Abstract Three Lagrangean methods for 0–1 quadratic programming are presented. The first one decomposes the initial problem into a continuous nonlinear subproblem and a 0–1 linear subproblem. The second decomposes the initial problem into a 0–1 quadratic problem without constraints and the same 0–1 linear subproblem. The last method is a Lagrangean relaxation which dualizes all the constraints. We try to show how a 0–1 quadratic problem may be analysed in order to choose the appropriate method. We also furnish reformulations of the dual problems which make them easier to solve. This is particularly true for the first decomposition for which we show that there is no need to solve the continuous nonlinear subproblem. We also give some new primal interpretations for the second decomposition and for the relaxation.
[1]
Michel Minoux,et al.
Graphes et algorithmes
,
1995
.
[2]
Monique Guignard-Spielberg,et al.
Lagrangean decomposition: A model yielding stronger lagrangean bounds
,
1987,
Math. Program..
[3]
A. M. Geoffrion.
Lagrangean Relaxation and Its Uses in Integer Programming
,
1972
.
[4]
Catherine Roucairol,et al.
Problème de la bipartition minimale d'un graphe
,
1987
.
[5]
P. Hansen,et al.
Best network flow bounds for the quadratic knapsack problem
,
1989
.
[6]
H. D. Ratliff,et al.
Minimum cuts and related problems
,
1975,
Networks.