Bug report: Trillinos-ilut and blockjacobi diverged but NS converged

[HadiWiputra]

Description

I am running the attached script for with different preconditioners.

But the result using Trillinos diverged, while I got a reasonable output using NS. I am not sure if it is an RCR boundary problem. I did test them with neumann conditions, and they seem to run okay.
RCR potential bug.zip

Reproduction

Run the script svFSI_rigid_NS.xml and svFSI_rigid_trillinos.xml to get the two outputs found in output_NS.out and output_trillinos.out respectively

Expected behavior

Both preconditioners should work okay since there were 3D pipe test that uses RCR and trllinos-blockjacobi

Additional context

No response

Code of Conduct

• I agree to follow this project's Code of Conduct and Contributing Guidelines

[HadiWiputra]

Hi All, an update on the script. There is a missing definition of the 0 velocity dirichlet boundary at the lumen_wall, making it fully permeable to flow. I added that line into the script, but the same problem as described above still persist.

I've also benchmarked the output from svFSI+ using NS vs svFSI fortran, they matched pretty well. I am uploading the updated zip file.
updated results.zip

Regards,
Hadi

[dcodoni]

Hi @HadiWiputra, in general we recommend to use the bipartition method (NS in svFSIplus) to run fluid problems, since it is specifically designed for fluid problems. In your case, we believe your application is more demanding from the computational point of view with respect to the pipe case, and thus GMRES (through Trilinos) is inappropriate for your problem and causing the convergence issues you're seeing. So our suggestion is to just keep using the NS method.

[dcodoni]

Hi @HadiWiputra did you try by any chance the GMRES with ILU preconditioner?