Deterministic and probabilistic discrepancies (Q1042558)

In the paper under review, the authors consider the discrepancy of point sets with respect to balls in the \(d\)-dimensional unit torus \(\mathbb{T}^d\). More precisely, they consider the standard lattice \[ L_M = \left\{ \left(\frac{r_1}{M}, \dots, \frac{r_d}{M} \right):~r_1, \dots, r_d \in \{0, 1, \dots, M-1\right\}, \] and probability measures \(d\mu\) translated by the points of \(p \in L_M\) to \(d\mu_p\). In the first part of the paper, the authors derive an explicit formula for the average \((L^2-)\)discrepancy \(D_{d\mu}^2\) (with respect to a measure \(d\mu\)). Then they compare the deterministic case \(d\mu=\delta_0\) (\(\delta_{0}\) being the Dirac measure centered at the origin) to the random case \(d\mu=d\lambda\), where \(\lambda\) denotes the normalized Lebesgue measure on the ``small cube'' \([-1/(2M),1/(2M)]^d\). Interestingly, the results depend on whether \(d \equiv 1 \bmod 4\) or \(d \not\equiv 1 \bmod 4\), a fact which has already been observed before see \textit{S. V. Konyagin, M. M. Skriganov} and \textit{A. V. Sobolev} [``On a lattice point problem arising in the spectral analysis of periodic operators'', Mathematika 50, No. 1-2, 87--98 (2003; Zbl 1101.11040)]. For the proofs, the authors use different (probabilistic and deterministic) methods from their earlier work.