OpenAI reports asymptotically minimal maxima for real Littlewood polynomials

For every eta > 0 and every sufficiently large integer N, there is a Littlewood polynomial with N consecutive coefficients in {−1,1} whose maximum modulus on the unit circle is at most (1+eta) sqrt(N).

Sources for this record

What was already known

Shapiro–Rudin constructions gave real-sign maxima of order sqrt(N), with a larger constant; complex unimodular coefficients admitted ultraflat constructions. A fixed relative gap above the Parseval lower bound was conjectured for real signs (companion manuscript introduction, pp. 1–2).

AI involvement AI-led

OpenAI credits itself as manuscript author and states that its internal frontier model produced the results in this release. Individual human input is not detailed enough to establish task-only autonomy.

OpenAI is named as author on the manuscript, and its release attributes the results to an internal frontier model. The individual human input is not detailed enough to establish autonomy.

Importance Major

Proves that the minimum normalized maximum tends to one through all integer lengths, refuting the real-sign fixed-gap conjecture.

Limitations

This is a one-sided maximum bound. It supplies no uniform positive lower bound for the modulus on the circle and does not establish two-sided real-sign ultraflatness. It is existential without an efficient signing algorithm.

Technical verification Formal (Lean 4)

OpenAI reports a Lean proof matching the selected claim.

Formal checking

The pinned Lean note gives a real Littlewood polynomial of every sufficiently large length with maximum at most (1+eta) sqrt(N), for each fixed eta > 0.

Checker
OpenAI-reported Lean 4 / Comparator
Scope
For every eta > 0 and all sufficiently large lengths N, a real-sign Littlewood polynomial of length N has circle maximum modulus at most (1+eta) sqrt(N).
Basis
Reported by the linked checker
Evidence source

Checking a formal statement does not establish that it matches the original problem.

Human understanding Few

Few is the default when evidence for a higher level is insufficient. No named person is documented as explaining the central argument for this assessed contribution.

Correction and publication history

2026-10-18 · publication

Primary companion manuscript, attribution, prior-work, and exact Lean-scope review for non-notifying local inclusion.

Read this published revision