[petsc-users] [petsc-maint] two-level Schwarz

[Smith, Barry F.]

   Note that, of course, you have to have a mechanism in the code to generate the coarse geometric grid, by for example using a DM object that knows how to refine or coarsen itself. 

   If all you have is code that generates a single grid and matrix associated there is no way to automatically use a geometric coarse grid but you could use two levels of algebraic multigrid by replacing the  -pc_type  mg with  -pc_type gamg.

   Barry


> On Mar 17, 2018, at 6:28 AM, Matthew Knepley <knepley at gmail.com> wrote:
> 
> On Fri, Mar 16, 2018 at 11:35 PM, John <johnlucassaturday at gmail.com> wrote:
> Hi PETSc team:
> 
> I am curious if the two-level Schwarz preconditioner, as is illustrated in R. Tuminaro's paper ``on a multilevel preconditioning module for unstructured mesh Krylov solvers: two-level Schwarz", is readily available from command line options?
> 
> Hopefully I understand from my quick look at the paper. It looks like 2 levels MG, using DD as the smoother for the fine grid:
> 
>   -pc_type mg -pc_mg_nlevels 2 -pc_mg_galerkin
>     -mg_levels_pc_type asm
> 
> The coarse grid will automatically be LU as in the paper. You can use PCTELESCOPE to move the coarse grid to a subset
> of processes or one process (PCREDUNDANT also works here). You do not have to use Galerkin projection, but could
> rediscretize on the coarse grid.
> 
>   Thanks,
> 
>      Matt
>  
> Thanks,
> 
> John
> 
> 
> 
> -- 
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener
> 
> https://www.cse.buffalo.edu/~knepley/