The sharp Kakeya maximal inequality in all dimensions

Get email updates

For n≥2n\ge 2 and 0<δ≤10<\delta\le1, Lewko obtains the LnL^n Kakeya maximal bound with factor (log⁡(2/δ))(n−1)/n(\log(2/\delta))^{(n-1)/n}. The weighted estimate bounds the integral of (∑jaj1Tj)q(\sum_j a_j 1_{T_j})^q by a dimension-dependent constant times log⁡(2/δ) δn−1∑jajq\log(2/\delta)\,\delta^{n-1}\sum_j a_j^q, where q=n/(n−1)q=n/(n-1), the weights are nonnegative, and the unit tubes have radius 2δ2\delta and δ\delta-separated directions.

What was already known

The argument builds on OpenAI’s local-smoothing development. Fernández-Delgado and de la Salle constructed Kakeya sets whose neighborhood volumes have the matching logarithmic order.

AI involvement AI-centered

Mark Lewko credits ChatGPT with refining the local-smoothing ideas to obtain the sharp estimate. Lewko directed the work and formulated the target. His disclosure leaves the attribution of individual proof steps open.

Importance Notable

The two estimates give the sharp logarithmic scale for the maximal bound and the corresponding Kakeya neighborhood-volume problem in every dimension at least two.

Technical verification Formal (Lean)

Palomar reports that Comparator replayed the two statements in Lean 4.35.0 with nanoda and con-ron.

Formal checking

Replayed sharpMaximalBound and sharpTubeBound at the pinned commit with Lean 4.35.0, nanoda and con-ron.

Checker
Palomar Comparator
Scope
For n≥2n\ge 2 and 0<δ≤10<\delta\le1, Lewko obtains the LnL^n Kakeya maximal bound with factor (log⁡(2/δ))(n−1)/n(\log(2/\delta))^{(n-1)/n}. The weighted estimate bounds the integral of (∑jaj1Tj)q(\sum_j a_j 1_{T_j})^q by a dimension-dependent constant times log⁡(2/δ) δn−1∑jajq\log(2/\delta)\,\delta^{n-1}\sum_j a_j^q, where q=n/(n−1)q=n/(n-1), the weights are nonnegative, and the unit tubes have radius 2δ2\delta and δ\delta-separated directions.
Basis
Reported by the linked checker

Human understanding Few

Lewko explains the volume consequence and the radial sharpness example. Ruixiang Zhang asks how the proof reduces the propagation step, and the discussion gives no explanation of that step.

Correction and publication history