OctoInformBeta

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

Lemma 3.2: Normalized tree gradients satisfy a unit bound, preserved under affine isometric restriction

Tree simplices have prescribed Euclidean volume.

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 Euclidean size of a simplex and the variation of a vertex potential can be measured using a spanning tree on its vertices.

Fix a nonnegative integer nn and the vertex set V={0,…,n}V=\{0,\ldots,n\}, with root 00. The full barycentric simplex Δn⊂ℝn+1\Delta_n\subset\mathbb R^{n+1} and the standard solid simplex Dn⊂ℝnD_n\subset\mathbb R^n are the sets in (1): Δn={p∈ℝn+1:pi≥0(i∈V),∑i=0npi=1},Dn={x∈ℝn:xi≥0(1≤i≤n),∑i=1nxi≤1}.\begin{aligned} \Delta_n&=\left\{p\in\mathbb R^{n+1}:p_i\geq0\ (i\in V),\ \sum_{i=0}^n p_i=1\right\},\\ D_n&=\left\{x\in\mathbb R^n:x_i\geq0\ (1\leq i\leq n),\ \sum_{i=1}^n x_i\leq1\right\}. \end{aligned}(1) The barycentric parametrization β:ℝn→ℝn+1\beta:\mathbb R^n\to\mathbb R^{n+1} is the affine map defined by (2): β(x)=(1−∑i=1nxi,x1,…,xn).\beta(x)=\left(1-\sum_{i=1}^n x_i,x_1,\ldots,x_n\right).(2)✦

Tree Coordinates and Areas

A spanning tree TT of a simple graph Γ\Gamma on VV is a tree on VV whose edges belong to Γ\Gamma; write 𝒯(Γ)\mathcal T(\Gamma) for the set of such trees.✦ For a tree TT on VV, choose a tree edge enumeration, a bijection from {1,…,n}\{1,\ldots,n\} to its edges, and choose an ordered pair of endpoints (aj,bj)(a_j,b_j) for its jjth edge.

The rooted vertex coordinate r(v)∈ℤnr(v)\in\mathbb Z^n of v∈Vv\in V is the vector in (3): r(v)i={1v=i,0v≠i,1≤i≤n.r(v)_i=\begin{cases}1&v=i,\\0&v\ne i,\end{cases} \qquad 1\leq i\leq n.(3)✦ The integer reduced incidence matrix BTℤB_T^{\mathbb Z} has as its columns the rooted coordinate differences of the ordered edge endpoints✦; its real reduced incidence matrix BTB_T is obtained by applying the inclusion ℤ↪ℝ\mathbb Z\hookrightarrow\mathbb R to each entry, as in (4): (BTℤ)ij=r(aj)i−r(bj)i,(BT)ij=(BTℤ)ij∈ℝ.(B_T^{\mathbb Z})_{ij}=r(a_j)_i-r(b_j)_i, \qquad (B_T)_{ij}=(B_T^{\mathbb Z})_{ij}\in\mathbb R.(4)

The tree parametrization PT:ℝn→ℝn+1P_T:\mathbb R^n\to\mathbb R^{n+1} is the affine map obtained by composing BTB_T with β\beta✦, and the tree solid simplex DT⊂ℝnD_T\subset\mathbb R^n is the inverse image of DnD_n under BTB_T: PT(x)=β(BTx),DT={x∈ℝn:BTx∈Dn}.P_T(x)=\beta(B_Tx),\qquad D_T=\{x\in\mathbb R^n:B_Tx\in D_n\}.(5) Both constructions in (5) use the usual Euclidean structure on ℝn\mathbb R^n.

Let ℋj\mathcal H^j denote jj-dimensional Euclidean Hausdorff measure. The tree area ajT(Q)a_j^T(Q) of a set Q⊂ℝn+1Q\subset\mathbb R^{n+1} is the measure of its inverse image under PTP_T.✦ For a simple graph Γ\Gamma on VV and a set Q⊂ℝn+1Q\subset\mathbb R^{n+1}, the averaged simplex area ajΓ(Q)a_j^\Gamma(Q) is the average of the tree areas ajT(Q)a_j^T(Q) over the spanning trees TT of Γ\Gamma, as in (6); both tree area and averaged simplex area take values in the extended nonnegative reals [0,∞][0,\infty].✦ ajT(Q)=ℋj(PT−1(Q)),ajΓ(Q)=∑T∈𝒯(Γ)ajT(Q)|𝒯(Γ)|.a_j^T(Q)=\mathcal H^j\bigl(P_T^{-1}(Q)\bigr),\qquad a_j^\Gamma(Q)=\frac{\displaystyle\sum_{T\in\mathcal T(\Gamma)}a_j^T(Q)} {|\mathcal T(\Gamma)|}.(6)

For n>0n>0, a vertex potential ℓ\ell on VV is a function from VV to ℝ\mathbb R. The normalized tree gradient gT(ℓ)∈ℝng_T(\ell)\in\mathbb R^n of a vertex potential ℓ:V→ℝ\ell:V\to\mathbb R on a tree TT is the vector of edge differences in (7), with its Euclidean norm: gT(ℓ)=1n(ℓ(aj)−ℓ(bj))j=1n.g_T(\ell)=\frac{1}{\sqrt n} \bigl(\ell(a_j)-\ell(b_j)\bigr)_{j=1}^n.(7)✦

The Claim

Theorem 1. Let TT be a tree on VV. The volume and tree area identities are (8) and (9), where vol⁡n\operatorname{vol}_n is Euclidean volume: vol⁡n(DT)=1n!.\operatorname{vol}_n(D_T)=\frac{1}{n!}.(8)✦ anT(Δn)=1n!.a_n^T(\Delta_n)=\frac{1}{n!}.(9)✦ For a connected simple graph Γ\Gamma on VV, the averaged area satisfies (10): anΓ(Δn)=1n!.a_n^\Gamma(\Delta_n)=\frac{1}{n!}.(10)✦ Suppose n>0n>0 and ℓ:V→ℝ\ell:V\to\mathbb R satisfies |ℓ(u)−ℓ(v)|≤1|\ell(u)-\ell(v)|\leq1 on every edge of TT. Its normalized tree gradient satisfies (11): ∥gT(ℓ)∥≤1.\|g_T(\ell)\|\leq1.(11)✦ Let EE and FF be real inner product spaces, let e:F→Ee:F\to E be an affine isometry, and take g∈Eg\in E and c∈ℝc\in\mathbb R. There is a vector g′∈Fg'\in F satisfying (12): ∥g′∥≤∥g∥,⟨g,e(x)⟩+c=⟨g′,x⟩+(⟨g,e(0)⟩+c)(x∈F).\|g'\|\leq\|g\|,\qquad \langle g,e(x)\rangle+c =\langle g',x\rangle+\bigl(\langle g,e(0)\rangle+c\bigr) \quad(x\in F).(12)✦

In the paper, the result reads: every top nn-simplex σ\sigma of a simplicial complex KK has volume 1/n!1/n! in each tree Euclidean structure, and, for every cc in the geometric realization |K||K|, every affine branch of fc=ρ(c,⋅)f_c=\rho(c,\cdot) on σ\sigma has gradient norm at most one, as does its restriction to any affine subspace of the span of σ\sigma, where ρ\rho is the dual transportation metric normalized by 1/n1/\sqrt n.

What Is Checked

The claims are stated over finite simple graphs and their spanning trees, the Euclidean measures defined by (6), vertex potentials with bounded tree edge differences, and affine isometries between real inner product spaces.