Paper 2012/294

Two grumpy giants and a baby

Daniel J. Bernstein and Tanja Lange

Abstract

Pollard's rho algorithm, along with parallelized, vectorized, and negating variants, is the standard method to compute discrete logarithms in generic prime-order groups. This paper presents two reasons that Pollard's rho algorithm is farther from optimality than generally believed. First, ``higher-degree local anti-collisions'' make the rho walk less random than the predictions made by the conventional Brent--Pollard heuristic. Second, even a truly random walk is suboptimal, because it suffers from ``global anti-collisions'' that can at least partially be avoided. For example, after (1.5+o(1))\sqrt(l) additions in a group of order l (without fast negation), the baby-step-giant-step method has probability 0.5625+o(1) of finding a uniform random discrete logarithm; a truly random walk would have probability 0.6753\ldots+o(1); and this paper's new two-grumpy-giants-and-a-baby method has probability 0.71875+o(1).