Spectral conditioning and pseudospectral growth (Q2642284)

For \(\varepsilon\geq 0\), the \(\varepsilon\)-pseudospectrum of a square matrix \(A\) is the set \[ \Lambda_\varepsilon(A)=\{ w\in \mathbb{C}\mid\underline{\sigma}(A-wI)\leq \varepsilon\}, \] where \(\underline{\sigma}(A)\) denotes the smallest singular value of \(A\). Using a result of \textit{M. Karow} [Geometry of spectral value sets, PhD thesis, University of Bremen (2003)] on the growth rate of pseudospectra, this paper refreshes the perturbation theory for eigenvalues with arbitrary Jordan structure due to \textit{V. B. Lidskii} [Zh. Vychislit. Mat. Mat. Fiz. 6, 52--60 (1966; Zbl 0166.40501)], and later reviewed by \textit{J. Moro, J. V. Burke} and \textit{M. L. Overton} [SIAM J. Matrix Anal. Appl. 18, No.~4, 793--817 (1997; Zbl 0889.15016)], where the Hölder condition number was introduced. Considering now larger perturbations and following \textit{R. Alam} and \textit{S. Bora} [Linear Algebra Appl. 396, 273--301 (2005; Zbl 1084.15017)], the authors provide a new proof for the characterization of the distance from a given matrix to the set of matrices with multiple eigenvalues in terms of the number of connected components of pseudospectra.