Try using a higher number of restart steps, say 500, -ksp_gmres_restart 500
Thanks. On Thu, May 29, 2014 at 7:00 PM, walter kou <[email protected]> wrote: > output from the -ksp_monitor_singular_value ? > > 250 KSP Residual norm 8.867989901517e-05 % max 1.372837955534e+05 min > 1.502965355599e+01 max/min 9.134195611495e+03 > > > > On Thu, May 29, 2014 at 5:54 PM, Vikram Garg <[email protected]> > wrote: > >> Try the following solver options: >> >> -pc_type lu -pc_factor_mat_solver_package superlu >> >> >> What was the output from the -ksp_monitor_singular_value ? >> >> >> Thanks. >> >> >> On Thu, May 29, 2014 at 6:52 PM, walter kou <[email protected]> >> wrote: >> >>> Hi Vikram, >>> >>> How to try using just lu or super lu? I am pretty ignorant in playing >>> with proper ksp options in the command line. >>> >>> Could you point out any introductory materials on this? >>> >>> Thanks so much. >>> >>> >>> >>> >>> On Wed, May 28, 2014 at 11:54 AM, Vikram Garg <[email protected]> >>> wrote: >>> >>>> Hey Walter, >>>> Have you tried using just lu or super lu ? You >>>> might also want to check and see whats the output for >>>> -ksp_monitor_singular_value and increase the gmres restart steps. >>>> >>>> Thanks. >>>> >>>> >>>> On Mon, May 26, 2014 at 1:57 PM, walter kou <[email protected]> >>>> wrote: >>>> >>>>> Hi all, >>>>> I ran a larger case: with elements = 200, and found for each >>>>> calculation in >>>>> the iteration, system.final_linear_residual() is about 0.5%. >>>>> >>>>> 1) Is the system.final_linear_residual() r = b -A X* (X* is the >>>>> solution) >>>>> ? right? >>>>> >>>>> 2) It seems final residual is too big, and the equation is not solved >>>>> well >>>>> (here |b| is about 1e-4). Does anyone have suggestion in playing with >>>>> solvers of Ax=b? >>>>> Here my case is on nonlinear elasticity, and A is almost symmetrical >>>>> positive definite (only components influenced by boundary conditions >>>>> will >>>>> break the symmetry). >>>>> >>>>> >>>>> Also, following the suggestion of Paul, I use -ksp_view and find my >>>>> solver >>>>> information is as below: >>>>> >>>>> KSP Object: 4 MPI processes >>>>> type: gmres >>>>> GMRES: restart=30, using Classical (unmodified) Gram-Schmidt >>>>> Orthogonalization with no iterative refinement >>>>> GMRES: happy breakdown tolerance 1e-30 >>>>> maximum iterations=250 >>>>> tolerances: relative=1e-08, absolute=1e-50, divergence=10000 >>>>> left preconditioning >>>>> using nonzero initial guess >>>>> using PRECONDITIONED norm type for convergence test >>>>> PC Object: 4 MPI processes >>>>> type: bjacobi >>>>> block Jacobi: number of blocks = 4 >>>>> Local solve is same for all blocks, in the following KSP and PC >>>>> objects: >>>>> KSP Object: (sub_) 1 MPI processes >>>>> type: preonly >>>>> maximum iterations=10000, initial guess is zero >>>>> tolerances: relative=1e-05, absolute=1e-50, divergence=10000 >>>>> left preconditioning >>>>> using NONE norm type for convergence test >>>>> PC Object: (sub_) 1 MPI processes >>>>> type: ilu >>>>> ILU: out-of-place factorization >>>>> 0 levels of fill >>>>> tolerance for zero pivot 2.22045e-14 >>>>> using diagonal shift to prevent zero pivot >>>>> matrix ordering: natural >>>>> factor fill ratio given 1, needed 1 >>>>> Factored matrix follows: >>>>> Matrix Object: 1 MPI processes >>>>> type: seqaij >>>>> rows=324, cols=324 >>>>> package used to perform factorization: petsc >>>>> total: nonzeros=16128, allocated nonzeros=16128 >>>>> total number of mallocs used during MatSetValues calls =0 >>>>> not using I-node routines >>>>> linear system matrix = precond matrix: >>>>> Matrix Object: () 1 MPI processes >>>>> type: seqaij >>>>> rows=324, cols=324 >>>>> total: nonzeros=16128, allocated nonzeros=19215 >>>>> total number of mallocs used during MatSetValues calls =0 >>>>> not using I-node routines >>>>> linear system matrix = precond matrix: >>>>> Matrix Object: () 4 MPI processes >>>>> type: mpiaij >>>>> rows=990, cols=990 >>>>> total: nonzeros=58590, allocated nonzeros=64512 >>>>> total number of mallocs used during MatSetValues calls =0 >>>>> not using I-node (on process 0) routines >>>>> >>>>> >>>>> >>>>> /********************************************************* >>>>> Thanks, >>>>> >>>>> Walter >>>>> >>>>> >>>>> On Thu, May 22, 2014 at 12:18 PM, Paul T. Bauman <[email protected]> >>>>> wrote: >>>>> >>>>> > >>>>> > >>>>> > >>>>> > On Thu, May 22, 2014 at 12:11 PM, walter kou <[email protected] >>>>> >wrote: >>>>> > >>>>> >> OK, but libMesh calls a library, defaulting to PETSc if it's >>>>> installed. >>>>> >> Which library are you using? >>>>> >> >>>>> >> PETSc-3.3 >>>>> >> >>>>> > >>>>> > I recommend checking out the PETSc documentation ( >>>>> > http://www.mcs.anl.gov/petsc/petsc-as/documentation/) and >>>>> tutorials. But >>>>> > you'll want to start with -ksp_view to get the parameters PETSc is >>>>> using. >>>>> > >>>>> >>>>> ------------------------------------------------------------------------------ >>>>> The best possible search technologies are now affordable for all >>>>> companies. >>>>> Download your FREE open source Enterprise Search Engine today! >>>>> Our experts will assist you in its installation for $59/mo, no >>>>> commitment. >>>>> Test it for FREE on our Cloud platform anytime! >>>>> >>>>> http://pubads.g.doubleclick.net/gampad/clk?id=145328191&iu=/4140/ostg.clktrk >>>>> _______________________________________________ >>>>> Libmesh-users mailing list >>>>> [email protected] >>>>> https://lists.sourceforge.net/lists/listinfo/libmesh-users >>>>> >>>> >>>> >>>> >>>> -- >>>> Vikram Garg >>>> Postdoctoral Associate >>>> Center for Computational Engineering >>>> Massachusetts Institute of Technology >>>> http://web.mit.edu/vikramvg/www/ >>>> >>>> http://www.runforindia.org/runners/vikramg >>>> >>> >>> >> >> >> -- >> Vikram Garg >> Postdoctoral Associate >> Center for Computational Engineering >> Massachusetts Institute of Technology >> http://web.mit.edu/vikramvg/www/ >> >> http://www.runforindia.org/runners/vikramg >> > > -- Vikram Garg Postdoctoral Associate Center for Computational Engineering Massachusetts Institute of Technology http://web.mit.edu/vikramvg/www/ http://www.runforindia.org/runners/vikramg ------------------------------------------------------------------------------ Time is money. Stop wasting it! Get your web API in 5 minutes. www.restlet.com/download http://p.sf.net/sfu/restlet _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
