On the convergence of quasi-random sampling/importance resampling (Q622171)

A sampling/importance resampling (SIR) algorithm is considered for generation of quasi-random sample \(X=(x_1,\dots,x_N)\) from a density \(h\) on \(\mathbb{R}^s\) known up to a multiplicative constant. This sample is used to approximate \(\int f(x)h(x)dx\) by \({1\over N}\sum_{n=1}^N f(x_n)\) for a function \(f\) with bounded variation on \(\mathbb{R}^s\) in the sense of Hardy and Krause. Two versions of sampling algorithm are considered. In the first one a deterministic quasi random sampling is used on the first stage with a random resampling on the second stage. The second algorithm is totally deterministic. The authors derive estimates for the rate of convergence of the empirical CDF of \(X\) to the target CDF in the uniform norm. Numerical examples are presented.