OctoInformBeta

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

Proposition 6.1 [Equality of infima with strict nonattainment]: Finite Monge costs and equal infima, with strict nonattainment for a Coulomb counterexample density

Compactly supported atomless marginals admit approximation from separated optimizers.

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

Three-particle transport with a common marginal compares probability measures on configurations with configurations prescribed by two maps of the first particle. The configuration space is (ℝ3)3(\mathbb R^3)^3, with points (x,y,z)(x,y,z); write p1p_1, p2p_2 and p3p_3 for its first, second and third coordinate projections.

The extended Coulomb cost cc on (ℝ3)3(\mathbb R^3)^3 is the function given by (1), with value +∞+\infty at every collision: c(x,y,z)=1|x−y|+1|x−z|+1|y−z|.c(x,y,z)=\frac1{|x-y|}+\frac1{|x-z|}+\frac1{|y-z|}.(1)✦ The inverse distance |x−y|−1|x-y|^{-1} between points x,y∈ℝ3x,y\in\mathbb R^3 is the reciprocal of their Euclidean distance, with value +∞+\infty when the points coincide.✦

A three-marginal self-coupling π\pi of a Borel probability measure μ\mu on ℝ3\mathbb R^3 is a probability measure on (ℝ3)3(\mathbb R^3)^3 whose three coordinate marginals equal μ\mu.✦ A measure-preserving map for μ\mu is a Borel map T:ℝ3→ℝ3T:\mathbb R^3\to\mathbb R^3 with T#μ=μT_\#\mu=\mu.✦ The coupling class Π(μ)\Pi(\mu) and the class 𝒜(μ)\mathcal A(\mu) of admissible pairs of maps are given by (2): Π(μ)={π a probability measure on (ℝ3)3:(pi)#π=μ(i=1,2,3)},𝒜(μ)={(T2,T3):T2,T3 are Borel,(T2)#μ=(T3)#μ=μ}.\begin{aligned} \Pi(\mu)&=\{\pi\text{ a probability measure on }(\mathbb R^3)^3: (p_i)_\#\pi=\mu\ (i=1,2,3)\},\\ \mathcal A(\mu)&=\{(T_2,T_3):T_2,T_3\text{ are Borel}, (T_2)_\#\mu=(T_3)_\#\mu=\mu\}. \end{aligned}(2)

For a Borel probability measure μ\mu on ℝ3\mathbb R^3, the coupling cost I(π)I(\pi) of π∈Π(μ)\pi\in\Pi(\mu) and the Monge transport cost Gμ(T2,T3)G_\mu(T_2,T_3) of (T2,T3)∈𝒜(μ)(T_2,T_3)\in\mathcal A(\mu) are the nonnegative extended integrals in (3). The Kantorovich value E*(μ)E_*(\mu) and the Monge value M(μ)M(\mu) of a Borel probability measure μ\mu on ℝ3\mathbb R^3 are the infima of the coupling and Monge transport costs, respectively, given by (4). I(π)=∫c(x,y,z)dπ(x,y,z),Gμ(T2,T3)=∫c(x,T2(x),T3(x))dμ(x).I(\pi)=\int c(x,y,z)\,d\pi(x,y,z),\qquad G_\mu(T_2,T_3)=\int c(x,T_2(x),T_3(x))\,d\mu(x).(3) E*(μ)=infπ∈Π(μ)I(π),M(μ)=inf(T2,T3)∈𝒜(μ)Gμ(T2,T3).E_*(\mu)=\inf_{\pi\in\Pi(\mu)}I(\pi),\qquad M(\mu)=\inf_{(T_2,T_3)\in\mathcal A(\mu)}G_\mu(T_2,T_3).(4)

For η>0\eta>0, a configuration (x,y,z)(x,y,z) in (ℝ3)3(\mathbb R^3)^3 is η\eta-separated when the inequalities in (5) hold: |x−y|≥η,|x−z|≥η,|y−z|≥η.|x-y|\geq\eta,\qquad |x-z|\geq\eta,\qquad |y-z|\geq\eta.(5)✦ A Borel probability measure μ\mu on ℝ3\mathbb R^3 admits finite-cost Monge approximation if, for every real ε>0\varepsilon>0, there is a pair (T2,T3)∈𝒜(μ)(T_2,T_3)\in\mathcal A(\mu) with finite cost at most E*(μ)+εE_*(\mu)+\varepsilon.✦

Counterexample Densities

The density measure μ\mu of a real function ρ\rho on ℝ3\mathbb R^3 is the measure obtained by weighting Lebesgue measure by ρ+=max⁡{ρ,0}\rho_+=\max\{\rho,0\}; for nonnegative ρ\rho write μ=ρdx\mu=\rho\,dx.✦ A smooth compact probability density ρ\rho on ℝ3\mathbb R^3 is a nonnegative real function such that ρ\rho and ρ\sqrt\rho are both smooth and compactly supported, and ∫ρdx=1\int\rho\,dx=1.✦

