Games Played by Exponential Weights Algorithms
Abstract
This paper studies the last-iterate convergence properties of the exponential weights algorithm with constant learning rates. We consider a repeated interaction in discrete time, where each player uses an exponential weights algorithm characterized by an initial mixed action and a fixed learning rate so that the mixed action profile played at stage t follows an homogeneous Markov chain. At first, we show that the probability to play a strict Nash equilibrium at the next stage converges almost surely to zero or one. Second, we show that the limit of , whenever it exists, belongs to the set of Nash equilibria with equalizing payoffs. Finally, we show that in strong coordination games, where the payoff of a player is positive on the diagonal and zero elsewhere, converges almost surely to one of the strict Nash equilibria.
Funding: This research benefited from financial support from the AI Institute Artificial and Natural Intelligence Toulouse Institute (ANITI) [Grants ANR-19-PI3A-0004 and ANR-23-IACL-P0002] and the ANR Chess [Grant ANR-17-EURE-0010]. J. Renault acknowledges funding from the Agence nationale de la recherche (ANR) [Grant MaSDOL-19-CE23-0017-01].

