Harmonic and refined extraction methods for the singular value problem, with applications in least squares problems (Q2566637)

The author presents the accurate approximation of the minimal singular triple (the smallest singular value and corresponding left and right singular vector) of a large and/or sparse matrix. He suggests three harmonic and refined approaches for the above problem with some specific variants as the finding the largest or interior singular triples and the applications of the determined smallest singular values in the approximations of the least squares solutions and in truncated singular value decomposition. As the conclusion of this paper (34 pages) with many comparative numerical experiments, the author proposes to use for the smallest and interior singular triples the refined extraction approach (the minimal left and right singular vector of matrix \(A\) minimize \(\| A^{*}x\| \) and \(\| Ay\| \), respectively) with the double-harmonic extraction method (harmonic Rayleigh-Ritz extraction introduced by \textit{C. C. Paige, B. N. Parlett} and \textit{H. A. Van der Vorst} [Numer. Linear Algebra Appl. 2(2), 115--133 (1995; Zbl 0831.65036)] as an alternative. For the largest singular triples is recommended standard extraction (Galerkin subspace method) because it is the cheapest.