On condition numbers of polynomial eigenvalue problems

Abstract: In this paper, we investigate condition numbers of eigenvalue problems of matrix polynomials with nonsingular leading coefficients, generalizing classical results of matrix perturbation theory. We provide a relation between the condition numbers of eigenvalues and the pseudospectral growth rate. We obtain that if a simple eigenvalue of a matrix polynomial is ill-conditioned in some respects, then it is close to be multiple, and we construct an upper bound for this distance (measured in the euclidean norm). We also derive a new expression for the condition number of a simple eigenvalue, which does not involve eigenvectors. Moreover, an Elsner-like perturbation bound for matrix polynomials is presented.

Review text

Condition numbers of eigenvalue problems of matrix polynomials with nonsingular leading coefficients are investigated. In Section 3 it is proved that if \(\lambda_0\) is a multiple eigenvalue of the matrix polynomial \(P(\lambda)\) and \(p_0\) is the dimension of the maximum Jordan blocks of \(\lambda_0\), then as \(\epsilon\to 0\), the Hausdorff distance between the connected component of the pseudospectrum \(\sigma_\epsilon(P)\) containing \(\lambda_0\) and the disc centered at \(\lambda_0\) of radius \(r=(\hat{k}(P,\lambda_0)\epsilon)^{1/p_0}\) is \(o(\epsilon^{1/p_0})\). Here \(\hat{k}(P,\lambda_0)\) is the condition number of the multiple eigenvalue \(\lambda_0\). An upper bound of the distance of a simple eigenvalue to the set of matrix polynomials having this eigenvalue with multiplicity bigger than 1 is given in Section 4. More precisely, it is proved that this distance is less than or equal to \[ \frac{c(P'(\lambda_0)) \parallel P(\lambda_0)\parallel} {k(P,\lambda_0) (\parallel y_0^* P'(\lambda_0)\parallel^2 - |y_0^* P'(\lambda_0)x_0|^2)^{1/2}}, \] where \(y_0\) and \(x_0\) are the left and right unit eigenvectors and \(c(A)\) is the condition number of \(A\). In Section 5 an expression of the condition number is obtained without eigenvectors and in Section 6 a bound of the minimum distance between an eigenvalue of a perturbed matrix to the eigenvalues of the original matrix is derived using a technique of \textit{L. Elsner} [Linear Algebra Appl. 71, 77--80 (1985; Zbl 0583.15009)]. All result are illustrated by examples.

Recommendations

Cites work

Cited in

(27)

Describes a project that uses

Uses Software

Eigtool

This page was built for publication: On condition numbers of polynomial eigenvalue problems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q972167)