VibeMathedMath problems solved with AI

Pilz's Extended 1-2-3 Conjecture

For every finite nonempty set AA of positive integers and every n≥1n\ge1, is ∣A △ 2A △⋯△ nA∣≥n|A\,\triangle\,2A\,\triangle\cdots\triangle\,nA|\ge n? Equivalently, are there at least nn integers mm with an odd number of representations m=ajm=aj with a∈Aa\in A and 1≤j≤n1\le j\le n?

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI co-developed
Method
Argument
Field
Multiplicative combinatorics (odd products, symmetric differences of dilates)
Posed by
G. F. Pilz, On polynomial near-ring codes (Contrib. Gen. Algebra 8, 1992); stated as a conjecture by Huang, Ke and Pilz (Proc. AMS 138, 2010)
Year posed
1992
Years open
34y
Solved
2026-10
Model
GPT-5.6 Sol, GPT-6 Astra (incl. Pro mode), Claude Opus 5.5
Vendor
OpenAI, Anthropic
Collaborators
Ben Cresswell
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
12 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

The conjecture is proved for every nn, in a stronger weighted form over every field: for every field KK and all nonzero c1,…,cn∈Kc_1,\dots,c_n \in K, every nonzero polynomial or Laurent-polynomial multiple of ∑a=1nca∏pxpvp(a)\sum_{a=1}^{n} c_a \prod_p x_p^{v_p(a)} has at least nn terms. Pilz's conjecture is the case K=F2K = \mathbb{F}_2, ca=1c_a = 1. The proof builds on Holdridge and Pach (arXiv:2607.00934), who proved the conjecture for all sufficiently large nn. A companion result proves the optimal density bound h(A)≥4/3h(A) \ge 4/3 (unless A={a}A = \{a\} or {a,2a}\{a,2a\}) that Holdridge and Pach expected in their Remark 1.9.

What the AI did

Under human direction (choice of problem, steering of the strategy, insistence on full formalisation), the models produced the proofs, the written papers and the complete Lean 4 formalisation. The models used are listed from the human author's recollection; the division of work between them was not recorded.

Verification

Read by this site on 15 November 2026, not rebuilt. Repository odd-products/extended-1-2-3-conjecture-of-Pilz at commit b6ab2c2 (Lean 4.24.0, mathlib, three projects): no sorry, admit, axiom declarations, native_decide, unsafe, implemented_by or kernel-bypass options. The headline theorem WeightedPilz.weighted_symmetric_difference states n <= |intervalSymmetricDifference n A| for every finite nonempty A of positive integers, where membership means lying in an odd number of the sets {a, 2a, ..., na}; that is Pilz's conjecture as posed. The check files reject any axiom beyond propext, Classical.choice and Quot.sound. GitHub Actions run 43065981090 on that commit built all three projects and ran the checks successfully. An exhaustive computation written here (all A within 1..16, n <= 10, plus 20,000 random cases) found no violation. The written proofs have not been checked by a human mathematician; the formal statement was audited here only.

Sources

Submitted by RapidKestrel829 on

Changelog2 changes

Discussion