1: %\documentclass[preprint,pre,aps,showpacs]{revtex4}

2: \documentclass[twocolumn,pre,aps,showpacs]{revtex4}

3: \usepackage{graphicx}

4: \usepackage{amsmath}

5: %

6: \def\bear{\begin{eqnarray}}

7: \def\eear{\end{eqnarray}}

8: %

9: \newcommand{\Fig}[1]{Fig.~\ref{#1}}

10: \newcommand{\Figs}[1]{Figs.~\ref{#1}}

11: \newcommand{\Eqn}[1]{Eq.~(\ref{#1})}

12: \newcommand{\Eqns}[1]{Eqs.~(\ref{#1})}

13: \newcommand{\Sec}[1]{Sec. \ref{#1}}

14: \newcommand{\Secs}[1]{Secs. \ref{#1}}

15: %\psdraft

16: \begin{document}

17: \bibliographystyle{revtex}

18: 

19: 

20: \title{On the conditions for synchronization in unidirectionally

21: coupled chaotic oscillators}

22: 

23: \author{

24: P.~Muruganandam%

25: }

26: \email{anand@bdu.ernet.in}

27: \affiliation{Centre for Nonlinear Dynamics, Department of Physics

28: Bharathidasan University, Tiruchirapalli 620 024, India}

29: \author{

30: S.~Parthasarathy%

31: }

32: \email{spartha@rect.ernet.in}

33: \affiliation{Department of Physics, Regional Engineering College,

34: Tiruchirapalli - 620 015, India}

35: \author{

36: M.~Lakshmanan%

37: }

38: \email{lakshman@bdu.ernet.in}

39: \affiliation{Centre for Nonlinear Dynamics, Department of Physics

40: Bharathidasan University, Tiruchirapalli 620 024, India}

41: \date{ }

42: 

43: \begin{abstract}

44: 

45: The conditions for synchronization in unidirectionally coupled chaotic

46: oscillators are revisited. We demonstrate with typical examples that

47: the conditional Lyapunov exponents (CLEs) play an important role in

48: distinguishing between intermittent and permanent synchronizations,

49: when the analytic conditions for chaos synchronization are not

50: uniformly obeyed. We show that intermittent synchronization can occur

51: when CLEs are very small positive or negative values close to zero

52: while permanent synchronization occurs when CLEs take sufficiently

53: large negative values. There is also strong evidence for the fact that

54: for permanent synchronization the time of synchronization is relatively

55: low while it is high for intermittent synchronization.

56: 

57: \end{abstract}

58: \pacs{05.45.Xt}

59: \maketitle

60: 

61: 

62: \section{Introduction}

63: 

64: Much attention has been focussed on the synchronization of chaotic

65: systems through different coupling schemes during the past decade or

66: so. It has potential applications not only in secure communication but

67: also in chaotic cryptography\cite{vaidya0}. In order to achieve

68: synchronization, coupled chaotic systems have to satisfy certain

69: conditions. According to Pecora and Carroll the coupled chaotic systems

70: can be regarded as master and slave systems which will perfectly

71: synchronize only if the sub-Lyapunov or conditional Lyapunov exponents

72: (CLEs) are all negative\cite{pecora1,pecora2}. However, recently there

73: are reports claiming that synchronization can be achieved even with

74: positive CLEs\cite{shuai,guemez,gutierrez}. Also there are situations

75: where one can observe intermittent or round-off induced synchronization

76: phenomenon\cite{baker,baker2,zhou} where locking occurs at certain time

77: intervals only, so that the synchronization of chaos need not be always

78: permanent. It is therefore important to understand clearly the

79: conditions under which chaos synchronization can occur and to know how

80: to distinguish between permanent and intermittent synchronizations

81: exhibited by coupled chaotic systems.

82: 

83: In this paper, the analytic conditions for synchronization in coupled

84: chaotic systems are revisited to show the difficulties in

85: distinguishing between intermittent and permanent synchronizations

86: which we illustrate by using simple chaotic systems. We demonstrate

87: that the conditional Lyapunov exponents (CLEs) play an important role

88: in distinguishing between intermittent and permanent synchronization,

89: when the analytic conditions for chaos synchronization are not

90: uniformly obeyed. We also point out that intermittent synchronization

91: can occur when the largest CLE has a value close to zero (either

92: positive or negative) while permanent synchronization occurs for

93: sufficiently large negative values of the CLEs.

94: 

95: In \Sec{sec2}, we describe briefly the notion of chaos synchronization

96: and the analytic conditions for synchronization which are to be

97: uniformly obeyed as well as the existence of negative conditional

98: Lyapunov exponents, with reference to a simple coupled system of

99: autonomous Duffing-van der Pol (ADVP) oscillators\cite{lakshman,

100: murali1}. \Sec{sec:cles} describes how the CLEs can be used to analyse

101: the nature of synchronization in coupled ADVP oscillators.  In

102: \Sec{sec3}, we show how conditional Lyapunov exponents play an

103: important role in distinguishing between the intermittent and permanent

104: synchronizations occurring in a class of unidirectionally coupled

105: nonlinear oscillators including coupled chaotic pendula\cite{baker},

106: coupled Duffing oscillators\cite{baker} and coupled

107: Murali-Lakshmanan-Chua (MLC) circuits\cite{lakshman} either in the

108: absence of the analytic criteria or when they are not uniformly obeyed.

109: We also show how the time of synchronization can be used qualitatively

110: to check the presence or absence of intermittent synchronization.

111: \Sec{sec4} summarizes the results.

112: 

113: \section{Chaos synchronization and the conditions}

114: \label{sec2}

115: 

116: In this section, we consider a set of two unidirectionally coupled

117: autonomous Duffing-van der Pol (ADVP) oscillators as a model system to

118: analyse the concept of chaos synchronization. Further, we show that the

119: derived analytic conditions for the occurrence of chaos synchronization

120: are uniformly obeyed in this system for a specific set of parametric

121: values (that is, the analytic conditions are shown to be valid for all

122: values of the variables, say, ($x_1, x_2, x_3) \in {\cal R}^3$ for all

123: $t$). We also indicate that when the usual analytic conditions are not

124: uniformly obeyed by a given system, then it may either exhibit

125: intermittent or permanent synchronization\cite{baker}. In this case,

126: the nature of the conditional Lyapunov exponents can be used profitably

127: to distinguish between intermittent and permanent synchronization.

