Iterative methods for least-square problems based on proper splittings. (Q1408395)

The paper is devoted to the linear least-squares problem. A linear system \(Ax=b\) is considered, where \(A\) is a large and sparse matrix. If the system is inconsistent, then the least-squares solution can be obtained using the Moore-Penrose generalized inverse of the matrix \(A\). In this case some problems arise because the matrix \(A'A\) is ill-conditioned and influenced greatly by roundoff error. To avoid this problem some iterative methods are used. They are similar to the iterative methods for nonsingular linear systems. It is very usual to use straightforward application of Cholesky or QR factorization, but it will lead to catastrophic fill in factor R. With the purpose to introduce an alternative solution to this problem, two iterative methods are introduced, to obtain the least-squares solution of the linear system \(Ax=b\). The authors handle the problem by iterative methods based on proper splittings. The first one is a two-stage method and the second one is a parallel iterative method. The convergence to the least-squares solution for any initial vector \(x\) is established, for the sequential two-stage iterative method and for the parallel stationary iterative method.