[petsc-dev] Is the ASM preconditioner algebraic or geometric in the context of DMPlex (unstructured) ?

[Mark Adams]

On Sun, May 8, 2022 at 12:44 AM markwinpe <markwinpe at 163.com> wrote:

> Dear PETSc’s developers,
>
>       In the context of unstructured applications developed by PETSc’s
> DMPlex, if the ASM is used for preconditioning, is it algebraic (based on
> matrix graph) or geometric (based on Metis partition)?
>

Algebraic. DMPlex is a topology thing and I don't know if it helps ASM.

ASM will have one subdomain per process by default, it does not try to
create subdomains for you.

You can give AMS domains that you can create with a partitioning (Metis the
common one to use). see
https://petsc.org/main/docs/manualpages/MatOrderings/MatPartitioningCreate

and

https://petsc.org/main/docs/manualpages/PC/PCASMSetTotalSubdomains/

Mark

    When the ASM preconditioner is associated with a specific DM object, I
> found there is  only the implementation of DMCreateDomainDecomposition
> <https://www.petsc.org/release/src/dm/impls/da/dadd.c.html#DMCreateDomainDecomposition_DA> for
> DA in PCSetUp_ASM, and there is no implementation for DMPlex. So I guess
> the ASM is algebraic for DMPlex, is that right ?
>
>
>
> Best,
>
>
> Wenpeng
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20220508/71d77032/attachment.html>