GAMG will coarsen the problem until it is small and fast to solve with a direct solver (LU). You can use preonly if you have a perfect preconditioner.
On Mon, Oct 2, 2017 at 9:08 AM, Matthew Knepley <knep...@gmail.com> wrote: > On Mon, Oct 2, 2017 at 8:30 AM, Wenbo Zhao <zhaowenbo.n...@gmail.com> > wrote: > >> Matt >> >> Because I am not clear about what will happen using 'preonly' for large >> scale problem. >> > > The size of the problem has nothing to do with 'preonly'. All it means is > to apply a preconditioner without a Krylov solver. > > >> It seems to use a direct solver from below, >> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/ >> KSP/KSPPREONLY.html >> > > However, I still cannot understand why you would change the default? > > Matt > > >> >> Thanks! >> Wenbo >> >> On Mon, Oct 2, 2017 at 5:09 PM, Matthew Knepley <knep...@gmail.com> >> wrote: >> >>> On Sun, Oct 1, 2017 at 9:53 PM, Wenbo Zhao <zhaowenbo.n...@gmail.com> >>> wrote: >>> >>>> Matt, >>>> Thanks for your reply. >>>> It DOES make no sense for this problem. >>>> But I am not clear about the 'preonly' option. Which solver is used in >>>> preonly? I wonder if 'preonly' is suitable for large scale problem such as >>>> 400,000,000 unknowns. >>>> So I tried 'gmres' option and found these error messages. >>>> >>> >>> I mean, why are you setting this at all. Just do not set the coarse >>> solver. The default should work fine. >>> >>> Thanks, >>> >>> Matt >>> >>> >>>> Could you give me some suggestions? >>>> >>>> Thanks. >>>> >>>> Wenbo >>>> >>>> >>>> On Mon, Oct 2, 2017 at 12:34 AM, Matthew Knepley <knep...@gmail.com> >>>> wrote: >>>> >>>>> On Sun, Oct 1, 2017 at 6:49 AM, Wenbo Zhao <zhaowenbo.n...@gmail.com> >>>>> wrote: >>>>> >>>>>> Hi, >>>>>> >>>>>> I met some questions when I use PETSC/SLEPC to solve two-group >>>>>> neutron diffusion equations with finite difference method. The grid is >>>>>> 3*3*3, when DOF on each points is 2. So the matrix size is 54*54. >>>>>> It is generalized eigenvalue problem Ax=\lamda Bx, where B is >>>>>> diagonally dominant matrix but not symmetry. >>>>>> EPS is set as below, >>>>>> ierr = EPSSetProblemType(eps,EPS_GNHEP);CHKERRQ(ierr);¬ >>>>>> ierr = EPSSetWhichEigenpairs(eps,EPS_LARGEST_REAL);CHKERRQ(ierr);¬ >>>>>> >>>>>> Krylovschur is used as eps sovler. GAMG is used as PC. >>>>>> I tried agg_nsmooths and mg_coarse_ksp_type. Only non-smooths and >>>>>> preonly is OK. >>>>>> >>>>> >>>>> Why are you setting the coarse solver. This makes no sense. >>>>> >>>>> Thanks, >>>>> >>>>> Matt >>>>> >>>>> >>>>>> >>>>>> Test 1 >>>>>> $ make NCORE=1 runkr_nonsmooth >>>>>> mpirun -n 1 ./step-41 \ >>>>>> -st_ksp_type gmres \ >>>>>> -st_pc_type gamg -st_pc_gamg_type agg -st_pc_gamg_agg_nsmooths 0 \ >>>>>> -st_ksp_view -mata AMAT.dat -matb BMAT.dat \ >>>>>> -st_mg_coarse_ksp_type preonly -st_mg_coarse_ksp_monitor \ >>>>>> -eps_nev 1 -eps_ncv 10 -eps_monitor -log_view > log_nonsmooth >>>>>> 2>&1 >>>>>> >>>>>> Test 2 >>>>>> $ make NCORE=1 runkr_smooth >>>>>> mpirun -n 1 ./step-41 \ >>>>>> -st_ksp_type gmres \ >>>>>> -st_pc_type gamg -st_pc_gamg_type agg -st_pc_gamg_agg_nsmooths 1 \ >>>>>> -st_ksp_view -mata AMAT.dat -matb BMAT.dat \ >>>>>> -st_mg_coarse_ksp_type preonly -st_mg_coarse_ksp_monitor \ >>>>>> -eps_nev 1 -eps_ncv 10 -eps_monitor -log_view > log_smooth 2>&1 >>>>>> makefile:43: recipe for target 'runkr_smooth' failed >>>>>> make: *** [runkr_smooth] Error 91 >>>>>> >>>>>> Test 3 >>>>>> $ make NCORE=1 runkr_gmres >>>>>> mpirun -n 1 ./step-41 \ >>>>>> -st_ksp_type gmres \ >>>>>> -st_pc_type gamg -st_pc_gamg_type agg -st_pc_gamg_agg_nsmooths 0 \ >>>>>> -st_ksp_view -mata AMAT.dat -matb BMAT.dat \ >>>>>> -st_mg_coarse_ksp_type gmres -st_mg_coarse_ksp_monitor >>>>>> -st_mg_coarse_ksp_rtol 1.0e-6 \ >>>>>> -eps_nev 1 -eps_ncv 10 -eps_monitor -log_view > log_gmres 2>&1 >>>>>> makefile:59: recipe for target 'runkr_gmres' failed >>>>>> make: *** [runkr_gmres] Error 91 >>>>>> >>>>>> Log files were attched. >>>>>> The matrix file were also attched as AMAT.dat and BMAT.dat. >>>>>> >>>>>> Is it correct? Or something wrong with my code or commad-line? >>>>>> >>>>>> Thanks! >>>>>> >>>>>> Wenbo >>>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> What most experimenters take for granted before they begin their >>>>> experiments is infinitely more interesting than any results to which their >>>>> experiments lead. >>>>> -- Norbert Wiener >>>>> >>>>> https://www.cse.buffalo.edu/~knepley/ >>>>> <http://www.caam.rice.edu/~mk51/> >>>>> >>>> >>>> >>> >>> >>> -- >>> What most experimenters take for granted before they begin their >>> experiments is infinitely more interesting than any results to which their >>> experiments lead. >>> -- Norbert Wiener >>> >>> https://www.cse.buffalo.edu/~knepley/ <http://www.caam.rice.edu/~mk51/> >>> >> >> > > > -- > What most experimenters take for granted before they begin their > experiments is infinitely more interesting than any results to which their > experiments lead. > -- Norbert Wiener > > https://www.cse.buffalo.edu/~knepley/ <http://www.caam.rice.edu/~mk51/> >