Bounds for the average \(L^p\)-extreme and the \(L^\infty\)-extreme discrepancy (Q2571297)

The extreme or unanchored discrepancy is the geometric discrepancy of point configurations in the \(d\)-dimensional unit cube with respect to the set system of all axis-parallel boxes. This kind of discrepancy appears more natural than the intensively studied star-discrepancy and has, like the latter one, applications to high-dimensional numerical integration. The paper provides upper bounds for the average \(L^p\)-extreme discrepancy for \(2\leq p <\infty\). These bounds are used to prove upper bounds for the inverse of \(L^\infty\)-extreme discrepancy with explicit constants and an optimal dependence on the dimension \(d\). The bounds given in the paper extend known results for the star-discrepancy of \textit{S. Heinrich, E. Novak, G. W. Wasilkowski} and \textit{H. Woźniakowski} [Acta Arith. 96, 279--302 (2001; Zbl 0972.11065)] to the extreme discrepancy. In contrast to the methods of Heinrich et al., the author uses the probabilistic concept of symmetrization and avoids the elaborate analysis of weighted products of Stirling numbers. This leads to improved and simplified results.