If you can use petsc-dev, there is a feature branch
(karpeev/ksp-matschurcomplement) that allows to assemble
the Schur complement used by PCFIELDSPLIT. While ordinarily this is a bad
idea, since it requires inversion of
the A00-block and results in a dense matrix, it can be done very cheaply,
if y
Hi All,
Probably a silly question, but is it possible to condense out the rows and
columns corresponding to dirichlet boundary conditions so that the
resulting system matrix to solve is smaller (using lagrange type elements).
I'm using a DirichletBoundary object to apply homogenous constraints to
