A new elementary proof of the Prime

Number Theorem
By Florian K. Richter
August 4, 2021
arXiv:2002.03255v3 [math.NT] 3 Aug 2021

Abstract
Let Ω(n) denote the number of prime factors of n. We show that for any bounded
f : N → C one has
N N
1 X 1 X
f (Ω(n) + 1) = f (Ω(n)) + oN →∞ (1).
N N
n=1 n=1

This yields a new elementary proof of the Prime Number Theorem.

1. Introduction
One of the most fundamental results in mathematics is the Prime Number Theorem, which
states that
|{p 6 N : p prime}|
lim = 1. (1.1)
N →∞ N/log N
It was conjectured independently by Gauß and Legendre towards the end of the 18th
century and proved independently by Hadamard and de la Vallée Poussin in the year
1896. Their proofs were similar in nature and relied on sophisticated analytic machinery
from complex analysis developed throughout the 18th and 19th century by the combined
effort of many great mathematicians of this era, including Euler, Dirichlet, Chebyshev,
and Riemann. This method of proving the Prime Number Theorem became known as the
analytic method. We refer the reader to [Apo00, Gol73b, Gol73a] for more details on the
history behind the analytic proof and to [New80] for an abridged version of it; see also
[Zag97].
Even though it was believed for a long time not to be possible, an elementary proof of
the Prime Number Theorem was eventually found by Erdős and Selberg in [Erd49, Sel49].
In this context, elementary does not necessarily mean simple, but refers to methods that
avoid using complex analysis and instead rely only on rudimentary facts from calculus
and basic arithmetic identities and inequalities. Their approach was based on Selberg’s
“fundamental formula”, which states that
X X
log2 (p) + log(p) log(q) = 2x log(x) + O(x). (1.2)
p6x pq6x

We refer to [Lev69] for a streamlined exposition of the Erdős-Selberg proof, and to [Gol04]
and [SG09] for the history behind it. See also [Sha50] for a short proof of (1.2) and
[Dia82, Gra10] for more general surveys on this topic. A novel and dynamically inspired
way of deriving the Prime Number Theorem from (1.2), which bears many similarities to
the argument that we present in Section 2, was recently and independently discovered by
McNamara [McN20].
Today we also know of other elementary ways of proving the Prime Number Theorem.
For instance, an alternative elementary proof was found by Daboussi in [Dab84], using
what he called the “convolution method” (cf. [Dab89, p. 1]). We refer the reader to
Chapter 4 in the book of Tenenbaum and Mendés France [TMF00] for a friendly rendition
of Daboussi’s argument. A third elementary proof, which is different from the proofs
of Erdős-Selberg and Daboussi, was provided by Hildebrand in [Hil86] and relies on a
corollary of the large sieve ([Mon71, Corollary 3.2]) as a starting point.
The purpose of this paper is to provide yet another elementary proof of the Prime Num-
ber Theorem. More precisely, we prove the following result, which contains an equivalent
form of the Prime Number Theorem as a special case:
Theorem 1.1. Let Ω(n) denote the number of prime factors of a positive integer n
(counted with multiplicities). Then for any bounded f : N → C one has
N N
1 X 1 X
f (Ω(n) + 1) = f (Ω(n)) + oN →∞ (1). (1.3)
N n=1 N n=1

Letting λ(n) = (−1)Ω(n) denote the classical Liouville function, it immediately follows
from Theorem 1.1 applied to the sequence f (n) = (−1)n that
N
1 X
lim λ(n) = 0. (1.4)
N →∞ N n=1

This is a well-known equivalent form of the Prime Number Theorem.1

Theorem 1.1 also recovers other results in multiplicative number theory. For instance,
by considering f (n) = ζ n where ζ 6= 1 is a m-th root of unity, we obtain from (1.3) that
N
1 X Ω(n)
lim ζ = 0.
N →∞ N
n=1

This implies a theorem of Pillai and Selberg [Pil40, Sel39], which says that for m ∈ N
and r ∈ {0, 1, . . . , m − 1} the set {n ∈ N : Ω(n) ≡ r mod m} has asymptotic density
1/m. In a similar vein, (1.3) applied to sequences of the form f (n) = e2πiαn for α ∈ R\Q
yields a classical result of Erdős and Delange (see [Erd46, p. 2, lines 4–5] and [Del58]),
asserting that (Ω(n)α)n∈N is uniformly distributed mod 1 for any irrational α. Theorem 1.1
also recovers several results recently obtained by the author in [BR20], including [BR20,
Theorem A].
The proof of Theorem 1.1 is self-contained (with the exception of Stirling’s approxi-
mation formula used in Section 3 without a proof) and was inspired by the author’s work
1
The validity of (1.4) was first observed by von Mangoldt in [vM97, p. 852] and the equivalence between
(1.4) and (1.1) was later realized by Landau (see [Lan99, §4] and [Lan11] and [Lan09, pp. 620–621] and
[Lan09, pp. 631–632]). See also [Ten95, p. 55].

2
in [BR20]. It is worth noting that this is the first proof of the Prime Number Theorem
that builds on Chebyshev’s original idea of estimating the number of primes between n
and 2n.

2. The proof
A well-known relation in number theory, which is often regarded as a corollary of the
Turán-Kubilius inequality (cf. [Dab75, Lemma 1], [Kát86, Eq. (3.1)], and [Ell79, Lemma
4.7]), states that for any finite set of primes P one has
N 2 !
1 X X X1 X1
lim sup 1p|n − = O , (2.1)
N →∞ N n=1 p p
p∈P p∈P p∈P

where 1p|n denotes the function that is 1 if p divides n and 0 otherwise. It is common
to interpret (2.1) using a probabilistic point of view: By considering {1, 2, . . . , N} as a
discrete
P probability space (with normalized counting measure as the probability measure)
and p∈P 1p|n as a random variable on this space, (2.1) says that for large N the expected
numberPof primes in P that divide a ”randomly chosen” n P ∈ {1, . . . , N} approximately
equals p∈P 1/p, with a standard deviation on the scale of ( p∈P 1/p)1/2 .
An important role in our proof of Theorem 1.1 is played by a generalization of (2.1)
where the finite set of primes P is replaced by an arbitrary finite set of positive integers
B ⊂ N.
Proposition 2.1. Suppose B ⊂ N is finite and non-empty. Then
N 2  2
1 X X X1 X X Φ(q, q ′ ) |B|
1q|n − = ′
+O , (2.2)
N n=1 q∈B q ′
qq N
q∈B q∈B q ∈B

where Φ : N×N → N ∪ {0} is the function Φ(m, n) := gcd(m, n) − 1.

Note that (2.2) implies (2.1), because for any finite set of primes P one has
!
X X Φ(p, p′ ) X 1 1
 X1
′
= 1− = O .
pp p p p
p∈P p′ ∈P p∈P p∈P
P
Proof of Proposition 2.1. Define a := q∈B 1/q. By expanding the square in (2.2) we get
1
PN P P 2 2 1
PN P
N n=1 | q∈B 1 q|n − q∈B 1/q| = S 1 − 2aS 2 + a , where S 1 := N n=1 q,q ′ ∈B 1q|n 1q ′ |n
N N
and S2 := N1 n=1 q∈B 1q|n . Since N1 n=1 1q|n = 1/q + O(1/N), we obtain S2 = a +
P P P
1
PN ′ ′
O(|B|/N). On the other hand, N n=1 1q|n 1q ′ |n = gcd(q, q )/qq + O(1/N) implies S1 =
′ ′ 2 2 2
P P
q∈B q ′ ∈B Φ(q, q )/qq +a +O(|B| /N). Substituting these estimates into S1 −2aS2 +a
finishes the proof of (2.2).
Before we proceed further, it will be convenient to rewrite (2.2) using the language of
averages. Given a finite set A ⊂ N and an arithmetic function f : A → C, we denote the

3
Cesàro average and the logarithmic average of f over A respectively by
P
1 X log n∈A f (n)/n
E f (n) :=
|A| n∈A
f (n) and E f (n) := P .
n∈A 1/n
n∈A n∈A

P let [x] abbreviate the set {1, 2, . . . , ⌊x⌋}. After dividing both sides of (2.2) by
Also
( q∈B 1/q)2 , we obtain the following equivalent version of it expressed in terms of av-
erages:

2
 2
E E q1q|n − 1 = E E′ Φ(q, q ) + O |B| .
n∈[N ]
log
q∈B
log log
q∈B
′
q ∈B
(2.3)
N
The following proposition is our main technical result. By combining it with (2.3), we
will be able to finish the proof of Theorem 1.1 rather quickly.
Proposition 2.2. For all η > 0, there exists k0 ∈ N such that for all k > k0 there exist
two finite, non-empty sets B1 , B2 ⊂ N with the following properties:
(a) all elements in B1 are primes and all elements in B2 are a product of exactly k primes;
(b) the sets B1 and B2 have the same cardinality and if B1 = {p1 < . . . < pt } and
B2 = {q1 < . . . < qt } then (1 − η)pj 6 qj 6 (1 + η)pj holds for all j = 1, . . . , t;
(c) Elog log
m∈Bi En∈Bi Φ(m, n) 6 η for i = 1, 2, where Φ is as in Proposition 2.1.

Proof of Theorem 1.1 assuming Proposition 2.2. Fix η > 0 and let k0 ∈ N be as guaran-
teed by Proposition 2.2. This means that for every k > k0 we can find two finite and
non-empty sets B1 , B2 ⊂ N satisfying properties (a), (b), and (c). For any g : N → C with
|g(n)| 6 1 we thus have
2 2
E g(Ω(n)) − Elog E g(Ω(qn)) = E g(Ω(n)) − Elog E q1q|ng(Ω(n)) + O 1
 
n∈[N ] q∈B2 n∈[N/q] n∈[N ] q∈B2 n∈[N ] N
2 1
log
E E


6 1 − q1q|n +O
n∈[N ] q∈B 2 N
1
6 η +O ,
N
where the second to last inequality follows from the Cauchy-Schwarz inequality and the
last inequality follows from property (c) combined with (2.3). Since Ω(nq) = Ω(n) + Ω(q),
we get

E g(Ω(n)) = Elog E g(Ω(n) + Ω(q)) + O η1/2 + N −1/2 .

(2.4)
n∈[N ] q∈B2 n∈[N/q ]

An analogous calculation carried out with B1 in place of B2 and g(n + k − 1) in place of

g(n) yields

E g(Ω(n) + k − 1) = Elog E g(Ω(n) + Ω(p) + k − 1) + O η1/2 + N −1/2 .

n∈[N ] p∈B1 n∈[N/p]

(2.5)
Recall that B1 consists only of primes and B2 only of k-almost primes, which means
Ω(p) = 1 for all p ∈ B1 and Ω(q) = k for all q ∈ B2 . This allows us to rewrite (2.4) and

4
(2.5) as
log
E g(Ω(n)) E E g(Ω(n) + k) + O η 1/2 + N −1/2 ,


= (2.6)
n∈[N ] q∈B n∈[ / ]
2 N q

log
E g(Ω(n) + k − 1) E E g(Ω(n) + k) + O η 1/2 + N −1/2 .


= (2.7)
n∈[N ] q∈B n∈[ / ]
1 N q

Finally, let B1 = {p1 < . . . < pt } and B2 = {q1 < . . . < qt } be enumerations of B1 and B2 .
Since (1 − η)pj 6 qj 6 (1 + η)pj , it follows that En∈[N/pj ] g(Ω(n) + k) = En∈[N/qj ] g(Ω(n) +
k) + O(η). Taking logarihtmic averages over B1 = {p1 , . . . , pt } and B2 = {q1 , . . . , qt }
(cf. [BR20, Lemma 2.3]) leaves us with
log log
E E g(Ω(n) + k)
p∈B n∈[ / ]
1 N p
= E E g(Ω(n) + k)
q∈B n∈[ / ]
2 N q
+ O(η). (2.8)

From (2.6), (2.7), and (2.8) it follows that En∈[N ] g(Ω(n)) = En∈[N ] g(Ω(n)+k−1)+O(η 1/2 +
N −1/2 ). This holds for all k > k0 and hence
E g(Ω(n) + k) = E g(Ω(n) + l) + O η1/2 + N −1/2


(2.9)
n∈[N ] n∈[N ]

for all k, l > k0 . Relation (1.3) now follows from (2.9) by taking k = k0 , l = k0 + 1,
g(n) = f (n − k0 ), and letting η go to 0.

3. Proof of Proposition 2.2

The starting point for our proof of Proposition 2.2 are Chebyshev-type estimates on the
number of primes in intervals. More precisely, we derive a rough lower bound on the
number of primes in (8x , 8x+1 ], as well as a rough upper bound on the number of primes
in (8x , 8x+ε ] for small ε.
Proposition 3.1. Let P be the set of primes. There are x0 > 1 and ε0 > 0 such that
8x
(i) |P ∩ (8x , 8x+1 ]| > √x
for all x > x0 , and
ε 8x
(ii) |P ∩ (8x , 8x+ε ]| 6 x for all x > x0 and ε ∈ (0, ε0].
The ideas used in the proof of Proposition 3.1 are classical and date back to Chebyshev.
We begin with the following lemma.
Lemma 3.2. We have the asymptotic estimate
x log(2)
P ∩ (1, x] > + O(1). (3.1)
log x
Proof. To obtain (3.1) for arbitrary positive reals x, it is enough to prove it for all even
natural numbers, i.e., x
= 2n. In this case, the key is to study the prime factorization of the
binomial coefficient 2n n
. Observe that there are ⌊m/p⌋ many numbers in {1, . . . , m} that
are divisible by p. Out of those, there are ⌊m/p2 ⌋ many divisible by p2 , and out of those
there are ⌊m/p3 ⌋ many divisible by p3 , and so on. Therefore, if ν is the largest exponent
for which pν 6 m, then the power of p in m! is equal to ⌊m/p⌋ + ⌊m/p2 ⌋ + . . . + ⌊m/pν ⌋. In
light of this observation, we see that the multiplicity of a prime p in the prime factorization

5
 2n

(2n)!
of n
= (n!)2
is given by the formula
X
⌊2n/pi ⌋ − 2⌊n/pi ⌋, (3.2)
16i6νp

is the largest exponent for which pνp 6 2n. Since ⌊2n/pi ⌋ − 2⌊n/pi ⌋ 6 1, we can
where νp P
νp
estimate i=1 ⌊2n/pi ⌋ − 2⌊n/pi ⌋ 6 νp . This yields
2n
  Y
6 pνp 6 (2n)|P∩(1,2n]| ,
n
p∈P∩(1,2n]

which after taking logarithms leaves us with

 2n 
log 6 log(2n)|P ∩ (1, 2n]|. (3.3)
n
Stirling’s approximation formula implies that log(m!) = m log(m) − m + O(log m). This
2n


