OctoInformBeta

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

Lemma 2.2: Mean estimates and explicit numerical conditions for finite families of convex sets

Simultaneous small holes at prescribed covolume.

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 a finite family of sets in Euclidean space, the question is whether a single horizontal lattice of prescribed covolume makes all their projections occupy most of the lattice torus. For a nonnegative integer mm, the coordinate space ℝm\mathbb{R}^m is the vector space of real-valued functions on {0,…,m−1}\{0,\ldots,m-1\}.✦ A full lattice Λ\Lambda in ℝm\mathbb{R}^m is a discrete integer submodule that spans ℝm\mathbb{R}^m; its determinant det⁡Λ\det\Lambda is its Haar covolume.✦

The torus 𝕋m\mathbb{T}^m is the product (ℝ/ℤ)m(\mathbb{R}/\mathbb{Z})^m.✦ A lattice basis (bi)i=0m−1(b_i)_{i=0}^{m-1} of a full lattice Λ\Lambda is the real vector-space basis obtained from a chosen integer basis of Λ\Lambda by extending scalars to ℝ\mathbb{R}.✦ The lattice quotient map πΛ:ℝm→𝕋m\pi_\Lambda\colon\mathbb{R}^m\to\mathbb{T}^m is given in these basis coordinates by (1): πΛ(∑i=0m−1tibi)=(ti+ℤ)i=0m−1.\pi_\Lambda\!\left(\sum_{i=0}^{m-1}t_i b_i\right) =(t_i+\mathbb{Z})_{i=0}^{m-1}.(1)✦

Let νm\nu_m be translation-invariant Haar measure on 𝕋m\mathbb{T}^m, normalized to have total mass one.✦ The hole hΛ(J)h_\Lambda(J) of a set J⊂ℝmJ\subset\mathbb{R}^m is the measure of the complement of its projected image, as in (2): hΛ(J)=νm(𝕋m\πΛ(J)).h_\Lambda(J)=\nu_m\bigl(\mathbb{T}^m\setminus\pi_\Lambda(J)\bigr).(2)✦

The real-valued volume vm(J)v_m(J) of a set J⊂ℝmJ\subset\mathbb{R}^m is the quantity in (3); for a family (Ja)a∈S(J_a)_{a\in S} indexed by a finite set SS and a target covolume Dh>0D_h>0, the normalized volume uau_a is the ratio in the same equation: vm(J)={vol⁡m(J),vol⁡m(J)<∞,0,vol⁡m(J)=∞,ua=vm(Ja)Dh.v_m(J)= \begin{cases} \operatorname{vol}_m(J),&\operatorname{vol}_m(J)<\infty,\\ 0,&\operatorname{vol}_m(J)=\infty, \end{cases} \qquad u_a=\frac{v_m(J_a)}{D_h}.(3) For each positive integer dd, the dimension threshold ηd\eta_d is the real number given by (4), with natural logarithms: ηd=d4log⁡2716−3log⁡d.\eta_d=\frac{d}{4}\log\frac{27}{16}-3\log d.(4)

The Claim

Theorem 1. Let SS be a finite index set, let Ja⊂ℝmJ_a\subset\mathbb{R}^m for a∈Sa\in S, and fix Dh>0D_h>0, with uau_a as in (3).

Mean holes. Let (X,μ)(X,\mu) be a probability space and let (Λx)x∈X(\Lambda_x)_{x\in X} be full lattices with det⁡Λx=Dh\det\Lambda_x=D_h for every xx. Let CR,η,τ,NC_R,\eta,\tau,N be real numbers with CR≥0C_R\geq0. Suppose that each function x↦hΛx(Ja)x\mapsto h_{\Lambda_x}(J_a) is measurable and satisfies ∫XhΛx(Ja)dμ(x)≤e−ua+CRe−η\int_X h_{\Lambda_x}(J_a)\,d\mu(x)\leq e^{-u_a}+C_Re^{-\eta}. Assume τ≤ua≤η\tau\leq u_a\leq\eta for every a∈Sa\in S, |S|≤N|S|\leq N, and (1+CR)Ne−τ/2<1(1+C_R)N e^{-\tau/2}<1. There is a full lattice Λh\Lambda_h satisfying (5): det⁡Λh=Dh,hΛh(Ja)≤e−ua/2(a∈S).\det\Lambda_h=D_h,\qquad h_{\Lambda_h}(J_a)\leq e^{-u_a/2}\quad(a\in S).(5)✦

Summed criterion. Suppose that m≥270m\geq270, that each JaJ_a is compact and convex, that ua≤ηmu_a\leq\eta_m for every a∈Sa\in S, and that 302∑a∈Se−ua/2≤1302\sum_{a\in S}e^{-u_a/2}\leq1. There is a full lattice Λh\Lambda_h satisfying (6): det⁡Λh=Dh,hΛh(Ja)≤e−ua/2(a∈S).\det\Lambda_h=D_h,\qquad h_{\Lambda_h}(J_a)\leq e^{-u_a/2}\quad(a\in S).(6)✦

Uniform lower bound. Suppose that m≥270m\geq270 and each JaJ_a is compact and convex. Let τ,N∈ℝ\tau,N\in\mathbb{R} satisfy τ≤ua≤ηm\tau\leq u_a\leq\eta_m for every a∈Sa\in S, |S|≤N|S|\leq N, and 302Ne−τ/2<1302N e^{-\tau/2}<1. There is a full lattice Λh\Lambda_h satisfying (7): det⁡Λh=Dh,hΛh(Ja)≤e−ua/2(a∈S).\det\Lambda_h=D_h,\qquad h_{\Lambda_h}(J_a)\leq e^{-u_a/2}\quad(a\in S).(7)✦

In the paper, the result reads: for convex bodies J1,…,JN⊂ℝmJ_1,\ldots,J_N\subset\mathbb{R}^m, Dh>0D_h>0 and ua=vol⁡m(Ja)/Dhu_a=\operatorname{vol}_m(J_a)/D_h, if max⁡aua≤ηm\max_a u_a\leq\eta_m and (1+CR)∑ae−ua/2<1(1+C_R)\sum_a e^{-u_a/2}<1, where CRC_R is the absolute constant in the stated mean-hole estimate, there is a lattice Λh\Lambda_h of determinant DhD_h with hΛh(Ja)≤e−ua/2h_{\Lambda_h}(J_a)\leq e^{-u_a/2} for every aa; in particular, this holds for all sufficiently large mm if min⁡aua≥m\min_a u_a\geq\sqrt m, max⁡aua=o(m)\max_a u_a=o(m) and log⁡N=o(m)\log N=o(\sqrt m), with the threshold depending only on these numerical bounds.

What Is Checked

The mean statement (5) is over arbitrary sets and probability ensembles with measurable hole functions, whereas (6) and (7) are over compact convex sets in dimension at least 270270. The lattice quotient map πΛ\pi_\Lambda is continuous and satisfies (8) for all x,y∈ℝmx,y\in\mathbb{R}^m: πΛ(0)=0,πΛ(x+y)=πΛ(x)+πΛ(y).\pi_\Lambda(0)=0,\qquad \pi_\Lambda(x+y)=\pi_\Lambda(x)+\pi_\Lambda(y).(8)