Thanks Ben, that's what I needed.

On Aug 22, 2008, at 12:02 PM, Benjamin Kirk wrote:
> If *all* the variables are decoupled, would you not want to put them  
> in
> different systems?

Heh - I thought that would come up.

As usual, I'm asking about one thing in order to accomplish something  
else.

In our JFNK code we use the system matrix as a preconditioner using - 
snes_mf_operator.  What we do is only fill in approximate jacobians  
for the diagonal blocks... then we hit that with something (usually  
AMG) and use it as a preconditioner.  This turns out to work pretty  
well in most circumstances (depending on how approximate your  
approximations are and how tightly coupled all of your physics is...  
as you are obviously throwing away the coupling sensitivities in your  
preconditioner).

So, I really did need a way of just preserving the block diagonals...  
but it does all need to be in one big matrix (as that's what's  
_actually_ being solved).

BTW - The Trilinos integration is going pretty well... I'm working on  
the solve steps now (getting segfaults for some reason).  Hopefully  
I'll get it worked out and checked in this afternoon.

Thanks again,
Derek


-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to