1: \documentclass[11pt,a4paper,twoside]{article}
2:
3: \topmargin 0mm \textwidth 150mm \textheight 230mm
4: \setlength{\oddsidemargin 6mm} \setlength{\evensidemargin 6mm}
5:
6: \usepackage{graphicx}
7: \usepackage{tabls}
8: \usepackage{amsmath}
9: \usepackage{amssymb}
10: \usepackage{bm}
11:
12: \newtheorem{theorem}{Theorem}[section]
13: \newtheorem{lemma}[theorem]{Lemma}
14: \newtheorem{definition}[theorem]{Definition}
15: \newtheorem{prop}[theorem]{Proposition}
16: \newtheorem{cond}[theorem]{Condition}
17: \newtheorem{corollary}[theorem]{Corollary}
18: \newtheorem{remark}[theorem]{Remark}
19: \newcounter{eqcounter}[section]
20:
21: \renewcommand\arraystretch{1.3}
22: \renewcommand\baselinestretch{1.6}
23:
24: \def\proof{{\it proof\hspace{0.5cm}}}
25:
26:
27: \begin{document}
28:
29: \title{\huge\bf Failure of Parameter Identification Based on Adaptive Synchronization Techniques}
30: \author{Wei LIN$^\ddagger$ ~~ and ~~ Huan-Fei MA}
31: \date{\it Key Laboratory of Mathematics for Nonlinear Sciences (Fudan
32: University), Ministry of Education, Research Center for Nonlinear
33: Sciences, School of Mathematical Sciences, Fudan University,
34: Shanghai 200433, China.}\maketitle
35:
36:
37: $^\ddagger$ To whom correspondence should be addressed:
38:
39: Dr. Wei LIN,
40:
41: School of Mathematical Sciences
42:
43: Fudan University
44:
45: 220 HanDan Rd.
46:
47: Shanghai 200433
48:
49: CHINA
50:
51: Phone: 86-21-5566-4901
52:
53: Fax: 86-21-6564-2342
54:
55: E-mail: \underline{\it wlin@fudan.edu.cn}
56:
57: \newpage
58:
59: {\bf \large Abstract}
60:
61: In the paper, several concrete examples, as well as their numerical
62: simulations, are given to show that parameter identification based
63: on the so-called adaptive synchronization techniques might be failed
64: if those functions with parameters pending for identification in
65: coupled systems are designed to be mutually linearly dependent or
66: approximately linearly dependent on the orbit in the synchronization
67: manifold. This failure might be emergent not only when the
68: synchronized orbit is selected to be some sort of equilibrium or
69: some sort of periodic oscillation, but also when it is taken as some
70: type of chaotic attractor produced by driving system. This result
71: implies that chaotic property of driving signal is not necessary to
72: achievement of parameter identification. The mechanism inducing
73: such a failure, as well as the bounded property of all trajectories
74: generated by coupled systems, is theoretically expatiated. New
75: synchronization techniques are proposed to rigorously realize the
76: complete synchronization and parameter identification in a class of
77: systems where the nonlinearity is not globally Lipschitz. In
78: addition, parameter identification are discussed for systems with
79: time delay.
80:
81: ~\\
82:
83: PACS: 05.45.Gg, 05.40.-a, 87.10.+e.
84:
85:
86: \newpage
87:
88: \section{Introduction}
89:
90: The most classical phenomenon in reference to synchronization is
91: owing to Huygens' observation about the synchrony of pendulum clocks
92: \cite{Huygens}. Since this historical discovery, synchronization
93: as an omnipresent technical issue has become a focal topic of great
94: importance in many applications. Moreover, the basic concept
95: related to chaos synchronization in coupled chaotic systems was
96: initially introduced by Pecora and Carrol in 1990 \cite{Pec}. Since
97: their seminal paper, chaos synchronization as an interesting
98: research topic of great potential application has been widely
99: investigated and consequently applied in plenty of fields, ranging
100: from secure communications to pattern recognitions, from complex
101: network dynamics to optimization of nonlinear systems, and even from
102: chemical reaction to brain activity analysis \cite{Syn-A}. In
103: particular, a wide varieties of synchronization approaches,
104: including traditional linear or nonlinear feedback coupling, impulse
105: coupling, invariant manifold method, adaptive design coupling
106: techniques, and white-noise-based coupling have been fruitfully
107: proposed \cite{Syn-B}-\cite{noise} and several types of
108: synchronization, including complete synchronization, generalized
109: synchronization, phase synchronization, and lag synchronization,
110: have been introduced in succession
111: \cite{complete}-\cite{generalized}.
112:
113: Among all the proposed coupling approaches for realization of
114: complete synchronization between coupled chaotic systems with or
115: without time delays, the newly developed adaptive design coupling
116: technique has aroused a great amount of attention from many
117: researchers \cite{Amri}-\cite{Zhou} simply due to the reported
118: success in unknown parameter identification. Their explorations
119: have shown that unknown parameters could be identified in several
120: well-known chaotic systems and even in some neural network models
121: with or without time delays. In particular, consider an
122: $n$-dimensional nonlinear system described by
123: \begin{equation}\label{x system}
124: \dot{\bm{x}}=\bm{F}(\bm{x},\bm{p}),
125: \end{equation}
126: where $\bm{x}=(x_1,x_2,\ldots,x_n)^T\in\mathbb{R}^n$,
127: $\bm{F}(x,\bm{p})=\bigg(F_1(\bm{x},\bm{p}),F_2(\bm{x},\bm{p}),\dots,F_n(\bm{x},\bm{p})\bigg)^T$,
128: and
129: \begin{equation}
130: F_i(\bm{x},\bm{p})=c_i(\bm{x})+\sum^{m}_{j=1}p_{ij}f_{ij}(\bm{x}), ~~~
131: i=1,2,\dots,n.
132: \end{equation}
133: Here, $c_i(\bm{x})$ and $f_{ij}(\bm{x})$ are, respectively, assumed
134: to be some kind of real valued functions, and
135: $\bm{p}=\left\{p_{ij}\right\}\in \mathcal{U}\subset \mathbb{R}^n$
136: are $(n\cdot m)$ parameters pending for identification, in which
137: $\mathcal{U}$ is some bounded set. Given the bounded driving signal
138: $\bm{x}(t)$ produced by system (1), its response system is designed
139: through
140: \begin{equation}\label{y system}
141: \begin{array}{l}
142: \dot{\bm{y}}=\bm{F}(\bm{y},\bm{q})+\bm{\epsilon}\cdot\bm{e},\\
143: \dot{\epsilon_i}=-r_i e_i^2, \quad \dot{q}_{ij}=-\delta_{ij}e_i f_{ij}(\bm{y}),\\
144: i=1,2,\dots,n, \ j=1,2,\dots,m,
145: \end{array}
146: \end{equation}
147: where the feedback coupling $\bm{\epsilon}\cdot\bm{e}$ is in the
148: form of $(\epsilon_1 e_1,\epsilon_2e_2,\dots,\epsilon_ne_n)^T$,
149: $e_i=(y_i-x_i)$, $\bm{q}=\left\{q_{ij}\right\}$, and both $r_i$ and
150: $\delta_{ij}$ are arbitrarily chosen positive constants.
151:
152: A question naturally arises: ``Is it possible to accurately identify
153: all the $(n\cdot m)$ parameters of the chaotic system provided that
154: the output time series of system (\ref{x system}) are experimentally
155: obtained?" When the response system is designed as (\ref{y
156: system}), the answer to this question, as mentioned above, is
157: reportedly positive. Concretely, the complete synchronization
158: between the driving system (\ref{x system}) and the response system
159: (\ref{y system}) could be always achieved; moreover, the varying
160: parameters $\bm{q}$ in (\ref{y system}), initiating from arbitrary
161: values, will be asymptotically convergent to the correct values of
162: the parameters $\bm{p}$ in (\ref{x system}) as time tends towards
163: positive infinity. Seemingly, their theoretical arguments are based
164: on a delicate design for Lyapunov function, on the well-known
165: Lyapunov Stability theorem, and even on the LaSalle invariance
166: principle.
