Wiki
Wiki

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

Updated

Stoll 2006 problem erdos graham concerning digits

../

corollary_3_5: Stoll's 2006 corollary identifying, for every integer starting value m outside {-1, 0}, the real number w whose binary digits the original recurrence u_{n+1} = floor(sqrt 2 (u_n + 1/2)) computes, through Beatty's theorem for the sequences floor(r(1 + 1/sqrt 2)) and floor(r(1 + sqrt 2)); it unifies the examples tabulated by Borwein and Bailey for 1 <= m <= 10.

fact_1: The Graham-Pollak identity as Stoll restates it in 2006: for the sequence u_1 = 1, u_{n+1} = floor(sqrt 2 (u_n + 1/2)), the difference u_{2n+1} - 2u_{2n-1} is the n-th binary digit of sqrt 2; the statement of Problem 482's first paragraph, held here in Stoll's restatement beside the 1970 note's original.

theorem_3_1: Stoll's general 2006 theorem: for every positive real w, every integer base g at least 2 and every integer triple (m, l, k) in six explicitly described cones with (g-1) dividing (k-1)l, the recurrence u_1 = m, u_{n+1} = floor(a (u_n + eps)) on odd steps and floor(b(u_n + l/(g-1))) on even steps has second differences u_{2n+1} - g u_{2n-1} equal to the base-g digits of w; the family behind the site's SOLVED label for Problem 482.

theorem_3_3: Stoll's two 2006 binary-digit families: for every positive real w and integer triples (m, l, k) with m outside {-1, 0}, k >= 0 and l in a range depending on m, a floor recurrence with the shift 1/2 on the odd steps (Theorem 3.3) or on the even steps (Theorem 3.4) whose second differences u_{2n+1} - 2u_{2n-1} are the binary digits of w; Theorem 3.3 at w = sqrt 2, eps = 1/2, (m, l, k) = (1, 0, 0) is the Graham-Pollak recurrence.


Thomas Stoll, On a problem of Erdős and Graham concerning digits, Acta Arith. 125 (2006), no. 1, 89--100. Received 16 March 2006 (the date printed on p. 100). The site's key St06.

The retained folder-name PDF is the journal's typeset file, 12 pages, printed pp. 89--100 (printed p. nn is PDF p. n−88n-88). Its text layer drops the plus signs, so the statements below were read on the rendered page images of printed pp. 89--90 and 92--94 (PDF pp. 1--2 and 4--6); the proofs (Section 4, pp. 94--99) were read in the text layer for their structure. Source: https://www.impan.pl/en/publishing-house/journals-and-series/acta-arithmetica/all/125/1/83728/on-a-problem-of-erdos-and-graham-concerning-digits. The file's text layer carries no copyright or license line; the journal's record offers the PDF under the download link "Free download under CC-BY license" and names no version or URL for it (https://www.impan.pl/en/publishing-house/journals-and-series/acta-arithmetica/all/125/1/83728/on-a-problem-of-erdos-and-graham-concerning-digits, read 2026-10-02): the Creative Commons Attribution license, with no version stated.

Read status: claims checked for Fact 1, the Erdős--Graham quotation, Example 1.1, Definitions 2.1--2.4, Theorem 3.1, Corollary 3.2, Theorems 3.3 and 3.4 and Corollary 3.5, read clause by clause on the page images; the inductive proofs of Section 4 were read for structure and not checked.

