Return to Colloquia & Seminar listing
The Importance of Being... Separable
PDE & Applied Mathematics| Speaker: | Claudia D'Ambrosio , Lab d'Informatique de l'Ecole polytechnique |
| Location: | 2112 MSB |
| Start time: | Mon, Nov 17 2014, 4:10PM |
Description
In this talk we present two exact approaches that exploit in
very different ways the separability of mixed integer nonlinear
programming (MINLP) problems.
On one side, a global optimization method for box-constrained polynomial
optimization is proposed. The main idea is to identify separable
under-estimators to obtain quickly good quality lower bounds employed in a
branch-and-bound scheme.
On the other side, a global optimization method for MINLP problems with
separable non-convexities is proposed. Convex (and linear) lower bounds
are proposed and employed in an iterative method that is proven to
converge to
(epsilon) optimality.
Computational results are provided so as to show the practical efficiency
of the two algorithms with respect to the classic resolution methods.
Also part of the Optimization Seminar Series