167:
168: As a matter of fact, those signals produced by these
169: driving-and-response systems could be completely synchronized;
170: nevertheless, the parameter identification might be failed if those
171: terms with parameters pending for identification are designed to be
172: mutually linearly dependent or approximately linearly dependent on
173: the synchronized orbit. In the paper, not only concrete examples
174: with their numerical simulations will be provided to illustrate such
175: a failure of parameter identification, but also the mechanism
176: inducing this failure will be anatomized. Also, the performed
177: analysis will show that {\it chaotic property of synchronized orbit
178: in synchronization manifold (or say, chaotic property of positive
179: limit set of driving signal) is not always necessary to achievement
180: of parameter identification}.
181:
182: The rest of the paper is organized as follows. In Section 2, three
183: concrete examples, as well as their numerical simulations, are
184: consecutively given to illustrate the possible occurrence of
185: parameter identification failure. The synchronized orbits in those
186: examples are, respectively, selected to be some sort of equilibrium,
187: periodic oscillation, and chaotic attractor. The mechanism inducing
188: this failure, as well as the the bounded property of all
189: trajectories generated by the coupled systems (\ref{x system}) and
190: (\ref{y system}), is theoretically expatiated in Section 3.
191: Furthermore, in Section 4, new synchronization techniques are
192: further proposed to realize complete synchronization and parameter
193: identification in a class of polynomial systems where the
194: nonlinearity is not globally Lipschitz. In Section 5, parameter
195: identification are discussed for systems with time delay. Finally,
196: the paper is closed with some concluded remarks.
197:
198:
199: \section{Examples Showing Failure of Parameter Identification}
200:
201: In this section, three groups of driving-and-response systems are
202: concretely presented to illustrate the possible occurrence of failed
203: parameter identification.
204:
205: First, consider the Lorenz system:
206: \begin{eqnarray}\label{countereg1}
207: \dot{x}_1&=&p_1(x_2-x_1), \nonumber\\
208: \dot{x}_2&=&p_2x_1-x_1x_3-x_2,\\
209: \dot{x}_3&=&x_1x_2-p_3x_3 \nonumber
210: \end{eqnarray}
211: as a driving system. And the corresponding response system becomes:
212: \begin{eqnarray}\label{rep1}
213: \dot{y}_1&=&q_1(y_2-y_1)+\epsilon_1(y_1-x_1),\nonumber\\
214: \dot{y}_2&=&q_2y_1-y_1y_3-y_2+\epsilon_2(y_2-x_2),\\
215: \dot{y}_3&=&y_1y_2-q_3y_3+\epsilon_3(y_3-x_3),\nonumber
216: \end{eqnarray}
217: where the updating laws of $\bm{q}=(q_1,q_2,q_3)$ and
218: $\bm{\epsilon}=(\epsilon_1,\epsilon_2,\epsilon_3)$ are,
219: respectively, taken as: $\dot{q}_1=-\delta_1(y_1-x_1)(y_2-y_1)$,
220: $\dot{q}_2=-\delta_2(y_2-x_2)y_2$,
221: $\dot{q}_3=-\delta_3(y_3-x_3)(-y_3)$, and
222: $\dot{\epsilon}_1=-r_1(y_1-x_1)^2$,
223: $\dot{\epsilon}_2=-r_2(y_2-x_2)^2$,
224: $\dot{\epsilon}_3=-r_3(y_3-x_3)^2$.
225:
226: Particularly, when the parameters are taken as $p_1=35$,
227: $p_2=\frac{8}{3}$, and $p_3=28$, the complete synchronization
228: between systems (\ref{countereg1}) and (\ref{rep1}) could be
229: achieved with time evolution, which is numerically shown by
230: Fig.1(a). This group of parameters, which are different from the
231: classical parameters inducing strange attractor of the Lorenz
232: system, simply make the synchronized orbit generated by system
233: (\ref{countereg1}) become an asymptotically stable equilibrium $E$,
234: as is shown by Fig.1(b). If the reported analytical results are
235: completely correct, it could be expected that the varying parameters
236: $\bm{q}$ will be eventually convergent to the accurate values of the
237: parameters $\bm{p}=(p_1,p_2,p_3)$. However, this is not the case.
238: As depicted in Fig.2, although the numerical simulation is
239: consistent with the expectation for the parameters $q_{2,3}$, it is
240: beyond the expectation for the parameters $q_1$. Concretely, the
241: value of $q_1$ does not approach but always keeps a distant from the
242: accurate value of $p_1$ with time evolution, which means failed
243: parameter identification does occur for $q_1$. Intuitively, there
244: must exist some mechanism inducing such differences between $q_1$
245: and $q_{2,3}$ when adaptive synchronization techniques are taken
246: into account.
247:
248: Secondly, construct a driving system based on the Chen's system
249: through:
250: \begin{eqnarray}\label{countereg2}
251: \dot{x}_1 &=& 30(x_2-x_1)+\mathcal{Q}(x_1,x_2,p_1,p_2), \nonumber\\
252: \dot{x}_2 &=& (28-30)x_1-x_1x_3+28x_2, \\
253: \dot{x}_3 &=& x_1x_2-3x_3. \nonumber
254: \end{eqnarray}
255: where the additional term
256: $$
257: \begin{array}{lll}
258: \mathcal{Q}(x_1,x_2,p_1,p_2) &=& 0.1\times
259: \displaystyle \left\{p_1\frac{\big(x_1\cos(0.9026)+x_2\sin(0.9026)\big)^2}{23.44^2}\right. \\
260: & & \displaystyle -p_2\left.\left[
261: \frac{\big(-x_1\sin(0.9026)+x_2\cos(0.9026)\big)^2}{7.19^2}-1\right]\right\},
262: \end{array}
263: $$
264: and both $p_1$ and $p_2$ are parameters expected to be identified.
265: As a matter of fact, without the term $\mathcal{Q}$, system
266: (\ref{countereg2}) becomes the original Chen's system admitting an
267: attractive periodic orbit. As displayed in Fig.3, the projection of
268: this attractive periodic orbit into the $x_1$-$x_2$ plane is
269: approximately looked upon as an ellipse. Thus, when $p_1=1$ and
270: $p_2=-1$, the term $\mathcal{Q}$ actually is an approximate formula
271: of this projection in the $x_1$-$x_2$ plane.
272:
273: Given a driving signal $(x_1,x_2,x_3)^T$ generated by system
274: (\ref{countereg2}), the corresponding response system is designed to
275: be in the form of
276: \begin{eqnarray}\label{rep2}
277: \dot{y}_1 &=& 30(y_2-y_1)+\mathcal{Q}(y_1,y_2,q_1,q_2)+\epsilon_1(y_1-x_1),\nonumber\\
278: \dot{y}_2 &=& (28-30)y_1-y_1y_3+28y_2+\epsilon_2(y_2-x_2), \\
279: \dot{y}_3 &=& y_1y_2-3y_3+\epsilon_3(y_3-x_3). \nonumber
280: \end{eqnarray}
281: in which, according to (\ref{y system}), the updating laws of the
282: two varying parameters are taken as:
283: $$
284: \begin{array}{lll}
285: \dot{q}_1 &=& \displaystyle -\delta_1(y_1-x_1)\left[\frac{
286: \big(y_1\cos(0.9026)+y_2\sin(0.9026)\big)^2}{23.44^2}\right], \\
287: \dot{q}_2 &=& \displaystyle
288: -\delta_2(y_1-x_1)\left[-\frac{\big(-y_1\sin(0.9026)+y_2\cos(0.9026)\big)^2}{7.19^2}+1\right],
289: \end{array}
290: $$
291: and the adaptive techniques of coupling strengths are, respectively,
292: chosen as: $\dot{\epsilon}_1=-r_1(y_1-x_1)^2$, $\dot{\epsilon}_2=
293: -r_2(y_2-x_2)^2$, and $\dot{\epsilon}_3=-r_3(y_3-x_3)^2$.
294:
295: Contrary to the expectation, both $q_1$ and $q_2$, starting from
296: almost every initial values, fail to approach the real values of the
297: parameters $p_1=1$ and $p_2=-1$, as shown in Fig.4. This example,
298: as well as the first example, shows that failure does occur for
299: parameter identification based on the adaptive synchronization
300: technique when the synchronized orbit is particularly selected to be
301: some type of steady dynamics, such as asymptotically stable
302: equilibrium and attractive periodic orbit.
303:
304: Instead of the above-mentioned steady synchronized orbit, the
305: existing numerical results \cite{Amri}-\cite{Zhou} show that
306: parameter identification could be always achieved when those
307: synchronized orbits in the synchronization manifold are designed to
308: be some type of chaotic attractor in advance. Then, a question
309: arises: ``Is chaotic property of synchronized orbit in
310: synchronization manifold necessary to achievement of parameter
311: identification based on the adaptive techniques?"
