Hi Toby,

you can use dealii::SolverGMRES with a PETSc block matrix, even in
parallel. There are many options to precondition this matrix. Block
triangular or block diagonal preconditioners using the schur
complement come to mind (elman, wathen, ...). I am not sure what is
done in elasticity, but that works for saddle point problems arising
from Stokes/Oseen/Navier-Stokes problems. For the individual blocks
(inner solves) you can again use PETSc solvers and preconditioners.
Also take a look at step-31 (even though it uses Trilinos).

--
Timo Heister
http://num.math.uni-goettingen.de/~heister



2010/7/24 Toby D. Young <[email protected]>:
>
>
> Greetings,
>
> Nothing new, in that I am trying to work with the PETSc and deal.II. Here
> goes with the problem I face...
>
> I have a block matrix (elastic-electric equations) which look like:
>
> (A B)
> (C D)
>
> where A and B are symmetric square matrices and B and C are rectangular.
> Obviously I would like to solve this system; though since D is not zero
> (such as in, for example, step-20) I can not use the techniques given by
> the deal.II tutorial. Jaka szkoda...   :-)
>
> (i) One option is to solve the entire matrix. This is fine for me to
> start, but it seems that the PETScWrappers do not allow this type of
> matrix (dealii::PETScWrappers::MPI::BlockSparseMatrix) to be passed to the
> GMRES solver; for example. Can this be fixed by template argument or
> instantiation in the petsc_solver.*? The use of "Blocks" is specific to
> deal.II right? All PETSc "sees" is another matrix to play with...
>
> (ii) A second option is to compute the Schur complement using a
> matrix-matrix solver, which can always be constructed, and is not too
> difficult to do so (PETSc has one of these and it only needs to provide
> an appropriate wrapper which would use the petsc_full_matrix class).
>
> Any guesses which is the way to go? (i) or (ii), both, or even a (iii)??
>
> I am interested in short-term solutions (getting a quick answer before
> the sun comes up tomorrow) and long term solutions (hacking and
> testing and then getting a fast, stable, and resusable algorithm).
>
> A few words of guidance, or a pointer to the lierature, if you can, would
> be greatly appreciated.
>
> Best,
>        Toby
>
> -----
>
> Toby D. Young
> Assistant Professor
>
> Institute of Fundamental Technological Problems
> Polish Academy of Sciences
> ul Adolfa Pawinskiego 5b
> 02-106 Warsaw
> Poland
>
> www:   http://www.ippt.gov.pl/~tyoung
> skype: stenografia
>
> _______________________________________________
> dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
>
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to