Skip to main content

Almost periodic solutions for SICNNs with time-varying delays in the leakage terms

Abstract

This paper is concerned with the shunting inhibitory cellular neural networks (SICNNs) with time-varying delays in the leakage (or forgetting) terms. Under proper conditions, we employ a novel argument to establish a criterion on the global exponential stability of almost periodic solutions by using Lyapunov functional method and differential inequality techniques. We also provide numerical simulations to support the theoretical result.

1 Introduction

It is well known that shunting inhibitory cellular neural networks (SICNNs) have been introduced as new cellular neural networks (CNNs) in Bouzerdout and Pinter in [13], which can be described by

x i j ( t ) = a i j ( t ) x i j ( t ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) , i = 1 , 2 , , m , j = 1 , 2 , , n ,
(1.1)

where C i j denotes the cell at the (i,j) position of the lattice. The r-neighborhood N r (i,j) is given as

N r (i,j)= { C k l : max ( | k i | , | l j | ) r , 1 k m , 1 l n } ,

N q (i,j) is similarly specified. x i j is the activity of the cell C i j , L i j (t) is the external input to C i j , the function a i j (t)>0 represents the passive decay rate of the cell activity, C i j k l (t) and B i j k l (t) are the connection or coupling strength of postsynaptic activity of the cell transmitted to the cell C i j , and the activity functions f() and g() are continuous functions representing the output or firing rate of the cell C k l , and τ(t)0 corresponds to the transmission delay.

Recently, SICNNs have been extensively applied in psychophysics, speech, perception, robotics, adaptive pattern recognition, vision, and image processing. Hence, they have been the object of intensive analysis by numerous authors in recent years. In particular, there have been extensive results on the problem of the existence and stability of the equilibrium point, periodic and almost periodic solutions of SICNNs with time-varying delays in the literature. We refer the reader to [49] and the references cited therein. Obviously, the first term in each of the right side of (1.1) corresponds to a stabilizing negative feedback of the system, which acts instantaneously without time delay; these terms are variously known as ‘forgetting’ or leakage terms (see, for instance, Kosko [10], Haykin [11]). It is known from the literature on population dynamics and neural networks dynamics (see Gopalsamy [12]) that time delays in the stabilizing negative feedback terms will have a tendency to destabilize a system. Therefore, the authors of [1319] dealt with the existence and stability of equilibrium and periodic solutions for neuron networks model involving leakage delays. However, to the best of our knowledge, few authors have considered the existence and exponential stability of almost periodic solutions of SICNNs with time-varying delays in the leakage terms. Motivated by the discussions above, in this paper, we consider the following SICNNs with time-varying leakage delays:

x i j ( t ) = a i j ( t ) x i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) ,
(1.2)

where i=1,2,,m, j=1,2,,n, a i j :R(0+), η i j ,τ:R[0+), and L i j , C i j k l , B i j k l :RR are almost periodic functions, η i (t) and τ(t) denote the leakage delay and transmission delay, respectively, the delay kernels K i j :[0,)[0,) are continuous and integrable, and η i j is a bounded continuous function.

The main purpose of this paper is to give the conditions for the existence and exponential stability of the almost periodic solutions for system (1.2). By applying Lyapunov functional method and differential inequality techniques, we derive some new sufficient conditions ensuring the existence, uniqueness and exponential stability of the almost periodic solution for system (1.2), which are new and complement previously known results. Moreover, an example with numerical simulations is also provided to illustrate the effectiveness of our results.

Throughout this paper, for ijJ:={11,12,,1n,21,22,,2n,,m1,m2,,mn}, from the theory of almost periodic functions in [20, 21], it follows that for all ϵ>0, it is possible to find a real number l=l(ϵ)>0, for any interval with length l(ϵ), there exists a number δ=δ(ϵ) in this interval such that

{ | a i j ( t + δ ) a i j ( t ) | < ϵ , | η i j ( t + δ ) η i j ( t ) | < ϵ , | C i j k l ( t + δ ) C i j k l ( t ) | < ϵ , | B i j k l ( t + δ ) B i j k l ( t ) | < ϵ , | τ ( t + δ ) τ ( t ) | < ϵ , | L i j ( t + δ ) L i j ( t ) | < ϵ ,
(1.3)

for all tR, kl,ijJ.

We set

{ x i j ( t ) } = ( x 11 ( t ) , , x 1 n ( t ) , , x i 1 ( t ) , , x i n ( t ) , , x m 1 ( t ) , , x m n ( t ) ) R m × n .

For x(t)={ x i j (t)} R m × n , we define the norm x(t)= max ( i , j ) {| x i j (t)|}. For the convenience, we shall introduce the notations

h + = sup t R | h ( t ) | ,

where h(t) is a bounded continuous function.

We also make the following assumptions.

(T1) There exist constants M f , M g , μ f and μ g such that

| f ( u ) f ( v ) | μ f | u v | , | f ( u ) | M f , | g ( u ) g ( v ) | μ g | u v | , | g ( u ) | M g , for all  u , v R .

(T2) For ijJ, a i j + η i j + <1,

δ i j ( t ) = { a i j ( t ) ( 1 2 a i j + η i j + ) | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | C k l N r ( i , j ) | C i j k l ( t ) | M f C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g } 1 1 a i j + η i j + > 0 ,
(1.4)

and there exist positive constants η>0 and λ such that

λ< a i j (t), 0 e λ u | K i j ( u ) | du<+,

and

[ ( a i j ( t ) λ ) ( 1 2 a i j + η i j + ) | a i j ( t ) e λ η i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | ] 1 1 a i j + η i j + + C k l N r ( i , j ) | C i j k l ( t ) | μ f e λ τ ( t ) 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N r ( i , j ) | C i j k l ( t ) | M f 1 1 a i j + η i j + + C k l N q ( i , j ) | B i j k l ( t ) | 0 e λ u | K i j ( u ) | d u μ g 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g 1 1 a i j + η i j + < η , for all  t 0 .
(1.5)

The initial conditions associated with system (1.2) are of the form

x i j (s)= φ i j (s),s(,0],ijJ,
(1.6)

where φ i j () denotes a real-valued bounded continuous function defined on (,0].

Definition 1.1 (See [20, 21])

Let u(t):R R m × n be continuous in t. u(t) is said to be almost periodic on R if for any ε>0, the set T(u,ε)={δ:u(t+δ)u(t)<ε,tR} is relatively dense, i.e., for any ε>0, it is possible to find a real number l=l(ε)>0, for any interval with length l(ε), there exists a number δ=δ(ε) in this interval such that u(t+δ)u(t)<ε for all tR.

2 Preliminary results

The following lemmas will be useful to prove our main results in Section 3.

