scientific article; zbMATH DE number 1500628

From MaRDI portal
Publication:4501650





It is well known that lack of existence of optimal controls for ordinary differential systems \(x'(t) = f(t, x(t), u(t))\) with compact control set \(U\) can be ``fixed by taking as control space the subset of the dual \(L^1(0, T; C(U))^*\) \((C(U) =\) continuous functions in \(U,\) supremum norm) consisting of functions \(t \to \mu(t)\) (= probability measure in \(U)\). The measures are \(\sigma\)-additive and the relaxed system is \[ x'(t) = \int_U f(t, x(t), u)\mu(t, du) . \] Extension of this result to infinite dimensional systems where the control set \(U\) is not even locally compact can be done in two ways. If \(U\) is a normal topological space, the definition above can be generalized verbatim replacing \(C(U)\) by \(BC(U)\) (= bounded continuous functions in \(U,\) supremum norm); the downside is, relaxed controls are merely \(t\)-dependent finitely additive measures. The other approach (due to Roubíček) is to imbed \(U\) in a compact or locally compact topological space, which changes the control set but produces controls whose values are \(\sigma\)-additive measures. As pointed out by the author in previous work, the Stone-Čech compactification works for general \(U,\) but if one restricts the control set then ``tighter (smaller) compactifications will do the job as well, may have a richer structure and may be more closely related to \(U.\) In this paper, \(U\) is a subset of \(L^p(\Omega, S),\) where \(\Omega\) is an \(n\)-dimensional domain and \(S\) a finite dimensional (or, more generally, separable) Banach space. The compactifications are subsets of locally convex spaces. The theory is applied to an optimal control problem for a semilinear system described by a partial differential equation. For the entire collection see [Zbl 0921.00020].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4501650)