An alternative solution to the \(H_{\infty}\)-optimal control problem (Q1107485): Difference between revisions

This paper considers the \(H^{\infty}\) optimization problem \(\min \| A+BXC\|_{\infty}:\) \(X\in H^{\infty}_{n\times n}\}\) where A, B and C are rational matrices in \(H^{\infty}_{n\times n}\). While the scalar case is classical and has its origins in the work of Nevanlinna and Pick in the early part of the century, the matrix case has received attention only since the mid-70's, especially lately due to its connection with the standard problem of \(H^{\infty}\) control theory. The usual approach in the control theory literature to solve the matrix problem has been to compute inner-outer and outer-inner factorizations of B and C respectively, which enables one to reduce the problem to a matrix Nehari problem. Then one can draw on the elegant one-step extension methods of Adamian-Arov-Krein or the state space formulas of Glover to obtain solutions. Recently there has been a search for a more direct method which avoids the computation of inner-outer factorizations. Several groups (Kimura, Limebeer-Anderson, Ball-Gohberg-Rodman) have found finite-step algorithms or formulas which come to terms with the intricate matrix structure of the problem to obtain analytically exact solutions of the problem. The authors, on the other hand, ignore much of the matrix structure and present a simple-minded iterative dynamic programming algorithm as the solution of the problem. The diagonally scaled version of the problem \((\inf \| D(A+BXC)D^{- 1}\|_{\infty}\), where the infimum is over D, X in \(H^{\infty}_{n\times n}\) with D diagonal), is also of interest in control theory; via the authors' approach, this requires a double dynamic programming algorithm.