[petsc-users] Poor multigrid convergence in parallel

[Lawrence Mitchell]

On 29/07/14 14:35, Jed Brown wrote:
> Lawrence Mitchell <lawrence.mitchell at imperial.ac.uk> writes:
>> So my coarse space is spanned by the fine one, so I copy coarse dofs to the corresponding fine ones and then linearly interpolate to get the coefficient value at the missing fine dofs.  
> 
> Good, and is restriction the transpose?

No, I'm L2-projecting (with mass-lumping) for the restriction.  So if I
weren't lumping, I think this is the dual of the prolongation.  The
lumping seems to be alright (although I'm aware that the preconditioned
operator is then not necessarily symmetric) in the sense that
convergence is good irrespective of mesh resolution/MG levels.

> Some people mistakenly use nodal injection and not surprisingly see poor
> convergence because the restriction operator aliases high frequencies at
> full amplitude and the prolongation is accurate for neither high or low
> frequencies.  (HPCG does this, for example.)

Sure.

Anyway, I think for now problem addressed.  Thanks for the continued advice!

Cheers,

Lawrence


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 455 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140729/4aaed6a3/attachment.pgp>