Complexity of a Root Clustering Algorithm

TL;DR

This paper analyzes the complexity of an unconditional root clustering algorithm for holomorphic functions, revealing exponential time behavior for simple functions and providing bounds on bit-precision used.

Contribution

It offers the first complexity bounds for an unconditional root clustering algorithm applied to holomorphic functions, extending previous polynomial-specific results.

Findings

The algorithm takes exponential time for simple transcendental functions.

A bound on the bit-precision used by the algorithm is derived.

The analysis introduces new geometric parameters for better understanding of the algorithm's behavior.

Abstract

Approximating the roots of a holomorphic function in an input box is a fundamental problem in many domains. Most algorithms in the literature for solving this problem are conditional, i.e., they make some simplifying assumptions, such as, all the roots are simple or there are no roots on the boundary of the input box, or the underlying machine model is Real RAM. Root clustering is a generalization of the root approximation problem that allows for errors in the computation and makes no assumption on the multiplicity of the roots. An unconditional algorithm for computing a root clustering of a holomorphic function was given by Yap, Sagraloff and Sharma in 2013. They proposed a subdivision based algorithm using effective predicates based on Pellet's test while avoiding any comparison with zeros (using soft zero comparisons instead). In this paper, we analyze the running time of their algorithm. We use the continuous amortization framework to derive an upper bound on the size of the subdivision tree. We specialize this bound to the case of polynomials and some simple transcendental functions such as exponential and trigonometric sine. We show that the algorithm takes exponential time even for these simple functions, unlike the case of polynomials. We also derive a bound on the bit-precision used by the algorithm. To the best of our knowledge, this is the first such result for holomorphic functions. We introduce new geometric parameters, such as the relative growth of the function on the input box, for analyzing the algorithm. Thus, our estimates naturally generalize the known results, i.e., for the case of polynomials.