English

On the period of the linear congruential and power generators

Number Theory 2015-06-26 v2

Abstract

We consider the periods of the linear congruential and the power generators modulo nn and, for fixed choices of initial parameters, give lower bounds that hold for ``most'' nn when nn ranges over three different sets: the set of primes, the set of products of two primes (of similar size), and the set of all integers. For most nn in these sets, the period is at least n1/2+ϵ(n)n^{1/2+\epsilon(n)} for any monotone function ϵ(n)\epsilon(n) tending to zero as nn tends to infinity. Assuming the Generalized Riemann Hypothesis, for most nn in these sets the period is greater than n1ϵn^{1-\epsilon} for any ϵ>0\epsilon >0. Moreover, the period is unconditionally greater than n1/2+δn^{1/2+\delta}, for some fixed δ>0\delta>0, for a positive proportion of nn in the above mentioned sets. These bounds are related to lower bounds on the multiplicative order of an integer ee modulo p1p-1, modulo λ(pl)\lambda(pl), and modulo λ(m)\lambda(m) where p,lp,l range over the primes, mm ranges over the integers, and where λ(n)\lambda(n) is the order of the largest cyclic subgroup of (Z/nZ)×(\Z/n\Z)^\times.

Keywords

Cite

@article{arxiv.math/0405120,
  title  = {On the period of the linear congruential and power generators},
  author = {P. Kurlberg and C. Pomerance},
  journal= {arXiv preprint arXiv:math/0405120},
  year   = {2015}
}

Comments

20 pages. One of the quoted results (Theorem 23 in the previous version) is stated for any unbounded monotone function psi(x), but it appears that the proof only supports the case when psi(x) is increasing rather slowly. As a workaround, we provide a modified version of Theorem 23, and change the argument in the proof of Theorem 27 (Theorem 25 in the previous version)