Selfishness need not be bad
This article studies the user selfish behavior in non-atomic congestion games (NCG). We prove that the price of anarchy of general NCGs tends to 1 as number of users tends to infinity. This generalizes a recent result in the literature. Although our result is general, the proof appears simpler. For routing games with BPR travel time functions, we prove that every system optimum strategy profile is an ϵ-approximate Nash equilibrium, where ϵ is a small constant depending on the travel demands. Moreover, we prove that the price of anarchy of these games equal 1+O(T^-β), where T is the total travel demand and β is the degree of the BPR functions. This confirms a conjecture proposed by O'Here et al. In addition, we proved that the cost of both, system optimum and Nash equilibrium, depends mainly on the distribution of users among OD pairs, when the total travel time is large. This does not only supply an approximate method for computing these cost, but also give insights how to reduce the total travel time, when the total travel demand is large. To empirically verify our theoretical findings, we have taken real traffic data within the 2nd ring road of Beijing as an instance in an experimental study. Our empirical results definitely validate our findings. In addition, they show that the current traffic in Beijing within that area is already far beyond saturation, and no route guidance policy can significantly reduce the total travel time for the current huge total travel demand. In summary, selfishness in a congestion game with a large number of users need not be bad. It may be the best choice in a bad environment.
READ FULL TEXT