A new iterative scheme that uses the residual vector as search direction is proposed and analyzed for solving large-scale nonsymmetric linear systems, whose matrix has a positive (or negative) definite symmetric part. It is closely related to Richardson's method, although the stepsize and some other new features are inspired by the success of recently proposed residual methods for nonlinear systems. Numerical experiments are included to show that, without preconditioning, the proposed scheme outperforms some recently proposed variations on Richardson's method, and competes with well-known and well-established Krylov subspace methods: GMRES and BiCGSTAB. Our computational experiments also show that, in the presence of suitable preconditioning strategies, residual iterative methods can be competitive, and sometimes advantageous, when compared with Krylov subspace methods.
linear systems; Richardson's method; Krylov subspace methods; spectral gradient method
