author's personal copyynucc.yu.ac.kr/~jessie/temp/fi1483.pdf · author's personal copy...
TRANSCRIPT
This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies areencouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Available online at www.sciencedirect.com
Journal of the Franklin Institute 349 (2012) 1699–1720
Synchronization criteria for coupled stochasticneural networks with time-varying delays
and leakage delay
M.J. Parka, O.M. Kwona,n, Ju H. Parkb, S.M. Leec, E.J. Chad
aSchool of Electrical Engineering, Chungbuk National University, 52 Naesudong-ro,
Heungduk-gu, Cheongju 361-763, Republic of KoreabDepartment of Electrical Engineering, Yeungnam University, 214-1 Dae-Dong,
Kyongsan 712-749, Republic of KoreacSchool of Electronic Engineering, Daegu University, Gyungsan 712-714, Republic of KoreadDepartment of Biomedical Engineering, School of Medicine, Chungbuk National University,
52 Naesudong-ro, Heungduk-gu, Cheongju 361-763, Republic of Korea
Received 7 November 2011; received in revised form 12 January 2012; accepted 5 February 2012
Available online 13 February 2012
Abstract
This paper proposes new delay-dependent synchronization criteria for coupled stochastic neural
networks with time-varying delays and leakage delay. By constructing a suitable Lyapunov–
Krasovskii’s functional and utilizing Finsler’s lemma, novel synchronization criteria for the networks
are established in terms of linear matrix inequalities (LMIs) which can be easily solved by using the
LMI toolbox in MATLAB. Three numerical examples are given to illustrate the effectiveness of the
proposed methods.
& 2012 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
1. Introduction
Complex networks have received increasing attention of researches from various fieldsof science and engineering such as the World Wide Web, social networks, electrical powergrids, global economic markets, and so on [1,2]. Also, in the real applications of systems,there exists naturally time-delay due to the finite information processing speed and the
www.elsevier.com/locate/jfranklin
0016-0032/$32.00 & 2012 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfranklin.2012.02.002
nCorresponding author. Tel.: þ82 43 261 2422; fax: þ82 43 263 2419.
E-mail address: [email protected] (O.M. Kwon).
Author's personal copy
finite switching speed of amplifiers. It is well known that time-delay often causesundesirable dynamic behaviors such as performance degradation, and instability of thesystems. Therefore, recently, the problem of synchronization of coupled neural networkswith time-delay which is one of the hot research fields of complex networks has been achallenging issue due to its potential applications such as information science, biologicalsystems and so on [3,4].On the other hand, in implementation of many practical systems such as aircraft,
chemical and biological systems and electric circuits, there exist occasionally stochasticperturbations. It is no less important than the time-delay as a considerable factor affectingdynamics in the systems. Therefore, stochastic modeling with time-delay plays animportant role in many fields of science and engineering applications. For this reason,various approaches to stability criteria for stochastic systems with time-delay have beeninvestigated in the literature [5–8]. Xu et al. [5] studied the problem of stability analysisfor stochastic systems with parameter uncertainties and a class of nonlinearities. In [6],by the Lyapunov–Krasovskii’s functional based on the delay fractioning approach, anexponential stability criterion for stochastic systems with time-delay was presented. Kwon[7] derived the delay-dependent stability criteria for uncertain stochastic dynamic systemswith time-varying delays via the Lyapunov–Krasovskii’s functional approach with twodelay fraction numbers. Moreover, by choosing the Lyapunov matrices in the decomposedintegral intervals, stability analysis for stochastic neural networks with time-varying delaywere proposed in [8]. Above this, the study on the problems for various forms of complexnetworks or stochastic systems have been addressed. For more details, see the literature[9–16] and references therein.Recently, the synchronization stability problem for a class of complex dynamical
networks with Markovian jumping parameters and mixed time delays had been studied in[17]. The considered model in [17] has stochastic coupling term and stochastic disturbanceto reflect more realistic dynamical behaviors of the complex networks that are affected bynoisy environment. Very recently, a leakage delay, which is the time delay in leakage termof the systems and a considerable factor affecting dynamics for the worse in the systems, isbeing put to use in the problem of stability for neural networks [18,19]. Balasubramaniamet al. [18] investigated the problem of passivity analysis for neutral type neural networkswith Markovian jumping parameters and time delay in the leakage term. By use of thetopological degree theory, delay-dependent stability conditions of neural networks ofneutral type with time delay in the leakage term was proposed in [19]. However, to the bestof authors’ knowledge, delay-dependent synchronization analysis of coupled stochasticneural networks with time-varying delay and leakage delay has not been investigated yet.Here, delay-dependent analysis has been paid more attention than delay-independent onebecause the sufficient conditions for delay-dependent analysis make use of the informationon the size of time delay [20]. That is, the former is generally less conservative than thelatter.Motivated by the above discussions, the problem of new delay-dependent synchroniza-
tion criteria for coupled stochastic neural networks with time-varying delays in networkcoupling and leakage delay is considered. The coupled stochastic neural networks arerepresented as a simple mathematical model by use of Kronecker product technique. Then,by construction of a suitable Lyapunov–Krasovskii’s functional and utilization of Finsler’slemma, new synchronization criteria are derived in terms of LMIs which can be solvedefficiently by standard convex optimization algorithms [21]. In order to utilize Finsler’s
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201700
Author's personal copy
lemma as a tool of getting less conservative synchronization criteria, it should be notedthat a new zero equality from the constructed mathematical model is devised. The conceptof scaling transformation matrix will be utilized in deriving zero equality of the method.Finally, three numerical examples are included to show the effectiveness of the proposedmethod.
Notation: Rn is the n-dimensional Euclidean space, and Rm�n denotes the set of m� n
real matrix. For symmetric matrices X and Y, X4Y (respectively, XZY ) means that thematrix X�Y is positive definite (respectively, nonnegative). In, 0n and 0m�n denote n� n
identity matrix, n� n and m� n zero matrices, respectively. En denotes the n� n matrixwhich all elements are 1. J � J refers to the Euclidean vector norm and the induced matrixnorm. lmaxð�Þ means the maximum eigenvalue of a given square matrix. diagf� � �g denotesthe block diagonal matrix. % represents the elements below the main diagonal of asymmetric matrix. For a given matrix X 2 Rm�n, such that rankðX Þ ¼ r, we define X? 2
Rn�ðn�rÞ as the right orthogonal complement of X; i.e., XX? ¼ 0. Let ðO,F,fFtgtZ0,PÞ becomplete probability space with a filtration fFtgtZ0 satisfying the usual conditions (i.e., it isright continuous and F0 contains all P-pull sets). Ef�g stands for the mathematicalexpectation operator with respect to the given probability measure P.
