1,2}x_{1,2}+0.165z_{1,2},\\
246: \dot{z}_{1,2}&=&0.2+z_{1,2}( x_{1,2}-10). \nonumber
247: %
248: \end{eqnarray}
249:
250: In Eqs.(\ref{Rossler}), $\omega _{1,2}=\omega _0 \pm \Delta$
251: represent the natural frequencies of the two chaotic oscillators,
252: $\omega _0=0.97$, $\Delta =0.02$ is the frequency mismatch and
253: $\epsilon>0 $ rules the coupling strength. As $\epsilon$
254: increases, the emergence of different synchronization features in
255: Eqs.(\ref{Rossler}) has been described and characterized in the
256: literature \cite{phase,intermittentlag}. Precisely, for $\epsilon
257: < 0.036$ no global synchronization (NS) is established, in terms
258: of the global indicators proposed up to now. For $0.036 \leq
259: \epsilon \preceq 0.11$ a phase synchronized (PS) regime emerges
260: characterized by the boundedness in time of the phase difference
261: $\Delta \phi \equiv \mid \phi_1 - \phi_2 \mid$ [$\phi_{1,2} \equiv
262: \arctan \left( \frac{y_{1,2}}{x_{1,2}} \right)$ being the phases
263: of the two oscillators], whereas the two chaotic amplitudes remain
264: almost uncorrelated \cite{phase}. At larger coupling strengths
265: ($\epsilon \geq 0.145$), lag synchronization (LS) is established,
266: corresponding to a collective motion wherein $\mid \text{{\bf
267: x}}(t) - \text{{\bf y}}(t-\tau) \mid$ is bounded over the whole
268: dynamical evolution ($\tau>0$ represents here a lag time)
269: \cite{intermittentlag}. In this regime, increasing $\epsilon$
270: results in gradually decreasing $\tau$, eventually ending with a
271: regime indistinguishable from complete synchronization (CS).
272:
273: Most of the transition points between these regimes were also
274: identified in Ref. \cite{intermittentlag}, by inspection of the
275: Lyapunov spectrum of Eqs.(\ref{Rossler}) as a function of the
276: coupling strength. Precisely, the NS to PS (PS to LS) transition
277: occurs for that value of $\epsilon$ for which a previously zero
278: (positive) Lyapunov exponent becomes negative. On the other hand,
279: the LS to CS transition is a smooth transition that can be tracked
280: by use of the time averaged similarity function
281: \cite{intermittentlag}. In the following we apply our method with
282: a threshold value of $\Theta =0.1$ for the discrimination of
283: whether or not the coupled systems display local functional
284: relationships.
285: %
286: \begin{figure}[!h]
287: \includegraphics[width=75mm]{feps.eps} \caption{ {\it SPP} indicator (see
288: text for definition) {\it vs.} coupling strength $\epsilon$. The
289: vertical dashed lines indicate the transition points between the
290: different synchronization regimes. The inset shows a zoom limited
291: to the range $0.05 < \epsilon <0.15$, where the PS to ILS
292: transition point is located at $\epsilon_{c} \simeq 0.10$. Notice
293: the two different slopes in the linear growth of {\it SPP} for
294: $\epsilon<\epsilon_{c}$ and $\epsilon>\epsilon_{c}$.}\label{fig2}
295: \end{figure}
296: %
297:
298: An intermediate synchronization regime between PS and LS exists in
299: the range $0.11 \preceq \epsilon < 0.145$, called intermittent lag
300: synchronization (ILS), where the system (\ref{Rossler}) displays
301: long epochs of LS evolution, interrupted by persistent bursts of
302: desynchronized motion. This has been observed numerically, and put
303: in relation with the system's trajectory passing through
304: configurations where one globally negative Lyapunov exponent has a
305: local positive value. Since ILS is an intimately local phenomenon,
306: its transition point has not been captured by those techniques
307: that measure time or ensemble averaged quantities. As a result, up
308: to now, studies on ILS have been limited to numerical
309: investigations \cite{intermittentlag}, or based upon the role in
310: the synchronization process played by the different unstable
311: periodic orbits visited by the dynamics \cite{pazozaks}. We will
312: show that our {\it SPP} indicator is able to discriminate between
313: ILS and PS regimes, as well as to directly identify the PS to ILS
314: transition point.