128: \begin{figure}[!ht]

129: \includegraphics[width=\linewidth]{advp_eta.eps}

130: 

131: \caption{Figure showing the variation of synchronization error $\eta$

132: as time elapses in ADVP oscillators [\Eqns{eqn:advp:1}] for $\nu=100$,

133: $\alpha=0.35$, $\beta=300$ and $\varepsilon = 1.35$.}

134: 

135: \label{ch5_f2}

136: \end{figure}

137: 

138: We consider the ADVP model in which Murali and Lakshmanan demonstrated

139: chaos synchronization \cite{lakshman,murali1}. The rescaled and

140: dimensionless version of the unidirectionally coupled ADVP oscillators

141: can be written as

142: \begin{subequations}

143: \label{eqn:advp:1}

144: \bear

145: \dot x_m & = & -\nu (x_m^3-\alpha x_m-y_m),\nonumber \\

146: \dot y_m & = & x_m-y_m-z_m, \nonumber \\

147: \dot z_m & = & \beta y_m ,

148: \label{eqn:advp:1a} \\

149: \nonumber \\

150: \dot x_s & = & -\nu (x_s^3-\alpha x_s-y_s) + \nu\varepsilon(x_m-x_s),\nonumber \\

151: \dot y_s & = & x_s-y_s-z_s, \nonumber \\

152: \dot z_s & = & \beta y_s.

153: \label{eqn:advp:1b}

154: \eear

155: \end{subequations}

156: Here $\nu$, $\alpha$, and $\beta$ are rescaled parameters, which are

157: fixed at $\nu=100$, $\alpha=0.35$ and $\beta=300$. One can define a

158: measure of the synchronization error, $\eta$, as

159: \begin{equation}

160: \eta = \sqrt{(x_m-x_s)^2+(y_m-y_s)^2+(z_m-z_s)^2}.

161: \label{eqn:eta}

162: \end{equation}

163: For synchronization, the above measure $\eta\to 0 $ as $t\to\infty$.

164: The synchronization error $\eta$ versus time is plotted in

165: \Fig{ch5_f2}.  The falling up of $\eta$ to zero is an indication of

166: synchronization at a finite time. However, this alone does not ensure

167: that the synchronization is permanent and one has to verify additional

168: criteria.  For this purpose, we will first discuss conditions for which

169: one can achieve permanent synchronization.

170: 

171: \renewcommand{\theenumi}{(\roman{enumi})}

172: \begin{enumerate}

173: \item

174: 

175: The criterion introduced by Fujisaka and Yamada \cite{fuji1, fuji2} for

176: high quality synchronization requires that the largest eigenvalue of

177: the Jacobian matrix corresponding to the flow evaluated on the

178: synchronization manifold be negative. In order to check this for the

179: system (\ref{eqn:advp:1}), let us consider the specific choice

180: $\varepsilon = 1+\alpha = 1.35$ in \Eqns{eqn:advp:1}. In this case, then one

181: can write the difference system of the ADVP oscillators for

182: $x^*=(x_m-x_s)$, $y^*=(y_m-y_s)$, $z^*=(z_m-z_s)$ in matrix form as

183: \begin{equation}

184: \left(

185: \begin{array}{c}

186: \dot x^* \cr \dot y^* \cr \dot z^*

187: \end{array}

188: \right)

189: =

190: \left(

191: \begin{array}{ccc}

192: -\nu (1+a) & \nu & 0 \cr

193: 1 & -1 & -1 \cr

194: 0 & \beta & 0

195: \end{array}

196: \right)

197: \left(

198: \begin{array}{c}

199: x^* \cr y^* \cr z^*

200: \end{array}

201: \right),

202: \label{eqn:advp:3}

203: \end{equation}

204: where $a = (x_m-x_s)^2+3x_mx_s \ge 0$. The

205: slave system (\ref{eqn:advp:1b}) synchronizes perfectly with the master

206: system (\ref{eqn:advp:1a}) only if all the eigenvalues of the above

207: linear system (\ref{eqn:advp:3}) possess negative real parts. One can

208: easily prove that this is indeed the case for (\ref{eqn:advp:3}).

209: 

210: \item

211: 

212: Recently, He and Vaidya~\cite{vaidya} developed a criterion for chaos

213: synchronization based on the notion of asymptotic stability of

214: dynamical systems, which refers to the condition for a given chaotic

215: system with master-slave configuration to reach the same eventual state

216: at a fixed (but sufficiently far enough) time irrespective of the

217: choice of initial conditions. One of the practical ways to establish

218: the asymptotic stability of the response subsystem is to find a

219: suitable Lyapunov function which can be defined as the square of the

220: magnitude of the vector describing the distance from the

221: synchronization manifold. Then the condition for all the perturbations

222: to decay to the synchronization manifold, without transient growth, is

223: that the time rate of the Lyapunov function has a negative magnitude

224: for all times\cite{gauthier}.

225: 

226: Now for the difference system (\ref{eqn:advp:3}), one can analytically

227: write a Lyapunov function in the following form for 

228: $\beta, \nu > 0$\cite{lakshman,murali1},

229: \begin{equation}

230: E = \frac{\beta}{2}\left[x^{*2}+\nu y^{*2}\right]+\frac{\nu}{2}z^{*2} \ge 0.

231: \end{equation}

232: Then the rate of change of $E$ along the trajectories is given by

233: \begin{equation}

234: \dot E = -\nu\beta \left [ ax^{*2} +(x^*-y^*)^2 \right] \le 0. 

