Wiki
Wiki

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

Updated


Statement

Setting (p. 145). For a set AA of integers, σ(n)=σA(n)\sigma(n)=\sigma_A(n) is the number of ordered pairs (a,a′)∈A2(a,a')\in A^2 with a+a′=na+a'=n.

Theorem 2 (p. 146, quoted). "There exists a set AA of nonnegative integers that forms a basis of order 2 (that is, σ(n)⩾1\sigma(n)\geqslant1 for all nn), and satisfies ∑n⩽Nσ(n)2=O(N)\sum_{n\leqslant N}\sigma(n)^2=O(N). (1.2)"

The abstract (p. 145) states the same result, printing the basis condition as "s(n)⩾1s(n)\geqslant1 [sic]". The introduction (p. 145) observes that boundedness in the first mean is equivalent to AA having O(N)O(\sqrt N) elements up to NN, which it calls well known to be possible; Theorem 2 is the square-mean version. The paper does not settle the question of Erdős and Turán whether some basis of order 2 has bounded σ(n)\sigma(n): Remark 1.2 (p. 146) says only the weaker Theorem 2 is obtained.

Source. Imre Z. Ruzsa, A Just Basis, Monatsh. Math. 109 (1990), 145--151, doi:10.1007/BF01302934. Labels and pages are those of the journal print: Theorem 2 on p. 146, Lemma 4.1 on p. 149, the proof in Section 4 on pp. 149--151. The edition read is identified on the source card.

Read depth. Claims checked: the definition and the statement were read clause by clause on the printed pages. The proof was read but not checked step by step. Nothing here is independently reviewed.

Proof pointer

Pages 149--151. Write D(X)=∑nσX(n)2D(X)=\sum_n\sigma_X(n)^2, the number of solutions of a+b=c+da+b=c+d in XX. Lemma 4.1 (p. 149): for a finite set XX of integers and a prime pp with (2p)=−1\left(\frac{2}{p}\right)=-1 there is a set Y⊂[p2/2,4p2)Y\subset[p^2/2,4p^2) with ∣Y∣≤12p|Y|\le12p, Y+Y⊃[4p2,5p2]Y+Y\supset[4p^2,5p^2] and D(X∪Y)≤D(X)+C(∣X∣3/p+p2)D(X\cup Y)\le D(X)+C(|X|^3/p+p^2) for an absolute constant CC; YY is a translate A+tA+t of the set of Theorem 1 by an integer tt, chosen by averaging over the solution counts. The proof of Theorem 2 (p. 151) takes primes pip_i with (2pi)=−1\left(\frac{2}{p_i}\right)=-1 and 1.1<pi+1/pi<5/21.1<p_{i+1}/p_i<\sqrt5/2, starts from X0=[0,4p12]X_0=[0,4p_1^2], adds the set YiY_i of Lemma 4.1 at each stage, and shows by induction that D(Xi)≪pi2D(X_i)\ll p_i^2. Sums up to NN use only elements of XiX_i when pi2≤2N<pi+12p_i^2\le2N<p_{i+1}^2, which gives (1.2).

Dependencies

Theorem 1 and Lemma 4.1 (p. 149).

Bears on

  • Problem 1192: the problem asks, for every r≥2r\ge2, for a basis of order rr with ∑n≤xfr(n)2≪x\sum_{n\le x}f_r(n)^2\ll x. Theorem 2 gives such a basis for r=2r=2, by the translation recorded on the claim page Ruzsa; the paper does not treat r≥3r\ge3.
  • Problem 28: the problem asserts that a set whose sumset contains all large integers has unbounded 1A∗1A1_A\ast1_A. Theorem 2 bounds the counts only in square mean and does not decide the problem.