0000000000451515
AUTHOR
Gerhard Reinelt
Branch-and-Bound
We now turn to the discussion of how to solve the linear ordering problem to (proven) optimality. In this chapter we start with the branch-and-bound method which is a general procedure for solving combinatorial optimization problems. In the subsequent chapters this approach will be realized in a special way leading to the so-called branch-and-cut method. There are further possibilities for solving the LOP exactly, e.g. by formulating it as dynamic program or as quadratic assignment problem, but these approaches did not lead to the implementation of practical algorithms and we will not elaborate on them here.
Branch-and-Cut
This chapter focuses on the approach for solving the LOP to optimality which can currently be seen as the most successful one. It is a branch-and-bound algorithm, where the upper bounds are computed using linear programming relax- ations.
The Linear Ordering Polytope
So far we developed a general integer programming approach for solving the LOP. It was based on the canonical IP formulation with equations and 3-dicycle inequalities which was then strengthened by generating mod-k-inequalities as cutting planes. In this chapter we will add further ingredients by looking for problem- specific inequalities. To this end we will study the convex hull of feasible solutions of the LOP: the so-called linear ordering polytope.