Core Concepts
The article presents a new Newton-type method for computing eigenvalues and eigenvectors of locally definite multiparameter eigenvalue problems based on the signed multiindex of the eigenvalues. The method exhibits local quadratic convergence and can even achieve global linear convergence for certain extreme eigenvalues.
Abstract
The article focuses on solving multiparameter eigenvalue problems (MEPs) efficiently using a Newton-type method. Key highlights:
Multiparameter eigenvalue problems combine linear systems of equations and eigenvalue problems, arising in various applications such as mathematical physics, delay differential equations, and optimization problems.
The authors introduce the concept of "local definiteness" as a weaker condition than right and left definiteness, which is often considered for MEPs. This condition is naturally satisfied for multiparameter Sturm-Liouville problems.
The core of the proposed method is to apply a semismooth Newton method to functions that have a unique zero, corresponding to the eigenvalues of the MEP. This approach provides local quadratic convergence and, for certain extreme eigenvalues, even global linear convergence.
The method gains an additional interpretation in the context of discretizing multiparameter Sturm-Liouville eigenvalue problems: it identifies eigenvalues for which the associated eigenfunctions exhibit the designated number of interior zeros.
The authors also discuss handling non-Hermitian matrices in the MEP by introducing diagonal scaling matrices, which can be useful when discretization or transformations compromise the self-adjoint property.
Numerical experiments demonstrate the performance of the proposed method, particularly in targeting a subset of eigenvalues associated with specific multiindices.