Author Topic: PETSc solvers  (Read 4018 times)

msodan

  • New Member
  • *
  • Posts: 1
PETSc solvers
« on: August 12, 2022, 08:03:20 AM »
Dear All,
I am dealing with a nonsymmetric sparse matrix (4000 elements - 32 000 degrees of freedom).
I have currently been using FEAP 8.6 and standard FEAP solver (UTAN including option OPTI), which seems to be too slow.

I compiled FEAP with PETSC just to test other solvers.

Can someone recommend me better solver for this type of matrix from PETSC library?

P.S.
I am only using PETSc for solvers so there is no any form or parallelization involved.

Thank you.

Prof. R.L. Taylor

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 2647
Re: PETSc solvers
« Reply #1 on: August 12, 2022, 12:22:04 PM »
If you have the Intel compiler you can use Pardiso -- which is much better, especially if the problem is 3-d.

Other options are umfpack or superlu


Prof. R.L. Taylor

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 2647
Re: PETSc solvers
« Reply #2 on: August 12, 2022, 01:11:33 PM »
There are 'umacr*.f' files for the solvers at the FEAP projects page.

Scroll down to Manuals for the link

Prof. S. Govindjee

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 1160
Re: PETSc solvers
« Reply #3 on: August 12, 2022, 01:59:22 PM »
For a serial solver with petsc, I usually try mumps and superlu.  I usually find superlu better of the two.
The pardisio option is quite good if you have the intel compiler.  It is available in FEAP if you use the mkl library
from intel.  I'll try to post the files and instructions if you need them.