On Mon, 2 Jun 2008, Derek Gaston wrote:

> On Jun 2, 2008, at 9:45 AM, Roy Stogner wrote:
>
> What advantages does FEMSystem have over using Ben's
> NonlinearImplicit system stuff using SNES?  I've used both, but I
> don't really see the difference.

For you, since the numerical jacobians aren't useful right now, one of
the biggest advantages is gone.  But having the time integration
factored out is nice for transient problems, and the interior_, side_,
etc. helper methods are nice for nonlinear problems.  But the biggest
remaining benefits are still "to be written", I think.  Threaded
assembly (without having to write your own weird TBB code) should be a
nice feature, even if we'll have to go through an API change to get
there.  And there are some other niceties (bubble-function based
physics-dependent error estimators, for example) that can really only
be implemented if the library can request residuals of one element at
a time.

> For us, we're using the NonlinearImplicit system because we hope to fill out 
> the Trilinos interfaces at some point and use NOX instead...

Is something wrong with the FEMSystem->SNES interface?  I admit that I
basically gave up on PetscDiffSolver when it turned out that my own
quasi-Newton code was faster on my apps, but there's at least a place
there to start.  If you're doing matrix-free work, the fact that
DiffSystem currently inherits from ImplicitSystem may be a stumbling
block, but NonlinearImplicitSystem has the same problem.
---
Roy

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to