\pdfrender

TextRenderingMode=2, LineWidth=0.1pt

A new elementary proof of the Prime Number Theorem

By Florian K. Richter

Abstract

Let $\Omega(n)$ denote the number of prime factors of $n$ . We show that for any bounded $f\colon\mathbb{N}\to\mathbb{C}$ one has

\frac{1}{N}\sum_{n=1}^{N}\,f(\Omega(n)+1)=\frac{1}{N}\sum_{n=1}^{N}\,f(\Omega(n))+{\rm o}_{N\to\infty}(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

\lim_{N\to\infty}\leavevmode\nobreak\ \frac{|\{p\leqslant N:p\leavevmode\nobreak\ \text{prime}\}|}{{N}/{\log N}}\leavevmode\nobreak\ =\leavevmode\nobreak\ 1.

(1.1)

It was conjectured independently by Gauß and Legendre towards the end of the 18 ${}^{\text{th}}$ 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 18 ${}^{\text{th}}$ and 19 ${}^{\text{th}}$ 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

\sum_{p\leqslant x}\log^{2}(p)\,+\,\sum_{pq\leqslant x}\log(p)\log(q)\leavevmode\nobreak\ =\leavevmode\nobreak\ 2x\log(x)\,+\,{\rm O}(x).

(1.2)

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 Number 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 $\Omega(n)$ denote the number of prime factors of a positive integer $n$ (counted with multiplicities). Then for any bounded $f\colon\mathbb{N}\to\mathbb{C}$ one has

\displaystyle\frac{1}{N}\sum_{n=1}^{N}\,f(\Omega(n)+1)\leavevmode\nobreak\ =\leavevmode\nobreak\ \frac{1}{N}\sum_{n=1}^{N}\,f(\Omega(n))\,+\,{\rm o}_{N\to\infty}(1).

(1.3)

Letting $\lambda(n)=(-1)^{\Omega(n)}$ denote the classical Liouville function, it immediately follows from 1.1 applied to the sequence $f(n)=(-1)^{n}$ that

\displaystyle\lim_{N\to\infty}\,\frac{1}{N}\sum_{n=1}^{N}\,\lambda(n)\,=\,0.

(1.4)

This is a well-known equivalent form of the Prime Number Theorem.¹¹1The 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].

1.1 also recovers other results in multiplicative number theory. For instance, by considering $f(n)=\zeta^{n}$ where $\zeta\neq 1$ is a $m$ -th root of unity, we obtain from (1.3) that

\lim_{N\to\infty}\,\frac{1}{N}\sum_{n=1}^{N}\,\zeta^{\Omega(n)}\,=\,0.

This implies a theorem of Pillai and Selberg [Pil40, Sel39], which says that for $m\in\mathbb{N}$ and $r\in\{0,1,\ldots,m-1\}$ the set $\{n\in\mathbb{N}:\Omega(n)\equiv r\bmod m\}$ has asymptotic density $1/m$ . In a similar vein, (1.3) applied to sequences of the form $f(n)=e^{2\pi i\alpha n}$ for $\alpha\in\mathbb{R}\backslash\mathbb{Q}$ yields a classical result of Erdős and Delange (see [Erd46, p. 2, lines 4–5] and [Del58]), asserting that $(\Omega(n)\alpha)_{n\in\mathbb{N}}$ is uniformly distributed mod $1$ for any irrational $\alpha$ . 1.1 also recovers several results recently obtained by the author in [BR20], including [BR20, Theorem A].

The proof of 1.1 is self-contained (with the exception of Stirling’s approximation formula used in Section 3 without a proof) and was inspired by the author’s work 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 $\mathcal{P}$ one has

\limsup_{N\to\infty}\,\frac{1}{N}\sum_{n=1}^{N}\,\Biggl{|}\,\sum_{p\in\mathcal{P}}1_{p\mid n}-\sum_{p\in\mathcal{P}}\text{\small$\dfrac{1}{p}$}\,\Biggr{|}^{2}\,=\,{\rm O}\Biggl{(}\sum_{p\in\mathcal{P}}\text{\small$\dfrac{1}{p}$}\Biggr{)},

(2.1)

where $1_{p\mid n}$ 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 $\{1,2,\ldots,N\}$ as a discrete probability space (with normalized counting measure as the probability measure) and $\sum_{p\in\mathcal{P}}1_{p\mid n}$ as a random variable on this space, (2.1) says that for large $N$ the expected number of primes in $\mathcal{P}$ that divide a ”randomly chosen” $n\in\{1,\ldots,N\}$ approximately equals $\sum_{p\in\mathcal{P}}1/p$ , with a standard deviation on the scale of $(\sum_{p\in\mathcal{P}}1/p)^{1/2}$ .

An important role in our proof of 1.1 is played by a generalization of (2.1) where the finite set of primes $\mathcal{P}$ is replaced by an arbitrary finite set of positive integers $B\subset\mathbb{N}$ .

Proposition 2.1.

Suppose $B\subset\mathbb{N}$ is finite and non-empty. Then

\frac{1}{N}\sum_{n=1}^{N}\,\Biggl{|}\,\sum_{q\in B}1_{q\mid n}-\sum_{q\in B}\text{\small$\dfrac{1}{q}$}\,\Biggr{|}^{2}\,=\,\sum_{q\in B}\sum_{q^{\prime}\in B}\text{\small$\dfrac{\Phi(q,q^{\prime})}{qq^{\prime}}$}+{\rm O}\biggl{(}\frac{|B|^{2}}{N}\biggr{)},

(2.2)

where $\Phi\colon\mathbb{N}{\mkern 0.0mu\times\mkern-0.3mu}\mathbb{N}\to\mathbb{N}\cup\{0\}$ is the function $\Phi(m,n)\coloneqq\gcd(m,n)-1$ .

