Building an effective solver for convex mixed integer nonlinear programs

Monday, November 17, 2008 - 9:00am - 10:00am
EE/CS 3-180
Jeff Linderoth (University of Wisconsin, Madison)
The most effective solvers for mixed integer linear programs (MILP)s
employ a variety of algorithmic refinements that have made previously
intractable models routinely solvable. Solvers for Mixed Integer
Nonlinear Programs (MINLP)s should be no different. In this talk, we
will discuss the impact of applying advanced branching rules, primal
heuristics, preprocessing, and cutting planes in algorithms to solve
(convex) mixed integer nonlinear programs. Many of the ideas have
been implemented in the solver FilMINT, and the talk contains
computational results to demonstrate the improvements that can be
obtained by applying traditional MILP techniques for MINLPs.
MSC Code: