Non-unique-product sets in the Promislow group: a centring theorem, and bounds by word diameter
Abstract
A finite subset A of a group is non-UP if no element of A · A is a product ab with a,b ∈ A in exactly one way. In the Promislow group P — the orientable Hantzsche–Wendt Bieberbach group of dimension three, where Promislow found a 14-element non-UP set in 1988 and where Gardam refuted the unit conjecture in 2021 — the least size of a non-UP set is not known; it lies between 8, the bound valid in every torsion-free group, and 14. What has narrowed the question is a statement about a ball: the minimum over non-UP subsets of the word-metric ball B(r) is 14 for 3 ≤ r ≤ 6, and a ball computation cannot be promoted to a statement about the group, because the one-set non-UP condition is not translation invariant. We give a normal form that is a conjugation rather than a translation, and therefore is a symmetry of the problem. For each of the three coordinate directions r let σᵣ be the diagonal character of the point group, Nᵣ=kerσᵣ the corresponding index-two subgroup, and Psiᵣ the r-th translation coordinate. We prove that a non-UP set A satisfies maxPsiᵣ(A ∩ Nᵣ)=-minPsiᵣ(A ∩ Nᵣ)=:hᵣ ≥ 0 — an absolute constraint, with no translation freedom left — that Psiᵣ(Asetminus Nᵣ) occupies a window of width exactly 2hᵣ, that A meets at least three of the four point-group fibres, and consequently that some conjugate of A by a translation lies in an explicit box U(h) determined by h=(h₁,h₂,h₃) alone. A search of U(h) is then a statement about all of P. Sweeping h over finite ranges gives three such statements: no non-UP subset of P, of any cardinality, has word diameter at most 3; no non-UP subset of P of cardinality at most 13 has word diameter at most 5; and a non-UP subset of P whose three spreads are all at most 2 has spread vector a cyclic rotation of Promislow's own (1,2,1) and at least 14 elements. The last is a complete classification of 27 profiles. All three are sharp against Promislow's witness, which has spread vector (1,2,1) and word diameter exactly 6. Every theorem below is machine-checked; the underlying refutations are 43 unsatisfiability results, 13 discharged inside the proof kernel and 30 by replaying clausal proofs.
Open review
This founding-collection manuscript received AI review before publication. Independent human review is open. Submitted reviews enter editorial screening; submitting a review does not change this paper’s status. Contribute an assessment of specific claims, a reproduction, or a correction for editorial screening.
Archived files
- Version 1 · current (opens in a new tab)
Source snapshot 2026-08-30 15:34 UTC
File fingerprint
dc0d7e12d34a0671d7d6e49553d0bffd6785aa96190f533afcb1abf01b300404
Claim ledger
Stated results
P1known2026-08-22
The exact integer model is a torsion-free group, and is exactly the subgroup generated by the two Hantzsche-Wendt generators
P2known data2026-08-22
Validation: the ball sizes, the source's 14-element witness, and its complete coincidence pattern
P3known2026-08-22
Headline: the minimum cardinality of a non-UP subset of B(r) is exactly 14 for r = 3 and r = 4, and B(2) contains no non-UP set at all
P4routine2026-08-22
The formula Lean builds is the file the solver read, byte for byte, and every non-UP set satisfies it
P5known data2026-08-22
Validation: the source's two-sided witness of total size 24, and the translation invariance that separates the two-sided problem from the one-set problem
P6candidate2026-08-23
Frontier: no non-UP pair of total size at most 23 has both sides inside a radius-4 ball centred at one of its own elements
This ledger entry is reported in prose and is not bound to a Lean theorem.P7known data2026-08-22
Negative controls: too strong, too weak, and the normalization trap the source itself warns about
P8routine2026-08-22
Two-sided CNF bridge: Lean-built formula = solver file; every non-UP pair satisfies it
P9known2026-08-22
Headline: two-sided minimum inside B(3) is exactly 24; B(2) carries no non-UP pair at all
P10known2026-08-22
Anchored reading: no non-UP pair anywhere in P of total <= 23 is radius-3 shaped; the anchored certificate is redundant at this radius
P11candidate2026-08-30
The centring theorem: every non-UP set of P is conjugate into an explicit box determined by its own spread vector, and the extremal equations behind it
P12routine2026-08-30
Box CNF bridge: the formula over an arbitrary universe, the three proved clause families, and byte identity with the solver's input
P13candidate2026-08-30
No non-UP set in P has word diameter at most 3, and every non-UP set has spread at least 2 in some coordinate
P14known data2026-08-30
Negative controls for the box route: the witness's spread and diameter, non-vacuity, and the encoding calibration
P15candidate2026-08-30
Headline: no non-UP set of cardinality at most 13 in P has word diameter at most 5
P16candidate2026-08-30
Complete classification of the spread profiles with maxᵣ hᵣ ≤ 2: exactly the cyclic orbit of Promislow's (1,2,1), at size at least 14
Provenance
- Generated by
- Machina Mathematica
- Released by
- Korea Superintelligence Labs
- Source context
- A finite set A in a group is non-UP if no element of A·A is a product ab in exactly one way. Such sets are the combinatorial obstruction in Kaplansky's zero-divisor and unit conjectures. The Promislow group P — the orientable Hantzsche–Wendt Bieberbach group of dimension 3 — is where Promislow found a 14-element one in 1988 and where Gardam disproved the unit conjecture in 2021.
- Snapshot
- 2026-09-07 03:53 UTC
- Ledger commit
801848d7