Note that (2.2) implies (2.1), because for any finite set of primes $\mathcal{P}$ one has

\sum_{p\in\mathcal{P}}\sum_{p^{\prime}\in\mathcal{P}}\text{\small$\dfrac{\Phi(p,p^{\prime})}{pp^{\prime}}$}\,=\,\sum_{p\in\mathcal{P}}\text{\small$\dfrac{1}{p}$}\Bigl{(}1-\text{\small$\dfrac{1}{p}$}\Bigr{)}\,=\,{\rm O}\Biggl{(}\sum_{p\in\mathcal{P}}\text{\small$\dfrac{1}{p}$}\Biggr{)}.

Proof of 2.1.

Define $a\coloneqq\sum_{q\in B}1/q$ . By expanding the square in (2.2) we get $\frac{1}{N}\sum_{n=1}^{N}|\sum_{q\in B}1_{q\mid n}-\sum_{q\in B}{1}/{q}|^{2}=S_{1}-2aS_{2}+a^{2}$ , where $S_{1}\coloneqq\frac{1}{N}\sum_{n=1}^{N}\sum_{q,q^{\prime}\in B}1_{q\mid n}1_{q^{\prime}\mid n}$ and $S_{2}\coloneqq\frac{1}{N}\sum_{n=1}^{N}\sum_{q\in B}1_{q\mid n}$ . Since $\frac{1}{N}\sum_{n=1}^{N}1_{q\mid n}=1/q+{\rm O}({1}/{N})$ , we obtain $S_{2}=a+{\rm O}({|B|}/{N})$ . On the other hand, $\frac{1}{N}\sum_{n=1}^{N}1_{q\mid n}1_{q^{\prime}\mid n}=\gcd(q,q^{\prime})/qq^{\prime}+{\rm O}({1}/{N})$ implies $S_{1}=\sum_{q\in B}\sum_{q^{\prime}\in B}\Phi(q,q^{\prime})/qq^{\prime}+a^{2}+{\rm O}(|B|^{2}/N)$ . Substituting these estimates into $S_{1}-2aS_{2}+a^{2}$ 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 $A\subset\mathbb{N}$ and an arithmetic function $f\colon A\to\mathbb{C}$ , we denote the Cesàro average and the logarithmic average of $f$ over $A$ respectively by

\underset{n\in A}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,f(n)\coloneqq\,\frac{1}{|A|}\sum_{n\in A}f(n)\qquad\text{and }\qquad\underset{n\in A}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,f(n)\coloneqq\,\frac{\sum_{n\in A}\,{f(n)}/{n}}{\sum_{n\in A}\,{1}/{n}}.

Also let $[x]$ abbreviate the set $\{1,2,\ldots,\lfloor x\rfloor\}$ . After dividing both sides of (2.2) by $(\sum_{q\in B}1/q)^{2}$ , we obtain the following equivalent version of it expressed in terms of averages:

\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\left|\,\underset{q\in B}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\big{(}q1_{q\mid n}-1\big{)}\,\right|^{2}\,=\leavevmode\nobreak\ \underset{q\in B}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{q^{\prime}\in B}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\Phi(q,q^{\prime})+{\rm O}\biggl{(}\text{\small$\dfrac{|B|^{2}}{N}$}\biggr{)}.

(2.3)

The following proposition is our main technical result. By combining it with (2.3), we will be able to finish the proof of 1.1 rather quickly.

Proposition 2.2.

For all $\eta>0$ , there exists $k_{0}\in\mathbb{N}$ such that for all $k\geqslant k_{0}$ there exist two finite, non-empty sets $B_{1},B_{2}\subset\mathbb{N}$ with the following properties:

(a)

all elements in $B_{1}$ are primes and all elements in $B_{2}$ are a product of exactly $k$ primes;
(b)

the sets $B_{1}$ and $B_{2}$ have the same cardinality and if $B_{1}=\{p_{1}<\ldots<p_{t}\}$ and $B_{2}=\{q_{1}<\ldots<q_{t}\}$ then $(1-\eta)p_{j}\leqslant q_{j}\leqslant(1+\eta)p_{j}$ holds for all $j=1,\ldots,t$ ;
(c)

$\mathbb{E}^{\log}_{m\in B_{i}}\mathbb{E}^{\log}_{n\in B_{i}}\Phi(m,n)\leqslant\eta$ for $i=1,2$ , where $\Phi$ is as in 2.1.

Proof of 1.1 assuming 2.2.