2. Problem statements
Consider the following neural networks with time-varying delays and leakage delay
_yðtÞ ¼�Ayðt�tÞ þW1gðyðtÞÞ þW2gðyðt�hðtÞÞÞ þ b, ð1Þ
where yðtÞ ¼ ½y1ðtÞ, . . . ,ynðtÞ�T 2 Rn is the neuron state vector, n denotes the number of
neurons in a neural network, gð�Þ ¼ ½g1ð�Þ, . . . ,gnð�Þ�T 2 Rn denotes the neuron activation
function vector with gð0Þ ¼ 0, b¼ ½b1, . . . ,bn�T 2 Rn means the constant external input
vector, A¼ diagfa1, . . . ,ang 2 Rn�nðak40,k¼ 1, . . . ,nÞ is the self-feedback matrix, Wk 2
Rn�nðk¼ 1; 2Þ are the connection weight matrices, and h(t) and t are time-varying delaysand leakage delay, respectively, satisfying
0rhðtÞrhM , _hðtÞrhD and 0ot, ð2Þ
where hM and t are positive constant scalars and hD is any constant one.In this paper, it is assumed that the activation functions satisfy the following
assumption:
Assumption 1. The neurons activation functions, gkð�Þ, are assumed to be nondecreasing,bounded and globally Lipschiz; that is
jgkðx1Þ�gkðx2Þjjx1�x2j
rlk, x1,x2 2 R, x1ax2, k¼ 1, . . . ,n, ð3Þ
where lk are positive constants.For simplicity, in stability analysis of the network (1), the equilibrium point
yn ¼ ½yn1, . . . ,y
nn �
T is shifted to the origin by utilization of the transformationxð�Þ ¼ yð�Þ�yn, which leads the system (1) to the following form:
_xðtÞ ¼ �Axðt�tÞ þW1f ðxðtÞÞ þW2f ðxðt�hðtÞÞÞ, ð4Þ
where xðtÞ ¼ ½x1ðtÞ, . . . ,xnðtÞ�T 2 Rn is the state vector of the transformed network.
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1701
Author's personal copy
Also, the function f ð�Þ ¼ ½f1ð�Þ, . . . ,fnð�Þ�T with fqð�Þ ¼ gqð� þ yn
qÞ�gqðynqÞ and fqð0Þ ¼ 0
satisfies Assumption 1.In this paper, a model of coupled stochastic neural networks with time-varying delays in
network coupling and leakage delay is considered as
dxiðtÞ ¼ ½�Axiðt�tÞ þW1f ðxiðtÞÞ þW2f ðxiðt�hðtÞÞÞ� dtþXN
j ¼ 1
gijGxjðt�hðtÞÞðdtþ do1ðtÞÞ
þsiðt,xiðtÞ,xiðt�hðtÞÞÞ do2ðtÞ, i¼ 1; 2, . . . ,N, ð5Þ
where N is the number of couple nodes, xiðtÞ ¼ ½xi1ðtÞ, . . . ,xinðtÞ�T 2 Rn is the state vector of
the ith node, G 2 Rn�n is the constant inner-coupling matrix of nodes, which represent theindividual coupling between the subnetworks, G¼ ½gij �N�N is the outer-coupling matrixrepresenting the coupling strength, the topological structure of the networks satisfy thediffusive coupling connections
gij ¼ gjiZ0 ðiajÞ, gii ¼�XN
j ¼ 1,iaj
gij ði,j ¼ 1; 2, . . . ,NÞ,
and okðtÞðk¼ 1; 2Þ are m-dimensional Wiener processes (Brownian motion) onðO,F,fFtgtZ0,PÞ which satisfies the following definition.
Definition 1 (Morters and Peres [22]). A real-valued stochastic process oðtÞ is called aBrownian motion or Wiener process if the following holds:
(i) oð0Þ ¼ 0 a.s.,(ii) oðtÞ has independent increment, i.e., for all 0rt1rt2r � � �rtn, the increments
oðt2Þ�oðt1Þ, . . . ,oðtnÞ�oðtn�1Þ are independent random variables,(iii) the increments oðtÞ�oðsÞ are N ð0,t�sÞ for all 0rsrt, where N ðm,n2Þ denote the
normal distribution with expected value m and variance n2,(iv) oðtÞ is almost surely continuous.
It should be noted that
EfdokðtÞg ¼ 0, Efdo2kðtÞg ¼ dt:
Here, o1ðtÞ and o2ðtÞ, which are mutually independent, are the coupling strengthdisturbance and the system noise, respectively. And the nonlinear uncertainties sið�, � ,�Þ 2Rn�mði¼ 1, . . . ,NÞ are the noise intensity functions satisfying the Lipschitz condition andthe following assumption:
sTi ðt,xiðtÞ,xiðt�hðtÞÞÞsiðt,xiðtÞ,xiðt�hðtÞÞÞrJH1xiðtÞJ
2þ JH2xiðt�hðtÞÞJ2, ð6Þ
where Hkðk¼ 1; 2Þ are constant matrices with appropriate dimensions.
Remark 1. In this paper, the activation functions and the noise intensity functions areassumed to be satisfied the conditions (3) and (6) which are Lipschitz continuous andsatisfy the linear growth condition as well. Under the assumptions, according to the workin Chapter 5 of [23], it is easy that the ith stochastic neural networks (5) expressed bydelayed stochastic differential equation has a unique continuous solution. One can alsoconfirm this fact in [24].
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201702
Author's personal copy
For the convenience of stability analysis for the networks (5), the following Kroneckerproduct and its properties are used.
Lemma 1 (Graham [25]). Let � denote the notation of Kronecker product. Then, the
following properties of the Kronecker product are easily established:
(i) ðaAÞ � B¼A� ðaBÞ.(ii) ðAþ BÞ � C ¼A� C þ B� C.(iii) ðA� BÞðC �DÞ ¼ ðACÞ � ðBDÞ.(iv) ðA� BÞT ¼AT � BT .
Let us define
xðtÞ ¼ ½x1ðtÞ, . . . ,xNðtÞ�T , fðxðtÞÞ ¼ ½f ðx1ðtÞÞ, . . . ,f ðxN ðtÞÞ�
T ,
rðtÞ ¼ ½s1ð�, � ,�Þ, . . . ,sN ð�, � ,�Þ�T :
Then, with Kronecker product in Lemma 1, the network (5) can be represented as
dxðtÞ ¼ ½�ðIN � AÞxðt�tÞ þ ðIN �W1ÞfðxðtÞÞ þ ðIN �W2Þfðxðt�hðtÞÞÞ� dt
þðG � GÞxðt�hðtÞÞðdtþ do1ðtÞÞ þ rðtÞ do2ðtÞ: ð7Þ
For stability analysis, Eq. (7) can be rewritten as
dxðtÞ ¼ gðtÞ dtþ .ðtÞ dxðtÞ, ð8Þ
where
gðtÞ ¼ �ðIN � AÞxðt�tÞ þ ðIN �W1ÞfðxðtÞÞ þ ðIN �W2Þfðxðt�hðtÞÞÞ
þðG � GÞxðt�hðtÞÞ,
.ðtÞ ¼ ½ðG � GÞxðt�hðtÞÞ,rðtÞ�, xT ðtÞ ¼ ½oT1 ðtÞ,o
T2 ðtÞ�:
The aim of this paper is to investigate the delay-dependent synchronization stabilityanalysis of network (8) with time-varying delays in network coupling and leakage delay.In order to do this, the following definition and lemmas are needed.
