No label defined (Q4035398)

[For the entire collection see Zbl 0752.00047.] Let \(X_ t\) be a controlled diffusion process governed by \[ dX_ t= b(X_ t,U_ t)dt+ dW_ t,\quad t\geq 0,\quad X_ 0= x_ 0, \] where \(\{W_ t\}\) is a standard Brownian motion, \(\{U_ t\}\) is a nonanticipative control taking values in a closed set \(U\subset \mathbb{R}^ n\), \(b(.)\) is assumed to be continuous and the cost functional is \[ C_ T= \int^ T_ 0 c(X_ t,U_ t)dt\quad\text{with a given } c(.,.)\geq 0. \] Admitting a class of piecewise continuous stationary control policies, the paper considers the cost \((C_ T-\theta T)/\sqrt T\) which is cost normalized with respect to the average cost. Subject to some conditions it is shown, that, asymptotically, the distribution of the normalized cost under the control 0 (the latter denotes an optimal stationary control with respect to the average cost and is assumed to exist) is optimal to the original control problem in a rather general sense. The proof extends methods presented in \textit{V. Dufková} [Adv. Appl. Probab. 9, 105-124 (1977; Zbl 0356.60024)].