Lemma 2.1 Let (T1) and (T2) hold. Suppose that x(t)={ x i j (t)} is a solution of system (1.2) with initial conditions

x i j (s)= φ i j (s), | φ i j ( t ) t η i j ( t ) t a i j ( s ) φ i j ( s ) d s | < ( L i j δ i j ) + ,s,t(,0],ijJ.
(2.1)

Then

| x i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s | ( L i j δ i j ) + ,for all t0,ijJ,
(2.2)

and

| x i j ( t ) | 1 1 a i j + η i j + ( L i j δ i j ) + ,for all t0,ijJ.
(2.3)

Proof Assume, by way of contradiction, that (2.2) does not hold. Then, there exist ijΛ, γ> ( L i j δ i j ) + and t >0 such that

| X i j ( t )|=γand| X i j (t)|<γfor all t(, t ),
(2.4)

where

X i j (t)= x i j (t) t η i j ( t ) t a i j (s) x i j (s)ds.

It follows that

| x i j ( t ) | | x i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s | + | t η i j ( t ) t a i j ( s ) x i j ( s ) d s | γ + a i j + η i j + sup s ( , t ] | x i j ( s ) | , for all  t ( , t ] .
(2.5)

Consequently, in view of (2.5) and the fact a i j + η i j + <1 (ijJ), we have

| x i j ( t ) | sup s ( , t ] | x i j ( s ) | 1 1 a i j + η i j + γ , for all  t ( , t ] .
(2.6)

From system (1.2), we derive

d d t ( x i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s ) = x i j ( t ) [ a i j ( t ) x i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) x i j ( t η i j ( t ) ) ] = [ a i j ( t ) x i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) x i j ( t η i j ( t ) ) ] + [ a i j ( t ) x i ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) ] = a i j ( t ) x i j ( t ) [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) ] x i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) = a i j ( t ) ( x i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s ) a i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) ] x i j ( t η i ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) , i j J .
(2.7)

Calculating the upper left derivative of | X i j (t)|, together with (2.4), (2.6), (2.7), (T1) and (T2), we obtain

0 D | X i j ( t ) | a i j ( t ) | X i j ( t ) | + | a i j ( t ) t η i j ( t ) t a i j ( s ) x i j ( s ) d s [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) ] x i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) | a i j ( t ) | X i j ( t ) | + a i j ( t ) t η i j ( t ) t a i j + | x i j ( s ) | d s + | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | | x i j ( t η i j ( t ) ) | + C k l N r ( i , j ) | C i j k l ( t ) | | f ( x k l ( t τ ( t ) ) ) | | x i j ( t ) | + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | | g ( x k l ( t u ) ) | d u | x i j ( t ) | + | L i j ( t ) | a i j ( t ) | X i j ( t ) | + a i j ( t ) t η i j ( t ) t a i j + | x i j ( s ) | d s + | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | | x i j ( t η i j ( t ) ) | + C k l N r ( i , j ) | C i j k l ( t ) | M f | x i j ( t ) | + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g | x i j ( t ) | + | L i j ( t ) | a i j ( t ) γ + a i j ( t ) a i j + η i j + 1 1 a i j + η i j + γ + | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | γ 1 a i j + η i j + + C k l N r ( i , j ) | C i j k l ( t ) | M f 1 1 a i j + η i j + γ + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g 1 1 a i j + η i j + γ + | L i j ( t ) | = { [ a i j ( t ) ( 1 2 a i j + η i j + ) | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | ] + C k l N r ( i , j ) | C i j k l ( t ) | M f + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g } 1 1 a i j + η i j + γ + | L i j ( t ) | δ i j ( t ) [ γ ( L i j δ i j ) + ] < 0 .

It is a contradiction, and it shows that (2.2) holds. Then, using a similar argument as in the proof of (2.5) and (2.6), we can show that (2.3) holds. The proof of Lemma 2.1 is now completed. □

Remark 2.1 In view of the boundedness of this solution, from the theory of functional differential equations with infinite delay in [22], it follows that the solution of system (1.2) with initial conditions (2.1) can be defined on [0,+).

Lemma 2.2 Suppose that (T1) and (T2) hold. Moreover, assume that x(t)={ x i j (t)} is a solution of system (1.2) with initial function φ i j () satisfying (2.1), and φ i j () is bounded continuous on (,0]. Then for any ϵ>0, there exists l=l(ϵ)>0, such that every interval [α,α+l] contains at least one number δ, for which there exists N>0 satisfying

x ( t + δ ) x ( t ) ϵ,for all t>N.
(2.8)

Proof For ijJ, set

ϵ i j ( δ , t ) = a i j ( t ) [ x i j ( t + δ η i j ( t + δ ) ) x i j ( t η i j ( t ) + δ ) ] [ a i j ( t + δ ) a i j ( t ) ] x i j ( t + δ η i j ( t + δ ) ) C k l N r ( i , j ) [ C i j k l ( t + δ ) C i j k l ( t ) ] f ( x k l ( t τ ( t + δ ) + δ ) ) x i j ( t + δ ) C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t + δ ) + δ ) ) f ( x k l ( t τ ( t ) + δ ) ) ] x i j ( t + δ ) C k l N q ( i , j ) [ B i j k l ( t + δ ) B i j k l ( t ) ] 0 K i j ( u ) g ( x k l ( t + δ u ) ) d u x i j ( t + δ ) + [ L i j ( t + δ ) L i j ( t ) ] .

By Lemma 2.1, the solution x(t)={ x i j (t)} is bounded and

| x i j ( t ) | 1 1 a i j + η i j + ( L i j δ i j ) + for all t[0,+),ijJ.
(2.9)

Thus, the right side of (1.2) is also bounded, which implies that x(t) is uniformly continuous on R. From (1.3), for any ϵ>0, there exists l=l(ϵ)>0, such that every interval [α,α+l], αR, contains a δ, for which

| ϵ i j ( δ , t ) | 1 2 η min i j ˜ J { 1 a i j ˜ + η i j ˜ + } ϵ,where ijJ,tR.
(2.10)

