Hi, Wolfgang! Just to be more precise. For the solution method using deal.II's CG I've used the following lines: SolverControl solver_control (dof_handler.n_dofs(), 1e-12); SolverCG<> cg (solver_control); cg.solve (system_matrix, solution, system_rhs, PreconditionIdentity());
For the PETSC MPI version: SolverControl solver_control (dof_handler.n_dofs(), 1e-12); PETScWrappers::SolverCG cg (solver_control, mpi_communicator); PETScWrappers::PreconditionBlockJacobi preconditioner(system_matrix); cg.solve (system_matrix, solution, system_rhs, preconditioner); Now I try to use deal's CG with PETSC MPI vectors as you suggested, I do the same code as for the first case, but it produce the error: /home/slawa/LinuxWork/Deal2Projects/HeatEq8-Explicit-Parallel/main.cpp|395|error: no matching function for call to ‘dealii::SolverCG<dealii::Vector<double> >::solve(dealii::PETScWrappers::MPI::SparseMatrix&, >dealii::PETScWrappers::MPI::Vector&, >dealii::PETScWrappers::MPI::Vector&)’| Question is - what should I change in this piece of code to make it run with deal's CG solver and PETSC MPI vectors and matrices? Regards, Slawa Saturday, June 27, 2009, 2:36, you wrote: WB> Slawa, >> Exactly! During first 5-10 steps its OK, but after that it starts >> to blow up and then after 2-3 more time steps its completely out of >> the game. And you are right - computed with deal's CG everything >> works fine, but with PETSC one this situation takes place. To get a >> working solution with PETSC I have to reduce time step 1 order of >> magnitude. >> >> One of the solutions - I've made a mistake somewhere in the code >> switching deal's vectors and solvers to the PETSC ones. But there >> are not that many places where one can make an error, furthermore >> with the smaller time step everything works fine. >> >> Another thing - for deal's CG I was using PreconditionIdentity(), >> for PETSC one: PreconditionBlockJacobi. WB> I am out of further suggestions on how to track down this problem :-( One WB> other thing you may try is to use PETSc vectors and matrices, and use them WB> with the deal.II SolverCG class. This way you can at least evaluate WB> whether it's the preconditioner or not. Other than that, I don't currently WB> have any further ideas how to solve this riddle... WB> Best WB> W. WB> ------------------------------------------------------------------------- WB> Wolfgang Bangerth email: [email protected] WB> www: WB> http://www.math.tamu.edu/~bangerth/ -- Best regards, 09.07.2009 Svyatoslav Gladkov, Kharkov, UKRAINE 13:11 _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