312:
313: To find out an answer to this question, consider a 4-dimensional
314: model developed from the original chaotic Lorenz system as a driving
315: system:
316: \begin{equation}\label{countereg3}
317: \begin{array}{lll}
318: \dot{x}_1 &=& p_1(x_2-x_1),\\
319: \dot{x}_2 &=& p_2x_1-x_1x_3-x_2,\\
320: \dot{x}_3 &=& x_1x_2-p_3x_3+p_4x_3(1+x^3_4),\\
321: \dot{x}_4 &=& ax_4+b(x_1-x_3),
322: \end{array}
323: \end{equation}
324: where $p_1=10$, $p_2=28$, and $p_3=\frac{8}{3}$ are three classical
325: parameters for the original Lorenz system to generate chaotic
326: attractor, and $a=-100$, $b=0.1$, $p_4=1$. Given these parameters,
327: the orbit produced by system (\ref{countereg3}) in the
328: synchronization manifold still exhibits chaotic property in the
329: phase plane, which is displayed by Fig.5(a)-(b). This chaotic
330: property is further verified by calculating the largest Lyapunov
331: exponent of the system ($\lambda_1\approx0.54274>0$), as is shown by
332: Fig.5(c).
333:
334: Provided with the driving signal produced by system
335: (\ref{countereg3}), the complete synchronization between systems
336: (\ref{countereg3}) and its response system could be numerically
337: achieved as long as the response system is designed through:
338: \begin{equation}\label{rep3}
339: \begin{array}{lll}
340: \dot{y}_1 &=& q_1(y_2-y_1)+\epsilon_1(y_1-x_1), \\
341: \dot{y}_2 &=& q_2y_1-y_1y_3-y_2+\epsilon_2(y_2-x_2), \\
342: \dot{y}_3 &=& y_1y_2-q_3y_3+q_4y_3(1+y_4^3)+\epsilon_3(y_3-x_3),\\
343: \dot{y}_4 &=& ay_4+b(y_1-y_3)+\epsilon_4(y_4-x_4),
344: \end{array}
345: \end{equation}
346: in which the updating laws of the parameters are taken as
347: $\dot{q}_1=-\delta_1(y_1-x_1)(y_2-y_1)$,
348: $\dot{q}_2=-\delta_2(y_2-x_2)y_1$,
349: $\dot{q}_3=-\delta_3(y_3-x_3)(-y_3)$,
350: $\dot{q}_4=-\delta_4(y_3-x_3)\left[y_3(1+y_4^3)\right]$, and the
351: adaptive coupling strengths are taken as
352: $\dot{\epsilon}_1=-r_1(y_1-x_1)^2$,
353: $\dot{\epsilon}_2=-r_2(y_2-x_2)^2$,
354: $\dot{\epsilon}_3=-r_3(y_3-x_3)^2$,
355: $\dot{\epsilon}_4=-r_4(y_4-x_4)^2$. In spite of the success in
356: complete synchronization and in parameter identification for
357: $q_{1,2}$, it is impossible to utilize $q_{3,4}$, initiating from
358: almost every points, to identify the accurate values of the
359: parameters $p_{3,4}$ in system (\ref{countereg3}). All these are
360: shown in Fig.6. Clearly, this example implies that the answer to the
361: above-posed question is negative.
362:
363: {\bf Remark 2.1} ~ The fourth-order Runge-Kutta scheme is used to
364: solve all the ordinary differential equations in our numerical
365: simulations.
366:
367:
368:
369:
370: \section{The Mechanism Inducing the Failure}
371:
372: On the one hand, three concrete examples in the last section show
373: that some parameter identification might be failed no matter what
374: kind of dynamical phenomenon is displayed in the synchronization
375: manifold. On the other hand, many existing numerical results always
376: show successful parameter identification. In order to clarify the
377: mechanism inducing such a seeming paradox, we perform a more
378: delicate argument by adopting the LaSalle invariance principle
379: \cite{LaSalle} as follows. Similar to \cite{Huang}, set a Lyapunov
380: function candidate by
381: \begin{equation}\label{Vfunction}
382: V(\bm{e},\bm{\epsilon},\bm{q})=
383: \frac{1}{2}\sum^n_{i=1}e^2_i+\frac{1}{2}\sum^n_{i=1}\sum^m_{j=1}\frac{1}{\delta_{ij}}(q_{ij}-p_{ij})^2
384: +\frac{1}{2}\sum^n_{i=1}\frac{1}{r_i}(\epsilon_i+L)^2.
385: \end{equation}
386: Then, the derivative of the function
387: $V(\bm{e},\bm{\epsilon},\bm{q})$ along with the coupled systems
388: (\ref{x system}) and (\ref{y system}) could be estimated by
389: $$
390: \dot{V}(\bm{e},\bm{\epsilon},\bm{q})\leqslant
391: (nl-L)\sum^n_{i=1}e^2_i.
392: $$
393: Here, it should be pointed out that $l$ is not the locally
394: Lipschitiz constant of the function $F_i(\bm{x},\bm{p})$ but the
395: uniformly Lipschitiz constant since the bounded property of the
396: trajectory $\bm{y}(t)$ generated by the newly response system
397: (\ref{y system}) are not confirmed but pending for confirmation yet.
398:
399: Now, {\it we contend that $\bm{e}(t)$, $\bm{\epsilon}(t)$, and
400: $\bm{q}(t)$ are bounded for all $t\geqslant t_0$, where $t_0$ is the
401: initial time}. Indeed, one of the three variables is supposed to be
402: unbounded on $[t_0,+\infty)$, so that
403: $V(\bm{e}(t),\bm{\epsilon}(t),\bm{q}(t))$ is also unbounded on
404: $[t_0,+\infty)$ according to (\ref{Vfunction}). On the other hand,
405: $V(\bm{e}(t),\bm{\epsilon}(t),\bm{q}(t))\leqslant
406: V(\bm{e}(t_0),\bm{\epsilon}(t_0),\bm{q}(t_0))$ simply due to
407: $\dot{V}(\bm{e},\bm{\epsilon},\bm{q})\leqslant 0$ for sufficiently
408: large $L$. This contradiction thus implies the bounded property of
409: $\bm{e}(t)$, $\bm{\epsilon}(t)$, and $\bm{q}(t)$ for all $t\geqslant
410: t_0$.
411:
412: Therefore, in light of the LaSalle invariance principle, the
413: trajectory $\big(\bm{e}(t), \bm{\epsilon}(t), \bm{q}(t)\big)$
414: initiating from any location in the phase plane will eventually
415: approach the largest invariant set $\mathcal{M}$ contained in the
416: set
417: $$
418: \mathcal{E}=\big\{(\bm{e},\bm{\epsilon},\bm{q})~\big|~\dot{V}(\bm{e},\bm{\epsilon},\bm{q})=0
419: \big\}.
420: $$
421: Then, the main concern becomes how to make a clear description of
422: the invariant set $\mathcal{M}$ ever contained in the set
423: $\mathcal{E}$ with respect to systems (\ref{x system}) and (\ref{y
424: system}). To this end, a combination of systems (\ref{x system})
425: and (\ref{y system}) yields
426: \begin{equation}\label{difference}
427: \begin{array}{l}
428: \displaystyle \dot{e}_i=\dot{x}_i-\dot{y}_i
429: =c_i(\bm{x})-c_i(\bm{y})+\sum^m_{j=1}p_{ij}f_{ij}(\bm{x})-\sum^m_{j=1}q_{ij}f_{ij}(\bm{y})-\epsilon_i(y_i-x_i) \\
430: \displaystyle =c_i(\bm{x})-c_i(\bm{y})+\sum^m_{j=1}\big[p_{ij}f_{ij}(\bm{x})-q_{ij}f_{ij}(\bm{x})\big]
431: +\sum^m_{j=1} \big[q_{ij}f_{ij}(\bm{x})- q_{ij}f_{ij}(\bm{y})\big]-\epsilon_i(y_i-x_i).
