Discrete-Time Mean Field Type Games: Probabilistic Setup

Grégoire Lambrecht, Mathieu LaurierearXiv 2026 · OpenReview

TL;DR

This paper introduces a general probabilistic framework for discrete-time Mean Field Type Games (MFTGs) that incorporates both global and team-specific common noise. The authors establish the existence of Nash equilibria in these complex multi-team games by reformulating them as Mean Field Markov Games (MFMGs), a technique that simplifies analysis and leverages existing game theory results.

What Problem Does This Paper Address?

Many real-world systems involve large populations of interacting agents, such as financial markets, traffic networks, or social platforms. Game theory provides tools to model these systems, but traditional approaches become computationally intractable as the number of agents grows. Mean Field Games (MFGs) address this by assuming a large number of agents, simplifying the problem to the interaction between a representative agent and the population's distribution.

However, many real-world scenarios involve not just large populations but also distinct groups or "teams" of agents who cooperate internally but compete with other teams. These teams might have different characteristics and face different external factors. Existing MFG frameworks often assume homogeneity across all agents. This paper tackles the challenge of modeling and analyzing games with multiple cooperating teams, known as Mean Field Type Games (MFTGs), especially when these teams are influenced by various sources of randomness, including shared global noise and team-specific noise. A key difficulty is proving the existence of stable outcomes (Nash equilibria) in such complex, multi-layered systems.

Methods

The core contribution of this paper is the development of a general probabilistic framework for discrete-time, infinite-horizon discounted MFTGs. The authors introduce a novel approach by:

  • Formalizing MFTGs with multilevel noise: They define MFTGs that incorporate both global common noise (affecting all teams) and team-specific common noises (affecting only agents within a particular team). Agents can also use randomized actions at both individual and team levels.
  • Connecting MFTGs to Mean Field Markov Games (MFMGs): A crucial step is reformulating the MFTG problem into an equivalent MFMG. This is achieved by considering a "lifted" state space that represents the distribution of agents' states across all teams. This reformulation allows them to leverage existing, powerful results from Markov game theory.
  • Proving Existence of Nash Equilibria: By using recent theoretical results for infinite-horizon discounted games with infinite compact state-action spaces, and by applying tools from Markov game theory to the reformulated MFMG, the authors establish the existence of a Nash equilibrium for the MFTG. This is a significant theoretical result, as existence proofs for such complex games are often challenging.
  • Introducing a "Mean Field Drift of Intentions" Example: To illustrate their framework and theoretical results, they present a specific example where team dynamics are strongly randomized, demonstrating the practical applicability of their approach.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Multilevel Noise is Crucial: The paper highlights that incorporating both global and team-specific common noise is essential for accurately modeling many real-world multi-team systems. These noise structures can significantly influence agent behavior and team coordination.
  • The Power of Reformulation: The key technical insight is the successful reformulation of the MFTG into an MFMG. This "lifting" of the state space to represent population distributions, combined with the use of common randomization, transforms a complex MFTG problem into a more tractable MFMG problem, allowing the application of established game theory tools.
  • Existence of Equilibria is Achievable: Despite the complexity introduced by multiple teams, common noises, and randomized actions, the paper demonstrates that Nash equilibria can exist under reasonable assumptions. This provides a theoretical foundation for analyzing and predicting outcomes in these systems.
  • Randomization as a Coordination Tool: The "Mean Field Drift of Intentions" example suggests that strong randomization, particularly through common noise, can be a mechanism for teams to coordinate their actions and achieve desired outcomes, even in the face of external perturbations.

Key Results

  • The paper provides a general probabilistic framework for discrete-time, infinite-horizon discounted MFTGs with multilevel common noise.
  • They establish a formal connection between closed-loop policies in MFTGs and Markov policies in MFMGs.
  • The authors prove the existence of a Nash equilibrium for MFTGs when state spaces are at most countable and action spaces are general Polish spaces.
  • An example, "Mean Field Drift of Intentions," is presented to satisfy the assumptions and demonstrate the existence of a Nash equilibrium.

Limitations and Future Work

  • Uniqueness of Nash Equilibrium: The paper focuses on the existence of Nash equilibria but does not address their uniqueness. Investigating uniqueness is suggested as a future research direction.
  • Algorithmic Aspects: While the paper establishes existence, it does not delve into algorithms for finding optimal policies. Developing such algorithms is another avenue for future work.
  • Real-World Applications: The authors note that applications with real-world data have not yet been explored, suggesting this as a promising area for future research, potentially leading to new engineering perspectives.
  • Assumptions: The existence results rely on specific assumptions, such as state spaces being at most countable and action spaces being Polish. Further research might explore relaxing these conditions.

Summary updated

Abstract

We introduce a general probabilistic framework for discrete-time, infinite-horizon discounted Mean Field Type Games (MFTGs) with both global common noise and team-specific common noises. In our model, agents are allowed to use randomized actions, both at the individual level and at the team level. We formalize the concept of Mean Field Markov Games (MFMGs) and establish a connection between closed-loop policies in MFTGs and Markov policies in MFMGs through different layers of randomization. By leveraging recent results on infinite-horizon discounted games with infinite compact state-action spaces, we prove the existence of an optimal closed-loop policy for the original MFTG when the state spaces are at most countable and the action spaces are general Polish spaces. We also present an example satisfying our assumptions, called Mean Field Drift of Intentions, where the dynamics are strongly randomized, and we establish the existence of a Nash equilibrium using our theoretical results.
BibTeX
@misc{lambrecht2026discrete,
  author = {Grégoire Lambrecht and Mathieu Lauriere},
  title = {{Discrete-Time Mean Field Type Games: Probabilistic Setup}},
  year = {2026},
  url = {https://openreview.net/forum?id=KlkCU2228l},
}

Loading related content...

Figure