View abstract

Session S20 - Applied Math and Computational Methods and Analysis across the Americas

No date set.

On the Adaptive Switching controller for avoiding stagnation on restarting $GMRES$

Juan Carlos Cabral

Scientific Computing and Applied Mathematics Group of the Polytechnic School/National University of, Paraguay   -   This email address is being protected from spambots. You need JavaScript enabled to view it.

The Restarted Generalized Minimal Residual method, denotes as GMRES($m$), is normally used for the solution of large, sparse, and nonsymmetric linear systems. In practice, it has the drawback of eventually presenting at certain re-starting cycles a stagnation or a slowdown rate of convergence. In this work, we are going to discuss strategies for avoiding stagnation and how a combination of them can exploit better their individual properties. The combination is implemented as a switching controller that changes the structure of the GMRES($m$) when stagnation is detected. The switching controller chooses conveniently from several techniques, how to augment the Krylov subspace for enriching it. Moreover, the controller varies the restarting parameter to modify the dimension of the Krylov subspace is needed. This strategy makes the adaptive switching controller competitive from the point of view of avoiding the stagnation and acceleration of the convergence respect to the number of iterations and the computational time. We are going to present computational experiments to show the advantages and the main issues raised from the perspective of the adaptive switching controller. For instance, when to perform the switching, what information is more important at each stage, and when to modify the restart parameter.

Joint work with Christian E. Schaerer (National University of Asuncion).

View abstract PDF