Abstract
A model of shunting inhibitory cellular neural networks with mixed delays is proposed. Applying appropriate differential inequality techniques, several sufficient conditions are derived to ensure the existence and exponential stability of weighted pseudo-anti-periodic solutions for the proposed neural networks. Moreover, numerical examples are provided to show the validity and the advantages of the obtained results
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
During the 1990s, Bouzerdoum and Pinter [1–3] proposed shunting inhibitory cellular neural networks (SICNNs) to describe a new class of biologically inspired cellular neural networks (CNNs) in which shunting inhibition mediates the synaptic interactions among neurons. Therefore, SICNNs have shown great potential as information processing systems [4–11]. Recently, the exponential stability of the anti-periodic solutions can describe the global dynamics of delay systems since the convergence rate can be estimated [12–14], and a lot of research work is focused on this topic of SICNNs with mixed delays [15–20]. In particular, the following dynamical system:
has been used to describe SICNNs with mixed delays involving time-varying delays \(\tau _{ij}(t)\) and unbounded distributed delay kernels \(\sigma _{ij }(u)\), where \(ij\in N=\{11, 12, \ldots , mn\},\) \(C_{ij}\) designates the cell at the (i, j) position of the lattice. The \( \varrho \) neighborhood \(N_\varrho (i, j)\) of \(C_{ij}\) is given as
\(x(t)=\{x_{ij}(t)\}=(x_{11}(t), x_{12}(t), \ldots , x_{mn}(t))^{T}\) corresponds to the state vector, \(d_{ij}(t)\) represents the rate of decay, and F and G are the signal transmission functions. The detailed biological accounts on the coefficients \(C_{ij}^{kl} (t)\) and \(B_{ij}^{kl}(t)\) can be found in [21].
As mentioned by Al-Islam et al. [22], the research of weighted pseudo-anti-periodic differential equations has academic significance in both dynamical theory and its practical application. Moreover, weighted pseudo-periodicity and weighted pseudo-anti-periodicity were first introduced in [22] to generalized the well-known notions of periodicity and that of anti-periodicity, respectively. In addition, in view of the biological mechanism of system (1.1), it is interesting and desirable to construct neural network models which are capable of producing weighted pseudo-anti-periodic solution. Nevertheless, the weighted pseudo-anti-periodic problem for SICNNs with mixed delays has not been adequately studied. For the above reasons, in this paper, we aim to provide a criterion to guarantee that all state vectors of (1.1) converge to a weighted pseudo-anti-periodic solution with a positive exponential convergence rate.
2 Preliminary results
To further our discussion, \({\mathbb{U}}\) designates the set of locally integrable functions (weights) \(\mu :{\mathbb {R}}\rightarrow (0, \ +\infty )\) satisfying
Define the following notations:
and
Furthermore, let \(BC({\mathbb {R}},{\mathbb {R}}^{mn})\) denote the bounded continuous function set, which is a Banach space with the supremum norm \(\Vert f\Vert _{\infty } := \sup \nolimits _{ t\in {\mathbb {R}}} \Vert f (t)\Vert\). Also, denote
and
A function \(W\in {BC({\mathbb {R}},{\mathbb {R}}^{mn})}\) is called weighted pseudo-anti-periodic if it can be expressed as
where \(Q_{1}\in {AP^{T}({\mathbb {R}},{\mathbb {R}}^{mn})}\) is the T-anti-periodic component and \(Q_{2}\in {PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}}^{mn})}\) is the ergodic perturbation. In particular, fixed \(\mu \in {\mathbb {U}}^{+}_{\infty }\), \((PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{mn}),\Vert .\Vert _{\infty })\) become a Banach space and \(AP^{T }({\mathbb {R}},{\mathbb {R}}^{mn})\) is a proper subset of \(PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{mn})\) [22].
