---
id: "2026-10-06-openai-borel-conjecture-dimension-four-free-group-not-good"
url: "https://postcutoff.com/e/2026-10-06-openai-borel-conjecture-dimension-four-free-group-not-good/"
as_of: "2026-10-30T13:42:00+01:00"
date: "2026-10-06"
date_precision: day
category: science
importance: 5
confidence: medium
status: [Awaiting review]
verification: Preprint only
sources: 7
editor: human editor
human_review: null
version: "2026-10-29"
---

As of: 2026-10-30 13:42 CET.
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/2026-10-06-openai-borel-conjecture-dimension-four-free-group-not-good/

# OpenAI claims 4D Borel counterexample and that F₂ is not good

Full title: OpenAI's release claims the topological Borel conjecture fails in dimension four, that the free group is not 'good' for Freedman's disk embedding, and a counterexample to Wall's PD4 realization question

Family 305 (three manuscripts, 24 Sept 2026) claims that the unrestricted four-dimensional disk embedding conjecture fails, so F₂, and every group containing it, is not good in the sense of Freedman–Quinn. It also claims a finitely presented PD4 group with a finite classifying space that is not the fundamental group of any closed aspherical 4-manifold. Family 320 (one manuscript, 4 Oct) uses these constructions to give closed aspherical topological 4-manifolds, with a common word-hyperbolic fundamental group, that are homotopy equivalent but not homeomorphic. The smooth analogue had been settled in 2024 (homeomorphic, not diffeomorphic, aspherical 4-manifolds; Davis, Hayden, Huang, Ruberman, Sunukjian). No Lean formalization, no withdrawal; no published expert assessment found.

## Key facts
- Disk embedding (abstract, verbatim): 'We disprove the four-dimensional disc embedding conjecture without a fundamental-group hypothesis… Consequently, the free group on two generators is not good in the sense of Freedman–Quinn, and neither is any group containing it as a subgroup.' A second manuscript gives a 'marked tensor' obstruction to the same conjecture
- Why it matters inside the field: Freedman's disk embedding theorem underlies 4D topological surgery and the s-cobordism theorem; it is proved for 'good' fundamental groups (e.g. groups of subexponential growth), and whether all groups, in particular F₂, are good has been the field's main open question since the 1980s
- Wall's question (verbatim): 'We construct a finitely presented integral Poincaré duality group of dimension four that has a finite classifying space but is not the fundamental group of any closed aspherical topological four-manifold. This gives a negative answer to Wall's manifold-realization question in dimension four.'
- Borel (abstract, verbatim): 'We construct closed connected aspherical topological four-manifolds that are homotopy equivalent but not homeomorphic, disproving the homeomorphism-existence formulation of the Borel conjecture in dimension four. Their common fundamental group is word-hyperbolic.' The paper also gives a self-homotopy equivalence not homotopic to a homeomorphism
- Prior state, as the Borel paper recounts it: rigidity known in dimensions ≤ 3 (dimension 3 via geometrization); Farrell–Jones and Bartels–Lück in dimensions ≥ 5 for hyperbolic and CAT(0) groups; in dimension four only with a good fundamental group or up to s-cobordism (Khan). The 2024 smooth examples of Davis–Hayden–Huang–Ruberman–Sunukjian are homeomorphic but not diffeomorphic
- Reaction: Xiaolei Wu (group theorist) on Proofs and Prompts, 8 Oct: 'So the Borel conjecture fails in dimension 4, the coarse Baum-Connes also fails…'. Vyacheslav Krushkal's 'collision rank' paper (arXiv 2609.25198, 21 Sept, before the release) enlarged the known class of good groups
- Status: manuscripts only; no Lean scope notes for families 305 or 320; Latent Seminar's exposition (7 Oct) explicitly disclaims independent verification

## What happened

In the topological category, four-dimensional manifolds are governed by Michael Freedman's disk embedding theorem (1980s). It makes
surgery and the s-cobordism theorem work in dimension four, but only when the fundamental group is "good", a class that includes
groups of subexponential growth. Whether every group is good, and in particular the free group on two generators, has been the main
open question of the subject since the 1980s.

OpenAI's release says the answer is no. Two manuscripts dated 24 Sept give obstructions to the disk embedding conjecture without a
group hypothesis, from which F₂ is not good. A third builds a four-dimensional Poincaré duality group that no closed aspherical
4-manifold realizes, a negative answer to a question of C. T. C. Wall. A manuscript dated 4 Oct combines these constructions into two
closed aspherical 4-manifolds that are homotopy equivalent but not homeomorphic. That contradicts the Borel conjecture, first raised by
Armand Borel in 1953, in dimension four. Rigidity holds in dimensions up to three and for broad classes of groups in dimensions five and
up. The smooth version was already known to fail in dimension four (2024), but these examples are homeomorphic, so they say nothing
about the topological statement.

## Why it matters

If correct, these results close the "is F₂ good?" problem in the negative and make dimension four an exception to topological
rigidity of aspherical manifolds. Both would change how 4-manifold topologists work. The claims rest on long, unrefereed,
unformalized AI manuscripts. We found no published assessment by 4-manifold topologists as of 29 Oct.

## What is disputed or not yet verified

- Verification: Unrefereed manuscripts; no formalization; no published expert assessment

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 159 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 98 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 189 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 128 days after its cutoff

## Sources

1. [OpenAI: A boundary-only obstruction to four-dimensional disk embedding (PDF)](https://github.com/openai/math/blob/main/preprints/A-boundary-only-obstruction-to-four-dimensional-disk-embedding-September-24-2026/paper.pdf) (github.com, paper)
2. [OpenAI: A marked tensor obstruction to four-dimensional disk embedding (PDF)](https://github.com/openai/math/blob/main/preprints/A-marked-tensor-obstruction-to-four-dimensional-disk-embedding-September-24-2026/paper.pdf) (github.com, paper)
3. [OpenAI: A PD4 group without an aspherical manifold model (PDF)](https://github.com/openai/math/blob/main/preprints/A-PD4-group-without-an-aspherical-manifold-model-September-24-2026/paper.pdf) (github.com, paper)
4. [OpenAI: Nonhomeomorphic closed aspherical four-manifolds with the same homotopy type (4 Oct 2026, PDF)](https://github.com/openai/math/blob/main/preprints/Nonhomeomorphic-closed-aspherical-four-manifolds-with-the-same-homotopy-type-October-4-2026/paper.pdf) (github.com, paper)
5. [Davis, Hayden, Huang, Ruberman, Sunukjian: Exotic aspherical 4-manifolds (arXiv 2411.19400, smooth case)](https://arxiv.org/abs/2411.19400) (arxiv.org, paper)
6. [Proofs and Prompts: 100+ reactions to 100+ solutions (Xiaolei Wu and others)](https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/) (proofsandprompts.com, discussion)
7. [oai-math-revised: community notes on the Borel manuscript](https://github.com/mathcommunity/oai-math-revised/blob/main/preprints/Nonhomeomorphic-closed-aspherical-four-manifolds-with-the-same-homotopy-type-October-4-2026/AGENT_NOTES.md) (github.com, discussion)

## Changes

- 2026-10-29 (filed): Created (families 251–378; four manuscripts' abstracts and the Borel paper's introduction read, Proofs and Prompts, oai-math-revised notes)

## Related

- 2026-10-24: [oai-math-revised: mathematicians launch a community hub to revise OpenAI's AI-written math manuscripts (guest post on Tao's blog)](https://postcutoff.com/e/2026-10-24-oai-math-revised-community-hub/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)