Topology identification of uncertain nonlinearly coupled...
Transcript of Topology identification of uncertain nonlinearly coupled...
Topology identification of uncertain nonlinearly coupled complex networkswith delays based on anticipatory synchronizationYanqiu Che, Ruixue Li, Chunxiao Han, Shigang Cui, Jiang Wang et al. Citation: Chaos 23, 013127 (2013); doi: 10.1063/1.4793541 View online: http://dx.doi.org/10.1063/1.4793541 View Table of Contents: http://chaos.aip.org/resource/1/CHAOEH/v23/i1 Published by the AIP Publishing LLC. Additional information on ChaosJournal Homepage: http://chaos.aip.org/ Journal Information: http://chaos.aip.org/about/about_the_journal Top downloads: http://chaos.aip.org/features/most_downloaded Information for Authors: http://chaos.aip.org/authors
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
Topology identification of uncertain nonlinearly coupled complex networkswith delays based on anticipatory synchronization
Yanqiu Che,1,2,a) Ruixue Li,1 Chunxiao Han,1 Shigang Cui,1 Jiang Wang,3 Xile Wei,3
and Bin Deng3
1Tianjin Key Laboratory of Information Sensing & Intelligent Control, Tianjin University of Technologyand Education, Tianjin 300222, People’s Republic of China2Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China3School of Electrical Engineering and Automation, Tianjin University, Tianjin 300072, China
(Received 20 September 2012; accepted 5 February 2013; published online 21 February 2013)
This paper presents an adaptive anticipatory synchronization based method for simultaneous
identification of topology and parameters of uncertain nonlinearly coupled complex dynamical
networks with time delays. An adaptive controller is proposed, based on Lyapunov stability theorem
and Barb�alat’s Lemma, to guarantee the stability of the anticipatory synchronization manifold
between drive and response networks. Meanwhile, not only the identification criteria of network
topology and system parameters are obtained but also the anticipatory time is identified. Numerical
simulation results illustrate the effectiveness of the proposed method. VC 2013 American Institute ofPhysics. [http://dx.doi.org/10.1063/1.4793541]
The research on complex networks, from natural to man-
made ones, pervades almost all scientific and technological
fields. Mathematically, complex networks can be described
as graphs composed of nodes representing individual units
that are linked by edges denoting physical or functional
connections. Among various problems in the study of com-
plex networks, topology identification is a significant one.
This paper investigates the topology identification of the
uncertain complex networks with unknown parameters,
nonlinear coupling and delays in states and coupling,
which are not fully considered in the existing research.
Moreover, in the previous widely used auto-synchroniza-
tion-based method, the stability of the complete synchroni-
zation (CS) manifold will be destroyed because of the
existence of time delays in real systems. As a result, the
estimation of the unknown topology and parameters will
be practically impossible. In this paper, the proposed novel
adaptive identification approach is based on anticipatory
synchronization (AS), that is, the states of the response sys-
tem synchronize with the future states of the drive system
with an adaptive determined anticipatory time, which is
much more practical. Numerical simulations are given to
illustrate the effectiveness of the proposed approach.
I. INTRODUCTION
Complex networks, which successfully describe many
natural and man-made systems including gene network, neu-
ral networks, World Wide Web, and communication net-
work, to name but a few, have gained growing attention in
the past decades.1–6 Various issues in complex networks
have been investigated, such as analysis of network dynam-
ics,2–4 small-world and scale-free characteristics,5,6 and syn-
chronization of large-scale complex networks with certain
types of topology.7–10 Recently, topology identification for
complex networks is emerging as an attractive topic of
research because of its essential role in lots of complex net-
works.11–20 Since a complex network is composed of a large
number of interacting nodes, it is clear that small perturba-
tions in system parameters of nodes may dramatically
change the cooperative behavior and network features.21,22
Moreover, due to the finite speed of signal transmission over
a distance, time delays ubiquitously exist in complex net-
works,23,24 such as neural networks and communication net-
works. Therefore, simultaneous identification of topology
and parameter for uncertain complex networks with time
delays is of theoretical and practical importance.
Thus far, a few methods have been proposed for topol-
ogy identification of complex networks. These methods can
basically be divided into two categories, i.e., perturbation-
based12,13,17,19 and synchronization-based11,14–16,18,20 ones.
