Re: Branch & bound for not everywhere defiend constraints
On 2009-09-01 14:00:14 -0400, Nate Hayes wrote:
> Arnold Neumaier wrote:
> > Correct is the following code, which uses Section 3.4 of the
> > Vienna proposal that must be considered whenever existence
> > statements are made (also in existence from an interval Newton
> > operator):
> >
> > if isEmpty(f(X')) delete X'
> > else if inf(f(X')) > 0 delete X'
> > else if PossiblyUndefined bisect or save X'
> > else if f(X') <= 0 accept X'
> > else bisect save X'
>
> Yes, but this requires global/local flags, which I think are to be avoided
> in data-parallel computing environments. In the context of this type of
> algorithm, the interval result is of no use when PossiblyUndefined is set,
> anyways. So it is better to return a "PossiblyUndefined" (or in some cases
> perhaps a "TotallyUndefined") NaI instead.
One could also have a flag attached to the returned interval.
Alternatively, one could have a function telling whether X' is
included in the domain of f.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arénaire project (LIP, ENS-Lyon)