This paper deals with the stability problem for a class of impulsive neural networks. Some sufficient conditions which can guarantee the globally exponential stability of the addressed models with given convergence rate are derived by using Lyapunov function and impulsive analysis techniques. Finally, an example is given to show the effectiveness of the obtained results.
1. Introduction
Recently, special interest has been devoted to the dynamics analysis of neural networks due to their potential applications in different areas of science. Particularly, there has been a significant development in the theory of neural networks with impulsive effects [1–9], since such neural networks with impulsive effect can be used as an appropriate description of the phenomena of abrupt qualitative dynamical changes of essential continuous time systems. Based on the theory of impulsive differential equations [10–17], some sufficient conditions guaranteeing the exponential stability are derived [18–24]. For example, in [8], the author has obtained a criterion of exponential stability for a Hopfield neural network with periodic coefficients; in [18], by constructing the extended impulsive delayed Halanay inequality and Lyapunov functional methods, authors have got some sufficient conditions ensuring exponential stability of the unique equilibrium point of impulsive Hopfield neural networks with time delays. They all have obtained exponential stability for some kinds of neural networks through different methods. However, most of the existing results about the exponential stability of impulsive neural networks have a common feature that the exponential convergence rate cannot be derived, or derived but not the given one [8, 18, 23, 24]. The purpose of this paper is to establish some criteria which can guarantee the globally exponential stability of impulsive neural networks with the given convergence rate by using Lyapunov function and impulsive analysis techniques. This work is organized as follows. In Section 2, we introduce some basic definitions and notations. In Section 3, the main results are presented. In Section 4, an example is discussed to illustrate the results.
2. Preliminaries
Let ℝ denote the set of real numbers, ℝ+ denote the set of nonnegative real numbers, ℤ+ denote the set of positive integers and ℝn denote the n-dimensional real space equipped with the Euclidean norm ∥·∥.
Consider the following impulsive neural networks:
(1)x˙i(t)=-ai(t)xi(t)+∑j=1nbij(t)fj(xj(t))+Ii(t),t≥t0,t≠tk,Δxi|t=tk=xi(tk)-xi(tk-),k∈ℤ+,i∈Λ,
where Λ={1,2,…,n}. n≥2 corresponds to the number of units in a neural network; the impulse times tk satisfy 0≤t0<t1<⋯<tk<⋯, limk→∞tk=∞; xi corresponds to the state of the neurons, fj denotes the measures of response to its incoming potentials of the unit j at time t; Ii(t) is the input of the unit i at time t. PC[I,ℝ]≜{φ:I→ℝ∣φ(t+)=φ(t) for t∈I, φ(t-) exists for t∈I, φ(t-)=φ(t) for all but points tk∈I}, where I⊂ℝ is an interval, φ(t+) and φ(t-) denote the left limit and right limit of function φ(t), respectively. ai(t)> 0, bij(t),Ii(t)∈PC[[t0,+∞),ℝ]. For given t0,x(t0)=x0=(x10,x20,…,xn0)∈ℝn, we denote by x(t) the solution of system (1) with initial value (t0,x(t0)).
In this paper, we assume that some conditions are satisfied so that the equilibrium point of system (1) does exist, see [16, 17]. Assume that x⋆=(x1⋆,x2⋆,…,xn⋆)T is an equilibrium point of system (1). Impulsive operator is viewed as perturbation of the point x⋆ of such system without impulsive effects. We assume that the following impulsive condition holds.
Furthermore, we will assume that the response function fi satisfies the following condition.
(H1) fi is globally Lipschizian with Lipschitz constant li>0, that is, |fi(s1)-fi(s2)|≤li|s1-s2|, for all s1,s2∈ℝ, i∈Λ.
Note that x⋆ is an equilibrium point of system (1), one can derive from system (1) that the transformation zi=xi-xi⋆, i∈Λ transforms such system into the following system:
(2)z˙i(t)=-ai(t)zi(t)+∑j=1nbij(t)Fj(zj(t)),t≥t0,t≠tk,zi(tk)=(1+σik)zi(tk-),i∈Λ,k∈ℤ+,
where Fj(zj(t))=fj(xj⋆+zj(t))-fj(xj⋆), and from the condition (H1), it holds that ∥Fj(zj(t))∥≤lj∥zj(t)∥, j∈Λ.
