[petsc-users] Local sizes when creating PETSC MatIS

[José Lorenzo]

Following the PETSC examples, I was trying something like:

  call ISLocalToGlobalMappingCreate(PETSC_COMM_WORLD, 1, nidx, idx_global,
PETSC_COPY_VALUES, ltog, ierr)

  call MatCreateIS(PETSC_COMM_WORLD, 1_ip, proedg(rank), proedg(rank),
PETSC_DETERMINE, PETSC_DETERMINE, ltog, ltog, An, ierr)

  call MatISSetPreallocation(An, d_nz, PETSC_NULL_INTEGER, o_nz,
PETSC_NULL_INTEGER, ierr)

with idx_global storing the global indices of both local and ghost edges,
and proedg(rank) the number of edges at the current processor.

The program crashes in the Preallocation subroutine:

[0]PETSC ERROR: --------------------- Error Message
--------------------------------------------------------------
[0]PETSC ERROR: Argument out of range
[0]PETSC ERROR: nnz cannot be less than 0: local row 4 value
-7493980938351671286
[0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for
trouble shooting.
[0]PETSC ERROR: #1 MatSeqAIJSetPreallocation_SeqAIJ() line 3670 in
PETSC_DIR/src/mat/impls/aij/seq/aij.c
[0]PETSC ERROR: #2 MatSeqAIJSetPreallocation() line 3646 in
PETSC_DIR/src/mat/impls/aij/seq/aij.c
[0]PETSC ERROR: #3 MatISSetPreallocation_IS() line 1795 in
PETSC_DIR/src/mat/impls/is/matis.c

And with your suggestion I get the same error during the preallocation step.

Thank you.


El vie., 14 jun. 2019 a las 14:24, Stefano Zampini (<
stefano.zampini at gmail.com>) escribió:

> Jose,
>
> The local size of the matrix (the one you use in MatSetSizes) tells Petsc
> what is the size of the local portion of the vectors involved in matrix
> vector products.
> It has nothing to do with the size of the local Neumann problem needed by
> MatIS. These are inferred from the local to global mapping of the degrees
> of freedom
> In your case, the  l2g object needs to map all the dofs that corresponds
> to the local mesh to their global dof numbering (you can take a look here
> https://epubs.siam.org/doi/abs/10.1137/15M1025785)
>
> MatCreate(comm,&A)
> MatSetSizes(A,PETSC_DECIDE,PETSC_DECIDE,N,N)
> MatSetType(A,MATIS)
> MatSetLocalToGlobalMapping(A,l2g,l2g)// this creates the local Neumann
> matrices
>
> For n = 1: number_of_local_elements
>   Construct element matrix
>   // you have two alternitives
>   MatSetValuesLocal(A, // using local ordering of dofs
>   //or MatSetValues(A, // using global ordering of dofs
> end
>
> On Jun 14, 2019, at 5:09 AM, José Lorenzo via petsc-users <
> petsc-users at mcs.anl.gov> wrote:
>
> I am working with Nedelec Finite Elements and using ghost vectors to store
> the ghost values of the solution in the edges that do not belong to current
> processor.
>
> Since I have a large contrast in material properties I would like to try
> the PCBDDC preconditioner in PETSC. Apparently, my matrix needs to be of
> type MatIS, but I don't get to understand how MatCreateIS would work in my
> particular case, although I did have a look at mat/ex3.c and ksp/ex59.c
>
> I set the local size m to the number of edges in current processor, but I
> do not know whether I should include the number of ghost edges. And for the
> local size n should I use the total number of DOFs or proceed as with m?
>
>
> Thanks,
>
> Jose
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20190614/75bba2ae/attachment.html>