Comment on "Algorithm for normal random numbers".

Abstract

We have performed some simple statistical tests on the recently proposed algorithm for generating normally distributed random numbers and we find that it has some serious flaws with regard to practical application. In particular, we find that there is a statistically significant difference between the sample mean and that expected when using the original authors' recommended parameters of the algorithm. Only by significantly increasing these parameters, at the cost of considerable extra computational effort, do we get satisfactory statistics.