Heuristic Improvements for Computing Maximum Multicommodity Flow and Minimum Multicut

摘要

We propose heuristics to reduce the number of shortest path computations required to compute a 1+ε approximation to the maximum multicommodity flow in a graph. Through a series of improvements we are able to reduce the number of shortest path computations significantly. One key idea is to use the value of the best multicut encountered in the course of the algorithm. For almost all instances this multicut is significantly better than that computed by rounding the linear program.