Electronic Journal of Differential Equations, Vol. 2007(2007), No. 50, pp. 1–16. ISSN: 1072-6691. URL: http://ejde.math.txstate.edu or http://ejde.math.unt.edu ftp ejde.math.txstate.edu (login: ftp) BEHAVIOR OF IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS WITH DISTRIBUTED DELAYS KELIN LI, ZUOAN LI, QIANKUN SONG Abstract. In this paper, we investigate a generalized model of fuzzy cellular neural networks with distributed delays and impulses. By employing the theory of topological degree, M -matrix and Lypunov functional, we find sufficient conditions for the existence, uniqueness and global exponential stability of both the equilibrium point and the periodic solution. Two examples are given to illustrate the results obtained here. 1. Introduction Since cellular neural networks (CNN) was introduced by Chua and Yang in [7, 8], many researchers have done extensive works on this subject due to their applications in classification of patterns, associative memories, image processing, quadratic optimization, and other areas, e.g., [2, 5, 6, 18, 19, 25]. However, in mathematical modelling of real world problems, we encounter inconveniences, namely, the complexity and the uncertainty or vagueness. In order to take vagueness into consideration, fuzzy theory is considered as a suitable setting. Based on traditional CNN, Yang and Yang proposed the fuzzy cellular neural networks (FCNN) [22, 23], which integrates fuzzy logic into the structure of the traditional CNN and maintains local connectedness among cells. Unlike previous CNN structures, FCNN has fuzzy logic between its template input and/or output besides the sum of product operation. FCNN is very useful paradigm for image processing problems, which is a cornerstone in image processing and pattern recognition. In such applications, it is of prime importance to ensure that the designed FCNN be stable. In [22, 23], the authors have obtained some conditions for the existence and the global stability of the equilibrium point of FCNN without delays. In [16], Liu and Tang have considered FCNN with either constant delays or time-varying delays, several sufficient conditions have been obtained to ensure the existence and uniqueness of the equilibrium point and its global exponential stability. Yuan, Cao and Deng have given several novel criteria of exponential stability and periodic solutions for FCNN with time-varying delays [24]. Recently, Huang has considered the stability of FCNN with diffusion terms and time-varying delay [14], at the same time, Huang has investigated the exponential stability of FCNN with distributed delay [13]. 2000 Mathematics Subject Classification. 92B20, 34K20, 34K13. Key words and phrases. Fuzzy cellular neural networks; impulses; distributed delays; global exponential stability; periodic oscillatory solution. c 2007 Texas State University - San Marcos. Submitted October 2, 2006. Published April 5, 2007. 1 2 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 However, besides delay effect, impulsive effect likewise exists in a wide variety of evolutionary processes in which states are changed abruptly at certain moments of time, involving such fields as medicine and biology, economics, mechanics, electronics and telecommunications, etc. Many interesting results on impulsive effect have been gained, e.g., Refs. [1, 3, 10, 11, 15, 17, 20, 21]. As artificial electronic systems, neural networks such as CNN, bidirectional neural networks and recurrent neural networks often are subject to impulsive perturbations which can affect dynamical behaviors of the systems just as time delays. Therefore, it is necessary to consider both impulsive effect and delay effect on the stability of neural networks. Motivated by the above discussions, in this paper, on the basis of the structure of FCNN, we consider a class of impulsive fuzzy neural networks with distributed delays described by the following system of integro-differential equations: n n X X dxi (t) ãij uj (t) + Ii (t) aij fj (xj (t)) + = −di xi (t) + dt j=1 