Matrix valued versions of a result of von Neumann with an application to time discretization (Q1062727): Difference between revisions

\textit{J. von Neumann} [Math. Nachr. 4, 258-281 (1951; Zbl 0042.123)] proved the following theorem: ''Let C(z) be a rational function satisfying \(| C(z)\| \leq 1\) for \(| z| \leq 1\) and let A be a linear operator in a Hilbert space H satisfying \(\| A\| \leq 1\). Then one has \(\| C(A)\| \leq 1.''\) This theorem has successfully been applied to the study of stability (and contractivity) of one-step methods for ordinary differential equations. The present paper treats a generalization of the above theorem to matrix-valued functions C(z). It is shown that the theorem of von Neumann remains valid, if \(| C(z)|\) is interpreted as the operator norm that corresponds to the Euclidean inner product, i.e. \(| C(z)| =\sup | C(z)\xi |_ 2/| \xi |_ 2\). \(\| C(A)\|\) has to be interpreted similarly. Furthermore, it is shown that in this generalization of von Neumann's result the Euclidean norm cannot be replaced by a norm, which is not defined by an inner product. However, by restricting the class of functions C(z) the author arrives to prove a similar theorem also for the max-norm. These results allow to extend the above mentioned stability investigations to multistep methods. An application to a special class of one-leg methods is given.