Preference Games and Personalized Equilibria, with Applications to Fractional BGP

We study the complexity of computing equilibria in two classes of network games based on flows - fractional BGP (Border Gateway Protocol) games and fractional BBC (Bounded Budget Connection) games. BGP is the glue that holds the Internet together and hence its stability, i.e. the equilibria of fractional BGP games (Haxell, Wilfong), is a matter of practical importance. BBC games (Laoutaris et al) follow in the tradition of the large body of work on network formation games and capture a variety of applications ranging from social networks and overlay networks to peer-to-peer networks. The central result of this paper is that there are no fully polynomial-time approximation schemes (unless PPAD is in FP) for computing equilibria in both fractional BGP games and fractional BBC games. We obtain this result by proving the hardness for a new and surprisingly simple game, the fractional preference game, which is reducible to both fractional BGP and BBC games. We define a new flow-based notion of equilibrium for matrix games -- personalized equilibria -- generalizing both fractional BBC and fractional BGP games. We prove not just the existence, but the existence of rational personalized equilibria for all matrix games, which implies the existence of rational equilibria for fractional BGP and BBC games. In particular, this provides an alternative proof and strengthening of the main result in [Haxell, Wilfong]. For k-player matrix games, where k = 2, we provide a combinatorial characterization leading to a polynomial-time algorithm for computing all personalized equilibria. For k >= 5, we prove that personalized equilibria are PPAD-hard to approximate in fully polynomial time. We believe that the concept of personalized equilibria has potential for real-world significance.