Abstract
In this work, we consider impulsive dynamical systems evolving on an infinite-dimensional space and subjected to external perturbations. We look for stability conditions that guarantee the input-to-state stability for such systems. Our new dwell-time conditions allow the situation, where both continuous and discrete dynamics can be unstable simultaneously. Lyapunov like methods are developed for this purpose. Illustrative finite and infinite dimensional examples are provided to demonstrate the application of the main results. These examples cannot be treated by any other published approach and demonstrate the effectiveness of our results.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Impulsive dynamical systems provide a mathematical modeling framework for practical processes where a combination of continuous and discrete dynamics takes place. Such a hybrid dynamics appears in many applications, for example, in case of mechanical collisions or in control systems involving a combination of analog and digital controllers. As well in pandemic systems, a mass vaccination can be modeled as an impulsive action meaning a (nearly instantaneous) transition of a large amount of susceptible individuals to become immune.
A combination of discrete and continuous dynamics leads to a higher complexity in the behavior of solutions compared with a purely discrete or purely continuous system. Such unexpected effects as instability [27] or chaos [14] can arise. In particular, such properties as stability and robustness are more difficult to investigate, especially in case of nonlinear systems.
Stability in the sense of Lyapunov of nonlinear impulsive systems has a long history of investigations, see [17, 23]. Later, more general stability notions were developed for hybrid systems, which include impulsive ones as a particular case, see [12, 28]. These notions use the generalized (hybrid) time concept, which allows to develop rather general results for a wide class of hybrid systems, including impulsive, switched and sampled data systems.
In case of systems having input signals, the notion of input-to-state stability (ISS) was introduced in [26] and was found very fruitful in many applications [16]. This framework was also successfully used for studying robust stability of impulsive systems, see [4, 6, 13, 18, 22]. In particular, [6] derives dwell-time conditions to establish the ISS property for nonlinear impulsive systems on infinite dimensional state spaces. This result is based on certain stability assumptions imposed either on the continuous or on the discrete dynamics. The ISS is assured by the stability property of either of both dynamics using a suitable dwell-time condition. Stability of interconnected impulsive systems is then studied in the case, when the ISS-Lyapunov functions are known for the subsystems, which leads to a combination of the dwell-time and small-gain conditions. The ISS property of impulsive systems where impulsive actions depend on time was studied in [4], where new and rather general dwell-time conditions were developed.
A small-gain theorem for \(n\ge 2\) interconnected hybrid systems was established in [22] for the case where not all subsystems are assumed to be ISS, which extends the results of [6]. Similar results for nonlinear interconnected impulsive systems were developed in [8] for the case of absent external perturbations. This work derives sufficient stability conditions for interconnected systems by means of vector Lyapunov functions, which leads to conditions similar to the small-gain ones.
In most of works (e.g., in those mentioned above) studying stability of impulsive systems by means of Lyapunov methods, it is assumed that the discrete and continuous dynamics share a common Lyapunov function V which decays either on jumps or along the continuous flow. A dwell-time condition allows to compensate the destabilizing effect of one type of dynamics by the stabilizing property of the other one. Certainly, if V increases in both cases, then the system is unstable. However, in general, it can happen that the whole system is asymptotically stable even in the case when both discrete and continuous dynamics are unstable. Identification of such systems needs a more refined consideration of the interaction between both dynamics types. It is expected that stability conditions become more involved in this case. It should be noted that there are not many stability results in the literature that cover the case of simultaneous instability of discrete and continuous dynamics: [1, 9,10,11, 19, 24], see also [3] and [2] for the linear case. However, these results cannot be extended directly to the case of nonlinear infinite-dimensional systems with inputs, also we note that only local stability was studied in the first group of these papers.
Vector Lyapunov functions were used in [19] to establish stability results, where second-order derivatives of Lyapunov functions along solutions enter to the stability conditions. These results were generalized in [9, 10], where higher-order derivatives of Lyapunov functions are employed. This approach cannot be used in case of systems with inputs as one would need to require infinitely smooth disturbances, which is very restrictive in real applications.
Averaged dwell-time conditions were considered in [1, 24] to establish stability of a linear impulsive system on a Banach space. Based on the identities from the commutator calculus new comparison theorems, constructions of Lyapunov functions and conditions for the local asymptotic stability were developed there. In this work, we are interested in global stability properties for nonlinear systems with inputs.
With an exception of [22], in the most of works devoted to investigations of the ISS-like properties of hybrid systems, it is assumed that either the continuous or discrete dynamics satisfies the ISS property. Hence, it is interesting to further develop the direct Lyapunov method for the case, when both types of dynamics fail to be ISS. Our paper contributes into this direction, providing a new approach and new stability conditions.
In this work, we improve the results of [9, 10] and extend them to the case of ISS for nonlinear impulsive systems. We derive stability conditions by means of a series of Lyapunov-like functions. Instead of higher-order derivatives of Lyapunov functions employed in [9, 10], we use an infinite sequence of auxiliary functions to provide estimates of the dwell-time in order to guarantee the ISS property. The obtained results are then applied to the studying of the global asymptotic stability of linear impulsive systems with continuous dynamics governed by a parabolic PDE. The ISS property is also studied for this type of systems. Moreover, we derive conditions for the ISS property of nonlinear locally homogeneous finite-dimensional impulsive systems.
The paper consists of six sections. Section 2 introduces the notation and several auxiliary inequalities used in the paper. The problem statement is described in Sect. 3. Section 4 contains the main results with their proofs. Application of the results to the investigation of GAS and ISS properties of linear impulsive systems in infinite-dimensional spaces and of nonlinear finite-dimensional systems is provided in Sect. 5. A brief discussion and conclusions are collected in Sect. 6. Proofs of several technical results are placed in Appendix.
2 Notation and preliminaries
We use the following classes of comparison functions:
\({\mathcal {K}}=\{\gamma \,:\,\mathbb R_+\rightarrow \mathbb R_+\,:\,\text {continuous, strictly increasing and} \,\,\gamma (0)=0\},\)
\({\mathcal {K}}_{\infty }=\{\gamma \,:\,\mathbb R_+\rightarrow \mathbb R_+\,:\gamma \in {\mathcal {K}},\quad \gamma (s)\rightarrow \infty \quad \text {for}\quad s\rightarrow \infty \},\)
\({\mathcal {L}}=\{\gamma \,:\,\mathbb R_+\rightarrow \mathbb R_+\,:\,\text {continuous, strictly decreasing and}\,\, \lim \nolimits _{t\rightarrow \infty }\gamma (t)=0\},\)
\({\mathcal {K}}{\mathcal {L}}=\{\beta \,:\,\mathbb R_+\times \mathbb R_+\rightarrow \mathbb R_+\,:\,\text {continuous}\,\, \beta (\cdot ,t)\in {\mathcal {K}},\,\,\beta (s,\cdot )\in {\mathcal {L}}\}.\)
By C[0, l] we denote the space of continuous functions, defined on [0, l] with values in \(\mathbb R\) with norm \(\Vert f\Vert _{C[0,l]}=\max \nolimits _{x\in [0,l]}|f(x)|\), \(C^k[0,l]\) denotes of k-times continuously differentiable functions with the norm \(\Vert f\Vert _{C^k[0,l]}=\max \nolimits _{p=0,\dots ,k}\max \nolimits _{x\in [0,l]}|f^{(p)}(x)|\). \(H^0[0,l]=L^2[0,l]\) is the Hilbert space of measurable square integrable functions with scalar product \((f,g)_{L^2[0,l]}=\int \limits _0^lf(z)g(z)\,dz\). Let \({\mathfrak {L}}(L^2[0,l])\) be the Banach algebra of linear bounded operators defined on \(L^2[0,l]\).