In the former category, researchers design various controller
to drive the complex dynamical network with unknown topol-
ogy to a steady-state, and then, by means of small perturba-
tion, transform the identification problem to solving a large
linear equation. In the latter category, researchers consider
the complex dynamical network with unknown topology as a
drive system, and construct a response system and design a
controller to make the states of the response system synchron-
ize with that of the drive one. During the process of synchro-
nization (CS,11,14–16,18 lag synchronization (LS),20 or AS),
the unknown topology can be identified. Most of the afore-
mentioned methods assume that (i) parameters of nodes are
known a priori, (ii) the nodes of complex networks are line-
arly coupled, or (iii) there exists only instant (at time t) ordelayed (at time t� s) information communication between
nodes. However, all these assumptions are not the usual cases
in real-world networks. Moreover, the CS-based method
requires complete synchronization between current states of
the drive and response system, which will be practically
impossible because of the presence of time delays during the
signal transmission. On the contrary, in many applications,a)Electronic mail: [email protected].
1054-1500/2013/23(1)/013127/7/$30.00 VC 2013 American Institute of Physics23, 013127-1
CHAOS 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
essential drive-response synchronization occurs regardless of
the natural coupling delays between drive and response sys-
tems,1,2 such as chemical transport delay between cells in a
living organism, optical delay in the synchronized movement
of swarms of birds, communication delay between multiple
robots in the execution of shared tasks, and transmission
delay in secure communication. In this regard, LS- or AS-
based methods, which require the response system to syn-
chronize with the past or future states of the drive system,
have important technological implications. Particularly, AS is
a state that the response system may anticipate the drive sys-
tem in real time, which is more reasonable and has been
observed in many practical systems such as lasers,25,26 elec-
tronic circuits,27 and neuronal systems.28 Thus, AS has
attracted a great deal of attention.29–36 However, to the best
of our knowledge, AS-based topology identification technol-
ogy has not been fully studied.
Motivated by the existing works, in this paper, using
adaptive control techniques, we propose an AS-based
approach to identify unknown network topological structure
and system parameters for general uncertain delayed nonli-
nearly coupled complex dynamical networks. Based on
Lyapunov stability theorem and Barb�alat’s Lemma,37
adaptive identification rules are proposed for the nonli-
nearly coupled networks with node and coupling delays.
During the AS process, not only the topology and parame-
ters but also the anticipatory time are identified, which
makes our approach much more practical.
The rest of this paper is organized as follows. Section II
introduces some useful mathematical preliminaries. Section III
presents the main theory for the identification of network topo-
logical structure and system parameters for nonlinearly coupled
uncertain complex dynamical networks with delays. After that,
in Sec. IV, a network of Hindmarsh-Rose (HR) neuronal
models38 is used to illustrate the effectiveness and reliability of
the suggested technique. Finally, Sec. V draws the conclusions.
II. PRELIMINARIES
Consider the following uncertain dynamical complex
networks consisting of N different nodes which are n-dimen-
sional dynamical units:
_xiðtÞ ¼ fiðxiðtÞ; xiðt� siÞ; aiÞ þXN
j¼1
aijgijðxjðtÞÞ
þXN
j¼1
bijhijðxjðt� sijÞÞ; i ¼ 1; 2;…;N (1)
where xi ¼ ðxi1; xi2;…; xinÞT 2Rn�1 and ai 2Rmi�1
for i¼ 1;…;N are the state vectors and unknown system
parameter vectors of the i-th node, respectively. fiðxiðtÞ;xiðt� siÞ;aiÞ 2Rn�1 is a known function of the i-th node dy-
namics. gijðxjÞ ¼ ðgij1ðxjÞ;gij2ðxjÞ;…;gijnðxjÞÞT and hijðxjÞ¼ ðhij1ðxjÞ;hij2ðxjÞ;…;hijnðxjÞÞT 2Rn�1 are known nonlinear
inner-coupling functions. A¼ ðaijÞN�N and B¼ ðbijÞN�N 2RN�N are unknown or uncertain outer-coupling configuration
matrices. si and sij are time-varying node delay of i-th node
and coupling delay from node j to node i, respectively.
In the following, we denote xs ¼ xðt� sÞ and omit the
time variable t whenever no confusion may arise.
Suppose that uncertain quantities in the i-th node
dynamic systems can be expressed linearly with respect to
unknown constant parameters ai, then rewrite system Eq. (1)
as
_xi ¼ Fiðx; xsii Þ þ �Fiðxi; x
sii Þai þ
XN
j¼1
aijgijðxjÞ
þXN
j¼1
bijhijðxsi
j
j Þ; i ¼ 1; 2;…;N (2)
where Fiðxi; xsii Þ 2 Rn�1 and �Fiðxi; x
sii Þ 2 Rn�mi are known
functions of the i-th node dynamics.
We now introduce some assumptions and lemmas that
will be required throughout this study.
