[petsc-users] about parallel preconditioned matrix-free gmres

[Matthew Knepley]

On Fri, Jun 29, 2012 at 7:25 PM, RenZhengYong <renzhengyong at gmail.com>wrote:

> Dear Petscs,
>
> Use the uniprocessor complex-value based version petsc, I recently
> successfully make a FETI_DP domain
> decomposition approach working for 3D electromagnetic induction (earth)
> problem.  The number of iteration of
> the interface problem seems to be scalable with regard to the number of
> sub-domains.
>
> To do this, I had two subroutines for petsc
>
> (1) int mat_vec_product_interface_problem(Mat A, Vec X, Vec Y) for
> matrix-free GMRES solver
> (2) int preconditioner_mat_vec(PC pc,Vec X,Vec Y) for shell
> preconditioner.
>
> Now, I want to solve the interface problem by paralleled GMRES solver so
> that I can solve real large-scale problems.  Could you please tell me the
> easiest way to accomplish it. Which specific data structures of petsc
> should be used.  I have been using Petsc for 3.5 years, I really want to
> have a try the real MPI-based Petsc.
>

1) The solver logic should be parallel already since it only uses calls to
Vec or Mat functions. The problems will be
    in building data structures.

2) It looks like your two items above are the things to be parallelized

3) Decide how to partition the problem

4) Use VecScatter() to communicate data along the interface of your
partitions

I don't think we can give better advice than that without more specific
questions. Note that there is
a current effort to put BDDCinto PETSc. You can see it in petsc-dev, as
PCBDDC.

  Thanks,

    Matt


> Thanks in advance.
> Have a nice weekeed
> Zhengyong
>
>
> --
> Zhengyong Ren
> AUG Group, Institute of Geophysics
> Department of Geosciences, ETH Zurich
> NO H 47 Sonneggstrasse 5
> CH-8092, Zürich, Switzerland
> Tel: +41 44 633 37561
> e-mail: zhengyong.ren at aug.ig.erdw.ethz.ch
> Gmail: renzhengyong at gmail.com
>



-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120630/0821cd90/attachment.html>