Semidefinite programming and its applications to NP problems

Abstract

The graph homomorphism problem is a canonical NP-complete problem in a sense that it generalizes various other well-studied problems such as graph coloring and finding cliques. To get a better insight into a combinatorial problem, one often studies relaxations of the problem. We define fractional homomorphisms and pseudo-homomorphisms as natural relaxations of graph homomorphisms. In their paper [4], Feige and Lovász defined a semidefinite relaxation of the homomorphism problem, which allowed them to obtain polynomial time algorithms for certain special cases of the problem. Their relaxation is defined in terms of the solution to a semidefinite program. Hence a characterization of their relaxation in terms of known combinatorial notions is desirable. We show that our pseudo-homomorphism is equivalent to the relaxation defined by Feige and Lovász [4]. Although general graph homomorphism does not admit a simple forbidden subgraph characterization, surprisingly we can show that there is a simple forbidden subgraph characterization of the fractional homomorphism (the forbidden subgraph is a clique in this case). As a byproduct, we obtain a simpler proof of the NP hardness of the fractional chromatic number, first proved by Grötschel, Lovász and Schrijver using the ellipsoid method [6] Finally, we briefly discuss how to apply these techniques to general NP problems and describe a unified setting in which a wide variety of seemingly disparate polynomial time problems can be decided.