Furthermore, let yi(t)=zi(t)eα(t-t0), i∈Λ, then system (2) becomes as follows:
(3)y˙i(t)=(α-ai(t))yi(t)+eα(t-t0)∑j=1nbij(t)Fj(yj(t)e-α(t-t0)),t≥t0,t≠tk,yi(tk)=(1+σik)yi(tk-),i∈Λ,k∈ℤ+.
To prove the stability of x⋆ of system (1), it is equal to prove the stability of the zero solution of system (2), and also equal to the boundedness of system (3).
In the following, the notion AT means the transpose of a square matrix A. We will use the notation A>0 (or A<0, A≥0, A≤0) to denote that the matrix A is a positive definite (negtive definite, positive semidefinite, an negative semidefinite) marix.
Let y(t)=(y1(t),y2(t),…,yn(t))T, A(t)=diag[a1(t),a2(t),…,an(t)], B(t)=(bij(t))n×n, L=diag[l1,l2,…,ln], I=diag[1,1,…,1], Dk=diag[1+σ1k,1+σ2k,…,1+σnk], F(y)=(F1(y1),F2(y2),…,andFn(yn))T then system (3) with initial condition becomes as follows:
(4)y˙(t)=(αI-A(t))y(t)+e-α(t-t0)B(t)F(y(t)e-α(t-t0)),t≥t0,t≠tk,y(tk)=Dky(tk-),i∈Λ,k∈ℤ+.
We introduce a definition as follows.
Definition 1.
Assume x⋆=(x1⋆,x2⋆,…,xn⋆)T∈ℝn is the equilibrium point of system (1), then the equilibrium point x⋆ of system (1) is said to be globally exponential stable with given convergence rate α>0. If for any initial data x(t0)=x0∈ℝ, there exists a constant M⩾1, such that
(5)∥x(t,t0,x0)-x⋆∥⩽∥x0-x⋆∥Me-α(t-t0),t⩾t0.
From the transformation zi=xi-xi⋆, i∈Λ, and z(t0)=z0=x0-x⋆, the globally exponential stability of the equilibrium point x⋆ of system (1) can be transformed into the globally exponential stability of trivial solution of system (2), so (5) can be rewritten as follows:
(6)∥z(t,t0,z0)∥≤∥z(t0)∥Me-α(t-t0),t⩾t0.
Furthermore, form the transformation y(t)=z(t,t0,z0)eα(t-t0), i∈Λ, the globally exponential stability of trivial solution of system (2) can be transformed into the boundedness of the solution of system (4) and it can be rewritten as follows:
(7)∥y(t)∥≤∥y(t0)∥M,t⩾t0.
3. Main ResultsTheorem 2.
Given constant α>0. The equilibrium point of the system (1) is globally exponentially stable with the given convergence rate α, if the conditions (H0) and (H1) are fulfilled; moreover, suppose that
αI-A(t)+B(t)L≤0, for all t>t0,
∏k=1∞maxi∈Λ(1+σik)<∞, and σik≥0, i∈Λ, k∈ℤ+.
Proof.
We only need to prove y(t)=z(t)eα(t-t0) is bounded when t≥t0, where z(t)=z(t,t0,z0) is a solution of (3) through (t0,z0).
Consider the Lyapunov function as follows:
(8)V(t)=y(t)Ty(t)=∑i=1nyi2(t).
Particularly, V(t0)=∑i=1nyi2(t0).
Then from conditions (H0)-(H1) and (i), we get the upper right-hand derivative of V(t) along the solutions of system (3), for t∈[tk,tk+1), k∈ℤ+(9)D+V(t)=2∑i=1nyi(t)y˙i(t)=2∑i=1nzi(t)e2α(t-t0)[z˙i(t)+αzi(t)]=2∑i=1nzi(t)e2α(t-t0)×[-ai(t)zi(t)+∑j=1nbij(t)Fj(zj(t))+αzi(t)]=2∑i=1ne2α(t-t0)×[(α-ai(t))zi2(t)+zi(t)∑j=1nbij(t)Fj(zj(t))]≤2∑i=1ne2α(t-t0)×[(α-ai(t))zi(t)2+zi(t)∑j=1nbij(t)lj∥zj(t)∥]=2∑i=1ne2α(t-t0)zi2(t)(α-ai(t))+2∑i=1ne2α(t-t0)zi(t)∑j=1nbij(t)lj∥zj(t)∥=2y(t)T[αI-A(t)+B(t)L]y(t)≤0,
which implies the functional V(t) is nonincreasing for t∈[tk,tk+1), k∈ℤ+. By condition (ii), it holds that
(10)V(tk)=∑i=1nyi2(tk)=∑i=1nzi2(tk)e2α(tk-t0)=∑i=1ne2α(t-t0)[zi(tk-)+Ji(zi(tk-))]2=∑i=1ne2α(t-t0)(1+σik)2zi2(tk-)≤maxi∈Λ(1+σik)2V(tk-).
For any t∈[t0,t1), since V(t) is nonincreasing, it holds that V(t)≤V(t0); moreover,
(11)V(t1)≤maxi∈Λ(1+σi1)2V(t1-)≤maxi∈Λ(1+σi1)2V(t0).
Similarly, for any t∈[t1,t2), it holds that V(t)≤V(t1)≤maxi∈Λ(1+σi1)2V(t0), and
(12)V(t2)≤maxi∈Λ(1+σi2)2V(t2-)≤maxi∈Λ(1+σi2)2maxi∈Λ(1+σi1)2V(t0).
Thus, it can be deduced that for t∈[tk,tk+1), k∈ℤ+,
(13)V(t)≤∏j=1kmaxi∈Λ(1+σij)2V(t0).
Hence, we obtain that for any t≥t0,
(14)V(t)≤∏j=1∞maxi∈Λ(1+σij)2V(t0),
which implies that
(15)∥y(t)∥≤M∥y(t0)∥,t≥t0,
where M=∏k=1∞maxi∈Λ(1+σik)<∞.
The proof of Theorem 2 is complete.
Remark 3.
Most of the existing results about the exponential stability of impulsive neural networks cannot effectively control the convergence rate. It is interesting to see that Theorem 2 can guarantee the globally exponential stability of impulsive neural networks with the given convergence rate.
Remark 4.
In particular, if A(t)≡A,B(t)≡B in Theorem 2, where A, B are constant matrices, then condition αI-A+BL<0 can be easily checked via Matlab.
Theorem 5.
Given constant α>0. The equilibrium point of the system (1) is globally exponentially stable with the given convergence rate α, if the conditions (H0)-(H1) are fulfilled; moreover, suppose that
there exists a constant λ>0, such that (α+λ)I-A(t)+B(t)L<0, for all t>t0,
τ≜mink∈ℤ+{tk+1-tk}, maxi∈Λ(1+σik)≤Mkeλτ, where 1≤Mk<∞ and ∏k=1∞Mk<∞, i∈Λ, k∈ℤ+.
Proof.
We only need to prove that y(t)=z(t)eα(t-t0) is bounded when t≥t0, where z(t)=z(t,t0,z0) is a solution of (3) through (t0,z0).
Consider the Lyapunov function as follows:
(16)V(t)=y(t)Ty(t)=∑i=1nyi2(t).
In particular, V(t0)=∑i=1nyi2(t0).
Then from conditions (H0)-(H1) and (i), we get the upper right-hand derivative of V(t) along the solutions of system (3), for t∈[tk,tk+1), k∈ℤ+(17)D+V(t)=2∑i=1nyi(t)y˙i(t)≤2y(t)T(αI-A(t)+B(t)L)y(t)<-2λV(t).
Thus,
(18)V(t)<V(tk)e-2λ(t-tk)<V(tk)e-2λτ,t∈[tk,tk+1),k∈ℤ+.
By condition (ii), it holds that
(19)V(tk)=∑i=1nyi2(tk)=∑i=1ne2α(t-t0)(1+σik)2zi2(tk-)≤Mk2e2λτ∑i=1ne2α(t-t0)zi2(tk-)≤Mk2e2λτV(tk-).
