Stochastic investigations of pseudo-random number generators (Q2640310)

The author proposes two new tests of the quality of pseudorandom number generators. The first one is based on the distribution of partial sums (it tests some local properties), the second uses a function of the first r sample moments (it tests global properties). The tests are applied to a quite large collection of generators. One generator was developed by the author. The majority of the generators are the linear congruential ones. Some Fibonacci like generators and shuffling generators are also tested. Very good properties (with respect to the mentioned tests) have the generators \((7^ 5X+1)(mod 2^{31}-1)\), 397204094 X(mod 2\({}^{31}-1)\) and the author's generator 1353776763018 X(mod 2365347734339). The Fibonacci generator does not pass the tests. Shuffling almost does not improve the quality of ``good'' generators. Fibonacci shuffling (shuffling table size 100) of Fibonacci generators \(X_{n+1}=X_ n+X_{n-1}(mod 2^{35})\) shows very good properties. The author gives examples of shuffling generators which have wrong properties (e.g. shuffling needs not produce a good generator). The paper contains a lot of numerical results. Formulas and results are taken without proofs from the author's paper [On a distribution encountered in the renewal process based on uniform distribution. Glasn. Mat. (to appear)] and his thesis.