[petsc-dev] controlling vector values while doing matrix free operations

Dmitry Karpeev karpeev at mcs.anl.gov
Wed Dec 14 21:34:29 CST 2011

On Wed, Dec 14, 2011 at 9:31 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

> On Dec 14, 2011, at 9:22 PM, Jed Brown wrote:
> > On Wed, Dec 14, 2011 at 17:55, Barry Smith <bsmith at mcs.anl.gov> wrote:
> >  What Matt is trying to tell you is that changing values at this level
> to prevent function evaluations outside of the function domain inside
> Newton is a total hack that rarely works well and should be avoided. You
> really don't want to do this.
> >
> > It sounds like for the problems Bobby is asking about, the continuum
> solutions never produce values at a "constraint". In gas dynamics, the
> continuum equations are PDEs, not variational inequalities, but the
> pressure and temperature are still never negative. If a bound was
> available, MatMFFD could potentially choose the sign of the differencing to
> stay within the feasible domain. This does seem quite limited, however,
> because the positive and negative versions could cause violations in
> different parts of the domain.
>    If you are truly near 0 in temperature or pressure and differencing can
> push you over the limit then I submit it is perfectly reasonable to solve
> them as VI.
>   Though actually our VI solvers are currently not really matrix-free so
> this is a bit academic.

We could support a matrix-free SNESVI if extracting a submatrix from a MF
matrix was implemented by pre-/post-multiplying it by a suitable P, no?


>   Barry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20111214/16e69225/attachment.html>

More information about the petsc-dev mailing list