English

On the optimization of discrepancy measures

Numerical Analysis 2025-08-08 v1 Numerical Analysis Optimization and Control

Abstract

Points in the unit cube with low discrepancy can be constructed using algebra or, more recently, by direct computational optimization of a criterion. The usual LL_\infty star discrepancy is a poor criterion for this because it is computationally expensive and lacks differentiability. Its usual replacement, the L2L_2 star discrepancy, is smooth but exhibits other pathologies shown by J. Matou\v{s}ek. In an attempt to address these problems, we introduce the \textit{average squared discrepancy} which averages over 2d2^d versions of the L2L_2 star discrepancy anchored in the different vertices of [0,1]d[0,1]^d. Not only can this criterion be computed in O(dn2)O(dn^2) time, like the L2L_2 star discrepancy, but also we show that it is equivalent to a weighted symmetric L2L_2 criterion of Hickernell's by a constant factor. We compare this criterion with a wide range of traditional discrepancy measures, and show that only the average squared discrepancy avoids the problems raised by Matou\v{s}ek. Furthermore, we present a comprehensive numerical study showing in particular that optimizing for the average squared discrepancy leads to strong performance for the L2L_2 star discrepancy, whereas the converse does not hold.

Cite

@article{arxiv.2508.04926,
  title  = {On the optimization of discrepancy measures},
  author = {François Clément and Nathan Kirk and Art B. Owen and T. Konstantin Rusch},
  journal= {arXiv preprint arXiv:2508.04926},
  year   = {2025}
}

Comments

22 pages, 3 Figures, 4 Tables

R2 v1 2026-07-01T04:38:13.932Z