The paper answers the Erdős--Graham remark that there "must be similar results for α\sqrt\alpha and other algebraic numbers but we have no idea what they are", quoted on p. 90 from "the closing paragraph of Chapter 9 of [2], 'Miscellaneous Problems'", with the locator "[2, p. 96]" on p. 89, for the Graham--Pollak sequence u1=mu_1=m, un+1=⌊2(un+1/2)⌋u_{n+1}=\lfloor\sqrt2(u_n+1/2)\rfloor (1.1), whose differences u2n+1−2u2n−1u_{2n+1}-2u_{2n-1} give the nnth binary digit of 2\sqrt2 when m=1m=1 (Fact 1). Theorems 3.1, 3.3 and 3.4 replace 2\sqrt2 by an arbitrary w∈R+w\in\mathbb R^+, the shift 1/21/2 by a parameter ε\varepsilon, and the base 22 by any g≥2g\ge2, producing infinite families of two-step floor recurrences indexed by integer triples (m,l,k)(m,l,k) in explicitly described sets Di±\mathcal D_i^{\pm} whose second differences read off the base-gg digits of ww. Theorems 3.3 and 3.4 treat the binary case separately; Theorem 3.3 recovers Graham--Pollak's result on specializing w=2w=\sqrt2, ε=1/2\varepsilon=1/2, (m,l,k)=(1,0,0)(m,l,k)=(1,0,0) (p. 90 and p. 93). Combined with Beatty's theorem, Theorems 3.3 and 3.4 show the original recurrence (1.1) yields binary digits for every integer m∉{−1,0}m\notin\{-1,0\}, and Corollary 3.5 characterizes exactly which number ww is represented, unifying the examples tabulated by Borwein and Bailey for 1≤m≤101\le m\le10. Example 1.1 illustrates the generality with a recurrence whose second differences give the ternary digits of ee. The proofs are inductive arguments on the normalized expansion t=w/g⌊log⁡gw⌋t=w/g^{\lfloor\log_gw\rfloor}. This is the source for problem 482, the Erdős--Graham digit question about the Graham--Pollak recurrence.

Contents

  • Section 1 (pp. 89--90): the recurrence (1.1), Graham and Pollak's closed form un=⌊τ(2(n−1)/2+2(n−2)/2)⌋u_n=\lfloor\tau(2^{(n-1)/2}+2^{(n-2)/2})\rfloor for n≥2n\ge2 with τ\tau the mmth smallest element of {1,2,3,…}∪{2,22,32,…}\{1,2,3,\ldots\}\cup\{\sqrt2,2\sqrt2,3\sqrt2,\ldots\}, Fact 1 (Graham--Pollak), the citation history, the Erdős--Graham quotation, the earlier partial results of Rabinowitz--Gilbert and of the author's 2005 paper, and Example 1.1: v1=3v_1=3, vn+1=⌊−3e+9(vn+π)⌋v_{n+1}=\lfloor-\tfrac{3}{e+9}(v_n+\pi)\rfloor for odd nn, ⌊−(e+9)(vn+1)⌋\lfloor-(e+9)(v_n+1)\rfloor for even nn; then v2n+1−3v2n−1v_{2n+1}-3v_{2n-1} is the nnth ternary digit of e=(2.201101121…)3e=(2.201101121\ldots)_3.
  • Section 2 (pp. 90--92): the normalized t=w/gMt=w/g^M, M=⌊log⁡gw⌋M=\lfloor\log_gw\rfloor; the cones Ω=Ω1∪Ω2\Omega=\Omega_1\cup\Omega_2 of pairs (m,l)(m,l), their six subcones A1,…,A6\mathcal A_1,\ldots,\mathcal A_6, the sets Di={(m,l,k):(m,l)∈Ai, 0<∣k∣<βi}\mathcal D_i=\{(m,l,k):(m,l)\in\mathcal A_i,\ 0<|k|<\beta_i\} with Di±\mathcal D_i^{\pm} by the sign of kk, and the interval endpoints γi±,δi±\gamma_i^{\pm},\delta_i^{\pm} for ε\varepsilon (Definitions 2.1--2.4).
  • Section 3 (pp. 92--94): Theorem 3.1 (the general family for every base g≥2g\ge2), Corollary 3.2 (odd bases g≥3g\ge3 with both shifts 1/21/2), Theorems 3.3 and 3.4 (two further binary families), the Beatty-theorem partition of Z∖{−1,0}\mathbb Z\setminus\{-1,0\} by S(1+1/2)S(1+1/\sqrt2) and S(1+2)S(1+\sqrt2), and Corollary 3.5 with the table of w/2w/2 for 1≤m≤101\le m\le10.
  • Section 4 (pp. 94--99): Proposition 4.1 (for 0<w<g0<w<g, the closed form u2n+1=mgn+⌊wgn−1⌋u_{2n+1}=mg^n+\lfloor wg^{n-1}\rfloor forced by digit differences) and the inductive proofs of the three theorems and of Corollary 3.5.

