Select a passage or formula to leave a suggestion here. Saved suggestions remain private until reviewed.

Multiple ergodic averages / Close Reading

Multiple ergodic averages: close reading

This topic begins with four preprints attributed to OpenAI and dated October 4, 2026. The labels A, B, C, and D follow their order in the official family 154 listing. The reading order follows the dependencies of the proofs. We retain the original theorem numbers and spell out the assumptions, deductions, and points that remain to be checked. The complete proof chains of the four main theorems and their companion manuscripts have not yet been independently verified.

The four main theorems

The common setting is a probability space $(X,\mathcal F,\mu)$ with an invertible measure-preserving transformation $T$ whose inverse $T^{-1}$ is also measurable. Mixing means that, for every pair of fixed measurable sets $E,F$ ,

\mu(E\cap T^{-n}F)\longrightarrow\mu(E)\mu(F) \quad (|n|\longrightarrow\infty).

All observables are fixed bounded complex-valued measurable functions. The underlying probability spaces need not be standard, and no rate of mixing is assumed. Each conclusion concerns all positive integer averaging lengths. The full-measure convergence set may depend on the fixed functions and, in C, on the chosen slopes.

A: Arbitrary finite order

Pointwise Multiple Ergodic Averages for Mixing Transformations, 25 pages, Theorem 1.1, p.1.

The paper claims that, for each fixed integer $d\ge2$ ,

\frac1N\sum_{n=1}^{N}\prod_{j=1}^{d}f_j(T^{jn}x) \longrightarrow\prod_{j=1}^{d}\int_X f_j\,\mathrm{d}\mu \quad\text{almost everywhere.}

Inserting the constant function $1$ at the missing positions formally yields the corresponding conclusion for any finite collection of distinct positive integer slopes. This consequence remains conditional on A's main theorem; it does not directly cover slopes of mixed signs at arbitrary order.

B: Fourfold averages

Pointwise convergence of fourfold ergodic averages for mixing transformations, 36 pages, Theorem 1.1, p.1.

The paper claims the conclusion above for $d=4$ . Its proof uses lengths selected from pointwise failure, joining distributions, three groups of auxiliary variables, tensor estimates, and stable conditional expectations. Appendix A of B separately treats the nonconsecutive triples of slopes needed in the argument. Its use of a triple-average result therefore involves more than the headline theorem for slopes $1,2,3$ .

C: Triple averages with distinct integer slopes

Triple ergodic averages with distinct integer slopes, 43 pages, Theorem 1.1, p.1.

The paper claims that, for every three pairwise distinct nonzero integers $\lambda_1,\lambda_2,\lambda_3$ ,

\frac1N\sum_{n=1}^{N}\prod_{j=1}^{3}f_j(T^{\lambda_jn}x) \longrightarrow\prod_{j=1}^{3}\int_X f_j\,\mathrm{d}\mu \quad\text{almost everywhere.}

The slopes may be negative, and the constants in the estimates may depend on these three fixed integers. The proof uses the entire four-point configuration $0,\lambda_1,\lambda_2,\lambda_3$ ; it is not obtained merely by replacing $T$ with a power in the basic triple-average theorem.

D: The basic triple average

Pointwise convergence of triple ergodic averages for mixing transformations, 61 pages, Theorem 1.1, p.2.

The paper claims C's conclusion for $(\lambda_1,\lambda_2,\lambda_3)=(1,2,3)$ . The main analytic task is to extend local estimates for fixed functions to functions that vary with depth, and then deduce a triple oscillation estimate.

Order of discussion

Unit 1: From oscillation estimates to all lengths

Source: C, Definition 3.1, Proposition 3.5, Lemmas 3.6–3.9, and the proof of Theorem 1.1 at the end of §3, pp.5, 8–13.

Assuming Proposition 3.5, this unit proves the implication from oscillation to convergence, the finite Hilbert-space inequality, the induction for multiple $L^2$ convergence under ordinary mixing, Abel summation for smooth weights, the correction of moments, and the recovery of all integer lengths. It also states the derivative conditions on smooth weights precisely, so that approximation does not change the original proposition.

C Theorem 2.2 and Lemmas 3.2–3.4 are not proved in this unit. What is completed is the ergodic-theoretic deduction conditional on the oscillation estimate, rather than an independent verification of the whole of C.

Unit 2: The local derivation of the oscillation estimate

Source: C Theorem 2.2, Lemmas 3.2–3.4, and the proof of Proposition 3.5.

This unit develops the four-direction cancellation kernel, reconstruction of the integer configuration, finite-orbit transference, and the depth-varying functions produced by linearizing the maximum. The adjoint partial sums and Abel summation in D treat a different depth-varying function problem. They should be distinguished from the Abel summation for smooth time weights in Unit 1.

Unit 3: Depth-varying functions and row estimates

Source: D §§4–7, pp.17–61. The exposition is organized into six stages according to their role in the proof:

Row estimates and actual projections: Unit 3 and continuations b–d, from abstract row estimates on a common index set to projection gain, finite restarts, and the earliest two precision increments.
Replacing the basic functions: continuations e–h, covering fixed predictor families, delayed replacement, high-precision functions, and three active patterns, reducing the problem to absolute sums with a minimum on each original cell.
Localization and control across depths: first read supplement s on D §6.1, then continuations i–m for main-projection closure, current groups, local-output variation, frozen mode rows, non-core energy, and representative counts.
At least two side positions: continuations n–p, covering the complete filter transfer, fixed Hilbert vectors, matched randomization, short-block comparison, and maximal averages on whole ancestor intervals.
At least three core positions: continuation q, stating the precise finite counting hypotheses, same-cell side-function windows, powers of complete counts, and summation over groups, amplitudes, and frequencies. The finite counting input, H Lemma 9.1 (October 5, 2026 version, pp.81–87), has a complete, independently reviewed proof in the extension “Finite counting on sparse scales.”
Completing the local estimate: continuation r, combining the costs, reassigning original summands, choosing the common $q$ for H and D and the initial precision, and taking the final finite-grid limit.