Fix $\eta>0$ and let $k_{0}\in\mathbb{N}$ be as guaranteed by 2.2. This means that for every $k\geqslant k_{0}$ we can find two finite and non-empty sets $B_{1},B_{2}\subset\mathbb{N}$ satisfying properties (a), (b), and (c). For any $g\colon\mathbb{N}\to\mathbb{C}$ with $|g(n)|\leqslant 1$ we thus have

	$\displaystyle\left\|\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(n))-\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(qn))\right\|^{2}$	$\displaystyle=\left\|\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(n))-\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,q1_{q\mid n}g(\Omega(n))\right\|^{2}+{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)}$
		$\displaystyle\,\leqslant\,\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\left\|\,\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\big{(}1-q1_{q\mid n}\big{)}\,\right\|^{2}+\,{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)}$
		$\displaystyle\,\leqslant\,\eta\,+\,{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)},$

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 $\Omega(nq)=\Omega(n)+\Omega(q)$ , we get

\displaystyle\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n))

\displaystyle\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+\Omega(q))\leavevmode\nobreak\ +\leavevmode\nobreak\ {\rm O}\left(\eta^{1/2}+N^{-1/2}\right).

(2.4)

An analogous calculation carried out with $B_{1}$ in place of $B_{2}$ and $g(n+k-1)$ in place of $g(n)$ yields

\displaystyle\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k-1)

\displaystyle\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{p\in B_{1}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{p}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+\Omega(p)+k-1)\leavevmode\nobreak\ +\leavevmode\nobreak\ {\rm O}\left(\eta^{1/2}+N^{-1/2}\right).

(2.5)

Recall that $B_{1}$ consists only of primes and $B_{2}$ only of $k$ -almost primes, which means $\Omega(p)=1$ for all $p\in B_{1}$ and $\Omega(q)=k$ for all $q\in B_{2}$ . This allows us to rewrite (2.4) and (2.5) as

	$\displaystyle\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n))$	$\displaystyle\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k)\leavevmode\nobreak\ +\leavevmode\nobreak\ {\rm O}\left(\eta^{1/2}+N^{-1/2}\right),$		(2.6)
	$\displaystyle\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k-1)$	$\displaystyle\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{q\in B_{1}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k)\leavevmode\nobreak\ +\leavevmode\nobreak\ {\rm O}\left(\eta^{1/2}+N^{-1/2}\right).$		(2.7)

Finally, let $B_{1}=\{p_{1}<\ldots<p_{t}\}$ and $B_{2}=\{q_{1}<\ldots<q_{t}\}$ be enumerations of $B_{1}$ and $B_{2}$ . Since $(1-\eta)p_{j}\leqslant q_{j}\leqslant(1+\eta)p_{j}$ , it follows that $\mathbb{E}_{n\in[N/p_{j}]}g(\Omega(n)+k)=\mathbb{E}_{n\in[N/q_{j}]}g(\Omega(n)+k)+{\rm O}(\eta)$ . Taking logarihtmic averages over $B_{1}=\{p_{1},\ldots,p_{t}\}$ and $B_{2}=\{q_{1},\ldots,q_{t}\}$ (cf. [BR20, Lemma 2.3]) leaves us with

\underset{p\in B_{1}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{p}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k)\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k)\leavevmode\nobreak\ +\leavevmode\nobreak\ {\rm O}(\eta).

(2.8)

From (2.6), (2.7), and (2.8) it follows that $\mathbb{E}_{n\in[N]}g(\Omega(n))=\mathbb{E}_{n\in[N]}g(\Omega(n)+k-1)+{\rm O}(\eta^{1/2}+N^{-1/2})$ . This holds for all $k\geqslant k_{0}$ and hence

\displaystyle\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+k)\leavevmode\nobreak\ =\leavevmode\nobreak\ \underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\,g(\Omega(n)+l)\,+\,{\rm O}\left(\eta^{1/2}+N^{-1/2}\right)

(2.9)

for all $k,l\geqslant k_{0}$ . Relation (1.3) now follows from (2.9) by taking $k=k_{0}$ , $l=k_{0}+1$ , $g(n)=f(n-k_{0})$ , and letting $\eta$ go to $0$ . ∎

3. Proof of 2.2

The starting point for our proof of 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 $(8^{x},8^{x+1}]$ , as well as a rough upper bound on the number of primes in $(8^{x},8^{x+\varepsilon}]$ for small $\varepsilon$ .

Proposition 3.1.

Let $\mathbb{P}$ be the set of primes. There are $x_{0}\geqslant 1$ and $\varepsilon_{0}>0$ such that

(i)

$|\mathbb{P}\cap(8^{x},8^{x+1}]|\geqslant\frac{8^{x}}{x}$ for all $x\geqslant x_{0}$ , and
(ii)

$|\mathbb{P}\cap(8^{x},8^{x+\varepsilon}]|\leqslant\frac{\sqrt{\varepsilon}\,8^{x}}{x}$ for all $x\geqslant x_{0}$ and $\varepsilon\in(0,\varepsilon_{0}]$ .

The ideas used in the proof of 3.1 are classical and date back to Chebyshev. We begin with the following lemma.

Lemma 3.2.

We have the asymptotic estimate

\big{|}\mathbb{P}\cap(1,x]\big{|}\,\geqslant\,\frac{x\log(2)}{\log x}+{\rm O}(1).

