Annihilator structure of a principal ideal: Relation to optimal compensators

Abstract

Given a linear, time-invariant, discrete-time plant, we consider the optimal control problem of minimizing, by choice of a stabilizing compensator, the seminorm of a selected closed-loop map in a basic feedback system. The seminorm can be selected to reflect any chosen performance feature and must satisfy only a mild condition concerning finite impulse responses. We show that if the plant has no poles or zeros on the unit circle, then the calculation of the minimum achievable seminorm is equivalent to the maximization of a linear objective over a convex set in a low-dimensional Euclidean space. Hence, for a wide variety of optimal control problems, one can compute the answer to an infinite-dimensional optimization by a finite-dimensional procedure. This allows the use of effective numerical methods for computation.