Let N 0 0 be sufficiently large such that t+δ0 for t N 0 , and denote u i j (t)= x i j (t+δ) x i j (t). We obtain

d u i j ( t ) d t = a i j ( t ) u i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) ( f ( x k l ( t τ ( t ) + δ ) ) f ( x k l ( t τ ( t ) ) ) ) x i j ( t + δ ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) ( x i j ( t + δ ) x i j ( t ) ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) ( g ( x k l ( t + δ u ) ) g ( x k l ( t u ) ) ) d u x i j ( t + δ ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u ( x i j ( t + δ ) x i j ( t ) ) + ϵ i j ( δ , t ) , for all  t N 0 , i j J ,

which yields

d d t ( e λ t u i j ( t ) t η i j ( t ) t a i j ( s ) e λ s u i j ( s ) d s ) = λ e λ t u i j ( t ) + e λ t u i j ( t ) [ a i j ( t ) e λ t u i j ( t ) ( 1 η i j ( t ) ) c i j ( t η i j ( t ) ) e λ ( t η i j ( t ) ) u i j ( t η i j ( t ) ) ] = λ e λ t u i j ( t ) [ a i j ( t ) e λ t u i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ ( t η i j ( t ) ) u i j ( t η i j ( t ) ) ] + e λ t { a i j ( t ) u i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) ( f ( x k l ( t τ ( t ) + δ ) ) f ( x k l ( t τ ( t ) ) ) ) x i j ( t + δ ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) ( x i j ( t + δ ) x i j ( t ) ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) ( g ( x k l ( t + δ u ) ) g ( x k l ( t u ) ) ) d u x i j ( t + δ ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u ( x i j ( t + δ ) x i j ( t ) ) + ϵ i j ( δ , t ) } = ( a i j ( t ) λ ) e λ t u i j ( t ) [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ η i j ( t ) ] e λ t u i j ( t η i j ( t ) ) + e λ t { C k l N r ( i , j ) C i j k l ( t ) ( f ( x k l ( t τ ( t ) + δ ) ) f ( x k l ( t τ ( t ) ) ) ) x i j ( t + δ ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) ( x i j ( t + δ ) x i j ( t ) ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) ( g ( x k l ( t + δ u ) ) g ( x k l ( t u ) ) ) d u x i j ( t + δ ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u ( x i j ( t + δ ) x i j ( t ) ) + ϵ i j ( δ , t ) } = ( a i j ( t ) λ ) ( e λ t a i j ( t ) t η i j ( t ) t a i j ( s ) e λ s u i j ( s ) d s ) ( a i j ( t ) λ ) t η i j ( t ) t a i j ( s ) e λ s u i j ( s ) d s [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ η i j ( t ) ] e λ t u i j ( t η i j ( t ) ) + e λ t { C k l N r ( i , j ) C i j k l ( t ) ( f ( x k l ( t τ ( t ) + δ ) ) f ( x k l ( t τ ( t ) ) ) ) x i j ( t + δ ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) ( x i j ( t + δ ) x i j ( t ) ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) ( g ( x k l ( t + δ u ) ) g ( x k l ( t u ) ) ) d u x i j ( t + δ ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u ( x i j ( t + δ ) x i j ( t ) ) + ϵ i j ( δ , t ) } , for all  t N 0 , i j J .
(2.11)

Set

U(t)= { U i j ( t ) } ,

where

U i j (t)= e λ t u i j (t) t η i j ( t ) t a i j (s) e λ s u i j (s)ds,ijJ.

Let ( i j ) t be such an index that

| U ( i j ) t ( t ) | = U ( t ) .
(2.12)

Calculating the upper left derivative of | U ( i j ) s (s)| along (2.11), we have

D ( | U ( i j ) s ( s ) | ) | s = t ( a ( i j ) t ( t ) λ ) | U ( i j ) t ( t ) | + | ( a ( i j ) t ( t ) λ ) t η ( i j ) t ( t ) t a ( i j ) t ( s ) e λ s u ( i j ) t ( s ) d s [ a ( i j ) t ( t ) ( 1 η ( i j ) t ( t ) ) a ( i j ) t ( t η ( i j ) t ( t ) ) e λ η ( i j ) t ( t ) ] e λ t u ( i j ) t ( t η ( i j ) t ( t ) ) + e λ t { C k l N r ( i , j ) t C ( i j ) t k l ( t ) ( f ( x k l ( t τ ( t ) + δ ) ) f ( x k l ( t τ ( t ) ) ) ) x ( i j ) t ( t + δ ) C k l N r ( i , j ) t C ( i j ) t k l ( t ) f ( x k l ( t τ ( t ) ) ) ( x ( i j ) t ( t + δ ) x ( i j ) t ( t ) ) C k l N q ( i , j ) t B ( i j ) t k l ( t ) 0 K ( i j ) t ( u ) ( g ( x k l ( t + δ u ) ) g ( x k l ( t u ) ) ) d u x ( i j ) t ( t + δ ) C k l N q ( i , j ) t B ( i j ) t k l ( t ) 0 K ( i j ) t ( u ) g ( x k l ( t u ) ) d u ( x ( i j ) t ( t + δ ) x ( i j ) t ( t ) ) + ϵ ( i j ) t ( δ , t ) } | .
(2.13)

Let

M(t)= sup s t { U ( s ) } .
(2.14)

It is obvious that U(t)M(t), and M(t) is non-decreasing. In particular,

e λ ρ | u i j ( ρ ) | | e λ ρ u i j ( ρ ) ρ η i j ( ρ ) ρ a i j ( s ) e λ s u i j ( s ) d s | + | ρ η i j ( ρ ) ρ a i j ( s ) e λ s u i j ( s ) d s | M ( t ) + a i j + η i j + sup θ ( , t ] e λ θ | u i j ( θ ) | , for all  t ρ , i j J .
(2.15)

Consequently, in view of (2.15) and the fact a i j + η i j + <1 (ijJ), we have

e λ s | u i j ( s ) | sup θ ( , t ] e λ θ | u i j ( θ ) | M ( t ) 1 a i j + η i j + ,where s(,t],ijJ.
(2.16)

Now, we consider two cases.

Case (i). If

M(t)> U ( t ) for all t N 0 .
(2.17)

Then, we claim that

M(t)M( N 0 ) is a constant for all t N 0 .
(2.18)

Assume, by way of contradiction, that (2.18) does not hold. Then, there exists t 1 > N 0 such that M( t 1 )>M( N 0 ). Since

U ( t ) M( N 0 )for all t N 0 .

There must exist β( N 0 , t 1 ) such that

U ( β ) =M( t 1 )M(β),

which contradicts (2.17). This contradiction implies that (2.18) holds. It follows from (2.16) that there exists t 2 > N 0 such that

u ( t ) = max i j J | u i j ( t ) | max i j J e λ t M ( t ) 1 a i j + η i j + = max i j J e λ t M ( N 0 ) 1 a i j + η i j + <ϵ,for all t t 2 .
(2.19)

Case (ii). If there is such a point t 0 N 0 that M( t 0 )=U( t 0 ). Then, in view of (1.5), (2.9), (2.10), (2.13), (2.16), (T1) and (T2), we get