(3.1)

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 ${2n}\choose{n}$ . Observe that there are $\lfloor m/p\rfloor$ many numbers in $\{1,\ldots,m\}$ that are divisible by $p$ . Out of those, there are $\lfloor m/p^{2}\rfloor$ many divisible by $p^{2}$ , and out of those there are $\lfloor m/p^{3}\rfloor$ many divisible by $p^{3}$ , and so on. Therefore, if $\nu$ is the largest exponent for which $p^{\nu}\leqslant m$ , then the power of $p$ in $m!$ is equal to $\lfloor m/p\rfloor+\lfloor m/p^{2}\rfloor+\ldots+\lfloor m/p^{\nu}\rfloor.$ In light of this observation, we see that the multiplicity of a prime $p$ in the prime factorization of $\binom{2n}{n}=\frac{(2n)!}{(n!)^{2}}$ is given by the formula

\sum_{1\leqslant i\leqslant\nu_{p}}\,\lfloor 2n/p^{i}\rfloor-2\lfloor n/p^{i}\rfloor,

(3.2)

where $\nu_{p}$ is the largest exponent for which $p^{\nu_{p}}\leqslant 2n$ . Since $\lfloor 2n/p^{i}\rfloor-2\lfloor n/p^{i}\rfloor\leqslant 1$ , we can estimate $\sum_{i=1}^{\nu_{p}}\leavevmode\nobreak\ \lfloor 2n/p^{i}\rfloor-2\lfloor n/p^{i}\rfloor\leavevmode\nobreak\ \leqslant\leavevmode\nobreak\ \nu_{p}$ . This yields

\mbinom{2n}{n}\,\leqslant\,\prod_{p\in\mathbb{P}\cap(1,2n]}p^{\nu_{p}}\,\leqslant\,(2n)^{|\mathbb{P}\cap(1,2n]|},

which after taking logarithms leaves us with

\log\mbinom{2n}{n}\,\leqslant\,\log(2n){|\mathbb{P}\cap(1,2n]|}.

(3.3)

Stirling’s approximation formula implies that $\log(m!)=m\log(m)-m+{\rm O}(\log m)$ . This can now be used to finish the proof by approximating $\log{}\binom{2n}{n}$ with $2\log(2)n+{\rm O}(\log n)$ in (3.3). ∎

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 3.1. For the proof of part (ii) we also need an upper bound.

Lemma 3.3.

Define $\beta(\sigma)\coloneqq\sigma\log(\sigma)-(\sigma-1)\log(\sigma-1)$ . Then for all $1<\sigma\leqslant 16$ ,

\big{|}\mathbb{P}\cap(x,\sigma x]\big{|}\leavevmode\nobreak\ \leqslant\leavevmode\nobreak\ \frac{\beta(\sigma)\,x}{\log x}+{\rm O}(1).

(3.4)

Proof.

For convenience, let us write $\binom{\sigma x}{x}$ for the quantity $\binom{\lfloor\sigma x\rfloor}{\lfloor x\rfloor}$ . Observe that every prime number in the interval $(x,\sigma x]$ divides the number $\binom{\sigma x}{x}$ . Therefore, the number $\binom{\sigma x}{x}$ is greater or equal than $\prod_{p\in\mathbb{P}\cap(x,\sigma x]}p$ . Using $\prod_{p\in\mathbb{P}\cap(x,\sigma x]}p\geqslant x^{|\mathbb{P}\cap(x,\sigma x]|}$ and taking logarithms, we obtain

\log\mbinom{\sigma x}{x}\,\geqslant\,\log(x){|\mathbb{P}\cap(x,\sigma x]|}.

(3.5)

Similarly as in the proof of 3.2, we can now use Stirling’s approximation formula, $\log(m!)=m\log(m)-m+{\rm O}(\log m)$ , to estimate that

	$\displaystyle\log\mbinom{\sigma x}{x}$	$\displaystyle\,=\,\lfloor\sigma x\rfloor\log(\lfloor\sigma x\rfloor)-\lfloor x\rfloor\log(\lfloor x\rfloor)-(\lfloor\sigma x\rfloor-\lfloor x\rfloor)\log(\lfloor\sigma x\rfloor-\lfloor x\rfloor)+{\rm O}(\log x)$
		$\displaystyle\,=\,\sigma x\log(\sigma x)-x\log(x)-(\sigma-1)x\log((\sigma-1)x)+{\rm O}(\log x)$
		$\displaystyle\,=\,\sigma x\log(\sigma)-(\sigma-1)x\log(\sigma-1)+{\rm O}(\log x).$

Together with (3.5), this proves (3.4). ∎

Proof of 3.1.

The proof of part (ii) simply follows from 3.3 (applied with $\sigma=8^{\varepsilon}$ ) and the fact that the order of magnitude of $\sqrt{\varepsilon}$ is much larger than the order of magnitude of $\beta(8^{\varepsilon})$ as $\varepsilon$ tends to $0$ .

For the proof of part (i), we start by rewriting the interval $(8^{x},8^{x+1}]$ in the form $(8^{x},8^{x+1}]=(1,8^{x+1}]\backslash\bigcup_{0\leqslant n\leqslant 3x}(\tfrac{8^{x}}{2^{n+1}},\tfrac{8^{x}}{2^{n}}].$ 3.2 gives the estimate $|\mathbb{P}\cap(1,8^{x+1}]|\geqslant{8^{x+1}}/{3(x+1)}+{\rm O}(1)$ , whereas 3.3 (applied with $\sigma=2$ ) gives the estimate $|\mathbb{P}\cap({8^{x}}/{2^{n+1}},{8^{x}}/{2^{n}}]|\leqslant 8^{x}/{2^{n+1}x}+{\rm O}(1)$ . Therefore

\displaystyle\big{|}\mathbb{P}\cap(8^{x},8^{x+1}]\big{|}

\displaystyle\leavevmode\nobreak\ \geqslant\leavevmode\nobreak\ \frac{8^{x+1}}{3(x+1)}\,-\,\sum_{0\leqslant n\leqslant 3x}\frac{8^{x}}{2^{n+1}x}\,+\,{\rm O}(x)\leavevmode\nobreak\ \geqslant\leavevmode\nobreak\ \frac{8^{x+1}}{3(x+1)}-\frac{8^{x}}{x}\,+\,{\rm O}(x).

