Monte Carlo efficiency improvement by multiple sampling of conditioned integration variables

We present a technique that permits to increase the efficiency of multidimensional Monte Carlo algorithms when the sampling of the first, unconditioned random variable consumes much more computational time than the sampling of the remaining, conditioned random variables while its variability contributes only little to the total variance. This is in particular relevant for transport problems in complex and randomly distributed geometries. The proposed technique is based on an new Monte Carlo estimator in which the conditioned random variables are sampled more often than the unconditioned one. A significant contribution of the present Short Note is an automatic procedure for calculating the optimal number of samples of the conditioned random variable per sample of the unconditioned one. The technique is illustrated by a current research example where it permits to increase the efficiency by a factor 100.