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 set A\mathcal A of positive integers, write s∧As^\wedge\mathcal A for the set of all sums of ss distinct elements of A\mathcal A; A\mathcal A is admissible when s∧A∩t∧A=∅s^\wedge\mathcal A\cap t^\wedge\mathcal A=\emptyset whenever s≠ts\ne t (printed p. 141). The paper says the notion "has been introduced by P. Erdős in 1962 (cf. [2]) and called admissibility by E.G. Straus in 1966 (cf. [5])".

Theorem 1 (printed p. 142). "There exists an integer N0N_0, effectively computable, such that for any integer N≥N0N\ge N_0 and any admissible subset A⊂[1,N]\mathcal A\subset[1,N] we have

Card⁡A≤2N+1/4−1.\operatorname{Card}\mathcal A\le2\sqrt{N+1/4}-1.

"

The introduction (p. 141) records: Erdős's conjecture that the largest size of an admissible subset of [1,N][1,N] is attained by a block of consecutive integers ending at NN; Straus's computation that {N−k+1,N−k+2,…,N}\{N-k+1,N-k+2,\ldots,N\} is admissible if and only if k≤2N+1/4−1k\le2\sqrt{N+1/4}-1; Straus's inequality ∣A∣≤(4/3+o(1))N|\mathcal A|\le(4/\sqrt3+o(1))\sqrt N; the slight reduction of the constant by Erdős, Nicolas and Sárközy (Théorème 1); and the authors' own ∣A∣≤(2+o(1))N|\mathcal A|\le(2+o(1))\sqrt N from part 1 (Israel J. Math. 92 (1995), 33--43), filed as deshouillers_1995_additive_problem_erdos_straus; its Theorem 1, card⁡A≤2N1/2+CN5/12\operatorname{card}\mathcal A\le2N^{1/2}+CN^{5/12}, is on printed p. 34 (PDF p. 2), read there clause by clause on the page image and paged on theorem_1. Theorem 1 therefore gives, for N≥N0N\ge N_0, the exact value max⁡∣A∣=⌊2N+1/4−1⌋\max|\mathcal A|=\lfloor2\sqrt{N+1/4}-1\rfloor, attained by Straus's block: the block gives the lower bound and Theorem 1 the matching upper bound. This one-line combination is made here; the paper states the theorem and the block computation separately.

Theorem 2 (p. 142, quoted from part 1). "Let A\mathcal A be an admissible set included in [1,N][1,N], such that Card⁡A>1.96N\operatorname{Card}\mathcal A>1.96\sqrt N. If NN is large enough, there exist C⊂A\mathcal C\subset\mathcal A and an integer qq having the following properties : (i) Card⁡C≤105N5/12\operatorname{Card}\mathcal C\le10^5N^{5/12}, (ii) for some tt the set t∧Ct^\wedge\mathcal C contains at least 3N5/63N^{5/6} terms in an arithmetic progression modulo qq, (iii) A∖C\mathcal A\setminus\mathcal C is included in an arithmetic progression modulo qq containing at most N7/12N^{7/12} terms."

Remark (p. 142). The authors state, without giving details, that their method also describes the admissible subsets of [1,N][1,N] of largest size: for N=n2N=n^2 or N=n2+nN=n^2+n with nn large enough, the Erdős–Straus block is the unique admissible subset of [1,N][1,N] of largest size.

Source. J.-M. Deshouillers and G. A. Freiman, On an additive problem of Erdős and Straus, 2, in Structure theory of set addition, Astérisque 258, Soc. Math. France (1999), 141–148 (the Numdam record, gives MR 1701192 and Zbl 0979.11005; the article's own DOI is 10.24033/ast.442, per its Crossref record read); the copy read is the Numdam file, 9 pages, printed p. nn on PDF p. n−139n-139. The introduction and Theorems 1 and 2 with the remark on printed pp. 141–142 (PDF pp. 2–3), read on the page images.

Read depth. Claims checked: the definition, the historical account, Theorem 1, Theorem 2 and the remark were read clause by clause on the page images. The proof (Sections 1–3, pp. 142–147) was read for its structure only; N0N_0 is not made explicit in the paper.

