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), Aq, QA and R(A;q) as defined in
Lemma 6;
QA consists of exact prime powers, and ω(n) counts distinct
prime divisors. Unqualified sums over q are sums over prime powers.
Statement (Lemma 5, pp. 13-15). There is an absolute constant c>0
with the following property. Let N≥M≥N1/2, with N
sufficiently large, and let k satisfy
1≤k≤cloglogN.
Suppose that A⊆[M,N] is a set of integers for which
ω(n)≤(logN)1/k(n∈A).
For every q∈QA such that
R(A;q)≥(logN)−1/2,
there is an integer d satisfying
qd>Mexp(−(logN)1−1/k),
ω(d)≤logk5loglogN,
and
n∈Aqqd∣n(qd,n/qd)=1∑nqd≫(logN)2/kR(A;q).
The endpoint k=1 in the printed statement is undefined; see the
proof scope immediately below.
Rewritten proof for 1<k≤cloglogN
The printed k=1 endpoint has no defined 1/logk bound and is not
asserted proved here. The subsequent Proposition 3 uses k→∞.
Fix an eligible prime power q, and define
y=exp((logN)1−2/k).
Let D consist of those positive integers d for which both of the following
hold:
pr∥d ⟹ pr>y,
and
qd∈(Mexp(−(logN)1−1/k),N].
For every n∈Aq, begin with n/q and remove all exact prime-power
components pr∥n/q having pr≤y. Since q has already
been separated, at most ω(n)−1
components are removed. Their product is strictly less than
yω(n)≤exp((logN)1−1/k).
The product of the components left behind is therefore some d∈D.
Moreover qd∣n and (qd,n/qd)=1, because q and all retained
components are exact prime-power components of n. It follows, after assigning
to each n such a d, that
R(A;q)≤d∈D∑d1n∈Aqqd∣n(qd,n/qd)=1∑nqd.(5.1)
Put
ω0=logk5loglogN.
For fixed d, discarding both the restriction n∈Aq and the coprimality
condition gives
n∈Aqqd∣n(qd,n/qd)=1∑nqd≤n≤Nqd∣n∑nqd≪logN.
Using $1_{\omega(d)\geq\omega_0}\leq
k^{\omega(d)-\omega_0}$ and an Euler-product majorant, the portion of the
right side of (5.1) with ω(d)>ω0 is at most
d∈Dω(d)>ω0∑d1n∈Aqqd∣n(qd,n/qd)=1∑nqd≪logNd: pr∥d⇒y<pr≤Nω(d)≥ω0∑d1≪k−ω0logNd: pr∥d⇒y<pr≤N∑dkω(d)≪C1kk−ω0logNy<p≤N∏(1+p−1k)≤k−ω0logN(C2logylogN)k≤C2kk−ω0(logN)3≤logN1(5.2)
for absolute constants C1,C2>0, provided c is sufficiently small and
N sufficiently large. For prime bases p≤y, every allowed exponent is at least two.
Their Euler factors have product at most
exp(k∑p∑a≥2p−a)≤ek.
For p>y, the factor is at most 1+k/(p−1), which is at most
(1−1/p)−k by Bernoulli's inequality for k>1.
Mertens' product estimate therefore gives the displayed bound; if
1<y<2, its logarithmic ratio is only larger, so the estimate still
holds with an absolute constant. The last line uses
k−ω0=(logN)−5,(logylogN)k=(logN)2,
and k≤cloglogN.
Since R(A;q)≥(logN)−1/2, the last quantity in (5.2) is at most
R(A;q)/2 once N is large. Thus (5.1) gives
21R(A;q)≤d∈Dω(d)≤ω0∑d1n∈Aqqd∣n(qd,n/qd)=1∑nqd.(5.3)
Another Euler-product estimate gives
d∈D∑d1≤d: pr∥d⇒y<pr≤N∑d1≪y<p≤N∏(1−p1)−1≪logylogN≪(logN)2/k.(5.4)
Comparing (5.3) and (5.4), at least one d∈D with
ω(d)≤ω0 has
n∈Aqqd∣n(qd,n/qd)=1∑nqd≫(logN)2/kR(A;q).
Membership in D supplies the required lower bound for qd, completing the
proof.
Dependencies and source details
Mertens' estimates, equations (1)–(2), p. 3, are external inputs, as
cited by Bloom to Montgomery and Vaughan, Chapter 2. The Euler product
in the unweighted estimate is written above in its ordinary form
∏(1−1/p)−1; the paper's displayed majorant
∏(1−1/(p−1))−1 is unnecessary and is undefined at p=2, which
lies in its range y<p whenever y<2.
The ordinary form handles the entire proof range k>1.
Bears on