OctoInformBeta

OpenAI preprint, September 2026 · §6. Equality of the two infima

§6 (setting): finite-cost approximation by preserving pairs: The definition requires finite Coulomb cost for every positive error

Approximating the Kantorovich value by preserving pairs.

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

Fix a Borel measure μ\mu on Euclidean space ℝ3\mathbb{R}^3. The costs to be compared are those of pairs of maps preserving μ\mu and of probability measures with three marginals equal to μ\mu.

The triple configuration space Ω=(ℝ3)3\Omega=(\mathbb{R}^3)^3 is the Cartesian product of three copies of ℝ3\mathbb{R}^3, with configurations written (x,y,z)(x,y,z).✦ The first, second and third coordinate projections p1,p2,p3p_1,p_2,p_3 of Ω\Omega are maps given by (1). p1(x,y,z)=x,p2(x,y,z)=y,p3(x,y,z)=z.p_1(x,y,z)=x,\qquad p_2(x,y,z)=y,\qquad p_3(x,y,z)=z.(1) A three-marginal self-coupling of μ\mu is a probability measure π\pi on Ω\Omega whose coordinate pushforwards equal μ\mu; the collection of these couplings is Γ(μ)\Gamma(\mu) in (2). Γ(μ)={π a probability measure on Ω:(p1)#π=(p2)#π=(p3)#π=μ}.\Gamma(\mu)= \left\{\pi\text{ a probability measure on }\Omega: (p_1)_\#\pi=(p_2)_\#\pi=(p_3)_\#\pi=\mu\right\}.(2)✦

The inverse distance |x−y|−1|x-y|^{-1} of points x,y∈ℝ3x,y\in\mathbb{R}^3 is the reciprocal of their Euclidean distance, with value +∞+\infty when x=yx=y.✦ The three-particle Coulomb cost c:Ω→[0,+∞]c:\Omega\to[0,+\infty] is defined by (3), with value +∞+\infty at every collision. c(x,y,z)=1|x−y|+1|x−z|+1|y−z|.c(x,y,z)=\frac{1}{|x-y|}+\frac{1}{|x-z|}+\frac{1}{|y-z|}.(3)✦ The Kantorovich value E*(μ)E_*(\mu) is the infimum of the nonnegative extended Coulomb integral over three-marginal self-couplings, as in (4). E*(μ)=infπ∈Γ(μ)∫Ωc(x,y,z)dπ(x,y,z).E_*(\mu)=\inf_{\pi\in\Gamma(\mu)}\int_{\Omega}c(x,y,z)\,d\pi(x,y,z).(4)✦

A measure-preserving map for μ\mu is a measurable map T:ℝ3→ℝ3T:\mathbb{R}^3\to\mathbb{R}^3 satisfying (5). T#μ=μ.T_\#\mu=\mu.(5)✦ For maps T2,T3:ℝ3→ℝ3T_2,T_3:\mathbb{R}^3\to\mathbb{R}^3, their Monge transport cost Cμ(T2,T3)C_\mu(T_2,T_3) is the nonnegative extended integral in (6). Cμ(T2,T3)=∫ℝ3c(x,T2(x),T3(x))dμ(x).C_\mu(T_2,T_3)= \int_{\mathbb{R}^3}c\bigl(x,T_2(x),T_3(x)\bigr)\,d\mu(x).(6)✦

The Definition

Definition 1. A measure μ\mu on ℝ3\mathbb{R}^3 admits finite-cost Monge approximation if, for every real ε>0\varepsilon>0, there are measure-preserving maps T2,T3:ℝ3→ℝ3T_2,T_3:\mathbb{R}^3\to\mathbb{R}^3 for μ\mu satisfying (7). Cμ(T2,T3)<+∞,Cμ(T2,T3)≤E*(μ)+ε.C_\mu(T_2,T_3)<+\infty, \qquad C_\mu(T_2,T_3)\leq E_*(\mu)+\varepsilon.(7)✦

In the paper, the result reads: for the constructed μ\mu, finite-cost preserving pairs approach E*(μ)E_*(\mu) arbitrarily closely, although no such pair attains it.

What Is Checked

The definition (7) is stated for an arbitrary Borel measure μ\mu on ℝ3\mathbb{R}^3, with probability self-couplings and measurable maps preserving μ\mu.