Optimal Stationary Value of Drift-Dominated Controlled Diffusion Processes: Computational Challenges and Effective Algorithms.

Aaron Palmer (Dept. of Mathematics, Cornell University)

Efficient numerical methods are well-known for deterministic optimal control problems and for diffusion-dominated controlled drift-diffusion problems. However, neither class of methods is efficient when the diffusion term is degenerate or the drift term is dominant. In this talk we explore adaptations of value and policy iteration algorithms along with Gauss-Seidel relaxation to compute approximate optimal value, and hence the optimal feedback control, in the region between deterministic and non-degenerate stochastic problems.
(Joint work with A. Vladimirsky.)