MONTE CARLO TECHNIQUES, widely used in computer simulations of physical systems, entail the wholesale generation of random numbers. A new study by scientists at the University of Georgia (Alan Ferrenberg, 404-542-8460) shows that even the most advanced random-number generators are biased under certain circumstances (A.M. Ferrenberg et al., 7 Dec. Physical Review Letters). Using one state-of-the-art program, the Marsaglia-Zeman random-number generator, Ferrenberg discovered that a simulated performance of the two-dimensional Ising model (which models the behavior of a plane of neighboring spins) did not agree with the results when calculated exactly by mathematical methods. He traced the discrepancy to the random- number generator. Other generators tried had differing faults. (Science News, 19 & 26 Dec.)
Can someone get the paper(s) and/or talk to the researcher? Does he have any programs he can throw into the pot for generating or testing random numbers? John