Assumption 1: The functions gijð�Þ and hijð�Þ are
Lipschitzian, that is, for any vectors x, y 2 Rn�1, there exist
nonnegative constants Lgij and Lhij for i ¼ 1; 2;…;N such that
jjgijðxÞ � gijðyÞjj � Lgijjjx� yjjjjhijðxÞ � hiðyÞjj � Lhijjjx� yjj:
(3)
Hereafter, the norm jjxjj of vector x is defined as jjxjj ¼ xTx.
Assumption 2: There exists a nonnegative constant Msuch that
jjfiðxi;xsi ;aiÞ � fiðyi;y
si ;aiÞjj2 �Mðjjxi� yijj
2þ jjxsi � ys
i jj2Þ:(4)
Remark 1. Note that the assumptions 1 and 2 hold as long
as @F@x
(here, F denotes gij, hij, or fi) are uniformly bounded.39
Almost all well-known finite-dimensional chaotic and hyper-
chaotic systems, such as Chua’s circuit, hyperchaotic L€u sys-
tem, generalized Lorenz system, and neuronal systems such
as FitzHugh-Nagumo (FHN) and HR models have the form of
Eq. (2), which meet the conditions of assumptions 1 and 2.
Assumption 3: Time-varying delays siðtÞ and sijðtÞ are dif-
ferentiable and satisfy 0� _siðtÞ�l<1 and 0� _sijðtÞ�g<1,
respectively, where l and g are constants.
Clearly, assumption 3 is certainly valid for constant sðtÞ.Lemma 1. For any vectors x; y 2 Rn�1, the matrix in-
equality 2xTy � xTxþ yTy holds.
Definition 1. Two systems of complex networks are said
to be AS if limt!1 jjyðt� sdðtÞÞ � xðtÞjj ¼ 0 for a positive
anticipatory time sdðtÞ, where x and y are the system states
of the drive and response complex networks, respectively.
The main goal of this paper is to identify the unknown
or uncertain coupling strengths aij and bij (for i; j ¼ 1;…;N),
namely, its network topological structure, and all unknown
system parameter vectors ai (for i ¼ 1;…;N) of its node dy-
namical systems.
III. GENERAL THEORY
In this paper, we consider system Eq. (2) as a drive net-
work, and construct another complex dynamical network
called the response network with actual control as follows:
013127-2 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
_xsd
i ¼ Fiðxsdi ; x
sdþsii Þ þ �Fiðxsd
i ; xsdþsii Þasd
i
þXN
j¼1
asdij gijðxsd
j Þ þXN
j¼1
bsd
ij hijðxsdþsi
j
j Þ þ ui;
ui ¼ �sd~xi; _sd ¼ cXN
i¼1
jj~xijj2;
_asd
i ¼ �b �FTi ðx
sdi ; x
sdþsii Þ~xi;
_asd
ij ¼ �da~xTi gijðxsd
j Þ;_bsd
ij ¼ �db~xTi hijðx
sdþsij
j Þ;
(5)
where xi 2 Rn�1 is the response state vector of the i-th node,
ui 2 Rn is its controller, ~xi ¼ xsdi � xi is the synchronization
error between drive and response networks. ai 2 Rmi and
aij; bij 2 R are the estimated values of parameter vector ai
and unknown coupling strengths aij and bij, respectively, for
i; j ¼ 1;…N. c, b, da and db are any positive constants.
Theorem 1. Suppose that assumptions 1, 2, and 3 hold.
Then, the uncertain parameter vectors faigNi¼1 and coupling
configuration matrices A ¼ ðaijÞN�N and B ¼ ðbijÞN�N of the
uncertain general complex dynamical network Eq. (2) can be
identified by the estimated values faigNi¼1; A ¼ ðaijÞN�N and
B ¼ ðbijÞN�N , respectively, via the response network Eq. (5).
See proof of Theorem 1 in Appendix.
IV. EXAMPLE
To verify the effectiveness of the proposed method in to-
pology and parameter identification of complex networks,
we consider a network consists of 5 non-identical HR neuro-
nal systems38 with instant electrical coupling and delayed
chemical synaptic coupling
_xi1 ¼ xi2 � aix3i1 þ bix
2i1 � xi3 þ Iext þ ge
XN
j¼1
aijxj1
þ gsðxi1 � VsÞXN
j¼1
bijCðxsi
j
j Þ;
_xi2 ¼ ci � dix2i1 � xi2;
_xi3 ¼ ri½siðxi1 � x0Þ � xi3�;
(6)
where xi1 are the membrane potentials, xi2 are associated with
the fast current Naþ or Kþ, and xi3 with the slow current, for
example, Ca2þ. The nominal values of the parameters
are ai¼1:0;bi¼3:0;ci¼1:0, di¼5:0;si¼4:0;ri¼0:006, and
x0¼�1:60. Iext is the external current input. In the range of
2:92< Iext<3:40, individual HR systems show chaotic spike-
bursting behaviors.40 ge and gs are electrical and chemical
synaptic coupling strengths, respectively. The reversal poten-
tial Vs>xi1ðtÞ, i.e., the synapse is excitatory. The sigmoidal
function CðxjÞ¼ 1=f1þexp½�10ðxj�HsÞ�g describes the
synaptic coupling function. The matrix (aij) describes the elec-
trical coupling, while the matrix (bij) describes the synaptic
coupling.