For \(M\subset \mathbb R\) and a Banach space X by \(L^{\infty }(M,X)\), we denote the space of mappings \(f\,:\,M\rightarrow X\) with the norm \(\Vert f\Vert _{L^{\infty }}=ess\sup _{m\in M}\Vert f(m)\Vert _{X}\). For \(M=\mathbb Z_+{:=\mathbb N\cup \{0\}}\), we write \(L^{\infty }(\mathbb Z_+,X)=l^{\infty }(X)\). By \(B_r(x_0)\), we denote the open ball of radius \(r>0\) in X centered at \(x_0\).
\(C^1(U,\mathbb R^n)\), \(U\subset \mathbb R\) is the set of continuously differentiable mappings \(f:U\rightarrow \mathbb R^n\). \(C^1(\mathbb R^n)\) is the space of continuously differentiable functions \(f:\mathbb R^n\rightarrow \mathbb R^n\), and for \(f\in C^1(\mathbb R^n)\) by \(\partial _x f(x)\), we denote the corresponding Jacobi matrix.
For a linear bounded operator A acting on a Banach space, \(\sigma (A)\) denotes the spectrum of A and \(r_{\sigma }(A)\) denotes its spectral radius.
\(\mathbb R^{n\times m}\) is the space of \(n\times m\)-matrices, for \(m=n\) the set \(\mathbb R^{n\times n}\) is then a Banach algebra. We use the norm on \(\mathbb R^{n\times n}\) induced by the Euclidean norm in \(\mathbb R^n\): \(\Vert A\Vert =\sup _{\Vert x\Vert =1}\Vert Ax\Vert =\lambda _{\max }^{1/2}(A^{T}A)\).
We will use the following well-known inequalities. For any \(a,b\in \mathbb R_+\) and \(\varrho \in {\mathcal {K}}\) holds
for \(p_1,p_2\in (1,\infty )\) with \(\frac{1}{p_1}+\frac{1}{p_2}=1\), the Young’s inequality is
3 Problem statement and related stability notions
We consider dynamical systems with inputs defined similarly to [5, 21, 26] as follows
Definition 1
Let X be the state space with the norm \(\Vert \cdot \Vert _X\) and \({\mathcal {U}}_1\subset \{f\,:\mathbb R\rightarrow U_1\}\) be the space of input signals normed by \(\Vert \cdot \Vert _{{\mathcal {U}}_1}\) with values in a nonempty subset \(U_1\) of some linear normed space and invariant under the time shifts, that is, if \(d_1\in {\mathcal {U}}_1\) and \(\tau \in \mathbb R\), then \({\mathcal {S}}_{\tau }d_1\in {\mathcal {U}}_1\), where \({\mathcal {S}}_s\,:\,{\mathcal {U}}_1\rightarrow {\mathcal {U}}_1\), \(s\in \mathbb R\) is the linear operator defined by \({\mathcal {S}}_s u(t)=u(t+s)\).
The triple \(\varSigma _c=(X,{\mathcal {U}}_1,\phi _c)\) is called dynamical system with inputs if the mapping \(\phi _c\,\,:\,(t,t_0,x,d_1)\mapsto \phi _c(t,t_0,x,d_1)\) defined for all \((t,t_0,x,d_1)\in [t_0,t_0+\epsilon _{t_0,x,d_1})\times \mathbb R\times X\times {\mathcal {U}}_1\) for some positive \(\epsilon _{t_0,x,d_1}\) and satisfies the following axioms
(\(\varSigma _c1\)) for \(t_0\in \mathbb R\), \(x\in X\), \(d_1\in {\mathcal {U}}_1\), \(t\in [t_0,t_0+\epsilon _{t_0,x,d_1})\), the value of \(\phi _c(t,t_0,x,d_1)\) is well-defined and the mapping \(t\mapsto \phi _c(t,t_0,x,d_1)\) is continuous on \((t_0,t_0+\epsilon _{t_0,x,d_1})\) with \(\lim _{t\rightarrow t_0+}\phi _c(t,t_0,x,d_1)=x\);
(\(\varSigma _c2\)) \(\phi _c(t,t,x,d_1)=x\) for any \((x,d_1)\in X\times {\mathcal {U}}_1\), \(t\in \mathbb R\);
(\(\varSigma _c3\)) for any \(t_0\in \mathbb R\), \((t,x,d_1)\in [t_0,t_0+\epsilon _{t_0,x,d_1})\times X\times {\mathcal {U}}_1\) and \(\widetilde{d}_1\in {\mathcal {U}}_1\) with \(d_1(s)=\widetilde{d}_1(s)\) for \(s\in [t_0,t]\), and it holds that \(\phi _c(t,t_0,x,d_1)=\phi _c(t,t_0,x,\widetilde{d}_1)\);
(\(\varSigma _c4\)) for any \((x,d_1)\in X\times {\mathcal {U}}_1\) and \(t\ge \tau \ge t_0\) with \(\tau \in [t_0,t_0+\epsilon _{t_0,x,d_1})\), \(t\in [\tau ,\tau +\epsilon _{\tau ,{\phi _c}(\tau ,t_0,x,d_1),d_1})\cap [t_0,t_0+\epsilon _{t_0,x,d_1})\) it holds that
(\(\varSigma _c5\)) for any \((x,d_1)\in X\times {\mathcal {U}}_1\) and \(t\in [t_0,t_0+\epsilon _{t_0,x,d_1})\), it holds that
Note that \((\varSigma _c5)\) implies that for all \(t\in [\tau ,\tau +\epsilon _{\tau ,x_0,d_1})\), \(\tau \le t\)
Systems with impulsive actions are defined as follows:
Definition 2
Let \({\mathcal {E}}=\{\tau _k\}_{k=0}^{\infty }, \tau _k\in \mathbb R\) be a strictly increasing time sequence of impulsive actions with \(\lim \limits _{k\rightarrow \infty }\tau _k=\infty \). Let \({\mathcal {U}}_2\subset \{f\,:\,\mathbb Z_+\rightarrow U_2\}\) be the space of input signals normed by \(\Vert \cdot \Vert _{{\mathcal {U}}_2}\) and taking values in a nonempty subset \(U_2\) of some linear normed space. Let \(g\,:\,X\times U_2\rightarrow X\) be a mapping defining impulsive actions and the mapping \(\phi \) be defined for all \((t,t_0,x,d_1,d_2)\in \mathbb R\times \mathbb R\times X\times {\mathcal {U}}_1\times {\mathcal {U}}_2\), \(t\ge t_0\).
The following data \(\varSigma =(X,\varSigma _c,{\mathcal {U}}_2, g,\phi ,{\mathcal {E}})\) defines an impulsive system if
\((\varSigma _1)\) for all \((k,x,d_1)\in \mathbb Z_+\times X\times {\mathcal {U}}_1\) the system \(\varSigma _c\) satisfies
where we denote \(p(t_0):=\min \{k\in \mathbb Z_+\,:\tau _k\in {\mathcal {E}}_{t_0}\}\) with \({\mathcal {E}}_{t_0}=[t_0,\infty )\cap {\mathcal {E}}\); and
\((\varSigma _2)\) the mapping \(\phi \) satisfies
We will denote for short
The conditions \((\varSigma _c1)\) and \((\varSigma _2)\) imply
and \((\varSigma _c4)\), \((\varSigma _c5)\), \((\varSigma _2)\) imply that for \(t\ge \tau \ge t_0\), \((x,d_1,d_2)\in X\times {\mathcal {U}}_1\times {\mathcal {U}}_2\), the following holds:
The system \(\varSigma _c\) describes the continuous dynamics of the impulsive system \(\varSigma \). One can also consider its discrete dynamics separately as a system \(\varSigma _d\) defined next
Definition 3
A discrete dynamical system with input \(\varSigma _d=(X,g,\phi _d,{\mathcal {U}}_2)\) is given by a normed state space \((X,\Vert \cdot \Vert _X)\); a space of input signals \({\mathcal {U}}_2\subset \{f\,:\,\mathbb Z_+\rightarrow U_2\}\) with norm \(\Vert \cdot \Vert _{{\mathcal {U}}_2}\) and values in a nonempty subset \(U_2\) of a linear normed space; a mapping \(g\,:\,X\times U_2\rightarrow X\); and a mapping \(\phi _d\,\,:(k,l,x,d_2)\mapsto \phi _d(k,l,x,d_2)\), for \((k,l,x,d_2)\in \mathbb Z_+\times \mathbb Z_+\times X\times {\mathcal {U}}_2\), \(k\ge l\) such that
\((\varSigma _d1)\) \({\phi _d}(k,k,x,d_2)=x\), \(\phi _d(k+1,l,x,d_2)=g(\phi _d(k,l,x,d_2),d_2(k))\) for all \(k\ge l\).
