A matrix pair of an almost diagonal skew-symmetric matrix and a symmetric positive definite matrix (Q1316177)

The authors study an almost diagonal pair \((A,B)\) with skew-symmetric \(A\) and symmetric positive definite \(B\). They also assume that the pair \((A,B)\) has multiple eigenvalues. Then they show that certain linear combinations composed of the elements of \(A\) and \(B\) are quadratically small with respect to the average off-diagonal elements of \(A\) and \(B\). The discovered property has a major influence on the asymptotic behavior of diagonalization methods for the generalized skew-symmetric eigenvalues problem. If \(B=I\), the identity matrix, the result describes a special structure of a skew-symmetric matrix almost in Murnaghan form. The authors then briefly discuss how this Paardekooper method for skew-symmetric matrices asymptotically behaves in the presence of multiple eigenvalues. The paper is organized in two sections. In Section 1 the authors derive general estimates for the pair \((A,B)\). In Section 2 they specialize the general result for the case \(B=I\). Then they use it to show the failure of the quadratic convergence of the Paardekooper method in the case of multiple eigenvalues. They provide examples which confirm their conclusion.