Problem creating a non-square MPIAIJ Matrix

[Satish Balay]

>    MatCreateMPIAIJ(PETSC_COMM_WORLD,PETSC_DECIDE,PETSC_DECIDE,
>            8,6,PETSC_DEFAULT,d_nnz ,PETSC_DEFAULT, o_nnz, &A);

Here you are asking PETSc to decide tine local 'm,n' partition sizes.

So you should use MatGetLocalSize() to get these values [and not
assume they are sequare blocks]

If you need to have the diagonal blocks square - then specify them
appropriately at the matrix creation time [instead of using
PETSC_DECIDE - for m,n parameters].

However - note that you'll have to create Vectors with matching
parallel layout [when using in MatVec()]

i.e in y = Ax

x should match column layout of A
y should match row layout of A

Satish

On Fri, 16 Feb 2007, Shi Jin wrote:

> Hi there,
> 
> I am found a very mysterious problem of creating a MxN
> matrix when M>N.
> Please take a look at the attached short test code I
> wrote to demonstrate the problem.
> 
> Basically, I want to create a 8x6 matrix. 
> A={1 2 3 4 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0
> 
> If I do it with 2 processes, I suppose the local
> submatrices should look like
> p0:
>    1 2 3 4 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
> p1:
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0;
>    0 0 0 0 0 0
> The problem is that on the first process, the diagonal
> portion of the local submatrix is
>    1 2 3 4 | 0 0;
>    0 0 0 0 | 0 0;
>    0 0 0 0 | 0 0;
>    0 0 0 0 | 0 0;
> So I need to set d_nnz[0]=4 on p0 since the first row
> has 4 diagonal nonzero entries. However, when I run
> the code by
> mpiexec -n 2 ./mpiaij
> I got error saying that
> [0]PETSC ERROR: --------------------- Error Message
> ------------------------------------
> [0]PETSC ERROR: Argument out of range!
> [0]PETSC ERROR: nnz cannot be greater than row length:
> local row 0 value 4 rowlength 3!
> 
> It seems that it is checking against the n parameter
> which is set by petsc to be n=N/2=3.
> But why should we do that? According the manual, the
> local submatrix is of dimension m by N.
> 
> Could you please help me understand the problem?
> Thank you very much.
> Shi
> 
> 
>  
> ____________________________________________________________________________________
> TV dinner still cooling? 
> Check out "Tonight's Picks" on Yahoo! TV.
> http://tv.yahoo.com/