Definition 2 (Liu and Chen [26]). The networks (5) are said to be asymptoticallysynchronized if the following condition holds:
limt-1
JxiðtÞ�xjðtÞJ¼ 0, i,j ¼ 1; 2, . . . ,N:
Lemma 2 (Cao et al. [3]). Let U ¼ ½uij �N�N , P 2 Rn�n, xT ¼ ½x1,x2, . . . ,xn�T , and
yT ¼ ½y1,y2, . . . ,yn�T . If U ¼UT and each row sum of U is zero, then
xT ðU � PÞy¼�X
1riojrN
uijðxi�xjÞT Pðyi�yjÞ:
Lemma 3 (Gu [27]). For any constant matrix M ¼MT40, the following inequality holds:
�hðtÞ
Z t
t�hðtÞ
_xT ðsÞM _xðsÞ dsrxðtÞ
xðt�hðtÞÞ
" #T�M M
% �M
� �xðtÞ
xðt�hðtÞÞ
" #: ð9Þ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1703
Author's personal copy
Lemma 4 (Finsler’s lemma [28]). Let z 2 Rn, F¼FT 2 Rn�n, and B 2 Rm�n such that
rankðBÞon. The following statements are equivalent:
(i) zTFzo0,8Bz¼ 0,za0,(ii) B?
TFB?o0,(iii) (X 2 Rn�m : Fþ XBþ BT X To0.
3. Main results
In this section, new synchronization criteria for network (8) will be proposed. For thesake of simplicity on matrix representation, eiði¼ 1, . . . ,7Þ 2 R7n�n and gij are defined asblock entry matrices (For example, e2 ¼ ½0n,In,0n,0n,0n,0n,0n�
T ) and the (i,j)th entry of amatrix GT G¼ ½gij�N�N , respectively. The notations of several matrices are defined as:
zTðtÞ ¼ ½xT ðtÞ,xT ðt�tÞ,xT ðt�hðtÞÞ,xT ðt�hM Þ,f
T ðxðtÞÞ,fT ðxðt�hðtÞÞÞ,gT ðtÞ�,
U¼ ½0nN ,�ðIN � AÞ,ðG � GÞ,0nN ,ðIN �W1Þ,ðIN �W2Þ,�InN �,
zijðtÞ ¼ xiðtÞ�xjðtÞ,f ðzijðtÞÞ ¼ f ðxiðtÞÞ�f ðxjðtÞÞ,ZijðtÞ ¼ ZiðtÞ�ZjðtÞ,
zTij ðtÞ ¼ ½z
Tij ðtÞ,z
Tij ðt�tÞ,z
Tij ðt�hðtÞÞ,zT
ij ðt�hM Þ,fT ðzijðtÞÞ,f
T ðzijðt�hðtÞÞÞ,ZTij ðtÞ�,
Uij ¼ ½0n,�A,�ðNgijGÞ,0n,W1,W2,�In�,
X1 ¼ e1P1eT7 þ e7P1e
T1 þ e3ðNgijG
T P1GÞeT3 þ e1ðrHT
1 H1ÞeT1 þ e3ðrHT
2 H2ÞeT3 ,
X2 ¼ e1ðQ1 þQ2ÞeT1�e3ð1�hDÞQ1e
T3�e4Q2e
T4 ,
X3 ¼ e7ðh2MR1Þe
T7 �ðe1�e3ÞRðe1�e3Þ
T�ðe3�e4ÞRðe3�e4Þ
T
�ðe1�e3ÞTðe3�e4ÞT�ðe3�e4ÞT
T ðe1�e3ÞT ,
X4 ¼ e1S1eT1 �e2S1e
T2 þ e7ðt2S2Þe
T7 �e1S2e
T1 þ e1S2eT
2 þ e2S2eT1�e2S2e
T2 ,
X5 ¼ e1LT1 D1L1e
T1�e5D1eT
5 ,
X6 ¼ e3LT2 D2L2e
T3�e6D2eT
6 ,
F¼X6l ¼ 1
Xl : ð10Þ
Then, the main results of this paper are presented as follows:
Theorem 1. For given scalars 0ohM , hD and 0ot, the network (8) is asymptotically
synchronized for 0rhðtÞrhM and _hðtÞrhD, if there exist a positive scalar r, positive definite
matrices P1, Q1, Q2, R, S1, S2, positive diagonal matrices D1, D2, and any matrix T satisfying
the following LMIs for 1riojrN:
P1�rInr0, ð11Þ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201704
Author's personal copy
R T
% R
� �Z0, ð12Þ
fððj�iÞUijÞ?gTFfððj�iÞUijÞ
?go0, ð13Þ
where Uij and F are defined in Eq. (10).
Proof. Define a matrix U as
U ¼ ½uij �N�N ¼NIN�EN : ð14Þ
Consider the following Lyapunov–Krasovskii’s functional candidate as
V ¼V1 þ V2 þ V3 þ V4, ð15Þ
where
V1 ¼ xT ðtÞðU � P1ÞxðtÞ,
V2 ¼
Z t
t�hðtÞ
xT ðsÞðU �Q1ÞxðsÞ dsþ
Z t
t�hM
xT ðsÞðU �Q2ÞxðsÞ ds,
V3 ¼ hM
Z t
t�hM
Z t
s
gT ðuÞðU � RÞgðuÞ du ds,
V4 ¼
Z t
t�txT ðsÞðU � S1ÞxðsÞ dsþ t
Z t
t�t
Z t
s
gT ðuÞðU � S2ÞgðuÞ du ds:
By the weak infinitesimal operator L in [29] and Lemma 2, the upper bound of LV1 can beobtained as
LV1 ¼ 2xT ðtÞðU � P1ÞgðtÞ þ .T ðtÞðU � P1Þ.ðtÞ
¼ 2xT ðtÞðU � P1ÞgðtÞ þ xT ðt�hðtÞÞðG � GÞT ðU � P1ÞðG � GÞxðt�hðtÞÞ|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}C1
þ rT ðtÞðU � P1ÞrðtÞ|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}C2
r2xT ðtÞðU � P1ÞgðtÞ þ xT ðt�hðtÞÞðGT NG � GT P1GÞxðt�hðtÞÞ
þr xT ðtÞðU �HT1 H1ÞxðtÞ þ xT ðt�hðtÞÞðU �HT
2 H2Þxðt�hðtÞÞ� �
¼X
1riojrN
zTij ðtÞX1zijðtÞ, ð16Þ
where X1 is defined in Eq. (10).Here, by properties of Kronecker product in Lemma 1 and UG ¼GU ¼NG, the term
C1 of LV1 is calculated as
C1 ¼ xT ðt�hðtÞÞðG � GÞT ðUG � P1GÞxðt�hðtÞÞ
¼ xT ðt�hðtÞÞðGT � GT ÞðNG � P1GÞxðt�hðtÞÞ
¼ xT ðt�hðtÞÞðGT NG � GT P1GÞxðt�hðtÞÞ, ð17Þ
and, if P1rrIn, then the upper bound of term C2 of LV1 is calculated as
C2rlmaxðP1ÞfrT ðtÞðU � InÞrðtÞg
rrfxT ðtÞðU �HT1 H1ÞxðtÞ þ xT ðt�hðtÞÞðU �HT
2 H2Þxðt�hðtÞÞg: ð18Þ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1705
Author's personal copy
Calculating LV2 gives
LV2rxT ðtÞðU �Q1ÞxðtÞ�ð1�hDÞxT ðt�hðtÞÞðU �Q1Þxðt�hðtÞÞ
þxT ðtÞðU �Q2ÞxðtÞ�xT ðt�hMÞðU �Q2Þxðt�hMÞ
¼X
1riojrN
zTij ðtÞX2zijðtÞ, ð19Þ
