Asymptotic behavior of average \(L_p\)-discrepancies (Q432762): Difference between revisions

Let \(p>0\), \(d\in \mathbb{N}\). Let \((\Omega_d, \mu_d)\) be a probability space and \(\{B(\mathbf{x}): \mathbf{x}\in \Omega_d\}\subset 2^{[0,1]^d}\) the allowed sets. The \(L_p\)-\(B\)-discrepancy is defined by \[ \text{disc}_{p}^{B}\left(\mathbf{t}_1,\ldots,\mathbf{t}_n\right)= \left(\int_{\Omega_d}\left|\lambda^d(B(\mathbf{x}))-\frac{1}{n}\sum_{i=1}^{n}\chi_{B(\mathbf{x})}(\mathbf{t}_i)\right|^p \text{d}\mu_d(\mathbf{x})\right)^{1/p}, \] where \(\lambda^d(B(\mathbf{x}))\) is the \(d\)-dimensional Lebesgue measure of \(B(\mathbf{x})\), and \(\chi_{B(\mathbf{x})}\) is the characteristic function on \(B(\mathbf{x})\). Define the average \(L_p\)-\(B\)-discrepancy by \[ \text{av}_{p}^{B}(n,d)=\left(\int_{[0,1]^{nd}}\text{disc}_{p}^{B} \left(\mathbf{t}_1,\ldots,\mathbf{t}_n\right)^p\text{d}\mathbf{t}\right)^{1/p}. \] This paper proves that \[ \lim_{n\rightarrow\infty}n^{p/2}\text{av}_{p}^{B}(n,d)^{p}=\frac{2^{p/2}}{\pi^{1/2}} \Gamma\left(\frac{1+p}{2}\right)\int_{\Omega_d}\left[\lambda^d(B(\mathbf{x}))\left(1-\lambda^d(B(\mathbf{x}))\right)\right]^{p/2} \text{d}\mu_d(\mathbf{x}). \]