OctoInformBeta

OpenAI preprint, September 2026 · §1. Introduction

§1 (setting): the universal coordinate dimension: The universal coordinate dimension dp(n,D)d_p(n,D) and the exponent γ(p)\gamma(p)

Coordinate dimension for finite LpL_p configurations.

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

For nn distinct points in a real LpL_p space, the dimension reduction question asks how few real coordinates can retain all pairwise distances within a prescribed factor DD, after a common change of scale. The representing map may be nonlinear.

The Definition

Definition 1. For p∈ℝp\in\mathbb R and d∈ℕd\in\mathbb N, the coordinate distance ρp:ℝd×ℝd→ℝ\rho_p:\mathbb R^d\times\mathbb R^d\to\mathbb R is the function defined by (1): ρp(u,v)=(∑a=1d|ua−va|p)1/p,u,v∈ℝd.\rho_p(u,v)=\left(\sum_{a=1}^{d}|u_a-v_a|^p\right)^{1/p}, \qquad u,v\in\mathbb R^d.(1)✦

Definition 2. Let p,D∈ℝp,D\in\mathbb R and n∈ℕn\in\mathbb N. A dimension d∈ℕd\in\mathbb N is admissible for (p,n,D)(p,n,D) if, for every measure space (Ω,μ)(\Omega,\mu) and every choice of distinct points x1,…,xnx_1,\ldots,x_n in the real Lebesgue space with exponent max⁡{p,0}\max\{p,0\}, there are points y1,…,yn∈ℝdy_1,\ldots,y_n\in\mathbb R^d and a scale s>0s>0 satisfying (2): s∥xi−xj∥Lp(μ)≤ρp(yi,yj)≤Ds∥xi−xj∥Lp(μ),1≤i,j≤n.s\|x_i-x_j\|_{L_p(\mu)} \leq \rho_p(y_i,y_j) \leq Ds\|x_i-x_j\|_{L_p(\mu)}, \qquad 1\leq i,j\leq n.(2)✦

Definition 3. Let 𝒜p(n,D)⊆ℕ\mathcal A_p(n,D)\subseteq\mathbb N be the set of dimensions admissible for (p,n,D)(p,n,D). The universal coordinate dimension dp(n,D)∈ℕd_p(n,D)\in\mathbb N is the natural number defined by (3): dp(n,D)={min⁡𝒜p(n,D),𝒜p(n,D)≠⌀,0,𝒜p(n,D)=⌀.d_p(n,D)= \begin{cases} \min\mathcal A_p(n,D), & \mathcal A_p(n,D)\ne\varnothing,\\ 0, & \mathcal A_p(n,D)=\varnothing. \end{cases}(3)✦

Definition 4. The exponent γ:ℝ→ℝ\gamma:\mathbb R\to\mathbb R is the function defined by (4): γ(p)={2−p,p<2,1−2p,p≥2.\gamma(p)= \begin{cases} 2-p, & p<2,\\ 1-\dfrac{2}{p}, & p\geq2. \end{cases}(4)✦

What Is Checked

The admissibility condition in (2) ranges over arbitrary measure spaces and all distinct nn-point configurations in their real LpL_p spaces, with target points in ℝd\mathbb R^d.