Asymptotic quadratic convergence of the parallel block-Jacobi EVD algorithm with dynamic ordering for Hermitian matrices (Q1631200)

The paper is devoted to the convergence behavior of the parallel two-sided block-Jacobi algorithm for the computation of the eigenvalue decomposition (EVD) of a Hermitian matrix is proved. Precise asymptotic quadratic convergence (AQC) of the algorithm with general spectra (simple and multiple eigenvalues, clusters of eigenvalues) are proved. Three numerical examples that illustrate the (AQC) for well-separated eigenvalues and for clusters are provided. The results are important when multiprocessor calculus is used.