Sorting Heuristics for the Feedback Arc Set Problem

The feedback arc set problem plays a prominent role in the four-phase framework to draw directed graphs, also known as the Sugiyama algorithm. It is equivalent to the linear arrangement problem where the vertices of a graph are ordered from left to right and the backward arcs form the feedback arc set. In this paper we extend classical sorting algorithms to heuristics for the feedback arc set problem. Established algorithms are considered from this point of view, where the directed arcs between vertices serve as binary comparators. We analyze these algorithms and afterwards design hybrid algorithms by their composition in order to gain further improvements. These algorithms primarily differ in the use of insertion sort and sifting and they are very similar in their performance, which varies by about 0.1%. The differences mainly lie in their run time and their convergence to a local minimum. Our studies extend related work by new algorithms and our experiments are conducted on much larger graphs. Overall we can conclude that sifting performs better than insertion sort.