[Trilinos-Users] 2D diffusion problem with MPI

[Heroux, Michael A]

Olivia,

I think the problems is with the version of the constructor you are using for the matrix A.  Normally, you don't need to specific the column map ("r" in your example).  You should instead let Epetra define it for you.

Try removing "r" and leaving it out of the constructor for "A".

Mike


On 6/26/08 2:31 PM, "Olivia Buzek" <obuzek at nist.gov> wrote:

I am trying to solve a 2D diffusion problem in parallel using MPI and
PyTrilinos.

The relevant pieces of the code I have been using are as follows:

(size is the dimension of the matrix, currently set to 25; bandwidth
is currently set to 3)

    m = Epetra.Map(size, 0, comm)
    r = Epetra.Map(size,size,0,comm)
    A= Epetra.CrsMatrix(Epetra.Copy, m, r, 3*bandwidth/2)
    A.ReplaceGlobalValues(...)
    A.FillComplete()
    b = Epetra.Vector(m)
    b.ReplaceGlobalValues(...[initial values]...)
    x = Epetra.Vector(m)

    solver = AztecOO.AztecOO(A, x, b)
    solver.SetAztecOption(AztecOO.AZ_solver, AztecOO.AZ_cg)
    solver.SetAztecOption(AztecOO.AZ_output, AztecOO.AZ_none)
    err = solver.Iterate(5000, 1e-10)

Trilinos is compiled properly for MPICH version: 1.2.7p1.  The problem
is being distributed across 2 processors.  Evidently, the matrix is
being built correctly.  However, upon reaching the first solver
statement (initializing AztecOO), the following error is returned:

AZ_extract_comm_info: Received elements must be stored after
                   all 12 local elements

Can anyone advise as to what is going on here?

Thanks.

Olivia Buzek

_______________________________________________
Trilinos-Users mailing list
Trilinos-Users at software.sandia.gov
http://software.sandia.gov/mailman/listinfo/trilinos-users


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://software.sandia.gov/mailman/private/trilinos-users/attachments/20080626/209af590/attachment.html