lhy11009

Secondly, I was thinking of the viscosity contrast in the model. Previously, I have already limited the viscosity in [1e21, 1e24] by setting the " Minimum viscosity", and the "Maximum...

Thirdly, A suggestion from John is to test "Use full A block as preconditioner" = true Which doesn't change the results, or the reported "final residual" in the "BlockSchurGMGPreconditioner::vmult".

Anyway, I'll try to get a simplified version of this. Here is the prm I have for now:

[test_ThD_gmg_mv1e21_peloton_case.prm.txt](https://github.com/geodynamics/aspect/files/9602831/test_ThD_gmg_mv1e21_peloton_case.prm.txt)

Sorry, I forget to mention. These are the versions. -- This is ASPECT, the Advanced Solver for Problems in Earth's ConvecTion. -- . version 2.5.0-pre (master_TwoD, 5a5f21131) -- . using...

Oh, right, let me check the previous cases

----------------------------------------------------------------------------- -- This is ASPECT, the Advanced Solver for Problems in Earth's ConvecTion. -- . version 2.4.0-pre (main, 875a17cf8) -- . using deal.II 9.3.1 -- . with 32 bit indices...

Sorry, sometimes I forget to include the full information.

Also, I suspect it's some issue with MPI. I have some clues here from the affinity test. Here, I ran the affinity test model I got from Rene with different...

Here is the test with Stampede2 "skx-normal" partition. This partition has 48 CPUs per core.  These are some differences in the header of the log file now that I...