English

Stronger Bounds for Weak Epsilon-Nets in Higher Dimensions

Computational Geometry 2023-12-27 v2 Combinatorics

Abstract

Given a finite point set PP in Rd{\mathbb R}^d, and ϵ>0\epsilon>0 we say that NRdN\subseteq{ \mathbb R}^d is a weak ϵ\epsilon-net if it pierces every convex set KK with KPϵP|K\cap P|\geq \epsilon |P|. We show that for any finite point set in dimension d3d\geq 3, and any ϵ>0\epsilon>0, one can construct a weak ϵ\epsilon-net whose cardinality is O(1ϵ2.558)\displaystyle O^*\left(\frac{1}{\epsilon^{2.558}}\right) in dimension d=3d=3, and o(1ϵd1/2)\displaystyle o\left(\frac{1}{\epsilon^{d-1/2}}\right) in all dimensions d4d\geq 4. To be precise, our weak ϵ\epsilon-net has cardinality O(1ϵαd+γ)\displaystyle O\left(\frac{1}{\epsilon^{\alpha_d+\gamma}}\right) for any γ>0\gamma>0, with αd={2.558if d=33.48if d=4(d+d22d)/2if d5.} \alpha_d= \left\{ \begin{array}{l} 2.558 & \text{if} \ d=3 \\3.48 & \text{if} \ d=4 \\\left(d+\sqrt{d^2-2d}\right)/2 & \text{if} \ d\geq 5. \end{array}\right\} This is the first significant improvement of the bound of O~(1ϵd)\displaystyle \tilde{O}\left(\frac{1}{\epsilon^d}\right) that was obtained in 1993 by Chazelle, Edelsbrunner, Grigni, Guibas, Sharir, and Welzl for general point sets in dimension d3d\geq 3.

Cite

@article{arxiv.2104.12654,
  title  = {Stronger Bounds for Weak Epsilon-Nets in Higher Dimensions},
  author = {Natan Rubin},
  journal= {arXiv preprint arXiv:2104.12654},
  year   = {2023}
}

Comments

Preliminary version accepted to STOC 2021. The exponent is corrected in dimension 3, and slightly improved in all dimensions $d\geq 4$. Submitted to a journal

R2 v1 2026-06-24T01:31:45.169Z