Open Access

Exponential stability for a stochastic delay neural network with impulses

Advances in Difference Equations20142014:250

https://doi.org/10.1186/1687-1847-2014-250

Received: 11 February 2014

Accepted: 12 September 2014

Published: 24 September 2014

Abstract

In this paper, we study the exponential stability for a stochastic neural network with impulses. By employing fixed point theory and some analysis techniques, sufficient conditions are derived for achieving the required result.

MSC:34K20.

Keywords

neural networkexponential stabilityimpulsefixed point theory

1 Introduction

Neural networks have found important applications in various areas such as combinatorial optimization, signal processing, pattern recognition, and solving nonlinear algebraic equations. We notice that a lot of practical systems have the phenomenon of time delay, and many scholars have paid much attention to time delay systems [18]. As is well known, stochastic functional differential systems which include stochastic delay differential systems have been widely used since stochastic modeling plays an important role in many branches of science and engineering [9]. Consequently, the stability analysis of these systems has received a lot of attention [1012].

In some applications, besides delay and stochastic effects, impulsive effects are also likely to exist [13, 14]; they could stabilize or destabilize the systems. Therefore, it is of interest to take delay effects, stochastic effects, and impulsive effects into account when studying the dynamical behavior of neural networks.

In [11], Guo et al. studied the exponential stability for a stochastic neutral cellular neural network without impulses and obtained new criteria for exponential stability in mean square of the considered neutral cellular neural network by using fixed point theory. To the best of the authors’ knowledge there are only a few papers where fixed point theory is used to discuss the stability of stochastic neural networks. In this paper, we will study the exponential stability for a stochastic neural network with impulses by the contraction mapping theorem and Krasnoselskii’s fixed point theorem.

2 Some preliminaries

Throughout this paper, unless otherwise specified, we let ( Ω , F , P ) be a complete probability space with a filtration { F t } t 0 satisfying the usual conditions, i.e. it is right continuous and F 0 contains all P-null sets, C F 0 b ( [ τ , 0 ] ; R n ) be the family of all bounded, F 0 -measurable functions. Let R n denote the n-dimensional real space equipped with Euclidean norm | | 1 . B = [ b i j ( t ) ] n × n denote a matrix, its norm is denoted by B ( t ) 3 = i , j = 1 n | b i , j ( t ) | .

In this paper, by using fixed point theory, we discuss the stability of the impulsive stochastic delayed neural networks:
{ d x ( t ) = [ ( A + A ( t ) ) x ( t ) + f ( t , x ( t ) , x ( t τ 1 ) ) ] d t d x ( t ) = + σ ( t , x ( t ) , x ( t τ 2 ) ) d ω ( t ) , t t k , x ( t k ) = I k ( x ( t k ) ) , t = t k , k Z + , x ( t ) = ψ ( t ) , τ t 0 ,
(2.1)

where τ = max { τ 1 , τ 2 } , τ 1 and τ 2 are positive constant. t ψ ( t ) C ( [ τ , 0 ] , L F 0 p ( Ω , R n ) ) with the norm defined by ψ 2 = sup τ θ 0 | ψ ( θ ) | 1 , x ( t ) = ( x 1 ( t ) , x 2 ( t ) , , x n ( t ) ) T is the state vector, A = diag ( a 1 , a 2 , , a n ) > 0 (i.e. a i > 0 , i = 1 , 2 , , n ) is the connection weight constant matrix with appropriate dimensions, and A ( t ) represents the time-varying parameter which is uncertain with | A ( t ) | 3 bounded.

Here f ( t , u 1 , u 2 ) C ( R × R n × R n ) is the neuron activation function with f ( t , 0 , 0 ) = 0 and ω ( t ) = ( ω 1 ( t ) , ω 2 ( t ) , , ω m ( t ) ) T R m is an m-dimensional Brownian motion defined on ( Ω , F , P ) . The stochastic disturbance term, σ ( t , u 1 , u 2 ) C ( R × R n × R n ) , can be viewed as stochastic perturbations on the neuron states and delayed neuron states with σ ( t , 0 , 0 ) = 0 . x ( t k ) = I k ( x ( t k ) ) = x ( t k + ) x ( t k ) is the impulse at moment t k , and t 1 < t 2 < is strictly increasing sequence such that lim k t k = + , x ( t k + ) and x ( t k ) stand for the right-hand and left-hand limit of x ( t ) at t = t k , respectively. I k ( x ( t k ) ) shows the abrupt change of x ( t ) at the impulsive moment t k and I k ( ) C ( L F t p ( Ω ; R n ) , L F t p ( Ω ; R n ) ) .

