The mathematics of how things change over time often relies on the idea of a system moving "downhill" toward lower energy, like a ball rolling down a slope. This works beautifully when there is a single energy function governing the whole system. But many real-world situations involve multiple interacting agents or species, each following its own energy landscape, with no single combined energy to minimize. Think of competing populations, interacting fluids, or strategic agents each pursuing their own objectives. The resulting dynamics cannot be captured by classical tools, and this paper develops a new mathematical framework to handle them.
The central idea is to replace the single energy function with a "bifunction," which assigns a value to every pair of states in a space. This bifunction encodes how much one state "dissipates" or improves relative to another. The authors impose a mild condition called dissipativity, which roughly says the bifunction cannot be simultaneously large in both directions between two points. Building on this, they design a time-stepping procedure called the Variational Movement Scheme, where at each small time step the system finds a saddle point of a penalized version of the bifunction. When applied to multiple interacting species, this saddle point becomes a Nash equilibrium, the kind of balanced outcome familiar from game theory. This bridges ideas from optimal transport and gradient flows on one side, and game theory on the other.
The paper then establishes rigorous results showing that this scheme actually produces well-behaved solutions as the time steps shrink to zero. The authors prove existence, stability, and regularity of solutions under general conditions that do not require the underlying space to have any linear structure. A key technical innovation is a new notion of convexity defined along "barycentric interpolations," which generalizes a tool previously used in optimal transport theory. The overall framework is broad enough to cover settings like populations of particles each minimizing their own interaction energy while coupled to others, offering a principled mathematical foundation for multi-agent dynamics in complex geometric spaces.