Given the nullspace, I'd modify your options slightly 1) -fieldsplit_pressure_lsc_pc_type cholesky or for something iterative 2) -fieldsplit_pressure_lsc_pc_type icc -fieldsplit_pressure_lsc_ksp_max_it 10
Kick out the mumps option. You should not have to perform a direct solve on this system to obtain good results. Cheers, Dave On 4 May 2011 15:34, Jed Brown <jed at 59a2.org> wrote: > On Wed, May 4, 2011 at 15:29, <domenico.borzacchiello at univ-st-etienne.fr> > wrote: >> >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_pc_type lsc \ >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_lsc_ksp_type fgmres \ >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_lsc_ksp_max_it 100 \ >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_lsc_ksp_constant_null_space \ >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_lsc_pc_type lu \ >> ? ? ? ? ? ? ? ?-fieldsplit_pressure_lsc_pc_factor_mat_solver_package mumps > > Perhaps this direct solve is the problem? That operator is expected to be > singular due to the constant null space so a direct solve is expected to > produce nonsense.
