[petsc-dev] default SNES rtol for single precision builds

[Matthew Knepley]

On Mon, Sep 19, 2011 at 1:13 PM, Lisandro Dalcin <dalcinl at gmail.com> wrote:

> On 19 September 2011 14:58, Matthew Knepley <knepley at gmail.com> wrote:
> > On Mon, Sep 19, 2011 at 12:27 PM, Lisandro Dalcin <dalcinl at gmail.com>
> wrote:
> >>
> >> After debugging a failing petsc4py testcase using single precision
> >> builds, I realized the default SNES relative tolerance is 1e-8. What
> >> should we do?
> >
> > In single, you can reach 1e-8, you just take more iterates since it looks
> > like
> > iterative refinement. I say we leave the tolerance where it is and make
> our
> > tests smarter.
> >    Matt
> >
>
> Well, this is a stupid testcase solving du/dt + u*u = 0 with TS. SNES
> failed at line search.


That is strange. This means the linear solve did not manage to find a
descent
direction, so its really about linear solve accuracy than SNES. I would not
have
thought that this equation would be that sensitive to precision. This is not
my
experience with other equations.

   Matt


>
> --
> Lisandro Dalcin
> ---------------
> CIMEC (INTEC/CONICET-UNL)
> Predio CONICET-Santa Fe
> Colectora RN 168 Km 472, Paraje El Pozo
> 3000 Santa Fe, Argentina
> Tel: +54-342-4511594 (ext 1011)
> Tel/Fax: +54-342-4511169
>



-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110919/0e5b4442/attachment.html>