The implementation follows division of search space by inclusion/exclusion of edges selected by a criterion that maximizes early cuts of subspaces of the search space.
A good algorithm explanation is at http://www.cs.berkeley.edu/~demmel/cs267/assignment4.html