Wiki
Wiki

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

Updated


Source. Thomas F. Bloom, On a density conjecture about unit fractions, arXiv:2112.03726v2 (12 October 2023). Printed and PDF page numbers agree.

Use R(A)R(A), AqA_q, QAQ_A and R(A;q)R(A;q) as defined in Lemma 6; QAQ_A consists of exact prime powers, and ω(n)\omega(n) counts distinct prime divisors. Unqualified sums over qq are sums over prime powers.

Statement (Lemma 4, p. 13). Let 0<ϵ<1/20<\epsilon<1/2, and let NN be sufficiently large in terms of ϵ\epsilon. Suppose that AA is a finite set of integers satisfying

R(A)≥(log⁡N)−ϵ/2R(A)\geq(\log N)^{-\epsilon/2}

and

(1−ϵ)log⁡log⁡N≤ω(n)≤2log⁡log⁡N(n∈A).(1-\epsilon)\log\log N\leq\omega(n)\leq2\log\log N \qquad(n\in A).

Then

∑q∈QA1q≥(1−2ϵ)e−1log⁡log⁡N.\sum_{q\in Q_A}\frac1q \geq(1-2\epsilon)e^{-1}\log\log N.

Rewritten proof. Set

σ=∑q∈QA1q,I=[(1−ϵ)log⁡log⁡N, 2log⁡log⁡N].\sigma=\sum_{q\in Q_A}\frac1q, \qquad I=[(1-\epsilon)\log\log N,\,2\log\log N].

Every n∈An\in A is the product of its distinct exact prime-power components, and their number is ω(n)∈I\omega(n)\in I. Summing over all possible collections of components, and then enlarging to all ordered choices, gives

R(A)≤∑t∈Iσtt!≤∑t∈I(eσt)t,R(A) \leq\sum_{t\in I}\frac{\sigma^t}{t!} \leq\sum_{t\in I}\left(\frac{e\sigma}{t}\right)^t,

where the last inequality uses t!≥(t/e)tt!\geq(t/e)^t and tt ranges over the integers in II.

If σ≥(1−ϵ)log⁡log⁡N\sigma\geq(1-\epsilon)\log\log N, the claimed weaker bound is immediate. Otherwise σ<t\sigma<t throughout II. The function (eσ/t)t(e\sigma/t)^t is then decreasing in tt, so there are at most 2log⁡log⁡N2\log\log N terms and

(log⁡N)−ϵ/2≤R(A)≤2log⁡log⁡N(σ(1−ϵ)e−1log⁡log⁡N)(1−ϵ)log⁡log⁡N.(\log N)^{-\epsilon/2} \leq R(A) \leq2\log\log N \left( \frac{\sigma}{(1-\epsilon)e^{-1}\log\log N} \right)^{(1-\epsilon)\log\log N}.

Taking the ((1−ϵ)log⁡log⁡N)((1-\epsilon)\log\log N)-th root yields

σ≥(1−ϵ)e−1log⁡log⁡N e−ϵ/(2(1−ϵ))(2log⁡log⁡N)−1/((1−ϵ)log⁡log⁡N).\sigma\geq (1-\epsilon)e^{-1}\log\log N\, e^{-\epsilon/(2(1-\epsilon))} (2\log\log N)^{-1/((1-\epsilon)\log\log N)}.

For 0<ϵ<1/20<\epsilon<1/2, e−ϵ/(2(1−ϵ))≥1−ϵe^{-\epsilon/(2(1-\epsilon))}\geq1-\epsilon. Choose NN so large that

(2log⁡log⁡N)2/log⁡log⁡N≤1+ϵ2.(2\log\log N)^{2/\log\log N}\leq1+\epsilon^2.

Since 1/(1−ϵ)≤21/(1-\epsilon)\leq2, the final factor in the preceding lower bound is at least (1+ϵ2)−1(1+\epsilon^2)^{-1}. Consequently

σ≥(1−ϵ)21+ϵ2e−1log⁡log⁡N≥(1−2ϵ)e−1log⁡log⁡N,\sigma\geq \frac{(1-\epsilon)^2}{1+\epsilon^2}e^{-1}\log\log N \geq(1-2\epsilon)e^{-1}\log\log N,

as required.

Dependencies

The elementary bound t!≥(t/e)tt!\ge(t/e)^t and prime-power factorization; no earlier numbered lemma is used.

Bears on