432: \end{array}
433: \end{equation}
434: Also, notice that $\dot{V}(\bm{e},\bm{\epsilon},\bm{q})=0$ implies
435: $\bm{e}=\bm{x}-\bm{y}=\bm{0}$, $\dot{\epsilon}_i(t)\equiv 0$, and
436: $\dot{q}_{ij}(t)\equiv 0$. It thus follows from (\ref{difference})
437: that, for every orbit
438: $\big(\bm{e}(t),\bm{\epsilon}(t),\bm{q}(t)\big)\in\mathcal{E}$ of
439: the coupled systems,
440: \begin{equation}\label{sum}
441: \sum^m_{j=1}\big[p_{ij}-q_{ij}(t)\big]f_{ij}(\bm{x}(t))=0,
442: \end{equation}
443: where each $q_{ij}(t)$ is identical to some constant $q_{ij}^*$. And
444: the largest invariant set contained in $\mathcal{E}$ with respect to
445: systems (\ref{x system}) and (\ref{y system}) is
446: $$
447: \mathcal{M}=\big\{(\bm{e},\bm{\epsilon},\bm{q})~\big|~\bm{e}=\bm{x}-\bm{y}=\bm{0},
448: ~\epsilon_i=\epsilon_i^*, ~q_{ij}=q_{ij}^* \big\},
449: $$
450: where $\bm{x}=\bm{x}(t)$ is the synchronized orbit, or
451: mathematically say, the positive limit set of the driving signal.
452: Thus, the question becomes: ``Is each $q_{ij}^*$ surely equal to
453: $p_{ij}$?" From (\ref{sum}), the answer to this question is
454: theoretically positive provided that [LIM]: {\it for any given $i$,
455: $\big\{f_{ij}(\bm{x}),j=1,2,\dots,m \big\}$ are linearly independent
456: on the synchronized orbit $\bm{x}=\bm{x}(t)$ in the synchronization
457: manifold}.
458:
459: For an accurate definition of linearly independent or linearly
460: dependent functions, refer to \cite{def}. Also, it is valuable to
461: mention that two functions might be linearly independent in a domain
462: but linearly dependent in some subset contained in this domain. For
463: example, functions $g_1(s,u)=s$ and $g_2(s,u)=u^2$ are obviously
464: linearly independent in $\mathbb{R}^2$ but they are linearly
465: dependent in a parabola-like subset $\mathcal{S}_\mu=\big\{
466: (s,u)\in\mathbb{R}^2~|~s=\mu u^2 \big\}\subset\mathbb{R}^2$ for some
467: nonzero constant $\mu$.
468:
469: If hypothesis [LIM] is not satisfied, for some $i=i_0$, either there
470: exist two nonzero functions $f_{i_0j_1}(\bm{x})$ and
471: $f_{i_0j_2}(\bm{x})$ linearly dependent on the orbit $\bm{x}(t)$ in
472: the synchronization manifold, or simply $f_{i_0j_1}(\bm{x}(t))\equiv
473: 0$. We focus on the former case since failure of parameter
474: identification could be easily illustrated in the latter case.
475: Accordingly, $f_{i_0j_1}(\bm{x}(t))=cf_{i_0j_2}(\bm{x}(t))$ for some
476: nonzero constant $c$, which at most implies that
477: $\big[p_{i_0j_1}-q_{i_0j_1}(t)\big]+c\big[p_{i_0j_2}-q_{i_0j_2}(t)\big]=0$.
478: Clearly, although $q_{i_0j_1}(t)$ and $q_{i_0j_2}(t)$ are,
479: respectively, identical to some constants $q_{i_0j_1}^*$ and
480: $q_{i_0j_2}^*$, it is not certain that $q_{i_0j_1}^*=p_{i_0j_1}$ and
481: $q_{i_0j_2}^*=p_{i_0j_2}$. Actually, they are totally distinct in
482: most cases. Therefore, parameter identification might always be
483: failed if hypothesis [LIM] is not strictly satisfied in the design
484: of driving-and-response systems.
485:
486: Next, by virtue of the argument performed above, the reason why
487: parameter identification fails in the three examples given in the
488: previous section is expatiated as follows.
489:
490: For the driving-and-response systems (\ref{countereg1}) and
491: (\ref{rep1}) with specific parameters, the synchronized orbit
492: $\bm{x}^*(t)=(x_1^*(t),x_2^*(t),x_3^*(t))^T$ in the synchronization
493: manifold, as shown in Fig.1, is a globally asymptotical equilibrium
494: $E=\bm{x}^*(t)=(6.8313,6.8313,1.6667)^T$. Substitution of
495: (\ref{countereg1}) into (\ref{sum}) gives
496: $$
497: [p_1-q_1(t)][x_2^*(t)-x_1^*(t)]=0, ~
498: [p_2-q_2(t)]x_1^*(t)=0,~
499: [p_3-q_3(t)]x_3^*(t)=0,
500: $$
501: where each $q_i(t)$ is identical to some constant $q_i^*$ in the
502: invariant set $\mathcal{M}$ ($i=1,2,3$). According to \cite{def},
503: each $x_i^*(t) ~(\not=0)$ is linearly independent and
504: $x_2^*(t)-x_1^*(t)~(\equiv 0)$ is linearly dependent. This implies
505: that $q^*_{2,3}$ is identical to $p_{2,3}$ but $q^*_1$ is not
506: necessarily identical to $p_1$. Therefore, $q_1(t)$, though
507: obeying the updating law, will not be surely convergent to $p_1$.
508: This illustrates the reason why parameter identification succeeds
509: for $q_{2,3}$ but always fails for $q_1$ as shown in Fig.2. However,
510: when the synchronized orbit $\bm{x}^*(t)$ with the classical
511: parameters is chaotic, $x_2^*(t)-x_1^*(t)$ is nonzero and thus is
512: linearly independent, which satisfies hypothesis [LIM]. Hence,
513: $q_1(t)$ will be convergent to $p_1$ almost surely, as is shown by
514: many existing numerical results. In addition, when the synchronized
515: orbit $\bm{x}^*(t)$ unfortunately becomes the unstable equilibrium
516: of the original chaotic systems, $x_2^*(t)-x_1^*(t)$ is still
517: identical to zero violating hypothesis [LIM]. So, $q_1(t)$ still
518: will not be surely convergent to the accurate value of $p_1$ in such
519: a case.
520:
521: For the coupled systems (\ref{countereg2}) and (\ref{rep2}), the
522: orbit $\bm{x}^*(t)$ in the synchronization manifold, as mentioned
523: above, is designed to be some kind of stable periodic orbit. Its
524: projection into the $x_1$-$x_2$ plane, which seems like an ellipse,
525: could be approximately expressed by the formula
526: $$
527: \frac{\big(x_1\cos(0.9026)+x_2\sin(0.9026)\big)^2}{23.44^2}+
528: \frac{\big(-x_1\sin(0.9026)+x_2\cos(0.9026)\big)^2}{7.19^2}-1 = 0.
529: $$
530: Also, substitution of (\ref{countereg2}) into (\ref{sum}) yields
531: $$
532: \begin{array}{l}
533: \displaystyle [p_1-q_1(t)]\frac{\big(x_1(t)\cos(0.9026)+x_2(t)\sin(0.9026)\big)^2}{23.44^2}+\\
534: \displaystyle
535: [-p_2+q_2(t)]\left\{\frac{\big(-x_1(t)\sin(0.9026)+x_2(t)\cos(0.9026)\big)^2}{7.19^2}-1\right\}=0.
536: \end{array}
537: $$
538: Thus, as long as the complete synchronization between systems
539: (\ref{countereg2}) and (\ref{rep2}) is achieved, the orbit
540: $\bm{x}(t)$, as well as $\bm{y}(t)$, will approximately approach the
541: stable periodic orbit $\bm{x}^*(t)$. Then, both functions
542: $$
543: \frac{\big(x_1^*(t)\cos(0.9026)+x_2^*(t)\sin(0.9026)\big)^2}{23.44^2}
544: ~\mbox{and}~
545: \frac{\big(-x_1^*(t)\sin(0.9026)+x_2^*(t)\cos(0.9026)\big)^2}{7.19^2}-1
546: $$
547: are approximately linearly dependent. This, according to the
548: argument performed above, means that both $q_1$ and $q_2$ are not
549: suitable for parameter identification, as is verified by the
550: simulation results in Fig.4.
551:
552: Unlike the steady dynamics exhibiting in synchronization manifold in
553: the previous two examples, the synchronized orbit $\bm{x}^*(t)$
554: generated by the driving system (\ref{countereg3}) is deliberately
555: designed to be chaotic in the sense of possessing positive Lyapunov
556: exponent. Analogously, substitution of (\ref{countereg3}) into
557: (\ref{sum}) produces
558: $$
559: [p_3-q_3(t)]x_3(t)+[p_4-q_4]x_3(t)\left\{1+[x_4(t)]^3\right\}=0.
