Decentralized control by rational controllers

A strongly decentralized control problem is considered where only local information is available on both states and goals (the desired states) of each subsystem. A theory is presented which allows one to optimize the behavior of a system with such an information structure.The development is based on a notion of rational controllers which are ergodic dynamical systems with some additional property of optimality. Analysis of the interacting rational controllers gives the conditions under which a system of rational controllers also exhibits a rational behavior. The knowledge of these conditions allows one to suggest a control technique for a strongly decentralized situation.Dedicated to R. Bellman