In the following simulations, we fix Iext ¼ 3:2;Hs ¼ �0:25;Vs ¼ 2; ge ¼ 0:1, and gs ¼ 0:2 and assume that
the parameters di (the extension to more than one unknown
parameters is immediate), aij, and bij are unknown and
sij ¼ 0:2 for i; j ¼ 1;…; 5.
Comparing Eq. (6) with Eq. (2), we have xi ¼ ðxi1; xi2;xi3ÞT ;FiðxÞ¼ ðxi2 � ax3
i1 þ bx2i1; c� xi2; r½sðxi1 � x0Þ� xi3�ÞT ;
�FiðxÞ ¼ ð0;�x2i1; 0Þ
T, ai ¼ di; gijðxjÞ ¼ ðgexj1; 0; 0ÞT ; hijðx
sij
j Þ¼ ðgsðxi1 � VsÞCðx
sij
j Þ; 0; 0ÞT, n¼ 3 and N¼ 5.
As is well-known, failures in components of a dynamical
system can be represented by abrupt changes in the system
parameters. Similarly, repairs of the failed components and
system reconfigurations also cause abrupt changes in the sys-
tem parameters.41 Taking the neuronal system as an example,
in the pathological condition, the coupling between different
neurons may become stronger or weaker than that in normal
states.42 Motivation for applicability of our method to a large
class of realistic problems, we consider system models with
abrupt (but infrequent) parameters changes in the simulations.
That is, when t < 1500, for i; j ¼ 1;…; 5, we set the system
parameter di ¼ ð4:9; 4:95; 5; 5:05; 5:1ÞT and the coupling ma-
trix (aij) and (bij)
A ¼
�3 0 1 1 1
0 �2 1 1 0
1 1 �4 1 1
1 1 1 �3 0
1 0 1 0 �2
26666664
37777775;
B ¼
0 1 0 1 1
1 0 1 1 0
1 1 0 0 1
0 1 1 0 1
1 0 1 1 0
26666664
37777775:
When t � 1500, we switch di to ð4:8; 4:9; 5; 5:1; 5:2ÞT ,
and switch (aij) and (bij) to
A ¼
�2 1 0 0 1
1 �3 1 0 1
0 1 �2 1 0
0 0 1 �2 1
1 1 0 1 �3
26666664
37777775;
B ¼
1 0 1 0 0
0 1 0 0 1
0 0 1 1 0
1 0 0 1 0
0 1 0 0 1
26666664
37777775:
The parameters and topology are estimated by means of
Theorem 1. The control gains are chosen as c ¼ b ¼ 1 and
da ¼ db ¼ 10. The initial values estimated for the unknown
parameters and coupling strengths are set to be zero. The ini-
tial states of the drive and response systems are randomly
given.
The simulation results are shown in Figs. 1–4. In Fig. 1,
waveforms of the states x11ðtÞ and x11ðtÞ, and the state errors
~xij ¼ xijðt� sdÞ � xijðtÞ (i ¼ 1;…; 5; j ¼ 1;…; 3) versus
time are shown in (a) and (b), respectively. The state errors
013127-3 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
tend to zero asymptotically as time evolves, which indicates
anticipatory synchronization between the drive and response
systems. Figs. 2 and 3 show the identification of unknown
system parameters di and network topology, respectively.
Figs. 2(a) and 2(b) describe the estimation values of di and
the corresponding estimation errors, respectively. In Fig. 3,
the examples of estimation values of a2j and b2j and the
corresponding estimation errors ~aij ¼ aijðt� sdÞ � aij and~bij ¼ bijðt� sdÞ � bij for i; j ¼ 1;…; 5 are shown in (a, c)
and (b, d), respectively. Clearly, the estimated values of pa-
rameters and coupling topology finally evolve to the real
values of the unknown parameters and topology. Note that
the dynamic evolution of the response system in Figs. 1–3 at
t¼ 1500 when the drive system undergoes abrupt parame-
ters changes anticipates the dynamics and parameters
changes of the drive system. This implies that the proposed
approach can be applied for online “anticipating” of the net-
work topology and system parameters. Fig. 4 shows the
time series of the anticipatory time sd which tends to a con-
stant as time evolves. The identification is very successful.
