Convergence analysis of the one-step iterative Krylov subspace methods (Q1049378)

For solving iteratively a linear operator equation in Banach space by Richardson's method and by the GMRES (Generalized Minimal Residual) method, the author combines operator spectral theory with optimization techniques to obtain faster convergence. There are different estimates for the convergence rate, and the iterations provided by Richardson's method and by the GMRES method may have a completely chaotic character (as the numerical experiments show). A detailed, exhaustive study of these iterations offers the possibility to obtain more sophisticated procedures having faster convergence. For a convex compact subset of complex numbers, it is proved the existence and the uniqueness of an optimal circle, containing the subset, not containing the origin, and with least ratio. Such a circle has been found for segments and sectors of a circle, and also for rectangles (any convex compact, not containing a fixed point, lies in a sector or in a segment). By equivalent renorming of the underlying space, the author proposes a procedure to accelerate the rate of convergence. One numerical example is solved by the proposed procedure.