Projected Chvátal-Gomory cuts for mixed integer linear programs

Recent experiments by Fischetti and Lodi show that the first Chvátal closure of a pure integer linear program (ILP) often gives a surprisingly tight approximation of the integer hull. They optimize over the first Chvátal closure by modeling the Chvátal–Gomory (CG) separation problem as a mixed integer linear program (MILP) which is then solved by a generalGérard Cornuéjols was supported in part by NSF grant DMI-0352885, ONR grant N00014-03-1-0188, and ANR grant BLAN 06-1-138894. Matteo Fischetti was supported in part by the EU projects ADONET (contract n. MRTN-CT-2003-504438) and ARRIVAL (contract n. FP6-021235-2). Andrea Lodi was supported in part by the EU projects ADONET (contract n. MRTN-CT-2003-504438) and ARRIVAL (contract n. FP6-021235-2). P. Bonami · S. Dash IBM T.J. Watson Research Center, P.O. Box 218, Yorktown Heights, NY 10598, USA e-mail: [email protected] S. Dash e-mail: [email protected] G. Cornuéjols (B) Tepper School of Business, Carnegie Mellon University, Pittsburgh, PA 15213, USA G. Cornuéjols LIF, Faculté des Sciences de Luminy, 13288 Marseille, France e-mail: [email protected] M. Fischetti DEI, University of Padova, via Gradenigo 6A, 35131 Padova, Italy e-mail: [email protected] A. Lodi DEIS, University of Bologna, viale Risorgimento 2, 40136 Bologna, Italy e-mail: [email protected] 242 P. Bonami et al. purpose MILP solver. Unfortunately, this approach does not extend immediately to the Gomory mixed integer (GMI) closure of an MILP, since the GMI separation problem involves the solution of a nonlinear mixed integer program or a parametric MILP. In this paper we introduce a projected version of the CG cuts, and study their practical effectiveness for MILP problems. The idea is to project first the linear programming relaxation of the MILP at hand onto the space of the integer variables, and then to derive Chvátal–Gomory cuts for the projected polyhedron. Though theoretically dominated by GMI cuts, projected CG cuts have the advantage of producing a separation model very similar to the one introduced by Fischetti and Lodi, which can typically be solved in a reasonable amount of computing time.