0 D ( | U ( i j ) s ( s ) | ) | s = t 0 ( a ( i j ) t 0 ( t 0 ) λ ) | U ( i j ) t 0 ( t 0 ) | + ( a ( i j ) t 0 ( t 0 ) λ ) t 0 η ( i j ) t 0 ( t 0 ) t 0 a ( i j ) t 0 + e λ s | u ( i j ) t 0 ( s ) | d s + | a ( i j ) t 0 ( t 0 ) e λ η ( i j ) t 0 ( t 0 ) ( 1 η ( i j ) t 0 ( t ) ) a ( i j ) t 0 ( t 0 η ( i j ) t 0 ( t 0 ) ) | e λ ( t 0 η ( i j ) t 0 ( t 0 ) ) | u ( i j ) t 0 ( t 0 η ( i j ) t 0 ( t 0 ) ) | + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | e λ t 0 | f ( x k l ( t 0 τ ( t 0 ) + δ ) ) f ( x k l ( t 0 τ ( t 0 ) ) ) | | x ( i j ) t 0 ( t 0 + δ ) | + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | | f ( x k l ( t 0 τ ( t 0 ) ) ) | e λ t 0 | x ( i j ) t 0 ( t 0 + δ ) x ( i j ) t 0 ( t 0 ) | + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 | K ( i j ) t 0 ( u ) | e λ t 0 | g ( x k l ( t 0 + δ u ) ) g ( x k l ( t 0 u ) ) | d u | x ( i j ) t 0 ( t 0 + δ ) | + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 | K ( i j ) t 0 ( u ) | | g ( x k l ( t 0 u ) ) | d u e λ t 0 | x ( i j ) t 0 ( t 0 + δ ) x ( i j ) t 0 ( t 0 ) | + | ϵ ( i j ) t 0 ( δ , t 0 ) | e λ t 0 ( a ( i j ) t 0 ( t 0 ) λ ) M ( t 0 ) + ( a ( i j ) t 0 ( t 0 ) λ ) a ( i j ) t 0 + η ( i j ) t 0 + M ( t 0 ) 1 a ( i j ) t 0 + η ( i j ) t 0 + + | a ( i j ) t 0 ( t 0 ) e λ η ( i j ) t 0 ( t 0 ) ( 1 η ( i j ) t 0 ( t ) ) a ( i j ) t 0 ( t 0 η ( i j ) t 0 ( t 0 ) ) | M ( t 0 ) 1 a ( i j ) t 0 + η ( i j ) t 0 + + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | μ f e λ τ ( t 0 ) e λ ( t 0 τ ( t 0 ) ) | u k l ( t 0 τ ( t 0 ) ) | 1 1 a ( i j ) t 0 + η ( i j ) t 0 + ( L ( i j ) t 0 δ ( i j ) t 0 ) + + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | M f e λ t 0 | u ( i j ) t 0 ( t 0 ) | + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 e λ u | K ( i j ) t 0 ( u ) | μ g e λ ( t 0 u ) | u k l ( t 0 u ) | d u 1 1 a ( i j ) t 0 + η ( i j ) t 0 + ( L ( i j ) t 0 δ ( i j ) t 0 ) + + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 | K ( i j ) t 0 ( u ) | d u M g e λ t 0 | u ( i j ) t 0 ( t 0 ) | + | ϵ ( i j ) t 0 ( δ , t 0 ) | e λ t 0 { [ ( a ( i j ) t 0 ( t 0 ) λ ) ( 1 2 a ( i j ) t 0 + η ( i j ) t 0 + ) | a ( i j ) t 0 ( t 0 ) e λ η ( i j ) t 0 ( t 0 ) ( 1 η ( i j ) t 0 ( t ) ) a ( i j ) t 0 ( t 0 η ( i j ) t 0 ( t 0 ) ) | ] 1 1 a ( i j ) t 0 + η ( i j ) t 0 + + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | μ f e λ τ ( t 0 ) 1 1 a k l + η k l + 1 1 a ( i j ) t 0 + η ( i j ) t 0 + ( L ( i j ) t 0 δ ( i j ) t 0 ) + + C k l N r ( i , j ) t 0 | C ( i j ) t 0 k l ( t 0 ) | M f 1 1 a ( i j ) t 0 + η ( i j ) t 0 + + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 e λ u | K ( i j ) t 0 ( u ) | d u μ g 1 1 a k l + η k l + 1 1 a ( i j ) t 0 + η ( i j ) t 0 + ( L ( i j ) t 0 δ ( i j ) t 0 ) + + C k l N q ( i , j ) t 0 | B ( i j ) t 0 k l ( t 0 ) | 0 | K ( i j ) t 0 ( u ) | d u M g 1 1 a ( i j ) t 0 + η ( i j ) t 0 + } M ( t 0 ) + η min i j J { 1 a i j + η i j + } ϵ e λ t 0 < η M ( t 0 ) + η min i j J { 1 a i j + η i j + } ϵ e λ t 0 ,
(2.20)

which yields that

U ( t 0 ) =M( t 0 )< min i j J { 1 a i j + η i j + } ϵ e λ t 0 ,
(2.21)

and

u ( t 0 ) max i j J e λ t 0 M ( t 0 ) 1 a i j + η i j + <ϵ.
(2.22)

For any t> t 0 , by the same approach used in the proof of (2.21) and (2.22), we have

U ( t ) = M ( t ) < min i j J { 1 a i j + η i j + } ϵ e λ t u ( t ) < ϵ } ,if M(t)= U ( t ) .
(2.23)

On the other hand, if M(t)>U(t) and t> t 0 , we can choose t 0 t 3 <t such that

M( t 3 )= U ( t 3 ) ,andM(s)> U ( s ) for all s( t 3 ,t],

which, together with (2.23), yields that

M( t 3 )= U ( t 3 ) < min i j J { 1 a i j + η i j + } ϵ e λ t 3 ,and u ( t 3 ) <ϵ.

Using a similar argument as in the proof of Case (i), we can show that

M(s)M( t 3 ) is a constant for all s( t 3 ,t],
(2.24)

which implies that

u ( t ) max i j J e λ t M ( t ) 1 a i j + η i j + = max i j J e λ t M ( t 3 ) 1 a i j + η i j + < max i j J e λ t min i j ¯ J { 1 a i j ¯ + η i j ¯ + } ϵ e λ t 3 1 a i j + η i j + <ϵ.

In summary, there must exist N>max{ t 0 , N 0 , t 2 } such that u(t)ϵ holds for all t>N. The proof of Lemma 2.2 is now complete. □

3 Main results

In this section, we establish some results for the existence, uniqueness and exponential stability of the almost periodic solution of (1.2).

Theorem 3.1 Suppose that (T1) and (T2) are satisfied. Then system (1.2) has exactly one almost periodic solution Z (t). Moreover, Z (t) is globally exponentially stable.

Proof Let v(t)={ v i j (t)} be a solution of system (1.2) with initial function φ i j v () satisfying (2.1), and ( φ i j v ( ) ) is bounded continuous on (,0].