j=1 Z t Z t n n ^ _ bij + Kij (t − s)fj (xj (s))ds + b̃ij Kij (t − s)fj (xj (s))ds + j=1 n ^ j=1 −∞ Tij uj (t) + j=1 n _ Hij uj (t), −∞ t 6= tk , j=1 − ∆xi (tk ) = xi (t+ k ) − xi (tk ) = ∆k (xi (tk )), t = tk , (1.1) for i = 1, 2, . . . , n. Where the fixed times tk satisfy t1 < t2 < . . . , limk→∞ tk = ∞. The first part (called continuous part) of model (1.1) describes the continuous processes of FCNN. n corresponds to the number of units in the neural network; xi corresponds to the state variable; fj (xj (t)) denotes the activation function of the jth neuron; ui and Ii (t) denote input and bias of the ith neuron, respectively. di represents the rate with which the ith unit will reset its potential to the resting state in isolation when disconnected from the networks and external inputs; aij and ãij are elements of feedback template and feedforward template; bij , b̃ij are elements of the distributed delay fuzzy feedback MIN template, the distributed delay fuzzy feedback MAX template, Tij and Hij are elements of fuzzy feedforward MIN template and fuzzy MAX template, respectively; Kij corresponds V feedforward W to the delay kernel. and denote the fuzzy AND and fuzzy OR operation, respectively. The second part (called discrete part) of model (1.1) describes that the evolution processes experience abrupt change of states at the moments of time tk (called impulsive moments). ∆xi (tk ) represents impulsive perturbations of the ith unit at time tk , and ∆k denotes the impulsive operator at time tk for k = 1, 2, . . . . To the best of our knowledge, few authors has considered dynamical behaviors of impulsive fuzzy neural networks with distributed delays. This paper studies the existence, uniqueness and global exponential stability of both the equilibrium point and the periodic solution for impulsive fuzzy neural networks with distributed delays. Several sufficient conditions ensuring the existence, uniqueness and global exponential stability of both the equilibrium point and the periodic solution for impulsive fuzzy neural networks with distributed delays will be established for the system (1.1). EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 3 The remainder part of this paper is organized as follows. some notations and preliminaries are given in section 2. In section 3, several sufficient conditions will be established ensuring model (1.1) to the existence, uniqueness and global exponential stability of equilibrium point. The existence, uniqueness and global exponential stability of the system (1.1) will be given in section 4. In section 5, two examples are given to illustrate our theory. 2. Preliminaries Throughout this paper we assume the following hypotheses: (H1) There exist constant scalers Fi > 0 such that |fi (x) − fi (y)| ≤ Fi |x − y|, i = 1, 2, . . . , n for any x, y ∈ R. (H2) The delay kernels Kij : [0, +∞) → [0, +∞) are piecewise continuous functionsRand satisfies: ∞ (i) 0 Kij (s)ds = 1, i, j = 1, 2, . . . , n. R∞ (ii) 0 sKij (s)ds < ∞, i, j = 1, 2, . . . , n. (iii) There exists a positive number µ such that Z ∞ seµs Kij (s)ds < ∞, i, j = 1, 2, . . . , n. 0 Let C = C((−∞, 0], Rn ) be the linear space of bounded and continuous functions which map (−∞, 0] into Rn . The initial conditions associated with model (1.1) are of the form xi (t) = ϕi (t), −∞ < t ≤ 0 (2.1) in which ϕi (·) are bounded continuous (i = 1, 2, . . . , n). First, we introduce some notation and recall some basic definitions. For an n × n matrix, |A| denotes the absolute value matrix given by |A| = (|aij |)n×n ; A−1 denotes the inverse of A. Let A, B be two n × n matrices, A > B represents aij > bij for all i, j = 1, 2, . . . , n. Let a vector norm kxkp (p = 1, ∞) (simply denoted by kxk) for x ∈ Rn be defined as kxk1 = n X |xi |, kxk∞ = max |xi |. i=1 1≤i≤n For ϕ ∈ C, kϕk∞ is defined as kϕk∞ = sup −∞<s≤0 kϕ(s)k∞ = sup max |ϕi (s)|. −∞<s≤0 1≤i≤n Definition 2.1. A function x : (−∞, +∞) → Rn is said to be the special solution of system (1.1) with initial condition (2.1) if the following two conditions