The QR iteration method for Hermitian quasiseparable matrices of an arbitrary order (Q2484401)

Based on earlier works of the first two authors on special classes of structured matrices the eigenvalue problem for Hermitian quasiseparable matrices is treated. It is well known that the \(QR\) iteration can be implemented very efficiently if it is applied to reduced matrices, e.g. to Hermitian tridiagonal matrices after reduction of Hermitian matrices. The authors show that the efficiency of the \(QR\) iteration can be carried over to Hermitian quasiseparable matrices. Quasiseparable matrices can be represented by generators which are computed applying a modifiction of the Dewilde-van der Veen method [cf. \textit{Y. Eidelman} and \textit{I. Gohberg}, Linear Algebra Appl. 343--344, 419--450 (2002; Zbl 1010.65013)]. The Hermitian quasiseparable structure is inherited under \(QR\) iteration. Proofs and complexity analysis are given. The last section contains numerical experiments and a description of the shift strategy and of the estimation of the accuracy. The numerical tests show that the total number of iterations to compute the eigenvalues of an \((n,n)\) Hermitian quasiseparable matrix is about \(3n\).