Sur l'approximation des réduites. (On the approximation of residues) (Q917159)

Let \((X_ t,{\mathcal F}_ t,P_ x)_{t\geq 0}\) be a continuous-time Markov process defined on a fixed probability space (\(\Omega\),\({\mathcal F},P)\) with values on a local compact with countable base space (E,\({\mathcal E})\), where \({\mathcal E}\) is the Borel \(\sigma\)-field. For given \(T>0\) and a continuous, non-bounded function f, define \[ u(x,T,f)=\sup_{\tau \in {\mathcal T}_ T}E_ xf(X_{\tau}), \] where \({\mathcal T}_ T\) is a set of stopping times with respect to (\({\mathcal F}_ t)_{t\geq 0}\) with values in [0,T]. The numerical calculation of the function u(\(\cdot,T,f)\) can be done by the analytic method based on the fact that this function is a solution of a variational inequality or by the probabilistic method based on an approximation of Markov processes by Markov chains and the calculation of the function u(\(\cdot,T,f)\) for Markov chains by backward induction. The justification of the second method is given in a different way than that one used by \textit{H. J. Kushner} [Probability methods for approximations in stochastic control and for elliptic equations. (1977; Zbl 0547.93076)]. It is shown that the convergence of infinitesimal generators implies the uniform convergence of suitable values of stopping problems. The result is based on the Trotter-Kato theorem which connects the convergence of infinitesimal operators and semigroups. The related problem for general continuous-time stochastic processes is also considered. In this case a lower semi-continuity property of u(\(\cdot,T,f)\) is proved under very weak assumptions.