[petsc-users] Multigrid preconditioning

[Jed Brown]

On Mon, Jul 30, 2012 at 4:39 PM, Michele Rosso <mrosso at uci.edu> wrote:

>  Thank you,
>
> I will try to use option 2 as you suggested.
> I'd prefer to implement the multigrid  preconditioner directly inside the
> code
> rather then using he command line options.
> Could you point me to an example where this (or something similar) is done?
>

You can use -help and grep to find the functions that provide the
functionality in command line options.

We recommend using command line options to explore which methods work. Once
you have a configuration that you really like, you can put it in the code
or (easier) just put it in an options file.


>
> Thank you,
>
> Michele
>
>
>  On 07/30/2012 02:18 PM, Jed Brown wrote:
>
> On Mon, Jul 30, 2012 at 4:06 PM, Michele Rosso <mrosso at uci.edu> wrote:
>
>>  Hi,
>>
>> I am solving a variable coefficients Poisson equation with periodic BCs.
>> The equation is discretized by using the standard 5-points stencil finite
>> differencing scheme.
>> I managed to solve the system successfully with the PCG method and now I
>> would like to add
>> a preconditioner to speed up the calculation. My idea is to use the
>> multigrid preconditioner.
>>
>> Example ex22f.F implements what I think I need.
>> If I understand correctly example ex22f.F, the subroutines "ComputeRHS"
>> and "ComputeMatrix" define how the
>> matrix and rhs-vector have to be computed at each level.
>> In my case tough, both the jacobian and the rhs-vector cannot be computed
>> "analytically", that is, they depend on variables
>> whose values are available only at the finest grid.
>>
>> How can I overcome this difficulty?
>>
>
>  Two possibilities:
>
> 1. homogenize on your own and rediscretize
>
>  2. use Galerkin coarse operators (possibly with algebraic multigrid)
>
>
>  Option 2 is much more convenient because it never
>
>  For geometric multigrid using DMDA, just use -pc_type mg -pc_mg_galerkin
>
>  For algebraic multigrid, use -pc_type gamg -pc_gamg_agg_nsmooths 1
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120730/65de1e8e/attachment-0001.html>