OctoInformBeta

OpenAI preprint, September 2026 · §3. A metric and volume bound for connected supports

Probability realization, transportation metric, and tree areas: Normalized transportation distance on finitely supported measures and averaged areas in tree coordinates

Probability geometry from transport and spanning trees.

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 geometry of connected supports in the Bass trace problem concerns barycentric mass on simplices and areas measured in tree coordinates. Let YY be a vertex set, and write ℝ(Y)\mathbb R^{(Y)} for the finitely supported real functions on YY. For a connected simple graph Γ\Gamma on YY, the graph metric dΓd_\Gamma is shortest-path distance.✦ Let KK be a simplicial complex on YY, with barycentric realization ℛ(K)⊆ℝ(Y)\mathcal R(K)\subseteq\mathbb R^{(Y)}.

For a pseudometric dd on YY, a function f:Y→ℝf:Y\to\mathbb R is 11-Lipschitz when |f(y)−f(z)|≤d(y,z)|f(y)-f(z)|\leq d(y,z) for all vertices y,zy,z. The normalization parameter is an integer n≥0n\geq0.

Tree Coordinates

For the area construction, let σ\sigma be a simplex with vertices {0,…,m}\{0,\ldots,m\} and let Γσ\Gamma_\sigma be a simple graph on these vertices. A spanning tree TT of Γσ\Gamma_\sigma is a connected acyclic graph on the vertices of Γσ\Gamma_\sigma whose edges belong to Γσ\Gamma_\sigma; write 𝒯(Γσ)\mathcal T(\Gamma_\sigma) for the finite set of these trees.✦ For each tree TT, enumerate its mm edges and choose ordered endpoints (uj,vj)(u_j,v_j) for the jjth edge. A tree edge enumeration of TT is a bijection from {1,…,m}\{1,\ldots,m\} to the edge set of TT.

The rooted vertex coordinate rv∈ℤmr_v\in\mathbb Z^m records the vertex vv after the coordinate of vertex 00 has been omitted.✦ The integer reduced incidence matrix CTC_T has the oriented edge coordinates as columns, as in (1)✦; its real reduced incidence matrix BTB_T is the same matrix viewed over ℝ\mathbb R.✦ rv(i)={1,v=i,0,v≠i,(CT)ij=ruj(i)−rvj(i),1≤i,j≤m.r_v(i)= \begin{cases}1,&v=i,\\0,&v\ne i,\end{cases} \qquad (C_T)_{ij}=r_{u_j}(i)-r_{v_j}(i), \qquad 1\leq i,j\leq m.(1)

The barycentric parametrization b:ℝm→ℝm+1b:\mathbb R^m\to\mathbb R^{m+1} and the tree parametrization ΦT:ℝm→ℝm+1\Phi_T:\mathbb R^m\to\mathbb R^{m+1} are the affine maps in (2), with the domain of ΦT\Phi_T given its usual Euclidean metric. b(x)=(1−∑i=1mxi,x1,…,xm),ΦT(x)=b(BTx).b(x)=\left(1-\sum_{i=1}^m x_i,x_1,\ldots,x_m\right), \qquad \Phi_T(x)=b(B_Tx).(2) For an integer j≥0j\geq0, write ℋEj\mathcal H_E^j for jj-dimensional Hausdorff measure on ℝm\mathbb R^m with its usual Euclidean metric.

The Definition

Definition 1. A finitely supported probability measure pp on YY is a nonnegative real function with finite support and total mass one; their set 𝒫f(Y)\mathcal P_f(Y) is given by (3). 𝒫f(Y)={p∈ℝ(Y):p(y)≥0 for every y∈Y,∑y∈Yp(y)=1}.\mathcal P_f(Y)= \left\{p\in\mathbb R^{(Y)}: p(y)\geq0\text{ for every }y\in Y,\quad \sum_{y\in Y}p(y)=1\right\}.(3)✦

For p,q∈𝒫f(Y)p,q\in\mathcal P_f(Y) and a real function ff on YY, the expectation 𝔼p[f]\mathbb E_p[f] and expectation difference Δp,q(f)\Delta_{p,q}(f) are the finite sums in (4). 𝔼p[f]=∑y∈Yp(y)f(y),Δp,q(f)=𝔼p[f]−𝔼q[f].\mathbb E_p[f]=\sum_{y\in Y}p(y)f(y), \qquad \Delta_{p,q}(f)=\mathbb E_p[f]-\mathbb E_q[f].(4) The dual transportation distance W1W_1 on 𝒫f(Y)\mathcal P_f(Y) for a pseudometric dd on YY is the supremum in (5). W1(p,q)=sup⁡{Δp,q(f):f:Y→ℝ is 1-Lipschitz}.W_1(p,q)=\sup\left\{\Delta_{p,q}(f): f:Y\to\mathbb R\text{ is }1\text{-Lipschitz}\right\}.(5)✦ The normalized transportation distance ρ\rho on 𝒫f(Y)\mathcal P_f(Y) is the function given by (6). ρ(p,q)=W1(p,q)n.\rho(p,q)=\frac{W_1(p,q)}{\sqrt n}.(6)✦ If n>0n>0 and d(y,z)≥1d(y,z)\geq1 for distinct vertices, the probability metric d𝒫f(Y)d_{\mathcal P_f(Y)} on 𝒫f(Y)\mathcal P_f(Y) is the metric specified in (7). d𝒫f(Y)(p,q)=ρ(p,q).d_{\mathcal P_f(Y)}(p,q)=\rho(p,q).(7)✦