where X2 is defined in Eq. (10).From Lemmas 2 and 3 and Theorem 1 in [30], the calculation of LV3 leads to
LV3 ¼ h2MgT ðtÞðU � RÞgðtÞ
�hM
Z t
t�hðtÞ
gT ðsÞðU � RÞgðsÞ dsþ
Z t�hðtÞ
t�hM
gT ðsÞðU � RÞgðsÞ ds
� �rh2
MgT ðtÞðU � RÞgðtÞ
�
R t
t�hðtÞgðsÞ dsR t�hðtÞ
t�hMgðsÞ ds
24
35T
1
hðtÞh�1M
ðU � RÞ 0nN
%
1
1�hðtÞh�1M
ðU � RÞ
26664
37775R t
t�hðtÞgðsÞ dsR t�hðtÞ
t�hMgðsÞ ds
24
35
¼ h2MgT ðtÞðU � RÞgðtÞ
�X
1riojrN
R t
t�hðtÞZijðsÞ dsR t�hðtÞ
t�hMZijðsÞ ds
24
35T
1
hðtÞh�1M
� �R 0n
%
1
1�hðtÞh�1M
� �R
26664
37775R t
t�hðtÞZijðsÞ dsR t�hðtÞ
t�hMZijðsÞ ds
24
35
rh2MgT ðtÞðU � RÞgðtÞ
�X
1riojrN
R t
t�hðtÞZijðsÞ dsR t�hðtÞ
t�hMZijðsÞ ds
24
35T
R T
% R
� � R t
t�hðtÞZijðsÞ dsR t�hðtÞ
t�hMZijðsÞ ds
24
35
¼ h2MgT ðtÞðU � RÞgðtÞ
�
R t
t�hðtÞgðsÞ dsR t�hðtÞ
t�hMgðsÞ ds
24
35T
ðU � RÞ ðU � TÞ
% ðU � RÞ
" # R t
t�hðtÞgðsÞ dsR t�hðtÞ
t�hMgðsÞ ds
24
35: ð20Þ
Since the following two zero equalities hold from [31]
xðtÞ�xðt�hðtÞÞ�
Z t
t�hðtÞ
gðsÞ ds�
Z t
t�hðtÞ
.ðsÞ dxðsÞ ¼ 0, ð21Þ
xðt�hðtÞÞ�xðt�hM Þ�
Z t�hðtÞ
t�hM
gðsÞ ds�
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ ¼ 0, ð22Þ
we have
�
Z t
t�hðtÞ
gðsÞ ds
� �T
ðU � RÞ
Z t
t�hðtÞ
gðsÞ ds
� �
r�ðxðtÞ�xðt�hðtÞÞÞT ðU � RÞðxðtÞ�xðt�hðtÞÞÞ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201706
Author's personal copy
þ2ðxðtÞ�xðt�hðtÞÞÞT ðU � RÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ
�
Z t
t�hðtÞ
.ðsÞ dxðsÞ
� �T
ðU � RÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ
� �, ð23Þ
�
Z t�hðtÞ
t�hM
gðsÞ ds
� �T
ðU � RÞ
Z t�hðtÞ
t�hM
gðsÞ ds
� �r�ðxðt�hðtÞÞ�xðt�hMÞÞ
TðU � RÞðxðt�hðtÞÞ
�xðt�hMÞÞ þ 2 xðt�hðtÞÞ�xðt�hM Þð ÞTðU � RÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
�
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
� �T
ðU � RÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
� �, ð24Þ
�2
Z t
t�hðtÞ
gðsÞ ds
� �T
ðU � TÞ
Z t�hðtÞ
t�hM
gðsÞ ds
� �r�2ðxðtÞ�xðt�hðtÞÞÞT ðU � TÞðxðt�hðtÞÞ
�xðt�hMÞÞ þ 2ðxðtÞ�xðt�hðtÞÞÞT ðU � TÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
þ 2ðxðt�hðtÞÞ�xðt�hM ÞÞTðU � TÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ
� xðtÞ�xðt�hðtÞÞ�
Z t
t�hðtÞ
gðsÞ ds
� �T
ðU � TÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
� xðt�hðtÞÞ�xðt�hMÞ�
Z t�hðtÞ
t�hM
gðsÞ ds
� �T
ðU � TÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ: ð25Þ
It should be noted that
�
Z t
t�hðtÞ
.ðsÞ dxðsÞ
� �T
ðU � RÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ
� �r0, ð26Þ
�
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
� �T
ðU � RÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
� �r0: ð27Þ
Then, an upper bound of LV3 can be rewritten as
LV3rgT ðtÞðU � h2MRÞgðtÞ�ðxðtÞ�xðt�hðtÞÞÞT ðU � RÞðxðtÞ�xðt�hðtÞÞÞ
�ðxðt�hðtÞÞ�xðt�hMÞÞTðU � RÞðxðt�hðtÞÞ�xðt�hMÞÞ
�2ðxðtÞ�xðt�hðtÞÞÞT ðU � TÞðxðt�hðtÞÞ�xðt�hMÞÞ þ x1ðdxðtÞÞ,
¼X
1riojrN
zTij ðtÞX3zijðtÞ þ x1ðdxðtÞÞ, ð28Þ
where X3 is defined in Eq. (10), and
x1ðdxðtÞÞ ¼ 2ðxðtÞ�xðt�hðtÞÞÞT ðU � RÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ þ 2ðxðt�hðtÞÞ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1707
Author's personal copy
�xðt�hMÞÞTðU � RÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ
þ xðt�hðtÞÞ�xðt�hMÞ þ
Z t�hðtÞ
t�hM
gðsÞ ds
� �T
ðU � TÞ
Z t
t�hðtÞ
.ðsÞ dxðsÞ
þ xðtÞ�xðt�hðtÞÞ þ
Z t
t�hðtÞ
gðsÞ ds
� �T
ðU � TÞ
Z t�hðtÞ
t�hM
.ðsÞ dxðsÞ:
Calculating LV4 obtains
LV4rxT ðtÞðU � S1ÞxðtÞ�xT ðt�tÞðU � S1Þxðt�tÞ
þt2gT ðtÞðU � S2ÞgðtÞ�
Z t
t�tgðsÞ ds
� �T
ðU � S2Þ
Z t
t�tgðsÞ ds
� �rxT ðtÞðU � S1ÞxðtÞ�x
T ðt�tÞðU � S1Þxðt�tÞþt2gT ðtÞðU � S2ÞgðtÞ�ðxðtÞ�xðt�tÞÞ
TðU � S2ÞðxðtÞ�xðt�tÞÞ þ x2ðdxðtÞÞ
¼X
1riojrN
zTij ðtÞX4zijðtÞ þ x2ðdxðtÞÞ, ð29Þ
where X4 is defined in Eq. (10)
x2ðdxðtÞÞ ¼ 2ðxðtÞ�xðt�tÞÞT ðU � S2Þ
Z t
t�t.ðsÞ dxðsÞ,
and the similar process of obtaining the upper bound of LV3 were utilized in calculation ofan upper bound of LV4.For any diagonal matrices Dk40ðk¼ 1; 2Þ and from Assumption 1, the following
inequalities hold:
0rX
1riojrN
fzTij ðtÞL
T1 D1L1zijðtÞ�f T ðzijðtÞÞD1f ðzijðtÞÞg
¼X
1riojrN
zTij ðtÞX5zijðtÞ, ð30Þ
0rX
1riojrN
fzTij ðt�hðtÞÞLT
2 D2L2zijðt�hðtÞÞ�f T ðzijðt�hðtÞÞÞD2f ðzijðt�hðtÞÞÞg
¼X
1riojrN
zTij ðtÞX6zijðtÞ, ð31Þ
where Xkðk¼ 5; 6Þ are defined in Eq. (10).From Eqs. (16)–(31) and by application of S-procedure [21], LV has a new upper