560: $$
561: It is obvious that functions $x_3$ and $x_3(1+x_4^3)$ are linearly
562: independent in the whole phase plane $\mathbb{R}^4$; nevertheless,
563: they are approximately linearly dependent on the synchronized orbit
564: $\bm{x}^*(t)$ because the cubic term $[x_4^*(t)]^3$ is almost equal
565: to zero as time $t$ is sufficiently large (see Fig.7). Thus, this
566: illustrates the reason why $q_3$ and $q_4$ initiating from a mass of
567: points will not be convergent to the accurate values of $p_3$ and
568: $p_4$, respectively, in concrete numerical simulations.
569:
570:
571: In addition, consider a case that parameter $b$ in both systems
572: (\ref{countereg3}) and (\ref{rep3}) is selected to be zero instead
573: of $0.1$. This case could be regarded as a very special example
574: where parameter identification may fail in spite of existence of
575: chaos. In such an example, because of $x_4^*(t)\equiv 0$, functions
576: $x_3$ and $x_3(1+x_4^3)$ are definitely linearly dependent on the
577: corresponding orbit $\bm{x}^*(t)$, which violates hypothesis [LIM].
578: Therefore, $q_3$ and $q_4$ can not be utilized to identify the
579: parameters $p_3$ and $p_4$. In a word, chaotic property of
580: synchronized orbit in synchronization manifold does not always
581: guarantee a success in parameter identification.
582:
583:
584:
585: \begin{remark}
586: {\rm In the last two examples, those functions on the orbits in the
587: synchronization manifold are approximately linearly dependent.
588: Rigorously, they are still linearly independent in a mathematical
589: sense, so that parameter identification might be theoretically
590: achieved for $q_i$ correspondingly with $p_i$. However, in real
591: application, discretization techniques, such as the Runge-Kutta
592: scheme and the Euler scheme, are always taken into account in
593: solving the coupled continuous differential systems. Thus, owing to
594: the precision limit, it is unavoidable that dynamics produced by the
595: discretized system may not be completely consistent with the true
596: dynamics generated by the original system. It is the approximate
597: dependence of those functions that poses some trap of local critical
598: point for $q_i$ and that leads to a failure of parameter
599: identification in the last two examples. Therefore, {\it not only
600: rigorous linear-dependence of those functions with parameters
601: pending for identification on the synchronized orbit but also
602: approximate linear-dependence on the synchronized orbit should be
603: always avoided whenever adaptive synchronization techniques are used
604: in practical parameter estimation and chaos communication.}}
605: \end{remark}
606:
607:
608: \section{Complete Synchronization without Globally Lipschitz Condition}
609:
610: In the previous section, it is pointed out that hypothesis [LIM] is
611: indispensable for a successful parameter identification. In
612: addition, the uniform Lipschitz condition for
613: $\bm{F}(\bm{x},\bm{p})$ is also important in the argument performed
614: above for obtaining a non-positive property of
615: $\dot{V}(\bm{e},\bm{\epsilon},\bm{q})$. As a matter of fact, this
616: uniform condition could be loosed if the bounded property of the
617: response system (\ref{y system}) could be priorly estimated.
618: However, this prior estimation could not directly follow from the
619: bounded property of driving system (\ref{x system}) since dynamics
620: of its response system with coupling term might be completely
621: different from the driving system which is, though, supposed to be
622: bounded in advance. Then it poses a question: ``Other than the
623: above coupling technique and uniform Lipschitz condition, under what
624: kind of coupling methods and conditions on $\bm{F}(\bm{x},\bm{p})$
625: can one obtain a successful parameter identification rigorously?"
626:
627: To this end, it is first assumed [HPT]: {\it each
628: $F_i(\bm{x},\bm{p})$ is homogeneous polynomial with degree no more
629: than two with respect to $\bm{x}$.} As a matter of fact, a large
630: quantities of nonlinear systems does not satisfy globally Lipschitz
631: condition but are consistent with this assumption [HPT], such as the
632: Lorenz system and the Chen's system.
633:
634: Next, notice that
635: $$
636: \begin{array}{lll}
637: 2y_ky_j- 2x_kx_j &=&
638: (y_k-x_k)(y_j+x_j)+(y_j-x_j)(y_k+x_k) \\
639: &=& 2 e_k e_j + 2x_j e_k + 2x_k e_j
640: \end{array}
641: $$
642: for arbitrary $k$ and $j$. Then, it is easy to verify that each
643: $e_i\big[F_i(\bm{y},\bm{p})-F_i(\bm{x},\bm{p})\big]$ can be written
644: as a homogeneous polynomial with degree no more than three with
645: respect to $\bm{e}=\bm{y}-\bm{x}$ if assumption [HPT] holds.
646:
647: Reasonably, the driving signal $\bm{x}(t)$ generated by system
648: (\ref{x system}) is supposed to be bounded in advance. In order to
649: obtain a rigorous synchronization in the system where the
650: nonlinearity only satisfies assumption [HPT], we re-designed the
651: response system as:
652: \begin{equation}\label{y new}
653: \begin{array}{l}
654: \dot{\bm{y}}=\bm{F}(\bm{y},\bm{q})+\bm{\epsilon}\cdot\bm{e}+\bm{\omega}\cdot\bm{e}^3,\\
655: \dot{\epsilon_i}=-r_i e_i^2 ,~~ \dot{\omega_i}=-s_i e_i^4, \\
656: \dot{q}_{ij}=-\delta_{ij}e_i f_{ij}(\bm{y}),\\
657: \end{array}
658: \end{equation}
659: where
660: $\bm{\omega}\cdot\bm{e}^3=\big(\omega_1e_1^3,\omega_2e_2^3,\cdots,\omega_n
661: e_n^3\big)^T$, each $s_i$ is arbitrarily positive constant, and
662: other states and parameters are the same as those defined in (\ref{y
663: system}).
664:
665: Set a Lyapunov function candidate by
666: $$
667: H(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})
668: = \frac{1}{2}\sum^n_{i=1} e^2_i+\frac{1}{2}\sum^n_{i=1}\sum^m_{j=1}\frac{1}{\delta_{ij}}(q_{ij}-p_{ij})^2
669: +\frac{1}{2}\sum^n_{i=1}\frac{1}{r_i}(\epsilon_i+M)^2+
670: \frac{1}{2}\sum^n_{i=1}\frac{1}{s_i}(\omega_i+N)^2.
671: $$
672: Thus, the derivative of this function along with the coupled systems
673: (\ref{x system}) and (\ref{y new}) yields
674: $$
675: \dot{H}(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})(t)
676: =\sum^n_{i=1}e_i(t)\big[F_i(\bm{y}(t),\bm{p})-F_i(\bm{x}(t),\bm{p})\big]-\sum^n_{i=1} M
677: e_i^2(t)
678: -\sum^n_{i=1} N e_i^4(t),
679: $$
680: where both $M$ and $N$ are positive numbers. By virtue of the
681: conclusion on each
682: $e_i\big[F_i(\bm{y},\bm{p})-F_i(\bm{x},\bm{p})\big]$ obtained above,
683: the elementary inequality
684: $$
685: e_i e_j e_ k\leqslant \frac{1}{6}\sum_{l=i,j,k}\big(e_l^2+e_l^4\big),
686: $$
687: and the assumed bounded property of the driving signal $\bm{x}(t)$
688: and parameter set $\mathcal{U}$, we can obtain that
689: $\dot{H}(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})(t)\leqslant 0$ for
690: sufficiently large numbers $M$ and $N$.
691:
692: By using the similar argument performed in the previous section, we
693: can easily prove that every trajectory generated by the coupled
694: systems (\ref{x system}) and (\ref{y new}) is not only bounded for
695: all $t\geqslant t_0$ but also approaching the largest invariant set
696: contained in
697: $$
698: \mathcal{E}'=\big\{(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})~\big|~\dot{H}(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})=0
699: \big\}
700: $$
701: with respect to these coupled systems. More precisely, the largest
702: invariant set becomes
703: $$
704: \mathcal{M}'=\big\{(\bm{e},\bm{\epsilon},\bm{\omega},\bm{q})~\big|~
705: \bm{e}=\bm{0},~\epsilon_i=\epsilon_i^*, ~\omega_i=\omega_i^*, ~
706: q_{ij}=q_{ij}^*
707: \big\},
708: $$
709: where $\epsilon_i^*$, $\omega_i^*$, and $q_{ij}^*$ are some
710: constants dependent on the initial values of the coupled systems.
