On Wed, May 31, 2017 at 10:18 AM, Salazar De Troya, Miguel < salazardet...@llnl.gov> wrote:
> Hi > > I am solving an elasticity problem with the following iterative solver > options: > > -ksp_monitor_true_residual > -ksp_converged_reason > -ksp_type cg > -log_view > -mg_levels_esteig_ksp_type cg > -mg_levels_ksp_chebyshev_esteig_steps 50 > -mg_levels_ksp_type chebyshev > -mg_levels_pc_type jacobi > -pc_type gamg > -snes_linesearch_type basic > -snes_atol 1e-6 > -ksp_atol 1e-7 > -ksp_rtol 1e-9 > -ksp_norm_type unpreconditioned > > I am passing the rigid body bodes to the KSP solver. There is a noticeable > increase in the number of iterations when the size ratio between biggest > and smallest element increases. This problem is also solving for two > materials with a high contrast in their properties, so this could have an > effect, but the iterations don’t increase as much when the mesh is uniform. > I wonder if there is some information about the mesh that could be passed > to the preconditioner to improve its performance. > I'm not sure exactly what type of multigrid preconditioner your mg_ options are setting up, but you might also want to try/compare your results with the options "-pc_type hypre -pc_hypre_type boomeramg" as well. Since you mention a ratio between biggest and smallest elements, are you referring to an adaptively refined grid, or just a continuously graded mesh? I don't know too much about MG, but there may be some special approach needed for meshes with a large hmax/hmin ratio. -- John ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
