A generalized formulation for vehicle routing problems

Different types of formulations are proposed in the literature to model vehicle routing problems. Currently, the most used ones can be fitted into two classes, namely vehicle flow formulations and set partitioning formulations. These types of formulations differ from each other not only due to their variables and constraints but also due to their main features. Vehicle flow formulations have the advantage of being compact models, so general-purpose optimization packages can be used to straightforwardly solve them. However, they typically show weak linear relaxations and have a large number of constraints. Branch-and-cut methods based on specialized valid inequalities can also be devised to solve these formulations, but they have not shown to be effective for large-scale instances. On the other hand, set partitioning formulations have stronger linear relaxations, but requires the implementation of sophisticate techniques such as column generation and specialized branch-and-price methods. Due to all these reasons, so far it is has been recognized in the vehicle routing community that these two types of formulations are rather different. In this paper, we show that they are actually strongly related as they correspond to special cases of a generalized formulation of vehicle routing problems.