Thus, the numerical simulations illustrate the effectiveness
of Theorem 1.
FIG. 1. Anticipatory synchronization of
Hindmarsh-Rose networks. (a) Time series of
states of drive and response network, x11ðtÞ(solid line) and x11ðtÞ (dotted-dashed line). (b)
Time series of anticipatory synchronization
errors ~xij ¼ x ijðt� sdÞ � xijðtÞ; i ¼ 1;…; 5;j ¼ 1;…; 3.
FIG. 2. Identification of system parameters
di (i ¼ 1;…; 5) in a Hindmarsh-Rose net-
work. (a) Time series of the estimation val-
ues d i and (b) time series of the estimation
errors ~di ¼ d iðt� sdÞ � di.
FIG. 3. Identification of network topology
aij and bij, i; j ¼ 1;…; 5 in a Hindmarsh-
Rose network. (a) and (c) are the time series
of the estimation values a2j and b2j, respec-
tively. (b) and (d) are the time series of the
estimation errors ~aij ¼ aijðt� sdÞ � aij and~bij ¼ bijðt� sdÞ � bij, respectively.
013127-4 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
V. CONCLUSIONS
We have proposed a new method based on adaptive an-
ticipatory synchronization to simultaneously identify the
unknown network topological structure and system parame-
ters of uncertain general complex dynamical networks with
nonlinear delayed coupling. According to the Lyapunov sta-
bility theorem and Barb�alat’s Lemma, the proposed method
guarantees the stable anticipatory synchronization between
drive and response systems, and thus ensures effective identi-
fication of the unknown network topological structure and
system parameters. This method also has the ability to dynam-
ically anticipate the parameters and topological structure.
Moreover, it operates in anticipatory synchronization, and the
anticipatory time is simultaneously identified, which makes it
superior than the previous methods that always require a com-
plete synchronization.11,14–16,18 The simulation results have
demonstrated the validity of the proposed method.
As in many previous studies,11,13–18,20 to keep things
simple, we also assume that all the states of each node are
completely available, in principle, which is not practical for
the case in which only partial state variables of partial nodes
are accessible. For example, in neuronal networks, usually
only the membrane potentials can be measurable. The pertur-
bation method12 works only for the network that automati-
cally reaches a stable stationary state, which fails for network
systems with complex dynamical behaviors (e.g., chaos).
Although the delayed feedback based steady state control
method19 accomplishes topology detection using only one
state variables of each node, its main disadvantages, including
careful calculation of control parameters, influence of dynam-
ical behavior of systems, and inability to time-varying topol-
ogy, limit its applications. On the contrary, our method does
not have these limitations and can be extended to practical
applications by combination with nonlinear state-observation
techniques. As verified by many theoretical and experimental
results, the unscented Kalman filter (UKF) is powerful in state
estimation of nonlinear system.43–45 An UKF aided AS-based
method for topology identification is under our investigation.
ACKNOWLEDGMENTS
This work was supported by The National Natural
Science Foundation of China (Grants Nos. 50907044,
60901035, 61072012, 61172009, and 61104032) and The
Science and Technology Development Program of Tianjin
Higher Education (Grants Nos. 20100819 and 20110834).
We are grateful to the editor and reviewers for their valuable
comments and suggestions. We would also acknowledge the
support of Tianjin University of Technology and Education
(Grants Nos. KYQD10009 and KJ11-04).
APPENDIX: PROOF OF THEOREM 1
Proof. Denote ~xi ¼ xsdi � xi; ~ai ¼ asd
i � ai; ~aij ¼ asdij
�aij; ~bij ¼ bsd
ij � bij. Then, the error dynamics is described as
_~x i ¼ Fiðxsdi ; x
sdþsii Þ þ �Fiðxsd
i ; xsdþsii Þasd
i � Fiðxi; xsii Þ � �Fiðxi; x
sii Þai
þXN
j¼1
asdij gijðxsd
j Þ þXN
j¼1
bsd
ij hijðxsdþsi
j
j Þ �XN
j¼1
aijgijðxjÞ �XN
j¼1
bijhijðxsi
j
j Þ þ ui
¼ fiðxsdi ; x
sdþsii ; aiÞ � fiðxi; x
sii ; aiÞ þ �Fiðxsd
i ; xsdþsii Þ~ai þ
XN
j¼1
~aijgijðxsdj Þ þ
XN
j¼1
aijðgijðxsdj Þ � gijðxjÞÞ
þXN
j¼1
~bijhijðxsdþsi
j
j Þ þXN
j¼1
bijðhijðxsdþsi
j
j Þ � hijðxsi
j
j ÞÞ þ ui: (A1)
In order to guarantee ~xiðtÞ ! 0; ~aiðtÞ ! 0; ~aijðtÞ ! 0, ~bijðtÞ ! 0, and sdðtÞ ! s�d as t!1, consider the following
Lyapunov function candidate:
2V ¼XN
i¼1
~xTi ~xi þ
1
b
XN
i¼1
~aTi ~ai þ
1
da
XN
i¼1
XN
j¼1
~a2ij þ
1
db
XN
i¼1
XN
j¼1
~b2
ij þ1
cðsd � s�dÞ
2
þ M
1� l
XN
i¼1
ðt
t�si
~xTi ðsÞ~xiðsÞdsþ j
XN
i¼1
XN
j¼1
ðt
t�sij
~xTj ðsÞ~xjðsÞds; (A2)
FIG. 4. Time series of anticipatory time sdðtÞ.
