No label defined (Q3974812)

[For the entire collection see Zbl 0728.00017.] The author uses functional occupation measures induced by the sample paths of a diffusion process (more usually called empirical processes) in order to analyze control problems with infinite horizon and ergodic averages as cost functions. The simplest result (for an uncontrolled process) is the following. Assume \((x(t), t\geq 0)\) is a diffusion process which is tight and has a unique invariant probability measure. Then the sequence of process valued random variables, \[ \widetilde P_ T=(1/T)\int^ T_ 0\delta_{x(t+\cdot)}dt,\quad\text{as } T\to\infty, \] converges weakly to the law of the stationary diffusion with the same transition as \(x\), for all but a negligible set of sample paths. An analogous result holds for a controlled diffusion together with control process. This allows to show the convergence of cost functions averaged on \((0,T)\) to the expected value of the cost for unit time of the limiting stationary diffusion driven by some control with stationary increments, as \(T\to\infty\). The same result is obtained for a singularly perturbed control problem in which a \(\delta\)-optimal control for the averaged system is used and for reflected diffusions. Some further discussions of these ideas can be found in [the author, Weak convergence methods and singularly perturbed stochastic control and filtering problems (Boston, 1990)].