Optimal control problem for a differential inclusion with rapidly oscillating coefficients on the semiaxis (Q6043663)

The authors consider the optimal control problem for a system of differential inclusions with rapidly oscillating coefficient: \(\overset{.}{x} \in f(\frac{t}{\varepsilon },x)+f_{1}(x)u(t)\) posed on the semiaxis, and the performance criterion \(J_{\varepsilon }[x,u]=\int_{0}^{\infty }(e^{-jt}A(t,x(t))+u^{2}(t))dt\), to be minimized. Here \(\varepsilon >0\) is a small parameter, \(j>0\) a fixed constant characterizing the discount, \(x\) a phase vector in \(\mathbb{R}^{d}\), and \(u(t)\) an \(m\)-dimensional control vector with \(u(\cdot )\in L_{2}([0,\infty ))\) that takes values in a set \( U\subset \mathbb{R}^{m}\). The initial condition \(x(0,u(0))=x_{0}\)\ is added. The multivalued function \(f\) is supposed to satisfy: \(\lim_{s\rightarrow \infty }\int_{0}^{s}f(t,x)dt=f_{0}(x)\), uniformly in \(x\in \mathbb{R}^{d}\), and growth conditions. The function \(A(t,s)\) defined for \(t\geq 0\), \(x\in \mathbb{R}^{d}\), \(u\in U\) is measurable in \(t\) and continuous in \(x\). It satisfies \(\exists C>0:A(t,x)\geq -C\), and the growth condition: \(\exists K>0:\left\vert A(t,x)\right\vert \leq K(1+\left\vert x\right\vert ^{p})\), for any \(t\geq 0\), \(x\in \mathbb{R}^{d}\), with \(p\geq 0\). The authors first prove that this problem is solvable. They consider a minimizing sequence to this problem on which they prove uniform estimates which allow to pass to the limit. The main result proves the convergence of the optimal trajectories (uniformly on each segment \([0,T]\) for any \(T>0\)), of the optimal controls (weakly in \(L_{2}([0,\infty ))\)), up to some subsequence, and of the optimal values of the above performance criterion to the corresponding parameters of the averaged problem: \(\overset{.}{y}\in f_{0}(x)+f_{1}(x)u(t)\), with the same initial condition and the performance criterion \(J_{0}[x,u]=\int_{0}^{ \infty }(e^{-jt}A(t,y(t))+u^{2}(t))dt\). Moreover, if the averaged problem has a unique solution, then the preceding convergences take place for all \( \varepsilon \rightarrow 0\). The proof is mainly based on the obtention of uniform estimates.