For any t∈[t0,t1), it holds that V(t)≤V(t0)e-2λτ, moreover
(20)V(t1)≤M12e2λτV(t1-)≤M12V(t0).
Similarly, for any t∈[t1,t2), it holds that V(t)≤V(t1)e-2λτ≤M12V(t0)e-2λτ, and
(21)V(t2)≤M2e2λτV(t2-)≤M22M12V(t0).
Without loss of generality, when t∈[tk,tk+1), k∈ℤ+, it can be deduced that
(22)V(t)≤∏j=1kMj2e-2λτV(t0).
Hence, we obtain that for any t≥t0,
(23)V(t)≤∏j=1∞Mj2V(t0),
which implies that
(24)∥y(t)∥≤M∥y(t0)∥,t≥t0,
where M=(∏j=1∞Mj)<∞.
The proof of Theorem 5 is complete.
Remark 6.
Although Theorem 5 enhances the restriction on condition (i), the impulsive restriction in (ii) is weaker; that is, σik is not necessary to converge to 0 as k is large enough, provided that the impulsive intervals are not too small.
Theorem 7.
Given constant α>0. The equilibrium point of the system (1) is globally exponentially stable with the given exponential convergence rate α, if the conditions (H0)-(H1) are fulfilled; moreover, suppose that
there exists a constant λ>0, such that (α-λ)I-A(t)+B(t)L<0, for all t>t0,
τ≜maxk∈ℤ+{tk+1-tk} and βkeλτ≤1, where βk=maxi∈Λ(1+σik), -1≤σik≤0, i∈Λ, k∈ℤ+.
Proof.
We only need to prove that y(t)=z(t)eα(t-t0) is bounded when t≥t0, where z(t)=z(t,t0,z0) is a solution of (3) through (t0,z0).
Consider the Lyapunov functional as follows:
(25)V(t)=y(t)Ty(t)=∑i=1nyi2(t).
Particularly, V(t0)=∑i=1nyi2(t0).
Then from conditions (H0)-(H1) and (i), we get the upper right-hand derivative of V(t) along the solutions of system (1), for t∈[tk,tk+1), k∈ℤ+(26)D+V(t)=2∑i=1nyi(t)y˙i(t)=2y(t)T(αI-A(t)+B(t)L)y(t)<2λV(t).
Thus,
(27)V(t)<V(tk)e2λ(t-tk)<V(tk)e2λτ,t∈[tk,tk+1),k∈ℤ+.
By condition (ii), it holds that
(28)V(tk)=∑i=1nyi2(tk)≤maxi∈Λ(1+σik)2V(tk-)=βk2V(tk-),k∈ℤ+.
By simple induction, we can prove that for any t∈[tk,tk+1), k∈ℤ+,
(29)V(t)≤∏j=1kβj2e2(k+1)λτV(t0)≤e2λτV(t0),
which implies that
(30)∥y(t)∥≤M∥y(t0)∥,t≥t0,
where M=eλτ.
The proof of Theorem 7 is complete.
4. Applications
The following illustrative example will demonstrate the effectiveness of our results.
Example 8.
Consider the following impulsive neural networks:
(31)x˙1(t)=-a1(t)x1(t)+b11(t)f1(x1(t))+b12(t)f2(x2(t))+I1(t),x˙2(t)=-a2(t)x2(t)+b21(t)f1(x1(t))+b22(t)f2(x2(t))+I2(t),x1(tk)=(1+σ1k)x1(tk-)-σ1k,k∈ℤ+,x2(tk)=(1+σ2k)x2(tk-)-σ2k,k∈ℤ+,
where f1(u)=f2(u)=(|u+1|-|u-1|)/2.
It is easy to see that fj, j=1,2 satisfying hypothesis (H1) with li=l2=1. We have a1(t)=3, a2(t)=2-(3/2)cos(2t), b11(t)=|sint|/2, b12(t)=1/2, b21(t)=2, b22(t)=-1-(cos(2t)/2), I1=(5/2)-(|sint|/2), I2=1-cos(2t), σ1k=1+(1/5k2)-1, σ2k=1+(1/6k2)-1.
