[petsc-dev] Revamping the multilevel KSP interface

[Jed Brown]

I'm updating the KSP interface to be consistent with what I did with SNES.
My question is how to manage the definition of the coarse level problem.
Does it actually make sense to have a callback for initial guess and right
hand side? PCMG arranges to only have the matrix assembled on levels by
modifying the DM, but I don't like that.

What is the use case where you want initial guess and right hand side to be
a callback?

I'm unsure of the usefulness now because DMComputeInitialGuess() does not
have a way to determine the forcing term, so the DM context would have to
have stored the right hand side somehow to be able to find a meaningful
initial guess. But then DMComputeInitialGuess() isn't even called again if
the RHS changes (KSP_SETUP_NEWRHS).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120317/7fed38bc/attachment.html>