scientific article; zbMATH DE number 4117742

From MaRDI portal
Publication:4732031





[For the entire collection see Zbl 0672.00008.] This paper is concerned with theoretical results and numerical procedures to determine\(\mu^*:=\inf \{\lambda \geq 0:\quad P(\lambda)\) not positive definite\(\}\) where \(P(\lambda)=A+\lambda B+\lambda^ 2C\) (\(\lambda\in {\mathbb{R}})\) is a quadratic \(\lambda\)-matrix with Hermitian matrices A, B, C and \(A=P(0)\) positive definite. The author's approach to determine \(\mu^*\) is based on the following theorem 3.1: Assume that there exists a (shift) function \(\phi\) : [0,\(\mu^*)\to (0,\infty]\) such that for each \(\lambda \in [0,\mu^*)\) the following properties hold: (i) P(\(\alpha)\) nonsingular for \(\alpha \in [\lambda,\lambda +\phi (\lambda)).\) (ii) There exists \(\epsilon >0\), \(\delta >0\) with \(\phi (\beta)>\delta\) for all \(\beta \in [0,\mu^*)\) and \(| \beta -\alpha | <\epsilon.\) Then the sequence \(\lambda_ 0:=0\), \(\lambda_{m+1}:=\lambda_ m+\phi (\lambda_ m)\) either satisfies \(\lambda_ k=\mu^*\) for some k or converges strongly monotonically to \(\mu^*.\) For the practical computation of \(\mu^*\) this paper proposes two methods: The first one is based on the computation of eigenvalues with smallest modulus of the shifted eigenvalue problem. The second one depends on tests of positive definiteness of auxiliary matrices derived through linearization with respect to \(\lambda\). Both methods are fast and reliable and were tested in several examples. In addition the authors study an approach suggested by R. Szilard using pseudolinearization. As this method turned out to fail occasionally, sufficient conditions for convergence are derived.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4732031)