Jed, I do see the drawbacks of this method, but it seems to have good scalability. That I want to test.
Thank you. On 17.08.2012 16:32, Jed Brown wrote: > On Fri, Aug 17, 2012 at 9:23 AM, Alexander Grayver > <agrayver at gfz-potsdam.de <mailto:agrayver at gfz-potsdam.de>> wrote: > > Right. > Could you please explain what do you mean by fragile? > > > Fragile means unreliable. The math is not right for Krylov methods in > general. You are relying on side-effects of particular methods and > preconditioner side combinations. Fragile is the opposite of "robust". > For a "robust" method, you should start with a compatible > discretization and/or choose a preconditioner that is stable on the > quasi-null space and/or filter in a way that is consistent with the > Krylov method (e.g. as described using FGMRES). > >> >> Looks a bit tricky. The number of fgmres iterations then >> defines number of the cycles correction will be applied, >> doesn't it? >> >> >> Yes. > > If I do this command line only setup: > > -pc_type composite -pc_composite_type multiplicative > -pc_composite_pcs ksp,shell > > How does petsc know what ShellPCApply routine to take? I guess I > have to specify this in the code anyway? > > > Yeah, to select your preconditioner in this way, you should name it > preconditioner something meaningful and use PCRegisterDynamic() just > like the native implementations. -- Regards, Alexander -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120817/05650d90/attachment.html>
