Equilibrium Computation in Atomic Splittable Routing Games with Convex Cost Functions

TL;DR

This paper introduces the first polynomial-time algorithms for computing equilibria in atomic splittable routing games with convex cost functions, expanding beyond previous affine or symmetric cases, and establishes NP-hardness results for general networks.

Contribution

The paper presents novel algorithms for equilibrium computation in general convex cost routing games and proves NP-hardness for certain decision problems in these settings.

Findings

First algorithms for convex cost functions in routing games.

Polynomial-time algorithms when either players or edges are few.

NP-hardness of equilibrium cost decision in general networks.

Abstract

We present polynomial-time algorithms as well as hardness results for equilibrium computation in atomic splittable routing games, for the case of general convex cost functions. These games model traffic in freight transportation, market oligopolies, data networks, and various other applications. An atomic splittable routing game is played on a network where the edges have traffic-dependent cost functions, and player strategies correspond to flows in the network. A player can thus split it's traffic arbitrarily among different paths. While many properties of equilibria in these games have been studied, efficient algorithms for equilibrium computation are known for only two cases: if cost functions are affine, or if players are symmetric. Neither of these conditions is met in most practical applications. We present two algorithms for routing games with general convex cost functions on parallel links. The first algorithm is exponential in the number of players, while the second is exponential in the number of edges; thus if either of these is small, we get a polynomial-time algorithm. These are the first algorithms for these games with convex cost functions. Lastly, we show that in general networks, given input $C$, it is NP-hard to decide if there exists an equilibrium where every player has cost at most $C$.