235: \end{equation}

236: Since $E$ is a positive definite function and $\dot E$ is negative

237: definite for $(x^*, y^*, z^*)\in {\cal R}^3$, according to Lyapunov

238: theorem $x^*$, $y^*$ and $z^* \to 0$ as $t\to \infty$.  Thus perfect

239: synchronization occurs as $t \to \infty$.

240: 

241: \item

242: 

243: In addition, if all the sub-Lyapunov or conditional Lyapunov exponents

244: (CLE) are negative then one \emph{may} have perfect synchronization

245: between the master and slave systems. The CLEs are the corresponding

246: Lyapunov exponents of the slave system. For the coupled ADVP

247: oscillators [\Eqns{eqn:advp:1}] the CLEs can be calculated numerically

248: using the standard algorithm\cite{wolf}. We find that for the system

249: (\ref{eqn:advp:1}) with $\varepsilon = 1+\alpha = 1.35$, the numerical

250: value of the largest CLE is $-0.3134$. Thus, one may expect that the

251: ADVP oscillators will synchronize perfectly for $\varepsilon = 1.35$

252: which is indeed true.

253: \end{enumerate}

254: Consequently the conditions for synchronization in coupled chaotic

255: systems may be atleast any one of the following:

256: \begin{enumerate}

257: 

258: \item 

259: \label{syn:cond:1}

260: 

261: The largest eigenvalue of the Jacobian matrix corresponding to the flow

262: evaluated on the synchronization manifold \emph{must always be

263: negative} [see \Eqn{eqn:advp:3}].

264: 

265: \item 

266: \label{syn:cond:2}

267: 

268: Existence of a suitable Lyapunov function for the difference system

269: as discussed earlier.

270: 

271: \item

272: \label{syn:cond:3}

273: 

274: The sub-Lyapunov exponents or CLEs are all negative.

275: 

276: \end{enumerate} 

277: Among these three conditions, it is obvious that either of the

278: conditions \ref{syn:cond:1} and \ref{syn:cond:2} is both necessary and

279: sufficient for perfect or permanent synchronization as the very

280: definition of the latter implies these conditions. On the other hand

281: the condition \ref{syn:cond:3} is only a necessary one, because the

282: CLEs pertain to finite time averages only. As we have discussed above

283: for the specific parametric ($\varepsilon = 1.35$) case of the coupled

284: ADVP oscillators (\ref{eqn:advp:1}), all the conditions

285: \ref{syn:cond:1} - \ref{syn:cond:3} are satisfied and so pemanent

286: synchronization indeed occurs. However, suppose for a given system, if

287: the first two conditions are not uniformly obeyed and only the third

288: condition is satisfied, the question is whether the negativity of CLEs

289: alone ensures permanent synchronization.

290: 

291: In fact, recent studies show that in certain specific systems the

292: observed synchronization is \emph{not always} a perfect one and rather

293: one can have an \emph{intermittent synchronization}\cite{baker, muru1}

294: even when all the CLEs are negative. Further, there are reports stating

295: that one can achieve synchronization in the presence of positive

296: conditional Lyapunov exponents as well\cite{shuai}. So one has to

297: clarify the conditions under which permanent synchronization can occur

298: in coupled chaotic systems and how one can distinguish between the

299: intermittent and permanent synchronization.

300: 

301: \section{Nature of CLEs in ADVP oscillators

302: [E\lowercase{qs.}~(\ref{eqn:advp:1})] for general $\varepsilon$}

303: \label{sec:cles}

304: Based on the analysis of the previous section, one understands that the

305: coupled ADVP oscillators (\ref{eqn:advp:1}) show perfect

306: synchronization for $\varepsilon=1.35$. Now what happens to other

307: values of $\varepsilon$ in the coupled system (\ref{eqn:advp:1})? It

308: does not seem to be feasible either to deduce analytically the nature

309: of the largest eigenvalue of the Jacobian corresponding to the flow

310: evaluated on the synchronization manifold or to obtain explicitly a

311: suitable Lyapunov function.

312: 

313: In order to understand the nature of synchronization in

314: \Eqns{eqn:advp:1} for $\varepsilon\ne 1.35$, let us calcuate the CLEs.

315: \Fig{advp_lyex} shows the variation of the largest CLE as a function of

316: \begin{figure}[!ht]

317: \begin{center}

318: \includegraphics[width=\linewidth]{advp_lyex.eps}

319: \end{center}

320: 

321: \caption{Figure showing the variation of largest CLE as a function of

322: coupling strength ($\varepsilon$) for ADVP oscillators.}

323: 

324: \label{advp_lyex}

325: \end{figure}

326: the coupling strength ($\varepsilon$). From the figure, one may expect

327: that the ADVP oscillators will synchronize for $\varepsilon \ge 0.65$.

328: However, the largest CLE value is almost zero in the range of

329: $\varepsilon\in (0.5, 0.65)$. On careful observation, by examining the

330: synchronization error $\eta$ with the addition of a small amount of

331: noise at every integration step, one finds that perfect synchronization

332: will occur in the coupled ADVP oscillators for $\varepsilon \ge 0.65$

333: only.

334: 

335: In addition, the Jacobian matrix corresponding to the flow evaluated on

336: the synchronization manifold can be given by

337: \begin{equation} 

338: \label{eq:advp:jacob}

339: J = \left( \begin{array}{ccc} -\nu

340: (3x_{\text{sm}}^2+c) & \nu & 0 \cr 1 & -1 & -1 \cr 0 & \beta & 0

341: \end{array} \right), 

342: \end{equation} 

343: where $\varepsilon = \alpha+c$ ($c=\text{constant}$), $\alpha = 0.35$. 

344: Then the eigenvalues take the form

345: \begin{eqnarray} 

346: \lambda_{1,2} & = &-\frac{1+\nu(3x_{\text{sm}}^2+c)}{2} \nonumber \\

347:  & &\pm\frac{\sqrt{[1+\nu(3x_{\text{sm}}^2+c)]^2

348: -4\nu(3x_{\text{sm}}^2+c-1)}}{2}, \nonumber\\

349:  \lambda_3 & = & 0.

