Residual iteration and accurate polynomial evaluation for shape-interrogation applications

Surface interrogation and intersection depend crucially on good root-finding algorithms, which in turn depend on accurate polynomial evaluation. Conventional algorithms for evaluation typically encounter difficulties near multiple roots, or roots that are very close, and this may lead to gross errors in the geometric computation, or even catastrophic failure. In this paper we study the cost and accuracy of several approaches to polynomial evaluation, explaining the reasons for non-convergence of certain methods, and supporting our subsequent conclusions with the results of benchmarking experiments.