Completely Integrable Gradient System on the bivariate beta statistical manifold

TL;DR

This paper explores the geometric and dynamical properties of a gradient system on the bivariate beta statistical manifold, revealing its Hamiltonian structure, integrability, and explicit geometric characteristics.

Contribution

It establishes the Hamiltonian nature and complete integrability of the gradient system on the bivariate beta manifold, linking information geometry with dynamical systems.

Findings

The vector field is Hamiltonian and admits a Lax pair representation.

The potential function defines a Riemannian metric equivalent to Fisher information.

The gradient flow is linearizable in dual affine coordinates.

Abstract

This paper investigates the geometry of a completely integrable gradient system defined on the three parameter bivariate beta statistical manifold of the first kind. We prove that the associated vector field is Hamiltonian and admits a Lax pair representation implying complete integrability. We show that the potential function derived from exponential family structure defines a Riemannian metric equivalent to the Fisher information metric. By applying Stirling's approximation to the gamma functions involved in the potential, we obtain an explicit expression that facilitates the study of the pseudo-riemannian geometry of the manifold. Furthermore, we demonstrate that the gradient flow is linearizable in dual affine coordinates, and we identify the Hamiltonian function whose gradient defines the flow. These results highlight the deep interplay between information geometry, dynamical systems, and asymptotic analysis.