[petsc-users] SNESFAS for an unstructured mesh

[Vijay S. Mahadevan]

>> Thanks for your response. Is DMMoad is other special kind of data structure in Petsc?

DMMoab is a concrete implementation of the DM object, which uses MOAB
mesh database underneath.

> Would it be possible to use FAS without using petsc DM/DMPlex by adding extra code?

If you have your own mesh handling code, as Lawrence suggests, use a
DMShell object and create appropriate wrapper routines that utilizes
your home grown API underneath. There is a linear solver example with
DMShell, which should help you get started. Instead of the DA object
reference used in the shell implementation, you can replace it with
your own structures and context.

http://www.mcs.anl.gov/petsc/petsc-current/src/ksp/ksp/examples/tutorials/ex65.c.html

Vijay

On Wed, Jun 29, 2016 at 1:58 PM, Lawrence Mitchell
<lawrence.mitchell at imperial.ac.uk> wrote:
>
>
>> On 29 Jun 2016, at 19:47, Khai Pham <khaipham at utexas.edu> wrote:
>>
>> Thanks for your response. Is DMMoad is other special kind of data structure in Petsc? I'd like to use the petsc FAS for the finite element code in our group and all the data structure for mesh topology is all set. Thus I try to avoid all petsc data structure. As far as I know the FAS will construct the coarse solution based on DM for structured grid and DMPlex for unstructured grid as in ex12. Would it be possible to use FAS without using petsc DM/DMPlex by adding extra code? How much effort would you estimate? Thanks !
>
> You can use a DMShell and just implement the needed grid transfer operations using your own data structures and then use the DMShellSetXXX interfaces to hook them up.
>
> Lawrence