Adaptive model trust region methods for generalized eigenvalue problems

Abstract

Computing a few eigenpairs of large-scale matrices is a significant problem in science and engineering applications and a very active area of research. In this paper, two methods that compute extreme eigenpairs of positive-definite matrix pencils are combined into a hybrid scheme that inherits the advantages of both constituents. The hybrid algorithm is developed and analyzed in the framework of model-based methods for trace minimization. © Springer-Verlag Berlin Heidelberg 2005.