Set

ϵ i j , k ( t ) = a i j ( t ) [ x i j ( t + t k η i j ( t + t k ) ) x i j ( t η i j ( t ) + t k ) ] [ a i j ( t + t k ) a i j ( t ) ] x i j ( t + t k η i j ( t + t k ) ) C k l N r ( i , j ) [ C i j k l ( t + t k ) C i j k l ( t ) ] f ( x k l ( t τ ( t + t k ) + t k ) ) x i j ( t + t k ) C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t + t k ) + t k ) ) f ( x k l ( t τ ( t ) + t k ) ) ] x i j ( t + t k ) C k l N q ( i , j ) [ B i j k l ( t + t k ) B i j k l ( t ) ] 0 K i j ( u ) g ( x k l ( t + t k u ) ) d u x i j ( t + t k ) + [ L i j ( t + t k ) L i j ( t ) ] , i j J ,
(3.1)

where { t k } is any sequence of real numbers. By Lemma 2.1, the solution v(t) is bounded and

| v i j ( t ) | 1 1 a i j + η i j + ( L i j δ i j ) + for all tR,ijJ,
(3.2)

which implies that the right side of (1.2) is also bounded, and v (t) is a bounded function on R. Thus, v(t) is uniformly continuous on R. Then, from the almost periodicity of a i j , η i j , τ, C i j k l and B i j k l , we can select a sequence { t k }+ such that

| a i j ( t + t k ) a i j ( t ) | 1 k , | C i j k l ( t + t k ) C i j k l ( t ) | 1 k , | B i j k l ( t + t k ) B i j k l ( t ) | 1 k | η i j ( t + t k ) η i j ( t ) | 1 k , | τ ( t + t k ) τ ( t ) | 1 k , | ϵ i j , k ( t ) | 1 k } ,
(3.3)

for all ij,klJ, tR.

Since { v ( t + t k ) } k = 1 + is uniformly bounded and equiuniformly continuous, by Arzala-Ascoli lemma and diagonal selection principle, we can choose a subsequence { t k j } of { t k }, such that v(t+ t k j ) (for convenience, we still denote by v(t+ t k )) uniformly converges to a continuous function Z (t)={ x i j (t)} on any compact set of R, and

| x i j ( t ) | 1 1 a i j + η i j + ( L i j δ i j ) + for all tR,ijJ.
(3.4)

Now, we prove that Z (t) is a solution of (1.2). In fact, for any t>0 and ΔtR, from (3.3), we have

x i j ( t + Δ t ) x i j ( t ) = lim k + [ v i j ( t + Δ t + t k ) v i j ( t + t k ) ] = lim k + t t + Δ t { a i j ( μ + t k ) v i j ( μ + t k η i j ( μ + t k ) ) C k l N r ( i , j ) C i j k l ( μ + t k ) f ( v k l ( μ + t k τ ( μ + t k ) ) ) v i j ( μ + t k ) C k l N q ( i , j ) B i j k l ( μ + t k ) 0 K i j ( u ) g ( v k l ( μ + t k u ) ) d u v i j ( μ + t k ) + L i j ( μ + t k ) } d μ = lim k + t t + Δ t { a i j ( μ ) v i j ( μ η i j ( μ ) + t k ) C k l N r ( i , j ) C i j k l ( μ ) f ( v k l ( μ τ ( μ ) + t k ) ) v i j ( μ + t k ) C k l N q ( i , j ) B i j k l ( μ ) 0 K i j ( u ) g ( v k l ( μ + t k u ) ) d u v i j ( μ + t k ) + L i j ( μ ) + ϵ i j , k ( μ ) } d μ = t t + Δ t { a i j ( μ ) x i j ( μ η i j ( μ ) ) C k l N r ( i , j ) C i j k l ( μ ) f ( x k l ( μ τ ( μ ) ) ) x i j ( μ ) C k l N q ( i , j ) B i j k l ( μ ) 0 K i j ( u ) g ( x k l ( μ u ) ) d u x i j ( μ ) + L i j ( μ ) } d μ + lim k + t t + Δ t ϵ i j , k ( μ ) d μ = t t + Δ t { a i j ( μ ) x i j ( μ η i j ( μ ) ) C k l N r ( i , j ) C i j k l ( μ ) f ( x k l ( μ τ ( μ ) ) ) x i j ( μ ) C k l N q ( i , j ) B i j k l ( μ ) 0 K i j ( u ) g ( x k l ( μ u ) ) d u x i j ( μ ) + L i j ( μ ) } d μ ,
(3.5)

which implies that

d d t { x i j ( t ) } = a i j ( t ) x i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) + L i j ( t ) , i j J .
(3.6)

Therefore, Z (t) is a solution of (1.2).

Secondly, we prove that Z (t) is a almost periodic solution of (1.2). From Lemma 2.2, for any ε>0, there exists l=l(ε)>0, such that every interval [α,α+l] contains at least one number δ for which there exists N>0 satisfying

| v i j ( t + δ ) v i j ( t ) | εfor all t>N,ijJ.
(3.7)

Then, for any fixed sR, we can find a sufficient large positive integer N 1 >N such that for any k> N 1

s+ t k >N, | v i j ( s + t k + δ ) v i j ( s + t k ) | ε,ijJ.
(3.8)

Let k+, we obtain

| x i j ( s + δ ) x i j ( s ) | ε,ijJ,
(3.9)

which implies that Z (t) is an almost periodic solution of (1.2).

Finally, we prove that Z (t) is globally exponentially stable.

Let Z (t)={ x i j (t)} be the positive almost periodic solution of system (1.2) with initial value φ ={ φ i j (t)}, and let Z(t)={ x i j (t)} be an arbitrary solution of system (1.2) with initial value φ={ φ i j (t)}, set y(t)={ y i j (t)}={ x i j (t) x i j (t)}=Z(t) Z (t). Then

y i j ( t ) = a i j ( t ) y i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t ) ) ) x i j ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) ] C k l N q ( i , j ) B i j k l ( t ) [ 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) ] ,

