[gecode-users] Limits on nonlinear constraints

[Javier Romero]

True, that makes total sense. This formulation of the problem is actually an
adaptation from floating point constraints to integer constraints; so
another way of solving it would be implementing something like FloatVar. But
i couldn't find anything about that either.

Thanks anyway for the great work you are doing with Gecode!
-- Javier Romero --


On Fri, Aug 20, 2010 at 5:49 PM, Guido Tack <tack at gecode.org> wrote:

> Javier Romero wrote:
>
> > Thanks for the fast answer. I tried decomposing the last constraint into
> mul, sqr and linear rules, but i got the same result. I attached the code. I
> compiled it by doing
> > g++ parallelepiped.cpp -lgecodeminimodel -lgecodeint -lgecodesearch
> -lgecodegist -lgecodedriver -o parallelepiped
>
> Ok, now it's easier to see what's going wrong.  The problem is that the
> equation may hold, but it cannot be expressed using IntVars because of the
> limited domains.  In your decomposition, the first constraint that exposes
> the problem is
>
>            sqr(*this,k,kk);
>
> At this point, k is already at least 2000000, but kk was initialized as an
> IntVar with domain Int::Limits::min..Int::Limits::max. Therefore, the
> problem has no solution, as 2000000*2000000 > Int::Limits::max.  This is
> expected and correct behavior.
> Unfortunately, the minimodel layer lets you express constraints that are
> (semantically) satisfiable, but the introduction of intermediate IntVars
> makes them unsatisfiable.
> For your actual problem, there's no way around this limitation except
> implementing IntVars with arbitrary precision (using e.g. the gmp library).
>  This would be possible, but it's not something we're planning to do anytime
> soon.
>
> Cheers,
>        Guido
>
> --
> Guido Tack, http://people.cs.kuleuven.be/~guido.tack/
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gecode.org/pipermail/users/attachments/20100820/e9caf5d8/attachment.htm>