OctoInformBeta

OpenAI preprint, September 2026 · §2. Sharpness on a fixed cube

§2 (setting): affine maxima and the three-atom cube family: The three-branch family and its pushforward measure

Affine maxima on the fixed cube.

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 setting is quadratic transport from a fixed cube to measures concentrated on finitely many sites. The cube family uses a central slab and two outer slopes, with parameters a,b∈ℝa,b\in\mathbb R; its sharpness specialization is b=a/2b=a/2.

Let FF be a real inner product space, let II be a finite nonempty index set, and let yi∈Fy_i\in F and ci∈ℝc_i\in\mathbb R be sites and intercepts. For the cube construction, take d≥2d\ge2 and identify the ambient Euclidean space with ℝd\mathbb R^d. The source cube K⊂ℝdK\subset\mathbb R^d is the set in (1). K=[−1,1]d={x∈ℝd:|xk|≤1 for every k}.K=[-1,1]^d=\{x\in\mathbb R^d: |x_k|\le1\text{ for every }k\}.(1)✦ The uniform measure ρ\rho on KK is normalized Lebesgue measure, given by (2), where ℒd\mathcal L^d denotes Lebesgue measure. ρ=ℒd(K)−1ℒd↾K.\rho=\mathcal L^d(K)^{-1}\,\mathcal L^d\!\restriction_K.(2)✦

Write eke_k for the kkth coordinate unit vector, with coordinates numbered from 00. For coordinate indices i,ji,j and real parameters a,ba,b, the central transport cell Pa,bi,j⊂ℝdP^{i,j}_{a,b}\subset\mathbb R^d is the set in (3). Pa,bi,j={x∈ℝd:|xi|≤a+bxj}.P^{i,j}_{a,b}=\{x\in\mathbb R^d: |x_i|\le a+b x_j\}.(3) The outer slope selection σi:ℝd→ℝd\sigma_i:\mathbb R^d\to\mathbb R^d is the map in (4). σi(x)={−ei,xi<0,ei,xi≥0.\sigma_i(x)= \begin{cases} -e_i,&x_i<0,\\ e_i,&x_i\ge0. \end{cases}(4)✦ For coordinate indices i,ji,j and real parameters a,ba,b, the slab transport map Sa,bi,j:ℝd→ℝdS^{i,j}_{a,b}:\mathbb R^d\to\mathbb R^d assigns to xx the slope of a maximizing branch among −xi-x_i, xix_i and a+bxja+b x_j, with a fixed selection at ties that sends points of the central cell Pa,bi,jP^{i,j}_{a,b} to bejb e_j and sends every other point xx to σi(x)\sigma_i(x).

The cube family chooses i=0i=0 as the first coordinate and j=1j=1 as the transverse coordinate. Its target sites y−,y+,y0∈ℝdy_-,y_+,y_0\in\mathbb R^d are given by (5). y−=−e0,y+=e0,y0=be1.y_-=-e_0,\qquad y_+=e_0,\qquad y_0=b e_1.(5)✦ With t+=max⁡{t,0}t_+=\max\{t,0\}, the target masses w−,w+,w0w_-,w_+,w_0 are the nonnegative weights in (6). w−=w+=(1−a2)+,w0=a+.w_-=w_+=\left(\frac{1-a}{2}\right)_+,\qquad w_0=a_+.(6)✦ The target measure μa,b\mu_{a,b} on ℝd\mathbb R^d is the atomic measure in (7), where δy\delta_y denotes the Dirac measure at yy. μa,b=w−δy−+w+δy++w0δy0.\mu_{a,b}=w_-\delta_{y_-}+w_+\delta_{y_+}+w_0\delta_{y_0}.(7)✦

The Claim

Definition 1. The finite affine potential u:F→ℝu:F\to\mathbb R determined by the sites yiy_i and intercepts cic_i is the function in (8). u(x)=maxi∈I(x⋅yi+ci).u(x)=\max_{i\in I}\bigl(x\cdot y_i+c_i\bigr).(8)✦

Definition 2. Let Y⊂ℝdY\subset\mathbb R^d be a target container. A measure μ\mu on ℝd\mathbb R^d is a three-atom probability measure on YY if it admits sites y1,y2,y3y_1,y_2,y_3 and weights w1,w2,w3w_1,w_2,w_3 satisfying (9). yk∈Y,yk≠yℓ(k≠ℓ),wk>0,∑k=13wk=1,μ=∑k=13wkδyk.\begin{gathered} y_k\in Y,\qquad y_k\ne y_\ell\quad(k\ne\ell),\qquad w_k>0, \qquad \sum_{k=1}^3w_k=1,\\ \mu=\sum_{k=1}^3w_k\delta_{y_k}. \end{gathered}(9)✦

Definition 3. For coordinate indices i,ji,j and real parameters a,ba,b, the three-plane potential va,bi,j:ℝd→ℝv^{i,j}_{a,b}:\mathbb R^d\to\mathbb R is the function in (10). va,bi,j(x)=max⁡{|xi|,a+bxj}.v^{i,j}_{a,b}(x)=\max\{|x_i|,a+b x_j\}.(10)✦

Theorem 1. Let d≥2d\ge2 and let a,b∈ℝa,b\in\mathbb R satisfy 0≤b≤a0\le b\le a and a+b≤1a+b\le1. The slab transport map has the pushforward law (11). (Sa,b0,1)#ρ=μa,b.(S^{0,1}_{a,b})_{\#}\rho=\mu_{a,b}.(11)✦

What Is Checked

The constructions concern finite nonempty affine families in real inner product spaces, three-atom measures in a target container Y⊂ℝdY\subset\mathbb R^d, and coordinate slab maps on KK, with (11) stated for d≥2d\ge2, 0≤b≤a0\le b\le a and a+b≤1a+b\le1.