Stability analysis of the method of seminormal equations for linear least squares problems (Q1821506)

For the linear least squares problem to minimize \(\| Ax-b\|_ 2\), different methods are compared with respect to their stability properties. In particular the method of seminormal equations with one step iterative correction is discussed. It is shown that this combination yields (under mild assumptions) a method which is as accurate as the QR- method. The method is used to obtain updating of a sparse R-factor in the QR-factorization of A. The different methods are compared with respect to accuracy on differently conditioned problems.