No influence of the Petsc options on the solution process

[Mari]

Hello,

I am using FEAP8.5 and Petsc 3.7.7. It seems that a parallelized calculation works with the command

Code: [Select]

mpirun -n 2 $FEAPHOME8_5/parfeap/feap -ksp_type cg -pc_type jacobi .
However, a variation of the (PETsc-) options has no influence on the solution process. For example, adding -ksp_monitor or -ksp_view does not create another file with the corresponding output.

If I still want to use a direct solver with e.g.

Code: [Select]

-ksp_monitor_short -ksp_type preonly -pc_type lu -pc_factor_mat_solver_package mumps  , this is apparently ignored.

Even if I don't pass on any options, a calculation is performed (probably with an iterative solver).


For your information: I configured PETsc with

Code: [Select]

./configure --with-cc=mpicc --with-cxx=mpicxx --with-fc=mpif90 --with-blas-lapack-dir=$MKLROOT --download-parmetis=1 --download-metis=1 --with-debugging=0 --download-mumps=1 --download-scalapack=1
Furthermore I utilize intel/17.0 and intelmpi/2019.5.281.

[Prof. S. Govindjee]

That is strange.  try adding -options_left at the end of your option list to see if they are being ignored or just not read.
also -ksp_monitor etc. shows to the screen/output file and not to a separate file.

If you look at the output to the screen or the output file you should also be able to see what solver is being used.

[Mari]

Adding -options_left does not help either. It seems that the options (-ksp_type, ...) are not passed on to PETsc at all.

[Prof. S. Govindjee]

The only thing I can think of is that mpirun is not functioning correctly  but this seems unlikely.  What happens if  you execute your runs using the make file in the parfeap directory?  so something like 'make feaprun' or 'make feaprun-mumps'?

[Mari]

You were right. I did something wrong when I installed PETsc and par feap, so the command didn't work properly. However, I cannot say exactly what I have done wrong  . But now it works  .