Computational experience with numerical methods for nonnegative least-squares problems

Abstract

We discuss the solution of large-scale box-constrained linear least-squares problems by two recent affine-scaling methods: a cyclic Barzilai–Borwein strategy and an Inexact Newton-like method where a preconditioning technique allows for an efficient computation of the steps. A robust globally and fast locally convergent method based on the combination of the two procedures is presented along with extensive numerical results. Copyright © 2010 John Wiley & Sons, Ltd.