Motivated by the importance and application of discrete dynamical systems, this paper presents a new Lyapunov characterization which is an extension of conventional Lyapunov characterization for multistable discrete-time nonlinear systems. Based on a new type stability notion of 𝒲-stability introduced by D. Efimov, the estimates of solution and the Lyapunov stability theorem and converse theorem are proposed for multi-stable discrete-time nonlinear systems.

1. Introduction

Theory of discrete-time systems is rapidly developed and widely applied to various fields (see three remarkable books [1–3]). In paper [4], motivated by a continuous second-order predator-prey ecological system of Lotka-Volterra type, Efimov introduces a new type notion, 𝒲-stable and presents Lyapunov characterization for multistable continuous nonlinear systems. For ecological systems, they usually considered that evolution and translation of populations is continuous. Thus continuous models [5, 6] are considered in many references. However, according to observing the translation process of population change, discrete models are better to represent ecological systems [7, 8]. Thus it is meaningful to study 𝒲-stability for discrete dynamical systems. This paper extends stability results given by Efimov about continuous multistable systems to discrete-time multistable systems.

Stability analysis is one of the main issues for research of control systems theory. A rapid progress has been made in local or global stability analysis for unique equilibrium [9], trajectories [10], close invariant set [11, 12], part of state variable [13], and so forth. In papers [14, 15], Sontag and Wang introduce Lyapunov characterization of input to state stability for continuous systems. Paper [16] by Jiang and Wang presents the property of input to state stability for discrete-time systems. Converse Lyapunov theorem is presented in paper [11] by Lin et al. and paper [17] by Jiang and Wang for continuous and discrete-time systems, respectively. In recent years, multistable systems have attracted considerable attention [8, 18–21]. There have many methods to deal with stability problem for multistable systems. Two popular modern approaches are based on density functions [19] and monotone systems [20]. The former approach substitutes conventional Lyapunov function with density function for establishing stability of stable set. The latter approach develops some constructive conditions based on monotone systems for establishing stability of the set of equilibriums. The above approaches are effective to handle the stability problem of multistable systems. The stability results obtained according to the above approaches are based on conventional stable notions. However, in the areas of theoretical biology and engineering, many systems that represent models are called multistable systems. The set of all invariant solutions of those systems contains stable subset and unstable subset.

In this paper, we firstly introduce the notion of 𝒲-stability to discrete-time multistable systems. Using some important approaches and techniques of stability analysis in papers [4, 11, 14–18], new Lyapunov characterizations are proposed for discrete-time multistable systems. Based on notions of 𝒲-Lyapunov function and weak 𝒲-Lyapunov function, the relation of two functions is presented, and a converse Lyapunov theorem is proved. Our main contribution is that Lyapunov characterizations presented in this paper contain the conventional Lyapunov characterization and it should be extensively applied.

The rest of the paper is organized as follows. Problem statement and mathematical preliminary are presented in Section 2. Stability results of multistable discrete-time nonlinear systems are proposed in Section 3. Finally, a brief conclusion is provided to summarize the paper in the final section.

2. Problem Statement and Mathematical Preliminary

Consider the following discrete nonlinear system:x(k+1)=f(x(k)),
where x(k)∈Rn is the system state vector at time instant k, k∈Z+={0,1,2,…}, f:Rn→Rn is locally Lipschitz continuous. x(k,x0) denotes the solution for any initial value x0=x(k0), and k0∈Z+. ∥x∥ denotes the Euclidean norm of vector x.

Let ℳ be a nonempty subset of Rn. The set is called (forward) invariant for system (2.1) ifx0∈ℳ⇒x(k,x0)∈ℳ,∀k∈Z+.
An invariant set is called minimal if it does not contain other smaller invariant sets. The distance of the set ℳ from a point x∈Rn is defined as∥x∥ℳ=dist(x,ℳ)=infξ∈ℳ∥x-ξ∥.