are satisfied (i) x is piecewise continuous with first kind discontinuity at the points tk , k = 1, 2, . . . . Moreover, x is left continuous at each discontinuity point. (ii) x satisfies model (1.1) for t ≥ 0, and x(s) = ϕ(s) for s ∈ (−∞, 0]. Especially, a point x∗ ∈ Rn is called an equilibrium point of model (1.1), if x(t) = x∗ is a solution of (1.1). Henceforth, we let x(t, ϕ) denote the special solution of (1.1) with initial condition ϕ ∈ C. 4 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 Definition 2.2. The periodic solution x(t, ϕ) of system (1.1) is said to be globally exponentially stable, if there exist positive constants α and M such that every solution x(t, φ) of (2.1) satisfies kx(t, φ) − x(t, ϕ)k∞ ≤ M kφ − ϕk∞ e−αt , ∀t ≥ 0. Definition 2.3 ([4]). A real matrix D = (dij )n×n is said to be a non-singular M -matrix if aij ≤ 0, i, j = 1, 2, . . . , n, i 6= j, and all successive principal minors of D are positive. For the non-singular M -matrix, we have the following result. Lemma 2.4 ([4]). Each of the following conditions is equivalent: (i) D is a nonsingular M -matrix. (ii) D−1 exists and D−1 is a nonnegative matrix. (iii) The diagonal elements of D are all positive and there exists a positive vector d such that Dd > 0 or DT d > 0. Lemma 2.5 ([22]). Suppose y and ȳ are two state of model (1.1), then we have n n n X ^ ^ α f (ȳ ) ≤ α f (y ) − αij · fj (yj ) − fj (ȳj ), ij j j ij j j j=1 j=1 j=1 n n n X _ _ βij fj (ȳj ) ≤ βij fj (yj ) − βij · fj (yj ) − fj (ȳj ). j=1 j=1 j=1 3. Global exponential stability of equilibrium point In this section, we will give several sufficient conditions on the global exponential stability of equilibrium point for the impulsive FCNN with distributed delays. Consider the case of model (1.1) as Ii (t) = Ii , ui (t) = ui , i = 1, 2, . . . , n, and let Pn Vn Wn I˜i = j=1 ãij uj + Ii + j=1 Tij uj + j=1 Hij uj , then model (1.1) becomes Z t n n ^ X dxi (t) bij aij fj (xj (t)) + Kij (t − s)fj (xj (s))ds = −di xi (t) + dt −∞ j=1 j=1 Z t n _ (3.1) + b̃ij Kij (t − s)fj (xj (s))ds + I˜i , t 6= tk , j=1 −∞ − ∆xi (tk ) = xi (t+ k ) − xi (tk ) = ∆k (xi (tk )), t = tk , for i = 1, 2, . . . , n. Theorem 3.1. Under assumptions (H1), (H2), the first equation in system (3.1) has a unique equilibrium point if D − (|A| + |B| + |B̃|)F is a nonsingular M matrix, where D = diag(d1 , d2 , . . . , dn ), A = (aij )n×n , B = (bij )n×n , B̃ = (b̃ij )n×n , F = diag(F1 , F2 , . . . , Fn ). Proof. Let x∗ = (x∗1 , x∗2 , . . . , x∗n )T denote an equilibrium point of the first equation in model (3.1). Then x∗ satisfies di x∗i − n X j=1 aij fj (x∗j ) − n ^ j=1 bij fj (x∗j ) − n _ j=1 b̃ij fj (x∗j ) − I˜i = 0, i = 1, 2, . . . , n. (3.2) EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 5 Let hi (xi ) = di xi − n X aij fj (xj )− j=1 n ^ bij fj (xj )− j=1 n _ b̃ij fj (xj )− I˜i = 0, i = 1, 2, . . . , n. j=1 Obviously, the solutions of the above system are the equilibrium point of model (3.1). Let us define homotopic mapping H(x, λ) = λh(x) + (1 − λ)x, where λ ∈ [0, 1], and h(x) = (h1 (x1 ), h2 (x2 ), . . . , hn (xn ))T , H(x, λ) = (H1 (x1 , λ), H2 (x2 , λ), . . . , Hn (xn , λ))T , then for i ∈ {1, 2, . . . , n}, from (H1) and Lemma 2.5, we have |Hi (x, λ)| n n n h i X ^ _ = λ di xi − aij fj (xj ) − bij fj (xj ) − b̃ij fj (xj ) − I˜i + (1 − λ)xi j=1 j=1 ≥ |λdi xi + (1 − λ)xi | − λ n X j=1 |aij ||fj (xj )| − λ j=1 −λ n X n X |bij ||fj (xj )| j=1 |b̃ij ||fj (xj )| − λ|I˜i | j=1 ≥ [1 + λ(di − 1)]|xi | − λ n X |aij |Fj |xj | − λ − λ |I˜i | + n X |aij ||fj (0)| + j=1 n X |bij |fj (0)| + j=1 = [1 + λ(di − 1)]|xi | − λ n X |bij |Fj |xj | − λ j=1 j=1 h n X n X n X |b̃ij |Fj |xj | j=1 i |b̃ij |fj (0)| . j=1 Fj |xj | |aij | + |bij | + |b̃ij | j=1 h − λ |I˜i | + n X i |fj (0)| |aij | + |bij | + |b̃ij | . j=1 Since D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix, there exist constants li > 0 such that n X li di − Fi lj |aji | + |bji | + |b̃ji | > 0, i = 1, 2, . . . , n, j=1 then, we have n X li |Hi (x, λ)| i=1 ≥ n X i=1 li (1 − λ)|xi | + λ n h X i=1 di li |xi | − li