scientific article; zbMATH DE number 3878236

From MaRDI portal
Publication:3343313





Matrices with a very ill-conditioned eigenvalue problem are studied. Explicit expressions are derived for perturbations that cause subsets of the eigenvalues to coalesce. In simple cases, the smallest possible perturbations are obtained, and it is indicated when those are relevant also in the general case. It is shown that cases with double eigenvalues generally occur for significantly smaller perturbations than cases with a whole group of eigenvalues coalescing. The rather involved paths described by the eigenvalues, when a curve of perturbations, \(A+\eta E\), \(0\leq \eta \leq \epsilon\), is followed, shows that just considering the Puiseux expansion around a multiple eigenvalue, as has been the case in most studies up to now, is not enough. The analysis given here is most significant for those developing software for ill-conditioned eigenvalue problems, giving indications on the questions which can be answered, when one is given a matrix of numbers of a limited precision. The tolerances accepted and bounds given by routines like JNF [cf. \textit{B. Kågström} and the reviewer, ACM Trans. Math. Software 6, 398-419 and 437-443 (1980; Zbl 0434.65020 and Zbl 0434.65021)] may have to be adjusted.




Cited in
(35)








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 Q3343313)