We define the following initial condition:
For \(kl,ij \in N,\) it will be supposed that \(\sigma _{ij}\in BC([0, +\infty ), {\mathbb {R}})\) , \(|\sigma _{ij}(s)|e^{\kappa s}\) is integrable on \([0, +\infty )\) for \(\kappa >0\), \(d_{ij}, C_{ij}^{kl}\), \(B_{ij}^{kl}\in C({\mathbb {R}},{\mathbb {R}} ), \ I_{ij}\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}} )\), \(\tau _{kl} \in C ^{1} ({\mathbb {R}}, \ [0, \ +\infty ) )\), and
where \(C_{ij}^{kl, h} , B_{ij}^{kl, h} \in {BC({\mathbb {R}},{\mathbb {R}} )}, C_{ij}^{kl,\varphi }, B_{ij}^{kl,\varphi }\in {PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}} )}\) satisfy
and
For \(ij \in N\), the following assumptions will be adopted:
\((S_0)\) there exist \(\tilde{d}_{ij} \in BC({\mathbb {R}}, (0, \ +\infty ) )\) and \(K_{ij}>0\) such that
\((S_1)\) F and G are global Lipschitz with Lipschitz constants \(L ^{F}\) and \(L^{G}\), and
\((S_{2})\) \(\mu \in {\mathbb {U}}^{+}_{\infty }\), and \({\mathbb {F}}(\alpha )=\sup \nolimits _{x\in \mathbb { R}}\frac{\mu (x+\alpha )}{\mu (x)}\) is bounded on arbitrary closed subinterval of \([0, \ +\infty ).\) \((S_3)\) there exist positive constants \(\gamma _{ij}\) and \(\delta\) such that
where \(I=\max \nolimits _{ij\in N} \left\{ K_{ij}\frac{I^{+}_{ij}}{\tilde{d}_{ij} ^{-}}\right\}\), and
Lemma 2.1
Assume that \(f\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}})\), \(\rho \in C^{1} ({\mathbb {R}}, \ {\mathbb {R}} )\) is \(T-\) periodic, \(\rho (s)\ge 0\) and \(\rho '(s)<1\), \(\forall s\in {\mathbb {R}}\). Then, \(f(s-\rho (s))\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}).\)
Proof
Let
Clearly, \(h(t-\rho (t))\in AP^{T }({\mathbb {R}},{\mathbb {R}}).\) In view of \((S_{2})\), we get
Letting \(\beta =\sup \nolimits _{t\in {\mathbb {R}}}\frac{1}{1- \rho '(t)} \times \sup \nolimits _{\alpha \in [\rho ^{-}, \ \rho ^{+}]}{\mathbb {F}}(\alpha )\) and \(s=t-\rho (t)\) give us
which, together with the fact that
implies that
This proves Lemma 2.1. \(\square\)
Lemma 2.2
If \(\varphi \in PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}} ) ,\) then, \(\int _{0}^{+\infty }|\sigma _{ij}(s)||\varphi (t-s)| \hbox {d}s\in PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}} ).\)
Proof
Obviously, one can obtain
Let \(M^{\varphi }=\sup \nolimits _{\theta \in {\mathbb {R}}}| \varphi (\theta )|\) and \(\sigma _{ij}^{*}=\int _{0}^{+\infty }|\sigma _{ij}(s)| \hbox {d}s\), we get
For any sequence \(\{r_{n}\}_{n=1}^{+\infty }\) satisfying
we denote
Then,
According to the Lebesgue dominated convergence theorem, we have
which entails that
Thus, \(\int _{0}^{+\infty }|\sigma _{ij}(s)|\varphi (t-s) \hbox {d}s\in PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}} ).\) This completes the proof. \(\square\)
Lemma 2.3
Let \(x_{ij}\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}})\) for all \(ij \in N\). Then,
and
Proof
From Lemma 2.1, we get
Furthermore, let
and
Then, for all \(t\in {\mathbb {R}}\), we obtain
Clearly, (2.4) gives us
and
Now, we choose constants \(\alpha _{ij}^{kl }, \ \beta _{ij}^{kl}\) and \(\eta _{ij}^{kl}\) such that
Consequently,
and
This, together with (2.6), leads to
Next, for \(ij, kl\in N\), we get
It follows from (2.5) and Lemma 2.2 that
and
Hence,
and
which, together with (2.7) and (2.8), imply that
This proves Lemma 2.3.
Lemma 2.4
Define a nonlinear operator Q by setting
Then, Q maps \(PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{mn})\) into itself.
Proof
With the application of the verification in Lemma 2.1 of [21], we can easily obtain that \(Q\varphi \in BC({\mathbb {R}}, {\mathbb {R}}^{mn})\) . From Lemma 2.3, we obtain that there are \(H_{ij} \in {AP^{T}({\mathbb {R}},{\mathbb {R}} )}\) and \(\Phi _{ij} \in {PAP_{0}^{\mu }({\mathbb {R}},{\mathbb {R}} )}\) such that
Arguing as in the verification of Lemma 2.2, one can show
Then,
and the fact that
imply that
and Q maps \(PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{mn})\) into itself. This ends the proof.
3 Main results
Theorem 3.1
Let \((S_{0})\), \((S_{1})\), \((S_{2})\) and \((S_{3})\) hold. Then, system (1.1) has exactly one weighted pseudo-anti-periodic solution \(x^{*}(t)\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{mn})\) , and all state vectors of (1.1) and (2.1) converge to \(x^{*}(t)\) with a positive exponential convergence rate \(\lambda \in \left(0, \ \min \left\{\kappa , \min \limits _{ij\in N}\tilde{d}_{ij}^{-}\right\}\right)\).
Proof
Set
and
Then,
and
Consequently, \((S_{3})\) entails
which implies that Q is a mapping from \(\Gamma\) to \(\Gamma\).
Furthermore, according to \((S_{0})\), \((S_{1})\), \((S_{2})\) and \((S_{3}),\) one can easily to see that
which, together with the fact that \(0<\max \nolimits _{ ij \in N}\{ ( 1-\frac{\gamma _{ij}}{2 \tilde{d}_{ij} ^{+}})\}<1,\) entails that the mapping Q: \(\Gamma \rightarrow \Gamma\) is a contraction mapping, and there exists a unique fixed point
which is a weighted pseudo-anti-periodic solution of (1.1).
Finally, with a similar proof in Theorem 3.2 of [21], one can pick constants \(\lambda \in (0, \ \min \{\kappa , \min \nolimits _{ij\in N}\tilde{d}_{ij}^{-}\})\) and \(M>\sum \nolimits _{ij=11}^{mn}K_{ij} +1\) such that
and
which proves Theorem 3.1.
4 Numerical simulations
Consider SICNNs (1.1) with the following parameters:
Obviously, one can choose
and
such that SICNNs (1.1) with (4.1) satisfy all the hypothesis mentioned in Section 2. Based on Theorem 3.1, we can conclude that system (1.1) has exactly one weighted pseudo-anti-periodic solution \(x^{*}(t)\in PAP^{T,\mu }({\mathbb {R}},{\mathbb {R}}^{4})\), and all state vectors of system (1.1) converge exponentially to \(x^{*}(t)\) as \(t\rightarrow +\infty\). Here, the exponential convergence rate \(\lambda \approx 0.02\). The time–response curve is given in Fig. 1, and there are three different groups initial values with \((1.1,-3.1,4.1,-3.1),(-3.2,4.1,-1.1,2.1),(3.1,-4.1,1.1,-2).\)
Remark 4.1
In the real world, there is little purely periodic phenomenon, and this motivates us to study the pseudo-almost-periodic and weighted pseudo-almost-periodic situations. In this work, we show that for the same assumptions in [21] plus other assumptions that we add to realize our demonstrations, allows us to show the dynamic characteristics of (1.1) in a weighted pseudo-anti-periodic set broader than the anti-periodic set in [21]. Since weighted pseudo-anti-periodic SICNNs with mixed delays has not been touched in [7–11, 21], our results improve and extend the corresponding ones in the above references.
5 Conclusions
In this manuscript, we have investigated shunting inhibitory cellular neural networks with mixed delays. With the aid of the contraction mapping fixed point theorem, differential inequality theory and the Lyapunov functional method, some sufficient criterion for the existence and global exponential stability of weighted pseudo-anti-periodic solutions of the system is established. In order to demonstrate the feasibility of the theoretical results, a numerical example is given. The established results were compared with those of recent methods existing in the literature. We expect to extend this work to other neural networks models with mixed delays. We will also study more types of weighted pseudo-almost-periodic solution problems on delayed neural networks models.