315:
316: We have performed long time simulations of Eqs.(\ref{Rossler}) at
317: several coupling strength values, and collected data points from
318: the two scalar outputs $x_1$ and $x_2$. For each $\epsilon$, data
319: points are collected over a time corresponding to $1.7\cdot 10^5$
320: R\"ossler cycles, with a sampling frequency of 10 points per
321: cycle. Simulations were performed with a standard fourth order
322: Runge-Kutta method, and with random initial conditions.
323: Furthermore, the standard embedding technique \cite{takens} was
324: used to reconstruct the three dimensional vector states {\bf x}
325: and {\bf y} from time-delayed coordinates of the scalar variables
326: $x_1$ and $x_2$, and calculation of the {\it SPP} indicator was
327: performed on the reconstructed spaces.
328:
329: \begin{figure}[!h]
330: \includegraphics[width=75mm]{fneps.eps}
331: \caption{a) {\it SPP} indicator {\it vs.} number $n$ of points
332: falling within $V$ for $\epsilon=0.01$ (triangles, NS),
333: $\epsilon=0.08$ (squares,PS) and $\epsilon=0.155$ (circles, LS).
334: b) {\it SPP} {\it vs.} $n$ within the LS regime for
335: $\epsilon=0.155$ (dotted line); $\epsilon=0.17$ (dashed line) and
336: $\epsilon=0.19$ (continuous line). For all cases, before
337: saturation ($n \leq 50$), {\it SPP} depends on $n$ with a scaling
338: law {\it SPP} $\sim n^\beta$ with $\beta \sim 0.85$. For $n > 50$
339: the three curves saturate to 100 \% of synchronization
340: points.}\label{fig3}
341: \end{figure}
342: %
343:
344:
345: Fig.\ref{fig2} reports the behavior of the {\it SPP} indicator
346: \textit{vs.} the coupling strength $\epsilon$, calculated by
347: fixing $\delta$ so as $n=150$ points are falling within $V$.
348: Fixing $n$ results in general in an error $\delta $ that is not
349: constant over the attractor. On the other side, if the measure is
350: strongly non homogeneous, fixing $\delta $ could generate
351: situations in which $n$ is so small that the statistics becomes
352: meaningless. These concerns do not apply however in the case of
353: the R\"ossler system for the parameters used in Eqs. (3), since
354: the density of points is roughly homogeneous over the attractor
355: and both choices lead to equivalent results. As one expects, {\it
356: SPP} increases monotonically as the coupling strength increases,
357: saturating to 1 when approaching the CS regime.
358:
359: Interesting novel information can be extracted by inspection of
360: {\it SPP} within those synchronization regimes, such as PS and ILS
361: that do not correspond to global synchronization features. In
362: particular, it is found that {\it SPP} is linearly increasing with
363: $\epsilon$ in both regimes, but with two different slopes (see the
364: inset of Fig.\ref{fig2}). The linear increase of the indicator
365: already within PS is a relevant result. Indeed, if and to which
366: extent PS implies weak correlations in the chaotic amplitudes was
367: yet unknown, and constituted an issue generating controversy. The
368: present result shows that PS does imply an increasing percentage
369: of local functional relationship, thus quantifying directly the
370: degree of amplitude synchronization within such a regime.
371: Furthermore, the crossover point between the slopes of the two
372: linear growths allows one to identify the PS to ILS transition
373: point at $\epsilon \simeq 0.10$, that none of the various
374: indicators used in previous works was capable to reveal.
375:
376: Finally, other novel information can be extracted from the scaling
377: behavior of {\it SPP} with $n$, that is with enlarging the radius
378: $\delta$ of the image box in the {\bf Y} subspace.
