Forcing to use the interior-point-convex algorithm which
produce more stable results. Still, it is not in the bounds.

Commented the normalization in the rational function class,
it seems to introduce some errors (weird).