can now be used to finish the proof by approximating log n with 2 log(2)n + O(log n)
in (3.3).
Lemma 3.2 gives a reasonable lower bound on the asymptotic number of primes in
(1, x], which is important for the proof for part (i) of Proposition 3.1. For the proof of
part (ii) we also need an upper bound.
Lemma 3.3. Define β(σ) := σ log(σ) − (σ − 1) log(σ − 1). Then for all 1 < σ 6 16,
β(σ) x
P ∩ (x, σx] 6 + O(1). (3.4)
log x
Proof. For convenience, let us write σx for the quantity ⌊σx⌋



x ⌊x⌋
. Observe that every
σx


prime

number in the interval Q (x, σx] divides the number x . Therefore, the number
σx Q |P∩(x,σx]|
x
is greater or equal than p∈P∩(x,σx] p. Using p∈P∩(x,σx] p > x and taking
logarithms, we obtain
 σx 
log > log(x)|P ∩ (x, σx]|. (3.5)
x
Similarly as in the proof of Lemma 3.2, we can now use Stirling’s approximation formula,
log(m!) = m log(m) − m + O(log m), to estimate that
σx
 
log = ⌊σx⌋ log(⌊σx⌋) − ⌊x⌋ log(⌊x⌋) − (⌊σx⌋ − ⌊x⌋) log(⌊σx⌋ − ⌊x⌋) + O(log x)
x
= σx log(σx) − x log(x) − (σ − 1)x log((σ − 1)x) + O(log x)
= σx log(σ) − (σ − 1)x log(σ − 1) + O(log x).
Together with (3.5), this proves (3.4).
Proof of Proposition 3.1. The proof of part (ii) simply follows
√ from Lemma 3.3 (applied
with σ = 8ε ) and the fact that the order of magnitude of ε is much larger than the order
of magnitude of β(8ε ) as ε tends to 0.
For the proof of part (i), we start by rewriting the interval (8x , 8x+1 ] in the form
8x x
(8x , 8x+1] = (1, 8x+1]\ 06n63x ( 2n+1 , 28n ]. Lemma 3.2 gives the estimate |P ∩ (1, 8x+1]| >
S

6
8x+1 /3(x + 1) + O(1), whereas Lemma 3.3 (applied with σ = 2) gives the estimate |P ∩
(8x /2n+1, 8x /2n ]| 6 8x /2n+1 x + O(1). Therefore
8x+1 X 8x 8x+1 8x
P ∩ (8x , 8x+1] > − n+1 x
+ O(x) > − + O(x).
3(x + 1) 06n63x
2 3(x + 1) x
8x
This implies that if x0 is sufficiently large then |P ∩ (8x , 8x+1]| > x
for all x > x0 .
Proposition 3.1 is the only number-theoretic component in our proof of Proposition 2.2
and, as we have mentioned above, it doesn’t use any ideas that weren’t already available
to Chebyshev. The rest of our argument is more combinatorial in nature.
The first part of Proposition 3.1 tells us that we can find a fair amount of primes in
any interval of the form (8n , 8n+1 ]. However, we will need a bit more control over where
these primes are within this interval. The following proposition tells us that in (8n , 8n+1 ]
we can find two smaller intervals (8x , 8x+δ ] and (8y , 8y+δ ] which are not too close together
but also not too far apart, and each containing a good amount of primes.
Lemma 3.4. Let x0 be as in Proposition 3.1. There exists ε1 > 0 such that for all
ε ∈ (0, ε1] and all δ ∈ (0, 1) there exists D = D(ε, δ) ∈ (0, 1) with the following property:
For all n > x0 there are x, y ∈ [n, n + 1) with ε4 < y − x < ε such that
D8n D8n
P ∩ (8x , 8x+δ ] > , and P ∩ (8y , 8y+δ ] > .
n n
Proof. As guaranteed by Proposition 3.1, the number of primes in (8n , 8n+1 ] is at least
8n /n. Therefore, by the Pigeonhole Principle, for some t ∈ [n, n + 1) the number of primes
in (8t , 8t+ε ] is at least ε8n /2n. We can then cover the interval (8t , 8t+ε ] by K := ⌈ε−3 ⌉
many smaller intervals in the following way:
4 4 4 4 4
(8t , 8t+ε ] = 8t , 8t+ε ∪ 8t+ε , 8t+2ε ∪ . . . ∪ 8t+(K−1)ε , 8t+Kε .
4 4
By Lemma 3.3, each of the intervals (8t+iε , 8t+(i+1)ε ] contains at most O(ε2 8n /n) many
primes, whereby the whole interval (8t , 8t+ε ] contains at least O(ε8n /n) many primes. It
follows that if ε is chosen sufficiently small, say smaller than some threshold ε1 , then we can
4 4
find two non-consecutive a, b ∈ {0, 1, . . . , K − 1} such that the intervals (8t+aε , 8t+(a+1)ε ]
4 4
and (8t+bε , 8t+(b+1)ε ] contain at least O(ε48n /n) many primes. Using the Pigeonhole
Principle once more we can then find for every δ ∈ (0, 1) some x ∈ [t + aε4 , t + (a + 1)ε4 )
and some y ∈ [t + bε4 , t + (b + 1)ε4 ) such that the intervals (8x , 8x+δ ] and (8y , 8y+δ ] contain
at least O(δε4 8n /n) many primes. Since a and b are non-consecutive, we have y − x > ε4 ,
and since x, y ∈ [t, t + ε) we have y − x < ε.
The final ingredient in our proof of Proposition 2.2 is a purely combinatorial lemma.
Lemma 3.5. Fix x0 > 1 and ε > 0. Suppose X is a subset of R with the property that
for every n > x0 there exist x, y ∈ X ∩ [n, n + 1) with ε4 < y − x < ε. Let k > ⌈2/ε4 ⌉.
Then for all n1 , . . . , nk ∈ {n ∈ N : n > x0 } there exist z, z1 , . . . , zk ∈ X such that
(I) zi ∈ [ni , ni + 1) for all 1 6 i 6 k;
(II) z1 + . . . + zk ∈ [z, z + ε).

