Solving Mean-Field Games with Monotonicity Methods in Banach Spaces

TL;DR

This paper introduces a unified Banach-space framework using monotonicity methods to establish the existence of strong solutions for stationary mean-field games, overcoming previous limitations of Hilbert-space approaches.

Contribution

It develops a low-order regularization approach with monotonicity techniques to prove existence of strong solutions for a broad class of mean-field games, including singular congestion models.

Findings

Established existence of strong solutions for standard and singular congestion MFGs.

Unified framework extends previous results and avoids high-order regularization.

Suitable for modern numerical algorithms.

Abstract

This paper develops a unified framework for proving the existence of solutions to stationary first-order mean-field games (MFGs) based on the theory of monotone operators in Banach spaces. We cast the coupled MFG system as a variational inequality, overcoming the limitations of prior Hilbert-space approaches that relied on high-order regularization and typically yielded only weak solutions in the monotone operator sense. In contrast, with our low-order regularization, we obtain strong solutions. Our approach addresses the non-coercivity of the underlying MFG operator through two key regularization strategies. First, by adding a low-order $p$-Laplacian term, we restore coercivity, derive uniform a priori estimates, and pass to the limit via Minty's method. This establishes, for the first time via monotonicity methods, the existence of strong solutions for models with both standard power-growth and singular congestion, with the latter requiring a careful restriction of the operator's domain. Second, for Hamiltonians with only minimal growth hypotheses, we regularize the Hamiltonian itself via infimal convolution to prove the existence of weak solutions. Our Banach-space framework unifies and extends earlier existence results. By avoiding high-order smoothing, it not only provides a more direct theoretical path but is also ideally suited for modern numerical algorithms.