Synergy of Adaptive Coarse Space and Krylov Subspace recycling for the BDDC method
Please login to view abstract download link
Means of acceleration of iterative methods for sequences of linear systems have been extensively studied in literature. A widely used approach is recycling the subspace within a Krylov method combined with deflation. Another approach is based on improving the preconditioner. In domain decomposition methods, adaptive selection of coarse space is the state of the art leading to powerful preconditioners. We compare these two approaches and study their combination for unsteady incompressible flow problems governed by the Navier-Stokes equations. These are solved by the pressure-correction scheme in connection with the finite element method. This approach leads to sequences of linear systems over the time steps. Our particular interest is the Poisson problem of pressure. Results for the problem of flow behind the sphere for Reynolds number 300 are presented. We demonstrate that by using these approaches we are able to save about half of the computational time.