The local Lipschitz condition and the linear growth condition on the function f ( t , , ) and σ ( t , , ) guarantee the existence and uniqueness of a global solution for system (2.1); we refer to [9] for detailed information. Clearly, system (2.1) admits a trivial solution x ( t ; 0 , 0 ) 0 .

Definition 2.1 System (2.1) is said to be exponentially stable in mean square for all admissible uncertainties if there exists a solution x of (2.1) and there exist a pair of positive constants β and μ with
E | x ( t ) | 1 2 μ E ψ 2 2 e β t , t 0 .

In order to prove the exponentially stability in mean square of system (2.1), we need the following lemma.

Lemma 2.1 (Krasnoselskii)

Suppose that Ω is Banach space and X is a bounded, convex, and closed subset of Ω. Let U , S : X Ω satisfy the following conditions:
  1. (1)

    U x + S y X for any x , y X ;

     
  2. (2)

    U is contraction mapping;

     
  3. (3)

    S is continuous and compact.

     

Then U + S has a fixed point in X.

Lemma 2.2 ( C p inequality)

If X L p ( Ω , R n ) , then
E | i = 1 n X i | p i = 1 n E | X i | p , for  0 < p 1
and
E | i = 1 n X i | p n p 1 i = 1 n E | X i | p , for  p > 1 .

3 Main results

Let ( ß , ß ) be the Banach space of all F 0 -adapted processes φ ( t , ω ) : [ τ , ) × Ω R n such that φ : [ τ , ) L F 0 p ( Ω , R n ) is continuous on t t k , lim t t k φ ( t , ) and lim t t k + φ ( t , ) exist, and lim t t k φ ( t , ) = φ ( t k , ) , k = 1 , 2 ,  ; we have
φ ß : = sup t 0 E | φ ( t ) | 1 2 , for  φ ß .

Let Λ be the set of functions φ ß such that φ ( s ) = ψ ( s ) on s [ τ , 0 ] and e α t E | φ ( t , ω ) | 1 2 0 as t . It is clear that Λ is a bounded, convex, and closed subset of ß.

To obtain our results, we suppose the following conditions are satisfied:
  • (H1) there exist μ 1 , μ 2 > 0 such that | f ( t , u , v ) f ( t , u ¯ , v ¯ ) | 1 μ 1 | u u ¯ | 1 + μ 2 | v v ¯ | 1 ;

  • (H2) there exist ν 1 , ν 2 > 0 such that | σ ( t , u , v ) σ ( t , u ¯ , v ¯ ) | 1 2 ν 1 | u u ¯ | 1 2 + ν 2 | v v ¯ | 1 2 ;

  • (H3) there exists an α > 0 such that α < min { a 1 , a 2 , , a n } ;

  • (H4) for k = 1 , 2 , 3 ,  , the mapping I k ( ) satisfies I k ( 0 ) 0 and is globally Lipschitz function with Lipschitz constants p k ;

  • (H5) there exists a constant ρ such that inf k = 1 , 2 , { t k t k 1 } ρ ;

  • (H6) there exists constant p such that p k p ρ , for i N and k = 1 , 2 ,  .

The solution x ( t ) : = x ( t ; 0 , ψ ) of system (2.1) is, for the time t, a piecewise continuous vector-valued function with the first kind discontinuity at the points t k ( k = 1 , 2 , ), where it is left continuous, i.e.,
x ( t k ) = x ( t k ) , x ( t k + ) = x ( t k ) + I k ( x ( t k ) ) , k = 1 , 2 , .
Theorem 3.1 Assume (H1)-(H6) hold and the following condition is satisfied:
( P 1 ) 3 { n 2 2 λ min ( A ) [ ( A ( t ) 3 + μ 1 + μ 2 ) 2 + ν 1 + ν 2 ] + n 2 p 2 ( 1 λ min ( A ) + ρ ) 2 } < 1 ,

then system (2.1) is exponentially stable in mean square for all admissible uncertainties, that is, e α t E | x ( t ) | 1 2 0 , as t .

