[petsc-users] Implementation of Multilevel preconditioner

Jed Brown jedbrown at mcs.anl.gov
Thu Apr 5 10:30:00 CDT 2012


On Thu, Apr 5, 2012 at 08:22, Abdul Hanan Sheikh <hanangul12 at yahoo.co.uk>wrote:

> For the system Au = f, I want to apply a multilevel preconditioner with a
> KSP (say FGMRES) . The preconditioner
> reads as
>
> Prec = C +  M^-1 (I - A*C) , where
>
>
>    -  C reads coarse grid correction operator i.e. C =  P*A_2h\R  [ R
>    restriction, A_2h coarse operator, P interpolation]
>    - M is say some sparse matrix (resembling to A)
>
> What makes it multilevel is I have to approximate coarse operator "A_2h"
> with few KSP iterations preconditioned by the above defined
> preconditioner "Prec" , but off course at the level 2h and hence at every
> coarse level until coarsest.
>

Hmm, if you use a constant number of iterations in C at each level, the
number of coarsest grid iterations will grow exponentially in the number of
levels. Is that really what you want?


> I only know, it can be implemented with PCMG. I am little afraid of DMMG.
>

Don't use DMMG, it is being removed.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120405/6be581ae/attachment-0001.htm>


More information about the petsc-users mailing list