which yields

d d t ( e λ t y i j ( t ) t η i j ( t ) t a i j ( s ) e λ s y i j ( s ) d s ) = λ e λ t y i j ( t ) + e λ t y i j ( t ) [ a i j ( t ) e λ t y i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ ( t η i j ( t ) ) y i j ( t η i j ( t ) ) ] = λ e λ t y i j ( t ) [ a i j ( t ) e λ t y i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ ( t η i j ( t ) ) y i j ( t η i j ( t ) ) ] + e λ t { a i j ( t ) y i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t ) ) ) x i j ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) ] C k l N q ( i , j ) B i j k l ( t ) [ 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) ] } = ( a i j ( t ) λ ) e λ t y i j ( t ) [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ η i j ( t ) ] e λ t y i j ( t η i j ( t ) ) + e λ t { C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t ) ) ) x i j ( t ) f ( x k l ( t τ ( t ) ) ) x i j ( t ) ] C k l N q ( i , j ) B i j k l ( t ) [ 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j ( t ) ] } = ( a i j ( t ) λ ) ( e λ t y i j ( t ) t η i j ( t ) t a i j ( s ) e λ s y i j ( s ) d s ) ( a i j ( t ) λ ) t η i j ( t ) t a i j ( s ) e λ s y i j ( s ) d s [ a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) e λ η i j ( t ) ] e λ t y i j ( t η i j ( t ) ) + e λ t { C k l N r ( i , j ) C i j k l ( t ) [ f ( x k l ( t τ ( t ) ) ) f ( x k l ( t τ ( t ) ) ) ] x i j ( t ) C k l N r ( i , j ) C i j k l ( t ) f ( x k l ( t τ ( t ) ) ) [ x i j ( t ) x i j ( t ) ] C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) [ g ( x k l ( t u ) ) g ( x k l ( t u ) ) ] d u x i j ( t ) C k l N q ( i , j ) B i j k l ( t ) 0 K i j ( u ) g ( x k l ( t u ) ) d u [ x i j ( t ) x i j ( t ) ] } ,
(3.10)

where ijJ.

Let

Y i j (t)= | e λ t y i j ( t ) t η i j ( t ) t a i j ( s ) e λ s y i j ( s ) d s | ,ijJ.

We define a positive constant M as follows:

M= max i j J { sup s ( , 0 ] Y i j ( s ) } .

Let K be a positive number such that

Y i j (t)M<M+1=Kfor all t(,0],ijJ.
(3.11)

We claim that

Y i j (t)<K,for all t>0,i=1,2,,n.
(3.12)

Otherwise, there must exist ijJ and θ>0 such that

Y i j (θ)=K, Y i j ˜ (t)<K,for all t(,θ), i j ˜ J.
(3.13)

It follows that

e λ t | y i j ˜ ( t ) | | e λ t y i j ˜ ( t ) t η i j ˜ ( t ) t a i j ˜ ( s ) e λ s y i j ˜ ( s ) d s | + | t η i j ˜ ( t ) t a i j ˜ ( s ) e λ s y i j ˜ ( s ) d s | K + a i j ˜ + η i j ˜ + sup s ( , θ ] e λ s | y i j ˜ ( s ) | , for all  t ( , θ ] , i j ˜ J .
(3.14)

Consequently, in view of (3.14) and the fact a i j ˜ + η i j ˜ + <1 ( i j ˜ J), we have

e λ t | y i j ˜ ( t ) | sup s ( , θ ] e λ s | y i j ˜ ( s ) | K 1 a i j ˜ + η i j ˜ + for all t(,θ], i j ˜ J.
(3.15)

Calculating the upper left derivative of Y i j (t), together with (2.3), (3.10), (3.13), (3.15), (T1) and (T2), we obtain

0 D Y i j ( θ ) ( a i j ( θ ) λ ) Y i j ( θ ) + | ( a i j ( θ ) λ ) θ η i j ( θ ) θ a i j ( s ) e λ s y i j ( s ) d s [ a i j ( θ ) ( 1 η i j ( θ ) ) a i j ( θ η i j ( θ ) ) e λ η i j ( θ ) ] e λ θ y i j ( θ η i j ( θ ) ) + e λ θ { C k l N r ( i , j ) C i j k l ( θ ) [ f ( x k l ( θ τ ( θ ) ) ) f ( x k l ( θ τ ( θ ) ) ) ] x i j ( θ ) C k l N r ( i , j ) C i j k l ( θ ) f ( x k l ( θ τ ( θ ) ) ) [ x i j ( θ ) x i j ( θ ) ] C k l N q ( i , j ) B i j k l ( θ ) 0 K i j ( u ) [ g ( x k l ( θ u ) ) g ( x k l ( θ u ) ) ] d u x i j ( θ ) C k l N q ( i , j ) B i j k l ( θ ) 0 K i j ( u ) g ( x k l ( θ u ) ) d u [ x i j ( θ ) x i j ( θ ) ] } | ( a i j ( θ ) λ ) Y i j ( θ ) + ( a i j ( θ ) λ ) K 1 a i j + η i j + a i j + η i j + + | a i j ( θ ) e λ η i j ( θ ) ( 1 η i j ( θ ) ) a i j ( θ η i j ( θ ) ) | e λ ( θ η i j ( θ ) ) | y i j ( θ η i j ( θ ) ) | + C k l N r ( i , j ) | C i j k l ( θ ) | μ f e λ τ ( θ ) e λ ( θ τ ( θ ) ) | y k l ( θ τ ( θ ) ) | | x i j ( θ ) | + C k l N r ( i , j ) | C i j k l ( θ ) | M f e λ θ | y i j ( θ ) | + C k l N q ( i , j ) | B i j k l ( θ ) | 0 e λ u | K i j ( u ) | μ g e λ ( θ u ) | y k l ( θ u ) | d u | x i j ( θ ) | + C k l N q ( i , j ) | B i j k l ( θ ) | 0 | K i j ( u ) | d u M g e λ θ | y i j ( θ ) | { [ ( a i j ( θ ) λ ) ( 1 2 a i j + η i j + ) | a i j ( θ ) e λ η i j ( θ ) ( 1 η i j ( θ ) ) a i j ( θ η i j ( θ ) ) | ] 1 1 a i j + η i j + + C k l N r ( i , j ) | C i j k l ( θ ) | μ f e λ τ ( θ ) 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N r ( i , j ) | C i j k l ( θ ) | M f 1 1 a i j + η i j + + C k l N q ( i , j ) | B i j k l ( θ ) | 0 e λ u | K i j ( u ) | d u μ g 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N q ( i , j ) | B i j k l ( θ ) | 0 | K i j ( u ) | d u M g 1 1 a i j + η i j + } K < η K < 0 ,

which is a contradiction, and it implies that (3.12) holds.

Consequently, using a similar argument as in ( 3.14)-(3.15), we know that

| y i j (t)| e λ t < K 1 a i j + η i j + ,for all tR,ijJ.

That is,

| x i j (t) x i j (t)| K 1 a i j + η i j + e λ t ,for all t>0, and ijJ.

This completes the proof. □

4 An example

In this section, we give an example with numerical simulations to demonstrate the results obtained in previous sections.

