Work

On the regularity threshold of uniformly positive quadratic stochastic operators

Year: 2026
Type: preprint
Abstract: A quadratic stochastic operator on the simplex of probability vectors in $\mathbb{R}^m$ is regular if all of its trajectories converge to one point. Ganikhodzhaev and Sarymsakov showed that the operator is regular when all of its heredity coefficients exceed $1/(2m)$, and Ganikhodzhaev and Rozikov asked for the least number $\lambda_m$ with this property. For $m=2$ we prove $\lambda_2=(3-\sqrt7)/4$. For every $m\ge3$ we prove that $\lambda_m$ lies between $(1-1/\sqrt{m+1})/(2m)$ and $2/(3m+2+\sqrt{m^2+12m+4})$, which is less than $1/(2m)$; in particular $2m\lambda_m\to1$. The upper bound comes from a contraction of the operator in the Fisher–Rao metric, which rests on two inequalities: the derivative of a quadratic stochastic operator at most doubles the Fisher information of a tangent vector, and the uniform-mixture map $q\mapsto\mu\mathbf 1+(1-m\mu)q$ multiplies it by at most $(1-m\mu)^2/[1-(m-2)\mu]$, where $\mu$ is the least heredity coefficient. The lower bound comes from an explicit family of operators with strictly positive coefficients and a two-cycle.
Language: English
Open Access status: green
OpenAlex ID: W7226003466
Cites:
Cited by:
Related to: 10