bound as
LVrX
1riojrN
zTij ðtÞFzijðtÞ þ x1ðdxðtÞÞ þ x2ðdxðtÞÞ, ð32Þ
where F, zijðtÞ and xkðdxðtÞÞðk¼ 1; 2Þ are defined in Eqs. (10), (28), and (29).Also, the network (8) with the augmented vector zijðtÞ can be rewritten asX
1riojrN
ðj�iÞUijzijðtÞ ¼ 0, ð33Þ
where Uij is defined as in Eq. (10).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201708
Author's personal copy
Here, in order to illustrate the process of obtaining Eq. (33), let us define
L¼ ½L1,L2, . . . ,LN � ¼ ½N,N�1, . . . ,1� � In 2 Rn�nN , ð34Þ
where Lkðk¼ 1, . . . ,NÞ 2 Rn�n.By Eqs. (8) and (14) and properties of Kronecker product in Lemma 1, we have the
following zero equality:
0¼LðU � InÞUzðtÞ
¼LðU � InÞ½0nN ,�ðIN � AÞ,ðG � GÞ,0nN ,ðIN �W1Þ,ðIN �W2Þ,�InN �zðtÞ¼L½0nN ,�ðUIN � InAÞ,ðUG � InGÞ,0nN ,ðUIN � InW1Þ,ðUIN � InW2Þ,�ðU � InÞ�zðtÞ
¼L½0nN ,�ðU � AÞ,ðNG � GÞ,0nN ,ðU �W1Þ,ðU �W2Þ,�ðU � InÞ�zðtÞ¼�LðU � AÞxðt�tÞ þ LðNG � GÞxðt�hðtÞÞ þ LðU �W1ÞfðxðtÞÞ
þLðU �W2Þfðxðt�hðtÞÞÞ�LðU � InÞgðtÞ, ð35Þ
where U and zðtÞ are defined in Eq. (10).By Lemma 2, the first term of Eq. (35) can be obtained as
�LðU � AÞxðt�tÞ ¼ �½NIn,ðN�1ÞIn, . . . ,In�|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}n�nN
ðU � AÞ|fflfflfflfflffl{zfflfflfflfflffl}nN�nN
½x1ðt�tÞ, . . . ,xN ðt�tÞ�T|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}nN�1
¼X
1riojrN
uijðLi�LjÞAðxiðt�tÞ�xjðt�tÞÞ
¼�X
1riojrN
ðLi�LjÞAðxiðt�tÞ�xjðt�tÞÞ
¼�X
1riojrN
ððN þ 1�iÞInÞ�ðN þ 1�jÞInÞÞAðxiðt�tÞ�xjðt�tÞÞ
¼�X
1riojrN
ðj�iÞInAðxiðt�tÞ�xjðt�tÞÞ,
¼�X
1riojrN
ðj�iÞAðxiðt�tÞ�xjðt�tÞÞ: ð36Þ
Similarly, the other terms of Eq. (35) are calculated as
LðNG � GÞxðt�hðtÞÞ ¼ ½NIn, . . . ,In�ðNG � GÞ½x1ðt�hðtÞÞ, . . . ,xNðt�hðtÞÞ�T
¼�X
1riojrN
NgijðLi�LjÞGðxiðt�hðtÞÞ�xjðt�hðtÞÞÞ
¼ �X
1riojrN
ðj�iÞðNgijGÞðxiðt�hðtÞÞ�xjðt�hðtÞÞÞ, ð37Þ
LðU �W1ÞfðxðtÞÞ ¼ ½NIn, . . . ,In�ðU �W1Þ½f ðx1ðtÞÞ, . . . ,f ðxNðtÞÞ�T
¼�X
1riojrN
uijðLi�LjÞW1ðf ðxiðtÞÞ�f ðxjðtÞÞÞ
¼X
1riojrN
ðj�iÞW1ðf ðxiÞÞ�f ðxjðtÞÞÞ, ð38Þ
LðU �W2Þfðxðt�hðtÞÞÞ ¼ ½NIn, . . . ,In�ðU �W2Þ½f ðx1ðt�hðtÞÞÞ, . . . ,f ðxN ðt�hðtÞÞÞ�T
¼�X
1riojrN
uijðLi�LjÞW2ðf ðxiðt�hðtÞÞÞ�f ðxjðt�hðtÞÞÞÞ
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1709
Author's personal copy
¼X
1riojrN
ðj�iÞW2ðf ðxiðt�hðtÞÞÞ�f ðxjðt�hðtÞÞÞÞ, ð39Þ
�LðU � InÞgðtÞ ¼�½NIn, . . . ,In�ðU � InÞ½Z1ðtÞ, . . . ,ZN ðtÞ�T
¼X
1riojrN
uijðLi�LjÞInðZiðtÞÞ�ZjðtÞÞ
¼ �X
1riojrN
ðj�iÞInðZiðtÞ�ZjðtÞÞ: ð40Þ
Then, Eq. (35) can be rewritten as
0¼LðU � InÞUzðtÞ¼ �LðU � AÞxðt�tÞ þ LðNG � GÞxðt�hðtÞÞ þ LðU �W1ÞfðxðtÞÞ
þLðU �W2Þfðxðt�hðtÞÞÞ�LðU � InÞgðtÞ
¼ �X
1riojrN
ðj�iÞAðxiðt�tÞ�xjðt�tÞÞ
�X
1riojrN
ðj�iÞðNgijGÞðxiðt�hðtÞÞ�xjðt�hðtÞÞÞ
þX
1riojrN
ðj�iÞW1ðf ðxiðtÞÞ�f ðxjðtÞÞÞ
þX
1riojrN
ðj�iÞW2ðf ðxiðt�hðtÞÞÞ�f ðxjðt�hðtÞÞÞÞ
�X
1riojrN
ðj�iÞInðZiðtÞ�ZjðtÞÞ
¼X
1riojrN
ðj�iÞUijzijðtÞ, ð41Þ
where Uij and zijðtÞ are defined in Eq. (10).Therefore, if Eq. (33) holds, then, a synchronization condition for network (8) isX
1riojrN
zTij ðtÞFzijðtÞo0 subject to
X1riojrN
ðj�iÞUijzijðtÞ ¼ 0, ð42Þ
where F is defined in Eq. (10).Here, if the inequality (42) holds, then there exist a positive scalar E such that
Fo�EI7n: ð43Þ
From Eqs. (32), (42), and (43), by utilization of the mathematical expectation on Eq. (32),we have
EfLVgrEX
1riojrN
zTij ðtÞFzijðtÞ þ x1ðdxðtÞÞ þ x2ðdxðtÞÞ
( )
oEX
1riojrN
zTij ðtÞð�EI7nÞzijðtÞ
( )
rEX
1riojrN
zTij ðtÞð�EInÞzijðtÞ
( )
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201710
Author's personal copy
¼ EX
1riojrN
ð�EJzijðtÞJ2Þ
( )
¼ EX
1riojrN
ð�EJxiðtÞ�xjðtÞJ2Þ
( ), ð44Þ
where EfxkðdxðtÞÞg ¼ 0ðk¼ 1; 2Þ according to work [5].By Lyapunov theorem and Definition 1, it can be guaranteed that the subnetworks in the
coupled stochastic neural networks (8) are asymptotically synchronized.Finally, by use of Lemma 4, the condition (42) is equivalent to the following inequality:X
1riojrN
fððj�iÞUijÞ?gTFfððj�iÞUijÞ
?go0: ð45Þ
From Eq. (45), if the LMIs (13) satisfy, then stability condition (42) holds. This completesour proof. &
When the information about _hðtÞ is unknown, then by setting Q1 ¼ 0n in the Lyapunov–Krasovskii’s functional (15), the following corollary can be obtained.
