First lower bounds above n for Naslund's crossing problem, found and proved in Lean by an agent team
A multi-agent harness of Claude models run by Arth Singh proved in Lean, over every field, that the quantity P(n) in Naslund's crossing problem is at least 4n/3, then 15n/11 and 67n/49 for larger n, and that P(4) = 7, the first lower bounds above the trivial n, though all are linear and leave Naslund's superlinear question open.
- Model
- Claude Fable 5.2 lead with Claude Opus 5.5 workers, in an in-house multi-agent harness
- Field
- Mathematics
- Date
- 2026-11-16
- Human collaborators
- Arth Singh
- Problem posed
- 2017 · open 9 yrs
What was found
is the least such that the tensor on is a sum of tensors of rank one along one pairing of the coordinates and along another; equivalently, the least over splittings of the swap matrix as , with the realignment. Naslund asked, as Problem 12 of his paper (arXiv 2017, JCTA 2020), whether . The repository proves for , for , for , for , , and , with an exact-arithmetic script showing the border value at is lower. The paper also shows that no argument of the slice kind used here can reach . The README's account of the run says one agent posed the lower-bound question and proved the trivial bound unprompted, a second improved it, the operator then set the goal of going further, and the bounds above followed within about 74 minutes.
Novelty check
arXiv, 2026-11-16, abs:Naslund returned 17 records, none with a lower bound on this quantity; the nearest are on partition rank and slice rank. The repository reports searching Naslund's paper, Karam (2310.11356), Hrubes, Wigderson and Yehudayoff (2011), Hrubes (2024), Brand, Kaski and Wang (2607.02241) and an appendix of Hexagon 2610.00577 without finding a bound above , and says this is a search result rather than a proof of novelty. Nothing in this registry covered the problem.
Caveats and known objections
Graded formal on the Palomar registration, trust level high, for the seven statements exactly as written in an 88-line Challenge.lean that imports one Mathlib module. Not formalized: the translation between Naslund's tensor form and the matrix form the Lean states, the upper bounds, and the remarks. Only the 4n/3 bound has been checked by hand, by the author's side and in one review of a draft; the others are Lean-only and, the README says, no person has read them in full. The Lean files are concatenated agent output running to thousands of lines, still carrying the agents' own names for the definitions with a bridge to the stated ones. The paper is unrefereed and the result is a day old. Autonomy is ai-led: the agents posed the question and found the first bounds before any person asked, but the operator then set the goals and directed the harness, which humans built.
Machine-checked elsewhere
PalomarLower bounds for Naslund's crossing problem
PALOMAR-2026-11-16-000011 · Arth-Singh/crossing-lower-bounds at 3d302668a9f3 · checked 2026-11-16
ProvedCrossing.four_thirdsCrossing.fifteen_eleventhsCrossing.sixtyseven_fortyninthsCrossing.add_threeCrossing.four_eq_sevenCrossing.five_ge_eightCrossing.six_ge_nine
A Lean proof that typechecks against the recorded statement at a pinned commit under a declared axiom set, checked mechanically rather than by human review. It does not certify that a result is new or of research interest: the only filter on that is a language-model screen, and Palomar states that it adds no human editorial step.
Trust level high. The repository reports that Palomar built the commit, ran Comparator and replayed the proofs, and that its automated review found no blocking problem.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Formally verified and AI-led.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded formally verified for verification and ai-led for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). First lower bounds above n for Naslund's crossing problem, found and proved in Lean by an agent team. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-11-16-naslund-crossing-lower-bounds
BibTeX
@misc{whataifound-aimintelligence-2026-bounds,
title = {First lower bounds above n for Naslund's crossing problem, found and proved in Lean by an agent team},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by AIM Intelligence. Verification: Formally verified. Autonomy: AI-led.},
url = {https://whataifound.org/finding/2026-11-16-naslund-crossing-lower-bounds}
}