013127-5 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
where s�d and j are sufficiently large positive constants to be determined.
Differentiating Eq. (A2) about time and noting Eqs. (5) and (A1), one has
_V jð5ÞðA1Þ ¼XN
i¼1
~xTi
_~x i þ1
b
XN
i¼1
~aTi
_asd
i þ1
da
XN
i¼1
XN
j¼1
~aij_asd
ij þ1
db
XN
i¼1
XN
j¼1
~bij_bsd
ij þ1
cðsd � s�dÞ _sd
þ M
2ð1� lÞXN
i¼1
½jj~xijj2 � ð1� _siÞjj~xsii jj
2� þ j2
XN
i¼1
XN
j¼1
½jj~xjjj2 � ð1� _sijÞjj~x
sij
j jj2�
¼XN
i¼1
~xTi ½fiðx
sdi ; x
sdþsii ; aiÞ � fiðxi; x
sii ; ai� � s�d
XN
i¼1
jj~xijj2
þXN
i¼1
XN
j¼1
aij~xTi ðgijðxsd
j Þ � gijðxjÞÞ þXN
i¼1
XN
j¼1
bij~xTi ðhijðx
sdþsij
j Þ � hijðxsi
j
j ÞÞ
þ M
2ð1� lÞXN
i¼1
½jj~xijj2 � ð1� _siÞjj~xsii jj
2� þ j2
XN
i¼1
XN
j¼1
½jj~xjjj2 � ð1� _sijÞjj~x
sij
j jj2�: (A3)
From assumptions 1, 2, and Lemma 1, one gets
~xTi ½fiðxsd
i ; xsdþsii ; aiÞ � fiðxi; x
sii ; ai� �
M þ 1
2jj~xijj2 þ
M
2jjxsi
i jj2; (A4)
aij~xTi ðgijðxsd
j Þ � gijðxjÞÞ � jaijj � jj~xjj � jjðgijðxsdj Þ � gijðxjÞjj � Lgijjaijj � jj~xjj � jj~xjjj �
Lgijjaijj2ðjj~xijj2 þ jj~xjjj2Þ; (A5)
bij~xTi ðhijðx
sdþsij
j Þ � hijðxsi
j
j ÞÞ � jbijj � jj~xijj � jjhijðxsdþsi
j
j Þ � hijðxsi
j
j Þjj � Lhijjbijj � jj~xijj � jj~xsi
j
j jj �Lhijjbijj
2ðjj~xijj2 þ jj~x
sij
j jj2Þ: (A6)
Denote aM ¼ maxi;j¼1;2;…;Nfjaijjg; bM ¼ maxi;j¼1;2;…;Nfjbijjg; LgM ¼ maxi¼1;2;…;NfLgijg, and LhM ¼ maxi¼1;2;…;NfLhijg.According to Eqs. (A4)–(A6), one obtains
_V jð5ÞðA1Þ �M þ 1
2þ M
2ð1� lÞ þ N LgMaM þLhMbM
2þ j
2
� �� s�d
� �XN
i¼1
jj~xijj2
þMð _si � lÞ2ð1� lÞ
XN
i¼1
jj~xsii jj
2 þXN
i¼1
XN
j¼1
LhMbM þ jð_sij � 1Þ
2jj~xsi
j
j jj2: (A7)
Letting
j ¼ LhMbM
1� g;
s�d ¼M þ 1
2þ M
2ð1� lÞ þ N LgMaM þLhMbM
2þ j
2
� �þ 1;
(A8)
noting assumption 3 and denoting eðtÞ ¼ ð~xT1 ðtÞ; ~xT
2 ðtÞ;…;~xT
NðtÞÞT 2 RnN�1, one has
_V jð5Þð7Þ � �eTe � 0: (A9)
From the above inequality, one has
0 � limt!1
ðt
0
eTðsÞeðsÞds � Vð0Þ � limt!1
VðtÞ: (A10)
Obviously, the right part of the above inequality is
bounded since both V(0) and V(t) are bounded according to
Eqs. (A2) and (A9). Therefore, eðtÞ 2L2. Moreover, eðtÞ is
bounded, so eðtÞ 2L1. According to Eq. (A1), _eðtÞ exists
and is bounded for t � 0. Thus, it follows from Barb�alat’s
Lemma37 that limt!1 eðtÞ ¼ 0. Then, one has limt!1_~x iðtÞ ¼ 0 and x
sd
i converges to
�i ¼�
xsdi : �Fiðxsd
i ; xsdþsii Þ~ai þ
XN
j¼1
~aijgijðxsdj Þ
þXN
j¼1
~bijhijðxsdþsi
j
j Þ ¼ 0
�as t!1; (A11)
where i ¼ 1; 2;…;N. Thus, one has ~aiðtÞ ! 0; ~aijðtÞ ! 0,
and ~bijðtÞ ! 0 as t!1.39 It implies that the unknown or
uncertain system parameter vectors ai and coupling matrices
013127-6 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions
A and B can be successfully identified using updating law
Eq. (5) in the process of anticipatory synchronization
between the drive and response systems.
