Wiki
Wiki

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

Updated


Statement

Notation (pp. 13--14). A subset of an Abelian group is sum-free when no a,b,ca,b,c in it, not necessarily distinct, satisfy a+b=ca+b=c. For a set BB, s(B)s(B) is the largest size of a sum-free subset of BB; for a sequence AA, s(A)s(A) is the largest length of a subsequence whose set of values is sum-free (see Proposition 1.2 for the definition in full).

Theorem 1.3 (p. 14, quoted). "For any finite Abelian group GG, every set BB of non-zero elements of GG satisfies s(B)>27∣B∣s(B)>\frac27|B|. The constant 27\frac27 is best possible. Similarly, every sequence AA of non-zero elements of GG satisfies s(A)>27∣A∣s(A)>\frac27|A|, and the constant 27\frac27 is optimal."

"Best possible" is shown by a family, not by a single group (p. 21): for G=Z7sG=\mathbb Z_7^s and B=G∖{0}B=G\setminus\{0\}, ∣B∣=7s−1|B|=7^s-1 and s(B)=2⋅7s−1s(B)=2\cdot7^{s-1}, so s(B)/∣B∣=2⋅7s−1/(7s−1)s(B)/|B|=2\cdot7^{s-1}/(7^s-1) exceeds 27\frac27 and tends to it as s→∞s\to\infty. No constant larger than 27\frac27 holds in every finite Abelian group, while each fixed group may admit a larger one.

The paper introduces the theorem as settling, for finite Abelian groups, the problem of Babai and Sós of estimating the largest sum-free subset of nn elements of a general group (p. 14), and calls it the paper's main result. The abstract (p. 13) states the set case.

Related statements in Section 4. For particular groups the constant improves (pp. 23--24): s(B)≥k+13k+1∣B∣s(B)\ge\frac{k+1}{3k+1}|B| for every sequence BB of nonzero elements of Zp\mathbb Z_p with p=3k+2p=3k+2 prime, s(B)≥13∣B∣s(B)\ge\frac13|B| for Zp\mathbb Z_p with p≡1(mod3)p\equiv1\pmod3 prime, s(B)≥2s−12s−1∣B∣s(B)\ge\frac{2^{s-1}}{2^s-1}|B| for Z2s\mathbb Z_2^s, each stated best possible, s(B)≥13∣B∣s(B)\ge\frac13|B| in Zn\mathbb Z_n when nn has no prime divisor congruent to 22 modulo 33, and Proposition 4.3 (p. 23): "For any prime p≡2(mod3)p\equiv2\pmod3 and any s≥1s\ge1, every sequence BB of non-zero elements of the cyclic group ZpsZ_{p^s} satisfies s(B)>13∣B∣s(B)>\frac13|B|", whose proof the paper omits (p. 24). Proposition 4.2 (p. 22), obtained by applying Theorem 1.3 (and Proposition 4.1) repeatedly, states that any set of nn nonzero elements of a finite Abelian group, and any set of nn nonzero reals, can be partitioned into O(log⁡n)O(\log n) sum-free subsets. Item 5 (p. 24) extends the lower bound 27\frac27, with the same optimal constant, to weakly sum-free sets, which exclude a1+a2=a3a_1+a_2=a_3 only for distinct a1,a2,a3a_1,a_2,a_3.

Source. N. Alon and D. J. Kleitman, Sum-free subsets, in: A Tribute to Paul Erdős (A. Baker, B. Bollobás and A. Hajnal, eds.), Cambridge Univ. Press (1990), 13--26, DOI 10.1017/CBO9780511983917.003, as described on the source card: the statement on p. 14, the proof in Section 3, pp. 18--21, the Section 4 statements on pp. 22--24.

Read depth. Claims checked: the statement, the optimality example and the Section 4 statements listed above were read clause by clause on the page images. The proof of the lower bound (pp. 18--20) was read for structure; the table of Lemma 3.1, whose proof the paper omits, was not recomputed. Nothing here is independently reviewed.

Proof pointer

Section 3, pp. 18--21. The lower bound is proved for sequences, which gives sets. In Zn\mathbb Z_n take the sum-free sets I1={x:13n<x≤23n}I_1=\{x:\frac13n<x\le\frac23n\} and I2={x:16n<x≤13n or 23n<x≤56n}I_2=\{x:\frac16n<x\le\frac13n\text{ or }\frac23n<x\le\frac56n\}. Lemma 3.1 (pp. 18--19) tabulates ∣dZn∩Ij∣/∣dZn∣|dZ_n\cap I_j|/|dZ_n| for the subgroups dZndZ_n, by n/dn/d modulo 66, and gives the weighted inequality (the paper's (3.1))

47 ∣dZn∩I1∣∣dZn∣+37 ∣dZn∩I2∣∣dZn∣≥27.\frac47\,\frac{|dZ_n\cap I_1|}{|dZ_n|}+\frac37\,\frac{|dZ_n\cap I_2|}{|dZ_n|}\ge\frac27 .

Embed GG in Zns\mathbb Z_n^s, map the terms bib_i to Zn\mathbb Z_n by a uniformly random homomorphism x↦∑jxjbijx\mapsto\sum_jx_jb_{ij}, whose image for each bib_i is uniform on a subgroup diZnd_iZ_n with di<nd_i<n, and compare the expected numbers M1,M2M_1,M_2 of terms landing in I1,I2I_1,I_2. The zero homomorphism lands no term in either set, so some homomorphism lands more than the average, and s(B)>Mjs(B)>M_j for j=1,2j=1,2; then (3.1) gives s(B)>47M1+37M2≥27∣B∣s(B)>\frac47M_1+\frac37M_2\ge\frac27|B| (p. 20). Optimality comes from Theorem 3.2 (p. 21), due to Rhemtulla and Street.

Dependencies

  • Theorem 3.2 (p. 21), cited from A. H. Rhemtulla and A. P. Street, Maximum sum-free sets in elementary Abelian pp-groups, Canad. Math. Bull. 14 (1971), 73--80: for a prime p=3k+1p=3k+1 and G=ZpsG=\mathbb Z_p^s, the largest sum-free subset of GG has kps−1kp^{s-1} elements. With p=7p=7 it gives the optimality example.
  • Lemma 3.1 (pp. 18--19), stated with its proof omitted as an easy case analysis.

Bears on

  • Problem 792: the problem concerns sets of integers and the theorem does not bound its f(n)f(n). The problem page uses the theorem to test a remark of Erdős's 1965 paper that his bound n/3n/3 holds in any finite Abelian group: by the optimality example no constant above 27\frac27, and so not 13\frac13, holds in every finite Abelian group.