OPERATIONAL CONDITIONS FOR RANDOM-NUMBER GENERATION

Ensemble theory is used to describe arbitrary sequences of integers, whether formed by the decimals of a or produced by a roulette or by any other means. Correlation coefficients of any range and order are defined as Fourier transforms of the ensemble weights. Competing definitions of random sequences are considered. Special attention is given to sequences of random numbers needed for Monte Carlo calculations. Different recipes for those sequences lead to correlations that vary in range and order, but the total amount of correlation is the same for all sequences of a given length (without internal periodicities). For maximum-length sequences produced by linear algorithms, most correlation coefficients are zero, but the remaining ones are of absolute value 1. In well-tempered sequences, these complete correlations are of high order or of very long range. General conditions to be obeyed by random-number generators are discussed and a qualitative method for comparing different recipes is given.