A note on the SOR and CG methods for large least squares problems (Q1124276)

The authors consider the successive overrelaxation (SOR) and the conjugate gradient (CG) methods for the least squares problem of minimizing the Euclidean norm of Ax-b, where A is an \(m\times n\) matrix with \(rank(A)=n.\) They show that the SOR and CG methods are preferable, with respect to storage and multiplication requirements, to the 2-block SOR given by \textit{T. L. Markham}, \textit{M. Neumann}, and \textit{R. J. Plemmons} [ibid. 69, 155-167 (1985; Zbl 0576.65026)] and the CG algorithm of \textit{D. M. Young}, \textit{L. J. Hayes}, and \textit{K. C. Jea} [CNA-Report No.162, Center for Numerical Analysis, University of Texas, Austin] when the problem is of large scale.