Documentary state: EDITORIALLY_ADMITTED. The manuscript has a protected editorial admission, not a theorem certification.
OAI-26 family 075: Almost-everywhere Fourier convergence in L log L
Editorial state: source-pinned documentary candidate. Mathematical state: NO_CERTIFICATION. All theorem claims below are attributed to the pinned publisher manuscript, not independently established by GCL.
The question in plain language
A periodic signal can be expressed as an infinite sum of waves with integer frequencies. Its ordinary Fourier partial sum retains only frequencies between minus an integer cutoff and plus that cutoff. Does this approximation converge to the original signal at almost every point, even when the signal contains large spikes? A classical conjecture asks for this conclusion at the integrability threshold usually called L log L. This means the average of the magnitude times the logarithm of two plus the magnitude is finite. It is a stronger condition than mere integrability, but weaker than membership in any fixed power-integrability class above one.
A small example
Consider a finite trigonometric polynomial such as a constant plus a cosine. Once the symmetric Fourier cutoff includes its highest frequency, every subsequent partial sum equals that polynomial at every point. This is elementary and says nothing about the hard case: an arbitrary measurable function may have infinitely many active frequencies and arbitrarily tall peaks. For such functions, convergence almost everywhere allows a function-dependent set of exceptions of measure zero, but not a positive-measure failure set.
What is known
The pinned September 23, 2026 manuscript, Almost-everywhere Fourier convergence in L log L, claims that for every complex-valued measurable function on the circle whose integral of magnitude times log of two plus magnitude is finite, all ordinary symmetric Fourier partial sums converge to that function almost everywhere along the full sequence of integer cutoffs. The claim is not limited to a subsequence, real-valued functions, or bounded functions. The source distinguishes this endpoint claim from Carleson's square-integrable theorem, Hunt's extension to each power-integrability class above one, Kolmogorov's counterexample for unrestricted integrable functions, and earlier near-endpoint results involving additional logarithms.
The source's introduction describes a route through a good-set maximal integral estimate. For a normalized nonnegative density of controlled height it bounds the integral of the supremum of the symmetric Fourier sums on a set where the spatial maximal density is not too large. The exceptional set is small; a decomposition into height layers makes these estimates summable at the claimed endpoint. Its finite dyadic argument combines delayed Haar sums, modulation frequencies, a noisy-observation and relative-entropy compression principle, positive correlations for noise densities, and separate budgets for information and representative-frequency selection. Translation and scale averaging then transfer dyadic estimates to the periodic Hilbert transform and Fourier projections.
The manuscript also describes a weak-type maximal estimate derived from its convergence theorem. That maximal consequence must be checked with the exact norms and hypotheses of its cited corollary, rather than assumed to be interchangeable with the good-set estimate.
GCL verification state
The source is locked to openai/math@1581d4cbdd69a0faf34b2089614fe2a814316a6a. The primary TeX abstract, introduction, main theorem, and proof roadmap were read directly at that commit. The publisher's Lean documentation describes a matching full-sequence statement, but neither the complete Lean dependency closure nor the manuscript's analytic proof has been independently replayed here. This is a source-fidelity editorial claim, not a mathematical certification.
What remains
Audit the noisy-frequency compression lemma and its distribution-uniform quantifiers, the thermal correlation inequality, path and information budgets, dyadic stopping rules, the translation/scale averaging into the Hilbert kernel, the exceptional-set measure estimate, and the layer summation yielding almost-everywhere full-sequence convergence. Independently replay the publisher comparator theorem and its dependencies, then distinguish formal source-statement matching from proof soundness. Documentary admission additionally requires a separate critical-review record, exact article SHA-256, compendium regression, strict documentation rendering, protected CI, merge-queue admission, and current-main readback.
Sources and claim boundaries
NO_CERTIFICATION.
Indexed primary manuscript paths:
- Almost-everywhere Fourier convergence in L log L —
preprints/Almost-everywhere-Fourier-convergence-in-L-log-L-September-23-2026/paper.pdf