On a related issue, I wanted to mention 3D linear elasticity example 
that I added: systems_of_equations_ex6. It seems very slow with the 
default iterative solver in PETSc (ILU + GMRES, right?). The example 
takes 26 seconds on my laptop.

But this is a symmetric problem, so you'd expect CG to be good. Calling
./systems_of_equations_ex6-opt -ksp_type cg
runs fast, but gives the wrong answer (I should add some code to check 
the final residual...)

The best thing for me is just a direct solver:
./systems_of_equations_ex6-opt -ksp_type preonly -pc_type lu 
-pc_factor_mat_solver_package umfpack

But I guess there must be a good KSP/PC combo for this simple problem, 
any recommendations?

Best,
David



On 09/13/2012 04:56 PM, John Peterson wrote:
> On Thu, Sep 13, 2012 at 2:49 PM, Paul T. Bauman <[email protected]> wrote:
>> This is what I do usually, but with the external packages. The internal
>> PETSc lu doesn't handle pivoting.
>
> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/PC/PCLU.html
>
>       -pc_factor_nonzeros_along_diagonal      - permutes the rows and columns 
> to
> try to put nonzero value along the diagonal.
>
> That's pivoting, right?
>


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to