Does mixing two observations force mixing any finite number

← All digests

Family 145 · Dynamical systems and ergodic theory
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-21

1. The question

Suppose a stationary system gradually forgets the relation between two observations made far apart in time. Must it also forget every relation among three, four, or any fixed finite number of well-separated observations?

For arbitrary random variables, the answer is no. Let $X$ and $Y$ be independent fair bits and set $Z=X+Y\pmod2$. Every pair is independent, but knowing two bits determines the third. Pairwise independence can hide a collective constraint.

Rokhlin’s multiple-mixing problem asks whether such a persistent higher-order constraint can survive when all observations come from repeated applications of one measure-preserving transformation. Family 145 claims that it cannot: ordinary mixing implies mixing of every finite order.

The phrase “one transformation” is essential. A system indexed by two independent integer directions can be mixing for pairs while retaining a three-way constraint. The proposed theorem concerns a single time axis, not all group actions.

2. Background and essential definitions

Mixing means asymptotic independence of fixed events

Let $(\Omega,\mathcal F,\mu)$ be a probability space and let $T$ be an invertible measurable transformation with measurable inverse. It preserves probability if $\mu(T^{-1}A)=\mu(A)$ for every measurable set $A$.

An event $A$ is a condition imposed on the current state. The preimage $T^{-n}B$ means that the state satisfies event $B$ after $n$ steps. Ordinary mixing is the requirement

$$\mu(A\cap T^{-n}B)\longrightarrow\mu(A)\mu(B) \qquad(|n|\to\infty)$$

for every fixed pair $A,B$. The limit on the right is what independence would give. The events themselves stay fixed while their observation times separate.

This is stronger than ergodicity, which only rules out invariant events of intermediate probability. It is also stronger than weak mixing, where suitable correlations vanish after time averaging. The theorem starts with ordinary mixing, not merely either of these weaker properties.

Higher orders test genuinely joint behavior

For an integer $k\geq2$, mixing of order $k$ means

$$\mu\left(\bigcap_{i=1}^kT^{-t_i}A_i\right) \longrightarrow\prod_{i=1}^k\mu(A_i)$$

whenever $t_1<\cdots<t_k$ are integer times and

$$\min_{1\leq i<k}(t_{i+1}-t_i)\longrightarrow\infty.$$

Thus all successive gaps grow, but they need not grow at comparable speeds. The theorem is not restricted to equally spaced times such as $0,n,2n$.

We count the order by the number of events: ordinary mixing is order two. Some sources call this multiplicity one and shift the terminology by one, so conventions must be checked when comparing papers.

A joining records a possible limiting dependence

A joining is a joint probability law for several copies of a system that retains the required marginal laws and is invariant under simultaneous time evolution. If higher-order mixing fails along a sequence of separated times, a limiting joint law can retain that failure.

Its proper subcollections may nevertheless be independent. The three-bit example illustrates why that information alone does not force the full law to be a product. The challenge is to use the additional structure imposed by a common time evolution.

3. Development history and previous partial results

Rokhlin introduced higher-order mixing in his work on compact commutative groups and established it for ergodic surjective continuous endomorphisms with Haar probability. The unrestricted implication from ordinary mixing became a major problem in ergodic theory.

The two-dimensional contrast emerged sharply in Ledrappier’s 1978 example. Its configurations are binary arrays indexed by $\mathbb Z^2$, subject to a local three-dot relation. Translating the array in two independent directions gives a mixing action with zero entropy that is not threefold mixing. This demonstrates a real obstruction, while leaving open whether a single generator can sustain it.

Several strong positive results required extra structure. Kalikow proved that twofold mixing implies threefold mixing for rank-one transformations, whose behavior can be approximated by one sequence of increasingly large towers. Host established all orders under a singular-spectrum hypothesis. Ryzhikov obtained all-orders results for mixing finite-rank automorphisms. More recently, Kanigowski and Ravotti treated mixing flows with a suitable local shearing property.

There were also broad results guaranteeing many favorable configurations of observation times. Bergelson and Zelada showed that higher-order good configurations form a strong Ramsey-large set for strongly mixing actions of countable abelian groups. Such abundance is different from convergence along every sequence whose time gaps diverge.

Joining methods, developed from Furstenberg’s work, supplied a language for possible obstructions. Earlier work described how a failure could be sought in a zero-entropy part of a system. De la Rue analyzed the twofold/threefold problem and explained why a direct analogue of the three-dot construction does not work in one dimension.

The new manuscript claims to complete the general single-transformation argument. It supplies an all-orders reduction and then addresses the residual joint dependence, without assuming rank, singular spectrum, group geometry, or a shearing model.

4. What the new manuscripts claim

Every finite order for one invertible transformation

The main theorem asserts that every invertible mixing probability-preserving transformation is mixing of every finite order. More explicitly, for fixed measurable sets $A_1,\ldots,A_k$,

$$\begin{aligned} &\mu\left(A_1\cap T^{-n_1}A_2\cap\cdots \cap T^{-(n_1+\cdots+n_{k-1})}A_k\right)\\ &\hspace{2em}\longrightarrow\prod_{i=1}^k\mu(A_i) \qquad\text{as }\min_i n_i\to\infty. \end{aligned}$$

The gaps $n_i$ are positive integers. No standardness or countable-generation assumption is imposed on the original probability space. The manuscript passes to a finite-symbol factor when its argument needs a more manageable measurable setting.

The conclusion is a limit for every diverging-gap configuration. It is not just convergence of a time average, or convergence after excluding a thin collection of times. No quantitative rate is asserted.

Noninvertible maps and flows

A corollary extends the conclusion to mixing measure-preserving endomorphisms: measurable maps that need not be invertible. Observation times are then nonnegative integers, and $T^{-t}A$ denotes a preimage under a forward iterate, not an inverse map.

Another corollary treats probability-preserving flows $(S_t)_{t\in\mathbb R}$. In addition to mixing, the manuscript assumes strong continuity of their Koopman operators on $L^1$. Concretely, for every measurable $A$,

$$\|\mathbf1_A\circ S_s-\mathbf1_A\|_1\longrightarrow0 \qquad(s\to0),$$

where $\mathbf1_A$ is the indicator of $A$, and the norm is the integral of absolute value. Under this hypothesis, the same product limit holds for arbitrary real observation times with diverging successive gaps. Jointly measurable probability-preserving flows on standard probability spaces satisfy the stated continuity condition.

5. Why the advance matters

Provisional importance: 5/5. The claim resolves a central structural problem: whether the simplest nontrivial mixing condition already controls every fixed finite pattern of widely separated observations along one time axis.

If correct, it removes an entire hierarchy of additional hypotheses from statements about a general mixing transformation. One would no longer need to establish threefold, fourfold, and higher mixing separately once ordinary mixing is known.

It also clarifies why pairwise-independent counterexamples from elementary probability do not settle the dynamical question. Their joint laws need not be compatible with the repeated action of one transformation and all the limiting configurations that action generates.

The result would have immediate consequences for other proposed examples. For instance, family 144 invokes this theorem to upgrade ordinary mixing of its smooth spectral construction to mixing of all orders. That application is logically separate from constructing the spectral model itself.

6. The main idea

Reduce a hypothetical failure to a small witness

Assume some finite order fails and choose the least such order $k$. All lower orders then mix correctly. The manuscript reduces the obstruction to a stationary process taking values in a finite alphabet, with centered one-coordinate functions whose joint correlation stays away from zero along a sequence of separated layouts.

The process can be chosen to have zero entropy rate. If $H(X_1,\ldots,X_n)$ is the Shannon entropy of an observed block, this means

$$\frac1nH(X_1,\ldots,X_n)\longrightarrow0.$$

It does not mean that the process is constant. It means that the typical long words require sublinear information per observed coordinate. This restriction becomes useful against a proposed persistent correlation.

Retain more than one limiting joint law

A single joining may hide the obstruction rather than resolve it. The manuscript therefore builds an ordered family of array laws using repeated limits at sums of observation layouts. Proper collections of coordinate planes retain independence, including specified conditional versions after additional information is recorded.

The order of these limits matters. The construction does not assume that swapping array axes leaves the law unchanged. Nor do the array axes become new independent time directions: the original system still evolves under one transformation.

An extension procedure adds information until certain conditional expectations can no longer improve. Conditional expectation is the best mean-square prediction from the information being retained. At this stage, the paper claims precise local relationships between row and column predictions.

Convert dependence into algebraic labels

The conditional-expectation operators satisfy identities associated with a square of coordinates. The manuscript uses these identities to extract orthogonal one-dimensional spaces of functions whose values have modulus one. Along suitable rows or columns, products of matching functions are constant.

Multiplication of these functions, ignoring constant phase factors, supplies abelian group labels. A measurable construction organizes relevant labels into finite blocks of fixed size. Time moves whole blocks according to one group automorphism. The argument does not require a uniform description of each individual label’s motion inside a block.

This stage translates a probabilistic dependence into algebraic data on which the single time direction imposes strong constraints.

Use one-dimensional time to force cancellation

The manuscript separates label orbits into two behaviors along arithmetic progressions. In one case, labels have enough algebraic independence to produce exponential estimates; zero entropy limits how many likely symbolic words can support a persistent correlation. In the other case, polynomial relations in the orbit allow a degree-reduction argument, with ordinary mixing as the starting case.

Both mechanisms make averaged absolute correlations with moving finite blocks tend to zero. This contradicts a positive quantity that the supposed failure forces time evolution to preserve.

The averaging is internal to the contradiction. It does not weaken the theorem’s final conclusion to an averaged form of higher-order mixing. The proposed argument rules out every least failing finite order.

7. Scope, limitations, and open questions

The theorem concerns repeated use of one transformation. It does not contradict Ledrappier’s action with two independent time generators, and it does not assert the same implication for arbitrary actions of $\mathbb Z^2$ or other groups.

The number of tested events is fixed before taking the separation limit. Nothing here gives a bound uniform in an increasing number of events, a convergence rate, or control when the events themselves vary with time.

Ordinary mixing remains an assumption. The claim does not upgrade every ergodic or weakly mixing system to ordinary mixing. Likewise, the continuous-time extension retains the specified continuity hypothesis.

The result concerns joint probabilities of observations. It should not be confused with pointwise convergence of multiple ergodic averages along almost every orbit, the different topic of family 154.

8. Sources and further reading

For earlier restricted implications, see Kalikow’s rank-one theorem, Host’s singular-spectrum theorem, and Ryzhikov’s finite-rank results. Bergelson and Zelada’s Strongly mixing systems are almost strongly mixing of all orders makes the distinction between abundant good configurations and unrestricted convergence especially relevant.

De la Rue’s account of twofold and threefold mixing explains the joining viewpoint and the contrast with Ledrappier’s example. Ryzhikov’s 75-year survey of Rokhlin’s problem supplies additional historical context. The source manuscript develops its claimed general argument and the endomorphism and flow corollaries.

Source manuscripts

• M1. OpenAI, Rokhlin’s multiple-mixing problem for one transformation (September 23, 2026).

Sources consulted on 21 October 2026 at repository revision 1581d4c. Download the manuscript citations.