Intrinsic noise in game dynamical learning.

Area of Science:

• Evolutionary Game Theory
• Mathematical Biology
• Statistical Physics

Background:

• Demographic noise, arising from discrete dynamics in finite populations, significantly influences evolutionary, population, and epidemiological models.
• Standard replicator dynamics often assume deterministic behavior, neglecting the impact of inherent stochasticity.

Purpose of the Study:

• To investigate the emergence of noise-sustained trajectories in game dynamical learning.
• To analyze the effects of finite opponent move sampling on stochastic dynamics in game theory.
• To understand how this novel source of noise impacts evolutionary game dynamics and attractors.

Main Methods:

• Developed a game dynamical learning model where agents sample a finite number of opponent moves.
• Utilized methods from statistical physics to analytically compute the characteristics of these fluctuations.
• Compared finite sampling dynamics with the deterministic modified replicator equations derived from infinite batch limits.

Main Results:

• Demonstrated that finite sampling of opponent moves in game dynamical learning introduces stochasticity, leading to noise-sustained trajectories.
• Showcased that these fluctuations can be analytically computed using statistical physics.
• Observed that the noise significantly affects attractors, potentially inducing sustained cycling or eliminating periodic orbits found in standard replicator dynamics.

Conclusions:

• Finite sampling in game dynamical learning generates a distinct form of demographic noise with significant consequences.
• This noise-sustained dynamics can alter the long-term behavior of evolutionary games, diverging from deterministic predictions.
• The findings highlight the importance of considering finite sampling effects in game theory models to accurately capture population dynamics.