Quasi-Monte Carlo algorithms for unbounded, weighted integration problems (Q1888375)

The authors investigate quasi-Monte-Carlo (QMC) methods for the integration of singular functions with respect to general measures. The functions are allowed to be singular at the left boundary of a multidimensional interval. Convergence of the QMC integration is proved under the condition that the improper integral exists and that the QMC points posses a sufficiently small discrepancy compared to the variation of the integrand on the subinterval with left boundary defined by the left most QMC points. These considerations extend similar results for the special case of Lebesgue measure on the multidimensional unit cube. Since the standard Hlawka-Mück scheme for the construction of low-discrepancy sequences with given density may produce points which fall on the left boundary, it is not directly suitable for the integration of singular functions. The authors propose a modification of the standard scheme which moves the left most points to some minimal distance in such a way that the low-discrepancy is not lost. In this way, QMC points suitable for singular integrands can be generated.