This implies that if $x_{0}$ is sufficiently large then $|\mathbb{P}\cap(8^{x},8^{x+1}]|\geqslant\frac{8^{x}}{x}$ for all $x\geqslant x_{0}$ . ∎

3.1 is the only number-theoretic component in our proof of 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 3.1 tells us that we can find a fair amount of primes in any interval of the form $(8^{n},8^{n+1}]$ . However, we will need a bit more control over where these primes are within this interval. The following proposition tells us that in $(8^{n},8^{n+1}]$ we can find two smaller intervals $(8^{x},8^{x+\delta}]$ and $(8^{y},8^{y+\delta}]$ which are not too close together but also not too far apart, and each containing a good amount of primes.

Lemma 3.4.

Let $x_{0}$ be as in 3.1. There exists $\varepsilon_{1}>0$ such that for all $\varepsilon\in(0,\varepsilon_{1}]$ and all $\delta\in(0,1)$ there exists $D=D(\varepsilon,\delta)\in(0,1)$ with the following property: For all $n\geqslant x_{0}$ there are $x,y\in[n,n+1)$ with $\varepsilon^{4}<y-x<\varepsilon$ such that

\big{|}\mathbb{P}\cap(8^{x},8^{x+\delta}]\big{|}\,\geqslant\,\frac{D8^{n}}{n},\qquad\text{and}\qquad\big{|}\mathbb{P}\cap(8^{y},8^{y+\delta}]\big{|}\,\geqslant\,\frac{D8^{n}}{n}.

Proof.

As guaranteed by 3.1, the number of primes in $(8^{n},8^{n+1}]$ is at least ${8^{n}}/{n}$ . Therefore, by the Pigeonhole Principle, for some $t\in[n,n+1)$ the number of primes in $(8^{t},8^{t+\varepsilon}]$ is at least $\varepsilon 8^{n}/2n$ . We can then cover the interval $(8^{t},8^{t+\varepsilon}]$ by $K\coloneqq\lceil\varepsilon^{-3}\rceil$ many smaller intervals in the following way:

(8^{t},8^{t+\varepsilon}]=\big{(}8^{t},8^{t+\varepsilon^{4}}\big{]}\cup\big{(}8^{t+\varepsilon^{4}},8^{t+2\varepsilon^{4}}\big{]}\cup\ldots\cup\big{(}8^{t+(K-1)\varepsilon^{4}},8^{t+K\varepsilon^{4}}\big{]}.

By 3.3, each of the intervals $(8^{t+i\varepsilon^{4}},8^{t+(i+1)\varepsilon^{4}}]$ contains at most ${\rm O}({\varepsilon^{2}8^{n}}/{n})$ many primes, whereby the whole interval $(8^{t},8^{t+\varepsilon}]$ contains at least ${\rm O}({\varepsilon 8^{n}}/{n})$ many primes. It follows that if $\varepsilon$ is chosen sufficiently small, say smaller than some threshold $\varepsilon_{1}$ , then we can find two non-consecutive $a,b\in\{0,1,\ldots,K-1\}$ such that the intervals $(8^{t+a\varepsilon^{4}},8^{t+(a+1)\varepsilon^{4}}]$ and $(8^{t+b\varepsilon^{4}},8^{t+(b+1)\varepsilon^{4}}]$ contain at least ${\rm O}({\varepsilon^{4}8^{n}}/{n})$ many primes. Using the Pigeonhole Principle once more we can then find for every $\delta\in(0,1)$ some $x\in[t+a\varepsilon^{4},t+(a+1)\varepsilon^{4})$ and some $y\in[t+b\varepsilon^{4},t+(b+1)\varepsilon^{4})$ such that the intervals $(8^{x},8^{x+\delta}]$ and $(8^{y},8^{y+\delta}]$ contain at least ${\rm O}({\delta\varepsilon^{4}8^{n}}/{n})$ many primes. Since $a$ and $b$ are non-consecutive, we have $y-x>\varepsilon^{4}$ , and since $x,y\in[t,t+\varepsilon)$ we have $y-x<\varepsilon$ . ∎

The final ingredient in our proof of 2.2 is a purely combinatorial lemma.

Lemma 3.5.

Fix $x_{0}\geqslant 1$ and $\varepsilon>0$ . Suppose $\mathcal{X}$ is a subset of $\mathbb{R}$ with the property that for every $n\geqslant x_{0}$ there exist $x,y\in\mathcal{X}\cap[n,n+1)$ with $\varepsilon^{4}<y-x<\varepsilon$ . Let $k\geqslant\lceil 2/\varepsilon^{4}\rceil$ . Then for all $n_{1},\ldots,n_{k}\in\{n\in\mathbb{N}:n\geqslant x_{0}\}$ there exist $z,z_{1},\ldots,z_{k}\in\mathcal{X}$ such that

(I)

$z_{i}\in[n_{i},n_{i}+1)$ for all $1\leqslant i\leqslant k$ ;
(II)

$z_{1}+\ldots+z_{k}\in[z,z+\varepsilon)$ .

Proof.

