- About distort_random "I suspect that when you call distort_random(), the mesh is different whether you run on one or four processes. Have you confirmed that the meshes are exactly identical?" You were right about this one. After importing a GMSH unstructured mesh the results are perfect and independent of the MPI processes used.
- About step 18 I plotted the stress in Paraview along a line for both MPI runs 1 and 3. The results matched. The MatrixTools::apply_boundary_values(boundary_values, system_matrix, tmp, system_rhs, false) function appears to be the culprit . With MPI 1 rhs.l2() before calling apply_boundary_values 3032.65 rhs.l2() after calling apply_boundary_values 1.86145e+10 with MPI 3 rhs.l2() before calling apply_boundary_values 3032.65 rhs.l2() after calling apply_boundary_values 3.67161e+10 Is it because apply_boundary_values is doing something special to ghost nodes? Best, Abbas On Friday, June 16, 2023 at 11:44:43 PM UTC+2 Wolfgang Bangerth wrote: > On 6/12/23 02:57, Abbas Ballout wrote: > > ** > > > > I am running step 18. > > > > This is the output I getting for a single quasi-tatic step with mpirun 1 > : > > > > /Cycle 0: > > Number of active cells: 3712 (by partition: 3712) > > Number of degrees of freedom: 17226 (by partition: 17226) > > Assembling system... norm of rhs is 1.88062e+10 > > Solver converged in 103 iterations. > > Updating quadrature point data... > > Cycle 1: > > Number of active cells: 12805 (by partition: 12805) > > Number of degrees of freedom: 51708 (by partition: 51708) > > Assembling system... norm of rhs is 1.86145e+10 > > Solver converged in 120 iterations. > > Updating quadrature point data... > > Moving mesh../. > > > > And this is the output I get when mpirun 3: > > > > /Timestep 1 at time 1 > > Cycle 0: > > Number of active cells: 3712 (by partition: 1360+1286+1066) > > Number of degrees of freedom: 17226 (by partition: 6651+5922+4653) > > Assembling system... norm of rhs is 1.88062e+10 > > Solver converged in 131 iterations. > > Updating quadrature point data... > > Cycle 1: > > Number of active cells: 12805 (by partition: 4565+4425+3815) > > Number of degrees of freedom: 51708 (by partition: 19983+17250+14475) > > Assembling system... norm of rhs is 3.67161e+10 > > Solver converged in 126 iterations. > > Updating quadrature point data... > > Moving mesh.../ > > > > The L2 norm in cycle 1 is different between runs with mpi 1 and mpi 3. > Is this > > normal? > > This is an interesting observation. It strikes me as a bug. > > Have you compared the solution you get after running either 1 or 3 MPI > processes? If you visualize the result, or compute some other indicator of > the > solution, in which ways do they differ? > > Best > W. > > -- > ------------------------------------------------------------------------ > Wolfgang Bangerth email: bang...@colostate.edu > www: http://www.math.colostate.edu/~bangerth/ > > > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/3b758adf-f6b4-43e9-9dfb-27d94f3d8040n%40googlegroups.com.
