Approximately Solving Continuous-Time Mean Field Games with Finite State Spaces

A team of researchers including Yannick Eich, Christian Fabian, Kai Cui, and Heinz Koeppl has published a significant paper titled 'Approximately Solving Continuous-Time Mean Field Games with Finite State Spaces' on arXiv. The work addresses a critical gap in game theory by developing practical computational methods for continuous-time mean field games (MFGs) with discrete state spaces. These games model large-scale multi-agent systems where individual agents' behaviors are governed by continuous-time Markov chains, with applications ranging from population dynamics to queueing networks and economic systems. While MFGs provide a powerful theoretical framework, efficient algorithms for finding Nash equilibria in continuous-time settings have remained underdeveloped, limiting their practical implementation.

The researchers' key innovation involves approximating classical Nash equilibria through regularization methods, making the solution algorithms computationally tractable. They specifically define regularized equilibria for continuous-time MFGs and extend two established algorithms—fixed-point iteration and fictitious play—to work with these equilibria. This approach bridges the gap between discrete-time methods that are computationally feasible and continuous-time models that are more realistic for many applications. The team validated their methods through illustrative numerical examples, demonstrating both effectiveness and practicality. This work represents an important step toward making complex multi-agent system modeling more accessible for real-world applications where continuous-time dynamics are essential.