Corollary 1. For given scalars 0ohM and 0ot, the network (8) is asymptotically
synchronized for 0rhðtÞrhM , if there exist a positive scalar r, positive definite matrices
P1, Q2, R, S1, S2, positive diagonal matrices D1, D2, and any matrix T satisfying the
following LMIs for 1riojrN:
P1�rInr0, ð46Þ
R T
% R
� �Z0, ð47Þ
fððj�iÞUijÞ?gT Ffððj�iÞUijÞ
?go0, ð48Þ
where Uij is defined in Eq. (10), and F ¼F�e1Q1eT1 þ ð1�hDÞe3Q1eT
3 .
Proof. The proof of Corollary 1 is very similar to the proof of Theorem 1, so it is omitted. &
Remark 2. In order to induce a new zero equality (33), the matrix L in Eq. (34) wasdefined. It is inspired by the concept of scaling transformation matrix. To reduce decisionvariable, Finsler’s lemma (ii) B?
TFB?o0 without free-weighting matrices was used. Atthis time, a zero equality is required. If the matrix L is not considered, then the followingdescription (see only Eq. (36) as example):
�f gðU � AÞxðt�tÞ ¼ �f gðU � AÞ½x1ðt�tÞ, . . . ,xN ðt�tÞ�T
¼X
1riojrN
f�gAðxiðt�tÞ�xjðt�tÞÞ ð49Þ
does not hold. Thus, the derivation of zero equality in Eq. (33) is impossible. Here, in orderto use Lemma 2, a suitable vector or matrix in the empty parentheses f g is needed.Therefore, by defining the matrix L, the induction of the zero equality (33) is possible.
Remark 3. The equality LðU � InÞUzðtÞ ¼ 0 in Eq. (35) is similar to the concept ofeigenvectors for the square matrix. For example, the nonzero vector x is an eigenvetor ofthe square matrix A if there is a scale factor l such that Ax¼ lx. The eigenvector of A are
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1711
Author's personal copy
not changed by the matrix A except for the scale factor l. That is, the equalityUzðtÞ ¼LðU � InÞUzðtÞ ¼ 0 with zðtÞa0 can be considered as an operator in the augmentedvector zðtÞ. The augmented vector zðtÞ is also not changed by the operator U except for thematrix LðU � InÞ.
Remark 4. In this paper, the problem of new delay-dependent synchronization criteria forcoupled stochastic neural networks with leakage delay is considered. By using Finsler’slemma without free-weighting matrices, the proposed synchronization criteria for thenetwork are formulated in terms of LMIs. Here, as mentioned in Introduction, the leakagedelay is the time delay in leakage or forgetting term of the systems and a considerablefactor affecting dynamics for the worse in the systems. The effect of the leakage delaywhich cannot be negligible is shown in Fig. 1. In addition, in the field of delay-dependentstability or synchronization analysis, one of major concerns is to get maximum delaybounds with fewer decision variables [7,32,33]. By utilizing Finsler lemma, one caneliminate free variables which were used in zero equalities in the works [7,33]. FromLemma 4, one can check that the B?
TFB?o0 is equivalent to the existence of X such thatFþ XBþ BT X To0 holds. Insertion of such an additional matrix X does not play a role toreduce the conservatism of B?
TFB?o0. It only increases the number of decision variables.Therefore, our proposed synchronization criteria are derived in the form of (ii) in Lemma4. To do this, the new zero equality (33) with Kronecker product technique was devised,which has not been deduced yet in the literature.
4. Numerical examples
In this section, we provide three numerical examples to illustrate the effectiveness of theproposed synchronization criteria in this paper.
Example 1. Consider the following coupled stochastic neural networks with three nodes:
dxiðtÞ ¼ ½�Axiðt�tÞ þW1f ðxiðtÞÞ þW2f ðxiðt�hðtÞÞÞ� dt
þX3j ¼ 1
gijGxjðt�hðtÞÞðdtþ do1ðtÞÞ þ siðt,xiðtÞ,xiðt�hðtÞÞÞ do2ðtÞ, ð50Þ
where
A¼3:5 0
0 1:5
� �, W1 ¼
�1 0:4
0 �0:1
� �, W2 ¼
0:4 �0:6
0:2 0:4
� �,
G¼0:2 0
0 0:2
� �, G¼
�2 1 1
1 �2 1
1 1 �2
264
375,
with
f ðxÞ ¼ 14ðjxþ 1j�jx�1jÞ, L1 ¼ L2 ¼ diagf0:5,0:5g,
sTi ðt,xiðtÞ, xiðt�hðtÞÞÞsiðt,xiðtÞ,xiðt�hðtÞÞÞrJxiðtÞJ
2þ Jxiðt�hðtÞÞJ2:
For the networks above, the results of maximum bounds of time-delay with different hD
and t by Theorem 1 are listed in Table 1. Also, when the value of the time-derivative of
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201712
Author's personal copy
time-delay is unknown, then by applying Corollary 1 to the above system (50), maximumdelay bounds for guaranteeing the asymptotic stability can be obtained as listed in Table 1.In order to confirm the obtained results with the time-varying delay conditions in Table 2,the simulation results for the synchronization errors trajectories, zi1ðtÞ ¼ xiðtÞ�x1ðtÞ
(i¼ 2; 3), of the networks (50) are shown in Figs. 1–3. These figures show that the networkswith the errors converge to zero for given initial values of the state by randomly choosing.Specially, the simulation results in Fig. 1 show synchronization errors trajectories for the
0 1 2 3 4 5−3
−2
−1
0
1
2
3
Time (Seconds)
z(t)
0 1 2 3 4 5−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time (Seconds)
z(t)
0 50 100 150−3
−2
−1
0
1
2
3
Time (Seconds)
z(t)
0 50 100 150−1.5
−1
−0.5
0
0.5
1
1.5
Time (Seconds)
z(t)
0 10 20 30 40 50−150
−100
−50
0
50
100
150
200
Time (Seconds)
z(t)
0 10 20 30 40 50−250
−200
−150
−100
−50
0
50
100
150
200
Time (Seconds)
z(t)
Fig. 1. Synchronization errors trajectories with C1-1: (top) t¼ 0:05, (middle) t¼ 0:48 and (bottom) t¼ 0:49(Example 1).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1713
Author's personal copy
values of leakage delay, t, by 0.05, 0.48 and 0.49 with fixed values hM¼1.4092 and hD¼0.2.It is easy to illustrate that the larger value of leakage delay gives the worse dynamicbehaviors of the networks.