379: Fig.\ref{fig3}a) shows {\it SPP} \textit{vs.} $n$ for the NS, PS
380: and LS regimes. In all cases, the {\it SPP} indicator increases
381: monotonically. For LS (circles) it fastly saturates to 1 (the same
382: value as CS). This is reflecting the fact that LS differs from CS
383: only due to the presence of a lag time $\tau$. Enlarging too much
384: the neighborhood size results in $V$ to fully overlap with all
385: images of points in $U$ shifted by a phase factor $\omega \tau$,
386: where $\omega$ is the mean frequency of the oscillator, thus
387: making indistinguishable LS from CS.
388:
389: More insights on this property can be extracted from
390: Fig.\ref{fig3}b), where {\it SPP} is reported \textit{vs.} $n$
391: within the LS regime for different values of $\epsilon$,
392: corresponding to different values of the lag time $\tau$. Here one
393: sees that, before saturation, {\it SPP} depends on $n$ with a
394: scaling law {\it SPP} $\sim n^\beta$ with a unique exponent $\beta
395: \sim 0.85$ for the three $\epsilon$ values. However, the three
396: curves saturate to 1 at three different values of $n$, reflecting
397: the behavior of $\tau$ within LS, that monotonically decreases as
398: $\epsilon$ increases.
399:
400: Coming again to Fig.\ref{fig3}a), one realizes that for both NS
401: (triangles) and PS (squares), the {\it SPP} indicator is always
402: bounded away from 1. This indicates that in these regimes a global
403: predictability of one subsystem's states from measurement in the
404: other subspace is never possible for any choice of resolution.
405: However, given a resolution $\delta$ in the image subspace (a
406: maximum error allowed in the prediction), our indicator quantifies
407: the number of states that can be locally predicted at that
408: resolution, thus revealing that local hidden synchronization
409: features can be extracted for prediction purposes, also in those
410: cases in which global dependencies are not established. This
411: feature might be relevant for detecting configurations where a
412: local prediction can be assessed, in many situations where a
413: global prediction procedure fails.
414:
415: In real data, the effect of noise is to reduce the resolution in
416: the phase space, so that the statistics relative to boxes
417: containing a small number of points is not reliable anymore. A
418: threshold in $n$ should therefore be introduced, typically
419: corresponding to $\delta$'s larger than the noise-induced
420: uncertainty.
421:
422: Authors are indebted with L. Moniz and L.M. Pecora for many
423: fruitful discussions. Work partially supported by EU Contract
424: HPRN-CT-2000-00158, and MIUR Project FIRB n. RBNE01CW3M\_001.
425: L.P. acknowledges support from contract MCFI-2000-01822.
426:
427:
428: \begin{thebibliography}{99}
429:
430: \bibitem{reviews}
431: For an overview on this matter we address the reader to: A.
432: Pikovsky, M. Rosenblum and J. Kurths, {\it Synchronization: A
433: Universal Concept in Nonlinear Sciences}, (Cambridge University
434: Press, 2001); S. Boccaletti, J. Kurths, G. Osipov, D. Valladares
435: and C. Zhou, Phys. Rep. {\bf 366}, 1, (2002).
436:
437: \bibitem{complete}
438: H. Fujisaka and T. Yamada, Prog. Theor. Phys. {\bf 69}, 32 (1983);
439: L.M. Pecora and T.L. Carroll, Phys. Rev. Lett. {\bf 64}, 821
440: (1990).
441:
442: \bibitem{generalized}
443: N.F. Rulkov, M.M. Sushchik, L.S. Tsimring and H.D.I. Abarbanel,
444: Phys. Rev. {\bf E51}, 980 (1995); L. Kocarev and U. Parlitz,
445: Phys. Rev. Lett. {\bf 76}, 1816, (1996).
446:
447: \bibitem{phase}
448: M.G. Rosenblum, A.S. Pikovsky and J. Kurths, Phys. Rev. Lett. {\bf
449: 76}, 1804 (1996).