Proof System (2.1) is equivalent to
x ( t ) = exp ( A t ) ψ ( 0 ) + 0 t exp A ( s t ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + 0 t exp A ( s t ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) + 0 < t k < t exp [ A ( t t k ) ] I k ( x ( t k ) ) .
(3.1)
Let
J 1 ( t ) : = exp ( A t ) ψ ( 0 ) , J 2 ( t ) : = 0 t exp A ( s t ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s , J 3 : = 0 t exp A ( s t ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) , J 4 ( t ) : = 0 < t k < t exp [ A ( t t k ) ] I k ( x ( t k ) ) .

Define an operator by ( Q x ) ( t ) = ψ ( t ) for t [ τ , 0 ] , and for t 0 , we define ( Q x ) ( t ) : = J 1 ( t ) + J 2 ( t ) + J 3 ( t ) + J 4 ( t ) (i.e. the right-hand side of (3.1)). From the definition of Λ, we have E | φ ( t ) | 2 < , for all t 0 , and φ Λ .

Next, we prove that Q Λ Λ . It is clear that ( Q x ) ( t ) is continuous on [ τ , 0 ] . For a fixed time t > 0 , it is easy to check that J 1 ( t ) , J 2 ( t ) , J 4 ( t ) are continuous in mean square on the fixed time t t k , for k = 1 , 2 ,  . In the following, we check the mean square continuity of J 3 ( t ) on the fixed time t t k ( k = 1 , 2 , ).

Let x Λ and r R such that | r | sufficiently small; we obtain
E | J 3 ( t + r ) J 3 ( t ) | 1 2 = E | 0 t + r exp A ( s t r ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) 0 t exp A ( s t ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) | 1 2 = E | 0 t [ exp A ( s t r ) exp A ( s t ) ] σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) + t t + r exp A ( s t r ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) | 1 2 2 ( 0 t [ exp A ( s t r ) exp A ( s t ) ] 3 2 × E | σ ( s , x ( s ) , x ( s τ 2 ) ) | 1 2 d s + t t + r exp A ( s t r ) 3 2 E | σ ( s , x ( s ) , x ( s τ 2 ) ) | 1 2 d s ) 0 as  r 0 .

Hence, ( Q x ) ( t ) is continuous in mean square on the fixed time t t k , for k = 1 , 2 ,  . On the other hand, as t = t k , it is easy to check that J 1 ( t ) , J 2 ( t ) , J 3 ( t ) are continuous in mean square on the fixed time t = t k .

Let r < 0 be small enough; we have
E | J 4 ( t k + r ) J 4 ( t k ) | 1 2 = E | 0 < t m < t k + r exp [ A ( t k + r t m ) ] I m ( x ( t m ) ) 0 < t m < t k exp [ A ( t k t m ) ] I m ( x ( t m ) ) | 1 2 E | [ exp ( A ( t k + r ) ) exp ( A t k ) ] × 0 < t m < t k exp ( A t m ) I m ( x ( t m ) ) | 1 2 ,

which implies that lim r 0 E | J 4 ( t k + r ) J 4 ( t k ) | 1 2 = 0 .

Let r > 0 be small enough, we have
E | J 4 ( t k + r ) J 4 ( t k ) | 1 2 = E | [ exp ( A ( t k + r ) ) exp ( A t k ) ] × 0 < t m < t k exp ( A t m ) I m ( x ( t m ) ) + exp ( A r ) I k ( x ( t k ) ) | 1 2 ,

which implies that lim r 0 + E | J 4 ( t k + r ) J 4 ( t k ) | 1 2 = E | I k ( x ( t k ) ) | 1 2 .

Hence, we see that ( Q x ) ( t ) : [ τ , ) L F 0 P ( Ω , R n ) is continuous in mean square on t t k , and for t = t k , lim t t k + ( Q x ) ( t ) and lim t t k ( Q x ) ( t ) exist. Furthermore, we also obtain lim t t k ( Q x ) ( t ) = ( Q x ) ( t k ) lim t t k + ( Q x ) ( t ) .

