Thanks for the suggestions guys. Barry’s field split approach gives sensible behavior. Now I’m working on tuning it. -gideon
> On Sep 21, 2015, at 9:21 AM, Jed Brown <[email protected]> wrote: > > Matthew Knepley <[email protected]> writes: >>> At the same time, the first block, as discussed, has >>> these zeros on the diagonal, and can’t use SOR. >>> >> >> Chebyshev/Jacobi is the obvious next try. > > Not really; the Chebyshev polynomial exceeds 1 left of the origin, so > it's pretty non-functional on indefinite operators. Pretending like > Gideon's saddle point problem is positive definite is not a solution.
