Reliable updated residuals in hybrid Bi-CG methods (Q1909599)

The authors propose some new algorithms (variants of Neumaier's strategy) in solving a linear system by an iterative method belonging to the class of Krylov-type methods (like Bi-CG, CGS, Bi-CGSTAB). This class of methods is characterized by the fact that the update for the residual vector is computed independently of the current approximation to the solution. In particular, the authors propose a more restrictive strategy for accumulating groups of updates for updating the residual and the approximation, and it is shown that this may improve the accuracy significantly and maintain the speed of convergence. This approach also allows more reliable stopping criteria. The updating conditions and the updating strategy discussed here lead to efficient methods and accurate residuals and are easily to implement. This strategy becomes clear by application to Bi-CG and to hybrid Bi-CG methods, like CGS, Bi-CGSTAB, and BiCG stab(I), but they seem to be applicable to many other iterative methods.