A three-point counterexample to a random lattice discrepancy comparison
Read the paper (PDF) · Download PDF · Supporting ZIP · Zenodo record
Plain-English explanation · Read on this page · Collaboration
Abstract
For three points {Δ}, {Z+Δ}, {2Z+Δ} with independent continuous uniform generator Z and shift Δ, the expected normalized anchored fourth discrepancy moment is 19/1620. The corresponding independent-point moment is 4/405=16/1620. This gives a negative answer to the universal expected-moment comparison posed by Novak, reporting joint work with Hinrichs, in 2004. We give an elementary inclusion-area proof and a separate empty-tail argument giving reversal for every point count n ≥ 4 in dimension one. Three is the smallest possible counterexample point count. Ferudun's preprint of 14 October 2026 also gives the same exact counterexample and broader results; this note records the proof in the source repository PR dated 10 October. First priority is not claimed.
The idea in plain English
Discrepancy measures how far a set of points departs from uniform distribution. A random lattice is evenly patterned once its generator and shift are chosen, but these points share dependencies. At the fourth moment, those dependencies can outweigh the benefits of the pattern.
The precise result
Choose Z and Δ independently and uniformly on [0,1). The lattice ensemble is the three points {Δ}, {Z + Δ}, {2Z + Δ}, where braces denote fractional parts. Compare it with three independent uniform points.
For points t₀,…,tn−1 in [0,1), the normalized anchored fourth discrepancy moment is
disc₄(t)⁴ = ∫₀¹ |(1/n)∑j=0n−1 1[0,x)(tⱼ) − x|⁴ dx.
In dimension one at n = 3,
E disc₄(lattice)⁴ = 19/1620 > 4/405 = E disc₄(independent)⁴.
The excess is 1/540 and the ratio is 19/16. For n ≤ 2 the complete ensemble laws agree in every dimension, so three is the smallest possible counterexample point count. A separate empty-tail argument gives reversal for every n ≥ 4 in dimension one.
Why the result holds
The three-point calculation expands the fourth power of the counting function and integrates the pair and triple inclusion probabilities. Pairs have the independent uniform law; the triple does not. Elementary inclusion-area calculations give the exact rational values above. The all-n argument uses the event that the entire lattice lies in a short interval, leaving an empty tail on which the discrepancy is large.
Scope and limitations
- The comparison averages over the full continuous uniform generator and shift, with the same 1/n normalization on both sides.
- The reversed quantity is the expected fourth power of anchored L₄ discrepancy. A comparison of expected L₄ norms is a separate question.
- The explicit three-point value and the all-n ≥ 4 reversal stated here are in dimension one. The equality of ensemble laws for n ≤ 2 holds in every dimension.
- The result makes no claim about selected good generators, finite rational generator grids, or a separate star-discrepancy existence problem.
- First priority and independent discovery relative to other authors are not established by the repository timestamps.
Background and prior work
The question is posed by Erich Novak, reporting joint work with Aicke Hinrichs, in Oberwolfach Report 13/2004. A. Ferudun’s preprint, fixed version 1.0 of 14 October 2026, gives the same exact counterexample and broader results. This overlap is explicit in the native paper.
The note also records the source proof in repository PR 1135, whose server creation date is 10 October 2026. This dated repository record does not by itself establish historical public visibility, firstness, awareness or mutual independence.
Status and AI use
This is an unrefereed, AI-assisted research note. The author is responsible for its contents. The linked native paper and supporting material are the source of record.
The original authored material is available under Creative Commons Attribution 4.0 International. Third-party material retains its original rights and attribution requirements.
Read the paper
Supporting materials
These are unchanged copies of the native files, in the order shown by the native record.
- random-lattice-counterexample.pdf — paper PDF. Original file on Zenodo.
- random-lattice-counterexample-source-and-verification.zip — supporting material. Original file on Zenodo.
Collaborate on this result
Researchers interested in independent assessment of this work or a substantive collaboration are welcome to contact Alec. Authorship and responsibilities would be discussed in relation to actual contributions.
Cite this paper
Alec Kriebel. A three-point counterexample to a random lattice discrepancy comparison. Version 1.0.0, 17 October 2026. Zenodo. 10.5281/zenodo.23442649. Download BibTeX.