Searching for good two-term multiple recursive random number generators using a backpropagation algorithm

Abstract

We describe an approximation algorithm for the problem of finding the maximum spectral value in a full period two-term multiple recursive generator (MRG). This algorithm is based on the principle of the backpropagation algorithm and random search. A welldesigned backpropagation algorithm is developed by an extensive set of computational comparisons of various parameters of the algorithm. Using this algorithm, a random search is conducted. Computational experiences show that our algorithm can find larger spectral value and shorter running time than forward/backward method for the full period twoterm MRG of order three. An application to the full period two-term MRG of order four is also discussed.