References
Bouzerdoum A, Pinter RB (1993) Shunting inhibitory cellular neural networks: derivation and stability analysis. IEEE Trans Circuits Syst 1 Fundam Theory Appl 40:215–221
Bouzerdoum A, Pinter RB (1991) Analysis and analog implementation of directionally sensitive shunting inhibitory cellular neural networks. Vis Inf Process Neurons Chips SPIE–1473:29–38
Bouzerdoum A, Pinter RB (1992) Nonlinear lateral inhibition applied to motion detection in the fly visual system. In: Pinter RB, Nabet B (eds) Nonlinear vision. CRC Press, Boca Raton, pp 423–450
Ou C (2009) Almost periodic solutions for shunting inhibitory cellular neural networks. Nonlinear Anal Real World Appl 10:2652–2658
Chen Z (2013) A shunting inhibitory cellular neural network with leakage delays and continuously distributed delays of neutral type. Neural Comput Appl 23:2429–2434
Wang W, Liu B (2014) Global exponential stability of pseudo almost periodic solutions for SICNNs with time-varying leakage delays. Abst Appl Anal 967328:1–17
Liu X (2015) Exponential convergence of SICNNs with delays and oscillating coefficients in leakage terms. Neurocomput 168:500–504
Zhao C, Wang Z (2015) Exponential convergence of a SICNN with leakage delays and continuously distributed delays of neutral type. Neural Process Lett 41:239–247
Jiang A (2015) Exponential convergence for shunting inhibitory cellular neural networks with oscillating coefficients in leakage terms. Neurocomput 165:159–162
Zhang A (2016) Pseudo almost periodic solutions for SICNNs with oscillating leakage coefficients and complex deviating arguments. Neural Process Lett. doi:10.1007/s11063-016-9518-x
Liu B (2016) Global exponential convergence of non-autonomous SICNNs with multi-proportional delays. Neural Comput Appl. doi:10.1007/s00521-015-2165-8
Liu B (2009) An anti-periodic LaSalle oscillation theorem for a class of functional differential equations. J Comput Appl Math 223:1081–1086
Li Y, Huang L (2009) Anti-periodic solutions for a class of Li\(\acute{e}\)nard-type systems with continuously distributed delays. Nonlinear Anal Real World Appl 10(4):2127–2132
Fan Q, Wang W, Yi X (2009) Anti-periodic solutions for a class of nonlinear nth-order differential equations with delays. J Comput Appl Math 230(2):762–769
Ou C (2008) Anti-periodic solution for high-order Hopfield neural networks. Comput Math Appl 56:1838–1844
Zhao C, Fan Q, Wang W (2010) Anti-periodic solutions for shunting inhibitory cellular neural networks with time-varying coefficients. Neural Process Lett 31:259–267
Gong S (2009) Anti-periodic solutions for a class of Cohen–Grossberg neural networks. Comput Math Appl 58:341–347
Wang W (2013) Anti-periodic solution for impulsive high-order Hopfield neural networks with time-varying delays in the leakage terms. Adv Differ Equ 2013(73):1–15
Peng L, Wang W (2013) Anti-periodic solutions for shunting inhibitory cellular neural networks with time-varying delays in leakage terms. Neurocomput 111(2):27–33
Zhou Q (2016) Anti-periodic solutions for cellular neural networks with oscillating coefficients in leakage terms. Int J Mach Learn Cyber. doi:10.1007/s13042-016-0531-1
Long Z (2016) New results on anti-periodic solutions for SICNNs with oscillating coefficients in leakage terms. Neurocomput 171(1):503–509
Al-Islam NS, Alsulami SM, Diagana T (2012) Existence of weighted pseudo anti-periodic solutions to some non-autonomous differential equations. Appl Math Comput 218:6536–6548
Acknowledgments
The authors would like to express their sincere appreciation to the editor and reviewers for their helpful comments in improving the presentation and quality of the paper. In particular, the authors express the sincere gratitude to Prof. Bingwen Liu (Jiaxing University) for the helpful discussion when this revision work was being carried out. This work was supported by the Natural Scientific Research Fund of Hunan Province of China (Grant Nos. 2016JJ6103, 2016JJ6104).
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Zhou, Q., Shao, J. Weighted pseudo-anti-periodic SICNNs with mixed delays. Neural Comput & Applic 29, 865–872 (2018). https://doi.org/10.1007/s00521-016-2582-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-016-2582-3
Keywords
- Weighted pseudo-anti-periodic solution
- Shunting inhibitory cellular neural network
- Exponential stability
- Mixed delay