Compiled scope

The whole paper was read (the statement pages on the page images, the proofs in the text layer). Fact 1, Theorem 3.1, Theorems 3.3--3.4 and Corollary 3.5 are compiled as statements with proof pointers; no step of the proofs was checked and nothing here is independently reviewed. The paper's own locator for the Erdős--Graham passage, printed p. 96 of the 1980 monograph, is what the site's source key for Problem 482 also gives.

Bears on. #482, as the second Stoll paper behind the site's SOLVED label: Fact 1 restates the Graham--Pollak identity of the problem's first paragraph, Theorem 3.1 gives, for every w>0w>0 and every base g≥2g\ge2, infinitely many recurrences of the same shape whose second differences are the base-gg digits of ww, Theorem 3.3 recovers the original recurrence as one member of a binary family, and Corollary 3.5 identifies the number whose binary digits the original recurrence with u1=mu_1=m produces for every integer m∉{−1,0}m\notin\{-1,0\}; the paper does not classify every recurrence with this property, which is what the site's "open-ended" qualification records.

Results to transcribe.

  • Fact 1 (p. 89): for m=1m=1, dn=u2n+1−2u2n−1d_n=u_{2n+1}-2u_{2n-1} is the nnth binary digit of 2=(1.011010100…)2\sqrt2=(1.011010100\ldots)_2.
  • Theorem 3.1 (p. 92): for w∈R+w\in\mathbb R^+, base g≥2g\ge2 and (m,l,k)∈Di±(m,l,k)\in\mathcal D_i^{\pm} with (g−1)∣(k−1)l(g-1)\mid(k-1)l, the recurrence u1=mu_1=m, un+1=⌊a(un+ε)⌋u_{n+1}=\lfloor a(u_n+\varepsilon)\rfloor for odd nn and ⌊b(un+l/(g−1))⌋\lfloor b(u_n+l/(g-1))\rfloor for even nn, with a=klg/((g−1)(t+mg))a=klg/((g-1)(t+mg)), b=g/ab=g/a and ε\varepsilon in the interval of Definition 2.4, has second differences u2n+1−gu2n−1u_{2n+1}-gu_{2n-1} equal to the base-gg digits of ww.
  • Theorem 3.3 (p. 93): the binary family with the shift 1/21/2 on odd steps, for m∉{−1,0}m\notin\{-1,0\}, k≥0k\ge0 and 0≤l≤m−10\le l\le m-1 if m≥1m\ge1 (resp. m+1≤l≤−1m+1\le l\le-1 if m≤−2m\le-2), with a=2k+1+(t+2l)/(t+2m)a=2k+1+(t+2l)/(t+2m), b=2/ab=2/a; specializing w=2w=\sqrt2, ε=1/2\varepsilon=1/2, (m,l,k)=(1,0,0)(m,l,k)=(1,0,0) recovers the Graham--Pollak fact. Theorem 3.4 (pp. 93--94): the second binary family, with the shift 1/21/2 on even steps and 1≤l≤m1\le l\le m (resp. m+1≤l≤−1m+1\le l\le-1), a=2k+1+2l/(t+2m)a=2k+1+2l/(t+2m).
  • Corollary 3.5 (p. 94): for each integer m∉{−1,0}m\notin\{-1,0\}, the number ww whose binary digits the original Graham--Pollak recurrence with u1=mu_1=m produces, via Beatty-type expressions in ⌊r2⌋\lfloor r\sqrt2\rfloor and ⌊r/2⌋\lfloor r/\sqrt2\rfloor.
  • Example 1.1 (p. 90): the sequence v1=3v_1=3, vn+1=⌊−3e+9(vn+π)⌋v_{n+1}=\lfloor-\tfrac{3}{e+9}(v_n+\pi)\rfloor for odd nn and ⌊−(e+9)(vn+1)⌋\lfloor-(e+9)(v_n+1)\rfloor for even nn satisfies v2n+1−3v2n−1=v_{2n+1}-3v_{2n-1}= the nnth ternary digit of ee.