[CIG-CS] Variable viscosity Stokes solver

[Walter Landry]

Wolfgang Bangerth <bangerth at math.tamu.edu> wrote:
> 
>> Now I am a bit confused.  You sent a me a modified version of step-32
>> which uses Trilinos explicitly, but the falling block test does not.
>> So, are they using the same type of solver?
> 
> Almost. step-32 uses Trilinos and AMG as preconditioner for the top left block 
> of the Stokes matrix. The modified step-22 we were discussing for the 
> variable-viscosity Stokes benchmark uses the same outer solver and the same 
> kind of preconditioner but instead of preconditioning the solution of the top 
> left block by AMG it just uses UMFPACK. Since it doesn't use AMG, there is no 
> need for Trilinos and so I use deal.II's native linear algebra.
> 
> That said, the solver and preconditioner is the same, just the implementation 
> is different.
> 
> Does that make sense?

Thanks.  So I presume that if I want to test how well the solver
works on 1000 cores, I should use the Trilinos solver?

Cheers,
Walter Landry
walter at geodynamics.org