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

[Matthew Knepley]

On Thu, Dec 6, 2012 at 7:04 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> On Wed, Dec 5, 2012 at 10:36 AM, Anton Popov <popov at uni-mainz.de> wrote:
>>
>> This is not exactly clear to me. Residuals for Dirichlet DOF should be
>> (set to) zero, should they? Because those are not unknowns. What do you mean
>> by Dirichlet boundary residuals?
>
>
> If you leave the variables in the vector (convenient for visualization, for
> example) then you have to write residuals for them. With most
> preconditioners, those residuals will be zero after the first iteration, but
> that is not required.
>
>>
>> Is there any SNES example demonstrating usage of Newton or Picard for a
>> problem with non-zero Dirichlet BC?
>
>
> The lid-driven cavity examples do the non-symmetric version. snes ex48 uses
> symmetric version enforcement and could trivially be inhomogeneous. When
> working with external applications, I almost always advocate doing this
> instead of removal or other ad-hoc boundary condition implementations.
>
> Here is a slightly more formal explanation:
>
> http://scicomp.stackexchange.com/a/3300/119

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.

   Matt

--
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which
their experiments lead.
-- Norbert Wiener