Affordable Access

A multishift, multipole rational qz method with aggressive early deflation

Authors
  • STEEL, T
  • CAMPS, D
  • MEERBERGEN, K
  • VANDEBRIL, R
Publication Date
Jan 01, 2021
Source
eScholarship - University of California
Keywords
License
Unknown
External links

Abstract

In the article "A Rational QZ Method"" by D. Camps, K. Meerbergen, and R. Vandebril [SIAM J. Matrix Anal. Appl., 40 (2019), pp. 943-972], we introduced rational QZ (RQZ) methods. Our theoretical examinations revealed that the convergence of the RQZ method is governed by rational subspace iteration, thereby generalizing the classical QZ method, whose convergence relies on polynomial subspace iteration. Moreover the RQZ method operates on a pencil more general than Hessenberg-upper triangular, namely, a Hessenberg pencil, which is a pencil consisting of two Hessenberg matrices. However, the RQZ method can only be made competitive to advanced QZ implementations by using crucial add-ons such as small bulge multishift sweeps, aggressive early deflation, and optimal packing. In this paper we develop these techniques for the RQZ method. In the numerical experiments we compare the results with state-of-the-art routines for the generalized eigenvalue problem and show that the presented method is competitive in terms of speed and accuracy.

Report this publication

Statistics

Seen <100 times