A projection method for general form linear least-squares problems

Abstract

One of the possible approaches for the solution of underdetermined linear least-squares problems in general form, for a chosen regularization operator L, projects the problem in the null space of L and in its orthogonal complement. In this paper, we show that the projected problem cannot be solved by the generalized singular value decomposition, and propose some approaches to overcome this issue. Numerical experiments ascertain the stability of the new procedures.