[petsc-users] DMDA ghosted boundary

[David Scott]

On 12/08/2022 18:15, Matthew Knepley wrote:
> This email was sent to you by someone outside the University.
> You should only click on links or attachments if you are certain that 
> the email is genuine and the content is safe.
> On Fri, Aug 12, 2022 at 9:27 AM David Scott <d.scott at epcc.ed.ac.uk> wrote:
>
>     On 12/08/2022 11:26, Matthew Knepley wrote:
>>     This email was sent to you by someone outside the University.
>>     You should only click on links or attachments if you are certain
>>     that the email is genuine and the content is safe.
>>     On Fri, Aug 12, 2022 at 2:21 AM Quynh Nguyen
>>     <nguyenquynh9537 at gmail.com> wrote:
>>
>>         Dear petsc developers and users,
>>
>>         I am a new petsc user ^_^.
>>         I have some questions about using DMDA_BOUNDARY_GHOSTED.
>>
>>         1. I have 2D Poisson code with domain (nx, ny), and I used 
>>         DMDA_BOUNDARY_NONE, then the size of the matrix is
>>         (nx*ny)X(nx*ny). */what is the size of the matrix if I used
>>         DMDA_BOUNDARY_GHOSTED? is it the same size as
>>         DMDA_BOUNDARY_NONE/*?
>>
>>
>>     Yes, the matrix should be the same size, but local vectors have
>>     extra ghost node slots.
>>
>>         2. if the matrix size of DMDA_BOUNDARY_GHOSTED isn't the same
>>         as DMDA_BOUNDARY_NONE, */how i can view the matrix
>>         including ghost points/*? I ask this one because I
>>         used MatView(P, PETSC_VIEWER_STDOUT_WORLD) to show the matrix
>>         but excluded ghost points.
>>
>>
>>     Ghost nodes are there so that stencil operations on local vectors
>>     work at boundaries. You fill in the slots with boundary information.
>>
>     I can see how this works if you are writing your own solver but
>     can it be used with a KSP? In the case of periodic boundary
>     conditions the boundary values are "filled in" for you by PETSc.
>     Is there any way for a user to supply boundary information to a
>     PETSc solver when using DMDA_BOUNDARY_GHOSTED?
>
>
> Right now, there is not. This is mainly due to the lack of consensus 
> for boundary condition specification. Periodic conditions are simple 
> in this respect.
>
> For example, Plex has such a thing
>
> https://petsc.org/main/docs/manualpages/DM/DMAddBoundary/
>
> that can be used to fill in these constrained spots
>
> https://petsc.org/main/docs/manualpages/DMPLEX/DMPlexInsertBoundaryValues/
>
> But there have been many complaints about this interface (and I 
> suspect any interface we choose).
>
> Maybe there is something that we could write that would make it easier 
> for you with DMDA?
>
The way I had imagined it working was that for each dimension the user 
could register a function to be to be called at the start of every 
iteration (in a similar fashion to adding a monitor). Not all of the 
dimensions would have to be specified as DMDA_BOUNDARY_GHOSTED and thus 
require a function to be registered. This function would have access to 
the DMDA and the local vector so it could determine if ghost locations 
were present and, if so, set appropriate values. Does this make sense?

Thanks,

David

>   Thanks,
>
>      Matt
>
>     Thanks,
>
>     David
>
>>       Thanks,
>>
>>          Matt
>>
>>         I hope someone can explain my wonder :)
>>
>>         P/S: maybe, those question is simple and a bit stupid ^_^.
>>
>>         Have a nice day,
>>
>>         -- 
>>         Quynh Nguyen
>>         Graduate student, Ocean Engineering.
>>         School of Naval Architecture and Ocean Engineering,
>>         University of Ulsan.
>>         -----------------------------------------------------------------------------------------
>>         Mobile: (+82) 010-9394-5595
>>         Email: nguyenquynh9537 at gmail.com; nguyenthiquynhb_t58 at hus.edu.vn
>>
>>
>>
>>     -- 
>>     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/>
>
>     The University of Edinburgh is a charitable body, registered in
>     Scotland, with registration number SC005336. Is e buidheann
>     carthannais a th’ ann an Oilthigh Dhùn Èideann, clàraichte an
>     Alba, àireamh clàraidh SC005336.
>
>
>
> -- 
> 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/20220815/dff4afa2/attachment-0001.html>