711: Furthermore, to achieve an accurate parameter identification between
712: systems (\ref{x system}) and (\ref{y new}), hypothesis [LIM] should
713: be still adopted. Then, the above performed argument could be
714: concluded as the following proposition.
715:
716: \begin{prop}
717: If assumptions [LIM] and [HPT] on $\bm{F}(\bm{x},\bm{p})$ are
718: satisfied, the complete synchronization between driving system
719: (\ref{x system}) and its response system (\ref{y new}) could be
720: surely achieved, and the parameter identification could be
721: accurately realized in a mathematical sense.
722: \end{prop}
723:
724: \begin{remark}
725: {\rm As mentioned above, in numerical experiment and even in real
726: application, not only hypothesis [LIM] should be strictly satisfied
727: but also the approximate linear-dependence attributed to precision
728: limit should be avoided.}
729: \end{remark}
730:
731: \begin{remark} {\rm Assumption [HPT] on
732: $\bm{F}(\bm{x},\bm{p})$ could be further generalized to some other
733: case where globally Lipschitz condition is not fulfilled. For
734: instance, one could further consider the system where the degree of
735: the homogeneous polynomials is larger than two, or some of those
736: polynomials are non-homogeneous. However, additional coupling terms
737: (e.g. $\bm{e}^{2v+1},~v=2,3,\cdots$) should be added into the
738: response systems in order to obtain a successful synchronization and
739: parameter identification in a rigorous sense.}
740: \end{remark}
741:
742: \begin{remark}
743: {\rm As a matter of fact, nonlinearities in the previous three
744: examples are not globally Lipschitz but polynomial. Theories and
745: coupling techniques (i.e. Proposition 4.1 and Remark 4.3) proposed
746: in this section should be utilized to deal with those systems for
747: obtaining a successful synchronization and parameter
748: identification.}
749: \end{remark}
750:
751:
752: \section{Parameter Identification in Systems with Time-Delay}
753:
754: Time-delay, as an omnipresent phenomenon, can not be neglected in
755: practice. So, in this section, complete synchronization and
756: parameter identification in time-delayed systems via adaptive
757: coupling techniques are further investigated. For simplicity,
758: consider a one-dimensional driving system:
759: \begin{equation}\label{delay-d}
760: \dot{x}(t)=a f\big(x(t)\big)+ b g\big(x(t-\tau)\big),
761: \end{equation}
762: where $\tau\geqslant 0$ is a time-delay, $a$ and $b$ are parameters
763: pending for identification, and functions $f$ and $g$ are assumed to
764: be globally Lipschitz continuous with Lipschitz constants $k_f$ and
765: $k_g$, respectively. Given driving signal $x(t)$ generated by
766: system (\ref{delay-d}), the response system is designed as
767: \begin{equation}\label{delay-r}
768: \begin{array}{l}
769: \dot{y}(t)=\alpha(t) f\big(y(t)\big)+ \beta(t) g\big(y(t-\tau)\big) + \eta(t) e(t)+ \omega(t) e(t-\delta), \\
770: \dot{\alpha}(t)= - f\big(y(t)\big) e(t),~~ \dot{\beta}(t)=-g\big(y(t-\tau)\big)e(t), \\
771: \dot{\eta}(t)= -e^2(t),~~\dot{\omega}(t)= -e(t)e(t-\delta),
772: \end{array}
773: \end{equation}
774: where $\delta\geqslant 0$ is a time-delay induced by coupling term,
775: error dynamics $e(t)=y(t)-x(t)$. The initial conditions for coupled
776: system (\ref{delay-d}) and (\ref{delay-r}) are chosen as $e=\phi$,
777: $\alpha=A$, $\beta=B$, $\eta=E$, $\omega=W\in
778: \mathcal{C}\triangleq\mathbb{C}([-\max\{\tau,\delta\},0],\mathbb{R})$,
779: in which $\mathcal{C}$ denotes the sets of all continuous functions
780: from $[-\max\{\tau,\delta\},0]$ to $\mathbb{R}$.
781:
782: Set a Lyapunov functional candidate by
783: $$
784: \begin{array}{lll}
785: \mathcal{V}(\phi,A,B,E,W) & = &
786: \displaystyle \frac{1}{2}\phi^2(0)+\frac{1}{2}[A(0)-a]^2+\frac{1}{2}[B(0)-b]^2+
787: \frac{1}{2}[E(0)+L]^2 \\
788: & &
789: \displaystyle +\frac{1}{2}[W(0)+M]^2 +
790: \bigg\{\int_{-\tau}^{0}+\int_{-\delta}^{0}\bigg\}\phi^2(s){\rm
791: d}s,
792: \end{array}
793: $$
794: where $L$, $M$ are some proper positive constant. Then, the
795: derivative of $\mathcal{V}$ along with coupled systems
796: (\ref{delay-d}) and (\ref{delay-r}) could be estimated by
797: $$
798: \begin{array}{lll}
799: \dot{\mathcal{V}}(\phi,A,B,E,W) &\leqslant&
800: \displaystyle \left(ak_f+2-\frac{L}{2}\right)\phi^2(0)+bk_g\cdot
801: |\phi(0)|\cdot|\phi(-\tau)|-\phi^2(-\tau) \\
802: & & \displaystyle-\frac{L}{2}\phi^2(0)-M
803: \phi(0)\phi(-\delta)-\phi^2(-\delta).
804: \end{array}
805: $$
806: Clearly, $\dot{\mathcal{V}}(\phi,A,B,E,W)$ becomes non-positive
807: provided $\displaystyle L>
808: \max\bigg\{\frac{M^2}{2},\frac{b^2k_g^2}{2}+2ak_f+4\bigg\}$. By
809: using a similar argument performed above, we can conclude that every
810: trajectory
811: $(e_t(\phi),\alpha_t(A),\beta_t(B),\eta_t(E),\omega_t(W))$, starting
812: from arbitrary initial condition, is surely bounded for all
813: $t\geqslant -\max\{\tau,\delta\}$.
814:
815: Then, according to the invariance principle for the systems with
816: time-delay \cite{Hale}, every trajectory, as time tends towards
817: positive infinity, approaches the largest invariant set
818: $\widetilde{\mathcal{M}}$ contained in
819: $$
820: \widetilde{\mathcal{E}}=\left\{(\phi,A,B,E,W)\in
821: \underbrace{\mathcal{C}\times\cdots\times\mathcal{C}}_{5}~\bigg|~\phi(0)=\phi(-\tau)=\phi(-\delta)=0
822: \right\}
823: $$
824: with respect to coupled systems (\ref{delay-d}) and (\ref{delay-r}).
825: This further implies that the first component of each element in
826: $\widetilde{\mathcal{M}}$ is identical to zero (i.e. $\phi\equiv0$)
827: and the others are some constant functions (i.e. $A\equiv A^*$,
828: $B\equiv B^*$, $E\equiv E^*$, and $W\equiv W^*$). The accurate
829: values of these constant functions rest on the initial conditions of
830: coupled systems (\ref{delay-d}) and (\ref{delay-r}).
831:
832: Parameter identification could be achieved if both equations $A^*=a$
833: and $B^*=b$ are valid. However, these equations are not always
834: tenable although $\phi \equiv 0$ indicates a successful complete
835: synchronization between systems (\ref{delay-d}) and (\ref{delay-r}).
