[petsc-users] DMPlex for cell centred finite volume

[Matthew Knepley]

On Tue, Aug 27, 2019 at 4:54 AM Edoardo alinovi <edoardo.alinovi at gmail.com>
wrote:

> Hello Matt,
>
> Thanks for your kind replay. Using DMplex and changing the data structure
> is not a straightforward task as you can imagine. Up to now I am in the
> preliminaries and I am trying to understand how to reuse most of all my
> subroutines (I am using fortran) such as fluxes and coeffs assembly,
> numerical schemes  and so o in order to avoid to rewrite all the code :p I
> will start with your first two suggestions and see if it is  an affordable
> task :)
>
> For the face value, I am not interested in Reimann solvers. Being
> incompressible I am using Central difference, QUICK, SOU, Upwind or
> TVD-family schemes, but maybe I can use my stuff to get it.
>
> I think I can be happy enough having the ghost cell centers available and
> some way to deal with the BC (I guess this will be the most tricky part).
> Basically I would like to use DMplex at low level just to have a support
> for the parallel mesh management.
> Have you got some example for P0 finite elements and overlap=1?
>

This is one group already using PETSc in exactly this way for a FV code.

Maybe it would help if I gave a high-level overview of this usage pattern.

1) Hopefully replacing your current mesh topology/geometry with Plex is not
hard. Let me know if you have any questions about this.

2) If you give overlap=1 on distribution, then support(face) for any
regular face gives the two cells on either side. You
    probably want to screen out faces between two ghost cells. There is a
DMLabel for doing that.

3) A PetscSection just gives the data layout over a mesh. The idea here is
to make it describe _exactly_ the layout you already have.
    The DMGetGlobalVector() and DMGetLocalVector() should produce vectors
in your existing order and you can reuse all your code.

  Thanks,

     Matt


> Thank you! :)
>
> Il giorno mar 27 ago 2019 alle ore 09:33 Matthew Knepley <
> knepley at gmail.com> ha scritto:
>
>> On Tue, Aug 27, 2019 at 4:07 AM Edoardo alinovi via petsc-users <
>> petsc-users at mcs.anl.gov> wrote:
>>
>>> Hello PETSc users and developers,
>>> I hope you are doing well! Today I have a general question about DMplex
>>> to see if it can be usefull to me or not.
>>>
>>> I have my fancy finite volume solver (cell centered, incompressible NS)
>>> which uses petsc as linear solver engine. Thanks to your suggestions so far
>>> it is now running well :)
>>>
>>> I would like to see if I can do another step incapsulatng in petsc also
>>> the mesh managment part. I think that Dmplex is the way to go since my code
>>> is fully unstructured.
>>>
>>> I have red some papers and lectures by Matt around the web, but still I
>>> have not found an answer to this question:
>>> Can dmplex dial with cell centered data arrangement and provide some
>>> support for basic operation (e. g. interpolation between partitions, face
>>> value calculation etc)?
>>>
>>
>> 1) Cell centered data
>>
>> Definitely yes. This is the same data layout as P0 finite elements. You
>> just assign the PetscSection k dofs per cell.
>>
>> 2) Interpolation between partitions
>>
>> I assume you mean ghost values from other parallel partitions. You can do
>> this by using overlap=1 in DMPlexDisrtibute().
>>
>> 3) Face value calculation
>>
>> You can do Riemann solves by looping over faces, grabbing values from the
>> neighboring cells, doing the calculation, and
>> updating cell values. We carry out this kind of computation in TS ex11.
>> That example attempts to do everything, so it is
>> messy, but we can help you understand what it is doing in any part that
>> is unclear.
>>
>> 4) FV boundary conditions
>>
>> You can use DMPlexConstructGhostCells() to put artificial cells around
>> the boundary that you use to prescribe fluxes.
>>
>>   Thanks,
>>
>>      Matt
>>
>>
>>> Thank you very much!
>>>
>>> Edo
>>>
>>
>>
>> --
>> What most experimenters take for granted before they begin their
>> experiments is infinitely more interesting than any results to which their
>> experiments lead.
>> -- Norbert Wiener
>>
>> https://www.cse.buffalo.edu/~knepley/
>> <http://www.cse.buffalo.edu/~knepley/>
>>
>

-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener

https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20190827/c108a848/attachment.html>