7
Proof. Let n1 , . . . , nk ∈ {n ∈ N : n > x0 }. According to the hypothesis of the lemma, we
can find for every i = 1, . . . , k two numbers xi , yi ∈ [ni , ni + 1) ∩ X with ε4 < yi − xi < ε.
Define, for every i = 0, 1, . . . , k, the number ui := x1 + . . . + xi + yi+1 + . . . + yk . Since
ε4 < yi − xi we have uk − u0 > kε4 > 2. This implies that there exists some z ∈ X with
u0 < z < uk . Since ui+1 − ui < ε, there is some i0 ∈ {0, 1, . . . , k} such that ui0 ∈ [z, z + ε).
Setting zi := xi for i 6 i0 and zi := yi for i > i0 , we obtain z1 + . . . + zk = ui0 ∈ [z, z + ε)
as desired.
Proof of Proposition 2.2. Let η ∈ (0, 1) be given. Let x0 and ε0 be as in Proposition 3.1
and let ε1 be as in Lemma 3.4. Pick any ε > 0 with ε < min{ε0 , ε1 , log(1 + η)/log(64)}
and set k0 := ⌈2/ε4 ⌉. We claim that k0 is as desired, meaning that for all k > k0 there
exist two finite, non-empty sets B1 , B2 ⊂ N satisfying properties (a), (b), and (c).
To verify this claim, fix k > k0 , set δ := ε/k, and let D = D(ε, δ) be as in Lemma 3.4.
Define
n o
X := x > x0 : P ∩ (8x , 8x+δ ] > D8⌊x⌋ /⌊x⌋ ,

and for every x ∈ X let Px be a subset of P ∩ (8x , 8x+δ ] with |Px | = ⌊D8⌊x⌋ /⌊x⌋⌋. The sets
Px are the building blocks from which we will construct B1 and B2 .
According to Lemma 3.4 the set X satisfies the hypothesis of Lemma 3.5, which allows
us to find for all ~n = (n1 , . . . , nk ) ∈ {n ∈ N : n > x0 }k numbers z~n , z1,~n , . . . , zk,~n ∈ X such
that
(I) zi,~n ∈ [ni , ni + 1) for all 1 6 i 6 k, and
(II) z1,~n + . . . + zk,~n ∈ [z~n , z~n + ε).
Note that Property (I) and the definition of Px imply
|Pzi,~n | = ⌊D8ni /ni ⌋ . (3.6)
Next, let N = N(D, η, k) be a constant that is to be determined later, and define
sets A1 , . . . , Ak ⊂ {n ∈ N : n > x0 } in the following way: Pick any s1P∈ N with s1 >
max{x0 , 2k} and let A1 be any finite subset of s1 N = {s1 n : n ∈ N} with n∈A1 1/n > N.
Then, assuming Ai has already been found, P take any si+1 > max(A1 + . . . + Ai ) and let
Ai+1 be any finite subset of si+1 N with n∈Ai+1 1/n > N. Following this procedure until
i = k, we end up with a family of finite sets A1 , . . . , Ak with the following property:
(A) For any (n1 , . . . , nk ) 6= (n′1 , . . . , n′k ) ∈ A1 × . . . × Ak the distance between n1 + . . . + nk
and n′1 + . . . + n′k is at least 2k.
We are now ready to define the sets B1 and B2 . The set B2 is defined as
[
B2 := Pz1,~n · . . . · Pzk,~n . (3.7)
n∈A1 ×...×Ak
~

According to (3.6) we have

D k 8n1 +...+nk D k 8n1 +...+nk D8⌊z~n ⌋
Pz1,~n · . . . · Pzk,~n 6 6 6 ,
n1 · . . . · nk n1 + . . . + nk ⌊z~n ⌋
where the last inequality follows from D k 6 D and n1 + . . . + nk 6 ⌊z~n ⌋. Therefore
|Pz1,~n · . . . · Pzk,~n | 6 |Pz~n |, which guarantees the existence of a set Q~n ⊂ Pz~n with |Q~n | =