It follows from (3.1) that
e α t E | ( Q x ) ( t ) | 1 2 4 e α t i = 1 4 E | J i ( t ) | 1 2 .

By (H3), it is easy to see e α t E | J 1 ( t ) | 1 2 0 , as t . Now, we prove e α t E | J 2 ( t ) | 1 2 0 , e α t E | J 3 ( t ) | 1 2 0 , and e α t E | J 4 ( t ) | 1 2 0 , as t .

Note, for any ϵ > 0 , there exists t > 0 such that s t τ implies that e α s E | x ( s ) | 1 2 < ϵ . Hence, we have from (H1), (H3)
e α t E | J 2 ( t ) | 1 2 = e α t E | 0 t exp A ( s t ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s | 1 2 e α t E 0 t exp A ( s t ) 3 2 | [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] | 1 2 d s = e α t E 0 t exp A ( s t ) 3 2 | [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] | 1 2 d s + e α t E t t exp A ( s t ) 3 2 | [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] | 1 2 d s e α t E 0 t exp A ( s t ) 3 2 [ A ( s ) 3 | x ( s ) | 1 + μ 1 | x ( s ) | 1 + μ 2 | x ( s τ 1 ) | 1 ] 2 d s + e α t E t t exp A ( s t ) 3 2 [ A ( s ) 3 | x ( s ) | 1 + μ 1 | x ( s ) | 1 + μ 2 | x ( s τ 1 ) | 1 ] 2 d s e α 2 λ min ( A ) t n 2 ( A ( s ) 3 + μ 1 + μ 2 ) 2 × E ( sup τ s t | x ( s ) | 1 2 ) 0 t e 2 λ min ( A ) s d s + 2 e α t n 2 t t [ ( A ( s ) 3 + μ 1 ) 2 e α s e α s E ( | x ( s ) | 1 2 ) + μ 2 2 e α ( s τ 1 ) e α ( s τ 1 ) E | x ( s τ 1 ) | 1 2 ] e 2 λ min ( A ) ( s t ) d s ,

where exp A ( s t ) 3 = i = 1 n e a i ( s t ) , λ min ( A ) represents the minimal eigenvalue of A. Thus, we have e α t E | J 2 ( t ) | 1 2 0 as t .

From (H2) and (H3), we have
e α t E | J 3 ( t ) | 1 2 = e α t E | 0 t exp A ( s t ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s ) | 1 2 e α t E 0 t exp A ( s t ) 3 2 | σ ( s , x ( s ) , x ( s τ 2 ) ) | 1 2 d s = e α t E 0 t exp A ( s t ) 3 2 | σ ( s , x ( s ) , x ( s τ 2 ) ) | 1 2 d s + e α t E t t exp A ( s t ) 3 2 | σ ( s , x ( s ) , x ( s τ 2 ) ) | 1 2 d s e α 2 λ min ( A ) t n 2 ( ν 1 + ν 2 ) E ( sup τ s t | x ( s ) | 1 2 ) 0 t e 2 λ min ( A ) s d s + e α t n 2 ( ν 1 + ν 2 ) t t e α s e α s E ( sup τ s t | x ( s ) | 1 2 ) e 2 λ min ( A ) ( s t ) d s 0 as  t .
As x ( t ) Λ , we have lim t e α t E | x ( t ) | 1 2 0 . Then, for any ϵ > 0 , there exists a nonimpulsive point T > 0 such that s T implies e α t E | x ( t ) | 1 2 < ϵ . It then follows from the conditions (H4)-(H6) that
e α t E | J 4 ( t ) | 1 2 = e α t E | 0 < t k < t exp [ A ( t t k ) ] I k ( x ( t k ) ) | 1 2 e α t E | 0 < t k < t exp A ( t k t ) 3 | I k ( x ( t k ) ) | 1 | 2 e α t E | 0 < t k < t n e λ min ( A ) ( t k t ) p k | x ( t k ) | 1 | 2 e α t E | 0 < t k < T n e λ min ( A ) ( t k t ) p k | x ( t k ) | 1 + T < t k < t n e λ min ( A ) ( t k t ) p ρ | x ( t k ) | 1 | 2 2 e α t [ E | 0 < t k < T n e λ min ( A ) ( t k t ) p k | x ( t k ) | 1 | 2 + E | T < t k < t n e λ min ( A ) ( t k t ) p ρ | x ( t k ) | 1 | 2 ] 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 2 e α t E | T < t r < t k n e λ min ( A ) ( t r t ) p ( t r + 1 t r ) | x ( t k ) | 1 + n e λ min ( A ) ( t k t ) p ρ | x ( t k ) | 1 | 2 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 2 e α t E | T t n e λ min ( A ) ( s t ) p | x ( s ) | 1 d s + n p ρ | x ( t ) | 1 | 2 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 4 e α t E | T t n e λ min ( A ) ( s t ) p | x ( s ) | 1 d s | 2 + 4 e α t E | n p ρ | x ( t ) | 1 | 2 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 4 n 2 p 2 e ( α 2 λ min ( A ) ) t T t e ( 2 λ min ( A ) α ) s e α s E | x ( s ) | 1 2 d s + 4 n 2 p 2 ρ 2 ϵ 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 4 n 2 p 2 e ( α 2 λ min ( A ) ) t ϵ T t e ( 2 λ min ( A ) α ) s d s + 4 n 2 p 2 ρ 2 ϵ 2 e ( α 2 λ min ( A ) ) t n 2 p k 2 E | 0 < t k < T e λ min ( A ) t k | x ( t k ) | 1 | 2 + 4 n 2 p 2 ϵ 2 λ min ( A ) α + 4 n 2 p 2 ρ 2 ϵ 0 as  t .

