The MIXMAX random number generator

Abstract: In this note, we give a practical solution to the problem of determining the maximal period of matrix generators of pseudo-random numbers which are based on an integer-valued unimodular matrix of size NxN known as MIXMAX and arithmetic defined on a Galois field GF[p] with large prime modulus p. The existing theory of Galois finite fields is adapted to the present case, and necessary and sufficient condition to attain the maximum period is formulated. Three efficient algorithms are presented. First, allowing to compute the multiplication by the MIXMAX matrix with O(N) operations. Second, to recursively compute the characteristic polynomial with O(N^2) operations, and third, to apply skips of large number of steps S to the sequence in O(N^2 log(S)) operations. It is demonstrated that the dynamical properties of this generator dramatically improve with the size of the matrix N, as compared to the classes of generators based on sparse matrices and/or sparse characteristic polynomials. Finally, we present the implementation details of the generator and the results of rigorous statistical testing.