9309154 Sennott The objective is the characterization and computation of optimal policies for the minimization of the average cost per unit time. Such policies enable complex systems to be efficiently controlled. Markov decision chains with countable state spaces are studied. Know axioms guarantee the existence of average cost optimal stationary policies. It will be determined whether these policies are also optimal under additional criteria, and when they may be computed by value iteration. In addition, the adaptive control case when certain system parameters are unknown will be treated. An important result of the research will be the development of an algorithm for the computation of optimal policies when the state space is denumerably infinite. The existence and computability of controls will be shown for a new class of stochastic control problems under a variety of average cost criteria. These extensions will have direct applicability to queuing systems, multiple-access communications, and vehicle control ***
