Lagrangian relaxation is commonly used to generate bounds for mixed-integer linear programming problems. However, when the number of dualized constraints is very large (exponential in the dimension of the primal problem), explicit dualization is no longer possible. In order to reduce the dual dimension, different heuristics were proposed. They involve a separation procedure to dynamically select a restricted set of constraints to be dualized along the iterations. This relax-and-cut type approach has shown its numerical efficiency in many combinatorial problems. We show Primal-dual convergence of such strategy when using an adapted subgradient method for the dual step.
Instituto Nacional de Matem\'atica Pura e Aplicada, 2008