We present results from a series of empirical tests of pseudorandom number generators. The tests cover a broad range of designs due to bit-oriented, efficient test statistics and a testing procedure inwhich we vary the sample size, dimension, and the statistics' resolution within vast bounds. Inversive generation methods pass the tests for a broader range of thest parameters thatn linear generators with equal period legth. The results exemplify how the lattice structure of linear generatorsf can affect a stochastic simulation and suggest the use of inversive generators for cross-checking the results.
展开▼
