Ill-Conditioned Matrices Are Componentwise Near to Singularity

Conditioning of matrices (15A12) Inequalities involving eigenvalues and eigenvectors (15A42) Numerical computation of matrix norms, conditioning, scaling (65F35)

Review text

For an \(n\times n\) matrix \(A\) and a nonnegative weight matrix \(E\), the componentwise distance to the nearest singular matrix is defined by \[ \sigma(A,E)=\min\{\varepsilon\geq 0; | A-A'| \leq\varepsilon E, A' \text{\;singular}\}. \] In his earlier paper [Linear Multilinear Algebra 42, No. 2, 93-107 (1997; Zbl 0877.15030)] the author proved the inequality \[ \sigma(A,E)<{(3+2\sqrt{2})n \over \rho(| A^{-1}| E)}, \] thereby solving in the affirmative a conjecture formulated by \textit{J. Demmel} [SIAM J. Matrix Anal. Appl. 13, No. 1, 10-19 (1992; Zbl 0749.65031)] and \textit{N. J. Higham} [Accuracy and stability of numerical algorithms (1996; Zbl 0847.65010)] (concerning the numerator in the right-hand side which is rather surprisingly independent of \(A\) and \(E\)). In the present paper the author gives another proof of this result using only basic facts from linear algebra.

Recommendations

Cited in

(21)

Describes a project that uses

Uses Software

mctoolbox

This page was built for publication: Ill-Conditioned Matrices Are Componentwise Near to Singularity

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4255826)