Fault-tolerant sampling designs for convex Lipschitz functions
Year: 2026
Type: preprint
Abstract: We find the optimal locations for sampling a univariate convex Lipschitz function when some measurements may be erased or arbitrarily corrupted. The minimax uniform error is explicit for every sample size and every prescribed minimum spacing, both when the extreme sites are free and when both endpoints must be sampled. We construct optimizers and determine exactly when distinct instruments can attain the zero-spacing infimum. For example, with six instruments and one hidden error, free placement has infimum 1/6 and requires limiting clusters, whereas a distinct endpoint design attains 1/5. For fixed observations with site-dependent error bounds we also construct a minimax reconstruction that belongs to the same convex data fibre.
Author: Xu, Zihuan
Institution: Capital Normal University
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7236697244