450:
451: \bibitem{intermittentlag}
452: M.G. Rosenblum, A.S. Pikovsky and J. Kurths, Phys. Rev. Lett. {\bf
453: 78}, 4193 (1997); S. Boccaletti and D.L. Valladares, Phys. Rev.
454: {\bf E62}, 7497 (2000).
455:
456: \bibitem{nature}
457: C. Schafer, M.G. Rosemblum, J. Kurths and H.H. Abel, Nature {\bf
458: 392}, 239 (1998); P. Tass, M.G. Rosemblum, M.G. Weule, J. Kurths,
459: A. Pikovsky, J. Volkmann, A. Schnitzler and H.J. Freund, Phys.
460: Rev. Lett. {\bf 81}, 3291 (1998); G.D. Van Wiggeren and R. Roy,
461: Science {\bf 279}, 1198 (1998); A. Neiman, X. Pei, D. Russell, W.
462: Wojtenek, L. Wilkens, F. Moss, H.A. Braun, M.T. Huber and K.
463: Voigt, Phys. Rev. Lett. {\bf 82}, 660 (1999); G. M. Hall, S. Bahar
464: and D.J. Gauthier, Phys. Rev. Lett. {\bf 82}, 2995 (1999);B.
465: Blasius, A. Huppert and L. Stone, Nature, {\bf 399}, 354 (1999);
466: D. J. DeShazer, R. Breban, E. Ott and R. Roy, Phys. Rev. Lett.
467: {\bf 87}, 044101 (2001).
468:
469: \bibitem{experiment}
470: C.M. Ticos, E. Rosa Jr., W.B. Pardo, J.A. Walkenstein and M.
471: Monti, Phys. Rev. Lett. {\bf 85}, 2929 (2000); D. Maza, A.
472: Vallone, H. Mancini and S. Boccaletti, Phys. Rev. Lett. {\bf 85},
473: 5567 (2000); E.Allaria, F.T Arecchi, A. Di Garbo and R. Meucci,
474: Phys. Rev. Lett. {\bf 86}, 791 (2001).
475:
476: \bibitem{previous}
477: I.I. Blekhman, A.L. Fradkov, H. Nijmeijer and A. Yu. Pogromsky,
478: Systems \& Control Letters {\bf 31}, 299 (1997); R. Brown and L.
479: Kocarev, Chaos {\bf 10}, 344 (2000).
480:
481: \bibitem{definition}
482: S. Boccaletti, Louis M. Pecora, A. Pelaez, Phys. Rev. {\bf E63},
483: 066219 (2001).
484:
485: \bibitem{disentangle}
486: S. Boccaletti, D.L. Valladares, L.M. Pecora, H.P. Geffert and T.
487: Carroll, Phys. Rev. {\bf E65}, 035204 (2002).
488:
489: \bibitem{abarbanel}
490: M.B. Kennel, R. Brown and H.D.I. Abarbanel, Phys. Rev. {\bf A45},
491: 3403 (1992); H.D.I. Abarbanel, {\it Analysis of Observed Chaotic
492: Data} (Springer-Verlag, New York, 1996).
493:
494:
495: \bibitem{PecCarHea95}
496: L. Pecora, T. Caroll, J. Heagy, Phys. Rev. {\bf E52}, 3420 (1995).
497:
498: \bibitem{ECCpaper}
499: L. Moniz and L. M. Pecora, private communication.
500:
501: \bibitem{pazozaks}
502: D. Paz\'o, M. A. Zaks and J. Kurths, Chaos {\bf 13}, 309 (2003).
503:
504: \bibitem{takens}
505: F. Takens, in {\it Detecting Strange Attractors in Turbulence},
506: ed. by D.A. Rand and L.S. Young, Lecture Notes in Mathematics,
507: Vol. 898 (Springer-Verlag, New York, 1981); T. Sauer, M. Casdagli
508: and J. A. Yorke, J. Stat. Phys. {\bf 65}, 579 (1991).
509:
510: \end{thebibliography}
511:
512: \end{document}
513: