Some Recent Advances in Mixed-Integer Nonlinear Programming
In this talk we review recent research activities in the context of the open source Bonmin (Basic Open-source Nonlinear Mixed-INteger programming) project at COIN-OR. The goal of this project is to explore a variety of solution techniques suitable for mixed-integer nonlinear programming (MINLP).
We discuss several algorithms, such as a "hybrid" algorithm (a branch-and-cut method using outer approximation cuts), a branch-and-bound method with different branching rules, and nonlinear heuristics. Numerical experiments include results from a parallel implementation with thousands of processors. We also consider a spatial branch-and-bound method for nonconvex problems that works with convex underestimators.
Andreas Wächter, IBM T.J. Watson Research Center