A quadratically convergent parallel Jacobi process for diagonally dominant matrices with nondistinct eigenvalues (Q756365)

A Jacobi-like algorithm for finding eigenvalues of non-Hermitian matrices is discussed. Similarity transformations, which are not necessarily unitary, are applied to successively eliminate off-diagonal elements. Elimination of certain elements is not performed when a multiple eigenvalue is detected. Quadratic convergence occurs, even in the presence of multiple eigenvalues, but the proof of this result is local in nature, that is, it depends on the starting matrix being nearly diagonal to begin with. As for other Jacobi-like methods, this algorithm can be implemented on a parallel computer in a straightforward way.