OctoInformBeta

OpenAI preprint, September 2026 · §6. From potentials to maps

Lemma 6.1: Bounded measurable subgradient selections on a convex body admit a geometric boundary error

Supporting maps obey a gradient interpolation bound.

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 passage from transport potentials to maps concerns the difference between their supporting slopes on the source domain. Let dd be a finite dimension and let μ\mu be a translation-invariant volume measure on ℝd\mathbb R^d. For a compact convex source body KK with nonempty interior, write |K||K| for its Lebesgue volume and ρ\rho for its normalized Lebesgue measure, defined by (1). ρ(A)=|A∩K||K|.\rho(A)=\frac{|A\cap K|}{|K|}.(1)

A global subgradient selection T:ℝd→ℝdT:\mathbb R^d\to\mathbb R^d of a function u:ℝd→ℝu:\mathbb R^d\to\mathbb R is a map satisfying the supporting inequalities (2). u(x)+⟨z−x,T(x)⟩≤u(z)(x,z∈ℝd).u(x)+\langle z-x,T(x)\rangle\leq u(z) \qquad (x,z\in\mathbb R^d).(2)✦

For a measurable set K⊆ℝdK\subseteq\mathbb R^d, a step h>0h>0 and a direction e∈ℝde\in\mathbb R^d, the outgoing boundary strip Ωh,e(K)\Omega_{h,e}(K) is the set in (3). Ωh,e(K)={x∈K:x+he∉K}.\Omega_{h,e}(K)=\{x\in K:x+he\notin K\}.(3)

The Claim

Theorem 1. Boundary strip interpolation. Let K⊆ℝdK\subseteq\mathbb R^d be measurable with μ(K)<∞\mu(K)<\infty, let L,C∈ℝL,C\in\mathbb R and let h>0h>0. Let T,S:ℝd→ℝdT,S:\mathbb R^d\to\mathbb R^d be measurable global subgradient selections of u,v:ℝd→ℝu,v:\mathbb R^d\to\mathbb R, respectively, with ∥T(x)∥,∥S(x)∥≤L\|T(x)\|,\|S(x)\|\leq L for every xx. Suppose that u−vu-v is measurable, (u−v)2(u-v)^2 is integrable on KK, and μ(Ωh,e(K))≤Ch\mu(\Omega_{h,e}(K))\leq Ch for every ∥e∥≤1\|e\|\leq1. Then ∫K∥T−S∥2dμ≤12dh2∫K(u−v)2dμ+28dL2Ch.\int_K\|T-S\|^2\,d\mu \leq \frac{12d}{h^2}\int_K(u-v)^2\,d\mu +28dL^2Ch.(4)✦

Convex body interpolation. Let K⊆ℝdK\subseteq\mathbb R^d be compact and convex with nonempty interior. For each L∈ℝL\in\mathbb R, there is a constant B≥0B\geq0, depending only on KK, μ\mu and LL, such that every pair of continuous functions u,v:ℝd→ℝu,v:\mathbb R^d\to\mathbb R with measurable global subgradient selections T,ST,S satisfying ∥T(x)∥,∥S(x)∥≤L\|T(x)\|,\|S(x)\|\leq L for every xx obeys (5) for every h>0h>0. ∫K∥T−S∥2dμ≤12dh2∫K(u−v)2dμ+Bh.\int_K\|T-S\|^2\,d\mu \leq \frac{12d}{h^2}\int_K(u-v)^2\,d\mu+Bh.(5)✦

In the paper, the result reads: for convex LL-Lipschitz functions u0,u1u_0,u_1 on KK, with f=u0−u1f=u_0-u_1 and the paper’s quantities AjA_j and PKP_K, only the almost-everywhere coordinate derivatives are needed for the estimate (6) for every h>0h>0. ∥∇u0−∇u1∥L2(ρ)2≤12dh2∥f∥L2(ρ)2+28L2PK|K|h.\|\nabla u_0-\nabla u_1\|_{L^2(\rho)}^2 \leq \frac{12d}{h^2}\|f\|_{L^2(\rho)}^2 +\frac{28L^2P_K}{|K|}h.(6)

What Is Checked

The claims are stated over the global subgradient selections of (2), with a directional boundary strip bound for (4) and a compact convex source body with nonempty interior for (5).