VibeMathedMath problems solved with AI

The Daykin–Frankl conjecture on convex subsets of the Boolean lattice

In 1983, Daykin and Frankl conjectured that if PP is a convex subset of QnQ_n, then it contains at leastP(nn/2)2n |P|\binom{n}{\lfloor n/2\rfloor}2^{-n} pairwise incomparable elements. We verify and communicate an LLM-generated proof of this conjecture.

Result
Proved(see note)
Status
Resolved
AI contribution
AI-discovered
Method
Argument
Field
Extremal set theory
Posed by
David E. Daykin, Peter Frankl
Year posed
1983
Years open
43y
Solved
2026-09-02
Model
GPT-5.6 Sol Pro
Vendor
OpenAI
Collaborators
Kada Williams
Verification
Unreviewed
Publication
Preprint
Significance
25 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

Let PQnP\subseteq Q_n be convex. Williams proves the stronger statement that for every k0k\ge0,w(P×Qk)w(Qn+k)P2n, w(P\times Q_k) \ge w(Q_{n+k})\,|P|\,2^{-n}, where ww denotes poset width.

Taking k=0k=0 givesw(P)P(nn/2)2n, w(P)\ge |P|\binom{n}{\lfloor n/2\rfloor}2^{-n}, which is exactly the Daykin-Frankl conjecture.

The proof proceeds by induction on nn, reducing the step to a structural lemma for a convex subset of R×Q1R\times Q_1 and carefully recombining antichains from its two layers.

What the AI did

Kada Williams explicitly credits ChatGPT 5.6 Sol Pro with generating the proof content of the note. The proof establishes a stronger product inequality for convex subsets of Boolean lattices and derives the original Daykin-Frankl conjecture as the case k=0k=0. Williams verifies, writes up, and takes responsibility for communicating the argument.

Verification

Unreviewed. arXiv 2609.03087 (four pages) read here: the note describes itself as verifying and communicating an LLM-generated proof, credits ChatGPT 5.6 Sol Pro, and gives the induction on dimension with the R x Q_1 convexity lemma in full. Checked by the human author, not by anyone independent; not peer reviewed; no formalization.

Sources

Submitted by VibeGene on

Changelog2 changes

Discussion