Effects of Time-Varying Impulses on the Synchronization of Delayed Dynamical Networks

Abstract

The effects of time-varying impulses on the synchronization of a class of general complex delayed dynamical networks are investigated. Different from the existing works, the impulses discussed here are time-varying, and both synchronizing and desynchronizing impulses are considered in the network model simultaneously. Moreover, the network topology is assumed to be directed and weakly connected with a spanning tree. By using the comparison principle, some simple yet generic globally exponential synchronization criteria are derived. It is shown that besides impulse strengths and impulsive interval, the obtained criteria are also closely related with topology structure of the network. Finally, numerical examples are given to demonstrate the effectiveness of the theoretical results.

1. Introduction

Recently, much efforts have been devoted to complex dynamical networks due to the wide and potential applications in many areas [1–3]. In general, a complex network is a large set of interconnected nodes, in which each node is a fundamental unit with specific dynamics and each edge represents the interactions between them. As a matter of fact, the structure of many real systems in nature including the Internet, food webs, biomolecular networks, and social networks can be described by complex networks [1, 2]. One significant and interesting phenomenon in a complex dynamical network is the synchronization of all its dynamical nodes. In the past few decades, synchronization of complex dynamical networks has been extensively studied from various fields of science and engineering, and a wide variety of synchronization criteria have been presented for various dynamical networks; see [2–22] and the references therein.

From the literature, there exist two common phenomena in many dynamical networks: delay effects and impulsive effects [4, 5, 23–29]. Due to the finite switching speed of amplifiers and finite signal propagation time, time delay is inevitably encountered in many dynamical networks and may cause undesirable dynamical behaviors such as oscillation and instability [4, 5, 23, 24]. On the other hand, the states of many complex systems and realistic networks are often subject to instantaneous perturbations and experience abrupt changes at certain instants, which may be caused by switching phenomenon, frequency change, or other sudden noise; that is, they exhibit impulsive effects [25–29]. Since time delays and impulses can heavily affect the dynamical behaviors of the networks, it is imperative to investigate both effects of time delays and impulses on the synchronization of dynamical networks.

In general, there are two types of impulses in terms of synchronization in dynamical networks. An impulsive sequence is said to be synchronizing if it can enhance the synchronization of dynamical networks. Conversely, an impulsive sequence is said to be desynchronizing if the impulsive effects can suppress the synchronization of dynamical networks. In recent years, many interesting results on the synchronization of complex dynamical networks with impulsive effects have been reported in the literature [6–16]. For instance, Lu et al. [11] established a unified synchronization criterion for impulsive dynamical networks subject to synchronizing impulses or desynchronizing impulses. In addition, Lu et al. [12] also investigated the exponential synchronization of coupled neural networks with impulsive disturbances. Zhang et al. [16] discussed the synchronization problem of coupled switched neural networks with mode-dependent impulsive effects. Unfortunately, in the most existing literature, it is implicitly assumed that the synchronizing and desynchronizing impulses occur separately. In practice, however, many electronic or biological networks are often subject to instantaneous disturbance and then exposed to time-varying impulsive strengths, and both synchronizing and desynchronizing impulses might exist in realistic networks simultaneously, which was widely overlooked in most of the existing results [30]. To the best of our knowledge, the effects of time-varying impulses on the synchronization of delayed dynamical networks are rarely addressed. Moreover, in much of the literature, time delays in the couplings are considered; however, time delays in the dynamical nodes [5, 9, 17, 18], which are more complex, are still relatively unexplored. In fact, numerous examples can be found in the real world which are characterized by delayed differential equations having time delays in the dynamical nodes [5, 9, 17, 18].

This paper aims at exploring the effects of time-varying impulses on the synchronization of general complex networks with time-varying delays dynamical nodes. Both synchronizing and desynchronizing impulses are considered in the network model simultaneously. The directed and weakly connected topology of networks is focused on. Based on the comparison principle, some simple yet generic globally exponential synchronization criteria are derived. Numerical examples are also provided to illustrate the effectiveness of the theoretical analysis.

2. Model and Preliminaries

Consider a directed complex network consisting of 𝑁 identical time-varying delays dynamical nodes: 𝑥̇_𝑖(𝑡)=𝑓𝑡,𝑥_𝑖(𝑡),𝑥_𝑖(𝑡−𝜏(𝑡))+𝑐^𝑁_𝑗=1𝑏_𝑖𝑗Γ𝑥_𝑗(𝑡),𝑖=1,…,𝑁, (1) where 𝑥_𝑖(𝑡)=(𝑥_𝑖1(𝑡),𝑥_𝑖2(𝑡),…,𝑥_𝑖𝑛(𝑡))^⊤∈ℝ^𝑛 is the state vector of node 𝑖; 𝑓∶ℝ×ℝ^𝑛×ℝ^𝑛→ℝ^𝑛 is a continuously vector-valued function governing the dynamics of isolated nodes; 𝜏(𝑡) is a time-varying delay satisfying 0≤𝜏(𝑡)≤𝜏; 𝑐>0 is the coupling strength; Γ>0 is the inner connecting matrix of nodes; 𝐵=(𝑏_𝑖𝑗)_𝑁×𝑁 is the coupling matrix representing the topological structure of the network. Without loss of generality, we assume that the matrix 𝐵 possesses the following properties: ∑^𝑁_𝑗=1𝑏_𝑖𝑗=0, 𝑏_𝑖𝑗≥0, 𝑖≠𝑗, and rank(𝐵)=𝑁−1 [9, 18]. It is worth mentioning that the coupling matrix 𝐵 can be regarded as the Laplacian matrix of a weighted graph with a spanning tree, and 𝐵 has an eigenvalue 0 with multiplicity 1 [9, 19].

Remark 1. The coupling matrix 𝐵 represents the topological structure of network (1). In this paper, the matrix 𝐵 is not restricted to be symmetric or irreducible. A general structure of the network is discussed; that is, the corresponding graph generated by the matrix 𝐵 can be directed and weakly connected with a spanning tree. Obviously, the network model (1) is a generalization of that discussed in [11, 12].
Due to switching phenomenon, frequency change, or other sudden noise, the states of nodes in many realistic networks are often subject to instantaneous perturbations and experience abrupt changes at certain instants [25–29]. Suppose at time instants 𝑡_𝑘, there are “sudden changes” (or “jumps”) in the state variable such that Δ𝑥_{𝑖𝑡=𝑡𝑘}≜𝑥_𝑖𝑡⁺_𝑘−𝑥_𝑖𝑡⁻_𝑘=𝐷_𝑘𝑥_𝑖𝑡⁻_𝑘,𝑖=1,2,…,𝑁, (2) where {𝑡₁,𝑡₂,𝑡₃,…} is an impulsive sequence satisfying 𝑡_𝑘−1<𝑡_𝑘 and lim_𝑘→∞𝑡_𝑘=+∞, 𝑥_𝑖(𝑡⁺_𝑘)=lim_𝑡→𝑡⁺_𝑘𝑥_𝑖(𝑡), 𝑥_𝑖(𝑡⁻_𝑘)=lim_𝑡→𝑡⁻_𝑘𝑥_𝑖(𝑡), and 𝐷_𝑘∈ℝ^𝑛×𝑛 is an impulsive gain matrix. For the sake of analytical simplification, we will choose 𝐷_𝑘=𝑑_𝑘𝐼_𝑛, where 𝑑_𝑘∈ℝ represents the strength of impulses and 𝐼_𝑛 is an 𝑛×𝑛 identity matrix. This simplification does not cause any loss of generality in the sense of synchronization analysis. In fact, when the constant matrix 𝐷_𝑘 is used to describe the impulsive gain, the matrix product can be used to describe synchronization criteria. Therefore, we can obtain the following impulsive delayed dynamical network: 𝑥̇_𝑖(𝑡)=𝑓𝑡,𝑥_𝑖(𝑡),𝑥_𝑖(𝑡−𝜏(𝑡))+𝑐^𝑁_𝑗=1𝑏_𝑖𝑗Γ𝑥_𝑗(𝑡),𝑡≠𝑡_𝑘,𝑡≥𝑡₀,Δ𝑥_𝑖=𝑥_𝑖𝑡⁺_𝑘−𝑥_𝑖𝑡⁻_𝑘=𝑑_𝑘𝑥_𝑖𝑡⁻_𝑘,𝑡=𝑡_𝑘,𝑘∈𝑍⁺,𝑖=1,2,…,𝑁, (3) where 𝑍⁺={1,2,…} denotes the set of positive integers. Without loss of generality, we assume that 𝑥_𝑖(𝑡) is right continuous at 𝑡=𝑡_𝑘; that is, 𝑥_𝑖(𝑡_𝑘)=𝑥_𝑖(𝑡⁺_𝑘). The initial conditions of (3) are given by 𝑥_𝑖(𝑡₀+𝑠)=𝜑_𝑖(𝑠)∈𝑃𝐶([−𝜏,0],ℝ^𝑛), where 𝑃𝐶([−𝜏,0],ℝ^𝑛) denotes the set of all functions of bounded variation and right-continuous on any compact subinterval of [−𝜏,0]. We always assume that (3) has an unique solution with respect to initial conditions.

Assumption 2 (see [18]). For the vector-valued function 𝑓(𝑡,𝑥(𝑡),𝑥(𝑡−𝜏(𝑡))), suppose that the uniform semi-Lipschitz condition with respect to the time 𝑡 holds; that is, for any 𝑥(𝑡), 𝑦(𝑡)∈ℝ^𝑛, there exist positive constants 𝐿₁>0 and 𝐿₂>0 such that 𝑥(𝑡)−𝑦(𝑡)^⊤×𝑓(𝑡,𝑥(𝑡),𝑥(𝑡−𝜏(𝑡)))−𝑓𝑡,𝑦(𝑡),𝑦(𝑡−𝜏(𝑡))≤𝐿₁𝑥(𝑡)−𝑦(𝑡)^⊤𝑥(𝑡)−𝑦(𝑡)+𝐿₂𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡))^⊤×𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡)). (4)

Remark 3. Assumption 2 gives some requirements for the dynamics of isolated node in network (1). If the function 𝑓(𝑡,𝑥(𝑡),𝑥(𝑡−𝜏(𝑡))) satisfies the uniform Lipschitz condition [17], that is, ‖𝑓(𝑡,𝑥(𝑡),𝑥(𝑡−𝜏(𝑡)))−𝑓(𝑡,𝑦(𝑡),𝑦(𝑡−𝜏(𝑡)))‖≤𝐾₁‖𝑥(𝑡)−𝑦(𝑡)‖+𝐾₂‖𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡))‖, one can choose 𝐿₁=𝐾₁+𝜔𝐾₂/2 and 𝐿₂=𝐾₂/(2𝜔) to satisfy Assumption 2, where 𝜔 is a positive constant. Moreover, it is easy to check that almost all the well-known chaotic systems with delays or without delays, such as the Lorenz system, Rössler system, Chen system, and Chua’s circuit as well as delayed Hopfield neural networks and delayed cellular neural networks (see [9, 18], and the references therein) also satisfy Assumption 2.