n X j=1 i Fj |xj | |aij | + |bij | + |b̃ij | 6 KELIN LI, ZUOAN LI, QIANKUN SONG −λ n n h i X X li |I˜i | + |fj (0)| |aij | + |bij | + |b̃ij | i=1 ≥λ n h X j=1 n X di li |xi | − li i=1 n X −λ h li |I˜i | + n h X i Fj |xj | |aij | + |bij | + |b̃ij | j=1 n X i |fj (0)| |aij | + |bij | + |b̃ij | i=1 =λ EJDE-2007/50 j=1 n i X l i d i − Fi lj |aji | + |bji | + |b̃ji | |xi | i=1 n X −λ h li |I˜i | + j=1 n X i=1 i |fj (0)| |aij | + |bij | + |b̃ij | j=1 ≥ λl0 kxk1 − λnI0 . Define n n o X lj |aji | + |bji | + |b̃ji | , l0 = min li di − Fi 1≤i≤n j=1 n n o X |fj (0)| |aij | + |bij | + |b̃ij | , I0 = max li |I˜i | + 1≤i≤n j=1 and let n(I0 + 1) . Γ = x : kxk1 ≤ l0 Then it follows that kxk1 = n(I0 + 1)/l0 for any x ∈ ∂Γ, and n X li |Hi (x, λ)| ≥ λl0 j=1 n(I0 + 1) − λnI0 > 0, l0 ∀λ ∈ (0, 1], that is F (x, λ) 6= 0, for any x ∈ ∂Γ, λ ∈ (0, 1]. Also, as λ = 0, H(x, λ) = id (x) = x 6= 0, for any x ∈ ∂Γ, here, id is identity mapping. Hence, we have H(x, λ) 6= 0, for any x ∈ ∂Γ, λ ∈ [0, 1]. From (H1), it is easy to prove deg(id , Γ, 0) = 1 thus we have from homotopy invariance theorem [9] that deg(h, Γ, 0) = deg(id , Γ, 0) = 1. By the topological degree theory, we can conclude that (3.1) has at least one solution in Γ. That is, model (3.1) has at least an equilibrium point. Suppose y ∗ = (y1∗ , y2∗ , . . . , yn∗ )T is also an equilibrium point of model (3.1), then we have di x∗i − di yi∗ − n X j=1 n X j=1 aij fj (x∗j ) − aij fj (yj∗ ) − n ^ j=1 n ^ j=1 bij fj (x∗j ) − bij fj (yj∗ ) − n _ j=1 n _ j=1 b̃ij fj (x∗j ) − I˜i = 0, b̃ij fj (yj∗ ) − I˜i = 0, EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 7 this implies di (x∗i − yi∗ ) = n X aij (fj (x∗j ) − fj (yj∗ )) + j=1 n _ b̃ij fj (x∗j ) − j=1 n X n _ n ^ bij fj (yj∗ ) j=1 b̃ij fj (yj∗ ) j=1 |aij ||fj (x∗j ) − fj (yj∗ )| + | j=1 +| bij fj (x∗j ) − j=1 + ≤ n ^ n ^ bij fj (x∗j ) − j=1 n _ b̃ij fj (x∗j ) − j=1 n _ n ^ bij fj (yj∗ )| j=1 b̃ij fj (yj∗ )| j=1 for i = 1, 2, . . . , n. By using (H1) and Lemma 2.5, we have di |x∗i − yi∗ | ≤ n X Fj |x∗j − yj∗ | |aij | + |bij | + |b̃ij | , i = 1, 2, . . . , n, j=1 which can be rewritten as (D − (|A| + |B| + |B̃|)F )(|x∗1 − y1∗ |, |x∗2 − y2∗ |, . . . , |x∗n − yn∗ |)T ≤ 0. Since D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix, (D − (|A| + |B| + |B̃|)F )−1 is a nonnegative matrix. Thus multiplying both sides of the above inequality by (D − (|A| + |B| + |B̃|)F )−1 does not change the inequality direction, it follows that (|x∗1 − y1∗ |, |x∗2 − y2∗ |, . . . , |x∗n − yn∗ |)T ≤ 0. This implies that x∗ = y ∗ . Therefore, the system (3.1) has one unique equilibrium point. Theorem 3.2. Assume that (H1), (H2) hold, and D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix, furthermore, suppose that the impulsive operator ∆k (xi (tk )) satisfies ∆k (xi (tk )) = −δik (xi (tk ) − x∗i ), 0 < δik < 2, Then the equilibrium point x∗ = (x∗1 , x∗2 , . . . , x∗n )T nentially stable. i = 1, 2, . . . , n, k = 1, 2, . . . . (3.3) of system (3.1) is globally expo- Proof. From (3.3), we have ∆k (x∗i ) = 0, and by Theorem 3.1, x∗ = (x∗1 , x∗2 , . . . , x∗n )T is only equilibrium point of the system (3.1). Let x(t) = (x1 (t), x2 (t), . . . , xn (t))T be an arbitrary solution of the system (3.1). From assumption (H1) and Lemma 8 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 2.5, we obtain that d+ |xi (t) − x∗i | dt d(xi (t) − x∗i ) dt n X ≤ −di |xi (t) − x∗i | + |aij ||fj (xj (t)) − fj (x∗j )| = sign(xi (t) − x∗i ) j=1 +| +| n ^ j=1 n _ Z t Kij (t − s)fj (xj (s))ds − bij −∞ Z t Kij (t − s)fj (xj (s))ds − b̃ij −∞ j=1 n ^ j=1 n _ Z n X Kij (t − s)fj (x∗j )ds| −∞ Z t Kij (t − s)fj (x∗j )ds| b̃ij j=1 ≤ −di |xi (t) − x∗i | + t bij −∞ |aij ||fj (xj (t)) − fj (x∗j )| j=1 +| +| n X j=1 n X Z t (3.4) Kij (t − s)|fj (xj (s)) − fj (x∗j )|ds |bij | −∞ Z t |b̃ij | Kij (t − s)|fj (xj (s)) − fj (x∗j )|ds −∞ j=1 ≤ −di |xi (t) − x∗i | + n X |aij |Fj |xj (t) − x∗j | j=1 + n X Z |bij | + |b̃ij | t Kij (t − s)Fj |xj (s) − x∗j |ds −∞ j=1 = −di |xi (t) − x∗i | + n X Fj |aij ||xj (t) − x∗j | j=1 + n X Z Fj |bij | + |b̃ij | +∞ Kij (s)|xj (t − s) − x∗j |ds 0 j=1 for t > 0, i = 1, 2, . . . , n, t 6= tk , k = 1, 2, . . . . Also, ∗ ∗ ∗ ∗ xi (t+ k ) − xi = −δik (xi (tk ) − xi ) + xi (tk ) − xi = (1 − δik )(xi (tk ) − xi ) for i = 1, 2, . . . , n, k = 1, 2, . . . . Hence ∗ ∗ ∗ |xi (t+ k ) − xi | ≤ |1 − δik ||xi (tk ) − xi | ≤ |xi (tk ) − xi | (3.5) for i = 1, 2, . . . , n, k = 1, 2, . . . . Since D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix, there exist constants li > 0 such that li di − Fi n X lj |aji | + |bji | + |b̃ji | > 0, j=1 i = 1, 2, . . . , n. (3.6) EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 9 Now, for i = 1, 2, . . . , n, we define the functions n h Z X h̃i (αi ) = li (di − αi ) − Fi lj |aji | + |bji | + |b̃ji | +∞ i eαi s Kji (s)ds , 0 j=1 where αi ∈ [0, +∞). Obviously, for i ∈ {1, 2, . . . , n}, h̃i (αi ) are continuous on [0, +∞), and from (3.6), we know that h̃i (0) > 0, for i ∈ {1, 2, . . . , n}. Also, for i ∈ {1, 2, . . . , n}, we have h̃i (αi ) → −∞ as αi → +∞. So there exists αi∗ such that h̃i (αi∗ ) = 0, i ∈ {1, 2, . . . , n}. Let α = min{α1 , α2 , . . . , αn }, we get n h Z +∞ i X h̃i (α) = li (di − α) − Fi lj |aji | + |bji | + |b̃ji | eαi s Kji (s)ds ≥ 0 (3.7) 0 j=1 for i = 1, 2, . . . , n. Let yi (t) = eαt |xi (t) − x∗i |, i = 1, 2, . . . , n. Then it follows from (3.4) that d+ yi (t) d+ |xi (t) − x∗i | = αeαt |xi (t) − x∗i | + eαt dt dt n X |aij |Fj yj (t) ≤ −(di − α)yi (t) + (3.8) j=1 + n X Z |bij | + |b̃ij | Fj +∞ eαs Kij (s)yj (t − s)ds 0 j=1 for t > 0, i = 1, 2, . . . , n, t 6= tk , k = 1, 2, . . . . Also, from (3.5), we have + αtk ∗ αtk yi (t+ |xi (t+ |xi (tk ) − x∗i | = yi (tk ) k)=e k ) − xi | ≤ e for i = 1, 2, . . . , n, k = 1, 2, . . . . Now, we construct the Lyapunov functional n n h Z +∞ Z t i X X αs V (t) = li yi (t) + Fj |bij | + |b̃ij | e Kij (s) yj (r)dr ds . (3.9) i=1 0 j=1 t−s The derivative of V (t) along with the trajectories of model (3.1) is D+ V (t) = ≤ Z +∞ n n h d+ y (t) X X i + Fj |bij | + |b̃ij | yj (t) eαs Kij (s)ds li dt 0 j=1 i=1 n Z +∞ i X − Fj |bij | + |b̃ij | eαs Kij (s)yj (t − s)ds 0 j=1 n X h li − (di − α)yi (t) + i=1 n X + j=1 + n X j=1 n X |aij |Fj yj (t) j=1 Fj Z |bij | + |b̃ij | +∞ eαs Kij (s)yj (t − s)ds 0 Z Fj |bij | + |b̃ij | yj (t) 0 +∞ eαs Kij (s)ds 10 KELIN LI, ZUOAN LI, QIANKUN SONG − =− + n X j=1 n X Fj Z |bij | + |b̃ij | +∞ eαs Kij (s)yj (t − s)ds li (di − α)yi (t) + n X n X li Fj |aij |yj (t) i=1 j=1 Z li Fj |bij | + |b̃ij | |yj (t) +∞ eαs Kij (s)ds 0 i=1 j=1 = i 0 i=1 n X n X n n X EJDE-2007/50 n h Z X − li (di − α) + Fi lj |aji | + |bji | + |b̃ji | i=1 +∞ eαs Kji (s)ds io yi (t) ≤ 0 0 j=1 for t > 0, t 6= tk , k = 1, 2, . . . . Also, V (t+ k) n n h Z X X + = li yi (tk ) + Fj |bij | + |b̃ij | ≤ i=1 n X h li yi (tk ) + i=1 j=1 n X Z e Kij (s) αs +∞ Z Fj |bij | + |b̃ij | t+ k t+ k −s 0 tk Z eαs Kij (s) i yj (r)dr ds i yj (r)dr ds tk −s 0 j=1 = V (tk ), +∞ k = 1, 2, . . . . So, we have V (t) ≤ V (0), for all t > 0. From (3.9), we obtain V (t) ≥ min {li } 1≤i≤n n X yi (t). (3.10) i=1 Also, V (0) n n h Z X X = li yi (0) + Fj |bij | + |b̃ij | i=1 ≤ max {li } 1≤i≤n ≤ max {li } 1≤i≤n +∞ Z e Kij (s) n X i=1 n h X j=1 n X yi (0) + yi (0) + i=1 i yj (r)dr ds −s 0 j=1 n h X 0 αs Fj Z |bij | + |b̃ij | Fj Z |bij | + |b̃ij | +∞ Z eαs Kij (s) 0 −s 0 +∞ seαs Kij (s)ds sup i yj (r) . −∞<r≤0 0 j=1 i yj (r)dr ds From the above inequality and (3.10), we have n X i=1 n yi (t) ≤ max1≤i≤n {li } X h yi (0) min1≤i≤n {li } i=1 n Z +∞ X + Fj |bij | + |b̃ij | seαs Kij (s)ds j=1 sup i yj (r) , −∞<r≤0 0 for t > 0. It follows from the definition of yi (t) and the above inequality that n X i=1 |xi (t) − x∗i | ≤ M e−αt sup n X −∞<s≤0 i=1 |ϕi (s) − x∗i | EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 11 for t > 0, where M= n Z +∞ i X max1≤i≤n {li } h 1 + max Fj |bij | + |b̃ij | seαs Kij (s)ds . 