English

Shallow Packings in Geometry

Computational Geometry 2014-12-18 v1 Discrete Mathematics

Abstract

We refine the bound on the packing number, originally shown by Haussler, for shallow geometric set systems. Specifically, let \V\V be a finite set system defined over an nn-point set XX; we view \V\V as a set of indicator vectors over the nn-dimensional unit cube. A δ\delta-separated set of \V\V is a subcollection \W\W, s.t. the Hamming distance between each pair \uu,\vv\W\uu, \vv \in \W is greater than δ\delta, where δ>0\delta > 0 is an integer parameter. The δ\delta-packing number is then defined as the cardinality of the largest δ\delta-separated subcollection of \V\V. Haussler showed an asymptotically tight bound of Θ((n/δ)d)\Theta((n/\delta)^d) on the δ\delta-packing number if \V\V has VC-dimension (or \emph{primal shatter dimension}) dd. We refine this bound for the scenario where, for any subset, XXX' \subseteq X of size mnm \le n and for any parameter 1km1 \le k \le m, the number of vectors of length at most kk in the restriction of \V\V to XX' is only O(md1kdd1)O(m^{d_1} k^{d-d_1}), for a fixed integer d>0d > 0 and a real parameter 1d1d1 \le d_1 \le d (this generalizes the standard notion of \emph{bounded primal shatter dimension} when d1=dd_1 = d). In this case when \V\V is "kk-shallow" (all vector lengths are at most kk), we show that its δ\delta-packing number is O(nd1kdd1/δd)O(n^{d_1} k^{d-d_1}/\delta^d), matching Haussler's bound for the special cases where d1=dd_1=d or k=nk=n. As an immediate consequence we conclude that set systems of halfspaces, balls, and parallel slabs defined over nn points in dd-space admit better packing numbers when kk is smaller than nn. Last but not least, we describe applications to (i) spanning trees of low total crossing number, and (ii) geometric discrepancy, based on previous work by the author.

Keywords

Cite

@article{arxiv.1412.5215,
  title  = {Shallow Packings in Geometry},
  author = {Esther Ezra},
  journal= {arXiv preprint arXiv:1412.5215},
  year   = {2014}
}
R2 v1 2026-06-22T07:34:15.518Z