HYPRE and Trilinos

[delcmo]

Hello,

I would like to get some more information on the compatibility of HYPRE with Trilinos packages. Is there any documentation I should review in order to get an understanding of the current capability? Any examples as well?

Thanks,

Marco

[cgcgcg]

We primarily provide algebraic multigrid in the form of the MueLu package. Hypre is available as a solver via Ifpack2, but there are several inefficiencies in the interface.

[delcmo]

Christian,

Thanks for taking the time to responding.

Is there any documentation about the current interface capabilities I can access? We are developing a Trilinos-based package and are collaborating with the HYPRE team for solvers. We would be interested in adding new HYPRE capabilities to our code and trying to figure out what would be the best approach?

Thanks,

Marco

[cgcgcg]

Everything about the interface is in the code. I cannot stress enough that relying on this interface might be bad. We do not test it, and it's been broken in the past for long stretches of time. What specific solver features do you need that Trilinos AMG does not provide?

[delcmo]

We have tried the following solvers: - Incomplete factorization (ILU), - Factorized Sparse approximate inverse (FSAI) - Algebraic multigrid (BoomerAMG) - Multigrid reduction (MGR) The MRG option seems to be working best with the equations we are trying to solve. Marco

…

On Fri, May 17, 2024 at 2:08 PM Christian Glusa ***@***.***> wrote: Everything about the interface is in the code. I cannot stress enough that relying on this interface might be bad. We do not test it, and it's been broken in the past for long stretches of time. What specific solver features do you need that Trilinos AMG does not provide? — Reply to this email directly, view it on GitHub <#13016 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABD4GCUO3CHYGI5PJJ4O7YTZCZBRXAVCNFSM6AAAAABH2OXQNWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMJYGE2DEMJRGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- Marc-Olivier Delchini

[cgcgcg]

Incomplete factorizations are available from Ifpack2. Plain AMG is provided in MueLu, and there are also variants for multiphysics.

[cgcgcg]

What type of problem are you trying to solve, @delcmo ?

[delcmo]

Sorry for the late reply but I got sidetracked. This task is part of a larger project that includes the HYPRE team. Using HYPRE solvers is one of the milestones. I have been using different solvers for Epetra and Tpetra: Ifpack, Belos, and Pardiso. I get good scaling in 2D but my preliminary work with 3D meshes has been disappointing. It is hard for me to find documentation on solvers and the associated parameters and I often rely on a trial-and-error approach to optimize the solver parameters. I am interesting in learning more about the Trilinos solver and their performance. Any documentation you could provide would be helpful. Thanks, Marco

…

On Thu, Jun 6, 2024 at 7:19 PM Christian Glusa ***@***.***> wrote: What type of problem are you trying to solve, @delcmo <https://github.com/delcmo> ? — Reply to this email directly, view it on GitHub <#13016 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABD4GCUNAULUFDCSQMXRY7TZGDVBVAVCNFSM6AAAAABH2OXQNWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNJTGU2DOOJXGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

-- Marc-Olivier Delchini