Decomposing the Train-Scheduling Problem into Integer-Optimal Polytopes

This paper presents conditions for which the linear relaxation for the train-scheduling problem is integer optimal. These conditions are then used to identify how to partition a general problem’s feasible region into integer-optimal polytopes. Such an approach yields an extended formulation that contains far fewer binary variables. Our computational experiments show that this approach results in significant computational savings. Moreover, this approach scales well when the train-scheduling problem is modeled using smaller time increments, allowing for higher fidelity models to be solved without significantly increasing the required computational time.