Optimal Hankel norm model reductions and Wiener-Hopf factorization II: The non-canonical case (Q1089884)

The model reduction problem is to approximate a given rational matrix function \(G(s)\) which is stable (i.e. all poles in the left half plane) by a stable rational matrix function \(\hat G(s)\) having a prescribed McMillan degree k smaller than that of \(G(s)\). The norm chosen for the approximation is the Hankel norm, or equivalently, the distance of the matrix function to the antistable functions (poles in the right half plane) in the \(L^{\infty}\)-metric (taken with respect to the imaginary axis). The minimum error of the approximation to G in the Hankel norm is known to equal the \((k+1)\)-st singular value of the Hankel operator \(H_ G\) induced by G. The authors produce state space formulas for a (degenerate) linear fractional map which parametrizes all best approximants \(\hat G+H\) (\(\hat G\) of McMillan degree k, H antistable). The approachis to use the approach of Ball-Helton to reduce the problem to one of symmetric Wiener-Hopf factorization. State space formulas for the Wiener-Hopf factors due to Kaashoek and Ran are then applied to obtain state space formulas for the desired linear fractional map. Earlier work of the authors handled the easier suboptimal problem by this same approach [Part I is to appear in Control Opt.]. The derivation gives an alternative more systematic approach to equivalent results obtained recently by Keith Glover.