Structured pseudospectra for nonlinear eigenvalue problems (Q2468130)

The authors consider a nonlinear eigenvalue problem, which includes a sum of matrices and corresponding nonlinear coefficient functions. The pseudospectrum contains all eigenvalues with respect to a set of perturbed problems. Thus pseudospectra characterise the sensitivity of the nonlinear eigenvalue problem. In the previous article \textit{W. Michiels, K. Green, T. Wagenknecht} and \textit{S.-I. Niculescu} [Linear Algebra Appl. 418, No. 1, 315--335 (2006; Zbl 1108.15010)], the authors investigate unstructured perturbations of the matrices, i.e., fully occupied defect matrices are added. In the article at hand, a class of structured perturbations is assumed, where the variations just take place within predetermined parts of the matrices. The aim is to achieve physically reasonable perturbations with respect to an underlying application. Furthermore, crucial properties or structures of the original matrices shall be preserved like singularities or block partitionings, for example. The article consists of two parts. In the first part, the authors introduce the structured perturbations and define an according structured pseudospectrum. It follows that the pseudospectrum exhibits a computable formula based on the singular values of specific defect matrices. In the second part, two test examples are discussed and numerical simulations are presented. Firstly, a mechanical system yields three linear ordinary differential equations of second order, which imply a quadratic eigenvalue problem by their characteristic matrix. Secondly, a system of three linear delay differential equations resulting from laser physics is examined, where the characteristic function features infinitely many eigenvalues. The computational effort for the simulation of structured perturbations is higher in comparison to unstructured perturbations in both cases. However, the results corresponding to the structured case reproduce the sensitivity of the problems qualitatively and quantitatively more accurate.