350: \end{eqnarray} 

351: In the above, the sign of the largest eigenvaule is determined by the

352: factor $(3x_{\text{sm}}^2+c-1)$ inside the square root sign. When $c=1$

353: (that is, $\varepsilon = 1.35$), the largest eigenvalue is negative as

354: discussed earlier. Further, it is evident that the largest eigenvalue

355: remains negative for $c>1$.

356: 

357: On the other hand, if $c < 1$ then one has to evaluate the eigenvalues

358: numerically in order to check the nature of synchronization. For this

359: purpose we examine the variation of the eigenvalues numerically.

360: \begin{figure}[!ht] 

361: \begin{center}

362: \includegraphics[width=\linewidth]{advp_lf.eps}

363: \end{center}

364: \caption{Figure showing the plot of the largest eigenvalue of the

365: Jacobian matrix (\ref{eq:advp:jacob}) versus time for ADVP oscillators:

366: (a) $\varepsilon=0.60$, (b) $\varepsilon=0.65$, (c) $\varepsilon=1.00$

367: and (d) $\varepsilon=1.35$.}

368: 

369: \label{advp_lf}

370: \end{figure}

371: \Figs{advp_lf} show the variation of the largest eigenvalue as a

372: function of time. \Fig{advp_lf}(a) illustrates the variation of the

373: largest eigenvalue for $\varepsilon=0.60$ ($c=0.25$). Here the largest

374: eigenvalue oscillates chaotically about zero with an average of the

375: order of $+10^{-1}$ and hence no synchronization is possible for

376: $\varepsilon=0.60$. For $\varepsilon=0.65$, the largest eigenvalue

377: still oscillates chaotically about zero [see \Fig{advp_lf}(b)] with an

378: average $\approx 10^{-2}$. In the latter case the largest eigenvalue

379: spends more time in the negative region despite the fact that its

380: average value is positive. In addition, for $\varepsilon=0.65$, the CLE

381: has a value of $-0.01298$ and thus it shows permanent synchronization.

382: Similar arguments hold good for $0.65 < \varepsilon < 1.35$.

383: \Figs{advp_lf}(c) and \ref{advp_lf}(d) depict the largest eigenvalue

384: for $\varepsilon=1.00$ and $\varepsilon=1.35$, respectively.

385: 

386: In addition to the above, one can also verify the condition

387: \ref{syn:cond:2}, by evaluating the time rate of the Lyapunov function

388: \begin{figure}[!ht] 

389: \begin{center}

390: \includegraphics[width=\linewidth]{advp_lyap.eps}

391: \end{center}

392: \caption{Figure showing the plot of $\dot E$ in the ADVP oscillators

393: [Eq.~\ref{eq:advp:edot}] as a function of time for (a)

394: $\varepsilon=0.64$ and (b) $\varepsilon=0.65$.}

395: \label{advp_lyap}

396: \end{figure}

397: corresponding to the square of the magnitude of the vector describing

398: the distance from the synchronization manifold\cite{gauthier}, which is

399: given by

400: \begin{eqnarray}

401: \label{eq:advp:edot}

402: \dot E & = & -\nu(3x_{\text{sm}}^2+c)x^{*2}-y^{*2} \nonumber \\

403: & & +(1+\nu)x^*y^*+(\beta+1)y^*z^* \le 0. 

404: \end{eqnarray}

405: \Fig{advp_lyap}(a) shows the variation of $\dot E$ for

406: $\varepsilon=0.64$. From the figure it is clear that the condition

407: \ref{syn:cond:2} is not uniformly satisfied. Thus, there is no

408: synchronization in the coupled ADVP oscillators (\ref{eqn:advp:1}) for

409: $\varepsilon=0.64$. However, $\dot E\to 0$ at a finite time for

410: $\varepsilon=0.65$ and $\varepsilon>0.65$ [see \Fig{advp_lyap}(b)] and

411: hence perfect synchronization does occur in the coupled ADVP

412: oscillators for $\varepsilon\ge 0.65$.

413: 

414: Thus from the above analysis and from the nature of the CLEs, one may

415: conclude that perfect synchronization in the coupled ADVP oscillators

416: (\ref{eqn:advp:1}) will occur for $\varepsilon \ge 0.65$. The role of

417: CLEs become even more important when both the analytic conditions

418: \ref{syn:cond:1} and \ref{syn:cond:2} are not uniformly satisfied. In

419: such cases, in the following, we will analyse how the nature of CLEs

420: can be effectively used to understand synchronization in typical

421: coupled chaotic systems.

422: 

423: \section{Distinguishing intermittent and permanent synchronization

424: using CLEs}

425: 

426: \label{sec3}

427: 

428: We now wish to clarify the conditions for which one can observe

429: intermittent and permanent synchronization, albeit in a qualitative

430: way, when the analytic conditions \ref{syn:cond:1} and \ref{syn:cond:2}

431: are not uniformly obeyed. In this regard we wish to analyse the nature

432: of synchronization on the basis of the conditional Lyapunov exponents

433: (CLEs) by considering three typical dynamical systems as examples. They

434: are namely (i) coupled chaotic pendula, (ii) coupled Duffing

435: oscillators and (iii) coupled MLC circuits. We find that all these

436: three systems do not satisfy the analytic conditions \ref{syn:cond:1}

437: and \ref{syn:cond:2} uniformly in contrast to the coupled ADVP

438: oscillators considered in Sec.~\ref{sec2}. We demonstrate how the

439: nature of the CLEs become important in distinguishing the intermittent

440: and permanent synchronizations exhibited by these systems.

