Adversarial Wronskians: A proof of the Kung--Traub conjecture

Abstract

We prove the Kung--Traub conjecture: a deterministic stationary root-finding method without memory, using at most $n$ real scalar function or derivative evaluations per iteration, cannot have local order greater than $2^{n-1}$ on every real analytic function with a simple zero. For $n\geq2$, the bound is attained by the methods of Kung and Traub. The local estimate is required at every sufficiently close starting point, with constants that may depend on the function. Evaluation points and derivative orders may be chosen adaptively, and the rules for choosing them and processing the data are unrestricted. The proof uses a localisation property of polynomial Wronskians to obtain a lower bound at each fixed scale. A diagonal construction then gives one entire function, with derivative bounded above and below on the real line, for which every larger order fails. We also obtain a sharper bound when evaluations are made in prescribed groups, and prove the corresponding result for complex scalar evaluations. The results of this paper are also formally verified in Lean.