A Borel probability measure μ\mu on ℝ3\mathbb R^3 has Kantorovich attainment if some π∈Π(μ)\pi\in\Pi(\mu) has I(π)=E*(μ)I(\pi)=E_*(\mu).✦ A Borel probability measure μ\mu on ℝ3\mathbb R^3 has Monge nonattainment if Gμ(T2,T3)>E*(μ)G_\mu(T_2,T_3)>E_*(\mu) for every (T2,T3)∈𝒜(μ)(T_2,T_3)\in\mathcal A(\mu).✦ A Coulomb counterexample density is a smooth compact probability density ρ\rho whose measure μ=ρdx\mu=\rho\,dx is a probability measure with finite E*(μ)E_*(\mu), Kantorovich attainment and Monge nonattainment.✦ A smooth compact probability density ρ\rho on ℝ3\mathbb R^3 satisfies the full Coulomb conclusion if ρ\rho is a Coulomb counterexample density and its measure μ=ρdx\mu=\rho\,dx satisfies M(μ)=E*(μ)M(\mu)=E_*(\mu) and admits finite-cost Monge approximation.✦

The Claim

Theorem 1. Let (X,dX)(X,d_X) be a metric space whose Borel σ\sigma-algebra makes it a standard Borel space, and let YY be a nonempty standard Borel space. Let α\alpha be a finite atomless measure on XX, β\beta a finite measure on YY, and f:Y→Xf:Y\to X a measurable map with f#β=αf_\#\beta=\alpha. Suppose K⊂XK\subset X is compact and α(X\K)=0\alpha(X\setminus K)=0. For every δ>0\delta>0 there is a measurable map Q:X→YQ:X\to Y satisfying (6): Q#α=β,dX(x,f(Q(x)))<δfor α-almost every x.Q_\#\alpha=\beta,\qquad d_X(x,f(Q(x)))<\delta\quad\text{for }\alpha\text{-almost every }x.(6)✦

Let μ\mu be a Borel probability measure on ℝ3\mathbb R^3, π∈Π(μ)\pi\in\Pi(\mu), and Q:ℝ3→(ℝ3)3Q:\mathbb R^3\to(\mathbb R^3)^3 measurable with Q#μ=πQ_\#\mu=\pi. Put T2=p2∘QT_2=p_2\circ Q and T3=p3∘QT_3=p_3\circ Q. Suppose η>0\eta>0, δ≤η/2\delta\leq\eta/2, π\pi is concentrated on η\eta-separated configurations, and |x−p1(Q(x))|≤δ|x-p_1(Q(x))|\leq\delta for μ\mu-almost every xx. The pair (T2,T3)(T_2,T_3) satisfies (7), where r+=max⁡{r,0}r_+=\max\{r,0\}: (T2,T3)∈𝒜(μ),Gμ(T2,T3)<+∞,Gμ(T2,T3)≤I(π)+(24δη2)+.\begin{gathered} (T_2,T_3)\in\mathcal A(\mu),\qquad G_\mu(T_2,T_3)<+\infty,\\ G_\mu(T_2,T_3)\leq I(\pi)+\left(\frac{24\delta}{\eta^2}\right)_+. \end{gathered}(7)✦

Let μ\mu be an atomless Borel probability measure on ℝ3\mathbb R^3. Suppose KK is a compact set with μ(ℝ3\K)=0\mu(\mathbb R^3\setminus K)=0, and π∈Π(μ)\pi\in\Pi(\mu) satisfies I(π)=E*(μ)I(\pi)=E_*(\mu) and is concentrated on η\eta-separated configurations for some η>0\eta>0. Then (8) holds: μ admits finite-cost Monge approximation.\mu\text{ admits finite-cost Monge approximation}.(8)✦

For any Borel probability measure μ\mu on ℝ3\mathbb R^3 admitting finite-cost Monge approximation, the infima satisfy (9): M(μ)=E*(μ).M(\mu)=E_*(\mu).(9)✦

Let ρ\rho be a Coulomb counterexample density and μ=ρdx\mu=\rho\,dx. Suppose π∈Π(μ)\pi\in\Pi(\mu) satisfies I(π)=E*(μ)I(\pi)=E_*(\mu) and is concentrated on η\eta-separated configurations for some η>0\eta>0. Then (10) holds: ρ satisfies the full Coulomb conclusion.\rho\text{ satisfies the full Coulomb conclusion}.(10)✦

In the paper, the result reads: for the probability measure μ\mu constructed in Theorem 1.1, the infimum of Gμ(T2,T3)G_\mu(T_2,T_3) over Borel maps T2,T3:ℝ3→ℝ3T_2,T_3:\mathbb R^3\to\mathbb R^3 with (T2)#μ=(T3)#μ=μ(T_2)_\#\mu=(T_3)_\#\mu=\mu equals E*(μ)E_*(\mu); every admissible pair has cost strictly above this common infimum, but finite-cost pairs approach it.

What Is Checked

The Coulomb assertions are stated over Borel self-couplings and measure-preserving maps of μ\mu, including the density measures μ=ρdx\mu=\rho\,dx, and the lifting assertion is over finite standard Borel measures with compact concentration of the atomless marginal.