A unitary Hessenberg QR-based algorithm via semiseparable matrices

摘要

In this paper, we present a novel method for solving the unitary Hessenberg eigenvalue problem. In the first phase, an algorithm is designed to transform the unitary matrix into a diagonal-plus-semiseparable form. Then we rely on our earlier adaptation of the QR algorithm to solve the dpss eigenvalue problem in a fast and robust way. Exploiting the structure of the problem enables us to yield a quadratic time using a linear memory space. Nonetheless the algorithm remains robust and converges as fast as the customary QR algorithm. Numerical experiments confirm the effectiveness and the robustness of our approach.