Example 4.1 Consider the following SICNNs with time-varying delays in the leakage terms:

d x i j d t = a i j ( t ) x i j ( t η i j ( t ) ) C k l N r ( i , j ) C i j k l f ( x k l ( t sin 2 t ) ) x i j ( t ) C k l N q ( i , j ) B i j k l 0 K i j ( u ) g ( x k l ( t u ) ) d u x i j + L i j ( t ) , i , j = 1 , 2 , 3 ,
(4.1)
[ a 11 a 12 a 13 a 21 a 22 a 23 a 31 a 32 a 33 ]=[ 1 1 3 3 1 3 3 1 3 ],
(4.2)
[ B 11 B 12 B 13 B 21 B 22 B 23 B 31 B 32 B 33 ]=[ C 11 C 12 C 13 C 21 C 22 C 23 C 31 C 32 C 33 ]=[ 0.1 0.2 0.1 0.2 0 0.2 0.1 0.2 0.1 ],
(4.3)
[ η 11 η 12 η 13 η 21 η 22 η 23 η 31 η 32 η 33 ]=0.01[ sin 2 3 t cos 2 3 t sin 2 2 t cos 2 5 t sin 2 5 t cos 2 2 t sin 2 2 t cos 2 3 t sin 2 2 t ],
(4.4)
[ L 11 L 12 L 13 L 21 L 22 L 23 L 31 L 32 L 33 ] = [ 0.7 + 0.24 sin 2 2 t 0.41 + 0.5 cos 2 t 0.74 + 0.2 sin 2 t 0.61 + 0.2 cos 2 t 0.67 + 0.2 sin 2 t 0.75 + 0.2 sin 2 t 0.59 + 0.4 cos 4 t 0.5 + 0.41 sin 2 t 0.76 + 0.2 cos 2 t ] .
(4.5)

Set

λ=0.001,r=q=1, K i j (u)=|sinu| e u ,i=1,2,3,j=1,2,3,

and

f(x)=g(x)= 1 40 ( | x 1 | | x + 1 | ) ,

clearly,

M f = M g = 0.05 , μ f = μ g = 0.05 , C k l N 1 ( 1 , 1 ) C 11 k l = C k l N 1 ( 1 , 1 ) B 11 k l = 0.5 , C k l N 1 ( 1 , 2 ) C 12 k l = C k l N 1 ( 1 , 2 ) B 12 k l = 0.8 , C k l N 1 ( 1 , 3 ) C 13 k l = C k l N 1 ( 1 , 3 ) B 13 k l = 0.5 , C k l N 1 ( 2 , 1 ) C 21 k l = C k l N 1 ( 2 , 1 ) B 21 k l = 0.8 , C k l N 1 ( 2 , 2 ) C 22 k l = C k l N 1 ( 2 , 2 ) B 22 k l = 1.2 , C k l N 1 ( 2 , 3 ) C 23 k l = C k l N 1 ( 2 , 3 ) B 23 k l = 0.8 , C k l N 1 ( 3 , 1 ) C 31 k l = C k l N 1 ( 3 , 1 ) B 31 k l = 0.5 , C k l N 1 ( 3 , 2 ) C 32 k l = C k l N 1 ( 3 , 2 ) B 32 k l = 0.8 , C k l N 1 ( 3 , 3 ) C 33 k l = C k l N 1 ( 3 , 3 ) B 33 k l = 0.5 , 1 a i j + 3 , η i j + 0.01 , a i j + η i j + 0.03 < 1 ,

where ijJ={11,12,13,21,22,23,31,32,33}. Then,

min i j J δ i j ( t ) = min i j J { { a i j ( t ) ( 1 2 a i j + η i j + ) | a i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | min i j J δ i j ( t ) = C k l N r ( i , j ) | C i j k l ( t ) | M f C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g } 1 1 a i j + η i j + } min i j J δ i j ( t ) > 0.5 > 0 , for all  t 0 ,
(4.6)
max i j J { [ ( a i j ( t ) λ ) ( 1 2 a i j + η i j + ) | a i j ( t ) e λ η i j ( t ) ( 1 η i j ( t ) ) a i j ( t η i j ( t ) ) | ] 1 1 a i j + η i j + + C k l N r ( i , j ) | C i j k l ( t ) | μ f e λ τ ( t ) 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N r ( i , j ) | C i j k l ( t ) | M f 1 1 a i j + η i j + + C k l N q ( i , j ) | B i j k l ( t ) | 0 e λ u | K i j ( u ) | d u μ g 1 1 a k l + η k l + 1 1 a i j + η i j + ( L i j δ i j ) + + C k l N q ( i , j ) | B i j k l ( t ) | 0 | K i j ( u ) | d u M g 1 1 a i j + η i j + } < ( 1 λ ) × ( 1 2 × 1 × 0.01 ) + ( 1 × e 0.01 λ 1 + 0.01 ) 1 1 0.01 + 1.2 × 0.05 × e λ × 1 1 1 0.03 1 1 0.03 × 2 + 1.2 × 0.05 × 1 1 0.03 + 1.2 × e λ 1 λ × 0.05 1 1 0.03 1 1 0.03 + 1.2 × 1 × 0.05 × 1 1 0.03 0.6538 < 0.5 , for all  t 0 .
(4.7)

It follows that system (4.1) satisfies all the conditions in Theorem 3.1. Hence, system (4.1) has exactly one positive almost periodic solution. Moreover, the almost periodic solution is globally exponentially stable. The fact is verified by the numerical simulation in Figures 1-3 and their three different initial values, which are φ 11 1, φ 12 3, φ 13 4, φ 21 2, φ 22 5, φ 23 3, φ 33 1, φ 32 2, φ 33 5; φ 11 2, φ 12 1, φ 13 5, φ 21 4, φ 22 2, φ 23 1, φ 33 3, φ 32 4, φ 33 3 and φ 11 2, φ 12 1, φ 13 5, φ 21 4, φ 22 2, φ 23 1, φ 33 3, φ 32 4, φ 33 3, respectively.

Figure 1
figure 1

Numerical solutions of system ( 4.1 ) for different initial values.

Figure 2
figure 2

Numerical solutions of system ( 4.1 ) for different initial values.

Figure 3
figure 3

Numerical solutions of system ( 4.1 ) for different initial values.

Remark 4.1 Since [49] only dealt with SICNNs without leakage delays, and [1419, 23, 24] give no opinions about the problem of almost periodic solutions for SICNNs with time-varying leakage delays. One can observe that all the results in this literature and the references therein can not be applicable to prove the existence and exponential stability of almost periodic solutions for SICNNs (4.1). In this present paper, we employ a novel proof to establish some criteria to guarantee the existence and exponential stability of almost periodic solutions for SICNNs system with time-varying coefficients and leakage delays. What will happen when one can increase the neuron’s order? Whether or not our results and method in this paper are available for this case, it is an interesting problem, and we leave it as our work in the future.