Definition 4. The impulsive delayed dynamical network (3) is said to be globally exponentially synchronized if there exist constants 𝜆₀>0 and 𝑀₀>0 such that for any initial conditions 𝜑_𝑖(𝑠)∈𝑃𝐶([−𝜏,0],ℝ^𝑛) (𝑖=1,2,…,𝑁) 𝑥_𝑖(𝑡)−𝑥_𝑗(𝑡)≤𝑀₀exp^{−𝜆0(𝑡−𝑡0)},∀𝑡≥𝑡₀. (5)
Define 𝑠(𝑡)=(1/𝑁)∑^𝑁_𝑙=1𝑥_𝑙(𝑡) and error vectors as 𝑒_𝑖(𝑡)=𝑥_𝑖(𝑡)−𝑠(𝑡), 𝑖=1,2,…,𝑁, then we have 𝑒_𝑖̇(𝑡)=𝑥̇_𝑖(𝑡)−𝑠(̇𝑡)=𝑓𝑡,𝑒_𝑖(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡))+𝑐^𝑁_𝑗=1𝑏_𝑖𝑗Γ𝑒_𝑗(𝑡)+𝐽,𝑡≠𝑡_𝑘,Δ𝑒_𝑖𝑡_𝑘=𝑒_𝑖𝑡⁺_𝑘−𝑒_𝑖𝑡⁻_𝑘=𝑥_𝑖𝑡⁺_𝑘−𝑥_𝑖𝑡⁻_𝑘−𝑠𝑡⁺_𝑘−𝑠𝑡⁻_𝑘=𝑥_𝑖𝑡⁺_𝑘−𝑥_𝑖𝑡⁻_𝑘−1𝑁^𝑁_𝑙=1𝑥_𝑙𝑡⁺_𝑘−𝑥_𝑙𝑡⁻_𝑘=𝑑_𝑘𝑥_𝑖𝑡⁻_𝑘−𝑑_𝑘𝑁^𝑁_𝑙=1𝑥_𝑙𝑡⁻_𝑘=𝑑_𝑘𝑒_𝑖𝑡⁻_𝑘,𝑡=𝑡_𝑘, (6) where 𝑓(𝑡,𝑒_𝑖(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡)))=𝑓(𝑡,𝑒_𝑖(𝑡)+𝑠(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡))+𝑠(𝑡−𝜏(𝑡)))−𝑓(𝑡,𝑠(𝑡),𝑠(𝑡−𝜏(𝑡))), and 𝐽=𝑓(𝑡,𝑠(𝑡),𝑠(𝑡−𝜏(𝑡)))−(1/𝑁)∑^𝑁_𝑙=1𝑓(𝑡,𝑥_𝑙(𝑡),𝑥_𝑙(𝑡−𝜏(𝑡)))−(𝑐/𝑁)∑^𝑁_𝑙=1∑^𝑁_𝑗=1𝑏_𝑙𝑗Γ𝑥_𝑗(𝑡).
Thus, the error dynamical system can be written as follows: 𝑒_𝑖̇(𝑡)=𝑓𝑡,𝑒_𝑖(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡))+𝑐^𝑁_𝑗=1𝑏_𝑖𝑗Γ𝑒_𝑗(𝑡)+𝐽,𝑡≠𝑡_𝑘,𝑡≥𝑡₀,𝑒_𝑖𝑡_𝑘=1+𝑑_𝑘𝑒_𝑖𝑡⁻_𝑘,𝑡=𝑡_𝑘,𝑘∈𝑍⁺,𝑖=1,2,…,𝑁. (7) Clearly, the globally exponential synchronization of the impulsive delayed dynamical network (3) is achieved if the zero solution of the error dynamical system (7) is globally exponentially stable.

Remark 5. When |(1+𝑑_𝑘)|<1, the impulses are beneficial for the synchronization of the impulsive delayed dynamical network (3) since the absolute values of the synchronization errors are reduced. Thus, the impulses are synchronizing impulses if −2<𝑑_𝑘<0. Conversely, when |(1+𝑑_𝑘)|>1, that is, the impulsive strengths 𝑑_𝑘>0 or 𝑑_𝑘<−2, the impulses are desynchronizing impulses since the absolute values of the synchronization errors are enlarged. In addition, when |(1+𝑑_𝑘)|=1, the impulses are neither beneficial nor harmful for the synchronization since the absolute values of the synchronization errors are unchanged. This type of impulses are called inactive impulses [11]. We will not consider inactive impulses here because they have no effect on the synchronization dynamics. In this paper, we focus on the synchronization problem of the dynamical network (3) with both synchronizing and desynchronizing impulses, which is different from most existing literature where the synchronizing and desynchronizing impulses are assumed to occur separately.

Lemma 6 (see [29]). Let 0≤𝜏(𝑡)≤𝜏 and 𝐹(𝑡,𝑥,𝑦)∶[𝑡₀,∞)×ℝ×ℝ→ℝ be nondecreasing in 𝑦 for fixed (𝑡,𝑥), and let 𝐼_𝑘(𝑥)∶ℝ→ℝ be nondecreasing in 𝑥. Suppose that 𝐷⁺𝑢(𝑡)≤𝐹(𝑡,𝑢(𝑡),𝑢(𝑡−𝜏(𝑡))),𝑡≥𝑡₀,𝑢𝑡_𝑘≤𝐼_𝑘𝑢𝑡⁻_𝑘,𝑘∈𝑍⁺,𝐷⁺𝑣(𝑡)>𝐹(𝑡,𝑣(𝑡),𝑣(𝑡−𝜏(𝑡))),𝑡≥𝑡₀,𝑣𝑡_𝑘≥𝐼_𝑘𝑣𝑡⁻_𝑘,𝑘∈𝑍⁺. (8) If 𝑢(𝑡)≤𝑣(𝑡), for 𝑡₀−𝜏≤𝑡≤𝑡₀, then 𝑢(𝑡)≤𝑣(𝑡),𝑡≥𝑡₀.

3. Main Results

In this section, globally exponential synchronization of delayed dynamical networks with time-varying impulses including both synchronizing and desynchronizing impulses simultaneously will be studied. The relationship between the strengths of synchronizing and desynchronizing impulses and the frequency of their occurrence will be established.

Let the matrix 𝐵 be defined as 𝐵≜(𝐵+𝐵^⊤)−Δ, where Δ=diag(𝛿₁,𝛿₂,…,𝛿_𝑁) with 𝛿_𝑗=∑^𝑁_𝑘=1𝑏_𝑘𝑗. Then the matrix 𝐵 is a symmetrical irreducible matrix with zero-sum and nonnegative off-diagonal elements [9, 20]. This implies that zero is an eigenvalue of 𝐵 with multiplicity 1, and all the other eigenvalues of 𝐵 are strictly negative [9, 19]. The eigenvalues of 𝐵 can be ordered as 0=𝜆₁>𝜆₂≥⋯≥𝜆_𝑁. Since both synchronizing and desynchronizing impulses are taken into the network model simultaneously, without loss of generality, we assume that the strengths of synchronizing and desynchronizing impulses take values from finite sets {𝑑^sy₁,𝑑^sy₂,…,𝑑^sy_𝑁1} and {𝑑^desy₁,𝑑^desy₂,…,𝑑^desy_𝑁2}, respectively, where −2<𝑑^sy_𝑖<0, 𝑑^desy_𝑗>0 or 𝑑^desy_𝑗<−2 for 𝑖=1,2,…,𝑁₁, 𝑗=1,2,…,𝑁₂. In addition, we assume that 𝑡^↓_𝑖𝑘 and 𝑡^↑_𝑗𝑘 represent the activation time of the synchronizing impulses and that of the desynchronizing impulses, respectively. Let 𝑑^sy_∗=max_{1≤𝑖≤𝑁1}𝑑^sy_𝑖, 𝑑^desy_∗=max_{1≤𝑗≤𝑁2}𝑑^desy_𝑗, 𝑇^↓_sup=sup{𝑡^↓_𝑖𝑘−𝑡^↓_𝑖𝑘−1}<∞, and 𝑇^↑_inf=inf{𝑡^↑_𝑗𝑘−𝑡^↑_𝑗𝑘−1}>0, where 𝑡^↓_𝑖𝑘,𝑡^↑_𝑗𝑘∈{𝑡₁,𝑡₂,𝑡₃,…}; then we have the following result.

