Sufficiency conditions for existence of an optimal feedback control in stochastic mechanics (Q1179522): Difference between revisions

In the framework of stochastic mechanics, the author considers the following optimal feedback control problem: in a set of admissible feedback controls \(v\), in the range of \(n\)-dimensional Euclidean space \(E^ n\), find such a control minimizing the expectation \[ E_{sx}\{\int_ s^ T L(t,\xi(t),v(t,\xi(t)))dt+W_ T(\xi(T))\}, \] for all \((s,x)\in [0,T)\otimes E^ n\), where \(L(t,x,v)=(1/2)mv^ 2- U(t,x)\) is the classical action integrand, \(\xi\) is an \(n\)-dimensional diffusion process (in weak sense) with drift \(v\) and diffusion coefficient \(D=\hbox{const}>0\), \(W_ T\) and \(U\) are given real functions. The difficulties of this apparently common problem of stochastic optimal control come essentially from the fact that the drift \(v\) does not satisfy the usual conditions for existence and uniqueness of a solution (i.e. linear growth and the Lipschitz condition). The author gives sufficiency conditions for the existence and uniqueness of an optimal feedback control to the above problem, using essentially the Girsanov transformation to prove the result.