8
|Pz1,~n · . . . · Pzk,~n |. Define B1 to be
[
B1 := Q~n . (3.8)
n∈A1 ×...×Ak
~

It remains to show that B1 and B2 satisfy properties (a), (b), and (c).
Proof that B1 and B2 satisfy (a). By construction, B1 consists only of primes and B2 only
of numbers that are a product of k primes. △
Proof that B1 and B2 satisfy (b). By definition we have Pzi,~n ⊂ (8zi,~n , 8zi,~n +δ ], and so
Pz1,~n · . . . · Pzk,~n ⊂ (8z1,~n , 8z1,~n +δ ] · . . . · (8zk,~n , 8zk,~n +δ ]
⊂ (8z1,~n +...+zk,~n , 8z1,~n +...+zk,~n +kδ ]
⊂ (8z~n , 8z~n+2ε ],
where the last inclusion follows from Property (II) and the fact that kδ 6 ε. Also by
definition we have Q~n ⊂ (8z~n , 8z~n +δ ] ⊂ (8z~n , 8z~n +2ε ]. Using the fact that |n1 +. . .+nk −z~n | 6
k + 1 and Property (A), we conclude that (Pz1,~n · . . . · Pzk,~n ) ∩ (Pz1,~n′ · . . . · Pzk,~n′ ) = ∅ and
Q~n ∩ Q~n′ = ∅ whenever ~n 6= ~n′ . Since |Pz1,~n · . . . · Pzk,~n | = |Q~n |, we immediately get that B1
and B2 have the same cardinality. Moreover, both Q~n and Pz1,~n · . . . · Pzk,~n belong to the
interval (8z~n , 8z~n +2ε ], which implies that the ratio between any element in Pz1,~n · . . . · Pzk,~n
and any element in Q~n lies between 8−2ε and 82ε . Since 82ε 6 (1 + η), it follows that
for enumerations Pz1,~n · . . . · Pzk,~n = {q1 < . . . < qr } and Q~n = {p1 < . . . < pr } we have
(1 − η)pj 6 qj 6 (1 + η)pj for all j = 1, . . . , r. This property now easily extends to
enumerations of B1 and B2 . △
Proof that B1 and B2 satisfy (c). First, let us show Elog log ′
p∈B1 Ep′ ∈B1 Φ(p, p ) 6 η. Since B1
consists only of primes, we have Φ(p, p′ ) = 0 unless p = p′ . Therefore
P Φ(p,p′ ) P 1
log log p,p ′ ∈B pp ′ p∈B1 p 1
E E′ Φ(p, p ) = P
p∈B1 p ∈B1
′ 1
1 6 P 1 =
P 1. (3.9)
p,p′ ∈B1 pp′ p,p′ ∈B1 pp′ p∈B1 p
P P P
Note, in view of (3.8) we have p∈B1 1/p = n∈A1 ×...×Ak
~ p∈Q~n 1/p and since any element
in Q~n is smaller than 8n1 +...+nk +k+1 we have
X 1 X |Q~n |
> .
p 8 8n1 +...+nk
k+1
p∈B1 n=(n1 ,...,nk )∈A1 ×...×Ak
~

Then, using (3.6) and |Q~n | = |Pz1,~n ·. . .·Pzk,~n | we can estimate |Q~n | > D k 8−k 8n1 +...+nk /(n1 ·
. . . · nk ), which implies
X 1 X Dk Dk N k
> 2k+1
> 2k+1
. (3.10)
p 8 n1 · . . . · nk 8
p∈B1 (n1 ,...,nk )∈A1 ×...×Ak

Thus, if N was chosen sufficiently large, then Elog log ′

p∈B1 Ep′ ∈B1 Φ(p, p ) 6 η follows from com-
bining (3.9) and (3.10).

9
 Next, let us show Elog log ′
q∈B2 Eq ′ ∈B2 Φ(q, q ) 6 η. In view of (3.7) we have
X Φ(q, q ′ ) X X X Φ(q, q ′ )
′
= ′
. (3.11)
qq qq
q,q ′ ∈B2 ~ n′ ∈A1 ×...×Ak
n,~ q∈Pz1,~n ···Pzk,~n q ′ ∈Pz1,~n′ ···Pzk,~n′

If q ∈ Pz1,~n · . . . · Pzk,~n and q ′ ∈ Pz1,~n′ · . . . · Pzk,~n′ are coprime then Φ(q, q ′ ) = 0. Hence, such
a pair does not contribute to (3.11). On the other hand, if q and q ′ are Q not coprime then
there must exist a finite non-empty set F ⊂ {1, . . . , k}, a number u ∈ i∈F Pzi,~n′ , and a
′ ′ ′ ′
Q
number u ∈ i∈F / Pzi,~n′ such that q = uu and gcd(q, q ) = u; note that this only happens
when ni = n′i for all i ∈ F because Pzi,~n and Pzi,~n′ are disjoint otherwise. In this case, we
have
Φ(q, q ′ ) u−1 1
′
= ′
6 ,
qq qq qu′
and therefore2
X Φ(q, q ′ ) X X X X 1
6 . (3.12)
qq ′ Q qu′
q,q ′ ∈B2 F ⊂{1,...,k} ~ n′ ∈A1 ×...×Ak
n,~ q∈Pz1,~n ···Pzk,~n u′ ∈ i∈F
/ Pzi,~n′
F 6=∅ ni =n′ , i∈F
i

