Variance minimization for constrained discounted continuous-time MDPs with exponentially distributed stopping times

This paper deals with minimization of the variances of the total discounted costs for constrained Continuous-Time Markov Decision Processes (CTMDPs). The costs consist of cumulative costs incurred between jumps and instant costs incurred at jump epochs. We interpret discounting as an exponentially distributed stopping time. According to existing theory, for the expected total discounted costs optimal policies exist in the forms of randomized stationary and switching stationary policies. While the former is typically unique, the latter forms a finite set whose number of elements grows exponentially with the number of constraints. This paper investigates the problem when the process stops immediately after the first jump. For costs up to the first jump we provide an index for selection of actions by switching stationary policies and show that the indexed switching policy achieves a smaller variance than the randomized stationary policy. For problems without instant costs, the indexed switching policy achieves the minimum variance of costs up to the first jump among all the equivalent switching policies.