Open-Loop Nash Limits for Caputo Differential Games: Preceding-Node L1 Discretization

In the world of complex systems, from the spread of a virus to the fluctuation of global markets, the future is rarely determined solely by the present moment. Often, what happens next depends heavily on what happened before. This is known as memory. In many physical and economic models, the current state of a system is not just a snapshot but a cumulative result of its entire history. When multiple decision-makers, or players, interact within such a system, trying to optimize their own outcomes while reacting to the actions of others, the situation becomes a game. These are called differential games. The challenge for scientists and mathematicians is that these games are incredibly difficult to solve directly because the "memory" of the system makes the equations continuous and infinitely complex. To make progress, researchers often break time into small, manageable steps, turning the smooth, flowing game into a series of discrete snapshots. They solve the game for these snapshots and hope that the solution they find is a good enough stand-in for the real, continuous world.

The critical question, however, has always been whether this shortcut actually works. Does the solution found in the stepped, discrete version of the game truly converge to the genuine equilibrium of the original, continuous game? For standard systems without memory, the answer is often yes. But for systems with memory, where a single player's change in strategy alters the entire historical record that influences the future, simply getting the path of the system to look similar is not enough. The strategic logic must also hold up. If a player tries to deviate from the plan in the discrete version, that deviation must remain valid and comparable in the continuous version. Until now, it was unclear if the complex, non-local nature of memory would break this link, causing the discrete solutions to drift away from the true strategic balance as the steps became smaller.

A team of researchers at the University of Peloponnese has now provided a definitive answer to this question for a specific and challenging class of these memory-dependent games. They focused on games governed by a type of memory known as Caputo memory, where the influence of past events fades over time but never completely disappears. The researchers developed a rigorous mathematical proof showing that, under reasonable conditions, the solutions found in these discrete, step-by-step games do indeed converge to the true open-loop Nash equilibrium of the continuous game. An open-loop Nash equilibrium is a state where no player can improve their outcome by unilaterally changing their strategy, assuming the others keep theirs fixed. The team proved that as the time steps in the simulation become infinitely small, the strategies and outcomes of the discrete players settle into a stable pattern that perfectly matches the continuous game's equilibrium.

To reach this conclusion, the authors had to overcome a significant hurdle: the "hereditary" nature of the memory. In these games, a player's current move changes the entire history of the system, which in turn changes the future for everyone. This creates a delicate situation where a small error in the discrete approximation could theoretically distort the strategic comparisons needed to define an equilibrium. The researchers demonstrated that by using a specific method called preceding-node L1 discretization, they could preserve the necessary strategic structure. This method ensures that when calculating the state at a specific moment, the model only uses information from the immediately preceding moment, maintaining a clear cause-and-effect chain. They proved that even with this limitation, the discrete solutions retain the ability to recover any possible deviation a player might make in the continuous world. This recovery is essential; without it, the discrete game would be solving a different problem than the continuous one.

The study goes beyond abstract theory by applying it to a concrete, fully specified scenario involving three players interacting in a nonlinear environment. The researchers constructed a detailed model of this three-player game, complete with specific numerical values for how the players interact and what they aim to achieve. They then used computer-assisted methods to verify that the mathematical conditions required for their proof held true for this specific case. They checked the curvature of the players' objective functions, ensuring that the game was well-behaved enough to have a unique best response for each player. Their calculations confirmed that for any sufficiently fine mesh of time steps, a pure global Nash equilibrium exists in the discrete version of the game. This means that for this specific three-player scenario, the discrete approximation does not just offer a vague guess; it guarantees a stable solution that the continuous game also possesses.

The findings are significant because they bridge the gap between the theoretical world of continuous differential games and the practical world of computer simulations. Before this work, it was possible that the strategic solutions found in computer models of memory-dependent systems were merely artifacts of the numerical method, drifting away from reality as the grid was refined. The authors have shown that this is not the case for this class of games. They proved that the discrete equilibria are not just approximations of the state trajectories but are genuine approximations of the strategic equilibrium itself. This gives researchers and engineers confidence that when they use these numerical methods to design strategies for complex systems with memory—such as resource management, economic policy, or biological control—they are finding solutions that are mathematically sound and converge to the true optimal behavior.

The researchers also established that the convergence happens in a specific, robust way. The strategies, which are functions of time, converge in a manner that preserves their essential properties, while the resulting states of the system converge uniformly, meaning the entire path of the system becomes indistinguishable from the continuous ideal as the time steps shrink. This dual convergence ensures that both the decisions made by the players and the physical or economic outcomes of those decisions are correctly captured. The proof relies on the fact that the memory kernel, which dictates how past events influence the present, has specific mathematical properties that allow for this stability. By leveraging these properties, the team showed that the "hereditary" effects do not destroy the equilibrium structure but rather allow it to be faithfully reconstructed in the limit.

In the specific three-player example analyzed, the researchers found that the game possesses a strong convexity property, which guarantees that each player has a single, clear best response to the actions of the others. This property was verified using a combination of analytical bounds and computer-assisted certification, ensuring that the results were not just theoretical possibilities but concrete mathematical facts for the given parameters. The study confirmed that for a mesh size smaller than a certain threshold, the discrete game is guaranteed to have a solution, and that solution will converge to the continuous equilibrium. This provides a solid foundation for using numerical methods in the study of fractional-order differential games, a field that has been growing in importance due to its ability to model complex, memory-rich phenomena in physics, biology, and economics.

The work also clarifies the role of the specific discretization method used. The "preceding-node" approach, where the current step depends only on the immediately previous state and control, is not just a computational convenience but a crucial feature that allows the proof to work. This causality ensures that the discrete model respects the flow of time and information, preventing circular dependencies that could invalidate the equilibrium conditions. The researchers showed that this specific structure allows them to recover any unilateral deviation in the continuous game by constructing a corresponding deviation in the discrete game. This recovery process is the key to proving that the strategic comparisons hold up in the limit, ensuring that the Nash condition is preserved.

Ultimately, this research provides a rigorous validation for a common practice in scientific computing: solving complex, continuous problems by breaking them into discrete steps. For games involving memory, where the stakes of getting the strategy wrong are high because of the long-term ripple effects, this validation is essential. The authors have demonstrated that the path from the discrete to the continuous is not a leap of faith but a mathematically guaranteed journey. Their results confirm that the strategic insights gained from these numerical simulations are reliable and that the equilibria found in the computer are indeed the equilibria of the real-world systems they are meant to represent. This gives a new level of confidence to the application of differential game theory in fields where history matters and where the future is inextricably linked to the past.