Controlled diffusions with constraints (Q1173639)

Controlled vector diffusion processes driven by a vector Wiener process are considered. It is desired to minimize the time average of a functional of the state and control, subject to a finite number of constraints of the same type. The authors first derive convexity and compactness properties for the invariant measures resulting from the application of relaxed Markov controls. It is then possible to characterize the optimal controls in terms of extreme points of the set of admissible controls. Finally, the authors prove that (if the set of admissible controls is non-empty) the constrained problem has an optimal Markov control solution, which moreover has a minimum connected with a Lagrange multiplier formulation involving the constraint functionals. The paper is completed by derivation of similar results for discounted, first exit time, and finite horizon cost criteria; for the latter, the optimal Markov control may depend on the initial state.