References

  1. Bouzerdoum A, Pinter RB: Shunting inhibitory cellular neural networks: derivation and stability analysis. IEEE Trans. Circuits Syst. I, Fundam. Theory Appl. 1993, 40: 215–221. 10.1109/81.222804

    Article  MathSciNet  Google Scholar 

  2. Bouzerdoum A, Pinter RB: Analysis and analog implementation of directionally sensitive shunting inhibitory cellular neural networks. SPIE 1473. Visual Information Processing: From Neurons to Chips 1991, 29–38.

    Chapter  Google Scholar 

  3. Bouzerdoum A, Pinter RB: Nonlinear lateral inhibition applied to motion detection in the fly visual system. In Nonlinear Vision. Edited by: Pinter RB, Nabet B. CRC Press, Boca Raton; 1992:423–450.

    Google Scholar 

  4. Chen A, Cao J, Huang L: Almost periodic solution of shunting inhibitory CNNs with delays. Phys. Lett. A 2002, 298: 161–170. 10.1016/S0375-9601(02)00469-3

    Article  MathSciNet  Google Scholar 

  5. Chen A, Huang X: Almost periodic attractor of delayed neural networks with variable coefficients. Phys. Lett. A 2005, 340(1–4):104–120. 10.1016/j.physleta.2005.04.021

    Article  Google Scholar 

  6. Cai M, Zhang H, Yuan Z: Positive almost periodic solutions for shunting inhibitory cellular neural networks with time-varying delays. Math. Comput. Simul. 2008, 78(4):548–558. 10.1016/j.matcom.2007.08.001

    Article  MathSciNet  Google Scholar 

  7. Shao J: Anti-periodic solutions for shunting inhibitory cellular neural networks with time-varying delays. Phys. Lett. A 2008, 372(30):5011–5016. 10.1016/j.physleta.2008.05.064

    Article  Google Scholar 

  8. Fan Q, Shao J: Positive almost periodic solutions for shunting inhibitory cellular neural networks with time-varying and continuously distributed delays. Commun. Nonlinear Sci. Numer. Simul. 2010, 15(6):1655–1663. 10.1016/j.cnsns.2009.06.026

    Article  MathSciNet  Google Scholar 

  9. Zhao C, Fan Q, Wang W: Anti-periodic solutions for shunting inhibitory cellular neural networks with time-varying coefficients. Neural Process. Lett. 2010, 31: 259–267. 10.1007/s11063-010-9136-y

    Article  Google Scholar 

  10. Kosko B: Neural Networks and Fuzzy Systems. Prentice Hall, New Delhi; 1992.

    Google Scholar 

  11. Haykin S: Neural Networks. Prentice Hall, New Jersey; 1999.

    Google Scholar 

  12. Gopalsamy K: Stability and Oscillations in Delay Differential Equations of Population Dynamics. Kluwer Academic, Dordrecht; 1992.

    Book  Google Scholar 

  13. Gopalsamy K: Leakage delays in BAM. J. Math. Anal. Appl. 2007, 325: 1117–1132. 10.1016/j.jmaa.2006.02.039

    Article  MathSciNet  Google Scholar 

  14. Li X, Cao J: Delay-dependent stability of neural networks of neutral type with time delay in the leakage term. Nonlinearity 2010, 23: 1709–1726. 10.1088/0951-7715/23/7/010

    Article  MathSciNet  Google Scholar 

  15. Li X, Rakkiyappan R, Balasubramaniam P: Existence and global stability analysis of equilibrium of fuzzy cellular neural networks with time delay in the leakage term under impulsive perturbations. J. Franklin Inst. 2011, 348: 135–155. 10.1016/j.jfranklin.2010.10.009

    Article  MathSciNet  Google Scholar 

  16. Balasubramaniam P, Vembarasan V, Rakkiyappan R: Leakage delays in T - S fuzzy cellular neural networks. Neural Process. Lett. 2011, 33: 111–136. 10.1007/s11063-010-9168-3

    Article  Google Scholar 

  17. Liu B: Global exponential stability for BAM neural networks with time-varying delays in the leakage terms. Nonlinear Anal., Real World Appl. 2013, 14: 559–566. 10.1016/j.nonrwa.2012.07.016

    Article  MathSciNet  Google Scholar 

  18. Chen Z: A shunting inhibitory cellular neural network with leakage delays and continuously distributed delays of neutral type. Neural Comput. Appl. 2012. 10.1007/s00521-012-1200-2

    Google Scholar 

  19. Zhang H: Global exponential stability of almost periodic solutions for SICNNs with continuously distributed leakage delays. Abstr. Appl. Anal. 2013., 2013: Article ID 307981 10.1155/2013/307981

    Google Scholar 

  20. Fink AM Lecture Notes in Mathematics 377. In Almost Periodic Differential Equations. Springer, Berlin; 1974.

    Google Scholar 

  21. He CY: Almost Periodic Differential Equation. Higher Education Publishing House, Beijing; 1992. (in Chinese)

    Google Scholar 

  22. Hino Y, Murakami S, Nai T Lecture in Mathematics 1473. In Functional Differential Equations with Infinite Delay. Springer, Berlin; 1991.

    Google Scholar 

  23. Zhang H, Shao J: Almost periodic solutions for cellular neural networks with time-varying delays in leakage terms. Appl. Comput. Math. 2013, 219(24):11471–11482. 10.1016/j.amc.2013.05.046

    Article  MathSciNet  Google Scholar 

  24. Balasubramaniam P, Vembarasan V: Global asymptotic stability results for BAM neural networks of neutral-type with time delays in the leakage term under impulsive perturbations. Int. J. Comput. Math. 2011, 88: 3271–3291. 10.1080/00207160.2011.591388

    Article  MathSciNet  Google Scholar 

Download references

Acknowledgements

This work was supported by the construct program of the key discipline in Hunan province (Mechanical Design and Theory), the Scientific Research Fund of Hunan Provincial Natural Science Foundation of PR China (Grant No. 11JJ6006), the Natural Scientific Research Fund of Hunan Provincial Education Department of PR China (Grants No. 11C0916, 11C0915), and the Natural Scientific Research Fund of Zhejiang Provincial Education Department of PR China (Grants No. LY12A01018, Y6110436).

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Jianying Shao.

Additional information

Competing interests

The authors declare that they have no competing interests.

Authors’ contributions

BL gave the proof of Lemmas 2.1-2.2 and drafted the manuscript. JS proved Theorem 3.1 and gave the example to illustrate the effectiveness of the obtained results. All authors read and approved the final manuscript.

Authors’ original submitted files for images

Below are the links to the authors’ original submitted files for images.

Authors’ original file for figure 1

Authors’ original file for figure 2

Authors’ original file for figure 3

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and permissions

About this article

Cite this article

Liu, B., Shao, J. Almost periodic solutions for SICNNs with time-varying delays in the leakage terms. J Inequal Appl 2013, 494 (2013). https://doi.org/10.1186/1029-242X-2013-494

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/1029-242X-2013-494

Keywords