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 a=(a1,a2,…)a=(a_1,a_2,\ldots) of numbers mod 11, the points a1,…,ana_1,\ldots,a_n cut the circle of circumference 11 into nn intervals; Mn1(a)M_n^1(a) and mn1(a)m_n^1(a) are the largest and smallest of their lengths, Λ1(a)=lim sup⁡nnMn1(a)\Lambda_1(a)=\limsup_n nM_n^1(a), λ1(a)=lim inf⁡nnmn1(a)\lambda_1(a)=\liminf_n nm_n^1(a), μ1(a)=lim sup⁡nMn1(a)/mn1(a)\mu_1(a)=\limsup_n M_n^1(a)/m_n^1(a), and Λ1\Lambda_1, λ1\lambda_1, μ1\mu_1 are the infimum, the supremum and the infimum of these over all sequences (Section 1, p. 14).

The r=1r=1 values (p. 14, proved in Section 2 and Sections 3--5).

Λ1=1log⁡2,λ1=1log⁡4,μ1=2.\Lambda_1=\frac1{\log2},\qquad \lambda_1=\frac1{\log4},\qquad \mu_1=2 .

The sequence ak=log⁡2(2k−1)a_k=\log_2(2k-1) reduced mod 11 attains all three: Λ1(a)=1/log⁡2\Lambda_1(a)=1/\log2, λ1(a)=1/log⁡4\lambda_1(a)=1/\log4, μ1(a)=2\mu_1(a)=2 (p. 15).

Source. N. G. de Bruijn and P. Erdős, Sequences of points on a circle, Proc. 52 (1949), 14--17; the values on printed p. 14 (PDF p. 2 of the TU/e portal PDF), Section 2 on pp. 14--15 (PDF pp. 2--3), read on the page images. The edition read is identified in the source digest.

Read depth. Claims checked: the definitions, the displayed values and the Section 2 computation were read clause by clause on the page images; the matching universal bounds are the r=1r=1 cases of the Section 3--5 results, whose proofs were read for structure only.

Proof pointer

Section 2 shows that a1,…,ana_1,\ldots,a_n sit on the circle in the same cyclic order as log⁡2n,log⁡2(n+1),…,log⁡2(2n−1)\log_2n,\log_2(n+1),\ldots,\log_2(2n-1), display (2.1), because reduction mod 11 pairs the aka_k with k≤nk\le n one-to-one with these numbers, the residues within each list being pairwise distinct. The nn intervals therefore have lengths log⁡2n+1n,log⁡2n+2n+1,…,log⁡22n2n−1\log_2\frac{n+1}n,\log_2\frac{n+2}{n+1},\ldots,\log_2\frac{2n}{2n-1} (the last one wrapping around), so

nMn1(a)=nlog⁡(1+1/n)log⁡2,nmn1(a)=nlog⁡(1−12n)−1log⁡2.nM_n^1(a)=\frac{n\log(1+1/n)}{\log2},\qquad nm_n^1(a)=\frac{n\log\bigl(1-\tfrac1{2n}\bigr)^{-1}}{\log2}.

As n→∞n\to\infty the first increases to 1/log⁡21/\log2, the second decreases to 1/log⁡41/\log4, and the ratio Mn1(a)/mn1(a)M_n^1(a)/m_n^1(a) increases to 22. The values are best possible because Section 3 gives Λ1(a′)≥1/log⁡2\Lambda_1(a')\ge1/\log2, Section 4 gives λ1(a′)≤1/log⁡4\lambda_1(a')\le1/\log4 and Section 5 gives μ1(a′)≥2\mu_1(a')\ge2 for every sequence a′a' (the r=1r=1 cases of the Section 3 bound, (4.3) and (5.7)).

Dependencies

Elementary properties of the logarithm; the universal bounds of Sections 3--5 for sharpness.

Bears on

  • Problem 1221: the site's commentary quotes these three values and the witness ak=log⁡2(2k−1)a_k=\log_2(2k-1); the r=1r=1 case of the problem's constants is exactly determined, and the question concerns the growth of the deviations for large rr.
  • Problem 480: background. The Chung--Graham chapter behind that problem attributes to this note the constant 1/log⁡41/\log4 for its clustering measure of a sequence in the unit interval; that constant is λ1\lambda_1 here, on the circle.