OctoInformBeta

OpenAI preprint, September 2026 · §3. Sheets and extremal derivatives

Lemma 3: At first coordinates close to the maximum AA, they are at most B+εB+\varepsilon; for B≥0B \ge 0 an affine barrier follows, and holds in mean

Second coordinates of a compact set near its lexicographic maximum.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

The lexicographic maximum of a nonempty compact set KK in the plane is the point (A,B)(A,B) of KK such that AA is the largest first coordinate in KK and BB is the largest second coordinate among the points of KK with first coordinate AA.

Let SS be a metric space with distance dd whose points are written (p,w)(p,w), with pp a point of the base plane ℝ2\mathbb{R}^2, and let ff be a map from SS to a finite-dimensional real space with d(z,z′)≤∥f(z)−f(z′)∥≤Dd(z,z′)d(z,z') \le \|f(z)-f(z')\| \le D\,d(z,z') for a constant D>0D>0. A sheet FF of ff is a map F(p)=f(p,w)F(p)=f(p,w) on the open set of base points pp with (p,w)∈S(p,w)\in S, for a fixed ww, and FxF_x, FyF_y are its partial derivatives in the horizontal and vertical directions. A good point pp of a sheet FF is a base point at which the derivative columns Fx(p)F_x(p) and Fy(p)F_y(p) are taken. The set KK is the closure of the pairs (∥Fx(p)∥2,∥Fy(p)∥2)(\|F_x(p)\|^2, \|F_y(p)\|^2) over all sheets FF of ff and their good points pp, and (A,B)(A,B) is its lexicographic maximum.

A random point of a set KK in the plane is a pair of integrable functions ss, tt on a measure space (X,μ)(X,\mu) with (s(x),t(x))∈K(s(x),t(x)) \in K for almost every xx. The mean of an integrable function hh on XX is 𝔼h=∫Xhdμ\mathbb{E}\,h = \int_X h\,d\mu.

How far can the second coordinate tt of a point (s,t)(s,t) of KK exceed BB when its first coordinate ss is close to AA, and what does a bound of that kind give for the means of a random point of KK?

The Claim

Theorem 1. Let KK be a compact set in the plane and AA, BB real numbers such that every point (s,t)(s,t) of KK has s≤As \le A, and has t≤Bt \le B when s=As = A. Let ε>0\varepsilon > 0.

  1. There is δ>0\delta > 0 such that t≤B+εfor every (s,t)∈K with s>A−δ.t \le B + \varepsilon \quad \text{for every } (s,t) \in K \text{ with } s > A - \delta .(1)✦

  2. If B≥0B \ge 0 and every point of KK has second coordinate at most MM, for some M≥0M \ge 0, there is C≥0C \ge 0 such that KK lies below the affine barrier t≤B+ε+C(A−s)for every (s,t)∈K.t \le B + \varepsilon + C\,(A - s) \quad \text{for every } (s,t) \in K .(2)✦

  3. Let now KK be any set in the plane lying below a barrier (2), for real numbers AA, BB, ε\varepsilon, CC of any sign. Then every random point (s,t)(s,t) of KK satisfies 𝔼t≤B+ε+C(A−𝔼s).\mathbb{E}\,t \le B + \varepsilon + C\,(A - \mathbb{E}\,s) .(3)✦

In the paper, the result reads: for D>0D > 0 and a nonempty compact set K⊆[0,D2]2K \subseteq [0,D^2]^2 with lexicographic maximum (A,B)(A,B), every ε>0\varepsilon > 0 has a δ>0\delta > 0 such that t≤B+εt \le B + \varepsilon whenever (s,t)∈K(s,t) \in K and s>A−δs > A - \delta, and consequently, if measurable pairs (sn,tn)(s_n,t_n) take values in KK and 𝔼(A−sn)→0\mathbb{E}(A - s_n) \to 0, then limsup⁡n𝔼tn≤B\limsup_n \mathbb{E}\,t_n \le B.

The library answers that limit with (3), applied to the barrier (2) for each ε>0\varepsilon > 0.

What Is Checked

The claim is stated over a compact set KK in the plane with real numbers AA, BB bounding its coordinates as its lexicographic maximum does, a bound M≥0M \ge 0 on its second coordinates together with B≥0B \ge 0, and ε>0\varepsilon > 0; the integrated form (3) is stated over any set in the plane below an affine barrier and a random point of it.