Accelerating Column Generation for a Survivable Network Design Problem

We consider a network design problem occurring in the extension of fiber optic networks on the last mile which generalizes the (Price Collecting) Steiner Tree Problem by introducing redundancy requirements on some customer nodes. In this work we present a formulation for this problem based on exponentially many variables and solve its linear relaxation by column generation. Using alternative dual-optimal solutions in the pricing problem we are able to significantly reduce the effects of typical efficiency issues of simplex based column generation. Computational results clearly show the advantages of our proposed strategy with respect to the number of pricing iterations needed as well as by means of required running times.