Theorem 7. Consider the impulsive delayed dynamical network (3) with both synchronizing and desynchronizing impulses simultaneously. Under Assumption 2, the impulsive delayed dynamical network (3) is globally exponentially synchronized if the following condition holds: 2ln1+𝑑_∗^𝑠𝑦𝑇^↓_sup+2ln1+𝑑_∗^{𝑑𝑒𝑠𝑦}𝑇^↑_inf+2𝐿₁+𝑟𝜆(𝑟)+2𝑑^∗𝐿₂<0, (9) where 𝑑^∗=1+𝑑_∗^{𝑑𝑒𝑠𝑦}1+𝑑_∗^𝑠𝑦2,𝑟=𝑐𝜆₂+max_{1≤𝑘≤𝑁}𝛿_𝑘(𝑟)=𝜆𝑤𝑖𝑡ℎ𝜆^max(Γ)𝜆,𝑖𝑓𝑟>0,0,𝑖𝑓𝑟=0,_min(Γ),𝑖𝑓𝑟<0. (10)

Proof. Let 𝑒(𝑡)=(𝑒^⊤₁(𝑡),𝑒^⊤₂(𝑡),…,𝑒^⊤_𝑁(𝑡))^⊤, construct a Lyapunov function 𝑉(𝑡)=12𝑒^⊤(𝑡)𝐼_𝑁⊗𝐼_𝑛𝑒(𝑡)=12^𝑁_𝑖=1𝑒^⊤_𝑖(𝑡)𝑒_𝑖(𝑡). (11)
Calculating the upper Dini derivative of 𝑉(𝑡) along the solution of (3), by using Assumption 2 and note that ∑^𝑁_𝑖=1𝑒_𝑖(𝑡)=0, we get 𝐷⁺𝑉(𝑡)=^𝑁_𝑖=1𝑒^⊤_𝑖(𝑡)×𝑓𝑡,𝑒_𝑖(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡))+𝑐^𝑁_𝑗=1𝑏_𝑖𝑗Γ𝑒_𝑗(𝑡)=+𝐽_𝑁𝑖=1𝑒^⊤_𝑖(𝑡)𝑓𝑡,𝑒_𝑖(𝑡),𝑒_𝑖(𝑡−𝜏(𝑡))+𝑐𝑒^⊤(𝑡)𝑒(𝑡)≤)(𝐵⊗Γ_𝑁𝑖=1𝐿₁𝑒^⊤_𝑖(𝑡)𝑒_𝑖(𝑡)+^𝑁_𝑖=1𝐿₂𝑒^⊤_𝑖(𝑡−𝜏(𝑡))𝑒_𝑖(𝑡−𝜏(𝑡))+𝑐𝑒^⊤(𝑡)(𝐵⊗Γ)𝑒(𝑡)≤2𝐿₁𝑉(𝑡)+2𝐿₂𝑉(𝑡−𝜏(𝑡))+𝑐𝑒^⊤(𝑡)(𝐵⊗Γ)𝑒(𝑡)=2𝐿₁𝑉(𝑡)+2𝐿₂𝑉((𝑡+𝑐𝑡−𝜏))2𝑒^⊤(𝑡)𝐵+𝐵^⊤⊗Γ𝑒(𝑡)=2𝐿₁𝑉(𝑡)+2𝐿₂𝑉((𝑡+𝑐𝑡−𝜏))2𝑒^⊤(𝑡)𝐵+Δ⊗Γ𝑒(𝑡),𝑡≠𝑡_𝑘. (12)
Since 𝐵 is a symmetrical matrix, there exists a unitary matrix 𝑈=(𝑢₁,𝑢₂,…,𝑢_𝑁) with 𝑈𝑈^⊤=𝐼_𝑁 such that 𝐵=𝑈diag(𝜆₁,𝜆₂,…,𝜆_𝑁)𝑈^⊤. Introduce a transformation 𝑍(𝑡)=(𝑈^⊤⊗𝐼_𝑛)𝑒(𝑡), where 𝑍(𝑡)=(𝑧^⊤₁(𝑡),𝑧^⊤₂(𝑡),…,𝑧^⊤_𝑁(𝑡))^⊤, 𝑧_𝑘∈ℝ^𝑛, then one has 𝑍^⊤(𝑡)𝑍(𝑡)=^𝑁_𝑖=1𝑒^⊤_𝑖(𝑡)𝑒_𝑖(𝑡). (13)
Notice that 𝜆₁=0 is an eigenvalue of the matrix 𝐵 and its corresponding eigenvector is 𝑢₁=(1/√𝑁,1/√𝑁,…,1/√𝑁)^⊤, then we have 𝑧₁(𝑡)=𝑢^⊤₁⊗𝐼_𝑛𝑒(𝑡)=1√𝑁^𝑁_𝑖=1𝑒_𝑖(𝑡)=0. (14)
According to (13)-(14) and the property of the Kronecker product of the matrices, we obtain 𝑐𝑒^⊤(𝑡)𝐵+Δ⊗Γ𝑒(𝑡)=𝑐𝑒^⊤(𝑡)𝑈⊗𝐼_𝑛×diag𝜆₁,𝜆₂,…,𝜆_𝑁+𝑈^⊤×𝑈Δ𝑈⊗Γ_⊤⊗𝐼_𝑛𝑒(𝑡)≤𝑐𝑍^⊤(𝑡)×diag𝜆₁,𝜆₂,…,𝜆_𝑁+𝜆_max𝑈^⊤Δ𝑈𝐼_𝑁×⊗Γ𝑍(𝑡)≤𝑐𝑍^⊤(𝑡)×diag𝜆₁,𝜆₂,…,𝜆_𝑁+max_{1≤𝑘≤𝑁}𝛿_𝑘𝐼_𝑁⊗Γ𝑍(𝑡)=𝑐^𝑁_𝑖=1𝜆_𝑖+max_{1≤𝑘≤𝑁}𝛿_𝑘𝑧^⊤_𝑖(𝑡)Γ𝑧_𝑖(𝑡)≤^𝑁_𝑖=1𝑐𝜆₂+max_{1≤𝑘≤𝑁}𝛿_𝑘𝑧^⊤_𝑖(𝑡)Γ𝑧_𝑖(𝑡)(𝑟)×≤𝑟𝜆_𝑁𝑖=1𝑒^⊤_𝑖(𝑡)𝑒_𝑖(𝑡)=2𝑟𝜆(𝑟)𝑉(𝑡). (15) Substituting (15) into (12) gives 𝐷⁺𝑉(𝑡)≤2𝐿₁+𝑟𝜆(𝑟)𝑉(𝑡)+2𝐿₂𝑉(𝑡−𝜏(𝑡)),𝑡≠𝑡_𝑘. (16)
When 𝑡=𝑡_𝑘, we have 𝑉𝑡_𝑘=12^𝑁_𝑖=1𝑒^⊤_𝑖𝑡_𝑘𝑒_𝑖𝑡_𝑘=12^𝑁_𝑖=11+𝑑_𝑘²𝑒^⊤_𝑖𝑡⁻_𝑘𝑒_𝑖𝑡⁻_𝑘=1+𝑑_𝑘²𝑉𝑡⁻_𝑘. (17)
Denote 𝑝=2𝐿₁+𝑟𝜆(𝑟) and 𝑞=2𝐿₂. For any 𝜖>0, let 𝜇_𝜖(𝑡) be unique solution of the following impulsive delayed dynamical system: 𝜇̇_𝜖(𝑡)=𝑝𝜇_𝜖(𝑡)+𝑞𝜇_𝜖(𝑡−𝜏(𝑡))+𝜖,𝑡≠𝑡_𝑘,𝑡≥𝑡₀,𝜇_𝜖𝑡_𝑘=1+𝑑_𝑘²𝜇_𝜖𝑡⁻_𝑘,𝑡=𝑡_𝑘,𝑘∈𝑍⁺,𝜇_𝜖(𝑡)=sup_{𝑡0−𝜏≤𝑠≤𝑡0}‖𝑉(𝑠)‖,𝑡₀−𝜏≤𝑡≤𝑡₀. (18) Let 𝑀₀=𝑑^∗sup_{𝑡0−𝜏≤𝑠≤𝑡0}‖𝑉(𝑠)‖ and 𝜂=−(2ln|(1+𝑑^sy_∗)|/𝑇^↓_sup+2ln|(1+𝑑^desy_∗)|/𝑇^↑_inf+𝑝). In the following, we will prove that condition (9) implies 𝜇_𝜖(𝑡)<𝑀₀exp^{−𝜆(𝑡−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝑡≥𝑡₀, (19) where 𝜆>0 is unique positive solution of the equation 𝜆−𝜂+𝑑^∗𝑞exp^𝜆𝜏=0. (20)