1≤i≤n min1≤i≤n {li } 0 j=1 The proof is completed. As a direct result of Theorems 3.1, 3.2, we have the following result. Corollary 3.3. Assume that (H1), (H2) hold, then model (3.1) has one unique equilibrium point, if any one of the following conditions is true: Pn (i) di > Fi j=1 |aji | + |bji | + |b̃ji | , i = 1, 2, . . . , n. Pn (ii) di > j=1 Fj |aij | + |bij | + |b̃ij | , i = 1, 2, . . . , n. (iii) There exists a positive vector l = (l1 , l2 , . . . , ln )T > 0 such that li di > n X lj Fj |aij | + |bij | + |b̃ij | , i = 1, 2, . . . , n. j=1 Furthermore, suppose that the impulsive operator ∆k (xi (tk )) satisfies ∆k (xi (tk )) = −δik (xi (tk ) − x∗i ), 0 < δik < 2, i = 1, 2, . . . , n, k = 1, 2, . . . . Then the equilibrium point x∗ = (x∗1 , x∗2 , . . . , x∗n )T of the system (3.1) is globally exponentially stable. Proof. In fact, any one of the conditions (i)-(iii) can assure, D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix. 4. Periodic oscillatory solution In the section, we discuss the existence, uniqueness and global exponential stability of the periodic oscillatory solution of model (1.1). Let Ii : R → R and ui : R → R be continuously periodic function with period ω, i.e. Ii (t + ω) = Ii (t), ui (t + ω) = ui (t) for i = 1, 2, . . . , n. Furthermore, we assume that (H3) There exists a positive integer q such that tk+q = tk + ω, δi(k+q) = δik , k = 1, 2, . . . , i = 1, 2, . . . , n, − where δik satisfy ∆k (xi (tk )) = xi (t+ k ) − xi (tk ) = −δik xi (tk ), 0 < δik < 2. Theorem 4.1. Under hypothesis (H1)–(H3), there exists exactly one ω-periodic solution of model (1.1) and all other solutions of model (1.1) converge exponentially to it as t → +∞, if D − (|A| + |B| + |B̃|)F is a nonsingular M -matrix. Proof. Let x(t, φ) = (x1 (t, φ), x2 (t, φ), . . . , xn (t, φ))T and let x(t, ϕ) = (x1 (t, ϕ), x2 (t, ϕ), . . . , xn (t, ϕ))T be an arbitrary pair of solutions of (1.1). Since D−(|A|+|B|+|B̃|)F is a nonsingular M -matrix, (3.6) and (3.7) hold. Let ỹ(t) = eαt |xi (t, φ) − xi (t, ϕ)|, i = 1, 2, . . . , n, 12 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 we easily obtain n X d+ ỹi (t) ≤ −(di − α)ỹi (t) + |aij |Fj ỹj (t) dt j=1 n Z +∞ X + Fj |bij | + |b̃ij | eαs Kij (s)ỹj (t − s)ds, 0 j=1 and (4.1) + + αtk ỹi (t+ |xi (t+ k)=e k , φ) − xi (tk , ϕ)| − = eαtk |xi (t− k , φ) − δik xi (tk , φ) − xi (tk , ϕ) + δik xi (tk , ϕ)| = eαtk |xi (tk , φ) − δik xi (tk , φ) − xi (tk , ϕ) + δik xi (tk , ϕ)| (4.2) = eαtk |1 − δik ||xi (tk , φ) − xi (tk , ϕ)| ≤ eαtk |xi (tk , φ) − xi (tk , ϕ)| = ỹi (tk ). Now, we construct the Lyapunov functional n n h Z +∞ Z X X V (t) = li ỹi (t) + Fj |bij | + |b̃ij | eαs Kij (s) i=1 j=1 0 t i ỹj (r)dr ds . t−s By a minor modification of the proof of Theorem 3.2, we can easily derive n n X X |φi (s) − ϕi (s)| |xi (t, φ) − xi (t, ϕ)| ≤ M e−αt sup −∞<s≤0 i=1 i=1 for t ≥ 0, where M ≥ 1 is constant, α = min1≤i≤n {αi } from (3.8). Therefore, we have kx(t, φ) − x(t, ϕ)k∞ ≤ M e−αt kφ − ϕk∞ . (4.3) Below, we prove that the system (1.1) has exactly one ω-periodic solution. For each solution x(t, φ) of (1.1) and each t ≥ 0, we define a function xt (φ) in this fashion: xt (φ)(s) = x(t + s, φ) for s ∈ (−∞, 0]. From (4.3), we can choose a positive integer N such that M e−αN ω ≤ 61 . Now, define a Poincare mapping C → C by P (ϕ) = xω (ϕ), then P N (ϕ) = xN ω (ϕ). Let t = N ω, then 1 kP N (φ) − P N (ϕ)k∞ ≤ kφ − ϕk∞ . 6 This implies that P N is a contraction mapping, hence there exists one unique fixed point ϕ∗ ∈ C such that P N (ϕ∗ ) = ϕ∗ . Since P N (P (ϕ∗ )) = P (P N (ϕ∗ )) = P (ϕ∗ ), P (ϕ∗ ) ∈ C is also a fixed point of N P , it follows that P (ϕ∗ ) = ϕ∗ , that is xω (ϕ∗ ) = ϕ∗ . Let x(t, ϕ∗ ) be the solution of model (1.1) through (0, ϕ∗ ), then x(t + ω, ϕ∗ ) is also a solution of model (1.1). Obviously xt+ω (ϕ∗ ) = xt (xω (ϕ∗ )) = xt (ϕ∗ ) for all t ≥ 0. Hence x(t + ω, ϕ∗ ) = x(t, ϕ∗ ). This shows that x(t, ϕ∗ ) is exactly one ω-periodic solution of model (1.1), and all solutions of model (1.1) converge exponentially to it as t → +∞. The proof is completed. EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 13 As a direct result of Theorem 4.1, we have following corollary. Corollary 4.2. Under hypothesis (H1)–(H3), there exists exactly one ω-periodic solution of model (1.1) and all other solutions of