On Tue, 25 Mar 2014 22:46:55 +0100 Miroslav Kuchta <[email protected]> wrote:
> On 03/25/2014 10:37 PM, Jan Blechta wrote: > > I have tracked it down to SuperLU_dist, which was few months ago > > picked as new default by Garth. It works fine with MUMPS. Still, it > > would be good to know if it is faulty install on my side or > > bug/numerical problem in solver. So, can anyone try? > > > > Jan > > > > > > On Tue, 25 Mar 2014 22:25:40 +0100 > > Jan Blechta <[email protected]> wrote: > > > >> Running > >> > >> $ mpirun -n 2 python demo_mixed-poisson.py > >> > >> on PETSc produces obviously wrong results. Adding line > >> > >> print norm(w) > >> > >> results in 1.3060017846e+11 instead of correct 0.611356580181. > >> Strange is that with > >> > >> parameters['linear_algebra_backend'] = 'Epetra' > >> > >> it gives me correct results. So is it my installation or are DOLFIN > >> PETSc wrappers broken? Ok, but are you running SuperLU_dist? solve(a == L, w, bc, solver_parameters={'linear_solver': 'superlu_dist'}) Jan > >> > >> Jan > >> _______________________________________________ > >> fenics-support mailing list > >> [email protected] > >> http://fenicsproject.org/mailman/listinfo/fenics-support > > _______________________________________________ > > fenics-support mailing list > > [email protected] > > http://fenicsproject.org/mailman/listinfo/fenics-support > I get correct norm with PETSc and about a week old version of master. > > MK > _______________________________________________ > fenics-support mailing list > [email protected] > http://fenicsproject.org/mailman/listinfo/fenics-support _______________________________________________ fenics-support mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics-support
