Abstract. We present an evaluation of a Lagrangean-based branch-andbound algorithm with additional valid inequalities for the capacitated network design problem. The focus is on two types of valid inequalities, the cover inequalities and local cuts. We show how these inequalities can be considered in a Lagrangean relaxation without destroying the computationally simple structure of the subproblems. We present an extensive computational study on a large set of benchmark data. The results show that the presented algorithm outperforms many other exact and heuristical solvers in terms of running time and solution quality.

Abstract. We present a fully polynomial-time approximation scheme for a multicommodity flow problem that yields lower bounds of the graph bisection problem. We compare the approximation algorithm with Lagrangian relaxation based cost-decomposition approaches and linear programming software when embedded in an exact branch&bound approach for graph bisection. It is shown that the approximation algorithm is clearly superior in this context. Furthermore, we present a new practical addition to the approximation algorithm which improves its performance distinctly. Finally, we prove the performance of the graph bisection algorithm using multicommodity flow approximation by computing formerly unknown bisection widths of some DeBruijn-and Shuffle-Exchange-Graphs.