mean-field analysis

This work addresses the computational challenge of efficiently solving for Nash equilibria in continuous-time, finite-state mean-field games—a setting where existing methods often struggle. The paper introduces, for the first time, the concept of regularization from discrete-time formulations into this continuous-time framework, thereby constructing a computable regularized equilibrium. It further extends fixed-point iteration and fictitious play algorithms to this setting, leveraging a continuous-time Markov chain model to significantly enhance computational tractability. Numerical experiments demonstrate the effectiveness and practical utility of the proposed algorithms in representative scenarios, confirming their potential for real-world applications in large-scale dynamic systems.

针对多智能体系统中交互与扩散核未知的问题，提出基于轨迹数据的稀疏回归方法，结合随机批采样与平均场近似进行有效识别。

This work addresses the challenge of approximating the mean-field dynamics of interacting particle systems exhibiting collective behavior using Transformer models, under the constraint of permutation equivariance arising from particle indistinguishability. Method: We propose the first theoretical Transformer framework that maps the vector field of a finite-particle system to its infinite-dimensional mean-field limit. Our approach integrates Transformer architecture with mean-field limit theory and the Cucker–Smale flocking model. Contribution/Results: We rigorously derive an explicit upper bound on the approximation error—expressed in terms of model and system parameters—and prove that the framework is inherently permutation-equivariant and consistent with the underlying mean-field PDE. Empirical validation is conducted on flocking dynamics and mean-field neural network training tasks. Numerical results closely match the theoretical error bounds, confirming both practical efficacy and theoretical soundness.

To address the inefficiency and instability of conventional forward-backward fixed-point iteration (FPI) in mean-field game (MFG) learning for large-scale multi-agent systems—caused by oscillatory behavior—this paper proposes a unified optimization framework that jointly treats policies and population distributions as co-optimizable control variables, enabling asynchronous joint updates. We introduce the first gradient-based MFG learning algorithm for continuous state-action spaces: population-aware linear function approximation (PA-LFA). Theoretically, we prove finite-time convergence to exact equilibria for contractive linear MFGs, asymptotic convergence to an equilibrium neighborhood under milder conditions, and derive approximation error bounds for nonlinear MFGs. Extensive evaluation across six benchmark tasks demonstrates the method’s effectiveness and robustness.

This paper addresses the stochastic game modeling of macro-market makers under price competition, extending the classical macro-market-making framework to a multi-agent stochastic game where each market maker benchmarks its optimal quote against rivals’ best responses. Methodologically, it introduces a novel multidimensional decoupling field framework, integrating forward-backward stochastic differential equations (FBSDEs), nonsmooth analysis, and stochastic game theory to establish equilibrium ordering properties and intrinsic dimension reduction in the linear case; for the nonlinear case, it develops a multidimensional characteristic equation theory and rigorously proves existence, uniqueness, and well-posedness of solutions. Key contributions include: (i) the first scalable mathematical framework for macro-market-making games; (ii) identification and characterization of structural equilibrium properties—particularly quote ordering and dimension-reduction mechanisms; and (iii) several new well-posedness results, providing a rigorous theoretical foundation for high-dimensional market microstructure modeling.

Finite-state mean-field games (MFGs) arise as limits of large interacting particle systems and are governed by an MFG system, a coupled forward-backward differential equation consisting of a forward Kolmogorov-Fokker-Planck (KFP) equation describing the population distribution and a backward Hamilton-Jacobi-Bellman (HJB) equation defining the value function. Solving MFG systems efficiently is challenging, with the structure of each system depending on an initial distribution of players and the terminal cost of the game. We propose an operator learning framework that solves parametric families of MFGs, enabling generalization without retraining for new initial distributions and terminal costs. We provide theoretical guarantees on the approximation error, parametric complexity, and generalization performance of our method, based on a novel regularity result for an appropriately defined flow map corresponding to an MFG system. We demonstrate empirically that our framework achieves accurate approximation for two representative instances of MFGs: a cybersecurity example and a high-dimensional quadratic model commonly used as a benchmark for numerical methods for MFGs.

This work addresses the failure of traditional mean-field approximations in modeling social behavior dynamics under high-noise networked settings by proposing a method based on the weak-form sparse identification of nonlinear dynamics (WSINDy). The approach directly learns continuous-time ordinary differential equation models from noisy trajectory data collected under multiple initial conditions, capturing the evolution of online-offline coupled social behaviors without relying on mean-field assumptions. By leveraging the weak formulation, the method recovers high-fidelity, interpretable dynamical models directly from stochastic process data. Experimental results demonstrate that incorporating only a few additional initial conditions substantially improves modeling accuracy in high-noise environments, significantly outperforming conventional mean-field approaches.

This work addresses the theoretical and practical limitations of traditional clustering methods in handling dynamic, nonparametric data evolution by proposing a mean-field game–based evolutionary clustering control framework. The approach models clustering as a population dynamics system governed by coupled Hamilton-Jacobi-Bellman and Fokker-Planck equations, where cluster structures evolve continuously under a variational cost functional regularized by time-averaged log-likelihood. By introducing mean-field games to clustering for the first time, the method enables nonparametric cluster evolution without assuming predefined statistical shapes, recovers the trajectory of the EM algorithm under Gaussian mixture models, and enforces mass conservation. Numerical experiments demonstrate the stability of the proposed framework and establish a novel pathway for tackling nonparametric dynamic clustering tasks that are challenging for conventional approaches.

This study addresses the problem of algorithmic sycophancy induced by large language models, which can trap populations in spirals of erroneous collective beliefs. To counteract this cognitive bias, the authors construct a networked stochastic dynamical system incorporating a minority of topologically central “teacher” nodes designed to correct group-level misperceptions. By employing degree-weighted mean-field approximation, they reduce the high-dimensional Langevin equations to a macroscopic drift equation, thereby offering the first analytical framework that integrates statistical physics with social network theory to elucidate AI-induced sycophancy. Key contributions include an analytical solution for the critical tipping time based on saddle-node bifurcation, a proof that centralized rapid intervention outperforms distributed slow strategies, and demonstration of universal data collapse and theoretical bounds across diverse network topologies. Under strict budget constraints, the work further derives an optimal intervention policy.

This work addresses the existence of equilibria in non-Markovian mean field games with unbounded control variables. By employing a weak formulation approach, it overcomes the conventional restrictions on boundedness of model parameters and time horizons, allowing the running cost to exhibit quadratic growth in the control. The authors establish a novel analytical framework for the existence and stability of quadratic-growth generalized McKean–Vlasov backward stochastic differential equations (BSDEs). Within this framework, they prove—under significantly more general conditions—the global existence of equilibrium solutions for such mean field games, thereby substantially extending the applicability of existing theoretical results.