VibeMathedMath problems solved by AI

Log-Concavity of Flats of Matroids

Mason conjectured the following: let MM be a matroid of rank rr, and let WiW_i denote the number of flats of MM of rank ii. Is it true that for all 1ir11 \leq i \leq r - 1, we have Wi2Wi+1Wi1W_i^2 \geq W_{i + 1}W_{i - 1}? This is false; a counterexample is given by a graphic matroid whose graph is a generalized theta graph with 7979 edges.

Result
Disproved(see note)
Status
Resolved
AI contribution
AI-discovered
Method
Construction
Field
Matroid theory
Posed by
J. H. Mason
Year posed
1972
Years open
54y
Solved
2026-07-02
Model
ChatGPT-5.5 Pro, Claude Opus 4.8
Vendor
OpenAI; Anthropic
Collaborators
Matt Larson
Verification
Unreviewed
Publication
Preprint
Significance
30 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

Refuting log-concavity of the flat counts is weaker than refuting their unimodality, since log-concavity is the stronger property. A counterexample to unimodality followed three weeks later and is tracked separately as Rota's Unimodality Conjecture for Matroid Flats; this paper came first.

What the AI did

The author prompted ChatGPT-5.5 Pro to search for counterexamples to Mason's conjecture. After finding none on at most 99 elements, the author expanded his search to consider matroids on large ground sets realizable over F5\mathbb{F}_5 and at failures of log-concavity at high indices. ChatGPT-5.5 Pro found a variant of the given counterexample. ChatGPT-5.5 Pro and Claude Opus 4.8 were used for proofreading and generating the figures.

Verification

arXiv preprint with explicit binary matroid construction; not yet peer-reviewed.

Sources

Submitted by HiddenPanther560 on

Changelog2 changes
  • Rasmus Lindahlapproved this entry
  • HiddenPanther560submitted this entry

Discussion