Rokhlin’s multiple-mixing problem

The explanation

Flip two fair coins and write 1 for heads, 0 for tails. Call the results X and Y. Now set a third digit Z. It is 1 when X and Y differ, and 0 when they match. Four outcomes are equally likely: 000, 011, 101 and 110. Each digit is 1 half the time. Each pair of digits is 1 together a quarter of the time. That is exactly ½ × ½, so any two digits look independent. Yet all three digits are never 1 together. Full independence would predict 1 time in 8.

That gap between pairs and triples sits at the heart of a question from 1949. The question is about a rule T that moves the points of a space, step by step. Picture stirring a drink. The rule is probability-preserving: it never shrinks or swells the share of space a region takes up. It is invertible: every step can be undone. The rule is mixing when long waits erase memory. Take two regions, A and B. Count the points that start in A and sit in B after n steps. As n grows, their share tends to the share of A times the share of B. If A and B each fill half the space, that share tends to ¼.

Mixing of order 3 asks the same for three regions visited at three times, with both gaps growing. Then the share should tend to the product of three shares, such as ⅛ for three halves. Mixing of order k asks it for k regions. V. A. Rokhlin introduced these higher orders in 1949. Thierry de la Rue presents his 1949 question as this: does ordinary mixing alone imply the next order? The coin example shows why the answer is not automatic. Pairs can be independent while a triple stays linked.

With two independent directions of time, the answer is no. In 1978, F. Ledrappier gave a mixing example on a grid that is not mixing of order 3. In 2006, de la Rue explained why a direct one-dimensional copy of it fails. Positive answers for one rule needed extra structure. S. A. Kalikow showed in 1984 that order 2 gives order 3 for one special class. B. Host in 1991 and V. V. Ryzhikov in 1993 reached all orders for other special classes. In 2024, A. Kanigowski and D. Ravotti did so for certain flows, which are motions in continuous time. A general mixing rule has none of that structureSource 1.

This paper proves the general answer is yes. Every invertible, probability-preserving, mixing rule is mixing of every finite orderSource 2. It assumes no special form for the ruleSource 3. It also imposes no technical regularity condition on the spaceSource 4.

The proof argues by contradictionSource 5. Suppose some order fails, and take the smallest one that does. Then every smaller group of times behaves independently, but the full group does not. That is the coin pattern again, and pairwise independence alone does not force full independence. The paper first trims the failure down to a very predictable process. Such a process has zero entropy: long stretches follow relatively few likely patternsSource 6,Source 7. After that step, the new difficulty is to control every leftover interaction at once, not just one chosen snapshotSource 8. It then shows any hidden link must ride on special wave-like functions. These come in finite blocks that time shuffles as wholesSource 9. Averaging over time makes those links cancel. That contradicts a positive quantity that time preservesSource 10.

So no mixing rule can hide a coin-style link between three far-apart moments. With three half-size regions and all gaps long, the triple share must approach ⅛. The same holds for any number of regions. The statement is about limits only. It names no rate, so it does not say how long the gaps must be. It concerns one rule repeated in time, which is exactly where Ledrappier's grid example does not applySource 11. The paper also extends the result to rules that cannot be undone, and to certain flowsSource 12.

How sure is this? A computer proof checker called Lean has checked a formal version of the main theorem. It covers every order from 3 up and every choice of measurable regions. The Lean scope note links it to a fixed reference statement for a checking tool called Comparator. The extensions to non-invertible rules and to flows are not in that scope note. No specialists have published a review yet.

What it doesn't show. This does not cover systems with two independent directions of time. Ledrappier's 1978 example shows those can fail.