Two-level least squares methods in Krylov subspaces (Q1662026)

The author presents two-level methods for accelerating the convergence of restarted iterations based on the least squares method (LSM) for decreasing the Euclidean norm of the residual, for large, sparse and ill-conditioned linear systems. In this respect, he also considers applications of less expensive conjugate residual (CR) algorithms and the Chebyshev acceleration. Numerical experiments are presented on a set of model Dirichlet problems for the convection-diffusion equation.