GRASP and VNS for Max-Cut

In this abstract, we study GRASP and VNS heuristics for Max-Cut, together with a combination of the two in a hybrid procedure. GRASP [3, 4, 5] is characterized by a construction phase and a local search phase. The iterative construction builds a solution, one element at a time, randomly selected from a restricted candidate list (RCL), whose elements are well-ranked according to a greedy function. Once a feasible solution is obtained, a local search procedure tries to improve it producing a solution that is locally optimal with respect to the specified neighborhood structure. The construction and the local search phases are repeatedly applied, and the best solution found is kept as an approximation of the optimal. VNS [9] is based on the exploration of a dynamic neighborhood model.