These parts give D's own conditional deduction from row estimates to the local conclusion. Thirteen local modules from H now have companion readings: scalar Fourier analysis, historical ball masks, fast-space propagation, two-coordinate detection, one-parameter rational fast splitting, same-slot detection, shared random coloring, fixed testing templates with the same tag, column-catalog termination under fixed inputs, fixed-triple scale estimates and active path budgets under supplied basic data, the original basic-column construction and path budget, actual search histories with conditional local far-band transfer, and short-depth-block comparison for fixed stacks. Each retains its stated inputs; the complete analytic and geometric proof in H has not yet been developed throughout. The finite sparse-counting deduction has been developed separately; its application still assumes the stated label separation, frequency gaps, positionwise counts, and height grouping.

D's local theorem concerns the four-point configuration $0,1,2,3$ . Unit 2 studies C and general four-point integer configurations. Their kernel constructions and transference methods correspond, but D's local conclusion does not by itself prove C's local theorem for general slopes.

Unit 4: Selected lengths and conditional joint laws for fourfold averages

Source: B §§2–6, pp.3–24, in six parts.

The six parts develop the probability construction after length selection, conditional laws given all distal parameters, tensor-projection counting, preservation of three auxiliary groups, stable extensions for specified conditional expectations, and the nonvanishing row–column pairing. Pointwise convergence for at most three factors on general systems is an assumed input here. The analytic proof in B Appendix A is not independently developed. The three-group definition always conditions on the full suspension phase array.

Unit 5: Minimal support and maximal auxiliary positions at arbitrary finite order

Source: A §§2–6, pp.3–21, in three parts. The projection count in Appendix A corresponds to B §4, already developed in Unit 4.

Part I recovers all lengths from smooth triple convergence on general systems and determines the three-state conditional law among arbitrarily many fixed positions. Part II develops globally minimal support, the maximal auxiliary-position count at fixed support, stable single-state conditional expectations, and regrouping into three groups. Part III computes the row–column pairing, proves that it is nonzero, and obtains the contradiction from one additional auxiliary position. The higher-order mixing and relative structure results invoked by the paper remain assumptions of the deduction.

Two essential external dependencies

H: An L3 bound for the trilinear Hilbert transform, October 5, 2026. D, C, and B cite its local quadrilinear estimates and several results within the proof. Boundedness of the main operator alone does not imply the oscillation estimates needed here.

R: Rokhlin’s multiple-mixing problem for one transformation, September 23, 2026. A Theorem 2.1 and the argument in B invoke its claim that ordinary mixing implies mixing of every finite order. A guide and five companion chapters now reconstruct its main proof in Sections 2–7; the guide specifies the scope. The proof of averaged $L^2$ convergence in Unit 1 does not use it.

The distal-structure arguments also require Theorem C of Huang–Shao–Ye and the nonergodic decomposition and relative-structure versions used in the manuscripts. Their precise scope is stated before use in Unit 4.

Source references

Page references use the printed PDF page numbers, which agree with the PDF page order in these four files. References identify the paper by letter and give page, theorem, and equation numbers as appropriate. Text outside quotation marks is our own exposition. The proofs on these pages discuss the original manuscripts; explanatory websites do not replace the primary proofs.

Fixed version: official repository commit fd4aeeb2.

Supplement: standard factors

Basic definitions and examples

One observable and its average

Let $(X,\mathcal F,\mu)$ be a probability space, with $\mu(X)=1$ . A transformation $T:X\to X$ preserves probability if $\mu(T^{-1}E)=\mu(E)$ for every measurable set $E$ . In these papers, $T$ is invertible and both $T$ and $T^{-1}$ are measurable, so $T^n$ is defined for every integer $n$ .

An observable is a measurable function $f$ , whose value at the state $x$ is $f(x)$ . Membership in $L^\infty$ means that a single bound holds outside a null set. Its first $N$ readings along an orbit have average

B_Nf(x)=\frac1N\sum_{n=1}^N f(T^n x).

Birkhoff's ergodic theorem states that, for $f\in L^1$ , these averages converge almost everywhere to $\mathbb E(f\mid\mathcal I)$ , where $\mathcal I$ is the sigma-algebra of invariant sets. If the system is ergodic, meaning that every invariant measurable set has measure zero or one, this limit is the constant $\int f\,\mathrm d\mu$ .

Almost-everywhere convergence means that one exceptional set $Z$ , with $\mu(Z)=0$ , can be removed so that the whole numerical sequence converges at every remaining point. It does not assert convergence at every starting point or a uniform rate.

Conditional expectation

Let $f\in L^1$ and let $\mathcal B\subseteq\mathcal F$ be a sub-sigma-algebra. The conditional expectation $\mathbb E(f\mid\mathcal B)$ is an integrable $\mathcal B$ -measurable function, unique almost everywhere, characterized by $\int_E\mathbb E(f\mid\mathcal B)\,\mathrm d\mu=\int_Ef\,\mathrm d\mu$ for every $E\in\mathcal B$ . For an $L^2$ function it is the orthogonal projection onto the closed subspace of $\mathcal B$ -measurable functions.