Paper: arXiv 2510.20017

Authors: Dena Firoozi, Anastasis Kratsios, Xuwei Yang

Abstract

Traditional mean-field game (MFG) solvers operate on an instance-by-instance basis, which becomes infeasible when many related problems must be solved (e.g., for seeking a robust description of the solution under perturbations of the dynamics or utilities, or in settings involving continuum-parameterized agents.). We overcome this by training neural operators (NOs) to learn the rules-to-equilibrium map from the problem data (``rules’’: dynamics and cost functionals) of LQ MFGs defined on separable Hilbert spaces to the corresponding equilibrium strategy. Our main result is a statistical guarantee: an NO trained on a small number of randomly sampled rules reliably solves unseen LQ MFG variants, even in infinite-dimensional settings. The number of NO parameters needed remains controlled under appropriate rule sampling during training. Our guarantee follows from three results: (i) local-Lipschitz estimates for the highly nonlinear rules-to-equilibrium map; (ii) a universal approximation theorem using NOs with a prespecified Lipschitz regularity (unlike traditional NO results where the NO’s Lipschitz constant can diverge as the approximation error vanishes); and (iii) new sample-complexity bounds for $L$-Lipschitz learners in infinite dimensions, directly applicable as the Lipschitz constants of our approximating NOs are controlled in (ii).

Complexity vs Empirical Score

  • Math Complexity: 9.5/10
  • Empirical Rigor: 2.0/10
  • Quadrant: Lab Rats — theoretically deep, empirically untested

Why this score: This paper presents highly advanced mathematical theory establishing PAC-learnability and Lipschitz bounds for neural operators applied to infinite-dimensional mean field games. However, it is entirely theoretical without empirical backtesting, code, or numerical experiments provided in the excerpt.

Research Flowchart

  flowchart TD
    A["Research Goal: Solve infinitely many LQ Mean Field Games (MFGs) simultaneously using Neural Operators"] --> B["Input Problem Data: LQ MFG Rules<br/>(Dynamics & Cost Functionals in Hilbert Spaces)"]
    B --> C["Core Methodology: Train Neural Operator (NO)<br/>to map Rules to Equilibrium Strategies"]
    C --> D["Theoretical Analysis & Bounds:<br/>1. Local-Lipschitz estimates of Rules-to-Equilibrium map<br/>2. Universal Approximation with bounded NO Lipschitz regularity<br/>3. Infinite-dimensional L-Lipschitz sample complexity"]
    D --> E["Key Outcome: Statistical Guarantee & Generalization<br/>NO reliably predicts equilibria for unseen, infinite-dimensional MFG variants"]

Cited References (87)