836: As a matter of fact, subtraction of (\ref{delay-d}) from the first
837: equation in (\ref{delay-r}) yields, in $\widetilde{\mathcal{M}}$,
838: $$
839: \begin{array}{lll}
840: 0 &=& \alpha(t)f\big(y(t)\big)-af\big(x(t)\big)+\beta(t)g\big(y(t-\tau)\big)-bg\big(x(t-\tau)\big)
841: \\
842: &=& [\alpha(t)-a]f\big(x(t)\big)+[\beta(t)-b]g\big(x(t-\tau)\big) \\
843: &=& [A^*-a]f\big(x(t)\big)+[B^*-b]g\big(x(t-\tau)\big),
844: \end{array}
845: $$
846: which follows from $e(t)=y(t)-x(t)\equiv 0$ in
847: $\widetilde{\mathcal{M}}$. Now, it is clear that when functions
848: $f\big(x(t)\big)$ and $g\big(x(t-\tau)\big)$, on the synchronized
849: orbit $x(t)$ in the synchronization manifold, are linearly
850: dependent, both $A^*=a$ and $B^*=b$ are not certainly tenable. More
851: concretely, (i) when the driving signal asymptotically tends towards
852: some equilibrium $x(t)\equiv x^*$ of system (\ref{delay-d}), two
853: constant functions $f\big(x(t)\big)\equiv f(x^*)$ and
854: $g\big(x(t-\tau)\big)\equiv g(x^*)$ becomes linearly dependent so
855: that parameter identification for $a$ and $b$ will be almost surely
856: failed; (ii) when the synchronized orbit $x(t)$ is periodic with
857: period $\tau$ and both functions $f$ and $g$ are linearly dependent
858: in $\mathbb{R}$, parameter identification also will be failed; (iii)
859: when $x(t)$ is chaotic, parameter identification will be achieved in
860: a mathematical sense for non-constant differential functions $f$ and
861: $g$, and even for $f=g$ (see an example shown in Fig.8(a) where both
862: $f$ and $g$ are taken as sinusoid functions). However, for case
863: (iii), parameter identification also might be failed in numerical
864: simulation or in real application. For example, in spite of chaotic
865: property of $x(t)$, it is likely that $x(t)\approx x(t-\tau)$ with a
866: small time-delay or that fluctuation of $x(t)$ seems to be
867: relatively steady in a macro scale. These extraordinary cases may
868: lead to approximate linear-dependence between functions
869: $f\big(x(t)\big)$ and $g\big(x(t-\tau)\big)$, which thus results in
870: failure of parameter identification in numerical simulation. See an
871: illustrative example in Fig.8 (b). In addition, $\tau=0$ could be
872: regarded as a special case where parameter identification is always
873: failed provided that functions $f$ and $g$ are linearly dependent on
874: $x(t)$.
875:
876: In conclusion, we have the following proposition on synchronization
877: and parameter identification for coupled systems (\ref{delay-d}) and
878: (\ref{delay-r}) with time-delay.
879:
880: \begin{prop}
881: The complete synchronization between driving system (\ref{delay-d})
882: and its response system (\ref{delay-r}) could be surely achieved via
883: adaptive coupling techniques. Furthermore, the parameter
884: identification could be accurately realized in a mathematical sense
885: provided that $f(x(t))$ and $g(x(t-\tau))$ are linearly independent
886: on the synchronized orbit $x(t)$ in the synchronization manifold.
887: \end{prop}
888:
889: \begin{remark} {\rm With analogous
890: arguments but more complicated notations, the results on the driving
891: system (\ref{delay-d}) could be further generalized to the case
892: where higher dimensional driving systems and multiple parameters
893: identification are taken into account. However, linear-independence
894: of all the functions with unknown parameters on the driving signal
895: is crucial to a successful parameter identification.}
896: \end{remark}
897:
898:
899: \section{Conclusion}
900:
901: In summary, concrete examples showing possible occurrence of failed
902: parameter identification have been numerically given in the paper.
903: The mechanism inducing this failure has been further rigorously
904: interpreted. It has been pointed out that chaotic property of
905: driving system is not always crucial to achievement of parameter
906: identification either in a mathematical argument or in a numerical
907: experiment. Actually, it is not the chaos but the hypothesis [LIM]
908: that guarantees a successful parameter identification based on
909: adaptive synchronization techniques. However, making good use of
910: chaotic property might easily lead to validity of hypothesis [LIM].
911: Apart from linear-dependence of functions, approximate
912: linear-dependence of functions with parameters pending for
913: identification on the synchronized orbit should be always avoided in
914: numerical simulation and even in real application.
915:
916: Furthermore, complete synchronization via new adaptive coupling
917: techniques in a class of polynomial systems where nonlinearity is
918: not globally Lipschitz has been theoretically investigated by
919: virtue of the LaSalle invariance principle. Also it has been
920: rigorously verified that every trajectory generated by the coupled
921: systems is bounded. By these derived theoretical results, our newly
922: proposed technique is convinced to be a rigorous and feasible
923: approach for realization of complete synchronization and parameter
924: identification in the Lorenz-like systems. Besides, adaptive
925: coupling techniques are also imported to realize parameter
926: identification in systems with time-delay. Those discussion also
927: shows the great importance of the condition that functions with
928: parameters pending for identification on the synchronized orbit
929: should be linearly independent.
930:
931:
932: \section{Acknowledgement}
933:
934: The authors are grateful to the learned referee and Prof. Jiong Ruan
935: for their helpful comments and suggestions. This research was
936: supported by the National Natural Foundation of China Grant
937: No.10501008.
938:
939:
940:
941:
942: {\small
943: \begin{thebibliography}{99}
944:
945: \bibitem{Huygens}
946: C. Huygens, Philo. Trans. Royal Soc. {\bf 4}, 937 (1669).
947: \bibitem{Pec}
948: L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. {\bf 64}, 821
949: (1990).
950: \bibitem{Syn-A}
951: E. Ott, C. Grebogi, and J. A. York, Phys. Rev. Lett. {\bf 64}, 1196 (1990);
952: N. F. Rulkov, M. M. Sushchik, L. S. Tsimring and H. D. I. Abarbanel, Phys. Rev. E. {\bf 51}, 980 (1995);
953: E. Jr. Rosa, E. Ott, and M. H. Hess, Phys. Rev. Lett. {\bf 80}, 1642 (1998).
954: \bibitem{Syn-B}
955: G. Chen, X. Dong, {\it From Chaos to Order: Methodologies,
956: Persperctives and Applications} (World Scientific, Singapore, 1998);
957: {\it Handbook of Chaos Control,} edited by H. G. Schuster (Wiley-VCH, Weinheim,
958: 1999); S. Boccaletti {\it et al.}, Phys. Rep. {\bf 366}, 1 (2002);
959: G. Chen, X. Yu, {\it Chaos Control: Theory and Applications} (Springer, Berlin, 2003);
960: E. M. Bollt, Int. J. of Bifur. Chaos, {\bf 13}, 269 (2003).
961: \bibitem{noise}
962: C. S. Zhou and J. Kurths, Phys. Rev. Lett. {\bf 88}, 230602 (2002);
963: Chaos {\bf 13}, 401 (2003); W. Lin, Y. He, Chaos {\bf 15}, 023705 (2005);
964: W. Lin, G. Chen, Chaos {\bf 16}, 013134 (2006); Z. Chen, W. Lin,
965: J. Zhou, Chaos, in press (2007).
966: \bibitem{complete}
967: H. Fujisaka and T. Yamada, Prog. Theor. Phys. {\bf 69}, 32 (1983);
968: T. Yamada and H. Fujisaka, Prog. Theor. Phys. {\bf 70}, 1240 (1983).
969: \bibitem{lag}
970: C. Grebogi, E. Ott and J. A. Yorke, Phys. Rev. Lett. {\bf 50}
971: 935 (1983); M. Rosenblum, A. Pikovsky, and J. Kurths, Phys. Rev. Lett. {\bf
972: 78}, 4193 (1997).
973: \bibitem{phase}
974: A. Pikovsky, G. Osipov, M. Rosenblum, M. Zaks, and J. Kurths, Phys. Rev. Lett.
975: {\bf 79}, 47 (1997).
976: \bibitem{generalized}
977: N.F. Rulkov, M.M. Sushchik, L.S. Tsimring, and H.D.I. Abarbanel, Phys. Rev. E
978: {\bf 51}, 980 (1995); L. Kocarev and U. Parlitz, Phys. Rev. Lett. {\bf 76},
979: 1816 (1996).
980: \bibitem{Amri}
981: A. Maybhate and R.E. Amritkar, Phys. Rev. E {\bf 59},284 (1999);
982: Phys. Rev. E {\bf 61}, 6461 (2000).
983: \bibitem{Huang}
984: D.B. Huang, Phys. Rev. Lett. {\bf 93}, 214101 (2003);
985: Phys. Rev. E {\bf 69}, 067201 (2004); Phys. Rev. E {\bf 71}, 037203
986: (2005); Phys. Rev. E {\bf 73}, 066204 (2006).
987: \bibitem{spain}
988: U. Parlitz, Phys. Rev. Lett. {\bf 76}, 1232 (1996).
989: \bibitem{Yassen}
990: M.T. Yassen, Appl. Math. Comput. {\bf 135}, 113 (2001).
991: \bibitem{Hegazi}
992: A.S. Hegazi {\it et al.}, Int. J. Bifucation and Chaos Appl. Sci.
993: Eng. {\bf 12}, 1579 (2002)
994: \bibitem{Chen}
995: S.H. Chen {\it et al.}, Phys. Lett. A. {\bf 321}, 50 (2004).
996: \bibitem{Cao}
997: J. Lu {\it et al.}, Chaos {\bf 15}, 043901 (2005).
998: \bibitem{Marino}
999: I.P. Mari$\tilde{\rm n}$o, J. Miguez, Phys. Rev. E {\bf 72},
1000: 057202 (2005).
1001: \bibitem{Zhou}
1002: J. Zhou {\it et al.}, Chaos, Solitons and Fractals {\bf 27},
1003: 905 (2006).
1004: \bibitem{LaSalle}
1005: J.P. LaSalle, IRE Trans. Circuit Theory {\bf CT-7}, 520 (1960); {\it An Invariance Principle in The Theory
1006: of Stability, in Differential Equations and Dynamical Systems}, edited by J.K. Hale, J.P. LaSalle,
1007: (Academic Press, 1967).
1008: \bibitem{def}
1009: A family of real valued functions $\big\{g_i(\bm{x}),i=1,2,\cdots,n\big\}$
1010: are said to be linearly independent in some set $\mathcal{S}$ if and
1011: only if $\displaystyle\sum_{i=1}^n c_i g_i(\bm{x})= 0$ (for all
1012: $\bm{x}\in\mathcal{S}$) implies each $c_i=0$; otherwise, they are
1013: said to be linearly dependent. In particular, function
1014: $g(\bm{x})\equiv 0$ is linearly dependent but single nonzero
1015: constant function is linearly independent.
1016: \bibitem{Hale}
1017: J.K. Hale, S.V. Lunel, {\it Introduction to Functional Differential
1018: Equations} in Applied Mathematical Sciences. (Berlin, Germany:
1019: Springer-Verlag, 1993); J.K. Hale, Resenhas IME-USP {\bf 3}, 55
1020: (1997).
1021: \end{thebibliography}
1022:
1023: }
1024:
1025: \newpage
1026:
1027:
1028: \begin{figure}[htp]
1029: \centering
1030: \includegraphics[scale=0.5]{f1-a.eps}{\small(a)}
1031: \includegraphics[scale=0.5]{f1-b.eps}{\small(b)}
1032: \renewcommand{\figurename}{Fig.}
1033: \caption{\small (Color online) A successful complete synchronization
1034: between the Lorenz system (\ref{countereg1}) and (\ref{rep1}) by
1035: means of the adaptive design coupling. Here, system
1036: (\ref{countereg1}) possesses an asymptotically stable equilibrium
1037: $E=(6.8313,6.8313,1.6667)^T$ instead of the strange attractor. The
1038: variation of the driving signal with the response state are shown in
1039: (a) and the evolution of response state in the phase plane are
1040: depicted in (b). Here, $r_i=15$, $\delta_i=2$ and all the initial
1041: values are simply chosen as $x_i^0=y_i^0=10$, $q_i^0=1$,
1042: $\epsilon_i^0=1$ ($i=1,2,3$). }\label{fig1}
1043: \end{figure}
1044:
1045:
1046: \begin{figure}[htp]
1047: \centering
1048: \includegraphics[scale=0.5]{f2.eps}
1049: \renewcommand{\figurename}{Fig.}
1050: \caption{\small The variation of the error between the parameters
1051: $q_i$ and $p_i$ with time initiating from 0 to 60 with step-size
1052: 0.01 ($i=1,2,3$). In particular, $q_1$ fails to identify the
1053: accurate value of $p_1$. All the parameters and initial values for
1054: coupling systems are the same as those given in Fig.1. }\label{fig2}
1055: \end{figure}
1056:
1057:
1058: \begin{figure}[htp]
1059: \centering
1060: \includegraphics[scale=0.5]{f3-a.eps} {\small(a)}
1061: \includegraphics[scale=0.5]{f3-b.eps} {\small(b)}
1062: \renewcommand{\figurename}{Fig.}
1063: \caption{\small The attractive periodic orbit generated by the
1064: original Chen's system (system (\ref{countereg2}) when
1065: $\mathcal{Q}\equiv 0$). The periodic orbit in the $x_1$-$x_2$-$x_3$
1066: phase plane (a) and its projection in the $x_1$-$x_2$ plane
1067: (b).}\label{fig3}
1068: \end{figure}
1069:
1070:
1071: \begin{figure}[thp]
1072: \centering
1073: \includegraphics[scale=0.5]{f4.eps}
1074: \renewcommand{\figurename}{Fig.}
1075: \caption{\small (Color
1076: online) The variation of $q_i$ ($i=1,2$) with time initiating from 0 to 15 with step-size
1077: 0.01 when its initial value $q_{i\varrho}$ is differently taken ($\varrho=a,b,c,d,e$). Here, $r_j=2$, $\delta_j=1$ and all the initial
1078: values are taken as $x_j^0=y_j^0=10$, $\epsilon_j^0=1$, $j=1,2,3$.
1079: .}\label{fig4}
1080: \end{figure}
1081:
1082:
1083:
1084: \begin{figure}[htp]
1085: \centering
1086: \includegraphics[scale=0.5]{f5-a.eps}{\small (a)}
1087: \includegraphics[scale=0.5]{f5-b.eps}{\small (b)}
1088: \includegraphics[scale=0.5]{f5-c.eps}{\small (c)}
1089: \renewcommand{\figurename}{Fig.}
1090: \caption{\small (Color
1091: online) The strange attractor produced by system
1092: (\ref{countereg3}) are, respectively, plotted in the
1093: $x_1$-$x_2$-$x_3$ plane (a) and in the time-state-$x_4$ plane
1094: (b). The chaotic property is verified by the Lyapunov exponent portrait (c),
1095: where the largest Lyapunov exponent $\lambda_1$ is above zero.}\label{fig5}
1096: \end{figure}
1097:
1098:
1099:
1100: \begin{figure}[htp]
1101: \centering
1102: \includegraphics[scale=0.5]{f6.eps}
1103: \renewcommand{\figurename}{Fig.}
1104: \caption{\small The variation of the error between the parameters
1105: $q_j$ and $p_j$ with time initiating from 0 to 25 with step-size
1106: 0.01. Indeed, $q_{3,4}$ fails to identify the accurate value of
1107: $p_{3,4}$, respectively. Here, $r_j=15$, $\delta_j=2$, and all the
1108: initial values are taken as $x_j^0=1$, $y_1^0=6$, $y_2^0=$,
1109: $y_3^0=10$, $y_4^0=1$, $q_j^0=0$, and $\epsilon_j^0=1$
1110: ($j=1,2,3,4$).}\label{fig6}
1111: \end{figure}
1112:
1113:
1114: \begin{figure}[htp]
1115: \centering
1116: \includegraphics[scale=0.5]{f7.eps}
1117: \renewcommand{\figurename}{Fig.} \caption{\small (Color
1118: online) The variation of $x_3$ and $x_3(1+x_4^3)$ on the
1119: synchronized orbit $\bm{x}^*(t)$ with time, respectively
1120: }\label{fig7}
1121: \end{figure}
1122:
1123:
1124: \begin{figure}[htp]
1125: \centering
1126: \includegraphics[scale=0.5]{f8-a.eps}{\small (a)}
1127: \includegraphics[scale=0.5]{f8-b.eps}{\small (b)}
1128: \renewcommand{\figurename}{Fig.}
1129: \caption{\small (a) Successful complete synchronization and
1130: parameter identification for chaotic driving signal when $a=-2$ and
1131: $b=4$; (b) Failed parameter identification when $a=2$ and $b=1$.
1132: This failure is simply due to an approximate dependence between
1133: $f\big(x(t)\big)$ and $g\big(x(t-\tau)\big)$ in a macro scale. Here,
1134: both $f$ and $g$ are taken as sinusoid functions, time-delays are
1135: taken as $\tau=10$, $\delta=2$, and time step size is 0.01.
1136: }\label{fig8}
1137: \end{figure}
1138:
1139:
1140:
1141: \end{document}
1142: