All optimal Hankel-norm approximations of linear multivariable systems and theirL,∞-error bounds†

From MaRDI portal
Publication:3332876





The problem is considered of approximating a linear system, with transfer function matrix \(G(s)\) of McMillan degree n, by a linear system with transfer function matrix \(\hat G(s)\) of McMillan degree \(k<n.\) A complete characterization is derived of all approximations that minimize the Hankel norm of the error system: \(\| G-\hat G\|_ H.\) The Hankel norm is an induced operator norm obtained by considering a linear system as a mapping between square integrable past inputs and square integrable future outputs. The key to the solution to this problem is the characterization of all matrices, with rational entries, in the form \(\hat G(s)+F(s)\) which minimize \[ \| G-\hat G-F\| =\sup_{\omega}{\bar \sigma}(G(j\omega)-\hat G(j\omega)-F(j\omega)), \] where \(\hat G(s)\) has McMillan degree k, \(F(s)\) is anti-causal and \({\bar \sigma}\)(.) denotes the maximum singular value of a matrix. The solution to this problem is obtaind using results on balanced realizations, all- pass functions and the inertia of matrices. An algorithm is presented for finding Hankel norm approximations, and various error bounds on the approximation error are given. For one class of approximants it is shown that \(\| G-\hat G\| \leq \sum^{n}_{i=k+1}\sigma_ i(G),\) where \(\sigma_ i(G)\) is the \(i^{th}\) largest Hankel singular value of G. Bounds such as this are important for predicting the performance of control schemes designed for \(\hat G\) but applied to G. Although the impetus for obtaining the results contained in this paper owes much to the pioneering work of \textit{V. M. Adamyan}, \textit{D. Z. Arov} and \textit{M. G. Krejn} [Mat. Sb., Nov. Ser. 86(128), 34-75 (1971; Zbl 0243.47023); Izv. Akad. Nauk Arm. SSR, Mat. 6, 87-112 (1971; Zbl 0311.15012)] on the approximation of Hankel matrices, the author has developed his own rather ingenious methods of obtaining closed form solutions to the finite dimensional case. The paper is selfcontained, and although rather long, is extremely well written and will reward its reader with many interesting results.




Cited in
(only showing first 100 items - show all)








This page was built for publication: All optimal Hankel-norm approximations of linear multivariable systems and theirL,∞-error bounds†

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3332876)