On the convergence of Stewart's QLP algorithm for approximating the SVD (Q1397933)

This paper tries to explain some of the success of Stewart's QLP algorithm [cf.\textit{G. W. Stewart}, SIAM J. Sci. Comput. 20, 1336--1348 (1999; Zbl 0939.65062)] in computing the singular values of a matrix. It discusses the tracking of singular values throughout the computation, and studies the convergence of the QLP. To be applicable, the theory developed requires a gap in the singular values, which does not yet explain everything. The paper closes with considering the asymptotic rate of convergence as well as some numerical examples.