An invariant set 𝒜 is said to be a locally attracting set if there exists an open neighborhood 𝒰 of 𝒜 such that, for any x0∈𝒰, limk→+∞∥x(k,x0)∥𝒜=0. An invariant set ℛ is said to be a locally repelling set if exists an open neighborhood 𝒰 of ℛ such that, for any x0∈𝒰-ℛ, limk→+∞∥x(k,x0)∥ℛ>0.

Let 𝒲⊂Rn be the set of all invariant solutions of system (2.1). Clearly, it is an invariant set. Assume the set 𝒲 is a closed and compact set and satisfies 𝒲=𝒜∪ℛ, where 𝒜 and ℛ denote attracting set and repelling set, respectively.

We first introduce the notion of 𝒲-asymptotical stability [4] for continuous nonlinear systems to discrete-time nonlinear systems (2.1).

Definition 2.1.

The system (2.1) is called 𝒲-stable with respect to 𝒲 if, for some given constant R≥0 and for each R≤ɛ≤+∞, there exists 0≤δ=δ(ɛ)<+∞, such that when ∥x0∥𝒲≤δ, it holds
∥x(k,x0)∥𝒲≤ɛ,∀k∈Z+.

Remark 2.2.

It is possible to exist unstable equilibriums in the set 𝒲 which contains all invariant solutions. When the trajectories of system initiate from a neighborhood of unstable equilibriums, it cannot ensure the trajectories in this neighborhood. Thus, the definition of 𝒲-stability need the existence R≥0 such that ɛ≥R. If ℛ is not empty then R>0. Particularly, when ℛ is empty (it implies R=0), 𝒲-stability is reduced to conventional stability (see [1, 2, 9]). The constant can be related with the radius of the set 𝒲 [4].

Definition 2.3.

The system (2.1) is called 𝒲-asymptotically stable with respect to 𝒲 if

it is 𝒲-stable;

it satisfies 𝒲-attracting property. There is a positive constant c, such that, for all ∥x0∥𝒲≤c, limk→+∞∥x(k,x0)∥𝒲=0; that is, for each 0<η<+∞ and ∥x0∥𝒲≤c, there exists T=T(x0,η)∈Z+ such that
∥x(k,x0)∥𝒲<η,∀k≥T(x0,η).

Remark 2.4.

T(x0,η) is dependent on x0, and η. T(·,η) is different for different initial values; that is, there exists no uniform time of convergence to 𝒲 of the trajectories which start from the neighborhood of different initial values due to the presence of unstable equilibriums.

Example 2.5.

Consider a second-order discrete system
x(k+1)=x(k)+hy(k),y(k+1)=y(k)+hy(k)(-y(k)+x(k)-x3(k)).
The set of all invariant solutions is 𝒲={(0,0),(1,0),(-1,0)}. Let h=0.1. Simulation results are shown in Figure 1.

𝒲-asymptotical stability of system (2.6) with different initial values. (a) and (b) show phase portraits of system (2.6) with initial value (−0.01, 0.01) and (0.01, −0.01), respectively.

According to Figure 1, we can get that (0,0) is unstable and (1,0) and (-1,0) is asymptotically stable. Thus 𝒜={(1,0),(-1,0)}, and ℛ={(0,0)}. System (2.6) is 𝒲-asymptotically stable by Definition 2.3.

Example 2.6.

Consider a second-order predator-prey system with a prey refuge in the following form:
x(k+1)=x(k)+hx(k)(β1-b1x(k)-a1(1-m)y(k)),y(k+1)=y(k)+hy(k)(β2-a2y(k)(1-m)x(k)),
where x(k), y(k) represent the prey and predator density, parameters β1 and β2 are the intrinsic growth rates of the prey and the predator, respectively. h is the step size. m∈[0,1) is a refuge protecting coefficient of the prey. The rest coefficients are positive constants. Let m=1/2,h=0.1, β1=β2=2, and a1=a2=b=1. The set of all invariant solutions of system (2.7) is 𝒲={(0,0),(2,0),(4/3,4/3)}. Simulation results are shown in Figure 2.

𝒲-asymptotical stability of system (2.7) with different initial values. (a) and (b) show phase portraits of system (2.7) with initial value (0.01,0.01) and (2.01,0.01), respectively.

