On the computation of the gap metric (Q1119226)

The paper presents an expression for the gap metric that is computable using well-known techniques. It also establishes the equivalence between the gap and the graph metrics. Let \(P_ i,i=1,2\), be linear, time-invariant systems, considered as operators acting on \({\mathbb{H}}_ 2\). For \(i=1,2\), the graph of \(P_ i\) is \(graph(P_ i)=G_ i{\mathbb{H}}_ 2\) where \(G_ i\) is an inner matrix function. The gap between \(P_ 1\) and \(P_ 2\) is \[ \delta (P_ 1,P_ 2):=\| {\mathbb{P}}_{G_ 1{\mathbb{H}}_ 2}-{\mathbb{P}}_{G_ 2{\mathbb{H}}_ 2}\| \] where \({\mathbb{P}}_{G_ i{\mathbb{H}}_ 2}\) denotes the orthogonal projection with range \(G_ i{\mathbb{H}}_ 2\). The above expression defines a metric on the space of (possibly unstable) systems. The gap metric has been introduced by Krein and Krasnosel'skij, and Sz.-Nagy [cf. \textit{N. I. Akhiezer} and \textit{I. M. Glazman}, Theory of Linear Operators in Hilbert Space, Vol. I. (New York 1961; for a review of the 1981 English edition see Zbl 0467.47001). In control theory, it has been introduced by \textit{G. Zames} and \textit{A. K. El-Sakkary} [Unstable Systems and feedback: The gap metric. Proc. Allerton Conf., Oct. 1980, pp. 380-385.] [see also \textit{A. K. El-Sakkary}, IEEE Trans. Autom. Control AC-30, 240-247 (1985; Zbl 0561.93047)] as being appropriate for the study of uncertainty in feedback systems. The gap metric induces the weakest topology in which feedback stability is a robust property. Using the commutant lifting theorem it is shown that \[ \delta (P_ 1,P_ 2)=\max \{\inf_{Q\in H_{\infty}}\| G_ 1-G_ 2Q\|_{\infty},\inf_{Q\in {\mathbb{H}}_{\infty}}\| G_ 2-G_ 1Q\|_{\infty}\}. \] The computation of the infima above has been studied extensively over the last few years and can be carried out using well-known techniques [cf. \textit{B. A. Francis}, A course in \({\mathbb{H}}_{\infty}\) control theory, Lect. Notes Control Inf. Sci. 88 (1987; Zbl 0624.93003), Ch. 8]. An alternate metric with similar properties has been introduced by \textit{M. Vidyasagar} [IEEE Trans. Autom. Control AC-29, 403-418 (1984; Zbl 0536.93042)]. This, known as the graph metric, is defined by: \[ d(P_ 1,P_ 2)=\max \{\inf_{Q\in {\mathbb{H}}_{\infty},\| Q\|_{\infty}\leq 1}\| G_ 1-G_ 2Q\|_{\infty},\inf_{Q\in {\mathbb{H}}_{\infty},\| Q\|_{\infty}\leq 1}\| G_ 2-G_ 1Q\|_{\infty}\}. \] It is shown that \(2\delta (P_ 1,P_ 2)\geq d(P_ 1,P_ 2)\geq \delta (P_ 1,P_ s)\).