Game approach to risk sensitive control for stochastic evolution systems
The author considers the system \(\xi\) governed by the following stochastic differential equation on some Sobolev space \(H\), \[ d\xi(t)= [A\xi(t)+ \delta(\xi(t),U(t))]dt+ \sqrt{\varepsilon} dM_t, \] where \(A\) is a uniformly elliptic differential operator, \(M(t)\) is an \(H\) valued random noise. The risk sensitive criterion \(Y^\varepsilon\) is defined by \[ Y^\varepsilon (t,\eta,U)= E\exp \frac{1}{\varepsilon} \Biggl[ \int_0^t f(\xi(s), U(s))ds+ q(\xi(t)) \Biggr], \] where \(U\) is an admissible control. In order to study a small noise asymptotic of the value function, the author deals with the logarithmic transformation, \(\nu^\varepsilon (t,\eta)= \varepsilon \log\inf_U Y^\varepsilon (t,\eta,U)\). The game representation of \(\nu^\varepsilon\) is considered. In fact, \(\nu^\varepsilon\) satisfies a Hamilton-Jacobi-Bellman equation in the viscosity sense, and this equation is equivalent to the upper Isaacs equation through a Legendre transformation. Further, the coincidence of \(\nu^\varepsilon\) with the upper value function (in Elliot-Kalton sense), say \(V^\varepsilon\), of a stochastic differential game is established. The existence of \(\lim_{\varepsilon\to 0} V^\varepsilon\) and the speed of convergence are studied. The limit function \(V^0\) is the unique viscosity solution of the paper Isaacs equation with \(\varepsilon=0\), and coincides with the upper value function of a deterministic differential game. As a consequence, the game has a value. For the entire collection see [Zbl 0905.00042].
This page was built for publication: Game approach to risk sensitive control for stochastic evolution systems
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4227199)
