[Trilinos-Users] Symmetric matrices

Erik Boman egboman at sandia.gov
Wed Nov 7 13:15:31 MST 2012


I believe the Amesos-MUMPS interface requires the full matrix since it 
also supports the unsymmetric case.
This should not be a great concern since typically, the L and U factors 
need a lot more memory than the original matrix.

Erik

Heroux, Michael A wrote:
> Bart,
>
> It is fairly straight-forward to write a simple wrapper around existing
> Epetra capabilities in order to provide a matrix-vector multiplication
> operation that works when storing just the upper or lower triangle. It is
> more challenging to do so with existing Trilinos preconditioners.  None of
> them takes advantage of symmetric storage at this time.
>
> Do you want Trilinos preconditioners for symmetric operators, or just the
> matrix-vector multiplication?
>
> Mike
>
> On 11/7/12 1:34 PM, "Bart Janssens" <bart.janssens at lid.kviv.be> wrote:
>
>   
>> Hello,
>>
>> I have a problem involving symmetric matrices, and to save memory I'd
>> like to only store the upper triangular part and the diagonal. I would
>> use the following solvers:
>> - MUMPS (through Amesos) on a matrix that stays the same during the
>> entire simulation, so I can reuse the factoring
>> - Belos Recycling Conjugate Gradient for a matrix that stays nearly or
>> completely the same for all time steps, but may be too big to factor
>> directly
>>
>> Is it possible to keep only the upper part of the matrix in these
>> cases, or do the solvers not support it? Sorry if I missed anything, I
>> couldn't find any references to symmetric matrix construction in the
>> tutorial.
>>
>> Cheers,
>>
>> -- 
>> Bart
>>
>> _______________________________________________
>> Trilinos-Users mailing list
>> Trilinos-Users at software.sandia.gov
>> http://software.sandia.gov/mailman/listinfo/trilinos-users
>>     
>
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users
>   



More information about the Trilinos-Users mailing list