Quasi-uniform low-discrepancy digital sequences

An infinite sequence in the s-dimensional unit cube is quasi-uniform if its first N points stay separated and cover the cube on the length scale N−1/sN^{-1}/^{s} for every N. Can a digital sequence combine that property with low discrepancy, meaning that point counts track volume in axis-aligned boxes?

Why it matters and what counts as progress

Why it matters: Quasi-Monte Carlo integration uses one sequence at many sample sizes. Low discrepancy controls how point counts are distributed, while separation and covering control clustering and large gaps.

What counts as progress: Construct one explicit digital sequence that works for every prefix, enlarge the range of bases or dimensions, or prove that a specified family cannot achieve quasi-uniformity. State the metric and any restriction on the sample sizes for explicit constants.

How the problem developed

  1. 26 Nov 2025 · Halton obstruction [1]

    Takashi Goda, Roswitha Hofer and Kosuke Suzuki Ruled out quasi-uniformity for ordinary Halton sequences in every dimension at least two, as well as some Halton-type sequences.

  2. 13 Feb 2026 · Digital-sequence question [2]

    Josef Dick, Takashi Goda and Kosuke Suzuki Asked for quasi-uniform digital constructions within low-discrepancy sequences.

  3. 29 Oct 2026 · Residue construction [3][4]

    Leonhard Gruenschloss with Claude Opus 5.5 For every s≥2s\ge 2 and prime p>2sp>2^{s}, gave a residue-based digital (0,s)-sequence with bounded Euclidean mesh ratio at all prefixes. Also ruled out the polynomial-base Halton-type family specified in the formal statement.

AI contribution

Low-discrepancy quasi-uniform digital sequences in every dimension

Leonhard Gruenschloss

For every dimension s≥2s \ge 2 and prime p>2sp > 2^{s}, a specified linear scramble of a generalized Faure digital sequence has bounded Euclidean mesh ratio across all prefixes. The paper also rules out quasi-uniformity for the stated Halton-type family with pairwise coprime nonconstant polynomial bases over FpF_{p} in dimensions at least two.

AI involvementAI-led
ImportanceNotable
Technical verificationFormal (Lean)
Human understandingPartial
Paper date:

Sources

  1. Disproving the quasi-uniformity of the Halton sequences and of some Halton-type sequences
  2. On the quasi-uniformity properties of quasi-Monte Carlo point sets and sequences – Part II
  3. Low-discrepancy quasi-uniform digital sequences in every dimension
  4. Exact Lean statements and provenance
  5. Lean comparison results
Publication history