441: 

442: \subsection{Coupled chaotic pendula}

443: \begin{figure*}[!ht]

444: \includegraphics[width=0.8\linewidth]{pen_cfig.eps}

445: 

446: \caption{(a) Variation of the conditional Lyapunov exponent (CLE) as a

447: function of $c$ and (b) synchronization error $\eta$ [see

448: \Eqn{eqn:pen:eta}] versus time in drive cycles for coupled pendula.}

449: 

450: \label{fig1}

451: \end{figure*}

452: 

453: First we consider a pair of coupled chaotic pendula defined by the

454: following set of equations and investigated by Baker, Blackburn and

455: Smith\cite{baker}

456: \begin{subequations}

457: \label{eq:pen1}

458: \bear

459: &&\ddot\theta_m+Q^{-1}\dot\theta_m =  \Gamma_0 \cos \Omega t ,

460: \label{eq:pen1a}\\

461: &&\ddot\theta_s+Q^{-1}\dot\theta_s  =  \Gamma_0 \cos \Omega t 

462: + c[\sin\theta_s - \sin\theta_m].

463: \label{eq:pen1b}

464: \eear

465: \end{subequations}

466: Here $\theta_m$ and $\theta_s$ correspond to the angular coordinates of

467: the master and slave systems, respectively. $Q^{-1}$ corresponds to the

468: damping factor, $\Gamma_0$ is the normalized drive torque, $\Omega$ is

469: the drive frequency and $c$ is the coupling strength. By fixing the

470: parameters at $Q=5.0$, $\Gamma_0=1.2$ and $\Omega=0.5$, the quality of

471: synchronization in the above coupled system (\ref{eq:pen1}) has been

472: studied. This has been facilitated by finding the synchronization error,

473: \begin{equation}

474: \eta = \sqrt{(\theta_s-\theta_m)^2+(\dot\theta_s-\dot\theta_m)^2}.

475: \label{eqn:pen:eta}

476: \end{equation}

477: As seen earlier, for synchronization $\eta\to 0$ as $t\to\infty$.

478: 

479: It has been shown by Baker, Blackburn and Smith\cite{baker} that

480: permanent synchronization in the above coupled identical chaotic

481: pendula (\ref{eq:pen1}) does not occur except as a numerical artifact

482: arising from finite computational precision. Further, they showed that

483: the synchronization of the above coupled pendula (\ref{eq:pen1}) is

484: always {\em intermittent}, for any value of the coupling coefficient,

485: by using numerical and analytical tests. However, contrary to the above

486: the present authors have reported that there exists atleast certain

487: range of coupling coefficient values for which one can observe

488: permanent synchronization by computing the CLEs\cite{muru1,muru2}. In

489: order to understand this we proceed as follows.

490: \begin{figure*}[!ht]

491: \includegraphics[width=\linewidth]{pen_eta.eps}

492: 

493: \caption{Figure showing the variation of synchronization error $\eta$

494: versus time  in the coupled chaotic pendula for (a) $c=0.80$, (b)

495: $c=0.81$, (c) $c=0.89$, (d) $c=0.92$, (e) $c=0.94$ and (f) $c=0.98$.

496: Fluctuation in $\eta$ shows the possibility of having intermittent

497: synchronization. In calculating $\eta$, tiny noise levels ($\approx

498: 10^{-15}$) are included at each time step to avoid round off induced

499: synchronization.}

500: 

501: \label{pen_eta}

502: \end{figure*}

503: 

504: For the coupled chaotic pendula (\ref{eq:pen1}), as shown in

505: Ref.~\cite{baker} the eigenvalues of the Jacobian matrix corresponding

506: to the flow evaluated on the synchronization manifold take the form

507: \begin{eqnarray}

508: \lambda_{1,2} & = & \frac{-1}{2Q}\left[

509: 1\pm\sqrt{1-2Q^2(1-c)\cos\theta_{\text{sm}}}

510: \right], \nonumber\\

511: \lambda_3 & = & 0.

512: \end{eqnarray}

513: Here $\theta_{\text{sm}}$ corresponds to the angle coordinate on the

514: synchronization manifold. On careful numerical observation, it has been

515: shown by Baker, Blackburn and Smith~\cite{baker} that the term inside

516: the square root sign varies chaotically about unity with an average

517: value which is slightly less than one for a range of $c$ values. This

518: implies that the largest eigenvalue is not always less than zero and

519: hence the condition \ref{syn:cond:1} is not obeyed uniformly.

520: 

521: Further, calculating the Lyapunov function which is the square of the

522: magnitude of the vector describing the distance from the

523: synchronization manifold\cite{gauthier}, the sufficient condition for

524: permanent synchronization can be written as\cite{baker}

525: \begin{equation}

526: \dot E = \left (1-c-\cos\theta_{\text{sm}}\right )\theta^{*}\omega^{*}

527: -Q^{-1}\omega^{*2} \le 0,

528: \end{equation}

529: where $\omega = \dot\theta$, $\theta^{*} = \theta_{m}-\theta_{s}$ and

530: $\omega^{*} = \omega_{m}-\omega_{s}$. The time series of the above

531: expression varies intermittently about zero with an average slightly

532: less than zero for a range of $c$ values [$c\in (0.79,1.0)$] and hence

533: the condition \ref{syn:cond:2} is also not obeyed for the coupled

534: chaotic pendula.  Thus, in the light of the above evidence, Baker,

535: Blackburn and Smith~\cite{baker,baker2} conclude that intermittent

536: synchronization can be a plausible behaviour in the coupled pendulum as

537: the locking occurs intermittently. In order to verify this assertion,

538: we have carried out an analysis\cite{muru1} by calculating the

539: conditional Lyapunov exponents of system (\ref{eq:pen1}) in the

540: parameter range $c\in (0.75,1.0)$. We find that there exists a range of

541: $c$ values for which one can indeed have permanent synchronization,

542: where Baker, Blackburn and Smith\cite{baker} have expected intermittent

543: synchronization. In our calculations we have used the standard Wolf et

544: al algorithm\cite{wolf} and used $5000$ drive cycles for calculations

545: after leaving out $5000$ drive cycles as transient.