Let $n_{1},\ldots,n_{k}\in\{n\in\mathbb{N}:n\geqslant x_{0}\}$ . According to the hypothesis of the lemma, we can find for every $i=1,\ldots,k$ two numbers $x_{i},y_{i}\in[n_{i},n_{i}+1)\cap\mathcal{X}$ with $\varepsilon^{4}<y_{i}-x_{i}<\varepsilon$ . Define, for every $i=0,1,\ldots,k$ , the number $u_{i}\coloneqq x_{1}+\ldots+x_{i}+y_{i+1}+\ldots+y_{k}.$ Since $\varepsilon^{4}<y_{i}-x_{i}$ we have $u_{k}-u_{0}\geqslant k\varepsilon^{4}\geqslant 2$ . This implies that there exists some $z\in\mathcal{X}$ with $u_{0}<z<u_{k}$ . Since $u_{i+1}-u_{i}<\varepsilon$ , there is some $i_{0}\in\{0,1,\ldots,k\}$ such that $u_{i_{0}}\in[z,z+\varepsilon)$ . Setting $z_{i}\coloneqq x_{i}$ for $i\leqslant i_{0}$ and $z_{i}\coloneqq y_{i}$ for $i>i_{0}$ , we obtain $z_{1}+\ldots+z_{k}=u_{i_{0}}\in[z,z+\varepsilon)$ as desired. ∎

Proof of 2.2.

Let $\eta\in(0,1)$ be given. Let $x_{0}$ and $\varepsilon_{0}$ be as in 3.1 and let $\varepsilon_{1}$ be as in 3.4. Pick any $\varepsilon>0$ with $\varepsilon<\min\{\varepsilon_{0},\varepsilon_{1},{\log(1+\eta)}/{\log(64)}\}$ and set $k_{0}\coloneqq\lceil 2/\varepsilon^{4}\rceil$ . We claim that $k_{0}$ is as desired, meaning that for all $k\geqslant k_{0}$ there exist two finite, non-empty sets $B_{1},B_{2}\subset\mathbb{N}$ satisfying properties (a), (b), and (c).

To verify this claim, fix $k\geqslant k_{0}$ , set $\delta\coloneqq{\varepsilon}/{k}$ , and let $D=D(\varepsilon,\delta)$ be as in 3.4. Define

\mathcal{X}\,\coloneqq\,\Big{\{}x\geqslant x_{0}:\big{|}\mathbb{P}\cap(8^{x},8^{x+\delta}]\big{|}\geqslant{D8^{\lfloor x\rfloor}}/{\lfloor x\rfloor}\Big{\}},

and for every $x\in\mathcal{X}$ let $P_{x}$ be a subset of $\mathbb{P}\cap(8^{x},8^{x+\delta}]$ with $|P_{x}|=\lfloor{D8^{\lfloor x\rfloor}}/{\lfloor x\rfloor}\rfloor$ . The sets $P_{x}$ are the building blocks from which we will construct $B_{1}$ and $B_{2}$ .

According to 3.4 the set $\mathcal{X}$ satisfies the hypothesis of 3.5, which allows us to find for all $\vec{n}=(n_{1},\ldots,n_{k})\in\{n\in\mathbb{N}:n\geqslant x_{0}\}^{k}$ numbers $z_{\vec{n}},z_{1,\vec{n}},\ldots,z_{k,\vec{n}}\in\mathcal{X}$ such that

(I)

$z_{i,\vec{n}}\in[n_{i},n_{i}+1)$ for all $1\leqslant i\leqslant k$ , and
(II)

$z_{1,\vec{n}}+\ldots+z_{k,\vec{n}}\in[z_{\vec{n}},z_{\vec{n}}+\varepsilon)$ .

Note that Property (I) and the definition of $P_{x}$ imply

\displaystyle|P_{z_{i,\vec{n}}}|\,=\,\left\lfloor{D8^{n_{i}}}/{n_{i}}\right\rfloor.

(3.6)

Next, let $N=N(D,\eta,k)$ be a constant that is to be determined later, and define sets $A_{1},\ldots,A_{k}\subset\{n\in\mathbb{N}:n\geqslant x_{0}\}$ in the following way: Pick any $s_{1}\in\mathbb{N}$ with $s_{1}>\max\{x_{0},2k\}$ and let $A_{1}$ be any finite subset of $s_{1}\mathbb{N}=\{s_{1}n:n\in\mathbb{N}\}$ with $\sum_{n\in A_{1}}1/n\geqslant N$ . Then, assuming $A_{i}$ has already been found, take any $s_{i+1}>\max(A_{1}+\ldots+A_{i})$ and let $A_{i+1}$ be any finite subset of $s_{i+1}\mathbb{N}$ with $\sum_{n\in A_{i+1}}1/n\geqslant N$ . Following this procedure until $i=k$ , we end up with a family of finite sets $A_{1},\ldots,A_{k}$ with the following property:

(A)

For any $(n_{1},\ldots,n_{k})\neq(n_{1}^{\prime},\ldots,n_{k}^{\prime})\in A_{1}\times\ldots\times A_{k}$ the distance between $n_{1}+\ldots+n_{k}$ and $n_{1}^{\prime}+\ldots+n_{k}^{\prime}$ is at least $2k$ .

We are now ready to define the sets $B_{1}$ and $B_{2}$ . The set $B_{2}$ is defined as

\displaystyle B_{2}\coloneqq\bigcup_{\vec{n}\in A_{1}\times\ldots\times A_{k}}P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}.

