Explicit Sequential Equilibria in LQ Deep Structured Games and Weighted Mean-Field Games
Abstract
In this paper, we investigate a class of nonzero-sum dynamic stochastic games, where players have linear dynamics and quadratic cost functions. The players are coupled in both dynamics and cost through a linear regression (weighted average) as well as a quadratic regression (weighted covariance matrix) of the states and actions, where the linear regression of states is called deep state. We study collaborative and non-collaborative games under three information structures: perfect sharing, deep state sharing, and no sharing for three different types of weights: positive, homogeneous and asymptotically vanishing weights. For perfect and deep state sharing information structures, we propose a transformation-based technique to solve for the best-response equations of players and identify a few sufficient conditions under which a unique subgame perfect Nash equilibrium exists. The equilibrium is linear in the local state and deep state, and the corresponding gains are obtained by solving a novel non-standard Riccati equation (whose dimension is independent of the number of players, thus making the solution scalable). When the information structure is no-sharing and the number of players is asymptotically large, one approximate population-size-dependent equilibrium and one approximate population-size-independent equilibrium (also called sequential weighted mean-field equilibrium) are proposed, and their convergence to the infinite-population limits are established. In addition, the main results are extended to infinite-horizon cost function, and generalized to multiple orthogonal linear regressions and heterogeneous sub-populations. A numerical example is provided to demonstrate the difference between the two proposed approximate equilibria.
Keywords
Cite
@article{arxiv.1912.03931,
title = {Explicit Sequential Equilibria in LQ Deep Structured Games and Weighted Mean-Field Games},
author = {Jalal Arabneydi and Amir G. Aghdam and Roland P. Malhamé},
journal= {arXiv preprint arXiv:1912.03931},
year = {2020}
}
Comments
Extended version with 22 pages. To appear in Automatica. Initial draft was submitted in 2018