1: \begin{abstract}
2: We present the first general bounds on the mixing time of the Markov chain associated to the logit dynamics for wide classes of strategic games. The
3: logit dynamics with inverse noise $\beta$ describes the behaviour of a complex system whose individual components act selfishly and keep responding according to some
4: partial (“noisy”) knowledge of the system, where the capacity of the agent to know the system and compute her best move is measured by the inverse of the parameter $\beta$.
5:
6: In particular, we prove nearly tight bounds for potential games and games with dominant strategies. Our results show that, for potential games,
7: the mixing time is upper and lower bounded by an exponential in the inverse of
8: the noise and in the maximum potential difference. Instead,
9: for games with dominant strategies, the mixing time cannot
10: grow arbitrarily with the inverse of the noise.
11:
12: Finally, we refine our analysis for a subclass of potential games called
13: graphical coordination games, a class of games that have been previously
14: studied in Physics and, more recently, in Computer Science
15: in the context of diffusion of new technologies. We give evidence that the
16: mixing time of the logit dynamics for these games strongly depends on the
17: structure of the underlying graph. We prove that the mixing time of the logit dynamics
18: for these games can be upper bounded by a function that is exponential in the cutwidth
19: of the underlying graph and in the inverse of noise.
20: Moreover, we consider two specific and popular network topologies, the clique and the
21: ring. For games played on a clique we prove an almost matching lower bound on the
22: mixing time of the logit dynamics that is exponential in the inverse of
23: the noise and in the maximum potential difference, while for games played on a ring
24: we prove that the time of convergence of the logit dynamics to its stationary distribution is significantly shorter.
25: \end{abstract}
26: