---
id: "2026-10-14-hardy-littlewood-chowla-mobius-shifted-primes-lu-wang-wu"
url: "https://postcutoff.com/e/2026-10-14-hardy-littlewood-chowla-mobius-shifted-primes-lu-wang-wu/"
as_of: "2026-10-15T10:09:00+02:00"
date: "2026-10-14"
date_precision: day
category: science
importance: 4
confidence: medium
status: [Awaiting review]
verification: Preprint only
sources: 3
editor: human editor
human_review: null
version: "2026-10-15"
---

As of: 2026-10-15 10:09 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-10-14-hardy-littlewood-chowla-mobius-shifted-primes-lu-wang-wu/

# Lü, Wang & Wu claim µ(p+2) = o(π(x)), a model case of the parity problem open since at least 1989, building on an OpenAI preprint with ChatGPT's help

On 14 Oct 2026 Guangshi Lü, Zhiwei Wang and Jie Wu posted arXiv 2610.20242 (52 pages). It claims a proof that Σ_{p≤x} µ(p+2) = o(π(x)): the Möbius function cancels on the shifted primes p + 2. The statement is a folklore case of the Hardy–Littlewood–Chowla conjecture, in print since Hildebrand (1989), and a standard model case for the parity problem in sieve theory. A corollary is that µ(p+2) = +1 and −1 each occur for half of the primes with p + 2 squarefree. The authors thank ChatGPT for help understanding the methods of an OpenAI Math Release preprint on prime predecessors, and for the appendix. The result is unrefereed.

## Key facts
- Paper: 'On the Hardy–Littlewood–Chowla conjecture', arXiv 2610.20242 (math.NT, 14 Oct 2026, 52 pages)
- Theorem 1: Σ_{p≤x} µ(p+2) = o(π(x)), equivalently Σ_{n≤x} Λ(n)µ(n+2) = o(x). The shift is fixed at 2 'for simplicity'
- Theorem 2: #{p ≤ x : µ(p+2) = κ} = (A₂/2 + o(1))π(x) for κ = ±1, with A₂ = ∏_{p>2}(1 − 1/(p(p−1))); this addresses the question in Sarnak's AIM problem 5.2 about both sign patterns along shifted primes
- Before: only averaged versions (Lichtman 2022; Lichtman–Teräväinen) or results conditional on Siegel zeros (Tao–Teräväinen 2022)
- AI statement (verbatim): 'The authors would like to thank ChatGPT for its assistance in understanding the methods in [11], and in establishing the Appendix A. The authors also thank ChatGPT for helping to check the manuscript.'
- Reference [11] is OpenAI, 'The Poisson–Dirichlet law for prime predecessors' (OpenAI Math Release, 24 Sep 2026 version), family 011 of the 6 Oct release, which claims the Ford–Konyagin–Luca conjecture. Appendix A reproves its Lemma 3.3 (a root-counting estimate) with an explicit error term 'so that this input can be verified within the present paper'
- Related OpenAI claim in the same family: infinitely many primes p with µ(p−1) = 1 ('Prime Predecessors with an Even Number of Prime Factors', 17 Sep 2026)

## What happened

Chowla's conjecture predicts that the Möbius function µ has no correlation with its own shifts. The Hardy–Littlewood conjecture predicts how often prime patterns occur. The mixed Hardy–Littlewood–Chowla conjecture combines the two. Its simplest case, Σ_{p≤x} µ(p+h) = o(π(x)), says that p + h is as likely to have an even as an odd number of prime factors when p runs over the primes. The paper says this "is a folklore conjecture and a well-known model case for the parity problem in sieve theory", in print at least since Hildebrand (1989). Earlier results covered only averages over the shift h (Lichtman; Lichtman–Teräväinen) or assumed that Siegel zeros exist (Tao–Teräväinen).

Lü, Wang and Wu claim the case h = 2 unconditionally ([arXiv 2610.20242](https://arxiv.org/abs/2610.20242)). They prove it with Type I/Type II estimates and a "parity conjugacy" bijection (Section 4). Their root-counting input comes from OpenAI's preprint "The Poisson–Dirichlet law for prime predecessors". To make the paper self-contained, they reprove that lemma in Appendix A.

AI statement: "The authors would like to thank ChatGPT for its assistance in understanding the methods in [11], and in establishing the Appendix A. The authors also thank ChatGPT for helping to check the manuscript."

No Hacker News thread was found on 15 Oct (Algolia). X was not searched directly. No expert reaction was found yet.

## Why it matters

If the proof holds, it gets past the parity barrier in one of the problem's simplest standard test cases. It also shows a route that is becoming common: human experts take the method of a machine-written OpenAI preprint, have a chatbot explain it, and turn it into a result OpenAI did not claim.

## What is disputed or not yet verified

- Verification: Unrefereed preprint; no Lean formalization

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 167 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 106 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 197 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 136 days after its cutoff

## Sources

1. [arXiv 2610.20242: On the Hardy–Littlewood–Chowla conjecture (Lü, Wang, Wu)](https://arxiv.org/abs/2610.20242) (arxiv.org, paper)
2. [OpenAI math release CONTENTS.md (family 011: prime-factor statistics of p − 1)](https://github.com/openai/math/blob/main/CONTENTS.md) (github.com, paper)
3. [AIM problem list: Sarnak's conjecture, problem 5.2](https://aimpl.org/sarnakconjecture/) (aimpl.org, docs)

## Changes

- 2026-10-15 (filed): Created from the arXiv PDF and the openai/math CONTENTS.md

## Related

- 2026-10-14: [Mid-October 2026: arXiv results with disclosed AI help](https://postcutoff.com/e/2026-10-14-ai-assisted-results-mid-october-2026/index.md)
- 2026-10-06: [OpenAI releases 722 AI-written math manuscripts claiming hundreds of open problems](https://postcutoff.com/e/2026-10-06-openai-math-release-722-manuscripts/index.md)
- 2026-10-06: [OpenAI model claims the quasi-Riemann hypothesis](https://postcutoff.com/e/2026-10-06-quasi-riemann-hypothesis-openai/index.md)