Discrepancy-based error estimates for Quasi-Monte Carlo. III: Error distributions and central limits

Abstract

In Quasi-Monte Carlo integration, the integration error is believed to be generally smaller than in classical Monte Carlo with the same number of integration points. Using an appropriate definition of an ensemble of quasi-randompoint sets, we derive various results on the probability distribution of the integration error, which can be compared to the standard Central Limit theorem for normal stochastic sampling. In many cases, a Gaussian error distribution is obtained.