Define 𝐻(𝜆)=𝜆−𝜂+𝑑^∗𝑞exp^𝜆𝜏. From condition (9), one has 𝐻(0)<0. Since 𝐻(+∞)>0 and 𝑑𝐻(𝜆)/𝑑𝜆>0, (20) must have unique positive solution 𝜆>0. By the formula for the variation of parameters [29], it follows from (18) that 𝜇_𝜖(𝑡)=𝑊𝑡,𝑡₀𝜇_𝜖𝑡₀+^𝑡_𝑡0𝑊(𝑡,𝑠)𝑞𝜇_𝜖(𝑠−𝜏(𝑠))+𝜖𝑑𝑠,𝑡≥𝑡₀, (21) where 𝑊(𝑡,𝑠),𝑡,𝑠≥𝑡₀ is the Cauchy matrix of linear system [29] 𝜓̇(𝑡)=𝑝𝜓(𝑡),𝑡≠𝑡_𝑘,𝑡≥𝑡₀,𝜓𝑡_𝑘=1+𝑑_𝑘²𝜓𝑡⁻_𝑘,𝑡=𝑡_𝑘,𝑘∈𝑍⁺. (22) According to the representation of the Cauchy matrix [29], we have 𝑊(𝑡,𝑠)=_{𝑠<𝑡𝑘≤𝑡}1+𝑑_𝑘²exp^{𝑝(𝑡−𝑠)},𝑡>𝑠≥𝑡₀. (23)
Suppose that there exist 𝜅_𝑖>0 synchronization impulses with the impulsive strength 𝑑^sy_𝑖 and 𝜅_𝑗>0 desynchronization impulses with the impulsive strength 𝑑^desy_𝑗 in the interval (𝑠,𝑡), then one can easily get that (𝜅_𝑖+1)𝑇^↓_sup≥𝑡−𝑠 and (𝜅_𝑗−1)𝑇^↑_inf≤𝑡−𝑠. Thus, it follows from (23) that 𝑊(𝑡,𝑠)=_{𝑠<𝑡𝑘≤𝑡}1+𝑑_𝑘²exp^{𝑝(𝑡−𝑠)}=^𝜅𝑖_𝑖=11+𝑑^sy_𝑖^2𝜅𝑗_𝑗=11+𝑑^desy_𝑗²exp^{𝑝(𝑡−𝑠)}≤1+𝑑^sy_∗^{2((𝑡−𝑠)/𝑇↓sup−1)}×1+𝑑^desy_∗^{2((𝑡−𝑠)/𝑇↑inf+1)}=1+𝑑^sy_∗⁻²1+𝑑^desy_∗²×exp^{(𝑝+2ln|(1+𝑑sy∗)|/𝑇↓sup+2ln|(1+𝑑desy∗)|/𝑇↑inf)(𝑡−𝑠)}=𝑑^∗exp^{−𝜂(𝑡−𝑠)}. (24) Substituting (24) into (21) gives 𝜇_𝜖(𝑡)≤𝑑^∗exp^{−𝜂(𝑡−𝑡0)}𝜇_𝜖𝑡₀+^𝑡_𝑡0𝑑^∗exp^{−𝜂(𝑡−𝑠)}𝑞𝜇_𝜖(𝑠−𝜏(𝑠))+𝜖𝑑𝑠=𝑀₀exp^{−𝜂(𝑡−𝑡0)}+^𝑡_𝑡0exp^{−𝜂(𝑡−𝑠)}𝑑^∗𝑞𝜇_𝜖(𝑠−𝜏(𝑠))+𝑑^∗𝜖𝑑𝑠,𝑡≥𝑡₀. (25) Since 𝜖,𝜆,𝜂−𝑑^∗𝑞>0, and 𝑑^∗>1, one has 𝜇_𝜖(𝑡)≤𝑑^∗sup_{𝑡0−𝜏≤𝑠≤𝑡0}‖𝑉(𝑠)‖<𝑀₀exp^{−𝜆(𝑡−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝑡₀−𝜏≤𝑡≤𝑡₀. (26) Now, we claim (19) holds; that is, 𝜇_𝜖(𝑡)<𝑀₀exp^{−𝜆(𝑡−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝑡≥𝑡₀. (27) If this is not true, from (26), then there must exists a 𝑡^∗>𝑡₀ such that 𝜇_𝜖𝑡^∗≥𝑀₀exp^{−𝜆(𝑡∗−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝜇_𝜖(𝑡)<𝑀₀exp^{−𝜆(𝑡−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝑡₀−𝜏≤𝑡<𝑡^∗. (28)(29) By (20), (25), and (29), we have 𝜇_𝜖𝑡^∗≤𝑀₀exp^{−𝜂(𝑡∗−𝑡0)}+^𝑡∗_𝑡0exp^{−𝜂(𝑡∗−𝑠)}𝑑^∗𝑞𝜇_𝜖(𝑠−𝜏(𝑠))+𝑑^∗𝜖𝑑𝑠≤exp^{−𝜂(𝑡∗−𝑡0)}×𝑀₀+^𝑡∗_𝑡0exp^{𝜂(𝑠−𝑡0)}×𝑑^∗𝑞𝑀₀exp^{−𝜆(𝑠−𝜏(𝑠)−𝑡0)}+𝜖𝑑^∗𝑞𝜂−𝑑^∗𝑞𝑑^∗−1+𝑑^∗𝜖𝑑𝑠≤exp^{−𝜂(𝑡∗−𝑡0)}𝑀₀+𝑑^∗𝑞𝑀₀×exp^{𝜆𝜏𝑡∗}_𝑡0exp^{(𝜂−𝜆)(𝑠−𝑡0)}+𝑑𝑠𝜖𝜂𝜂−𝑑^∗𝑞𝑑^∗−1×^𝑡∗_𝑡0exp^{𝜂(𝑠−𝑡0)}𝑑𝑠<𝑀₀exp^{−𝜆(𝑡∗−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1. (30) Since 𝑉(𝑡)≤sup_{𝑡0−𝜏≤𝑠≤𝑡0}‖𝑉(𝑠)‖=𝜇_𝜖(𝑡), for 𝑡₀−𝜏≤𝑡≤𝑡₀, it follows from (16)−(18) and Lemma 6 that 𝑉(𝑡)≤𝜇_𝜖(𝑡)<𝑀₀exp^{−𝜆(𝑡−𝑡0)}+𝜖𝜂−𝑑^∗𝑞𝑑^∗−1,𝑡≥𝑡₀. (31) Letting 𝜖→0⁺, then we have 𝑉(𝑡)≤𝑀₀exp^{−𝜆(𝑡−𝑡0)},𝑡≥𝑡₀. (32) This means the zero solution of the error system (7) is globally exponentially stable. The proof of Theorem 7 is thus completed.

Remark 8. Theorem 7 provides a simple yet generic globally exponential synchronization criterion for directed impulsive delayed dynamical network (3) with time-varying impulses. It should be stressed that here both synchronizing and desynchronizing impulses are considered in the model simultaneously. This is different from previous studies on impulsive dynamical networks in [6–15], where the two kinds of impulses are assumed to occur separately. Thus, the impulses discussed in [6–15] can be viewed as a special case of our proposed time-varying impulses.

Remark 9. It can be seen from Theorem 7 that the derived criterion depends mainly on the synchronization and desynchronizing impulsive strengths 𝑑^sy_𝑖 and 𝑑^desy_𝑗, the upper and lower bounds of the synchronization and desynchronizing impulsive intervals 𝑇^↓_sup and 𝑇^↑_inf, and the eigenvalue 𝜆₂. Just as stated in [21, 22], the synchronizability of the dynamical network can also be characterized by the second largest eigenvalue 𝜆₂ of the specific matrix 𝐵. Therefore, the result shows that the network topology also has a great impact on synchronization dynamics of the impulsive delayed dynamical network, which is different from the results presented in [6–8, 14–16].

Remark 10. Condition (9) in Theorem 7 indicates that the sum ((2ln|(1+𝑑^sy_∗)|/𝑇^↓_sup)+(2ln|(1+𝑑^desy_∗)|/𝑇^↑_inf))=2ln(|(1+𝑑^sy_∗)|^1/𝑇↓sup|(1+𝑑^desy_∗)|^1/𝑇↑inf) plays an important role in the synchronization criterion. For convenience, let 𝑑=|(1+𝑑^sy_∗)|^1/𝑇↓sup|(1+𝑑^desy_∗)|^1/𝑇↑inf; then condition (9) can be rewritten as ln𝑑+𝐿₁+𝑟𝜆(𝑟)2+𝑑^∗𝐿₂<0. (33) Since |(1+𝑑^sy_∗)|<1 and |(1+𝑑^desy_∗)|>1, the following two cases will appear. (i)If 0<𝑑<1, that is, |(1+𝑑^desy_∗)|^1/𝑇↑inf<|(1+𝑑^sy_∗)|^{−1/𝑇↓sup}, then ln𝑑< 0. In this case, the restriction condition 𝐿₁+𝑟𝜆(𝑟)/2<−𝐿₂, that is, −𝑝>𝑞≥0, is not required in inequality (33). According to the Halanay differential inequality on delayed dynamical systems [13, 27] and (16), this implies that the underlying delayed dynamical network (1) without impulses might be asynchronous. Therefore, the result shows that even if desynchronization impulses occur frequently, the underlying delayed dynamical networks without impulses which may be asynchronous itself can also be globally exponentially synchronized if synchronization impulses can prevail over the influence of desynchronization impulsive effects. Such statement will be further verified through numerical examples.(ii)If 𝑑≥1, that is, |(1+𝑑^desy_∗)|^1/𝑇↑inf≥|(1+𝑑^sy_∗)|^{−1/𝑇↓sup}, then 𝑝+𝑑^∗𝑞<0 since ln𝑑≥0. Note that 𝑑^∗>1, one has −𝑝>𝑞>0, which means the underlying delayed dynamical network (1) without impulses in fact is globally exponential synchronization itself in this case [13, 27]. Thus, inequality (33) provides a criterion under which globally exponential synchronization of the original delayed dynamical network (1) can be preserved when desynchronization impulses prevail over the influence of synchronization impulsive effects.
In the following, for simplicity, we consider that both the strengths of the synchronization and desynchronization impulses are time-invariant; that is, 𝑑^sy_𝑖≡𝑑^sy, 𝑑^desy_𝑗≡𝑑^desy for 𝑖=1,2,…,𝑁₁, 𝑗=1,2,…,𝑁₂. Then, the following result can be obtained readily from Theorem 7.

