[Paper Review] A Coupling Argument for the Random Transposition Walk

Olena Bormashenko|arXiv (Cornell University)|Sep 18, 2011

Stochastic processes and statistical mechanics15 references11 citations

TL;DR

This paper presents a coupling argument proving that the random transposition walk on the symmetric group $S_n$ mixes in $O(n\log n)$ time, resolving a long-standing open problem. By projecting the walk to conjugacy classes and applying Bubley-Dyer path coupling with techniques from Schramm's work on random transpositions, the authors establish a contraction rate of $1 - \frac{1}{n}$ on average, enabling the $O(n\log n)$ bound via coupling.

ABSTRACT

This paper explores the mixing time of the random transposition walk on the symmetric group. While it has long been known that this walk mixes in order n*log(n) time, this result has not previously been attained using coupling. A coupling argument showing the correct order mixing time is presented. This is accomplished by first projecting to conjugacy classes, and then using the Bubley-Dyer path coupling construction. In order to obtain appropriate bounds on the time it takes the path coupling to meet, ideas from Schramm's paper "Compositions of Random Transpositions" are used.

Motivation & Objective

• To resolve the open problem of establishing an $O(n\log n)$ mixing time for the random transposition walk using a coupling argument.
• To overcome the limitation of Markovian couplings, which cannot achieve better than $O(n^2)$ bounds, by employing a non-Markovian coupling strategy.
• To project the random transposition walk onto conjugacy classes, transforming it into a split-merge random walk, which simplifies analysis while preserving mixing time properties.
• To apply the Bubley-Dyer path coupling framework to the split-merge walk by establishing a favorable contraction coefficient of $1 - \frac{1}{n}$ on average.
• To use tools from Schramm's analysis of random transposition compositions to bound the time to meeting in the coupling process.

Proposed method

• Project the random transposition walk on $S_n$ to conjugacy classes, resulting in a split-merge random walk that is Markovian and preserves the mixing time behavior.
• Apply the Bubley-Dyer path coupling technique to the split-merge walk, using a distance function based on cycle structure to measure contraction between coupled chains.
• Establish that the expected contraction per step is $1 - \frac{1}{n}$ by analyzing cycle merging and splitting probabilities under the coupling scheme.
• Use Schramm’s results on the emergence of large cycles after time $\frac{n}{2}$ to bound the time until the coupled chains meet with high probability.
• Construct a non-Markovian coupling by carefully aligning transitions in the coupled chains $X_t$ and $Y_t$ to maximize meeting probability, particularly when large cycles form.
• Leverage bounds on the probability that the sum of cycle sizes above a threshold increases, ensuring sufficient contraction to achieve $O(n\log n)$ mixing time.

Experimental results

Research questions

• RQ1 Can a coupling argument achieve the $O(n\log n)$ mixing time bound for the random transposition walk on $S_n$, despite the failure of Markovian couplings?
• RQ2 What is the appropriate contraction rate for the split-merge walk on conjugacy classes that enables path coupling to yield $O(n\log n)$ mixing time?
• RQ3 How can Schramm’s analysis of random transposition compositions be adapted to bound the meeting time of coupled chains in the path coupling framework?
• RQ4 Why does the standard path coupling fail when applied directly to the split-merge walk, and how can this be overcome?
• RQ5 What role do large cycles play in accelerating the coupling time, and how can their emergence be quantitatively leveraged?

Key findings

• The paper establishes that the random transposition walk on $S_n$ mixes in $O(n\log n)$ time via a coupling argument, resolving a long-standing open problem.
• The coupling argument relies on projecting the walk to conjugacy classes and applying path coupling to the resulting split-merge process.
• The average contraction coefficient of the split-merge walk is shown to be $1 - \frac{1}{n}$, enabling the $O(n\log n)$ bound via path coupling.
• The proof uses Schramm’s results on cycle composition to show that large cycles emerge after time $\frac{n}{2}$, which accelerates the coupling process.
• The coupling is non-Markovian, as required by theoretical limitations, and achieves a meeting time bound of $O(n\log n)$ for all initial configurations.
• The constant in the $O(n\log n)$ bound is large, indicating that the coupling does not prove the cut-off phenomenon, though the order is correct.