model (1.1) converge exponentially to it as t → +∞, if any one of the following conditions is true: Pn (i) di > Fi j=1 |aji | + |bji | + |b̃ji | , i = 1, 2, . . . , n. Pn (ii) di > j=1 Fj |aij | + |bij | + |b̃ij | , i = 1, 2, . . . , n. (iii) There exists a positive vector l = (l1 , l2 , . . . , ln )T > 0 such that li di > n X lj Fj |aij | + |bij | + |b̃ij | , i = 1, 2, . . . , n. j=1 5. Examples Example 5.1. Consider the model 2 2 X X dx1 (t) ã1j uj + I1 a1j fj (xj (t)) + = −d1 x1 (t) + dt j=1 j=1 + 2 ^ Z + −(t−s) b1j e fj (xj (s))ds + −∞ j=1 2 ^ t T1j uj + j=1 2 _ Z H1j uj , e−(t−s) fj (xj (s))ds −∞ j=1 2 _ t b̃1j t ≥ 0, t 6= tk j=1 1 25 sin(1 + k))(x1 (tk ) − ), k = 1, 2, . . . , 2 32 2 2 X X dx2 (t) = −d2 x2 (t) + a2j fj (xj (t)) + ã2j uj + I2 dt j=1 j=1 ∆x1 (tk ) = −(1 + + 2 ^ j=1 + 2 ^ j=1 Z t b2j −(t−s) e fj (xj (s))ds + −∞ T2j uj + 2 _ Z H2j uj , e−(t−s) fj (xj (s))ds −∞ j=1 2 _ t b̃2j t ≥ 0, t 6= tk j=1 ∆x2 (tk ) = −(1 + 1 2 cos(2k))(x2 (tk ) − ), 3 2 k = 1, 2, . . . , (5.1) where 0 < t1 < t2 < . . . is a strictly increasing sequence such that limt→+∞ tk = +∞; fi (x) = 12 (|x + 1| + |x − 1|), i = 1, 2; 1 1 1 1 1 4 0 −1 D= , A= 4 , B = 22 21 , B̃ = 41 22 , 0 5 −3 +1 3 3 3 3 1 2 Ã = 31 13 , I1 = I2 = 2, u1 = u2 = 1, T = (Tij ) = E, H = (Hij ) = E. 2 2 We can easily check that (H1) and (H2) hold, and for any x1 , x2 ∈ R, we have |f1 (x1 ) − f2 (x2 )| ≤ |x1 − x2 |, i = 1, 2, 14 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 hence F1 = F2 = 1. It follows that D − (|A| + |B| + |B̃|)F = 3 −2 −4 3 is a nonsingular M -matrix. Also, α1k = 1 + 21 sin(1 + k), α2k = 1 + 32 cos(2k) such that 0 < αik < 2, i = 1, 2, k = 1, 2, . . . . From Corollary 3.3, we know that neural network model (5.1) has one unique equilibrium point, which is globally exponentially stable. Using MATLAB software, we can get the unique equilibrium 1 T ∗ point x = 25 32 , 2 ) . Example 5.2. Consider the following impulsive neural network model with distributed delays 2 2 X X dx1 (t) = −d1 x1 (t) + a1j fj (xj (t)) + ã1j uj + I1 dt j=1 j=1 + 2 ^ Z + e−2(t−s) fj (xj (s))ds + −∞ j=1 2 ^ t b1j T1j uj + 2 _ Z H1j uj , e−2(t−s) fj (xj (s))ds −∞ j=1 2 _ t b̃1j t ≥ 0, t 6= tk j=1 j=1 1 sin(1 + k))(x1 (tk )), tk = 0.3 + 2(k − 1)π, 2 2 2 X X dx2 (t) = −d2 x2 (t) + a2j fj (xj (t)) + ã2j uj + I2 dt j=1 j=1 ∆x1 (tk ) = −(1 + + 2 ^ Z + e−2(t−s) fj (xj (s))ds + −∞ j=1 2 ^ t b2j T2j uj + j=1 2 _ Z H2j uj , t e−2(t−s) fj (xj (s))ds b̃2j −∞ j=1 2 _ k = 1, 2, . . . , t ≥ 0, t 6= tk j=1 ∆x2 (tk ) = −(1 + 2 cos(2k))(x2 (tk )), 3 tk = 0.3 + 2(k − 1)π, k = 1, 2, . . . , (5.2) where 0 < t1 < t2 < . . . is a strictly increasing sequence such that limt→+∞ tk = +∞; fi (x) = 1+e1−x , i = 1, 2; 1 1 1 4 0 −1 4 D= , A= , B = 22 12 , 0 5 −3 +1 1 1 3 3 3 3 B̃ = 41 22 , Ã = 21 21 , 3 3 2 2 I1 = I2 = 2 sin t, u1 = u2 = cos t, T = (Tij ) = E, H = (Hij ) = E. We can easily check that (H1) and (H2) hold, and that for any x1 , x2 ∈ R, we have |f1 (x1 ) − f2 (x2 )| ≤ |x1 − x2 |, i = 1, 2, hence F1 = F2 = 1. It follows that D − (|A| + |B| + |B̃|)F = 3 −2 −4 3 EJDE-2007/50 IMPULSIVE FUZZY CELLULAR NEURAL NETWORKS 15 is a nonsingular M -matrix. Also, α1k = 1 + 12 sin(1 + k), α2k = 1 + 32 cos(2k) such that 0 < αik < 2, i = 1, 2, k = 1, 2, . . . . From Theorem 4.1, we conclude that there exists exactly one 2π-periodic solution of model (5.2), and all other solutions converge exponentially to this solution as t → +∞. Conclusions. Stability and periodic oscillatory behavior are important in the applications and theories of neural networks. By employing the theory of topological degree, M -matrix and Lypunov functional, We have obtained some sufficient conditions ensuring the existence, uniqueness and global exponential stability of both the equilibrium point and the periodic solution for a class of impulsive fuzzy cellular neural networks with distributed delays. It is believed that these results are significant and useful for the design and applications of the fuzzy cellular neural networks. Acknowledgments. This work was supported by grant 2006A109 