An L^1 estimate for half-space discrepancy

TL;DR

This paper constructs point distributions within a cube that achieve near-optimal L^1 discrepancy bounds for half-space intersections, extending previous results by Beck and the first author.

Contribution

It generalizes earlier discrepancy bounds by providing a new distribution of points with improved L^1 discrepancy estimates for half-space intersections.

Findings

Established existence of point distributions with discrepancy bounded by c_d(log N)^d

Extended previous discrepancy results to more general settings

Improved understanding of geometric discrepancy in high dimensions

Abstract

For every unit vector $\sigma\in\Sigma_{d-1}$ and every $r\ge0$, let % % \begin{displaymath} P_{\sigma,r}=[-1,1]^d\cap\{t\in\Rr^d:t\cdot\sigma\le r\} \end{displaymath} % % denote the intersection of the cube $[-1,1]^d$ with a half-space containing the origin $0\in\Rr^d$. We prove that if $N$ is the $d$-th power of an odd integer, then there exists a distribution $\PPP$ of $N$ points in $[-1,1]^d$ such that % % \begin{displaymath} \sup_{r\ge0} \int_{\Sigma_{d-1}}\vert\card(\PPP\cap P_{\sigma,r})-N2^{-d} \vert P_{\sigma,r}\vert\vert\,\dd\sigma \le c_d(\log N)^d, \end{displaymath} % % generalizing an earlier result of Beck and the first author.