This completes the proof.
1A. Pikovsky, M. rosenblum, and J. Kurths, Synchronization: A UniversalConcept in Nonlinear Science (Cambridge University Press, 2001).
2G. V. Osipov, J. Kurths, and C. Zhou, Synchronization in OscillatoryNetworks (Springer, 2007).
3S. Boccaletti, V. Latora, Y. Moreno, M. Ghavez, and D. U. Hwang,
“Complex networks: Structure and dynamics,” Phys. Rep. 424, 175–308
(2006).4H. Jeong, B. Tombor, R. Albert, Z. N. Oltvai, and A.-L. Barab�asi, “The
large-scale organization of metabolic networks,” Nature 407, 651–654
(2000).5D. J. Watts and S. H. Strogatz, “Collective dynamics of “small-world”
networks,” Nature 393, 440 (1998).6A. L. Barab�asi and R. Albert, “Emergence of scaling in random networks,”
Science 286, 509 (1999).7M. Barahona and L. M. Pecora, “Synchronization in small-world systems,”
Phys. Rev. Lett. 89, 054101 (2002).8L. Kocarev and P. Amato, “Synchronization in power-law networks,”
Chaos 15, 024101 (2005).9C. S. Zhou and J. Kurths, “Dynamical weights and enhanced synchroniza-
tion in adaptive complex networks,” Phys. Rev. Lett. 96, 164102 (2006).10A. Arenas, A. Diaz-Guilera, J. Kurths, Y. Moreno, and C. S. Zhou,
“Synchronization in complex networks,” Phys. Rep. 469, 93 (2008).11D. Yu, M. Righero, and L. Kocarev, “Estimating topology of network,”
Phys. Rev. Lett. 97, 188701 (2006).12M. Timme, “Revealing network connectivity from response dynamics,”
Phys. Rev. Lett. 98, 224101 (2007).13D. Yu and U. Parlitz, “Driving a network to steady states reveals its coop-
erative architecture,” Europhys. Lett. 81, 48007 (2008).14X. Wu, “Synchronization-based topology identification of weighted gen-
eral complex dynamical networks with time-varying coupling delay,”
Physica A 387, 997–1008 (2008).15H. Liu, J. A. Lu, J. L€u, and D. J. Hill, “Structure identification of uncertain
general complex dynamical networks with time delay,” Automatica 45,
1799–1807 (2009).16L. Chen, J. Lu, and C. K. Tse, “Synchronization: an obstacle to identifica-
tion of network topology,” IEEE Trans. Circuits Syst., II: Express Briefs
56, 310–314 (2009).17D. Yu, “Estimating the topology of complex dynamical networks by
steady state control: Generality and limitation,” Automatica 46, 2035–
2040 (2010).18J. Zhao, Q. Li, J. A. Lu, and Z.-P. Jiang, “Topology identification of com-
plex dynamical networks,” Chaos 20, 023119 (2010).19D. Yu and U. Parlitz, “Inferring network connectivity by delayed feedback
control,” PLos ONE 6, e24333 (2011).20Y. Q. Che, R. X. Li, C. X. Han, J. Wang, S. G. Cui, B. Deng, and X. L.