Thus we conclude that Q : Λ Λ .

Finally, we prove that Q is a contraction mapping. For any φ , ϕ Λ , we obtain
sup t τ E | ( Q φ ) ( t ) ( Q ϕ ) ( t ) | 1 2 = sup t τ { E | 0 t exp A ( s t ) [ A ( s ) ( φ ( s ) ϕ ( s ) ) + ( f ( s , φ ( s ) , φ ( s τ 1 ) ) f ( s , ϕ ( s ) , ϕ ( s τ 1 ) ) ) ] d s + 0 t exp A ( s t ) [ σ ( s , φ ( s ) , φ ( s τ 2 ) ) σ ( s , ϕ ( s ) , ϕ ( s τ 2 ) ) ] d ω ( s ) + 0 < t k < t e A ( t t k ) [ I k ( φ ( t k ) ) I k ( ϕ ( t k ) ) ] | 1 2 } 3 sup t τ { E 0 t exp A ( s t ) 3 2 [ A ( s ) 3 | φ ( s ) ϕ ( s ) | 1 + μ 1 | φ ( s ) ϕ ( s ) | 1 + μ 2 | φ ( s τ 1 ) ϕ ( s τ 1 ) | 1 ] 2 d s + E 0 t exp A ( s t ) 3 2 [ ν 1 | φ ( s ) ϕ ( s ) | 1 2 + ν 2 | φ ( s τ 2 ) ϕ ( s τ 2 ) | 1 2 ] d s + E | 0 < t k < t exp A ( t k t ) 3 p k | φ ( t k ) ϕ ( t k ) | 1 | 2 } 3 { n 2 2 λ min ( A ) [ ( A ( t ) 3 + μ 1 + μ 2 ) 2 + ν 1 + ν 2 ] + n 2 p 2 E | 0 < t k < t e λ min ( A ) ( t k t ) ρ | 1 2 } sup t τ E | φ ( s ) ϕ ( s ) | 1 2 3 { n 2 2 λ min ( A ) [ ( A ( t ) 3 + μ 1 + μ 2 ) 2 + ν 1 + ν 2 ] + n 2 p 2 E | 0 < t r < t k e λ min ( A ) ( t r t ) ( t r + 1 t r ) + e λ min ( A ) ( t k t ) ρ | 1 2 } sup t τ E | φ ( s ) ϕ ( s ) | 1 2 3 { n 2 λ min ( A ) [ ( A ( t ) 3 + μ 1 + μ 2 ) 2 + ν 1 + ν 2 ] + n 2 p 2 ( 1 λ min ( A ) + ρ ) 2 } sup t τ E | φ ( s ) ϕ ( s ) | 1 2 .

