Subspace Search Method for a Class of Least Squares Problem

Abstract

A subspace search method for solving a class of least squares problem is presented in the paper. The original problem is divided into many independent subproblems, and a search direction is obtained by solving each of the subproblems, as well as a new iterative point is determined by choosing a suitable steplength such that the value of residual norm is decreasing. The convergence result is also given. The numerical test is also shown for a special problem. \u00a0