Well-Supported versus Approximate Nash Equilibria: Query Complexity of Large Games
Abstract
We study the randomized query complexity of approximate Nash equilibria (ANE) in large games. We prove that, for some constant , any randomized oracle algorithm that computes an -ANE in a binary-action, -player game must make payoff queries. For the stronger solution concept of well-supported Nash equilibria (WSNE), Babichenko previously gave an exponential lower bound for the randomized query complexity of -WSNE, for some constant ; the same lower bound was shown to hold for -ANE, but only when . Our result answers an open problem posed by Hart and Nisan and Babichenko and is very close to the trivial upper bound of . Our proof relies on a generic reduction from the problem of finding an -WSNE to the problem of finding an -ANE, in large games with actions, which might be of independent interest.
Keywords
Cite
@article{arxiv.1511.00785,
title = {Well-Supported versus Approximate Nash Equilibria: Query Complexity of Large Games},
author = {Xi Chen and Yu Cheng and Bo Tang},
journal= {arXiv preprint arXiv:1511.00785},
year = {2015}
}
Comments
10 pages