[petsc-dev] Fwd: rename SNES methods ls, tr etc

[Jed Brown]

Matt and I have talked about this many, many times. Removing Dirichlet dofs
entirely is inconvenient for structured grids and does not allow any data
structure reuse in cases of phase change or similar where a node changes
from Dirichlet to Neumann or vice-versa. It is also much less convenient
for visualization because you need to embed the solution in the larger
space before visualizing.

I'm not a big fan of having two ways to do things, so since it's sometimes
necessary, I usually recommend always leaving the Dirichlet dofs in using
the procedure explained above. Regardless, the implementation is definitely
not more complicated in defect correction mode than in "non-defect Picard".
Dave May asserts that it is simpler to think about and to implement in
defect correction mode.

On Thu, Dec 6, 2012 at 8:39 AM, Matthew Knepley <knepley at gmail.com> wrote:

> I know Jed likes this, but I think it is simpler to think about the
> problem with
> Dirichlet unknowns removed from the solve. We can do both.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20121206/0fdc2c06/attachment.html>