′ ′
Next, we can use Pzi,~n ⊂ (8ni , 8ni+1 ] and Pzi,~n′ ⊂ (8ni , 8ni+1 ] to deduce that
k
! !
X X 1 X X Y 1 Y 1
′
6 ni n′
Q qu Q 8 8 i
q∈Pz1,~n ···Pzk,~n u′ ∈ i∈F
/ Pzi,~
q∈Pz1,~n ···Pzk,~n u′ ∈ i∈F
/ Pzi,~
i=1 i∈F
/
n′ n′

k
! !
Y |Pzi,~n | Y |Pz |
n′
i,~
= n′i
.
8ni 8
i=1 i∈F
/

Thereafter, using (3.6), we get

k
! ! k
! !
Y |Pzi,~n | Y |Pzi,~n′ | Y D YD
′ 6 .
8ni 8ni ni n′i
i=1 i∈F
/ i=1 i∈F
/

All together, this implies that

k
! !
X Φ(q, q ′ ) X X Y D YD
6
qq ′ ni n′i
q,q ′ ∈B2 F ⊂{1,...,k} n′ ∈A1 ×...×Ak
n,~
~ i=1 i∈F
/
F 6=∅ ni =n′ , i∈F
i
k X
! !
X Y 1 YX 1
= D 2k−|F | .
ni n′i
F ⊂{1,...,k} i=1 n∈Ai / n∈Ai
i∈F
F 6=∅

Dk
P 1
Qk P 1
A calculation similar to (3.10) shows that q∈B2 q > 83k i=1 n∈Ai n . Putting every-
2
In the published version of this paper the restriction ni = n′i , i ∈ F , in the subscript of the second
summation on the right hand side of equation (3.12) is missing, leading to a mistake that carries through
the rest of the argument. This is corrected here. We thank Corentin Darreye for bringing this issue to
our attention.

10
thing together gives
P Φ(q,q ′ )
log log q,q ′ ∈B2 qq ′
E E
q∈B q ∈B
2
′
′
Φ(q, q ) = P 1
2
q,q ′ ∈B2 qq ′
Q P  Q
P 2k−|F | k 1
P 1


F ⊂{1,...,k}, F 6=∅ D i=1 n∈Ai n i∈F
/ n∈Ai n
6  Qk P  Qk P 
Dk 1 Dk 1
83k i=1 n∈Ai n 83k i=1 n∈Ai n
6k −|F |
X 8 D
6 Q P 1

.
F ⊂{1,...,k} i∈F n∈Ai n
F 6=∅

P log log ′
Finally, since n∈Ai 1/n > N, we see that Eq∈B2 Eq ′ ∈B2 Φ(q, q ) 6 η as long as N was
chosen sufficiently large. △
This completes the proof of Proposition 2.2.

Acknowledgments: The author thanks Vitaly Bergelson and Redmond McNamara for
commenting on an earlier version of this paper, and the anonymous referee for providing
numerous helpful suggestions. This work was supported by the National Science Founda-
tion under grant number DMS 1901453.

References
[Apo00] T. M. Apostol, A centennial history of the prime number theorem,
in Number theory, Trends Math., Birkhäuser, Basel, 2000, pp. 1–14.
https://doi.org/10.1007/978-3-0348-7023-8 1.

[BR20] V. Bergelson and F. K. Richter, Dynamical generalizations of the Prime

Number Theorem and disjointness of additive and multiplicative semigroup ac-
tions, ArXiv e-prints (2020). Available at https://arxiv.org/abs/2002.03498.

[Dab75] H. Daboussi, Fonctions multiplicatives presque périodiques B, in Journées

Arithmétiques de Bordeaux (Conf., Univ. Bordeaux, Bordeaux, 1974),
tome 24–25, Soc. Math. France, Paris, 1975, pp. 321–324. Available at
http://www.numdam.org/item/AST 1975 24-25 321 0.

[Dab84] H. Daboussi, Sur le théorème des nombres premiers, C. R. Acad. Sci. Paris
Sér. I Math. 298 no. 8 (1984), 161–164.

[Dab89] H. Daboussi, On the prime number theorem for arithmetic

progressions, J. Number Theory 31 no. 3 (1989), 243–254.
https://doi.org/10.1016/0022-314X(89)90071-1.

[Del58] H. Delange, On some arithmetical functions, Illinois J. Math. 2 (1958), 81–87.

Available at http://projecteuclid.org/euclid.ijm/1255380835.

11
[Dia82] H. G. Diamond, Elementary methods in the study of the distribution of
prime numbers, Bull. Amer. Math. Soc. (N.S.) 7 no. 3 (1982), 553–589.
https://doi.org/10.1090/S0273-0979-1982-15057-1.

[Ell79] P. D. T. A. Elliott, Probabilistic number theory. I, Grundlehren der

Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sci-
ence] 239, Springer-Verlag, New York-Berlin, 1979, Mean-value theorems.
https://doi.org/10.1007/978-1-4612-9989-9.

[Erd46] P. Erdős, On the distribution function of additive functions, Ann. of Math.

(2) 47 (1946), 1–20. https://doi.org/10.2307/1969031.

[Erd49] P. Erdős, On a new method in elementary number theory which leads to an

elementary proof of the prime number theorem, Proc. Nat. Acad. Sci. U.S.A.
35 (1949), 374–384. https://doi.org/10.1073/pnas.35.7.374.

[Gol04] D. Goldfeld, The elementary proof of the prime number theorem: an histori-
cal perspective, in Number theory (New York, 2003), Springer, New York, 2004,
pp. 179–192. https://doi.org/10.1007/978-1-4419-9060-0 10.

[Gol73a] L. J. Goldstein, Correction to: “A history of the prime number theorem”

(Amer. Math. Monthly 80 (1973), 599–615), Amer. Math. Monthly 80 (1973),
1115. https://doi.org/10.2307/2318546.

[Gol73b] L. J. Goldstein, A history of the prime number theorem, Amer. Math.

Monthly 80 (1973), 599–615. https://doi.org/10.2307/2319162.

[Gra10] A. Granville, Different approaches to the distribution of primes, Milan J.

Math. 78 no. 1 (2010), 65–84. https://doi.org/10.1007/s00032-010-0122-7.

[Hil86] A. Hildebrand, The prime number theorem via the large sieve, Mathematika
33 no. 1 (1986), 23–30. https://doi.org/10.1112/S002557930001384X.

[Kát86] I. Kátai, A remark on a theorem of H. Daboussi, Acta Math. Hungar. 47

no. 1-2 (1986), 223–225. https://doi.org/10.1007/BF01949145.
P∞ µ(k)
[Lan99] E. Landau, Neuer Beweis der Gleichung k=1 k , Ph.D. the-
sis, Friedrich-Wilhelms-Iniversität zu Berlin, 1899. Available at
http://resolver.sub.uni-goettingen.de/purl?PPN317979566.

[Lan09] E. Landau, Handbuch der Lehre von der Verteilung der Primzahlen -
Zweiter Band, 2, Leipzig Und Berlin: B. G. Teubner, 1909. Available at
http://name.umdl.umich.edu/ABV2766.0001.001.

[Lan11] E. Landau, Über die äquivalenz zweier Hauptsätze der analytischen Zahlenthe-
orie, Sitzungsberichte der Kaiserlichen Akadamie der Wissenschaften in Wien,
Mathematisch-Naturwissenschaftliche Klasse 120 no. 2a (1911), 1–16.

12
[Lev69] N. Levinson, A motivated account of an elementary proof of the
prime number theorem, Amer. Math. Monthly 76 (1969), 225–245.
https://doi.org/10.2307/2316361.
P∞ µ(k)
[vM97] H. von Mangoldt, Beweis der Gleichung k=1 k = 0, Sitzungs-
berichte der Königlich Preussischen Akademie der Wissenschaften zu Berlin,
Zweiter Halbband – Juli bis Dezember (1897), 835–852. Available at
https://biodiversitylibrary.org/page/29982413.
[McN20] R. McNamara, A dynamical proof of the Prime Number Theorem, ArXiv
e-prints (2020). Available at https://arxiv.org/abs/2002.04007.
[Mon71] H. L. Montgomery, Topics in multiplicative number theory, Lecture Notes in
Mathematics, Vol. 227, Springer-Verlag, Berlin-New York, 1971.
[New80] D. J. Newman, Simple analytic proof of the prime number theorem, Amer.
Math. Monthly 87 no. 9 (1980), 693–696. https://doi.org/10.2307/2321853.
[Pil40] S. S. Pillai, Generalisation of a theorem of Mangoldt, Proc. Indian Acad. Sci.,
Sect. A. 11 (1940), 13–20.
[Sel49] A. Selberg, An elementary proof of the prime-number theorem, Ann. of Math.
(2) 50 (1949), 305–313. https://doi.org/10.2307/1969455.
[Sel39] S. Selberg, Zur Theorie der quadratfreien Zahlen, Math. Z. 44 no. 1 (1939),
306–318. https://doi.org/10.1007/BF01210655.
[Sha50] H. N. Shapiro, On a theorem of Selberg and generalizations, Ann. of Math.
(2) 51 (1950), 485–497. https://doi.org/10.2307/1969336.
[SG09] J. Spencer and R. Graham, The elementary proof of the prime num-
ber theorem, Math. Intelligencer 31 no. 3 (2009), 18–23, With a note
on the controversy by E. G. Straus and a postscript by Carl Pomerance.
https://doi.org/10.1007/s00283-009-9063-9.
[Ten95] G. Tenenbaum, Introduction to analytic and probabilistic number theory, Cam-
bridge Studies in Advanced Mathematics 46, Cambridge University Press, Cam-
bridge, 1995, Translated from the second French edition (1995) by C. B. Thomas.
[TMF00] G. Tenenbaum and M. Mendès France, The prime numbers and their dis-
tribution, Student Mathematical Library 6, American Mathematical Society,
Providence, RI, 2000, Translated from the 1997 French original by Philip G.
Spain. https://doi.org/10.1090/stml/006.
[Zag97] D. Zagier, Newman’s short proof of the prime number theorem, Amer. Math.
Monthly 104 no. 8 (1997), 705–708. https://doi.org/10.2307/2975232.

Florian K. Richter
Northwestern University
[email protected]

13