(3.7)

According to (3.6) we have

\big{|}P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}\big{|}\,\leqslant\,\frac{D^{k}8^{n_{1}+\ldots+n_{k}}}{n_{1}\cdot\ldots\cdot n_{k}}\,\leqslant\,\frac{D^{k}8^{n_{1}+\ldots+n_{k}}}{n_{1}+\ldots+n_{k}}\,\leqslant\,\frac{D8^{\lfloor z_{\vec{n}}\rfloor}}{\lfloor z_{\vec{n}}\rfloor},

where the last inequality follows from $D^{k}\leqslant D$ and $n_{1}+\ldots+n_{k}\leqslant\lfloor z_{\vec{n}}\rfloor$ . Therefore $|P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}|\,\leqslant\,|P_{z_{\vec{n}}}|$ , which guarantees the existence of a set $Q_{\vec{n}}\subset P_{z_{\vec{n}}}$ with $|Q_{\vec{n}}|=|P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}|$ . Define $B_{1}$ to be

\displaystyle B_{1}\coloneqq\bigcup_{\vec{n}\in A_{1}\times\ldots\times A_{k}}Q_{\vec{n}}.

(3.8)

It remains to show that $B_{1}$ and $B_{2}$ satisfy properties (a), (b), and (c).

Proof that $B_{1}$ and $B_{2}$ satisfy (a).

By construction, $B_{1}$ consists only of primes and $B_{2}$ only of numbers that are a product of $k$ primes. ∎

Proof that $B_{1}$ and $B_{2}$ satisfy (b).

By definition we have $P_{z_{i,\vec{n}}}\subset(8^{z_{i,\vec{n}}},8^{z_{i,\vec{n}}+\delta}]$ , and so

	$\displaystyle P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}$	$\displaystyle\,\subset\,(8^{z_{1,\vec{n}}},8^{z_{1,\vec{n}}+\delta}]\cdot\ldots\cdot(8^{z_{k,\vec{n}}},8^{z_{k,\vec{n}}+\delta}]$
		$\displaystyle\,\subset\,(8^{z_{1,\vec{n}}+\ldots+z_{k,\vec{n}}},8^{z_{1,\vec{n}}+\ldots+z_{k,\vec{n}}+k\delta}]$
		$\displaystyle\,\subset\,(8^{z_{\vec{n}}},8^{z_{\vec{n}}+2\varepsilon}],$

where the last inclusion follows from Property (II) and the fact that $k\delta\leqslant\varepsilon$ . Also by definition we have $Q_{\vec{n}}\subset(8^{z_{\vec{n}}},8^{z_{\vec{n}}+\delta}]\subset(8^{z_{\vec{n}}},8^{z_{\vec{n}}+2\varepsilon}]$ . Using the fact that $|n_{1}+\ldots+n_{k}-z_{\vec{n}}|\leqslant k+1$ and Property (A), we conclude that $(P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}})\cap(P_{z_{1,\vec{n}^{\prime}}}\cdot\ldots\cdot P_{z_{k,\vec{n}^{\prime}}})=\emptyset$ and $Q_{\vec{n}}\cap Q_{\vec{n}^{\prime}}=\emptyset$ whenever $\vec{n}\neq\vec{n}^{\prime}$ . Since $|P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}|=|Q_{\vec{n}}|$ , we immediately get that $B_{1}$ and $B_{2}$ have the same cardinality. Moreover, both $Q_{\vec{n}}$ and $P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}$ belong to the interval $(8^{z_{\vec{n}}},8^{z_{\vec{n}}+2\varepsilon}]$ , which implies that the ratio between any element in $P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}$ and any element in $Q_{\vec{n}}$ lies between $8^{-2\varepsilon}$ and $8^{2\varepsilon}$ . Since $8^{2\varepsilon}\leqslant(1+\eta)$ , it follows that for enumerations $P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}=\{q_{1}<\ldots<q_{r}\}$ and $Q_{\vec{n}}=\{p_{1}<\ldots<p_{r}\}$ we have $(1-\eta)p_{j}\leqslant q_{j}\leqslant(1+\eta)p_{j}$ for all $j=1,\ldots,r$ . This property now easily extends to enumerations of $B_{1}$ and $B_{2}$ . ∎

Proof that $B_{1}$ and $B_{2}$ satisfy (c).

First, let us show $\mathbb{E}^{\log}_{p\in B_{1}}\mathbb{E}^{\log}_{p^{\prime}\in B_{1}}\Phi(p,p^{\prime})\leqslant\eta$ . Since $B_{1}$ consists only of primes, we have $\Phi(p,p^{\prime})=0$ unless $p=p^{\prime}$ . Therefore

\displaystyle\underset{p\in B_{1}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{p^{\prime}\in B_{1}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\Phi(p,p^{\prime})\,=\,\frac{\sum_{p,p^{\prime}\in B_{1}}\frac{\Phi(p,p^{\prime})}{pp^{\prime}}}{\sum_{p,p^{\prime}\in B_{1}}\frac{1}{pp^{\prime}}}\,\leqslant\,\frac{\sum_{p\in B_{1}}\frac{1}{p}}{\sum_{p,p^{\prime}\in B_{1}}\frac{1}{pp^{\prime}}}\,=\,\frac{1}{\sum_{p\in B_{1}}\frac{1}{p}}.