From the condition (P1), we find that Q is a contraction mapping. Hence, by the contraction mapping principle, we see that Q has a unique fixed point x ( t ) , which is a solution of (2.1) with x ( t ) = ψ ( t ) as t [ τ , 0 ] and e α t E | x ( t ) | 1 2 0 as t . □

The second result is established using Krasnoselskii’s fixed point theorem.

Theorem 3.2 Assume (H1)-(H6) hold and the following condition is satisfied:
( P 2 ) n 2 2 λ min ( A ) ( ν 1 + ν 2 ) < 1 ,

then system (2.1) is exponentially stable in mean square for all admissible uncertainties, that is, e α t E | x ( t ) | 1 2 0 , as t .

Proof For x X , define the operators U : X X and S : X X , respectively, by
( U x ) ( t ) = exp ( A t ) ψ ( 0 ) + 0 t exp A ( s t ) σ ( s , x ( s ) , x ( s τ 2 ) ) d ω ( s )
and
( S x ) ( t ) = 0 t exp A ( s t ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + 0 < t k < t exp [ A ( t t k ) ] I k ( x ( t k ) ) .

By the proof of Theorem 3.1, we can verify that S x + U y Λ when x , y Λ and S is mean square continuous.

Next, we show that U is a contraction mapping. For x , y Λ , we have
sup t τ | U ( x ) ( t ) U ( y ) ( t ) | 1 2 = sup t τ E | 0 t exp A ( s t ) [ σ ( s , x ( s ) , x ( s τ 2 ) ) σ ( s , y ( s ) , y ( s τ 2 ) ) ] d ω ( s ) | 1 2 sup t τ E 0 t exp A ( s t ) 3 2 [ ν 1 | x ( s ) y ( s ) | 1 2 + ν 2 | x ( s τ 2 ) y ( s τ 2 ) | 1 2 ] d s n 2 2 λ min ( A ) sup t τ E | x ( t ) y ( t ) | 1 2 .

From the condition (P2), we find that U is a contraction mapping.

Finally, we prove that S is compact.

Let D Λ be a bounded set: | x | 1 M , x D , we have
| ( S x ) ( t ) | 1 = | 0 t exp A ( s t ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + 0 < t k < t exp [ A ( t t k ) ] I k ( x ( t k ) ) | 1 0 t exp A ( s t ) 3 [ A ( s ) 3 | x ( s ) | 1 + μ 1 | x ( s ) | 1 + μ 2 | x ( s τ 1 ) | 1 ] d s + 0 < t k < t exp A ( t k t ) p ρ | x ( t k ) | 1 ( A ( s ) 3 + μ 1 + μ 2 ) M 0 t e λ min A ( s t ) d s + n p ρ M 1 λ min ( A ) ( A ( s ) 3 + μ 1 + μ 2 + n p ρ ) M .

Therefore, we can conclude that Sx is uniformly bounded.

Further, let x D and t ̲ , t ¯ [ t k 1 , t k ] , with t ̲ < t ¯ ; we have
| ( S x ) ( t ¯ ) ( S x ) ( t ̲ ) | 1 = | 0 t ¯ exp A ( s t ¯ ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + 0 < t m < t ¯ exp [ A ( t ¯ t m ) ] I m ( x ( t m ) ) 0 t ̲ exp A ( s t ̲ ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s 0 < t m < t ̲ exp [ A ( t ̲ t m ) ] I m ( x ( t m ) ) | 1 = | 0 t ̲ [ exp A ( s t ¯ ) exp A ( s t ̲ ) ] × [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + t ̲ t ¯ exp A ( s t ¯ ) [ A ( s ) x ( s ) + f ( s , x ( s ) , x ( s τ 1 ) ) ] d s + 0 < t m < t ¯ [ exp A ( t m t ¯ ) exp A ( t m t ̲ ) ] I m ( x ( t m ) ) | 1 0 as  t ̲ t ¯ .