Proof pointer

Section 1 (pp. 142–143) proves Proposition 1, a local lemma: for integers r,s,t,a,qr,s,t,a,q with t≥2s−qt\ge2s-q, s≥4r+3+qs\ge4r+3+q and 0≤a<q0\le a<q, if D\mathcal D is a set of tt integers congruent to aa modulo qq spanning (t−1+r)q(t-1+r)q, then among any 2r+12r+1 consecutive integers congruent to sasa modulo qq in the range of s∧Ds^\wedge\mathcal D, at least r+1r+1 lie in s∧Ds^\wedge\mathcal D. Section 2 (pp. 144–145) proves Theorem 3, the structure of an admissible A⊂[1,N]\mathcal A\subset[1,N] with ∣A∣=2N1/2+O(N5/12)|\mathcal A|=2N^{1/2}+O(N^{5/12}): the modulus qq of Theorem 2 is O(N5/12)O(N^{5/12}), and, for A={a1<⋯<a∣A∣}\mathcal A=\{a_1<\cdots<a_{|\mathcal A|}\}, some u∈[N11/24,2N11/24]u\in[N^{11/24},2N^{11/24}] has a∣A∣−u−au+1=q(2N1/2+O(N11/24))a_{|\mathcal A|-u}-a_{u+1}=q(2N^{1/2}+O(N^{11/24})), a span estimate for the middle elements. Section 3 (pp. 146–147) takes A\mathcal A of maximal cardinality and applies Proposition 1 with s=σs=\sigma and s=σ+qs=\sigma+q, where σ=[(t−q)/2]\sigma=[(t-q)/2], to the middle part D=A∩[au+1,a∣A∣−u]\mathcal D=\mathcal A\cap[a_{u+1},a_{|\mathcal A|-u}] of t=∣A∣−2ut=|\mathcal A|-2u elements, uu as in Theorem 3, and derives Theorem 1 in the form (∣A∣+1)2≤4N+1(|\mathcal A|+1)^2\le4N+1. Not reconstructed here.

Dependencies

Theorem 2 of the authors' first paper (Israel J. Math. 92 (1995), 33--43, DOI 10.1007/BF02762069), quoted as Theorem 2 here; the paper is filed as deshouillers_1995_additive_problem_erdos_straus, and its Theorem 2 is on printed p. 34 (PDF p. 2), read there clause by clause on the page image on 2026-09-22 and paged on theorem_2; the quotation above matches it apart from "modulo qq" for the original's "with difference dd". Straus's block computation (J. Math. Sci. 1 (1966), 77–80, not held), quoted on p. 141.

Bears on

  • Problem 874: the status-defining source. The problem's k(N)k(N) is the largest admissible subset of {1,…,N}\{1,\ldots,N\}; Theorem 1 with Straus's block computation gives k(N)=⌊2N+1/4−1⌋k(N)=\lfloor2\sqrt{N+1/4}-1\rfloor for N≥N0N\ge N_0, hence k(N)=2N1/2+O(1)k(N)=2N^{1/2}+O(1) and k(N)∼2N1/2k(N)\sim2N^{1/2}, the affirmative answer to the site's question; the uniqueness remark is the site's "in some cases the largest such AA has the form (N−k,N]∩N(N-k,N]\cap\mathbb N".
  • Problem 875: for an infinite admissible A={a1<a2<⋯ }A=\{a_1<a_2<\cdots\}, Theorem 1 applied to A∩[1,x]A\cap[1,x] gives A(x)≤2x+1/4−1A(x)\le2\sqrt{x+1/4}-1 for x≥N0x\ge N_0, so an≥n(n+2)/4a_n\ge n(n+2)/4 for large nn, and a gap bound an+1−an≤nca_{n+1}-a_n\le n^c for all large nn forces c≥1c\ge1; both are deductions made here from the theorem.
  • Problem 789: the paper's introduction attests Straus's (4/3+o(1))N(4/\sqrt3+o(1))\sqrt N bound and the naming of admissibility; the problem's h(n)h(n) is bounded by the largest admissible subset of {1,…,n}\{1,\ldots,n\}.