Corollary 11. Consider the impulsive delayed dynamical network (3) with time-varying impulses. Under Assumption 2, the impulsive delayed dynamical network (3) is globally exponentially synchronized if the following condition holds: ln1+𝑑^{𝑠𝑦1/𝑇↓sup}1+𝑑^{𝑑𝑒𝑠𝑦1/𝑇↑inf}+𝐿₁+𝑟𝜆(𝑟)2+𝑑^∗𝐿₂<0, (34) where 𝑑^∗=(1+𝑑^{𝑑𝑒𝑠𝑦})²/(1+𝑑^𝑠𝑦)² and the other parameters are given in Theorem 7.

Furthermore, if we assume the impulsive intervals of the synchronization and desynchronization impulses are equidistant and equal to each other, that is, 𝑡^↓_𝑖𝑘−𝑡^↓_𝑖𝑘−1≡Δ𝑡^↓, 𝑡^↑_𝑗𝑘−𝑡^↑_𝑗𝑘−1≡Δ𝑡^↑, and Δ𝑡^↓=Δ𝑡^↑=Δ𝑡, then Corollary 11 is reduced to the following.

Corollary 12. Consider the impulsive delayed dynamical network (3) with time-varying impulses. Under Assumption 2, the impulsive delayed dynamical network (3) is globally exponentially synchronized if the following condition holds: ln1+𝑑^𝑠𝑦1+𝑑^{𝑑𝑒𝑠𝑦}Δ𝑡+𝐿₁+𝑟𝜆(𝑟)2+𝑑^∗𝐿₂<0. (35)

Remark 13. Corollary 12 shows that if the impulsive intervals of the synchronization and desynchronization impulses are equidistant and equal to each other, when |(1+𝑑^sy)(1+𝑑^desy)|<1, that is, when the absolute value of product of (1+𝑑^sy) and (1+𝑑^desy) is less than 1, the whole impulsive effects are synchronizing; when the absolute value is equal to 1, the whole impulsive effects are inactive; and when the absolute value is more than 1, the whole impulsive effects are desynchronizing.

4. Numerical Examples

In this section, two numerical examples are given to illustrate the results derived in this work. The delayed Hopfield neural network is chosen as the isolated node of network (1), which can be described by the following [9]: 𝑥̇(𝑡)=𝑓(𝑡,𝑥(𝑡),𝑥(𝑡−𝜏(𝑡)))=𝐴₁𝑥(𝑡)+𝐸₁𝑔₁(𝑥(𝑡))+𝐹₁𝑔₂(𝑥(𝑡−𝜏(𝑡))), (36) where 𝑥(𝑡)=(𝑥₁(𝑡),𝑥₂(𝑡))^⊤∈ℝ², 𝑔₁(𝑥(𝑡))=𝑔₂(𝑥(𝑡))=(tanh(𝑥₁),tanh(𝑥₂))^⊤, 𝜏(𝑡)=1, and 𝐴₁=−1.000−1.0,𝐸₁=,𝐹2.0−0.1−5.03.0₁=.−1.5−0.1−0.2−2.5 (37)

The single delayed Hopfield neural network (36) has a chaotic attractor as shown in Figure 1. By using the inequality |𝑎𝑏|≤(1/2)𝑎²+(1/2)𝑏², it is easy to check that 𝑥(𝑡)−𝑦(𝑡)^⊤×𝑓((𝑡)((𝑡(𝑡)((𝑡≤𝑥(𝑡)(𝑡)𝑡,𝑥,𝑥𝑡−𝜏)))−𝑓𝑡,𝑦,𝑦𝑡−𝜏))−𝑦^⊤𝐴₁𝑥(𝑡)(𝑡)+()𝑥−𝑦4.55+0.8𝜔₁(𝑡)−𝑦₁(𝑡)²+(5.55+1.35𝜔)𝑥₂(𝑡)−𝑦₂(𝑡)²+0.85𝜔𝑥₁(𝑡−𝜏(𝑡))−𝑦₁(𝑡−𝜏(𝑡))²+1.3𝜔𝑥₂(𝑡−𝜏(𝑡))−𝑦₂(𝑡−𝜏(𝑡))²≤𝜆_max𝐴₁𝑥(𝑡)−𝑦(𝑡)^⊤𝑥(𝑡)−𝑦(𝑡)+𝜆_max𝐴₂𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡))^⊤×𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡))=𝐿₁𝑥(𝑡)−𝑦(𝑡)^⊤𝑥(𝑡)−𝑦(𝑡)+𝐿₂𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡))^⊤×𝑥(𝑡−𝜏(𝑡))−𝑦(𝑡−𝜏(𝑡)), (38) where 𝐴₁=diag((3.55+0.8𝜔),(5.55+1.35𝜔)), 𝐴₂=diag((0.85/𝜔),(1.3/𝜔)), 𝐿₁=𝜆_max(𝐴₁), 𝐿₂=𝜆_max(𝐴₂), and 𝜔 is a positive constant. Thus, the condition of Assumption 2 is satisfied.

Figure 1 

Chaotic attractor of the isolate delayed Hopfield neural network (36) with initial conditions 𝑥1(0)=0.2 and 𝑥2(0)=0.5.

For the reason of convenient explanation, we here consider 𝑑^sy_𝑖≡𝑑^sy, 𝑑^desy_𝑗≡𝑑^desy, 𝑖=1,2,…,𝑁₁, 𝑗=1,2,…,𝑁₂, 𝑡^↓_𝑖𝑘−𝑡^↓_𝑖𝑘−1=𝑡^↑_𝑗𝑘−𝑡^↑_𝑗𝑘−1=Δ𝑡, and the impulsive strength 𝑑_𝑘 (𝑘∈𝑍⁺) satisfies 𝑑_𝑘=𝑑^sy()𝑑,ifmod𝑘,2≠0,_desy,ifmod(𝑘,2)=0. (39)

Example 1. In this example, a nearest-neighbor unidirectional coupled impulsive delayed dynamical network (3) with time-varying impulses is considered. It is well known that the synchronization of the nearest-neighbor coupled dynamical network is difficult to achieve if the number of network nodes is large enough [21]. The coupling matrix 𝐵 of this network is of the form 𝐵=−110⋯00−11⋯0⋮⋮⋱⋱⋮0⋯0−1110⋯0−1_𝑁×𝑁. (40) Clearly, the coupling matrix 𝐵 is a asymmetrical Laplacian matrix of a weighted graph with a spanning tree [9]. In this simulation, choosing 𝑁=100, then one has 𝜆₂=−0.0039 and max_{1≤𝑘≤𝑁}𝛿_𝑘=0. Let the coupling strength 𝑐=1 and the inner coupling matrix Γ=𝐼₂; then 𝑟𝜆(𝑟)=−0.0039.

Select 𝜔=2; one has 𝐿₁=8.25 and 𝐿₂=0.65. Let 𝑑^sy=−0.5 and 𝑑^desy=0.2, by Corollary 12, then it can be concluded that the nearest-neighbor unidirectional coupled impulsive delayed dynamical network can be globally exponentially synchronized if Δ𝑡<−ln1+𝑑^sy1+𝑑^desy𝐿₁+𝑟𝜆(𝑟)/2+𝑑^∗𝐿₂=0.0525. (41)

Let the equidistant impulsive interval be taken as Δ𝑡=0.05. Figure 2 shows the time-varying impulses sequence. Figure 3 visualizes the change process of the state variables of the nearest-neighbor unidirectional coupled delayed dynamical network without impulses, which clearly indicates desynchronization of the underlying delayed dynamical network without impulses. Error trajectories of the nearest-neighbor unidirectional coupled impulsive delayed dynamical network are plotted in Figure 4. It can be seen that the network achieves quickly synchronization under the time-varying impulses. Since here |(1+𝑑^sy)(1+𝑑^desy)|=0.6<1, the whole impulsive effects are synchronizing. This example verifies the above statement that even if desynchronization impulses occur frequently, an unsynchronized delayed dynamical network (1) can also be globally exponentially synchronized under time-varying impulses if synchronization impulses can prevail over the influence of desynchronization impulsive effects.

Figure 2 

Time-varying impulsive sequence with 𝑑sy=−0.5, 𝑑desy=0.2, and Δ𝑡=0.05 in time interval 02.

Figure 3 

Change process of the state variables of the nearest-neighbor unidirectional coupled delayed dynamical network without impulses in time interval 020.

Figure 4 

Error trajectories of the nearest-neighbor unidirectional coupled delayed dynamical network with the time-varying impulses in time interval 02.

Example 2. A BA scale-free [31] coupled impulsive delayed dynamical network (3) is taken as the second example. The parameters of the BA model are given by 𝑚₀=𝑚=4 and 𝑁=100. We generate 100 BA scale-free networks randomly, and the network with 𝜆₂=−4.3532 is selected for simulation in this example. The remaining 99 scale-free networks can be analyzed similarly [10]. By simple calculation, we get that max_{1≤𝑘≤𝑁}𝛿_𝑘=0. Let the coupling strength 𝑐=5 and the inner coupling matrix Γ=𝐼₂. Select 𝜔=1.5; one has 𝐿₁=7.575 and 𝐿₂=0.8667. Then, 2𝐿₁+𝑟𝜆(𝑟)+2𝐿₂=−4.8827<0; that is, −𝑝>𝑞>0. Thus, the underlying delayed dynamical network without impulses is globally exponentially synchronized itself in this example, as shown in Figure 5.

Figure 5 

Change process of the state variables of the scale-free coupled delayed dynamical network without impulses in time interval 050.

Let 𝑑^sy=−0.1 and 𝑑^desy=0.4. By Corollary 12, we can derive the following condition ensuring globally exponential synchronization of the scale-free coupled delayed dynamical network with time-varying impulses will be maintained Δ𝑡>−ln1+𝑑^sy1+𝑑^desy𝐿₁+𝑟𝜆(𝑟)/2+𝑑^∗𝐿₂=0.1909. (42)

Select Δ𝑡=0.2; Figure 6 represents the time-varying impulses sequence, and the synchronization process of the scale-free coupled impulsive delayed dynamical network is plotted in Figure 7. In this example, the whole impulsive effects are desynchronizing because |(1+𝑑^sy)(1+𝑑^desy)|=1.26>1. We can see that globally exponential synchronization of the underlying delayed dynamical network without impulses is preserved under such time-varying impulses.

Figure 6 

Time-varying impulsive sequence with 𝑑sy=−0.1, 𝑑desy=0.4, and Δ𝑡=0.2 in time interval 05.

Figure 7 

Synchronization of the scale-free coupled delayed dynamical network with the time-varying impulses in time interval 05.

5. Conclusion

In this paper, a detailed analysis has been carried out for the synchronization of directed complex networks with time-varying delays dynamical nodes and time-varying impulsive effects. Both synchronizing and desynchronizing impulses were considered in the network model simultaneously. Without assuming symmetry and irreducibility of coupling structure, some globally exponential synchronization criteria for the proposed impulsive delayed dynamical networks with weakly connected topology have been established by using the comparison principle. The obtained results showed that even if desynchronization impulses occur frequently, the underlying delayed dynamical networks without impulses which may be asynchronous itself can also be globally exponentially synchronized if synchronization impulses can prevail over the influence of desynchronization impulsive effects. Two numerical examples and their simulations have been given to verify the effectiveness of the theoretical results.

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

The authors are grateful to the reviewers and the editor for their valuable comments and suggestions. This work was supported by the National Science Foundation of China (Grant no. 11172158), National Science Foundation of China, Tian Yuan Special Foundation (Grant no. 11326193), the Research Foundation for Advanced Talents of Jiangsu University (Grant no. 13JDG027), and the First-Class Discipline of Universities in Shanghai.