[petsc-dev] Using PCFieldSplitSetIS

[Matthew Knepley]

Each process calls FieldSplitSetIS() with the indices for THAT FIELD that
are owned by that process. Each
process call FieldSplitSetIS() n+1 times if you have n+1 fields. Identifying
fields with processes is a mistake.

   Matt

On Wed, Mar 9, 2011 at 1:50 PM, Thomas Witkowski <
Thomas.Witkowski at tu-dresden.de> wrote:

> As I already asked on petsc-users, I want to implement some kind of
> iterative substructuring algorithm in my fem code. It was suggested to me to
> switch to the dev version of petsc and to make use of PCFieldSplit. So far I
> have installed petsc-dev and read a little bit about the PCFieldSplit. It
> sounds great. But I'm not really sure how to make use of. In my code I want
> to build n blocks (where n is also the number of processors), each for the
> interior domain of one rank. Okay, this seems to be easy. I make just one
> call to PCFieldSplitSetIS on each rank with IS being the global indices of
> the ranks interior nodes. But what about the n+1 block, which should contain
> all the nodes of the boundaries between the subdomains? Each rank
> contributes to this block. So how show PCFieldSplitSetIS should be called?
> May be some small example: Assume we have two ranks, each with 100 nodes in
> its interior domain and each rank contributes with 10 nodes to the interior
> boundary (so the overall interior boundary contains 20 nodes). So rank 0
> owns global indices 0 to 109, with 100 to 109 being the nodes of the first
> part of the interior boundary), and rank 1 owns global indices 110 to 219,
> with 210 to 219 being the nodes of the second part of the interior boundary.
> When I understood the idea behind PCFieldSplit correctly, it should be
> possible to generate the three blocks [0-99],[110-209],[100-109,210-219]
> using PCFieldSplitSetIs. But how to call it correctly?
>
> Thomas
>



-- 
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-dev/attachments/20110309/b2496736/attachment.html>