arXiv study maps when exponential weights algorithms settle on Nash equilibria
A new analysis of repeated games shows constant learning rates drive exponential weights play toward strict equilibria in coordination settings, with caveats.
In plain words
Researchers studied a classic learning rule where players in a repeated game adjust their strategy mixtures using a fixed step size. They prove that, under certain conditions, this process settles on stable strategy profiles known as strict Nash equilibria, though not always.
What was announced
The paper, posted on arXiv, examines the last-iterate convergence of the exponential weights algorithm with constant learning rates in repeated discrete-time interactions. Each player holds an initial mixed action and a fixed learning rate, making the sequence of played mixed actions a homogeneous Markov chain. The authors first show that when a strict Nash equilibrium exists, the probability of playing one at the next stage converges almost surely to either 0 or 1. They further show that any limit of the mixed action profile, when it exists, lies within the set of Nash Equilibria with Equalizing Payoffs.
For strong coordination games, where a player's payoff is positive on the diagonal and zero elsewhere, the analysis gives a sharper result: the mixed action profile converges almost surely to one of the strict Nash equilibria. The authors close by listing open questions rather than claiming the general case is settled.
What it does not show
- The evidence is the paper's abstract only; full proofs, assumptions and the open questions are not covered here.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.