A sharp dyadic threshold in a characteristic-two random-cone construction
A finite geometry can be made smaller without changing the recipe it supports. This note identifies the precise dyadic threshold for exponential squared-word decay in one random-cone model: the projective-plane order can be reduced from 128 to 32. It also carries the full existential construction through at that order.
The foundational group construction is OpenAI’s. Its October 4 manuscript supplies the torsion-free group-algebra construction and its probabilistic and geometric framework. This note’s contribution is the exact threshold for the unchanged seven-extra model and the parameter analysis that carries that construction to order 32. The embedding, idempotent and module mechanisms below are classical or inherited.
Original abstract
OpenAI's October 4 construction uses a projective plane of order q=128 and seven extra letters to obtain a finitely presented torsion-free group with a finite two-dimensional classifying complex and a characteristic-two group algebra that is not directly finite. We determine the least dyadic order q≥4 at which its unchanged squared turn weights have exponential word-mass decay: it is q=32. A three-class reduction and exact rational certificates prove contraction at 32 and exponential growth at 4,8,16. Checking the remaining parameter-dependent estimates yields the same construction on a 532-edge rose. This is a threshold for this specified model, rather than an optimum among all group-algebra examples. Classical HNN embedding also gives a two-generator realization. We spell out the inherited scalar idempotent and nonzero absorbed cyclic projective module, which remain valid over every characteristic-two field. The construction is existential; no numerical matching or group-ring multiplication certificate is claimed.
What the threshold measures
Take q=2k with k≥2 and v=q2+q+1 ordinary letters, the points of the projective plane PG(2,q). Add seven extra letters. Each extra letter is paired with a distinct ordinary inverse letter; the remaining ordinary letters are paired among themselves. This is the fixed signed alphabet T of the construction.
For consecutive letters t,u, write inv(t) for t’s inverse letter and put p=(q+1)/v. The original turn weights are:
| Turn | Weight |
|---|---|
| u=inv(t) | 0 |
| u≠inv(t), with inv(t) and u both ordinary | 1/(q+1) |
| inv(t) and u of different kinds | p |
| u≠inv(t), with inv(t) and u both extra | 1/2 |
For a word W=t1⋯th, multiply these turn weights to obtain P(W). They are turn-frequency weights, not probabilities assigned to words. For every allowed inverse pairing, the squared-turn matrix M=(w(t,u)2) has spectral radius below one first at the dyadic order q=32, among q≥4.
At q=32, for every h≥1: ∑W∈ThP(W)2 ≤ (5376/5)(987/1000)h−1.
At q=4,8,16, the same squared-word sum instead has a positive lower bound Cq(503/500)h−1. Thus the decay fails for those smaller dyadic orders in this model. This does not rule out good matchings, idempotents, a different type design or another method at smaller parameters.
From order 32 to a group
At q=32 there are 1057 ordinary letters and seven extras, giving 1064 directed letters and a rose with 532 geometric edges. The parameter analysis carries the entire random-cone argument through: balanced outgoing labels, large-girth conditioned matchings, component diameter, bounded-path estimates, planar arrangement exclusion, asphericity and protection of the distinguished roots.
The resulting existential construction gives a finitely presented torsion-free group G, generated by at most 532 elements, with a finite two-dimensional classifying complex K(G,1). Its characteristic-two group algebra contains finite scalar sums a,b,c satisfying
ab=1, ac=0, c≠0 in 𝔽2[G].
These relations force ba≠1. A classifying complex here has fundamental group G and contractible universal cover. The 532 count describes this particular rose construction; it is not a smallest possible generating set for all examples.
Classical embedding and ring consequences
A classical HNN embedding places G inside a two-generator group H while retaining a finite two-dimensional classifying complex and torsion-freeness. The same scalar relations survive under the induced group-algebra injection. The note spells out this specialization; it does not claim a new embedding mechanism.
Set e=1−ba in 𝔽2[H]. Then e is a scalar idempotent with e2=e and e≠0,1. For every characteristic-two field K, the natural coefficient embedding keeps e nontrivial in S=K[H]. The right ideal P=eS is a nonzero cyclic projective right module, and
SS ≅ SS ⊕ P, [P]=0 in K0(S).
The zero K0 class does not make P the zero module or make K0(S) vanish. It records a failure of cancellation for finitely generated projective modules. These idempotent and module consequences were already recorded in the author’s earlier public triage and are included for interpretation, not offered as a new discovery.
A torsion-free group generated by at most one element is trivial or infinite cyclic, and its group algebra over a field is a domain. This explains the stated two-generator minimum for such an idempotent example. It does not turn the order-32 threshold into an optimum across all construction models.
Scope and attribution
The order-32 result is specific to the unchanged seven-extra types, inverse pairing rule, turn weights and squared-word-decay argument. The construction is existential: it selects a sufficiently large matching outcome without supplying a numerical size threshold, selected matching, relator list, reduced group-ring support or target multiplication certificate.
The source’s central group construction and breakthrough belong to OpenAI, with its September 23 zero-divisor construction also part of the background. HNN embedding is classical. The defect-idempotent reduction and right-module splitting are classical algebraic mechanisms. No first-priority claim is made.
The conclusions concern characteristic-two fields. No characteristic-zero or odd-characteristic conclusion, reduced group C*-algebra projection, full proof-assistant formalization or conventional human refereeing is claimed.
Read the paper and materials
Open the original PDF · Download PDF · Zenodo record · BibTeX
If the embedded PDF is unavailable, open it directly or download the unchanged original.
The manuscript prints 6 October 2026 without a numeric version. The Zenodo record identifies version 1.0 and declares publication on 6 October; the record was created on 7 October UTC. The original files follow in the native record’s returned order:
- paper.pdf (91,584 bytes) · Zenodo original
- source-and-audit.zip (219,261 bytes) · Zenodo original
The native record is CC BY 4.0. The package’s actual rights statement licenses newly authored prose, proofs, explanatory documents, audits and data representation under CC BY 4.0, and newly authored Python code under MIT. Cited material retains its original authorship and rights, including the pinned upstream OpenAI source’s Apache 2.0 terms.
Citation
Alec Kriebel, A sharp dyadic threshold in a characteristic-two random-cone construction, version 1.0, 6 October 2026. doi:10.5281/zenodo.23205181. Download BibTeX.
Status and collaboration
AI tools were used extensively in research, drafting and verification. This is an unrefereed preprint without conventional human peer review. Automated checking is not human endorsement or full formal verification. Citation of OpenAI and other authors does not imply affiliation, coauthorship or endorsement.
Substantive independent verification, mathematical extensions and journal preparation would be useful contributions. Collaboration and authorship should reflect the work actually contributed.
Selected foundations
- OpenAI, A Torsion-Free Group Algebra That Is Not Directly Finite, 4 October 2026, Theorem 1.1 and construction. Pinned original source.
- G. Higman, B. H. Neumann and H. Neumann, Embedding theorems for groups, Journal of the London Mathematical Society 24 (1949), 247–254. DOI.
- Giles Gardam, The Kaplansky conjectures, Proposition
kaplansky_relations. Pinned lecture source. - P. Ara, K. R. Goodearl, K. C. O’Meara and E. Pardo, Separative cancellation for projective modules over exchange rings, report 142 (1996), p. 9. Institutional full text.
- Charles A. Weibel, The K-book (2013), Chapter II, §2. Author’s chapter.