For a family 𝒮\mathcal S of finite subsets of YY, the probability realization 𝒫(𝒮)\mathcal P(\mathcal S) is the subset of 𝒫f(Y)\mathcal P_f(Y) in (8). 𝒫(𝒮)={p∈𝒫f(Y):supp(p)⊆τ for some τ∈𝒮}.\mathcal P(\mathcal S)= \left\{p\in\mathcal P_f(Y): \operatorname{supp}(p)\subseteq\tau\text{ for some }\tau\in\mathcal S\right\}.(8)✦ For v∈ℛ(K)v\in\mathcal R(K), its barycentric probability measure pvp_v has coefficients pv(y)=v(y)p_v(y)=v(y).✦ Given a connected graph Γ\Gamma and a center c∈𝒫f(Y)c\in\mathcal P_f(Y), the function rc:ℝ(Y)→ℝr_c:\mathbb R^{(Y)}\to\mathbb R measuring distance from cc is defined by (9), using dΓd_\Gamma in ρ\rho. rc(v)={ρ(c,pv),v∈ℛ(K),0,v∉ℛ(K).r_c(v)= \begin{cases} \rho(c,p_v),&v\in\mathcal R(K),\\ 0,&v\notin\mathcal R(K). \end{cases}(9)✦

For a tree TT on the vertices of σ\sigma, an integer j≥0j\geq0 and a subset Q⊆ℝm+1Q\subseteq\mathbb R^{m+1}, the tree area ajT(Q)a_j^T(Q) is the Hausdorff measure of the inverse image in (10). ajT(Q)=ℋEj(ΦT−1(Q))∈[0,∞].a_j^T(Q)=\mathcal H_E^j\bigl(\Phi_T^{-1}(Q)\bigr)\in[0,\infty].(10)✦ The averaged simplex area ajσ(Q)a_j^\sigma(Q) for Γσ\Gamma_\sigma is the extended nonnegative real value in (11). ajσ(Q)=∑T∈𝒯(Γσ)ajT(Q)|𝒯(Γσ)|∈[0,∞].a_j^\sigma(Q)= \frac{\displaystyle\sum_{T\in\mathcal T(\Gamma_\sigma)}a_j^T(Q)} {|\mathcal T(\Gamma_\sigma)|}\in[0,\infty].(11)✦

In the paper, the result reads: for a real radius ss, writing Lip⁡(f)\operatorname{Lip}(f) for the Lipschitz constant of ff and relint⁡σ\operatorname{relint}\sigma for the relative interior of σ\sigma, |K||K| consists of finitely supported probability vectors with simplicial support, ρ(p,q)=1nsup⁡Lip⁡(f)≤1∑yf(y)(p(y)−q(y))\rho(p,q)=\frac1{\sqrt n} \sup_{\operatorname{Lip}(f)\leq1}\sum_y f(y)(p(y)-q(y)) uses graph distance, Ls(c)={p∈|K|:ρ(c,p)≤s}L_s(c)=\{p\in|K|:\rho(c,p)\leq s\}, Us(c)={p∈|K|:ρ(c,p)<s}U_s(c)=\{p\in|K|:\rho(c,p)<s\} and Ss(c)={p∈|K|:ρ(c,p)=s}S_s(c)=\{p\in|K|:\rho(c,p)=s\} are respectively the closed ball, open ball and sphere, and on a top simplex σ\sigma the measure ajσa_j^\sigma is the average of jj-dimensional Hausdorff measures, restricted to relint⁡σ\operatorname{relint}\sigma, over the Euclidean structures in which the oriented edge vectors of each spanning tree are orthonormal.

What Is Checked

The probability constructions are stated over a pseudometric vertex set, with n>0n>0 and distances between distinct vertices at least one for the probability metric; the center function uses a connected graph and a simplicial complex. The area constructions are stated over a graph on the vertices of a finite simplex, an integer j≥0j\geq0 and an arbitrary subset of the real coordinate space.