  1. Learning algorithms for mean field optimal control (2025) — link
  2. Unsupervised solution operator learning for mean-field games (2025) — link
  3. Physics-Informed Graph Neural Operator for Mean Field Games on Graph: A Scalable Learning Approach (2024)
  4. A Mean Field Game Approach to Equilibrium Pricing with Market Clearing Condition (2022)
  5. A mean-field game approach to price formation in electricity markets (2021)
  6. A mean-field game approach to equilibrium pricing in solar renewable energy certificate markets (2022)
  7. Modern statistical estimation via oracle inequalities (2006)
  8. Deep nonparametric regression on approximate manifolds: Nonasymptotic error bounds with polynomial prefactors (2023)
and 79 more
  1. Foundations of machine learning (2018)
  2. Optimal approximation rate of ReLU networks in terms of width and depth (2022)
  3. Designing universal causal deep learning models: The case of infinite-dimensional dynamical systems from stochastic analysis (2022) — link
  4. Bridging the gap between approximation and learning via optimal approximation by ReLU MLPs of maximal regularity (2024) — link
  5. Low-dimensional approximations of the conditional law of Volterra processes : a non-positive curvature approach — → 2405.20094
  6. Notes on the Wasserstein metric in Hilbert spaces (1989)
  7. Out-of-distributional risk bounds for neural operators with applications to the Helmholtz equation (2024)
  8. Designing universal causal deep learning models: The geometric (hyper) transformer (2024)
  9. Neural Operators Can Play Dynamic Stackelberg Games — → 2411.09644
  10. Probabilistic Analysis of Mean-Field Games (2013) — link
  11. Probabilistic Theory of Mean Field Games with Applications I Mean Field FBSDEs, Control, and Games (2018)
  12. Simultaneously Solving FBSDEs with Neural Operators of Logarithmic Depth, Constant Width, and Sub-Linear Rank (2024)
  13. Linear-Quadratic Mean Field Games in Hilbert spaces (2024)
  14. On Mean Field Games in Infinite Dimension (2024) — link
  15. Hilbert Space-Valued LQ Mean Field Games: An Infinite-Dimensional Analysis (2025)
  16. Integral probability metrics pac-bayes bounds (2022)
  17. Learning from One Graph: Transductive Learning Guarantees via the Geometry of Small Random Worlds (2025) — link
  18. Neural Operators Can Play Dynamic Stackelberg Games — → 2411.09644
  19. Dynamic programming approach to stochastic evolution equations (1979)
  20. Some remarks on the R iccati equation arising in an optimal control problem with state-and control-dependent noise (1992)
  21. Quadratic Optimal Control of Graphon Q-noise Linear Systems (2024) — link
  22. Linear-Quadratic-Gaussian Mixed Games with Continuum-Parametrized Minor Players (2012) — link
  23. Large-population cost-coupled LQG problems with nonuniform agents: individual-mass behavior and decentralized $\varepsilon$- Nash equilibria (2007)
  24. Large-population LQG games involving a major player: the Nash certainty equivalence principle (2010)
  25. Nonlinear approximation and (deep) ReLU networks (2022)
  26. Nonlocal techniques for the analysis of deep ReLU neural network approximations (2025) — link
  27. Tighter generalization bounds on digital computers via discrete optimal transport (2024) — link
  28. Instance-dependent generalization bounds via optimal transport (2023)
  29. Approximation theory for 1-Lipschitz ResNets (2025) — link
  30. Generating Rectifiable Measures through Neural Networks (2024) — link
  31. Operator learning with PCA-Net: upper and lower complexity bounds (2023)
  32. PCANet: A simple deep learning baseline for image classification? (2015)
  33. Markovian lifting and asymptotic log-Harnack inequality for stochastic Volterra integral equations (2024)
  34. Generalized Feller processes and Markovian lifts of stochastic Volterra processes: the affine case (2020)
  35. Markovian lifts of positive semidefinite affine Volterra-type processes (2019)
  36. Physics-informed neural operator for learning partial differential equations (2024)
  37. Exponential convergence of deep operator networks for elliptic partial differential equations (2023)
  38. Optimal deep learning of holomorphic operators between Banach spaces (2024)
  39. Physics-informed deep neural operator networks (2023)
  40. Generic bounds on the approximation error for physics-informed (and) operator learning (2022)
  41. Neural operators for accelerating scientific simulations and design (2024)
  42. Deep learning architectures for nonlinear operator functions and nonlinear inverse problems (2022)
  43. Learning the solution operator of parametric partial differential equations with physics-informed DeepONets (2021)
  44. A convergence rate for the Newton’s method for mean field games with non-separable Hamiltonians (2023) — link
  45. Constructive approximation: advanced problems
  46. Simultaneously solving fbsdes with neural operators of logarithmic depth, constant width, and sub-linear rank — → 2410.14788
  47. Connecting GANs, mean-field games, and optimal transport (2024)
  48. Deep learning methods for mean field control problems with delay (2020)
  49. Numerical resolution of McKean-Vlasov FBSDEs using neural networks (2022)
  50. Convergence analysis of machine learning algorithms for the numerical solution of mean field control and games I: The ergodic case (2021)
  51. On the implementation of a primal-dual algorithm for second order time-dependent mean field games with local couplings (2019)
  52. Cemracs 2017: numerical probabilistic approach to MFG (2019)
  53. Individual vaccination as Nash equilibrium in a SIR model with application to the 2009–2010 influenza A (H1N1) epidemic in France (2015)
  54. Density flow in dynamical networks via mean-field games (2016)
  55. Mean field games (2007)
  56. Mean field games and applications: Numerical aspects (2020)
  57. Learning in mean field games: A survey (2022) — link
  58. Mean-field theory of two-layers neural networks: dimension-free bounds and kernel limit (2019)
  59. On the mean field theory of Ensemble Kalman filters for SPDEs (2025)
  60. Opinion dynamics in social networks through mean-field games (2016)
  61. A mean field theory of nonlinear filtering (2008)
  62. The mean-field ensemble Kalman filter: near-Gaussian setting (2024)
  63. Applications of mean field games in financial engineering and economic theory (2020) — link
  64. Deep learning algorithms for hedging with frictions (2023) — link
  65. An Adaptive Shifted Power Method for Computing Generalized Tensor Eigenpairs (2014) — link
  66. SIAM Style Manual: For journals and books (2013) — link
  67. A Call for Better Indexes (2014) — link
  68. Accelerating Community Detection by Using K -core Subgraphs (2014) — link
  69. Numerical Solution of the Bloch Equations Provides Insights into the Optimum Design of PARACEST Agents for MRI (2005) — link
  70. Properties of Highly Clustered Networks (2003) — link
  71. Clawpack Software (2015) — link
  72. Mathematics Subject Classification (2010) — link
  73. \LaTeX: A Document Preparation System (1986)
  74. The \LaTeX\ Companion (2004)
  75. Matrix Computations (2013)
  76. Paul’s Online Math Notes: Calculus I — Notes — link
  77. User’s Guide for the \texttt amsmath Package (Version 2.0) (2002) — link
  78. Short Math Guide for \LaTeX (2002) — link
  79. Manual for Package \texttt PGFPLOTS (2015) — link

4 of 87 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.