CMU-IBM
Open source MINLP Project

 



Goal
of this project is to produce novel open source software for solving mixed-integer nonlinear programs (MINLP) with convex relaxation.

The main objectives of this effort are:
(i) a new publicly available library of test instances of convex MINLPs
(ii) a new software package bonmin that contains classical methods such as branch-and-bound and outer approximation, and a new family of hybrid algorithms of which branch-and-bound and outer-approximation are two extreme cases
(iii) the new framework is open source and uses existing software in COIN-OR. In particular Clp, Cbc and Ipopt are used as building blocks.
(iv) computational results on the new and other available libraries of test problems.
 

Participants


Professor Gerard Cornuejols
Professor Larry Biegler
Professor Ignacio Grossmann
Professor Francois Margot

Lehigh University
Dr. Pietro Belotti


Dr. Pierre Bonami
Dr. Andrew R Conn
Dr. Jon Lee
Dr. Laszlo Ladanyi
Dr. Andreas Waechter

Dr. Carld Laird, Texas A&M

Dr. Nick Sawaya, ExxonMobil

Additional collaborators:

Dr. Leo Liberti, Ecole Polytechnique, Palaiseau (FR)

Dr. Andrea Lodi, Bologna

 

Test Problems

Batch (GAMS) (.NL)
CLay (GAMS) (.NL)
FLay (GAMS) (.NL)
fo (GAMS) (.NL)
RSyn (GAMS) (.NL)
SLay (GAMS) (.NL)
Syn (GAMS) (.NL)
trimloss (.NL)
Water (.NL)

IBM Research Reports:
"An Algorithmic Framework for Convex Mixed Integer Nonlinear Programs"

"A Feasibility Pump for Mixed Integer Nonlinear Programs"

"Branching and Bounds Tightening Techniques for Non-Convex MINLP"

 

Useful Links

Software used in this project:
bonmin
CLP COIN-OR
CBC COIN-OR
IPOPT

Other Software:
alpha-ECP Abo Akademi
MINLP Argonne
GAMS World-MINLP
MINOPT-Princeton

Participating Institutions:
IBM Research
IBM Mathematical Sciences