Example 2. Consider the following coupled stochastic neural networks with five nodes:
dxiðtÞ ¼ ½�Axiðt�tÞ þW1f ðxiðtÞÞ þW2f ðxiðt�hðtÞÞÞ� dt
þX5j ¼ 1
gijGxjðt�hðtÞÞðdtþ do1ðtÞÞ þ siðt,xiðtÞ,xiðt�hðtÞÞÞ do2ðtÞ, ð51Þ
Table 1
Maximum bounds hM of time-delay with different hD and t (Example 1).
t hD
0 0.2 0.5 0.8 Unknown
0.05 1.6602 1.4092 1.1565 0.9849 0.8826
0.1 1.4844 1.2572 1.0300 0.8411 0.7662
0.2 1.1043 0.8767 0.4820 0.4243 0.4243
0.29 0.0504 0.0436 0.0436 0.0436 0.0436
0.3 Infeasible
Table 2
The conditions of simulation in Example 1.
No. t hM hD h(t)
0.05
C1-1 0.48 1.4092 0.2 1:4092 sin2ð0:14tÞ
0.49
C1-2 0.2 0.4820 0.5 0:4820 sin2ð1:03tÞ
C1-3 0.29 0.0436 Unknown 0:0436jsinðtÞj
0 1 2 3 4 5−3
−2
−1
0
1
2
3
Time (Seconds)
z(t)
0 1 2 3 4 5−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time (Seconds)
z(t)
Fig. 2. Synchronization errors trajectories with C1-2 (Example 1).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201714
Author's personal copy
where
A¼1:4 0
0 1
� �, W1 ¼
�1:5 0:4
0:1 �0:2
� �, W2 ¼
0:5 �0:2
0:1 �0:5
� �,
G¼0:1 0:1
0 0:2
� �, G¼
�2 1 0 0 1
1 �3 1 1 0
0 1 �2 1 0
0 1 1 �3 1
1 0 0 1 �2
26666664
37777775,
with
f ðxÞ ¼ 110ðjxþ 1j�jx�1jÞ,L1 ¼ L2 ¼ diagf0:2,0:2g,
sTi ðt,xiðtÞ,xiðt�hðtÞÞÞsiðt,xiðtÞ,xiðt�hðtÞÞÞrJxiðtÞJ
2þ Jxiðt�hðtÞÞJ2:
The results of maximum bounds of time-delay with different hD and fixed t¼ 0:01 byTheorem 1 and Corollary 1 are listed in Table 3. Figs. 4–7 show for the state responses,xiðtÞði¼ 1; 2,3; 4,5Þ, and the synchronization errors trajectories, zi1ðtÞ ¼ xiðtÞ�x1ðtÞ
(i¼ 2; 3,4; 5) of system (51) with the conditions of time-delay in Table 4 when initialvalues of the state are xT
1 ð0Þ ¼ ½1,�1�, xT2 ð0Þ ¼ ½�1:5,1:2�, xT
3 ð0Þ ¼ ½2:0,�1:6�, xT4 ð0Þ ¼
½0:7,1:1� and xT5 ð0Þ ¼ ½�1:2,�1:8�. Figures show that the system (51) with the errors
converge to zero. This implies the synchronization of the network (51).
0 1 2 3 4 5−3
−2
−1
0
1
2
3
Time (Seconds)
z(t)
0 1 2 3 4 5−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time (Seconds)
z(t)
Fig. 3. Synchronization errors trajectories with C1-3 (Example 1).
Table 3
Maximum bounds hM of time-delay with different hD and fixed t¼ 0:01 (Example 2).
hD 0 0.2 0.5 0.8 Unknown
hM 0.5847 0.4827 0.3712 0.3712 0.3712
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1715
Author's personal copy
0 0.5 1 1.5 2 2.5 3−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
Time (Seconds)
x(t)
Fig. 4. State responses with C2-1 (Example 2).
0 0.5 1 1.5 2 2.5 3−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−2.5
−2
−1.5
−1
−0.5
0
0.5
Time (Seconds)
z(t)
Fig. 5. Synchronization errors trajectories with C2-1 (Example 2).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201716
Author's personal copy
0 0.5 1 1.5 2 2.5 3−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
x(t)
Fig. 6. State responses with C2-2 (Example 2).
0 0.5 1 1.5 2 2.5 3−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
z(t)
0 0.5 1 1.5 2 2.5 3−2.5
−2
−1.5
−1
−0.5
0
Time (Seconds)
z(t)
Fig. 7. Synchronization errors trajectories with C2-2 (Example 2).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1717
Author's personal copy
Example 3. Recall the coupled stochastic neural networks in Example 1 with the followingchanged parameters:
G¼ 2In, H1 ¼ 4In, H2 ¼H1:
The result of maximum bound of time-delay with hD¼0.2 and t¼ 0:05 by Theorem 1 is0.0874. With the condition of time-varying delay C3 as hðtÞ ¼ 0:0874 sin2ð2:28tÞ and
Table 4
The conditions of simulation in Example 2.
No. t hM hD h(t)
C2-1 0.01 0.4827 0.2 0:2 sinðtÞ þ 0:2827C2-2 0.01 0.3712 Unknown 0:3712jsinðtÞj
0 0.2 0.4 0.6 0.8 1−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
Time (Seconds)
x(t)
Fig. 8. State responses with C3 (Example 3).
0 0.2 0.4 0.6 0.8 1−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
Time (Seconds)
z(t)
0 0.2 0.4 0.6 0.8 1−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Time (Seconds)
z(t)
Fig. 9. Synchronization errors trajectories with C3 (Example 3).
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201718
Author's personal copy
t¼ 0:05, the simulation results for the state responses, xi(t) (i¼ 1; 2,3), and thesynchronization errors trajectories, zi1ðtÞ ¼ xiðtÞ�x1ðtÞ (i¼ 2; 3), of the networks areshown in Figs. 8 and 9.
5. Conclusions
In this paper, the delay-dependent synchronization criteria for the coupled stochastic neuralnetworks with time-varying delays in network coupling and leakage delay have been proposed.To do this, a suitable Lyapunov–Krasovskii’s functional was used to investigate the feasibleregion of stability criteria. By establishment of a new zero equality and utilization of Finsler’slemma, sufficient conditions for guaranteeing asymptotic synchronization for the concernednetworks have been derived in terms of LMIs. Three numerical examples have been given toshow the effectiveness and usefulness of the presented criteria.
Acknowledgments
This research was supported by Basic Science Research Program through the NationalResearch Foundation of Korea (NRF) funded by the Ministry of Education, Science andTechnology (2011-0001045), and by a grant of the Korea Healthcare Technology R&DProject, Ministry of Health & Welfare, Republic of Korea (A100054).
References
[1] S.H. Strogatz, Exploring complex networks, Nature 410 (2001) 268–276.
[2] S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, D.U. Hwang, Complex networks: structure and dynamics,
Physical Report 424 (2006) 175–308.
[3] J. Cao, G. Chen, P. Li, Global synchronization in an array of delayed neural networks with hybrid coupling,
IEEE Transactions on Systems, Man, and Cybernetics, Part B 38 (2008) 488–498.
[4] W. Yu, J. Cao, G. Chen, J. Lu, J. Han, W. Wei, Local synchronization of a complex network model, IEEE
Transactions on Systems, Man, and Cybernetics, Part B: 39 (2009) 230–241.
[5] S. Xu, J. Lam, X. Mao, Y. Zou, A new LMI condition for delay dependent robust stability of stochastic
time-delay systems, Asian Journal of Control 7 (2005) 419–423.
[6] R. Yang, P. Shi, H. Gao, New delay-dependent stability criterion for stochastic systems with time delays,
IET Control Theory & Applications 2 (2008) 966–973.
[7] O.M. Kwon, Stability criteria for uncertain stochastic dynamic systems with time-varying delays,
International Journal of Robust and Nonlinear Control 21 (2011) 338–350.
[8] S. Lakshmanan, P. Balasubramanian, Linear matrix inequality approach for robust stability analysis for
stochastic neural networks with time-varying delay, Chinese Physics B 20 (2011) 0402041–04020411.
[9] X. Li, J. Cao, Adaptive synchronization for delayed neural networks with stochastic perturbation, Journal of
The Franklin Institute 345 (2008) 779–791.
[10] Y. Tang, Z. Wang, J.-A. Fang, Pinning control of fractional-order weighted complex networks, Chaos
19 (2009) 013112.
[11] Y. Tang, S.Y.S. Leung, W.K. Wong, J.-A. Fang, Impulsive pinning synchronization of stochastic discrete-
time networks, Neurocomputing 73 (2010) 2132–2139.
[12] G. Wang, J. Cao, L. Wang, Global dissipativity of stochastic neural networks with time delay, Journal of the
Franklin Institute 346 (2009) 794–807.
[13] X. Li, C. Ding, Q. Zhu, Synchronization of stochastic perturbed chaotic neural networks with mixed delays,
Journal of The Franklin Institute 347 (2010) 1266–1280.
[14] Q. Zhu, X. Yang, H. Wang, Stochastically asymptotic stability of delayed recurrent neural networks with
both Markovian jump parameters and nonlinear disturbances, Journal of The Franklin Institute 347 (2010)
1489–1510.
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–1720 1719
Author's personal copy
[15] Y. Tang, Z. Wang, W.K. Wong, J. Kurths, J.-A. Fang, Multiobjective synchronization of coupled systems,
Chaos 21 (2011) 025114.
[16] Z. Xiang, C. Qiao, M.S. Mahmoud, Finite-time analysis and H1 control for switched stochastic systems,
Journal of The Franklin Institute (2011), doi:10.1016/j.jfranklin.2011.10.021.
[17] H. Li, D. Yue, Synchronization of Markovian jumping stochastic complex networks with distributed time
delays and probabilistic interval discrete time-varying delays, Journal of Physics A: Mathematics and
Theoretical 43 (2010) 105101.
[18] P. Balasubramanian, G. Nagamani, R. Rakkiyappan, Passivity analysis for neural networks of neutral type
with Markovian jumping parameters and time delay in the leakage term, Communications in Nonlinear
Science and Numerical Simulation 16 (2011) 4422–4437.
[19] X. Li, J. Cao, Delay-dependent stability of neural networks of neutral type with time delay in the leakage
term, Nonlinearity 23 (2010) 1709–1726.
[20] S. Xu, J. Lam, A survey of linear matrix inequality techniques in stability analysis of delay systems,
International Journal of Systems Science 39 (2008) 1095–1113.
[21] S. Boyd, L.E. Ghaoui, E. Feron, V. Balakrishnan, Linear Matrix Inequalities in System and Control Theory,
SIAM, Philadelphia, 1994.
[22] P. Morters, Y. Peres, Brownian Motion. Cambridge Series in Statistical and Probabilistic Mathematics,
Cambridge University Press, UK, 2010.
[23] X. Mao, Stochastic Differential Equations and Their Applications, Horwood, Chichester, UK, 1997.
[24] J. Lu, D.W.C. Ho, Z. Wang, Pinning stabilization of linearly coupled stochastic neural networks via
minimum number of controllers, IEEE Transactions on Neural Networks 20 (2009) 1617–1629.
[25] A. Graham, Kronecker Products and Matrix Calculus: With Applications, John Wiley & Sons, Inc.,
New York, 1982.
[26] X. Liu, T. Chen, Exponential synchronization of the linearly coupled dynamical networks with delays,
Chinese Annals of Mathematics 28B (2007) 737–746.
[27] K. Gu, An integral inequality in the stability problem of time-delay systems, in: Proceedings of 39th IEEE
Conference on Decision Control, Sydney, Australia, December 2000, pp. 2805–2810.
[28] M.C. de Oliveira, R.E. Skelton, Stability Tests for Constrained Linear Systems, Springer-Verlag, Berlin,
2001, pp. 241–257.
[29] R.Z. Khasminskii, Stochastic Stability of Differential Equations, Alphen aan den Rjin, Sjithoffand Noor,
Khasminskiidhoff, 1980.
[30] P. Park, J.W. Ko, C.K. Jeong, Reciprocally convex approach to stability of systems with time-varying delays,
Automatica 47 (2011) 235–238.
[31] L. Arnold, Stochastic Differential Equations: Theory and Applications, Wiley, New York, 1972.
[32] T. Li, L. Guo, L. Wu, Simplified approach to the asymptotical stability of linear systems with interval time-
varying delay, IET Control Theory & Applications 3 (2009) 252–260.
[33] O.M. Kwon, Ju.H. Park, S.M. Lee, Secure communication based on chaotic synchronization via interval
time-varying delay feedback control, Nonlinear Dynamics 63 (2011) 239–252.
M.J. Park et al. / Journal of the Franklin Institute 349 (2012) 1699–17201720