Accelerating true orbit pseudorandom number generation using Newton's method

TL;DR

This paper introduces an efficient Newton's method-based approach for generating high-quality pseudorandom binary sequences from irrational algebraic numbers, significantly improving speed and passing rigorous statistical tests.

Contribution

It presents a novel Newton's method technique for exact binary expansion computation of quadratic algebraic integers, enhancing the efficiency of true orbit pseudorandom number generation.

Findings

Method accelerates sequence generation compared to previous approaches

Generated sequences pass all TestU01 statistical tests

Complexity is comparable to integer multiplication

Abstract

The binary expansions of irrational algebraic numbers can serve as high-quality pseudorandom binary sequences. This study presents an efficient method for computing the exact binary expansions of real quadratic algebraic integers using Newton's method. To this end, we clarify conditions under which the first $N$ bits of the binary expansion of an irrational number match those of its upper rational approximation. Furthermore, we establish that the worst-case time complexity of generating a sequence of length $N$ with the proposed method is equivalent to the complexity of multiplying two $N$-bit integers, showing its efficiency compared to a previously proposed true orbit generator. We report the results of numerical experiments on computation time and memory usage, highlighting in particular that the proposed method successfully accelerates true orbit pseudorandom number generation. We also confirm that a generated pseudorandom sequence successfully passes all the statistical tests included in RabbitFile of TestU01.