Computation of optimal backward perturbation bounds for large sparse linear least squares problems (Q5960913)

The backward perturbation bound is estimated for large sparse linear least squares problems. The algorithm is based on Lanczos bidiagonalization and requires \(O((m+n)l)\) operations where the matrix is \(m\times n\) and \(l\ll\min(m,n)\). Illustrative examples are selected from the Harwell-Boeing collection of test matrices.