546: 

547: Figure \ref{fig1}(a) shows the variation of CLE as a function of $c$

548: for the coupled pendula and it takes negative values for $0.796 \le c

549: <1.0$ only. The CLE value for $c=0.79$ is $+0.00320$ ($\approx

550: 10^{-3}$), for which synchronization can not occur and hence the

551: observed intermittent synchronization (see Fig.~1 in Ref~\cite{baker})

552: is a computer artifact. However, we find that for $c=0.932$, the CLE

553: value becomes the lowest ($-0.06825$) and it is relatively a large

554: negative value for which intermittent synchronization is absent as

555: shown in \Fig{fig1}(b). The absence of the intermittent synchronization

556: has been verified upto $2\times 10^6$ drive cycles even with the

557: addition of tiny noise levels showing that permanent synchronization

558: does occur for this value of $c$. The same phenomenon persists over a

559: range of $c$ values close to $0.932$. \Figs{pen_eta} show the variation

560: of the synchronization error for various $c$ values. From the figures

561: it is clear that one can indeed have intermittent synchronization for a

562: range of $c$ values as noted by Baker, Blackburn and

563: Smith\cite{baker}.  For example, the system definitely exhibits

564: intermittent synchronization for $c=0.80$, $0.81$ and $0.98$, where we

565: find that in this range of $c$ values (that is, $c\le 0.88$ or $c\ge

566: 0.95$) the largest CLEs take values very close to zero (either positive

567: or negative). But more interestingly we also find that there exists

568: another range of $c$ values from $0.89$ to $0.94$ where one can have

569: permanent synchronization. In calculating $\eta$ for

570: \Figs{pen_eta}(a)-(f), we have added a small amount of noise ($\approx

571: 10^{-15}$) at each time step to avoid the effect of round off induced

572: \begin{figure}[!h]

573: \includegraphics[width=\linewidth]{pen_tser.eps}

574: \caption{Figure showing the variation of time of synchronization 

575: $t_s$ (in drive cycles) versus coupling strength for coupled chaotic

576: pendula.}

577: \label{pen_ts}

578: \end{figure}

579: synchronization. We note that the CLEs are negative and the largest CLE

580: is relatively away from zero in this range. Thus we note that the

581: actual values of the CLEs, including the largest CLE, seem to

582: distinguish between permanent and intermittent synchronization.

583: 

584: Next we calculate the time of synchronization ($t_s$), that is, the

585: time taken by the system for which the synchronization error becomes

586: zero. \Fig{pen_ts} depicts the variation of $t_s$ as a function of the

587: coupling coefficient. It can be easily seen that the value of $t_s$ is

588: very low for the range of $c$ values where the CLEs are large negative

589: which corresponds to permanent synchronization. Similarly, the value of

590: $t_s$ becomes high in the regions where the largest CLE has a value

591: close to zero (either negative or positive) which corresponds to

592: intermittent synchronization, as there exists large fluctuations in the

593: finite time Lyapunov exponents here.

594: 

595: It is clear from the above analysis on the coupled chaotic pendula

596: (\ref{eq:pen1}) that there exists intermittent synchronization for

597: certain ranges of coupling strength values. However, permanent

598: synchronization does occur for certain other range of $c$ values. In

599: particular, we have pointed out that when the CLEs become very close to

600: zero (either positive or negative) one can have intermittent

601: synchronization while permanent synchronization does occur when the

602: CLEs become large negative. Thus the CLEs play a crucial role in

603: distinguishing between intermittent and permanent synchronization. In

604: the following subsections we will examine the existence of similar

605: intermittent synchronization in the coupled Duffing equations and

606: coupled MLC circuits.

607: 

608: \subsection{Coupled Duffing oscillators}

609: 

610: Now we consider the case of two coupled Duffing oscillators described

611: by the following set of equations,

612: \begin{subequations}

613: \bear

614: &&\ddot x_m + \alpha \dot x_m + \beta x_m^3 = f\cos \omega t, \\

615: &&\ddot x_s + \alpha \dot x_s + \beta x_s^3 = f\cos \omega t + c(x_s - x_m),

616: \eear

617: \end{subequations}

618: where $\alpha = 0.1$, $\beta = 1.0$, $f=10$ and $\omega=1.0$ and $c$ is

619: the coupling strength. By setting $\dot x = y$, $z = \omega t$, one can 

620: rewrite the above equations as a set of first order ordinary differential

621: equations of the form

622: \begin{subequations}

623: \label{duf:syn1} 

624: \bear

625: \dot x_m & = & y_m \nonumber\\

626: \dot y_m & = & -\alpha y_m -\beta x_m^3 + f\cos z_m \nonumber\\

627: \dot z_m & = & \omega

628: %\eear

629: \\ \nonumber \\

630: %\bear

631: \dot x_s & = & y_s \nonumber\\

632: \dot y_s & = & -\alpha y_s -\beta x_s^3 + f\cos z_s + c(x_s - x_m) \nonumber\\

633: \dot z_s & = & \omega

634: \eear

635: \end{subequations}

636: 

637: First, we study the nature of synchronization in the above coupled

638: Duffing equations by analysing the conditions \ref{syn:cond:1} and

639: \ref{syn:cond:2} of Sec.~\ref{sec2} as in the case of chaotic pendula

640: \begin{figure}[!ht]

641: \includegraphics[width=0.9\linewidth]{duf_cfig}

642: \caption{

643: (a) Variation of the conditional Lyapunov exponent (CLE) as a function

644: of $c$ and (b) synchronization error $\eta$ for coupled Duffing oscillators.

