Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement

For a sequence (xk)k≥1(x_k)_{k\ge1} on the circle S1≅[0,1)S^1\cong[0,1), the first nn terms cut the circle into intervals; an rr-span is the total length of rr consecutive intervals.

Theorem 2 (p. 2). Some sequence (xk)k≥1(x_k)_{k\ge1} on S1≅[0,1]S^1\cong[0,1], together with one universal constant cc with 0<c<∞0<c<\infty, has this property: for each r∈Nr\in\mathbb N there is a threshold, depending on rr, such that for every n∈Nn\in\mathbb N beyond it the intervals cut by x1,…,xnx_1,\dots,x_n satisfy

largest r-spansmallest r-span ≤ 1+clog⁡rr.\frac{\text{largest $r$-span}}{\text{smallest $r$-span}} \ \le\ 1+\frac{c\log r}{r}.

The paper proves it for two sequences, the golden-ratio Kronecker sequence xk={kφ}x_k=\{k\varphi\} with φ=(1+5)/2\varphi=(1+\sqrt5)/2 and the base-2 van der Corput sequence 12,14,34,18,…\frac12,\frac14,\frac34,\frac18,\ldots, and says the statement was conjectured in Brethouwer's 2024 Ph.D. thesis.

Source. F. Clément and S. Steinerberger, Balanced stick breaking, arXiv:2511.14637v1 (18 November 2025), Theorem 2 on p. 2 and Theorem 3 on p. 2 of that arXiv version, read in the text layer. The edition read is identified in the source digest.

Read depth. Claims checked: the statement was read clause by clause, with Theorem 3 and the remark that it implies Theorem 2; the proofs (Sections 2--3, pp. 3--11) were not read. Unrefereed; not independently reviewed.

Proof pointer

Theorem 3 (p. 2): for either sequence there is a universal c>0c>0 such that for all r∈Nr\in\mathbb N, all nn sufficiently large depending on rr, and all 0≤x≤1−r/n0\le x\le1-r/n,

∣#{1≤k≤n: x≤xk≤x+rn}−r∣≤clog⁡r,\Bigl|\#\bigl\{1\le k\le n:\ x\le x_k\le x+\tfrac rn\bigr\}-r\Bigr|\le c\log r,

and the paper says the bound holds for all intervals of length r/nr/n on S1S^1 (p. 3). A short-interval discrepancy bound of this kind controls every rr-span from both sides, which gives Theorem 2 (the derivation is on p. 9). Section 2 (pp. 3--9) proves Theorem 3 for the van der Corput sequence through an ordering lemma (Lemma 1, p. 4) and Lemmas 2--4 (p. 5); Section 3 (pp. 9--11) proves it for the golden-ratio sequence through the three-distance structure of the Kronecker sequence (Lemma 5, p. 9). Not reconstructed here. The derivation of Theorem 2 from Theorem 3, with the quantifiers made explicit and the literal r=1r=1 case of both statements recorded as false, is reconstructed (author-recorded, unreviewed) at its reconstruction page; the proofs of Theorem 3 remain unreconstructed.

Dependencies

Classical properties of the van der Corput and golden-ratio Kronecker sequences; the paper's own short-interval discrepancy estimates.

Bears on

  • Problem 1221: with μr=inf⁡alim sup⁡nMnr(a)/mnr(a)\mu_r=\inf_a\limsup_nM_n^r(a)/m_n^r(a) this gives μr≤1+clog⁡r/r\mu_r\le1+c\log r/r for every r≥2r\ge2 (an authored one-line remark; at r=1r=1 the printed statement would give μ1≤1\mu_1\le1, while μ1=2\mu_1=2), so r(μr−1)≤clog⁡rr(\mu_r-1)\le c\log r: the third expression of the de Bruijn--Erdős conjecture grows at most logarithmically if it grows at all. The 1949 lower bound is (5.7), μr≥1+1/r\mu_r\ge1+1/r.