Multiple colour interacting urns on complete graphs
Benito Pires, Rafael A. Rosales
TL;DR
The paper studies a multi-colour interacting Pólya urn model on a complete graph with a general reinforcement rule defined by a smooth function $\varphi$ and parameter $\alpha$. Using stochastic approximation and a Hopfield-inspired strict Lyapunov function, it proves convergence of the colour-proportion vector $X(n)$ to fixed points of the induced map $\pi$, with convergence to a single fixed point when Fix$(\pi)$ is finite. For small $|\alpha|$ the unique equilibrium is the uniform distribution $x^*=(1/d,\dots,1/d)$, while large $|\alpha|$ can yield multiple stable fixed points, characterized by the Jacobian $J\pi(x)$. An explicit triangle example demonstrates multiple stable fixed points and validates the theoretical framework through eigenvalue analysis. Overall, the work provides a rigorous route to understanding long-run equilibria in non-gradient flows arising from interactions on graphs and reinforces the role of Hopfield-like energy in analysing complex reinforcement dynamics.
Abstract
We present a multiple colour generalisation of the model of graph interacting urns studied by Benaim et. al., Random Struct. Alg., 46: 614-634, 2015. We show that for complete graphs and for a broad class of reinforcement functions governing the addition of balls in the urns, the process of colour proportions at each urn converges almost surely to the fixed points of the reinforcement function.
