On multiplicative Chung--Diaconis--Graham process

TL;DR

This paper analyzes the mixing time of a specific lazy Markov chain on finite fields, revealing it grows subexponentially with the size of the field, and applies findings to a Sidon-type set problem.

Contribution

It introduces a new multiplicative Markov chain model on finite fields and determines its subexponential mixing time, connecting probabilistic and additive combinatorics.

Findings

Mixing time is (\u221a{\log p / \log \log p})

Provides bounds on convergence rate of the chain

Applies results to a Sidon-type set problem

Abstract

We study the lazy Markov chain on $\mathbf{F}_p$ defined as $X_{n+1}=X_n$ with probability $1/2$ and $X_{n+1}=f(X_n) \cdot \varepsilon_{n+1}$, where $\varepsilon_n$ are random variables distributed uniformly on $\{ \gamma^{}, \gamma^{-1}\}$, $\gamma$ is a primitive root and $f(x) = \frac{x}{x-1}$ or $f(x)=\mathrm{ind} (x)$. Then we show that the mixing time of $X_n$ is $\exp(O(\log p / \log \log p))$. Also, we obtain an application to an additive--combinatorial question concerning a certain Sidon--type family of sets.