Assumption
For any \(\tau \ge 0\), there exist \(\xi ,\,\xi _{\tau }\in {\mathcal {K}}_{\infty }\) and \(\chi _{\tau },\chi \in {\mathcal {K}}_{\infty }\) such that for all \((x,d_1,d_2)\in X\times {\mathcal {U}}_1\times {\mathcal {U}}_2\), it holds that
and
This assumption is not restrictive; for example, if g is continuous, then (6) is satisfied.
We are interested in the stability properties of the system \(\varSigma \) and its robustness with respect to the input signals \(d_1\) and \(d_2\). To this end, we use the notion of input-to-state stability (ISS). It was originally introduced in [25] for time invariant finite-dimensional systems. In our case, we adapt it as follows for \(\varSigma \)
Definition 4
For a fixed time sequence \({\mathcal {E}}\) of impulsive actions, the system \(\varSigma \) is called ISS if there exist \(\beta _{t_0}\in {\mathcal {K}}{\mathcal {L}}\), \(\gamma _{t_0}\in {\mathcal {K}}_{\infty }\), such that for any initial state \(x\in X\), any \(t\ge t_0\) and any \((d_1,d_2)\in {\mathcal {U}}_1\times {\mathcal {U}}_2\), it holds that
Next, we introduce a class of functions that we will use as Lyapunov functions to study the ISS property.
Definition 5
A function \(V:\,[t_0,\infty )\times X\rightarrow \mathbb R\) is said to be of class \({\mathcal {V}}({\mathcal {T}}_0)\), where \({\mathcal {T}}_0:=[t_0,\infty )\setminus {\mathcal {E}}\), if it satisfies the following properties:
-
(1)
V is continuous at any point \((t,x)\in {\mathcal {T}}_0\times X\), left continuous in t for \(t\in {\mathcal {E}}\), that is \(\lim \nolimits _{h\rightarrow 0-}V(t+h,x)=V(t,x)\) for all \((t,x)\in {\mathcal {E}}\times X\) and such that there exists \(\lim \nolimits _{h\rightarrow 0+}V(t+h,x):=V(t+0,x)\) for all \((t,x)\in {\mathcal {E}}\times X\);
-
(2)
The Lie derivative \({\dot{V}}(t,x,\zeta )\) exists for all \((t,x,\zeta )\in {\mathcal {T}}_0\times X\times U_1\) which is defined by
$$\begin{aligned} {\dot{V}}(t,x,\zeta ):=\lim \limits _{h\rightarrow 0+}\frac{1}{h}(V(t+h,\phi _c(t+h,t,x,\zeta ))-{V(t,x)}), \end{aligned}$$(8) -
(3)
For \((t,x,\zeta )\in {\mathcal {E}}\times X\times U_1\) the limits \({\dot{V}}(t\pm 0,x,\zeta )=\lim \nolimits _{h\rightarrow 0+}{\dot{V}}(t\pm h,x,\zeta )\) exist and \({\dot{V}}(t-0,x,\zeta )={\dot{V}}(t,x,\zeta )\)
4 Main results
In this section, we provide sufficient conditions to guarantee the ISS property for \(\varSigma \). Complementary stability conditions are given in Theorems 1 and 2, respectively. Application of these results are illustrated in Sect. 5.
Theorem 1
Assume that for \(\varSigma \) there are functions \(V_i\in {\mathcal {V}}({\mathcal {T}}_0)\), \(i\in \mathbb Z_+\) with
-
(1)
for some \(\alpha _1\), \(\alpha _2\in {\mathcal {K}}_{\infty }\), it holds that
$$\begin{aligned} \alpha _1(\Vert x\Vert )\le V_0(t,x)\le \alpha _2(\Vert x\Vert ),\quad \text {for all}\quad (t,x)\in [t_0,\infty )\times X; \end{aligned}$$(9) -
(2)
there is a sequence \(\eta _p\in {\mathcal {K}}_{\infty }\), \(p\in \mathbb Z_+\) such that for all \((t,x,\zeta )\in {\mathcal {T}}_0\times X\times U_1\) holds
$$\begin{aligned} {\dot{V}}_0(t,x,\zeta )\le & {} V_1(t,x)+\eta _0(\Vert \zeta \Vert _{U_1}),\nonumber \\ -{\dot{V}}_p(t,x,\zeta )\le & {} V_{p+1}(t,x)+\eta _p(\Vert \zeta \Vert _{U_1}),\quad p\in \mathbb N; \end{aligned}$$(10) -
(3)
there are \(\eta \in {\mathcal {K}}_{\infty }\) and \(W_k\,:X\rightarrow \mathbb R\), \(k\in \mathbb Z_+\) so that \(\forall \) \((k,x,\zeta )\in \mathbb Z_+\times X\times U_2\)
$$\begin{aligned} V_0(\tau _k+0,g(x,\zeta ))-V_0(\tau _k,x)\le W_k(x)+\eta (\Vert \zeta \Vert _{U_2}); \end{aligned}$$(11)holds;
-
(4)
there exists \(\delta \in {\mathcal {K}}_{\infty }\) such that for all \((k,x)\in \mathbb Z_+\times X\), it holds that
$$\begin{aligned} G_{k+1}(x):=W_{k+1}(x)+\sum \limits _{p=1}^{\infty }V_p(\tau _{k+1},x)\frac{(\tau _{k+1}-\tau _k)^p}{p!}\le -\delta (\Vert x\Vert ); \end{aligned}$$(12) -
(5)
for any \(\rho >0\) exists \(q_{\rho }\in [0,1)\) such that \(\lim \nolimits _{p\rightarrow \infty }\frac{\eta _p(s)}{(p+1)\eta _{p-1}(s)}=q_{\rho }\) uniformly for \(s\in [0,\rho ]\) and for any \(k\in \mathbb Z_+\), there is \(\omega _k\in {\mathcal {K}}_{\infty }\) such that \(|V_p(s,x)|\le \omega _k(\Vert x\Vert )\) for all \((p,s)\in \mathbb Z_+\times (\tau _k,\tau _{k+1}]\).
Then, \(\varSigma \) satisfies the ISS property.
Remark 1
The inequality (12) in condition (4) of the theorem restricts the time intervals between jumps, which is a dwell-time condition. If both continuous and discrete dynamics are stable, the existence of \(W_k\) and \(V_p\) so that \(W_{k+1}(x)\) and \(\sum \nolimits _{p=1}^{\infty }V_p(\tau _{k+1},x)\frac{(\tau _{k+1}-\tau _k)^p}{p!}\) are negative is guaranteed, so that (12) implies no restrictions on \({\mathcal {E}}\). However if one of the dynamics is unstable, the dwell-time is restricted. For example, if the discrete dynamics is stable but the continuous one is not, we have \(W_{k+1}(x)<0\), but the second summand in (12) can be positive, and hence it needs to be small enough in order to satisfy (12). This implies that the time distances \(\tau _{k+1}-\tau _k\) need to be small enough. Moreover, (12) allows the situation, where both dynamics types are not stable. This will be illustrated in our examples.
Let us fix any \((t_0,x_0)\in \mathbb R\times X\) as the initial data and assume without loss of generality that \(t_0\le \tau _0\). For any disturbances \((d_1,d_2)\in {\mathcal {U}}_1\times {\mathcal {U}}_2\), the corresponding solution to \(\varSigma \) will be denoted by \(x(t)=\phi (t,t_0,x_0,d_1,d_2)\).
Lemma 1
Under the conditions of Theorem 1 for any \(k\in \mathbb Z_+\), \(n\ge 2\) the following inequality holds true:
The proof can be found in Appendix.
Corollary 1
Under the conditions of Theorem 1, the following estimate is true:
Proof
The estimate follows after taking the limit for \(n\rightarrow \infty \) in the inequality (13), which is possible under the conditions of Theorem 1. \(\square \)
Proof
(of Theorem 1) The combination of (11) and (14) implies that
Let \({\widehat{\eta }}(s):=\sum \nolimits _{p=1}^{\infty }\frac{\eta _{p-1}(s)\theta ^p}{p!}+\eta (s)\), then \({\widehat{\eta }}\in {\mathcal {K}}_{\infty }\). Recall that \(d=\Vert d_1\Vert _{{\mathcal {U}}_1}+\Vert d_2\Vert _{{\mathcal {U}}_2}\), then for all \(k\in \mathbb Z_+\), we have
From this inequality together with condition (4) of Theorem 1, it follows that for all \(k\in \mathbb Z_+\)
Let \(\epsilon \in (0,1)\), \(r=\delta ^{-1}(\frac{{\widehat{\eta }}(d)}{1-\epsilon })\). Let us show by contradiction that for some \(k^*\in \mathbb Z_+\), it holds that \(x(\tau _{k^*})\in B_r(0)\). Assume that this is not true, that is for all \(k\in \mathbb Z_+\), we have \(\Vert x(\tau _k)\Vert \ge r\), then from (15) follows:
This means that the bounded from below sequence \(\{V_0(\tau _k+0,x(\tau _k+0))\}_{k\in \mathbb Z_+}\) is decreasing, hence there exists the limit \( m:=\lim \nolimits _{k\rightarrow \infty }V_0(\tau _k+0,x(\tau _k+0))\). From (15) follows then \(\lim \sup _{k\rightarrow \infty }\delta (\Vert x(\tau _{k+1})\Vert )\le {\widehat{\eta }}(d)\). This implies that
which leads to a contradiction. Hence, for some \(k^*\), it holds that \(\Vert x(\tau _{k^*})\Vert <r\). Let \(R:=\max \{(\alpha _1^{-1}\circ \alpha _2)(\xi (r)+\chi (d)),r\}\). We show that for \(k\ge k^*\), the inequality \(\Vert x(\tau _k+0)\Vert \le R\) is true. Indeed, if for some \(m\ge k^*\)
where \(1\le j(m)\le \infty \), then from (15) follows that
hence, by condition (1) of Theorem 1, we obtain
Taking (6) into account, we obtain \(\Vert x(\tau _m+0)\Vert \le \xi (\Vert x(\tau _m)\Vert )+\chi (d)\). Hence,
Let \(S_r:=\{k\in \mathbb Z\,\,:\,\,\forall l,\,\,0\le l\le k\,\,\Vert x(\tau _l)\Vert \ge r\}\) and
We need to consider the case \(N\ge 1\). Let \(k\in \mathbb N \) be such that \(1\le k\le N\), then \({\widehat{\eta }}(d)\le (1-\epsilon )\delta (\Vert x(\tau _k)\Vert )\) and from (15), it follows that
The inequality (6) implies that
It is easily seen that \(\varphi \in {\mathcal {K}}_{\infty }\), hence \(\Vert x(\tau _k)\Vert \ge \varphi ^{-1}(\Vert x(\tau _{k}+0)\Vert )\) and from (16) follows
We denote \(\delta _1:=\delta \circ \varphi ^{-1}\circ \alpha _2^{-1}\in {\mathcal {K}}_{\infty }\), \(v_k:=V_0(\tau _{k}+0,x(\tau _{k}+0))\), and conclude that for all k, \(1\le k\le N\), the inequality (17) can be written as
Let us define the sequence \({\widehat{v}}_k\) for \(k\in \mathbb N\) by \({\widehat{v}}_k:=v_k\) for \(1\le k\le N\) and so that for \(k\ge N+1\) the \({\widehat{v}}_k\) satisfies the difference equation \( {\widehat{v}}_k-{\widehat{v}}_{k-1}= -\epsilon \delta _1({\widehat{v}}_k). \)
Hence for all \(k\in \mathbb N\), the sequence \(\{{\widehat{v}}_k\}_{k\in \mathbb N}\) satisfies the inequality
Together with the inequality (19), we consider the comparison equation
First, let us show that for all \(k\in \mathbb Z_+\), the inequality \({\widehat{v}}_k\le w_k\) is true. Indeed, if for some \(k_1\in \mathbb N\) \({\widehat{v}}_{k_1-1}\le w_{k_1-1}\) and \({\widehat{v}}_{k_1}> w_{k_1}\), then from \({\text {id}}\,+\epsilon {\delta }_1\in {\mathcal {K}}_{\infty }\), it follows that
which is a contradiction. From (20) follows \(w_{k}=({\text {id}}\,+\epsilon \delta _1)^{-1}(w_{k-1})\), \(k\in \mathbb N\). By the properties of comparison functions, there exists \(\delta _2\in {\mathcal {K}}_{\infty }\) such that \(({\text {id}}\,+\epsilon \delta _1)^{-1}={\text {id}}\,-\delta _2\), hence (20) can be written as
Let \({\widehat{\delta }}_2(s):=\min \{s,\delta _2(s)\}\), then
We define \(\varDelta _{v_0}(s):=\int \limits _{s}^{v_0}\frac{d\tau }{\widehat{\delta _2}(\tau )}\), then \(\varDelta _{v_0}(s)\rightarrow +\infty \) for \(s\rightarrow 0+\). By the mean value theorem for some \(s^*\in (w_k,w_{k-1})\) from (22), we obtain
Hence, \(\varDelta _{v_0}(w_k)\ge \varDelta _{v_0}(v_0)+k=k\), which implies the estimate
This means that for all k, \(0\le k\le N\), we can estimate \( \Vert x(\tau _k+0)\Vert \le (\alpha _1^{-1}\circ \varDelta ^{-1}_{v_0})(k). \) From (5), we have that for all \(k\in \mathbb Z_+\) follows
That is for all \(t\in (\tau _k,\tau _{k+1}]\), \(0\le k\le N-1\), the following inequality holds:
Let \(\vartheta _k:=(\xi _2\circ \alpha _1^{-1}\circ \varDelta ^{-1}_{v_0})(k)\), and we define the function
It is easily seen that \(\widetilde{\beta }\in {\mathcal {K}}{\mathcal {L}}\). From (23), we get for all \(t\in (\tau _0,\tau _N]\), the inequality
Applying the estimates
and the inequality (1), we obtain that for some functions \(\widetilde{\xi }_{\tau _0-t_0}\), \(\widetilde{\chi }_{\tau _0-t_0}\in {\mathcal {K}}_{\infty }\)
Applying again (1), we derive
From (24) using again (1), we get that for all \(t\in (\tau _0,\tau _N]\), the inequality
holds, where we have denoted
It easy to check by definition that \({\widehat{\beta }}_{t_0}\in {\mathcal {K}}{\mathcal {L}}\), \({\widehat{\chi }}_{t_0}\in {\mathcal {K}}_{\infty }\). There exists a function \(\beta _{t_0}\in {\mathcal {K}}{\mathcal {L}}\) such that \(\beta _{t_0}(s,t)\ge {\widehat{\beta }}_{t_0}(s,t)\) for \(t\in (\tau _0,\tau _N]\) and \(\beta _{t_0}(s,t)\ge \widetilde{\xi }_{\tau _0-t_0}(s)\) for \(t\in [t_0,\tau _0]\). Hence, from (25), it follows that for all \(t\in [t_0,\tau _N]\)
Recall that for \(k\ge N\), we have \(\Vert x(\tau _k+0)\Vert \le R\). Hence from the estimate \(\Vert x(t)\Vert \le \xi _{\theta }(\Vert x(\tau _k+0)\Vert )+\chi _{\theta }(d)\), \(t\in (\tau _k,\tau _{k+1}]\) it follows that
where \({\hat{\chi }}\in {\mathcal {K}}_{\infty }\) by definition. Combining the inequalities (26) and (27), we see that for some \(\gamma _{t_0}\in {\mathcal {K}}_{\infty }\), the following estimate holds
which proves the theorem. \(\square \)
Remark 2
Condition (5) of Theorem 1 assures the convergence of three last terms in (14) for \(n\rightarrow \infty \). However, if \(V_{n}(t,x)\le 0\) for some \(n\ge 1\), then we can set \(V_p(t,x)\equiv 0\) for all \(p>n\) and this condition (5) can be dropped. Stability investigation in this case is essentially easier because we deal with a finite number of auxiliary functions instead of an infinite sequence. The class of systems, where this simplification is possible becomes wider due to the next result (see the difference in the sign before \({\dot{V}}_p\) in conditions (2) of the previous and the next theorem). Such simplification will be used in some of our examples later.
Theorem 2
Assume that for the system \(\varSigma \), there are \(V_i\in {\mathcal {V}}({\mathcal {T}}_0)\) such that
-
(1)
for some \(\alpha _1\), \(\alpha _2\in {\mathcal {K}}_{\infty }\) it holds that
$$\begin{aligned} \alpha _1(\Vert x\Vert )\le V_0(t,x)\le \alpha _2(\Vert x\Vert ),\quad \text {for all}\quad (t,x)\in [t_0,+\infty )\times X, \end{aligned}$$(28) -
(2)
there is a sequence \(\eta _p\in {\mathcal {K}}_{\infty }\), \(p\in \mathbb Z_+\) such that \(\forall \,(t,x,\zeta )\in {\mathcal {T}}_0\times X\times U_1\)
$$\begin{aligned} {\dot{V}}_p(t,x,\zeta )\le V_{p+1}(t,x)+\eta _p(\Vert \zeta \Vert _{U_1}),\quad p\in \mathbb Z_+, \end{aligned}$$(29) -
(3)
there are \(\eta \in {\mathcal {K}}_{\infty }\) and \(W_k\,:X\rightarrow \mathbb R\), \(k\in \mathbb Z_+\) so that \(\forall \,(k,x,\zeta )\in \mathbb Z_+\times X\times U_2\)
$$\begin{aligned} V_0(\tau _k+0,g(x,\zeta ))-V_0(\tau _k,x)\le W_k(x)+\eta (\Vert \zeta \Vert _{U_2}), \end{aligned}$$(30) -
(4)
there are \(Q_{kp}\,:\,X\rightarrow \mathbb R\) and \(\pi _p\in {\mathcal {K}}_{\infty }\), \((k,p)\in \mathbb Z_+\times \mathbb N\), \((x,\zeta )\in X\times U_2\) with
$$\begin{aligned} V_p(\tau _k+0,g(x,\zeta ))-V_p(\tau _k,x)\le Q_{kp}(x)+\pi _p(\Vert \zeta \Vert _{U_2}), \end{aligned}$$(31) -
(5)
there is \(\delta \in {\mathcal {K}}_{\infty }\) such that for all \((k,x)\in \mathbb Z_+\times X\), the next inequality holds
$$\begin{aligned} G_{k}(x):=W_{k}(x)+\sum \limits _{p=1}^{\infty }(V_p(\tau _{k},x)+Q_{kp}(x))\frac{(\tau _{k+1}-\tau _k)^p}{p!}\le -\delta (\Vert x\Vert ), \end{aligned}$$(32) -
(6)
For any \(\rho >0\), there exists \(q_{\rho }\in [0,1)\) such that \( \lim \nolimits _{p\rightarrow \infty }\frac{\eta _p(s)}{(p+1)\eta _{p-1}(s)}\le q_{\rho } \) and \( \lim \nolimits _{p\rightarrow \infty }\frac{\pi _p(s)}{(p+1)\pi _{p-1}(s)}\le q_{\rho } \) exist uniformly for \(s\in [0,\rho ]\), and for each \(k\in \mathbb Z_+\) exists \(\omega _k\in {\mathcal {K}}_{\infty }\) such that
$$\begin{aligned} |V_p(s,x)|\le \omega _k(\Vert x\Vert ),\quad |Q_{kp}(x)|\le \omega _k(\Vert x\Vert ) \end{aligned}$$for all \((p,s,x)\in \mathbb Z_+\times (\tau _k,\tau _{k+1}]\times X\). Then, system \(\varSigma \) is ISS.
Lemma 2
Under the conditions of Theorem 2, we have for all \(k\in \mathbb Z_+\), \(n\ge 2\)
Corollary 2
Under the conditions of Theorem 2, the following holds:
Proof
Condition (6) of Theorem 2 assures the possibility to take the limit for \(n\rightarrow \infty \) in (33) which implies the assertion. \(\square \)
Proof
(of Theorem 2) From (30) and (34) follows
Taking (31) into account, we obtain
Let \( {\widehat{\eta }}(s):=\eta (s)+\sum \nolimits _{p=1}^{\infty }\frac{(\pi _p(s)+\eta _{p-1}(s))\theta ^p}{p!}. \) Obviously \({\widehat{\eta }}\in {\mathcal {K}}_{\infty }\) and from (32) and (36) follows (recall that \(d=\Vert d_1\Vert _{{\mathcal {U}}_1}+\Vert d_{2}\Vert _{{\mathcal {U}}_2}\))
Let \(\epsilon \in (0,1)\), \(r=\delta ^{-1}(\frac{{\widehat{\eta }}(d)}{1-\epsilon })\). First we show by contradiction that there exists \(k^*\in \mathbb Z_+\) such that \(\Vert x(\tau _{k^*})\Vert <r\). Indeed, otherwise, for all \(k\in \mathbb Z_+\) \(\Vert x(\tau _k)\Vert \ge r\) and hence from (37), it follows that
This means that the sequence \(\{V_0(\tau _{k},x(\tau _{k}))\}_{k\in \mathbb Z_+}\) is strictly decreasing and is bounded from below, hence it possesses a nonnegative limit \(m^*=\lim _{k\rightarrow \infty }V_0(\tau _{k},x(\tau _{k}))\). From (37) follows
which leads to a contradiction.
We denote \(R:=\max \{(\alpha _1^{-1}\circ \alpha _2)(\xi _{\theta }(\xi (r)+\chi (d))+\chi _{\theta }(d)),r\}\). Let us show that for \(k\ge k^*\) the inequality \(\Vert x(\tau _k)\Vert \le R\) is true. Indeed, if for some \(m\ge k^*\)
where \(1\le j(m)\le \infty \), then from (37), we have that for \(i\ge 2\)
and hence from (5) and by the condition (1) of Theorem 2, it follows that
Taking (6) into account, we obtain
Similarly, for \(i=1\) we get
Let \(S_r:=\{k\in \mathbb Z_+\,\,:\,\,\forall l,\,\,0\le l\le k\,\,\Vert x(\tau _l)\Vert \ge r\}\) and
It is enough to consider the case \(N\ge 1\). Let k be such that \(0\le k\le N\), then \({\widehat{\eta }}(d)\le (1-\epsilon )\delta (\Vert x(\tau _k)\Vert )\), and from (37), the next inequality follows:
By condition (1) of Theorem 2, we obtain \(\Vert x(\tau _k)\Vert \ge \alpha _2^{-1}(V_0(\tau _{k},x(\tau _{k})))\) and hence (38) can be written as
where \(\delta _1=\delta \circ \alpha _2^{-1}\in {\mathcal {K}}_{\infty }\). Let us denote \(v_k:=V_0(\tau _{k},x(\tau _{k}))\), then for all \(k\in \mathbb Z_+\), we have the inequality \( v_{k+1}-v_k\le -\epsilon \delta _1(v_k). \) Let \({\widehat{\delta }}_1(s):=\min \{s,\delta _1(s)\}\in {\mathcal {K}}_{\infty }\), then
We define the \(\{{\widehat{v}}_k\}_{k\in \mathbb Z_+}\) sequence recurrently setting \({\widehat{v}}_k:=v_k\) for \(0\le k\le N\) and
for \(k\ge N\). From this definition follows that for all \(k\in \mathbb Z_+\)
and \(0\le {\widehat{v}}_k\le v_0\).
Let \(\varDelta _{v_0}(s):=\int \limits _s^{v_0}\frac{d\tau }{{\widehat{\delta }}_1(\tau )}\). By the mean value theorem \(\exists \,s^*\in ({\widehat{v}}_{k+1},{\widehat{v}}_{k})\) with
Hence, \(\varDelta _{v_0}({\widehat{v}}_k)\ge k\epsilon \) for all \(k\in \mathbb Z_+\). From its definition, it follows that \(\varDelta _{v_0}(s)\) is strictly growing with respect to \(v_0\) for a fixed s and strictly decreasing in s for a fixed \(v_0\) with \(\varDelta _{v_0}(s)\rightarrow \infty \) for \(s\rightarrow 0+\). This implies that
The desired ISS estimate follows similarly as in the proof of Theorem 1, which proves Theorem 2. \(\square \)
Remark 3
If for some \(n\ge 1\) we have \(V_{n}(t,x)\le 0\), then we can take \(V_p(t,x)\equiv 0\) for all \(p>n\), and in this case, condition (5) of Theorem 2 can be dropped.
5 Examples
5.1 Heat equation with variable coefficients
Let X be the normed linear vector space defined by
with \(l>0\), \(0<2\delta <l\) and the norm \(\Vert \cdot \Vert _X:=\Vert \cdot \Vert _{L^2[0,l]}\).
Consider the following linear system with impulsive actions
with the following initial and boundary conditions
where \((z,t)\in [0,l]\times [0,\infty )\), \(u\,\,:\,\,[0,l]\times [0,\infty )\rightarrow \mathbb R\). We assume that \(a>0\), \(b\,:[0,l]\rightarrow \mathbb R\) is piecewise continuous bounded function which can be discontinuous only at \(z=2\delta \), and \(c\,:[0,l]\rightarrow \mathbb R\), \(\,c\in C^2[0,l]\).
We are interested in the properties of classical solutions to (40)–(41) defined next. Let us denote \({\mathcal {T}}_T=[0,T]\cap {\mathcal {T}}_0\), \({\mathcal {E}}_T=[0,T]\cap {\mathcal {E}}\) for \(T>0\).
Definition 6
A function \(u\,:[0,l]\times [0,T]\rightarrow \mathbb R\) is called classical solution to (40)-(41), if
For \(z=2\delta \), it holds that
and u satisfies the initial and boundary conditions (41).
Since \(c\in C^2[0,l]\), it follows that the state space X is invariant under jumps, that is, \(f\in X\) implies \((1+c)f\in X\). It is also clear that the jump operator is consistent with the boundary conditions. Theorem 1 from [15] implies the existence and uniqueness of solutions to (40)—(41) so that \(u(\cdot ,t)\in X\) for \(t\ge 0\).
We consider the question of asymptotic stability of solutions to (40)–(41) with respect to the norm in \(L^2[0,l]\). The next proposition verifies the estimates (5)–(6).
Proposition 1
Solutions of (40)—(41) satisfy the estimates (5)–(6) with the following choice of functions
where \(b_{\max }=\sup \nolimits _{z\in [0,l]}|b(z)|\), \(c_{\max }=\sup \nolimits _{z\in [0,l]}|c(z)|\).
The proof can be found in Appendix.
To study stability properties, we introduce the following Lyapunov function:
for which we have
Let us calculate the full time derivative of \(V_1(u(\cdot ,t))\) along solutions to (40)–(41):
Hence, we can apply Theorem 1 taking \(V_2=0\). From (40) for \(t\in {\mathcal {E}}\), we obtain
With help of Theorem 1, we arrive to the following conditions for the global asymptotic stability
for some \(\varepsilon _0>0\), where \(T_k=\tau _{k+1}-\tau _k\) denotes the dwell-time between two impulsive actions. By means of the Friedrich’s inequality, we arrive to the following condition guaranteeing the GAS property of the system (40)–(41)
For illustration, we apply this condition to the following particular case of (40)-(41): \(a=1\), \(l=\pi \) and
where \(\varepsilon >0\), \(\delta <\pi /2\), \(c\in C^2\), \(-2\varepsilon \le c(z)\le \varepsilon \) . We will see that for some choice of parameters \(\varepsilon ,\delta \) both continuous and discrete dynamics have unstable behavior.
In this case, the dwell-time condition (45) reduced to the following two inequalities:
or equivalently to the following explicit condition applied on the dwell-times
Let us consider continuous and discrete dynamics of (40)–(41) separately.
First, we consider the stability properties of the differential equation
with the following boundary and initial conditions
The corresponding self-adjoint spectral problem is
We can show that at least for \(\varepsilon \) small enough there exists some critical \(\delta ^*\), (\(\delta ^*\approx 0.651331\)) such that for all \(\delta <\delta ^*\) the linear system (47)–(48) is not stable, and for \(\delta >\delta ^*\), this system is asymptotically stable.
This follows immediately from the next proposition proved in Appendix:
Proposition 2
The largest eigenvalue \(\lambda _{\max }(\varepsilon )\) of the spectral equation (49) can be represented asymptotically as
Indeed, the function \(f(\delta )=\frac{3}{4}\sin (4\delta )+\frac{\pi }{2}-3\delta \) is decreasing on \([0,\pi /2]\), and we have \(f(0)f(\pi /2)<0\), hence the equation \(f(\delta )=0\) has a unique solution in \([0,\pi /2]\), which we denote by \(\delta ^*\). Then for \(\delta <\delta ^*\) for sufficiently small \(\varepsilon \), we have \(\lambda _{\max }(\varepsilon )>0\), and for \(\delta >\delta ^*\), we have \(\lambda _{\max }(\varepsilon )<0\).
This mens that for sufficiently small \(\varepsilon \), the parameter \(\delta \) can be chosen so that the partial differential equation (47)–(48), which describes the continuous dynamics of the original impulsive system (40)–(41), is unstable.
We consider the difference equation in the state space X, that describes the discrete dynamics of the original impulsive system (40)—(41), and demonstrate its unstable behavior. Let \(C\psi :=(1+c(z))\psi \) for \(\psi \in X\), then \(C\in {\mathfrak {L}}(X)\) and \( \Vert C^n\Vert =\sup \nolimits _{\Vert \psi \Vert _{L^2[0,\pi ]}=1}\Vert (1+c(z))^n\psi \Vert _{L^2[0,l]} \ge \sup \nolimits _{\Vert \psi \Vert _{L^2[0,\pi ]}=1,{\text {supp}}\,\psi \subset [0,\delta ]}\Vert (1+\varepsilon )^n\psi \Vert _{L^2[0,l]}=(1+\varepsilon )^n, \) which implies that \(r_{\sigma }(C)\ge 1+\varepsilon >1\), showing the instability property.
Remark 4
Let us note that in case of an unstable scalar ODE of the first order subjected to destabilizing impulsive actions, the overall dynamics of the whole impulsive system is always unstable. In contrary to this, our example shows that for the considered unstable PDE with destabilizing impulses the overall dynamics is stable under certain condition on dwell times.
5.2 Nonlinear impulsive ODE
Consider the following nonlinear ODE system with impulsive actions
where \(x\in \mathbb R^n\), \(f,g\,:\mathbb R^n\rightarrow \mathbb R^n\) are homogeneous functions of the order \(m>1\) with odd m, that is, \(f(\lambda x)=\lambda ^m f(x)\), \(g(\lambda x)=\lambda ^m g(x)\) for all \((\lambda ,x)\in \mathbb R^{n+1}\), \(f\in C^1(\mathbb R^n)\). Let \(U_1=U_2=\mathbb R^n\), \({\mathcal {U}}_1=L^{\infty }(\mathbb R_+,\mathbb R^n)\), \({\mathcal {U}}_2=L^{\infty }(\mathbb Z_+,\mathbb R^n)=l^{\infty }(\mathbb R^n)\).
Let there exist a positive definite matrix P such that for all \(x\in \mathbb R^n\) such that the function f, g satisfies the inequalities
for some positive constants \(b_0\), \(c_0\), \(\nu _0\) and \(\nu _1\). For short, we denote
Proposition 3
Solutions of the system (50) satisfy the estimates (5)—(6) with functions
where \(\omega _i\), \(i=1,2\) are certain positive functions (provided explicitly in the proof).
The proof can be found in Appendix.
Stability properties will be studied with help of the following Lyapunov function \(V_0(x)=x^{{\text {T}}\,}Px\), \(P\succ 0\). Its time derivative along solutions to (50) for \(t\ne k\theta \) is
By means of the Young’s inequality (2), we have
for any \(\tau >0\). Hence from (52), we obtain
We chose
On the jumps due to impulsive actions from the definition of \(V_0\), we calculate
Again by the Young’s inequality (2), we obtain
From (51), it follows that \(|g^{{\text {T}}\,}(x)Pg(x)|\le \Vert P\Vert b_0^2\Vert x\Vert ^{2m}\). Hence from (55), the following estimate follows:
where
Now, consider \({\dot{V}}_1(x,d_1)\) for \(t\ne k\theta \)
By means of the Cauchy inequality and from (51), it follows that
Hence,
Now, we apply the Young’s inequality with the parameter \(\tau >0\)
and obtain
Denoting
we can write
For sufficiently small \(\tau >0\), we have \(\vartheta _1(\tau )>0\), \(\vartheta _2(\tau )>0\). Since \(V_2(x)\le 0\), we have
and we can set \(V_3(x)=0\).
In order to apply Theorem 1 (see Remark 2), we need to estimate the function
From the estimates (54) and (56), we obtain
where \(\sigma _3(\tau )=2c_0-2\Vert P\Vert (\theta +1)\frac{\tau ^{m+1}}{m+1}\), \(\sigma _4(\tau )=\Vert P\Vert (b_0^2+b_0\tau )\).
Proposition 4
If
then for \(\tau >0\) small enough and for all \(s\in \mathbb R_+\), the next inequality is true
The proof can be found in Appendix.
Proposition 5
Let system (50) satisfy the conditions of (51) as well as the inequality
Then, system (50) is ISS.
Proof
Number \(\tau >0\) can be chosen small enough, so that from the inequality
the following estimation \(G(x)\le -\delta (\Vert x\Vert )\) c \(\delta (s)=\varepsilon _1s^{m+1}/\sqrt{1+s^{2m}}\), follows where \(\varepsilon _1>0\) is small enough. Hence, system (50) satisfies all conditions of Theorem 1 (see Remark 2) from which desired assertion follows. \(\square \)
To illustrate this result, we take
\(P=I\), \(\theta =1\), then \(b_0^2=1.003329653\), \(c_0=0.5185185185\), \(\nu _0=0.2511999583\), \(\nu _1=0.08780118723\).
It is easy to see that conditions of the last proposition are satisfied. Hence, the nonlinear impulsive system (50) is ISS. Note that in this case, both continuous and discrete dynamics are unstable.
6 Discussion and conclusions
The main results of this paper are given by Theorems 1 and 2. They allow us to study the ISS property of nonlinear impulsive systems with different assumptions imposed on the discrete and continuous dynamics. These theorems can establish the ISS property even for the case when neither discrete nor continuous dynamics are ISS. Our approach enables usage of a wider class of ISS-Lyapunov functions to study the ISS property of nonlinear impulsive systems. One advantage of our approach, demonstrated in the examples, is that a rather simple (energetic) Lyapunov function equipped with a sequence of auxiliary functions allows to derive desired stability condition to assure the ISS or GAS property.
An interesting direction for future research would be to develop an approach of stability investigation of nonlinear impulsive systems by means of a combination of a Lyapunov and Chetaev functions, which was used for studying local stability of finite-dimensional systems without inputs [11]. This can provide stability conditions alternative to the ones developed in the current paper. First steps in this direction can be found in [7].
Another interesting direction of research is to explore an extension possibility of our results to the strong ISS notion in the context of time varying impulsive systems (see [20]).
References
Bivzyuk VO, Slynko VI (2019) Sufficient conditions for the stability of linear differential equations with periodic impulse action. Mat Sb 210(11):3–23
Briat C (2013) Convex conditions for robust stability analysis and stabilization of linear aperiodic impulsive and sampled-data systems under dwell-time constraints. Automatica J IFAC 49(11):3449–3457
Briat C, Seuret A (2012) A looped-functional approach for robust stability analysis of linear impulsive systems. Syst Control Lett 61(10):980–988
Dashkovskiy S, Feketa P (2017) Input-to-state stability of impulsive systems and their networks. Nonlinear Anal Hybrid Syst 26:190–200
Dashkovskiy S, Mironchenko A (2013) Input-to-state stability of infinite-dimensional control systems. Math Control Signals Syst 25(1):1–35
Dashkovskiy S, Mironchenko A (2013) Input-to-state stability of nonlinear impulsive systems. SIAM J Control Optim 51(3):1962–1987
Dashkovskiy S, Slynko V (2021) Dwell-time stability conditions for infinite dimensional impulsive systems, arXiv:2106.11224
Dvirnyi AI, Slynko VI (2004) Criteria for the stability of quasilinear impulse systems. Int Appl Mech 40(5):137–144
Dvirnyi AI, Slyn’ko VI (2014) Application of Lyapunovs’ direct method to the study of the stability of solutions to systems of impulsive differential equations. Math Notes 96(1–2):26–37
Dvirnyi AI, Slyn’ko VI (2014) Application of Lyapunov’s direct method to the study of the stability of solutions to systems of impulsive differential equations. Transl Mat Zametki 96(1):22–35
Dvirnyĭ AI, Slynko VI (2011) On the stability of solutions of nonlinear nonstationary systems of differential equations with impulse action in a critical case. Nelīnīĭnī Koliv. 14(4):445–467
Goebel R, Sanfelice RG, Teel AR (2012) Hybrid dynamical systems. Princeton University Press, Princeton
Hespanha JP, Liberzon D, Teel AR (2008) Lyapunov conditions for input-to-state stability of impulsive systems. Automatica J IFAC 44(11):2735–2744
Jiang G, Lu Q, Qian L (2007) Chaos and its control in an impulsive differential system. Chaos Solitons Fractals 34(4):1135–1147
Kamynin LI (1963) The method of heat potentials for a parabolic equation with discontinuous coefficients. Sibirsk Mat Ž 4:1071–1105
Karafyllis I, Jiang Z-P (2011) Stability and stabilization of nonlinear systems. Communications and Control Engineering Series, Springer, London
Lakshmikantham V, Bainov DD, Simeonov PS (1989) Theory of impulsive differential equations, vol 6. Series in Modern Applied Mathematics. World Scientific Publishing Co., Inc, Teaneck
Li X, Li P, Wang Q-G (2018) Input/output-to-state stability of impulsive switched systems. Syst Control Lett 116:1–7
Liu X, Willms A (1995) Stability analysis and applications to large scale impulsive systems: a new approach. Canad Appl Math Quart 3(4):419–444
Mancilla-Aguilar JL, Haimovich H, Feketa P (2020) Uniform stability of nonlinear time-varying impulsive systems with eventually uniformly bounded impulse frequency. Nonlinear Anal Hybrid Syst 38(100933):16
Mironchenko A, Karafyllis I, Krstic M (2019) Monotonicity methods for input-to-state stability of nonlinear parabolic PDEs with boundary disturbances. SIAM J Control Optim 57(1):510–532
Mironchenko A, Yang G, Liberzon D (2018) Lyapunov small-gain theorems for networks of not necessarily ISS hybrid systems. Automatica J IFAC 88:10–20
Samoĭlenko AM, Perestyuk NA (1995) Impulsive differential equations, volume 14 of World Scientific Series on Nonlinear Science. Series A: Monographs and Treatises. World Scientific Publishing Co., Inc., River Edge, NJ, 1995. With a preface by Yu. A. Mitropolskiĭ and a supplement by S. I. Trofimchuk, Translated from the Russian by Y. Chapovsky
Slyn’ko VI, Tunç O, Bivziuk VO (2019) Application of commutator calculus to the study of linear impulsive systems. Syst Control Lett 123:160–165
Sontag ED (1989) Smooth stabilization implies coprime factorization. IEEE Trans Automat Control 34(4):435–443
Sontag ED (2013) Mathematical control theory: deterministic finite dimensional systems, vol 6. Springer, Berlin
van der Schaft A, Schumacher H (2000) An introduction to hybrid dynamical systems, vol 251. Lecture Notes in Control and Information Sciences, Springer, London
Ye H, Michel AN, Hou L (1998) Stability analysis of systems with impulse effects. IEEE Trans Automat Control 43(12):1719–1723
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research was supported by the German Research Foundation (DFG), grant No. DA 767/12-1.
Appendix
Appendix
1.1 Proof of Lemma 1
The proof is by induction as follows. First, we check that the statements is true for \(n=2\) from the obvious equality
and the first inequality of (10) for \(s\in (\tau _k,\tau _{k+1}]\):
it follows that
From the next two obvious relations
we obtain the desired inequality
Now, let the statement be true for \(n=q\), that is
Similarly to the previous step, we write
and from the inequalities (10), we obtain
We substitute this estimate into (59) and obtain the desired inequality
which proves the Lemma.
1.2 Proof of Lemma 2
The proof is again by the mathematical induction. Let \(n=2\), then from (29), it follows that
Applying (29) for \(s\in (\tau _k,\tau _{k+1}]\), we obtain
which implies the desired inequality:
Now, we assume that the Lemma 2 is true for \(n=q\); then from (29), it follows for \(s_{q-1}\in (\tau _k,\tau _{k+1}]\) that
using the induction assumption we obtain the desired inequality for \(n=q+1\):
The Lemma is proved.
1.3 Proof of Proposition 1
We take \(V(u(\cdot ,t))=\int \limits _0^lu^2(z,t)\,dz\) as a Lyapunov function, and calculate its derivative along the flow \(\phi _c\) for \(t\in [0,\tau ]\):
which implies the first of the needed inequalities: \(\Vert u(\cdot ,t)\Vert _{L^2[0,l]}\le e^{\tau b_{\max }}\Vert u(\cdot ,0)\Vert _{L^2[0,l]}\). The second inequality becomes trivial.
1.4 Proof of Proposition 2
We consider the eigenvalue problem (49) where b depends on \(\varepsilon \) treated as a small parameter. It is easy to see that \(\lambda <\max _{z\in [0,\pi ]}b(z)=1+\varepsilon \). For \(\varepsilon =0\) we have \(\lambda _{\max }(0)=0\), hence for the largest eigenvalue, we can write \(\lambda _{\max }(\varepsilon )=\lambda _1\varepsilon +O(\varepsilon ^2)\). The solution of (49) for \(\varepsilon >0\) small enough is given by
where \(C_l\) and \(C_r\) are some real constants, \(\omega _l:=\sqrt{1-2\varepsilon -\lambda }\), \(\omega _r:=\sqrt{1+\varepsilon -\lambda }\), and must satisfy the continuity conditions
These conditions are satisfied if \( \omega _l\text {ctg}(2\omega _l\delta )=-\omega _r\text {ctg}(\omega _r(\pi -2\delta )). \)
We use the following two basic facts from analysis
that enable us to write for \(\varepsilon \rightarrow 0+\) the next equalities.
From which we finally conclude that
From which we calculate \( \lambda _1=\frac{2}{\pi }\Big (\frac{3}{4}\sin (4\delta )+\frac{\pi }{2}-3\delta \Big ). \) This proves the statement.
1.5 Proof of Proposition 3
From the condition of Proposition 1, the following estimate follows:
By the Young’s inequality (2)
we further obtain
Since \(\sup _{s\in \mathbb R_+}\frac{s^{m-1}}{\sqrt{1+s^{2m}}}=\frac{(m-1)^{(m-1)/2m}}{\sqrt{m}}<\infty \), denoting \( \mu _0=\big (a_0+\frac{1}{m+1}\big )\frac{(m-1)^{(m-1)/2m}}{\sqrt{m}}, \) we arrive to the following differential inequality
Hence by the comparison principle, we obtain
for \(t\in [0,\tau ]\). The inequality (5) follows then immediately, if we take \(\omega _1(\tau )=e^{\mu _0\tau }\), \(\omega _2(\tau )=\sqrt{\frac{m}{\mu _0(m+1)}(e^{2\mu _0\tau }-1)}\). The estimate (6) for \(t=k\theta \) can be obtained easier:
This finishes the proof.
1.6 Proof of Proposition 4
From the condition of the proposition, we have that \( \sigma _4(\tau )\frac{(m-1)^{(m-1)/2m}}{\sqrt{m}}<\sigma _3(\tau ), \) is satisfied for \(\tau =0\). By the continuity, the inequality is also true for some small enough \(\tau >0\). A simple calculation shows that \(\max _{s\in \mathbb R_+}\frac{s^{m-1}}{\sqrt{1+s^{2m}}}=\frac{(m-1)^{(m-1)/2m}}{\sqrt{m}}\), so that from the last inequality, we conclude
which proves the Proposition 4.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Dashkovskiy, S., Slynko, V. Stability conditions for impulsive dynamical systems. Math. Control Signals Syst. 34, 95–128 (2022). https://doi.org/10.1007/s00498-021-00305-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00498-021-00305-y
Keywords
- Stability
- Robustness
- Impulsive systems
- Infinite-dimensional systems
- Nonlinear systems
- Lyapunov methods
- Input-to-state stability