Wiki
Wiki

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

Updated


Statement

Theorem 1 (printed p. 617). If the integers 1≤a1<⋯<an1\le a_1<\cdots<a_n have pairwise distinct subset sums, that is, the 2n2^n sums ∑i=1nεiai\sum_{i=1}^n\varepsilon_ia_i with each εi∈{0,1}\varepsilon_i\in\{0,1\} all differ, then their reciprocals sum to less than 22:

∑i=1n1ai<2.\sum_{i=1}^n\frac1{a_i}<2.

The paper introduces the theorem through the divisors of an integer: nn has property P if the 2k2^k sums ∑εidi\sum\varepsilon_id_i over its divisors did_i are distinct, and Theorem 1 gives σ(n)/n<2\sigma(n)/n<2 for such nn; "We conjectured this and the simple and ingenious proof is due to C. Ryavec" (p. 617).

Refinement (printed p. 619, after the proof). The paper says that the same argument, under the same distinct-subset-sum hypothesis, gives the sharper bound

∑i=1n1ai≤2−12n−1,\sum_{i=1}^n\frac1{a_i}\le2-\frac1{2^{n-1}},

with equality only for the powers of two, ai=2i−1a_i=2^{i-1} for i=1,…,ni=1,\ldots,n. (The set {1,2,4,…,2n−1}\{1,2,4,\ldots,2^{n-1}\} attains the bound, its reciprocal sum being 2−21−n2-2^{1-n}; the converse direction is what is printed.) The page then recalls Erdős's conjecture that distinct subset sums force an>2n−Ca_n>2^{n-C}, with the offer of a prize (the site's Problem 1).

Source. S. J. Benkoski and P. Erdős, On weird and pseudoperfect numbers, Math. Comp. 28 (1974), no. 126, 617–623, DOI 10.1090/S0025-5718-1974-0347726-9 (Crossref record read); the copy read is a seven-page scan, printed p. nn on PDF p. n−616n-616. Theorem 1 on printed p. 617 (PDF p. 1), the proof on pp. 617–619 (PDF pp. 1–3), the refinement on p. 619 (PDF p. 3), read on the page images.

Read depth. Claims checked: Theorem 1 and the refinement were read clause by clause on the page images. The one-page proof was read for its structure (below); it is not reconstructed or independently reviewed here.

Proof pointer

Pages 617–619 (Ryavec's argument). For 0<x<10<x<1 the distinctness of the 2n2^n subset sums gives ∏i=1n(1+xai)<∑k≥0xk=1/(1−x)\prod_{i=1}^n(1+x^{a_i})<\sum_{k\ge0}x^k=1/(1-x), hence ∑ilog⁡(1+xai)<−log⁡(1−x)\sum_i\log(1+x^{a_i})<-\log(1-x); dividing by xx and integrating over (0,1)(0,1) (display (1)), then substituting y=xaiy=x^{a_i} in each term, ∑i1ai∫01log⁡(1+y)y dy<−∫01log⁡(1−x)x dx\sum_i\frac1{a_i}\int_0^1\frac{\log(1+y)}y\,dy<-\int_0^1\frac{\log(1-x)}x\,dx, that is ∑i1ai⋅π212<π26\sum_i\frac1{a_i}\cdot\frac{\pi^2}{12}<\frac{\pi^2}6, so ∑1/ai<2\sum1/a_i<2. The refinement is stated as following from "the same argument" with no further detail.

Dependencies

None beyond the two classical integrals.

Bears on

  • Problem 350: the status-defining source. The problem's dissociated set (all subset sums distinct) is the theorem's hypothesis (subsets of AA correspond to the vectors (εi)(\varepsilon_i); a dissociated set contains no 00, so its elements are 1≤a1<⋯<an1\le a_1<\cdots<a_n), and the conclusion is the problem's ∑n∈A1/n<2\sum_{n\in A}1/n<2. The refinement is the site's "∑1/n≤2−21−∣A∣\sum1/n\le2-2^{1-|A|}, with equality if and only if A={1,2,…,2k}A=\{1,2,\ldots,2^k\}", where the source writes the extremal set as ai=2i−1a_i=2^{i-1}, the powers of two up to 2n−12^{n-1}.
  • Problem 469 (not linked from this page; the source card carries its row): the property P context in which the theorem is stated.