Fast enclosure of matrix eigenvalues and singular values via rounding mode controlled computation (Q5929749)

The author derives two methods to enclose matrix eigenvalues and one method to enclose singular values. The entries of the matrix are assumed to be floating point numbers. The computations are executed in a floating point system using MATLAB combined with directed rounding. The first method starts with a Bauer-Fike like perturbation theorem for general complex \(n\times n\) matrices \(A\). The second assumes \(A\) to be real and symmetric and is based on a Weyl type perturbation theorem which also induces a perturbation theorem for singular values of a real \(m\times n\) matrix. It is shown how these theorems are to be applied in order to get guaranteed lower and upper bounds for the items to be enclosed. MATLAB-like programs are given, their efficiency is tested using matrices of various size.