According to Figure 2, the trajectories of solutions from the neighborhood of points (0,0) and (2,0) are convergent to point (4/3,4/3). We have (0,0) and (2,0) are unstable and (4/3,4/3) is asymptotically stable. System (2.7) is 𝒲-asymptotical stable.

Remark 2.7.

For system (2.7), there have been many important and interesting results, such as the global stability, periodic solutions, almost periodic solutions, and chaos (see [4–8, 14, 15, 22]). Here our interesting focus is on 𝒲-asymptotical stability of the set of all invariant solutions.

Throughout this paper, assume 𝒜 and ℛ are not empty. In this case, we can exclude an open set containing ℛ from admissible set of initial value. Then there exists a uniform convergent time T in the definition of attracting property. That is, 𝒲-attracting property is similar to the definition of attracting property in [9, 18]. Define a hyper-surfaceΣ={x∈Rn:∥x∥ℛ-∥x∥𝒜=0}.
Then we obtainΣ-={x∈Rn:∥x∥ℛ-∥x∥𝒜<0},Σ+={x∈Rn:∥x∥ℛ-∥x∥𝒜>0}.
Let an open neighborhood 𝒰 of ℛ be 𝒰=ℬρ={x∈Rn:∥x∥ℛ<ρ}. We can choose some ρ* which ensure the properties ℬρ⊂Σ-, ℬρ∩Σ+=∅ hold only for ρ<ρ*.

Definition 2.8.

System (2.1) satisfies 𝒲-attracting property. Choose any open set ℬρ of ℛ. For each η>0 and r>0 there exists T=T(r,η)∈[0,+∞) such that for any x0∈{x:∥x∥𝒲<r}-ℬρ∥x(k,x0)∥𝒲<η,∀k≥T(r,η).

3. Stability Results3.1. Estimates of SolutionLemma 3.1.

The system (2.1) is 𝒲-stable with respect to 𝒲 for some given R≥0 if and only if for any constant r≥0 there exists a class 𝒦 function α such that for any ∥x0∥𝒲≤r∥x(k,x0)∥𝒲≤α(∥x0∥𝒲+R̃),∀k∈Z+,
where R̃=α-1(R).

Proof.

Sufficiency. For system (2.1), when ∥x0∥𝒲≤r, (3.1) holds for any k∈Z+. For each ɛ=α(r+R̃)≥R, choose δ=min{r,α-1(ɛ)-R̃}, when ∥x0∥𝒲≤δ, ∥x(k,x0)∥𝒲≤ɛ holds for any k∈Z+. 𝒲-stability is ensured.

Necessity. Assume system (2.1) is 𝒲-stable, that is, for some given R≥0 and for each ɛ≥0, there exists δ≥0, such that
∥x0∥𝒲≤δ⇒∥x(k,x0)∥𝒲≤ɛ+R,k∈Z+.
For fixed ɛ and R, let δ¯(ɛ) be the supremun of all applicable δ(ɛ). Clearly, the function δ¯(r) is positive and nondecreasing. So there exist a class 𝒦 function φ and a constant λ∈(0,1) such that φ(r)≤φ(r+R)≤λ(δ¯(r)+R̃), where R̃=φ(R). Let α(r)=φ-1(r). Then α is a function of class 𝒦. Let c=limr→+∞α(r). Given ∥x0∥𝒲≤c-R̃, let ɛ+R=α(∥x0∥𝒲+R̃). Then we have ∥x0∥𝒲≤φ(ɛ+R)-R̃≤δ¯(ɛ) and ∥x(k,x0)∥𝒲≤α(∥x0∥𝒲+R̃)=ɛ+R. Thus 𝒲-stability implies the property as in inequality (3.1).

Remark 3.2.

The construction approach of class 𝒦 function α is similar to its in [9, 11]. When R=0, the result of Lemma 3.1 is the same as the corresponding result in [9, 11].

Lemma 3.3.

The system (2.1) is 𝒲-asymptotically stable with respect to 𝒲 if and only if for some constant R≥0 and any constant r≥0 there exist a class 𝒦ℒ function β and a class 𝒦 function μ such that for any ∥x0∥𝒲≤r∥x(k,x0)∥𝒲<β(∥x0∥𝒲+R,k),∀k∈Z+.

Proof.

Sufficiency. Suppose there is a class 𝒦ℒ function β such that inequality (3.3) is satisfied. With fixed k=k1, we have
∥x(k,x0)∥𝒲<β(∥x0∥𝒲+R,k1).
By Lemma 3.1, system (2.1) is 𝒲-stable. For any ∥x0∥𝒲≤ζ, it yields
∥x(k,x0)∥𝒲<β(ζ+R,k).
It implies ∥x(k,x0)∥𝒲→0 as k→+∞. Attracting property is satisfied. Thus, system (2.1) is 𝒲-asymptotically stable.

Necessity. Suppose that system (2.1) is 𝒲-asymptotically stable. According to Lemma 3.1, there exists a class 𝒦 function α such that for any ∥x0∥𝒲≤r∥x(k,x0)∥𝒲≤α(∥x0∥𝒲+R̃).
Moreover, choose an arbitrary small constant ρ which satisfies 0<ρ<ρ*. For any x0∈{x:∥x∥𝒲≤r}-ℬρ and given η>0, there exists T(r,η) such that
∥x(k,x0)∥𝒲<η,∀k≥T(r,η).
Let T¯(r,η) be the infimun of T(r,η). The function T¯(r,η) is nonnegative in r and nonincreasing in η, and T¯(r,η)=0 for all η>α(r+R). Let
ψr(η)=2η∫η/2ηT¯(r,s)ds>T¯(r,η).
The function ψr(η) is positive and has the following properties:

for each fixed r, ψr(η) is continuous, strictly decreasing, and ψr(η)→0 as η→+∞;

for each fixed η, ψr(η) is strictly increasing in r.

Take χr(η)=ψr-1(η). Then χr(η) also satisfies the above two properties and T¯(r,χr(η))<ψr(χr(η))=η. So
∥x(k,x0)∥𝒲≤χr(k),∀k∈Z+,∀x0∈{∥x0∥𝒲≤r}-ℬρ.
According to inequalities (3.6) and (3.9) we get
∥x(k,x0)∥𝒲≤α(∥x0∥𝒲+R̃)χr(k)2=β(∥x0∥𝒲+R,k).
Thus, according to arbitrariness of ρ, there exists a class 𝒦ℒ function β such that inequality (3.3) is satisfied.3.2. Stability TheoremDefinition 3.4.

A continuous function 𝒲:Rn→R is a 𝒲-Lyapunov function with respect to 𝒲 for system (2.1) if

there exist class 𝒦 functions α1 and α2 and a constant R≥0 such that for any k∈Z+ and x0∈Rnα1(∥x0∥𝒲)≤𝒲(x(k,x0))≤α2(∥x0∥𝒲+R);

there exists a class 𝒦 function α3 such that for any k∈Z+ and x0∈Rn𝒲(f(x(k,x0)))-𝒲(x(k,x0))≤-α3(∥x0∥𝒲).

Assume system (2.1) has output y(k)=h(x(k)), where h:Rn→Rm is a continuous function. The system (2.1) with output y(x(k)) is 𝒲-detectable if for any k∈Z+ and x0∈Rny(x(k,x0))=h(x(k,x0))≡0⇒∥x(k,x0)∥𝒲→0,k→+∞.Definition 3.5.

A continuous function 𝒲:Rn→R is a weak 𝒲-Lyapunov function with respect to 𝒲 for system (2.1) if

there exist class 𝒦 functions α1 and α2 and a constant R≥0 such that for any k∈Z+ and x0∈Rnα1(∥x0∥𝒲)≤𝒲(x(k,x0))≤α2(∥x0∥𝒲+R);

there exists a continuous function α3:Rm→[0,∞), with α3(0)=0 and α3(s)>0 for all s≠0 such that for any ∈Z+ and x0∈Rn𝒲(f(x(k,x0)))-𝒲(x(k,x0))≤-α3(∥h(x(k,x0))∥)≤0;

system (2.1) with output y(x(k)) is 𝒲-detectable.

Theorem 3.6.

Weak 𝒲-Lyapunov function implies 𝒲-asymptotical stability.

Proof.

Suppose system (2.1) has a weak 𝒲-Lyapunov function. Using inequality (3.15), we have
𝒲(f(x(k,x0)))-𝒲(x(k,x0))≤0.
It implies 𝒲 is bounded and
α1(∥x0∥𝒲)≤𝒲(x(k,x0))≤𝒲(x0)≤α2(∥x0∥𝒲+R),∀k∈Z+.
Thus
∥x(k,x0)∥𝒲≤α1-1(α2(∥x0∥𝒲+R)).
According to Lemma 3.1, system (2.1) is 𝒲-stable with respect to 𝒲.

Since the compactness of set 𝒲, there exist a class 𝒦 function μ and a positive constant c such that
∥x(k)∥≤μ(∥x(k)∥𝒲+c),
which shows x(k) is bounded. Thus, for any solution of system (2.1) we can find a forward invariant attracting compact set Ω(x(0)). Choosing x0∈Ω(x(0)), we have x(k,x0)∈Ω(x(0)) for any k∈Z+.

Furthermore, by 𝒲(f(x(k,x0)))-𝒲(x(k,x0))≤0 we get 𝒲(x(k,x0)) is nonincreasing. However, by 𝒲(x(k,x0))≥α1(∥x0∥𝒲) we obtain 𝒲(x(k)) is nondecreasing. Thus, there exists a positive constant d such that
𝒲(x(k,x0))=d,
which implies
𝒲(f(x(k,x0)))-𝒲(x(k,x0))=0.
Then α3(h(x(k,x0)))=0. Using the detectability of system (2.1) with y(x(k)), we have ∥x(k,x0)∥𝒲→0 as k→+∞. That is, 𝒲-attracting property holds.

The proof is completed.

Lemma 3.7 is given by Sontag in [15] which is useful for proof of Theorem 3.8.

Lemma 3.7.

Assume that β is a function of class 𝒦ℒ. Then there exist two class 𝒦∞ functions φ1 and φ2 such that
β(s,r)≤φ1(φ2(s)e-r),∀s≥0,∀r≥0.

Theorem 3.8.

Considering system (2.1), the following is equivalent:

there exists a 𝒲-Lyapunov function;

there exists a weak 𝒲-Lyapunov function;

there is 𝒲-asymptotically stable.

Proof.

(a)⇒(b). Let h(x(k))=α3(𝒲(x(k,x0))). Because α3 is a class 𝒦 function, h(x(k)) is continuous and inequality (3.15) is satisfied. Furthermore, when h(x(k,x0))≡0, we have
∥x(k,x0)∥𝒲→0,k→+∞.
The Property (iii) of Definition 3.5 holds. Thus (a) implies (b).

(b)⇒(c). The proof is given in Theorem 3.6.

(c)⇒(a). Assume that system (2.1) is 𝒲-asymptotically stable. According to Lemma 3.3, there exist a class 𝒦ℒ function β and a class 𝒦 function μ such that for any x0∈Rn∥x(k,x0)∥𝒲≤β(∥x0∥𝒲+R,k),∀k∈Z+.
By Lemma 3.1, there exist two class 𝒦∞ functions φ1 and φ2 such that
∥x(k,x0)∥𝒲≤β(∥x0∥𝒲+R,k)≤φ1(φ2(∥x0∥𝒲+R)e-k).
Let θ=φ1-1. We have
θ(∥x(k,x0)∥𝒲)≤φ2(∥x0∥𝒲+R)e-k.
Define 𝒲(x(k,x0))=∑k=0∞θ(∥x(k+k0,x0)∥𝒲). Clearly, 𝒲(x(k,x0)) is continuous since φ1 is a class 𝒦∞ function. By the definition of 𝒲, it yields
θ(∥x0∥𝒲)≤𝒲(x(k,x0))≤∑k=0∞φ2(∥x0∥𝒲+R)e-k≤ee-1φ2(∥x0∥𝒲+R).𝒲 satisfies the property of inequality (3.11) due to inequality (3.27).

In the following we show 𝒲(x(k)) satisfies the property as in inequality (3.12). Arbitrarily choose k0∈Z+ and x0=x(k0)∈Rn. Then x(k0+1,x(k0))=f(x(k0,x(k0))) and
𝒲(x(k0+1,x(k0+1)))=∑k=0∞θ(∥x(k+k0+1,x(k0+1))∥𝒲).
Since function f(x(k)) is Lipschitz continuous, the solution of system (2.1) is unique for arbitrary initial value. Then we have
x(k+k0+1,x(k0+1))=x(k+k0+1,x(k0)),∀k∈Z+.
Considering (3.28), we can get
𝒲(x(k0+1,x(k0+1)))=∑k=0∞θ(∥x(k+k0+1,x(k0+1))∥𝒲)=∑k=0∞θ(∥x(k+k0+1,x(k0))∥𝒲)=∑k=1∞θ(∥x(k+k0,x(k0))∥𝒲)=∑k=0∞θ(∥x(k+k0,x(k0))∥𝒲)-θ(∥x(k0,x(k0))∥𝒲)≤𝒲(x(k0,x(k0)))-θ(∥x(k0,x(k0))∥𝒲).
That is,
𝒲(x(k0+1,x(k0,x(k0))))-𝒲(x(k0,x(k0)))≤-θ(∥x0∥𝒲).
Due to the arbitrariness of k0, we have
𝒲(f(x(k,x0)))-𝒲(x(k,x0))≤-θ(∥x0∥𝒲).
Function 𝒲 satisfies the property as in inequality (3.12).

4. Conclusion

We conclude with a brief discussion. The notion of 𝒲-stability introduced by Efimov is different from conventional notion of stability. It is required to consider the set of all invariant solutions of systems. However, the set of all invariant solutions can contain not only stable invariant solutions but also unstable invariant solutions. If it does not contain unstable invariant solutions, 𝒲-stability is conventional stability. Thus our results should have an more extensive application than those corresponding results in the sense of conventional stability.

Acknowledgment

This work was supported by the Natural Science Foundation of China under the Contract no. 60874006.

AgarwalR. P.LakshmikanthamV.TrigianteD.La SalleJ. P.EfimovD.On global Lyapunov characterization of multi-stable nonlinear systemsProceedings of the 48th IEEE Conference on Decision and Control2009Shanghai, China62996304ChenF.The permanence and global attractivity of Lotka-Volterra competition system with feedback controlsLaurenM.LellershohnN.Multi-stability: a major means of differentiation and evolution in biological systemsSongY.YuanS.ZhangJ.Bifurcation analysis in the delayed Leslie-Gower predator-prey systemÇelikC.DumanO.Allee effect in a discrete-time predator-prey systemKhalilH. K.SepulchreR.JankovićM.KokotovićP. V.LinY.SontagE. D.WangY.A smooth converse Lyapunov theorem for robust stabilityChengG. F.MuX. W.Finite-time stability with respect to a closed invariant set for a class of discontinuous systemsVorotnikovV. I.SontagE. D.WangY.New characterizations of input-to-state stabilitySontagE. D.Comments on integral variants of ISSJiangZ. P.WangY.Input-to-state stability for discrete-time nonlinear systemsJiangZ. P.WangY.A converse Lyapunov theorem for discrete-time systems with disturbancesMalisoffM.MazencF.Constructions of strict Lyapunov functions for discrete time and hybrid time-varying systemsAngeliD.FerrellF. E.SongtagE. D.Detection of multistability, bi-furcations and hysteresis in a large class of biological positive-feedback systemsEncisoG.SontagE. D.Monotone systems under positive feedback: multistability and a reduction theoremAngeliD.angeli@dsi.unifi.itSontagE. D.sontag@math.rutgers.eduOscillations in I/O monotone systems under negative feedbackZhuangK.WenZ.Dynamical behaviors in a discrete predator-prey model with a prey refuge