Thus, the equicontinuity of S is obtained. According to the PC-type Ascoli-Arzela lemma [[15], Lemma 2.4], S ( D ) is relatively compact in Λ. Therefore S is compact. By Lemma 2.1, U + S has a fixed point x in Λ and we note x ( s ) = ( U + S ) ( s ) on [ τ , 0 ] and e α t E | x ( t ) | 1 2 0 as t . This completes the proof. □

Declarations

Acknowledgements

This research was supported by the National Nature Science Foundation of China (No. 11371027); National Nature Science Foundation of China, Tian Yuan (No. 11326115); Research Project for Academic Innovation (No. yfc100002); Program of Natural Science of Colleges of Anhui Province (No. KJ2013A032); Program of Natural Science Research in Anhui Universities (No. KJ2011A020 and No. KJ2012A019); Special Research Fund for the Doctoral Program of the Ministry of Education of China (No. 20123401120001 and No. 20103401120002).

Authors’ Affiliations

(1)
School of Mathematical Science, Anhui University
(2)
Department of Mathematics and Physics, Anhui Jianzhu University

References

  1. Huang H, Cao JD: Exponential stability analysis of uncertain stochastic neural networks with multiple delays. Nonlinear Anal., Real World Appl. 2007, 8: 646-653. 10.1016/j.nonrwa.2006.02.003MathSciNetView ArticleGoogle Scholar
  2. Wang ZD, Liu YR, Liu XH: On global asymptotic stability analysis of neural networks with discrete and distributed delays. Phys. Lett. A 2005, 345(5-6):299-308.View ArticleGoogle Scholar
  3. Wang ZD, Lauria S, Fang JA, Liu XH: Exponential stability of uncertain stochastic neural networks with mixed time-delays. Chaos Solitons Fractals 2007, 32: 62-72. 10.1016/j.chaos.2005.10.061MathSciNetView ArticleGoogle Scholar
  4. Wan L, Sun J: Mean square exponential stability of stochastic delayed Hopfield neural networks. Phys. Lett. A 2005, 343: 306-318. 10.1016/j.physleta.2005.06.024View ArticleGoogle Scholar
  5. Zheng Z: Theory of Functional Differential Equations. Anhui Education Press, Hefei; 1994.Google Scholar
  6. Wei J: The Degeneration Differential Systems with Delay. Anhui University Press, Hefei; 1998.Google Scholar
  7. Wei J: The constant variation formulae for singular fractional differential systems with delay. Comput. Math. Appl. 2010, 59(3):1184-1190. 10.1016/j.camwa.2009.07.010MathSciNetView ArticleGoogle Scholar
  8. Wei J: Variation formulae for time varying singular fractional delay differential systems. Fract. Differ. Calc. 2011, 1(1):105-115.MathSciNetView ArticleGoogle Scholar
  9. Mao X: Stochastic Differential Equations and Applications. Horwood, New York; 1977.Google Scholar
  10. Mao X: Razumikhin-type theorems on exponential stability of stochastic functional differential equations. Stoch. Process. Appl. 1996, 65: 233-250. 10.1016/S0304-4149(96)00109-3View ArticleGoogle Scholar
  11. Guo C, O’Regan D, Deng F, Agarwal RP: Fixed points and exponential stability for a stochastic neutral cellular neural network. Appl. Math. Lett. 2013, 26: 849-853. 10.1016/j.aml.2013.03.011MathSciNetView ArticleGoogle Scholar
  12. Luo J: Exponential stability for stochastic neutral partial functional differential equations. J. Math. Anal. Appl. 2009, 355: 414-425. 10.1016/j.jmaa.2009.02.001MathSciNetView ArticleGoogle Scholar
  13. Peng S, Jia B: Some criteria on p th moment stability of impulsive stochastic functional differential equations. Stat. Probab. Lett. 2010, 80: 1085-1092. 10.1016/j.spl.2010.03.002MathSciNetView ArticleGoogle Scholar
  14. Zhang YT, Luo Q: Global exponential stability of impulsive cellular neural networks with time-varying delays via fixed point theory. Adv. Differ. Equ. 2013., 2013: Article ID 23Google Scholar
  15. Bainov DD, Simeonov PS: Impulsive Differential Equations: Periodic Solutions and Applications. Longman, New York; 1993.Google Scholar

Copyright

© Zhang et al.; licensee Springer. 2014

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.