645: }

646: \label{duf_cfig}

647: \end{figure}

648: discussed above.  Here the eigenvalues of the Jacobian matrix

649: corresponding to the flow evaluated on the synchronization manifold can

650: be explicitly written as

651: \begin{equation}

652: \lambda_{1,2} = \frac{-\alpha\pm\sqrt{\alpha^2 

653: + 4(3\beta x_{\text{sm}}^2+c)}}{2\alpha},\;\;\;\;

654: \lambda_3 = 0,

655: \end{equation}

656: where $x_{\text{sm}}$ denotes the $x$-variable evaluated on the

657: synchonization manifold. We found that the largest of eigenvalue

658: oscillates chaotically about zero with an average value very close to

659: \begin{figure*}[!ht]

660: \includegraphics[width=0.9\linewidth]{duf_eta.eps}

661: \caption{Figure showing the variation of synchronization error $\eta$

662: versus time  in the coupled Duffing oscillators for (a) $c=2.74$, (b)

663: $c=2.92$, (c) $c=3.36$ and (d) $c=3.84$ with the inclusion of tiny noise 

664: levels of the order of $10^{-15}$ at each integration step.}

665: \label{duf_eta}

666: \end{figure*}

667: zero ($\approx\pm 10^{-3}$) for two ranges of $c\in (1.48,1.64)$ and $c\in

668: (2.74,3.84)$. Thus the condition \ref{syn:cond:1} is not uniformly

669: obeyed. In addition, one can write the condition \ref{syn:cond:2} as

670: \begin{equation}

671: \dot E = (1-3\beta x_{\text{sm}}^2-c)x^{*}y^{*} -\alpha y^{*2} \le 0,

672: \end{equation}

673: with $x^{*} = x_m - x_s$ and $y^{*} = y_m - y_s$. On actual numerical

674: simulation, it has been found that the time series of the above

675: expression oscillates chaotically about zero with an average of the

676: order of $\pm 10^{-3}$. This confirms that the condition

677: \ref{syn:cond:2} is also not satisfied uniformly. Thus, the only way to

678: \begin{figure}[!h]

679: \includegraphics[width=\linewidth]{duf_tser.eps}

680: 

681: \caption{Figures showing the variation of time of synchronization $t_s$

682: versus $c$ in the coupled Duffing oscillators.}

683: 

684: \label{ts_duf}

685: \end{figure}

686: understand the nature of synchronization in \Eqns{duf:syn1} is to

687: analyse the CLEs. We have again calculated the largest CLE as a

688: function of the coupling strength. \Fig{duf_cfig} shows a plot of the

689: largest CLE versus coupling strength ($c$). We find that the CLE takes

690: small negative values close to zero ($\approx 10^{-3}$) for

691: $c\in(1.48,1.64)$, where synchronization is essentially intermittent

692: and numerical artifact does arise in this range. In addition, there

693: exists another range of $c$ values, $2.74\le c\le 3.84$, where the CLEs

694: are again negative (cf.  \Fig{duf_cfig}(a)).  We find that for

695: $c=3.06$, the CLE takes the lowest value of $-0.05967$ (large negative

696: value) for which persistent synchronization occurs and the intermittent

697: synchronization is absent (cf. \Fig{duf_cfig}(b)).  The variation of

698: synchronization error for a range of selected $c$ values is depicted in

699: \Figs{duf_eta}(a)-(d). It can be noted from the figures that for

700: $c=2.74$ and $3.84$ (\Fig{duf_eta}(a) and (d)) the system can exhibit

701: intermittent synchronization while for $c=2.92$ and $3.36$

702: (\Fig{duf_eta}(b) and (c)) the system exhibits permanent

703: synchronization. Here also tiny noise levels in all the variables were

704: included at each integration step to ensure the absence of round off

705: induced synchronization.

706: 

707: As in the case of coupled pendula, we have calculated the time of

708: synchronization ($t_s$) by varying the coupling strength. \Fig{ts_duf}

709: shows the variation of $t_s$ as a function of the coupling strength for

710: this case. From the figure, it can be noted again that $t_s$ takes

711: relatively low values for a region of $c$ values where the CLEs have

712: large negative values which corresponds to permanent synchronization.

713: Similarly, the values of $t_s$ becomes high in the regions where the

714: largest CLE has a value close to zero or positive which corresponds to

715: intermittent synchronization.

716: \begin{figure*}[!ht]

717: \includegraphics[width=0.8\linewidth]{mlc_lyex.eps}

718: 

719: \caption{Figure showing (a) the variation of largest CLE as a function

720: of the coupling coefficient $\varepsilon$ for the coupled MLC circuits

721: and (b) blow up of a small region in (a) indicating the fluctuations of

722: the largest CLE near zero.}

723: 

724: \label{mlc_lyex}

725: \end{figure*}

726: \begin{figure*}[!ht]

727: \includegraphics[width=\linewidth]{mlc_eta.eps}

728: 

729: \caption{Figure showing the variation of synchronization error $\eta$

730: versus time in the coupled MLC circuits for (a) $\varepsilon = 0.0377$,

731: (b) $\varepsilon = 0.0380$, (c) $\varepsilon = 0.0390$, (d)

732: $\varepsilon = 0.0400$, (e) $\varepsilon=0.0500$ and (f) a small region

733: in (e) with the inclusion of tiny noise levels at each integration

734: step.}

735: 

736: \label{mlc_eta}

737: \end{figure*}

738: 

739: \subsection{Coupled MLC circuits}

740: 

741: Finally we consider the case of the coupled Murali-Lakshmanan-Chua

742: (MLC) circuits, represented by the following set of

743: equations\cite{lakshman}

744: \begin{subequations}

745: \label{eqn:mlc}

746: \bear

747: \dot x_m & = & y_m-h(x_m), \nonumber \\

748: \dot y_m & = & -\beta x_m +\sigma y_m + F\sin z_m \nonumber \\

749: \dot z_m & = & \omega 

750: \eear

751: %\\

752: %\nonumber \\

753: \bear

754: \label{eqn:mlc1}

755: \dot x_s & = & y_s-h(x_s)+\varepsilon (x_m-x_s), \nonumber \\

756: \dot y_s & = & -\beta x_s +\sigma y_s + F\sin z_s \nonumber \\

757: \dot z_s & = & \omega

758: \label{eqn:mlc2}

759: \eear

760: \end{subequations}

761: where $\varepsilon$ is the coupling strength and $h(x) = bx - 0.5(a-b)

762: (|x+1| - |x-1|)$. The uncoupled system of the above MLC circuits has

763: been shown to exhibit chaos for the choice of parameter values

764: $\beta=1$, $\sigma=1.015$, $a=-1.02$ and $b=-0.55$. We are interested

765: to analyse the existence of intermittent synchronization in this case

766: also.

