Skip to content

Documentary state: DRAFT_UNACCEPTED. This draft has NOT been editorially admitted. Mathematical claims require independent checking.

OAI-26 family 007: two-point Chowla and corrected binary Elliott correlations

Documentary status: DRAFT_UNACCEPTED. NO_CERTIFICATION. This article describes the claims in a pinned September 2026 manuscript; it does not independently certify the argument.

The question in plain language

Prime factorization creates arithmetic patterns that can seem random. One way to test that intuition is to assign each positive integer a sign according to whether it has an even or odd number of prime factors, counted with multiplicity. This gives the Liouville function. The two-point Chowla conjecture predicts cancellation when two related integer inputs are tested together, provided the associated affine forms are not proportional.

The broader Elliott correlation question replaces the Liouville function with other bounded multiplicative functions. These functions respect multiplication on coprime inputs, but can imitate simple periodic or oscillatory arithmetic patterns. The corrected conjecture therefore needs a precise nonpretentiousness hypothesis excluding that obstruction.

The pinned OAI-26 overview claims that Family 007 proves both an ordinary quantitative two-point Chowla estimate and a corrected binary Elliott theorem under an explicitly uniform nonpretentiousness condition.

A small example

Define the Liouville function by

\[ \lambda(n)=(-1)^{\Omega(n)}, \]

where \(\Omega(n)\) counts prime factors with repetition. Thus \(\lambda(6)=1\), since \(6=2\cdot3\) has two prime factors, while \(\lambda(8)=-1\), since \(8=2^3\) has three.

A simple pair of affine forms is \(n\) and \(n+1\). The two-point question asks whether the products \(\lambda(n)\lambda(n+1)\) cancel on average. If they behaved like largely independent signs, their sum through \(X\) would be much smaller than \(X\). That analogy motivates the conjecture but is not proof.

The publisher's claimed bound is stronger than merely saying the normalized correlation tends to zero: it gives a logarithmic saving of the schematic form

\[ \left|\sum_{n\leq X}\lambda(a_1n+b_1)\lambda(a_2n+b_2)\right| \leq C\frac{X}{(\log X)^c}, \]

for a positive absolute exponent \(c\) and an implied constant depending on the fixed admissible forms. This display illustrates the indexed claim, not an independently checked transcription of its full hypotheses. Inputs of the forms must be positive where the functions are evaluated.

What is known

The pinned source catalogue names one manuscript, Ordinary two-point correlations of multiplicative functions (September 24, 2026). Its indexed synopsis claims:

  • The ordinary two-point Chowla conjecture for fixed nonproportional affine forms, with a bound of order \(X/(\log X)^c\) for some absolute \(c>0\).
  • A more general corrected binary Elliott conjecture for complex-valued multiplicative functions bounded in magnitude by one, when one function is uniformly nonpretentious against each fixed Dirichlet character multiplied by \(n^{it}\), for real \(t\) of size at most \(X\).

The nonproportionality restriction excludes degenerate pairs whose values are tightly linked by a constant multiple. The uniformity and character-twist clauses matter because a multiplicative function may track a character or an archimedean phase rather than display cancellation.

These statements are publisher-reported manuscript results. Their precise quantified constants, possible exceptional primes, treatment of zero or negative arguments, and proof dependencies require an exact-PDF source audit before editorial acceptance.

GCL verification state

The GCL Family 007 bibliographic entry and publisher overview are pinned to openai/math@1581d4cbdd69a0faf34b2089614fe2a814316a6a. This draft verifies the existence and stated scope of the one indexed manuscript, not its internal proof.

No independent validation of the logarithmic saving, nonpretentiousness quantifiers, or transfer from the general Elliott statement to Chowla is established here. No Family 007-specific Lean theorem, Comparator result, semantic-equivalence audit, or MATHCERT certificate has been identified. The existence of a paper and a source-locked catalogue entry is not mathematical proof verification.

What remains

A source critic must read the pinned PDF, locate the exact theorem numbers and definitions, and test that the two affine forms satisfy all positivity and nondegeneracy hypotheses. The critic should check whether the exponent \(c\) is truly absolute, which parameters the implicit constant may depend on, and whether the result is on the ordinary interval average rather than only a logarithmically weighted average.

For the corrected Elliott theorem, verify the precise range and order of quantifiers for characters and \(t\), the nonpretentiousness distance employed, and any assumptions on both multiplicative functions. An editorial reviewer should also distinguish a logarithmic saving from stronger power savings or uniformity in forms that the source does not claim.

After source-critical review, run the prescribed mathematical Markdown, compendium and rendered HTML checks, record content-bound reviewer evidence, and merge through protected governance. Documentary admission does not establish the claimed theorems.

Sources and claim boundaries

Boundary: NO_CERTIFICATION. This article records a claim of progress on two-point multiplicative correlations; it is not a validated proof, mathematical certification, or assertion of stronger quantitative estimates.


Indexed primary manuscript paths: