On extremal solutions to stochastic control problems (Q1180333)

This paper proves the conjecture by \textit{U. G. Haussmann} [Stochastic differential systems, Proc. 3rd Bad Honnef Conf. 1985, Lect. Notes Control Inf. Sci. 78, 171-186 (1986; Zbl 0593.93068)] and \textit{N. El Karoui, D. Huu Nguyen} and \textit{M. Jeanblanc-Pique} [Stochastics 20, 169- 219 (1989; Zbl 0613.60051)] that the solutions of the optimal Markov control in the control of degenerate diffusions must be Markov. Consider a \(d\)-dimensional controlled diffusion process \(X(\cdot)\) satisfying the stochastic differential equations \[ X(t)=X_ 0+\int^ t_ 0 m(X(s),u(s))ds+\int^ t_ 0\sigma(X(s))dW(s) \] with some Wiener process \(W(\cdot)\). The author first establishes some technical lemmas concerning extremal measures on a product Polish space with a prescribed marginal. Then it is shown that the set of attainable laws of \((X(\cdot,u(\cdot))\) under all admissible controls with the law of \(X_ 0\) being kept fixed at \(\pi\) is compact. Using these results, every extremal class is proved to be a Markov class.