Wei, “Adaptive lag synchronizationbased topology identification scheme
of uncertain general complex dynamical networks,” Eur. Phys. J. B 85,
265 (2012).21A. Roxin, H. Riecke, and S. A. Solla, “Self-sustained activity in a small-
world network of excitable neurons,” Phys. Rev. Lett. 92, 198101 (2004).
22X. Wang, Y. Lu, M. Jiang, and Q. Ouyang, “Attraction of spiral waves by
localized inhomogeneities with small-world connections in excitable
media,” Phys. Rev. E 69, 056223 (2004).23K. Pyragas, “Synchronization of coupled time-delay systems: analytical
estimations,” Phys. Rev. E 58, 3067–3071 (1998).24Q. Zhang, J. Lu, J. L€u, and C. Tse, “Adaptive feedback synchronization of
a general complex dynamical network with delayed nodes,” IEEE Trans
Circuits and Syst., II: Express Briefs 55, 183–187 (2008).25T. Heil, I. Fischer, W. Els€asser, J. Mulet, and C. R. Mirasso, “Chaos syn-
chronization and spontaneous symmetry-breaking in symmetrically delay-
coupled semiconductor lasers,” Phys. Rev. Lett. 86, 795–798 (2001).26S. Tang and J. M. Liu, “Experimental verification of anticipated and re-
tarded synchronization in chaotic semiconductor lasers,” Phys. Rev. Lett.
90, 194101 (2003).27N. J. Corron, J. N. Blakely, and S. D. Pethel, “Lag and anticipating syn-
chronization without time-delay coupling,” Chaos 15, 023110 (2005).28M. Ciszak, F. Marino, R. Toral, and S. Balle, “Dynamical mechanism of
anticipating synchronization in excitable systems,” Phys. Rev. Lett. 93,
114102 (2004).29H. U. Voss, “Anticipating chaotic synchronization,” Phys. Rev. E 61, 5115
(2000).30H. Huijberts, H. Nijmeijer, and T. Oguchi, “Anticipating synchronization
of chaotic lur’e systems,” Chaos 18, 019901 (2008).31G. Ambika and R. E. Amiritkar, “Anticipatory synchronization with vari-
able time delay and reset,” Phys. Rev. E 79, 056206 (2009).32Q. Han, C. Li, and T. Huang, “Anticipating synchronization of a class of
chaotic systems,” Chaos 19, 023105 (2009).33Q. Han, C. Li, and J. Huang, “Anticipating synchronization of chaotic sys-
tems with time delay and parameter mismatch,” Chaos 19, 013104 (2009).34H. Wei and L. Li, “Estimating parameters by anticipating chaotic syn-
chronization,” Chaos 20, 023112 (2010).35F. S. Matias, P. V. Carelli, and C. R. Mirasso, “Anticipated synchroniza-
tion in a biologically plausible model of neuronal motifs,” Phys. Rev. E
84, 021922 (2011).36C. Mayol, C. R. Mirasso, and R. Toral, “Anticipated synchronization and
the predict-prevent control method in the fitzhugh-nagumo model system,”
Phys. Rev. E 85, 056216 (2012).37J.-J. E. Slotine and W. Li, Applied Nonlinear Control (Prentice Hall, NJ,
1991).38J. L. Hindmarsh and R. M. Rose, “A model of neuronal bursting using 3
coupled 1st order differential-equations,” Proc. R. Soc. Lond. B 221,
87–102 (1984).39H. K. Khalil, Nonlinear Systems, 3rd ed. (Prentice Hall, NJ, 2002).40Y. Q. Che, J. Wang, K. M. Tsang, and W. L. Chan, “Unidirectional syn-
chronization for hindmarsh-rose neurons via robust adaptive sliding mode
control,” Nonlinear Anal. Real World Appl. 11, 1096–1104 (2010).41A. S. Willsky, “A survey of design methods for failure detection in
dynamic systems,” Automatica 12, 601–611 (1976).42T. I. Netoff, R. Clewley, S. Arno, T. Keck, and J. A. White, “Epilepsy in
small-world networks,” J. Neurosci. 24, 8075–8083 (2004).43S. J. Julier and J. K. Uhlmann, “Unscented filtering and nonlinear
estimation,” Proc. IEEE 92, 401–422 (2004).44G. Ullah and S. J. Schiff, “Tracking and control of neuronal hodgkin-
huxley dynamics,” Phys. Rev. E 79, 040901(R) (2009).45G. Ullah and S. J. Schiff, “Assimilating seizure dynamics,” PLOS
Comput. Biol. 6, e1000776 (2010).
013127-7 Che et al. Chaos 23, 013127 (2013)
Downloaded 28 Aug 2013 to 159.226.228.14. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://chaos.aip.org/about/rights_and_permissions