767: 

768: First let us check the validity of the conditions \ref{syn:cond:1} and 

769: \ref{syn:cond:2}. In this case, the eigenvalues of the Jacobian matrix

770: of the flow evaluated on the synchronization manifold can be written as

771: \begin{eqnarray}

772: \lambda_{1,2} & = & -\frac{1}{2}

773: \left[h'(x_{\text{sm}})+\beta-\varepsilon\right]\nonumber \\

774: & & \pm\frac{1}{2}\sqrt{\left[h'(x_{\text{sm}})+\beta-\varepsilon\right]^2 

775: - 4[(h'-\varepsilon)\beta+\sigma]},\nonumber\\

776: \lambda_3 & = & 0.

777: \end{eqnarray}

778: Here $h'(x_{\text{sm}})$ corresponds to the derivative of $h(x)$ with

779: respect to $x$ variable evaluated on the synchronization manifold. From

780: numerical analysis, we have found that the largest of the above

781: eigenvalues oscillates chaotically about zero with an average slightly

782: less than zero for a range of $\varepsilon\in (0.036, 0.04)$ and thus

783: the condition \ref{syn:cond:1} is not uniformly obeyed. On the other

784: hand, the condition \ref{syn:cond:2} for the coupled MLC circuits can

785: be written as

786: \begin{equation}

787: \dot E = \left [\varepsilon-h'(x_{\text{sm}})\right ]x^{*2}

788: -\beta y^{*2}+(1-\sigma)x^{*}y^{*} \le  0,

789: \end{equation}

790: with $x^{*} = x_m - x_s$ and $y^{*} = y_m - y_s$. Again by numerical

791: analysis, we find that the above inequality is not uniformly obeyed and

792: hence the condition \ref{syn:cond:2} is not satisfied. However, the

793: coupled system (\ref{eqn:mlc}) has been shown to exhibit perfect

794: synchronization\cite{lakshman}. In order to analyse this we have

795: evaluated the CLEs as in the case of chaotic pendula and Duffing

796: oscillators.

797: 

798: The variation of the CLEs as a function of the coupling strength

799: ($\varepsilon$) is shown in \Fig{mlc_lyex}. The largest CLE changes its

800: sign from positive to negative values at $\varepsilon = 0.0377$ and it

801: is slightly less than zero over the range $0.0377 < \varepsilon \lesssim

802: 0.04$. In this range the synchronization completely depends on the

803: choice of initial conditions and the computer precision used. This

804: implies that the observed synchronization is essentially due to the

805: finite computational precision and accordingly locking ($\eta=0$)

806: occurs after relatively long time, leading to an intermittent

807: synchronization in this range of coupling strength.

808: 

809: We have also calculated the synchronization error ($\eta$) for various

810: values of coupling strength as a function of time as in the case of

811: previous systems and the results are depicted in \Fig{mlc_eta}. From

812: \Figs{mlc_eta}(a)-(d), it is evident that one can have intermittent

813: synchronization for $\varepsilon=0.0377$, $0.0380$, $0.0390$ and

814: $0.04$. \Fig{mlc_eta}(e) and \Fig{mlc_eta}(f) indicates the presence of

815: permanent synchronization for $\varepsilon=0.05$ where the sharp and

816: quick fall off in the synchronization error towards zero occurs (see

817: \Fig{mlc_eta}(f)).

818: 

819: \begin{figure}[!ht]

820: \includegraphics[width=\linewidth]{mlc_tser.eps}

821: \caption{Figure showing the variation of time of synchronization 

822: $t_s$ versus coupling strength $\varepsilon$ for coupled MLC 

823: circuits.}

824: \label{mlc_ts}

825: \end{figure}

826: 

827: \Fig{mlc_ts} shows the variation of time of synchronization ($t_s$) as

828: a function of the coupling strength ($\varepsilon$). As in the case of

829: coupled pendula and Duffing oscillators, $t_s$ takes relatively low

830: values for the range of $\varepsilon$ values where the CLEs take large

831: negative values corresponding to permanent synchronization. However,

832: the values of $t_s$ becomes high in the regions where the largest CLE

833: has a value close to zero or positive which corresponds to intermittent

834: synchronization. Thus one can conclude that permanent synchronization

835: in coupled MLC circuits (\ref{eqn:mlc}) occur for $\varepsilon >

836: 0.04$.

837: 

838: \section{Summary and conclusions}

839: \label{sec4}

840: 

841: By considering typical examples of coupled oscillator models, we have

842: discussed the quality of synchronization using various analytical and

843: numerical tests. It has been noted that one can have an

844: \emph{intermittent} as well as \emph{permanent} synchronization in the

845: coupled oscillator systems depending on the choice of coupling

846: strength. We have pointed out that the Conditional Lyapunov Exponents

847: (CLEs) can be used to distinguish between the intermittent and

848: permanent synchronization when the other criteria for asymptotic

849: stability are not uniformly obeyed. Particularly, we find that

850: intermittent synchronization can occur when CLEs are very small

851: positive or negative values close to zero while persistent

852: synchronization occurs when CLEs take sufficiently large negative

853: values. This fact is further supported by the relative time taken for

854: the system to approach the synchronization manifold.

855: 

856: The present work is mainly concerned with the qualitative analysis of

857: chaos synchronization in unidirectionally coupled systems based on

858: conditional Lyapunov exponents. In order to understand the entire

859: dynamics one has to analyse the nature of the attractors that exist in

860: the coupled system and their bifurcations. It is obvious that we have

861: not made any quantitative analysis on \emph{how small} or \emph{large}

862: be the magnitude of the largest conditional Lyapunov exponent in

863: distinguishing the intermittent and permanent synchronizations. Work is

864: in progress along these directions.

865: 

866: \begin{acknowledgments}

867: 

868: This work was supported by Department of Science and Technology, 

869: Govt. of India and National Board for Higher Mathematics

870: (Department of Atomic Energy) in the form of research projects.

871: 

872: \end{acknowledgments}

873: 

874: \bibliography{pap2e}

875: 

876: \end{document}

877: