Abstract Generated abstract
This note studies an averaging procedure for nonlinear integro-differential equations with a small parameter, in which the memory term is replaced by an integral over an infinite interval and the resulting vector field is time averaged. Under continuity, Lipschitz, boundedness, and growth assumptions on the kernel and averaged field, the authors prove that solutions of the original and averaged systems remain close on intervals of order the reciprocal of the small parameter. A second result gives closeness for all positive time when the averaged solution is uniformly asymptotically stable and related uniform averaging conditions hold. The paper also indicates how the method applies to more general coupled systems reducible to standard form and to Volterra integral equations after differentiation.
Full Text
Reports of the Academy of Sciences of the USSR
1970. Vol. 192, No. 4
UDC 517.948.34
MATHEMATICS
A. N. FILATOV, L. I. TALIPOVA
ON ONE VARIANT OF AVERAGING IN INTEGRO-DIFFERENTIAL EQUATIONS
(Presented by Academician A. A. Dorodnitsyn on 17 XI 1969)
1. Consider the system of nonlinear integro-differential equations
\[ \dot{x}=\varepsilon X\left(t,x,\int_0^t \varphi(t,s,x(s))\,ds\right). \tag{1} \]
Here \(\varepsilon>0\) is a small parameter, \(x=(x_1,x_2,\ldots,x_n)\).
In papers \((^1,^2)\), for equations of the form (1) the following averaging procedure was proposed. Let
\[ \psi(t,x)=\int_0^\infty \varphi(t,s,x)\,ds, \tag{2} \]
\[ X_0(x)=\lim_{T\to\infty}\frac{1}{T}\int_0^T X(t,x,\psi(t,x))\,dt. \tag{3} \]
Then to the system (1) there is put in correspondence an averaged system of the form
\[ \dot{\xi}=\varepsilon X_0(\xi). \tag{4} \]
In the present note theorems are proved on the closeness of the solutions of systems (1) and (4), both on a finite and on an infinite interval.
Theorem 1. Let the functions \(X(t,x,y)\) and \(\varphi(t,s,x)\) be defined and continuous in the domain \(Q\{t\geq 0,\ s\geq 0,\ x\in D,\ y\in E_n\}\), and suppose that in this domain the following conditions are fulfilled:
1)
\[
\left|X(t,x',y')-X(t,x'',y'')\right|
\leq \lambda\{|x'-x''|+|y'+y''|\},
\]
\[ \left|\varphi(t,s,x')-\varphi(t,s,x'')\right| \leq \mu(t,s)|x'-x''|, \]
\[ \int_0^t d\tau\int_0^\tau \mu(\tau,s)\,ds\leq At^\alpha, \qquad A>0,\quad 0\leq \alpha<1,\quad \lambda=\text{const}. \]
2) The function \(\psi(t,x)\), defined by equality (2), satisfies the Lipschitz condition
\[ |\psi(t,x')-\psi(t,x'')|\leq \nu |x'-x''|, \qquad \nu=\text{const}. \]
3) At each point \(x\) of the domain \(D\) the limit (3) exists, and the function \(X_0(x)\) is bounded \((|X_0|\leq M)\) and satisfies the Lipschitz condition.
4) The solution \(\xi=\xi(t)\), \(\xi(0)=x(0)\in D\), of the averaged system is defined for all \(t\geq 0\) and lies in the domain \(D\) together with some \(\rho\)-neighborhood.
5) Along the trajectory \(\xi(t)\),
\[ \int_0^t d\tau\left|\int_\tau^\infty \varphi(\tau,s,\xi(\tau))\,ds\right| \leq Bt^\beta, \qquad B>0,\quad 0\leq \beta<1. \]
Then for any \(\eta>0\) and \(L>0\) one can specify an \(\varepsilon_0\) such that, for \(\varepsilon<\varepsilon_0\), on the interval \(0\leqslant t\leqslant L\varepsilon^{-1}\) the inequality
\[
|x(t)-\xi(t)|<\eta
\]
will hold.
Proof. As in \((3)\), it is shown that for any \(a>0\) one can specify an \(\bar\varepsilon\) such that, for \(\varepsilon<\bar\varepsilon\), on the interval \(0\leqslant t\leqslant L\varepsilon^{-1}\) the inequality
\[
\varepsilon\left|\int_0^t\left[
X\left(\tau,\xi(\tau),\int_0^\infty \varphi(\tau,s,\xi(\tau))\,ds\right)
-
X_0(\xi(\tau))
\right]\,d\tau\right|<a
\]
will hold. Hence,
\[
\begin{aligned}
|x-\xi| \leqslant{}& a+\varepsilon\lambda\int_0^t |x(\tau)-\xi(\tau)|\,d\tau
+\varepsilon\lambda\int_0^t d\tau\int_0^\tau \mu(\tau,s)|x(s)-\xi(s)|\,ds \\
&+\varepsilon\lambda\int_0^t d\tau\left|\int_\tau^\infty \varphi(\tau,s,\xi(\tau))\,ds\right|
+\varepsilon\lambda\int_0^t d\tau\int_0^\tau \mu(\tau,s)|\xi(s)-\xi(\tau)|\,ds .
\end{aligned}
\]
From this, taking into account the conditions of the theorem, we find on the interval \(0\leqslant t\leqslant L\varepsilon^{-1}\):
\[
|x-\xi|\leqslant
\left(a+\lambda M A L^{1+\alpha}\varepsilon^{1-\alpha}
+\lambda B L^{1+\beta}\varepsilon^{1-\beta}\right)
e^{\lambda L+A L\alpha\varepsilon^{1-\alpha}} .
\]
The assertion of the theorem follows from the last inequality.
Remark 1. Consider the system
\[
\dot x=\varepsilon X(t,x)+\varepsilon\int_0^t Y(t,s,x(s))\,ds.
\tag{*}
\]
If the functions \(X\) and \(Y\) are bounded, then the solution \(x(t)\) of system \((*)\) will change substantially already on an interval of order \(\varepsilon^{-1/2}\). Therefore, in particular, for systems of the form \((*)\) one can formulate an averaging theorem analogous to Theorem 1, establishing the closeness of the solutions of the original and averaged systems on the interval \(0\leqslant t\leqslant L\varepsilon^{-1/2}\). In this case the parameters \(\alpha\) and \(\beta\) will vary within the limits \(0\leqslant \alpha<2\), \(0\leqslant \beta<2\).
Theorem 2. Replace condition (3) of Theorem 1 by the following:
a) at every point \(x\) of the domain \(D\), uniformly with respect to \(t\), there exists the limit
\[
\lim_{T\to\infty}\frac1T\int_t^{t+T} X(t,x,\psi(t,x))\,dt=X_0(x),
\tag{5}
\]
and the function \(X_0(x)\) is bounded and satisfies the Lipschitz condition;
b) the solution \(\xi=\xi(\tau)\), \(\tau=\varepsilon t\), of the averaged system
\[
\dot\xi=\varepsilon X_0(\xi),\qquad \xi(0)=x(0)
\]
is uniformly* asymptotically stable;
c) equation (1) has no singular points.
Then for any \(0<\eta<\rho\) one can specify an \(\varepsilon_0\) such that, for \(\varepsilon<\varepsilon_0\), for all \(t>0\) the inequality
\[
|x(t)-\xi(t)|<\eta
\]
will hold.
Proof. The proof is carried out by the methods set forth in \((4,6)\).
* For the notion of uniform asymptotic stability, see, for example, \((5)\).
Remark 2. Let us note that in many cases the requirement of uniform asymptotic stability appearing in Theorem 2 can be weakened.
2. Consider a system of a more general form
\[ \dot{x}=\varepsilon X\left(t,x,\dot{x},y,\dot{y},\int_{0}^{t}\varphi(t,s,x(s),\dot{x}(s),y(s),\dot{y}(s))\,ds\right), \]
\[ \dot{y}=Y_0(t,x,y)+\varepsilon Y_1\left(t,x,\dot{x},y,\dot{y},\int_{0}^{t}\psi(t,s,x(s),\dot{x}(s),y(s),\dot{y}(s))\,ds\right). \tag{6} \]
If the general solution of system (6) for \(\varepsilon=0\) is known, then, as was shown in (4), this system can be reduced to standard form. After this, the averaging procedure under consideration can be applied to the resulting system.
3. Let a system of nonlinear integral equations of Volterra type be given:
\[ u(t)=\varepsilon\int_{0}^{t}\Phi(t,s,u(s))\,ds. \tag{7} \]
Differentiating (7), we find
\[ \dot{u}=\varepsilon\Phi(t,t,u)+\varepsilon\int_{0}^{t}\frac{\partial \Phi(t,s,u(s))}{\partial t}\,ds. \tag{8} \]
System (8) is integro-differential, and averaging Theorems 1 and 2 are applicable to it.
Institute of Cybernetics with Computing Center
Academy of Sciences of the Uzbek SSR
Tashkent
Received
13 XI 1969
CITED LITERATURE
¹ A. N. Filatov, L. I. Talipova, Differentsial’n. uravn., 5, No. 5 (1969).
² A. N. Filatov, L. I. Talipova, Izv. AN UzSSR, Ser. Tekhn. Nauk, No. 2 (1969).
³ G. S. Larionov, A. N. Filatov, ibid.
⁴ A. A. Ilyushin, G. S. Larionov, A. N. Filatov, DAN, 188, No. 1 (1969).
⁵ B. N. Krasovskii, Some Problems in the Theory of Stability of Motion, Moscow, 1959.
⁶ C. Banfi, Boll. Unione Mat. Ital., 22 (1967).