VibeMathedMath problems solved with AI

The L'vov-Kaplansky conjecture

Let KK be a field and let f∈K⟨X1,…,Xm⟩f\in K\langle X_1,\ldots,X_m\rangle be a multilinear noncommutative polynomial. Is the image f(Mn(K)m)={f(A1,…,Am):Ai∈Mn(K)}f(M_n(K)^m)=\{f(A_1,\ldots,A_m): A_i\in M_n(K)\} a vector subspace of Mn(K)M_n(K) for every n≥1n\ge1?

Result
Proved(see note)
Status
Partial result
AI contribution
AI-discovered
Method
Argument
Field
Ring Theory, Noncommutative Algebra
Posed by
I. V. L'vov, Dniester Notebook, Problem 1.98 (4th ed., Novosibirsk 1993); the question is also associated with Kaplansky
Year posed
1993
Years open
33y
Solved
2026-11-08
Model
GPT-6 Pro (research runs); Codex (proof audit)
Vendor
OpenAI
Collaborators
Qiyue Tang
Verification
Unreviewed
Publication
Preprint
Significance
30 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 1.1: for each d≥2d\ge2 there is an integer B(d)B(d) such that for every field KK, every nonzero ordinary homogeneous multilinear polynomial ff of degree dd over KK and every n≥B(d)n\ge B(d), f(sln(K)d)=f(Mn(K)d)f(\mathfrak{sl}_n(K)^d)=f(M_n(K)^d), and this image is sln(K)\mathfrak{sl}_n(K) if ff is formally traceless (every cyclic coefficient sum vanishes) and Mn(K)M_n(K) otherwise. So the image is a vector space for all n≥B(d)n\ge B(d). B(d)B(d) is not made explicit, and the orders n<B(d)n<B(d) remain open.

What the AI did

From the Hexagon record's provenance statement: "The main results were developed over ten GPT-6-Pro research runs with a combined scheduled duration of 62.5 hours. Codex was used to audit the proofs and polish the manuscript. A Rethlas-OSV review will be added in a later version." The record is labelled "A mix of human-written and AI-generated text" and "Human understanding: some parts". The PDF itself carries no AI statement.

Verification

Read by this site on 15 November 2026. The source is a Hexagon deposit (hexagon:2611.00206v1, deposited 8 November, version 1 of 10 November) by Qiyue Tang, who is identified by ORCID with no affiliation; the record labels the text as mixed human and AI-written and the author's understanding as covering only some parts. Theorem 1.1 was read and matches the result note. The proof (33 pages) was not checked. A small computation written here over GF(7) for degrees 3 and 4 with n = d agreed in 80 of 80 random cases that the trace of f vanishes identically exactly when f is formally traceless, the elementary half of the dichotomy; the surjectivity for n at least B(d) was not tested, since B(d) is not explicit. No formalisation.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion