Critical-Grid Debiasing and Transport Certificates for Supercritical Minimax Wasserstein Distance Estimation

Read full text Summarise with VIRAT
Authors: Максим Чурилов
Cited by: 0
References: 16

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Let P, Q ∈ P([0, 1]d) be observed through two independent samples of size N , and consider minimax estimation of Wp(P, Q) in the supercritical regime d > 2p. The empirical plug-in estimator gives the scale N −1/d, while the Niles–Weed– Rigollet lower-bound mechanism gives the smaller candidate scale ηN = (N log N )−1/d. The manuscript develops critical-grid debiasing, multiscale polynomial transport estimation, transport certificates, and finite-LP curvature methods for the sharp law. The unrestricted upper bound is reduced, with constants and no loss of scale, to an adaptive finite Kantorovich linear-program value on a Euclidean grid with ≍ N log N atoms. The target law is proved on a large family of critical subclasses retaining the same N log N -alphabet difficulty as the lower-bound construction. The first positive engine is an exact rooted total-variation skeleton principle. It converts the Wasserstein value into finite weighted sums of large-alphabet L1 distances, where the effective N log N gain from functional estimation is available. This yields exact minimax laws on paired Euclidean grids, finite-band and packed direct sums, dyadic pair-isolation models, critical laminar hierarchies, hierarchical tree classes, sparse-shortcut graph classes, continuum blob lifts, partition lifts, full-support paired cores, contiguous split shells, martingale/Haar shells, smooth and real-analytic lower cores, and local block models.

Fields of Study & Subjects

Geometric Analysis and Curvature Flows Random Matrices and Applications Point processes and geometric inequalities

Publication details

Publication date
25 April 2026
Language
EN
Document type
Preprint