On 09/13/2012 07:22 PM, Jed Brown wrote:
> On Thu, Sep 13, 2012 at 4:16 PM, David Knezevic 
> <[email protected] <mailto:[email protected]>> wrote:
>
>     >     Calling
>     >     ./systems_of_equations_ex6-opt -ksp_type cg
>     >     runs fast, but gives the wrong answer
>     >
>     >
>     > That's alarming...
>     >
>     > Is ICC the default preconditioner with CG? If not, give that a try
>     > (maybe with -pc_factor_levels 4 or something)
>
>
>     ahh, "-pc_type icc" did the trick, thanks! I'll add that to the
>     Makefile.
>
>
> Use -ksp_monitor_true_residual. If ILU does something meaningfully 
> different from ICC, the system is almost certainly nonsymmetric. Check 
> boundary conditions.
>
> But for elasticity, it's well worth getting algebraic multigrid 
> working. Hypre does classical AMG which isn't much good for systems. 
> Use petsc-3.3 or later and call MatSetNearNullSpace() to provide the 
> rigid body modes, then use either -pc_type gamg or -pc_type ml. Either 
> should do fine (and can usually be tuned to do about the same 
> algorithm, but have different defaults). If you use a nodal basis, you 
> can stuff the nodal coordinates in a vector with suitable block size 
> and call MatNullSpaceCreateRigidBody().

OK, thanks, I'll look into this.

David


------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://ad.doubleclick.net/clk;258768047;13503038;j?
http://info.appdynamics.com/FreeJavaPerformanceDownload.html
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to