from the Scientific Research Fund of Sichuan Provincial Education Department. References [1] M. U. Akhmetov, A. Zafer, Stability of the zero solution of impulsive differential equations by the Lyapunov second method, J. Math. Anal.Appl. 248 (2000) 69-82. [2] S. Arik, V. Tavsanoglu, Equilibrium analysis of delayed CNNs, IEEE Transactions on Circuits Systems I, 45(2)(1998): 168-171. [3] D. D. Bainov, P. S. Simenov, Systems with Impulse Effect: Stability Theory and Applications, Ellis Horwood, Chichester, 1989. [4] A. Berman, R. J. Plemmons, Nonnegative Matrices in Mathematical Sciences, Academic Press, New York, 1979. [5] J. D. Cao, Global stability analysis in delayed cellular neural networks, Physical Review E, 59(5)(1999): 5940-5944. [6] J. D. Cao, A set of stability criteria for delayed cellular neural networks, IEEE Transactions on Circuits and Systems I, 48(2001): 494-498. [7] L. O. Chua and L. Yang, Cellular neural networks: Theory, IEEE Transactions on Circuits and Systems I. 35(10)(1988): 1257-1272. [8] L. O. Chua and L. Yang, Cellular neural networks: applications, IEEE Transactions on Circuits and Systems I. 35(10)(1988): 1273-1290. [9] J. Cronin, Fixed Points and Topological Degree in Nonlinear Analysis, Providence, RI: Amer. Math. Soc., 1964. [10] K. Gopalsamy, B.G. Zhang, On delay differential equation with impulses, J. Math. Anal. Appl. 139 (1989) 110-122. [11] Z.H. Guan, G. Chen, On delayed impulsive Hopfield neural networks, Neural Networks 12 (1999) 273-280. [12] R. A. Horn, C. R. Johnson, Matrix Analysis, World Publishing, Beijing, 1991. [13] T. W. Huang, Exponential stability of fuzzy cellular neural networks with distributed delay, Physics letters A, 351(1-2)(2006): 48-52. [14] T. W. Huang, Exponential stability of delayed fuzzy cellular neural networks with diffusion, Chaos, Solitons and Fractals, 31(3) (2007):658-664. [15] V. Lakshmikantham, D.D. Bainov, P.S. Simeonov, Theory of Impulsive Differential Equations, World Scientific, Singapore, 1989. [16] Y. Q. Liu, W. S. Tang, Exponential stability of fuzzy cellular neural networks with constant and time-varying delays, Physics Letters A, Vol. 323(3-4)(2004): 224-233. [17] X. Liu, G. Ballinger, Uniform asymptotic stability of impulsive delay differential equations, Comput. Math. Appl. 41 (2001) 903-915. [18] T. Roska and L. O. Chua, Cellular neural networks with nonlinear and delay-type template elements nonumiform grids, Int. J. Cricuit Theory Appl., 20 (1992): 469-481. [19] V. Singh, A generalized LMI-based approach to the global asymptotic stability of delayed cellular neural networks, IEEE Transactions on Neural Networks, 15(1)(2004): 223-225. 16 KELIN LI, ZUOAN LI, QIANKUN SONG EJDE-2007/50 [20] D. Y. Xu, Z. C. Yang, Impulsive delay differential inequality and stability of neural networks, J. Math. Anal. Appl. 305 (2005): 107-120. [21] J. Yan, Stability theorems of perturbed linear system with impulse effect, Portugal. Math. 53 (1996) 43-51. [22] T. Yang, L. B. Yang, The global stability of fuzzy celluar neural networks, IEEE Transactions on Cricuit and Systems I, 43(10)(1996): 880-883. [23] T. Yang, L. B. Yang, C. W. Wu, L. O. Chua, Fuzzy celluar neural networks: Theory, in: Proceedings of IEEE International Workshop on Cellular Neural Networks and Applications, 1996: 181-186. [24] K. Yuan, J.D. Cao, J.M. Deng, Exponential stability and periodic solutions of fuzzy cellular neural networks with time-varying delays, Neurocomputing, 69(13-15)(2006):1619-1627. [25] Q. Zhang, X.P. Wei, J. Xu, Stability analysis for cellular neural networks with variable delays, Chaos, Solitons and Fractals, 28(2)(2006): 331-336. Kelin Li Department of Mathematics, Sichuan University of Science & Engineering, Sichuan 643000, China E-mail address: lkl@suse.edu.cn Zuoan Li Department of Mathematics, Sichuan University of Science & Engineering, Sichuan 643000, China E-mail address: lizuoan@suse.edu.cn Qiankun Song Department of Mathematics, Chongqing Jiaotong University, Chongqing, 400074, China E-mail address: qiankunsong@163.com