Randomly shifted lattice rules with the optimal rate of convergence for unbounded integrands (Q964919)

High-dimensional integrals of the form \[ \int_{\mathbb{R}^d} f(\mathbf{x}) \rho(\mathbf{x}) d(\mathbf{x}) \] are considered, where \(\rho(\mathbf{x})\) is a multivariate probability density function being the product of univariate functions. It is assumed that \(f\) belongs to a weighted tensor-product reproducing kernel Hilbert space of functions whose mixed first derivatives, when multiplied by a weight function \(\Psi\), have bounded \(L_2\)-norms. The problem of quasi-Monte Carlo estimation of such integrals is considered after mapping the integral to the unit cube \((0,1)^d\) by changing the integration variable for each component of the vector \(\mathbf{x}\) to the cummulative probability distribution, corresponding to that variable. It is related to a well known difficulty that this transformation often results in integrands which are either unbounded or have very large derivatives near the boundary. The authors recently have addressed this problem by assuming that \(f\) belongs to certain function spaces and constructing randomly shifted lattice rules to achieve a worst case error of order \(\mathcal{O} \left(n^{-1/2} \right)\). In the present work, it is shown how a rate of convergence close to the optimal order of \(\mathcal{O} \left(n^{-1} \right)\) can be achieved, by using a more clever proof technique together with more restrictive assumptions. The results are illustrated for a Poisson time series maximum likelihood integral.