Let α=3. It can be deduced that αI-A(t)+B(t)L≤0 and ∏k=1∞maxi=1,2|1+σik|<∞. Hence, all the conditions of Theorem 2 are satisfied; then the equilibrium point x⋆=(1,1) of the above system (31) is globally exponentially stable with the given convergence rate α=3.
Acknowledgments
This work was jointly supported by the Project of Shandong Province Higher Educational Science and Technology Program (J12LI04), Research Fund for Excellent Young and Middle-aged Scientists of Shandong Province (BS2012DX039), and National Natural Science Foundation of China (11226136, 11171192).
YangZ.XuD.Stability analysis of delay neural networks with impulsive effects200552151752110.1109/TCSII.2005.849032CaoJ.Global stability analysis in delayed cellular neural networks19995955940594410.1103/PhysRevE.59.5940ShenJ.LiuY.LiJ.Asymptotic behavior of solutions of nonlinear neutral differential equations with impulses200733211791892-s2.0-3414713504310.1016/j.jmaa.2006.09.078KoskoB.1992New Delhi, IndiaPrentice HallCaoJ.On stability of delayed cellular neural networks19992615-630330810.1016/S0375-9601(99)00552-6GopalsamyK.Stability of artificial neural networks with impulses200415437838132-s2.0-304275006110.1016/S0096-3003(03)00750-1SamiduraiR.Marshal AnthoniS.BalachandranK.Global exponential stability of neutral-type impulsive neural networks with discrete and distributed delays2010411031122-s2.0-7035056791210.1016/j.nahs.2009.08.004LisenaB.Exponential stability of Hopfield neural networks with impulses2011124192319302-s2.0-7995552379210.1016/j.nonrwa.2010.12.008WuS.LiC.LiaoX.DuanS.Exponential stability of impulsive discrete systems with time delay and applications in stochastic neural networks: a Razumikhin approach201282293610.1016/j.neucom.2011.09.029BermanA.PlemmonsR. J.1979New York, NY, USAAcademic PressMR544666BaĭnovD. D.SimeonovP. S.1989New York, NY, USAHalsted PressMR1010418LakshmikanthamV.BaĭnovD. D.SimeonovP. S.1989SingaporeWorld ScientificMR1082551FuX. L.YanB. Q.LiuY. S.2005Beijing, ChinaScience PressStamovaI. M.2009New York, NY, USAWalter de GruyterLiuX.WangQ.The method of Lyapunov functionals and exponential stability of impulsive systems with time delay2007667146514842-s2.0-3384607907110.1016/j.na.2006.02.004LiX. D.New results on global exponential stabilization of impulsive functional differential equations with infinite delays or finite delays2010115419442012-s2.0-7795551610910.1016/j.nonrwa.2010.05.006LiuX. Z.Stability of impulsive control systems with time delay2004394-551151910.1016/S0895-7177(04)90522-5MR2046538FuX.LiX.Global exponential stability and global attractivity of impulsive Hopfield neural networks with time delays200923111871992-s2.0-6734918258810.1016/j.cam.2009.02.094LiX.FuX.BalasubramaniamP.RakkiyappanR.Existence, uniqueness and stability analysis of recurrent neural networks with time delay in the leakage term under impulsive perturbations2010115409241082-s2.0-7795551706510.1016/j.nonrwa.2010.03.014LiX. D.ChenZ.Stability properties for Hopfield neural networks with delays and impulsive perturbations2009105325332652-s2.0-6404909127410.1016/j.nonrwa.2008.10.028LiX. D.BohnerM.Exponential synchronization of chaotic neural networks with mixed delays and impulsive effects via output coupling with delay feedback2010525-66436532-s2.0-7795417614110.1016/j.mcm.2010.04.011AkçaH.AlassarR.CovachevV.CovachevaZ.Al-ZahraniE.Continuous-time additive Hopfield-type neural networks with impulses200429024364512-s2.0-1074422533110.1016/j.jmaa.2003.10.005HuangZ. T.YangQ. G.LuoX. S.Exponential stability of impulsive neural networks with time-varying delays20083547707802-s2.0-3474884140010.1016/j.chaos.2006.05.089StamovaI. M.IlarionovR.On global exponential stability for impulsive cellular neural networks with time-varying delays20105911350835152-s2.0-7795313745910.1016/j.camwa.2010.03.043