OctoInformBeta

OpenAI preprint, September 2026 · §2. Geometric and horizontal preparations

Fibers, Gaussian comparison and horizontal holes: In horizontal dimension mm, with standard Gaussian comparison and the explicit dimension parameter ηm\eta_m

Coordinate sections and horizontal holes.

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

Let m,Dm,D be nonnegative integers, and split the coordinates of ℝm+D\mathbb R^{m+D} into a horizontal variable x∈ℝmx\in\mathbb R^m and a transverse variable y∈ℝDy\in\mathbb R^D. The horizontal geometry of a set K⊆ℝm+DK\subseteq\mathbb R^{m+D} is considered at fixed yy, with a full lattice in ℝm\mathbb R^m.

For a nonnegative integer rr, the real coordinate space ℝr\mathbb R^r consists of real vectors with rr coordinates.✦ The coordinate torus 𝕋r\mathbb T^r in dimension rr is the product of rr copies of ℝ/ℤ\mathbb R/\mathbb Z in (1). 𝕋r=(ℝ/ℤ)r.\mathbb T^r=(\mathbb R/\mathbb Z)^r.(1)✦ Let μr\mu_r be Haar measure on 𝕋r\mathbb T^r, normalized to have total mass one.✦

The Definition

Definition 1. For a set K⊆ℝm+DK\subseteq\mathbb R^{m+D} and a transverse vector y∈ℝDy\in\mathbb R^D, the coordinate fiber Ky⊆ℝmK_y\subseteq\mathbb R^m is the set in (2). Ky={x∈ℝm:(x,y)∈K}.K_y=\{x\in\mathbb R^m:(x,y)\in K\}.(2)✦

For a DD-dimensional real normed space FF, the Gaussian comparison function γF:F→ℝ\gamma_F:F\to\mathbb R is given by (3). On Euclidean space, write γD\gamma_D for the standard Gaussian density, with the coordinate product in the same equation. γF(y)=(2π)−D/2exp⁡(−∥y∥22),γD(y)=(2π)−D/2exp⁡(−∥y∥222)=∏i=1D(2π)−1/2exp⁡(−yi2/2).\begin{aligned} \gamma_F(y)&=(2\pi)^{-D/2}\exp\!\left(-\frac{\|y\|^2}{2}\right),\\ \gamma_D(y)&=(2\pi)^{-D/2}\exp\!\left(-\frac{\|y\|_2^2}{2}\right) =\prod_{i=1}^{D}(2\pi)^{-1/2}\exp(-y_i^2/2). \end{aligned}(3)✦

A full lattice Λ⊆ℝm\Lambda\subseteq\mathbb R^m is an additive subgroup satisfying (4). Λ is discrete,span⁡ℝΛ=ℝm.\Lambda\text{ is discrete},\qquad \operatorname{span}_{\mathbb R}\Lambda=\mathbb R^m.(4)✦

Fix a lattice basis b1,…,bmb_1,\ldots,b_m of Λ\Lambda. The lattice quotient map πΛ:ℝm→𝕋m\pi_\Lambda:\mathbb R^m\to\mathbb T^m is defined by its basis coordinates in (5). πΛ(∑i=1mtibi)=(ti+ℤ)i=1m.\pi_\Lambda\!\left(\sum_{i=1}^{m}t_i b_i\right) =(t_i+\mathbb Z)_{i=1}^{m}.(5)✦

For a set J⊆ℝmJ\subseteq\mathbb R^m and a full lattice Λ⊆ℝm\Lambda\subseteq\mathbb R^m, the horizontal hole hΛ(J)h_\Lambda(J) of JJ with respect to Λ\Lambda is the uncovered torus measure in (6). hΛ(J)=μm(𝕋m\πΛ(J)).h_\Lambda(J)=\mu_m\!\left(\mathbb T^m\setminus\pi_\Lambda(J)\right).(6)✦

The dimension parameter ηm∈ℝ\eta_m\in\mathbb R is the explicit mean-hole cutoff in (7), where log\log denotes the natural logarithm. ηm={0,m=0,m4log⁡(2716)−3log⁡m,m≥1.\eta_m= \begin{cases} 0,&m=0,\\[2pt] \displaystyle\frac{m}{4}\log\!\left(\frac{27}{16}\right)-3\log m, &m\geq1. \end{cases}(7)✦

What Is Checked

The definitions apply to arbitrary subsets KK and JJ in nonnegative coordinate dimensions and to discrete full lattices; the Gaussian comparison function is stated on any finite-dimensional real normed space. The lattice basis is a real basis of the ambient space, and the lattice projection in (5) is a continuous additive map.