Counterexamples to two conjectures about matroids

Matt Larson

A graphic matroid formed from four paths between two vertices, three of length 26 and one of length 1, has flat counts W₇₄²=967,813,250,625 < W₇₃W₇₅=982,359,393,275. This refutes Mason’s plain log-concavity conjecture for matroid flats.

Sources for this record

What was already known

Mason’s flat-count log-concavity conjecture was open. Results for characteristic-polynomial coefficients and independent-set counts do not prove it.

AI involvement AI-centered

ChatGPT found a variant of the graph after Matt Larson directed a search toward large F5-representable matroids and high-index failures. Larson supplied the counting proof and paper.

The disclosed model search found the counterexample variant under Larson’s directions; the proof and exposition are human work.

Importance Notable

The explicit graphic counterexample resolves the flat-count log-concavity question while leaving the weaker unimodality question separate.

Limitations

This example does not itself refute unimodality. A later human-led paper develops nonunimodal examples using additional constructions.

Technical verification Formal (human)

Larson supplies and claims authorship of a complete human combinatorial count for the AI-found construction, ending in an explicit strict inequality. This is a source-reported completed proof check of the construction, not a reproduced independent calculation.

Independent checking

Human dual-matroid cyclic-set enumeration and exact integer inequality in Theorem 2.3

Checker
Matt Larson, author
Scope
Example 2.2 violates Mason’s flat-count log-concavity at index 74
Basis
Reported by the linked checker
Evidence source

Human understanding Full

Larson explains the generalized-theta graph, its dual parallel classes, the cyclic-set counts and why unequal class sizes create failure.

Correction and publication history

2026-10-18 · publication

The explicit graphic counterexample resolves the flat-count log-concavity question while leaving the weaker unimodality question separate.

Read this published revision