(3.9)

Note, in view of (3.8) we have $\sum_{p\in B_{1}}1/p=\sum_{\vec{n}\in A_{1}\times\ldots\times A_{k}}\sum_{p\in Q_{\vec{n}}}1/p$ and since any element in $Q_{\vec{n}}$ is smaller than $8^{n_{1}+\ldots+n_{k}+k+1}$ we have

\sum_{p\in B_{1}}\text{\small$\dfrac{1}{p}$}\,\geqslant\,\sum_{\vec{n}=(n_{1},\ldots,n_{k})\in A_{1}\times\ldots\times A_{k}}\text{\small$\dfrac{|Q_{\vec{n}}|}{8^{k+1}8^{n_{1}+\ldots+n_{k}}}$}.

Then, using (3.6) and $|Q_{\vec{n}}|=|P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}|$ we can estimate $|Q_{\vec{n}}|\geqslant D^{k}8^{-k}8^{n_{1}+\ldots+n_{k}}/(n_{1}\cdot\ldots\cdot n_{k})$ , which implies

\displaystyle\sum_{p\in B_{1}}\text{\small$\dfrac{1}{p}$}\,\geqslant\,\sum_{(n_{1},\ldots,n_{k})\in A_{1}\times\ldots\times A_{k}}\text{\small$\dfrac{D^{k}}{8^{2k+1}\,n_{1}\cdot\ldots\cdot n_{k}}$}\,\geqslant\,\text{\small$\dfrac{D^{k}N^{k}}{8^{2k+1}}$}.

(3.10)

Thus, if $N$ was chosen sufficiently large, then $\mathbb{E}^{\log}_{p\in B_{1}}\mathbb{E}^{\log}_{p^{\prime}\in B_{1}}\Phi(p,p^{\prime})\leqslant\eta$ follows from combining (3.9) and (3.10).

Next, let us show $\mathbb{E}^{\log}_{q\in B_{2}}\mathbb{E}^{\log}_{q^{\prime}\in B_{2}}\Phi(q,q^{\prime})\leqslant\eta$ . In view of (3.7) we have

\displaystyle\sum_{q,q^{\prime}\in B_{2}}\text{\small$\dfrac{\Phi(q,q^{\prime})}{qq^{\prime}}$}\leavevmode\nobreak\ =\leavevmode\nobreak\ \sum_{\vec{n},\vec{n}^{\prime}\in A_{1}\times\ldots\times A_{k}}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{q\in P_{z_{1,\vec{n}}}\cdots P_{z_{k,\vec{n}}}}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{q^{\prime}\in P_{z_{1,\vec{n}^{\prime}}}\cdots P_{z_{k,\vec{n}^{\prime}}}}\text{\small$\dfrac{\Phi(q,q^{\prime})}{qq^{\prime}}$}.

(3.11)

If $q\in P_{z_{1,\vec{n}}}\cdot\ldots\cdot P_{z_{k,\vec{n}}}$ and $q^{\prime}\in P_{z_{1,\vec{n}^{\prime}}}\cdot\ldots\cdot P_{z_{k,\vec{n}^{\prime}}}$ are coprime then $\Phi(q,q^{\prime})=0$ . Hence, such a pair does not contribute to (3.11). On the other hand, if $q$ and $q^{\prime}$ are not coprime then there must exist a finite non-empty set $F\subset\{1,\ldots,k\}$ , a number $u\in\prod_{i\in F}P_{z_{i,\vec{n}^{\prime}}}$ , and a number $u^{\prime}\in\prod_{i\notin F}P_{z_{i,\vec{n}^{\prime}}}$ such that $q^{\prime}=uu^{\prime}$ and $\gcd(q,q^{\prime})=u$ ; note that this only happens when $n_{i}=n_{i}^{\prime}$ for all $i\in F$ because $P_{z_{i,\vec{n}}}$ and $P_{z_{i,\vec{n}^{\prime}}}$ are disjoint otherwise. In this case, we have

\frac{\Phi(q,q^{\prime})}{qq^{\prime}}\,=\,\frac{u-1}{qq^{\prime}}\,\leqslant\,\frac{1}{qu^{\prime}},

and therefore²²2In the published version of this paper the restriction $n_{i}=n_{i}^{\prime}$ , $i\in F$ , 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.

\displaystyle\sum_{q,q^{\prime}\in B_{2}}\text{\small$\dfrac{\Phi(q,q^{\prime})}{qq^{\prime}}$}

\displaystyle\leavevmode\nobreak\ \leqslant\leavevmode\nobreak\ \sum_{F\subset\{1,\ldots,k\}\atop F\neq\emptyset}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{\vec{n},\vec{n}^{\prime}\in A_{1}\times\ldots\times A_{k}\atop{n_{i}=n_{i}^{\prime},\leavevmode\nobreak\ i\in F}}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{q\in P_{z_{1,\vec{n}}}\cdots P_{z_{k,\vec{n}}}}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{u^{\prime}\in\prod_{i\notin F}P_{z_{i,\vec{n}^{\prime}}}}\text{\small$\dfrac{1}{qu^{\prime}}$}.

(3.12)

Next, we can use $P_{z_{i,\vec{n}}}\subset(8^{n_{i}},8^{n_{i}+1}]$ and $P_{z_{i,\vec{n}^{\prime}}}\subset(8^{n_{i}^{\prime}},8^{n_{i}^{\prime}+1}]$ to deduce that

	$\displaystyle\sum_{q\in P_{z_{1,\vec{n}}}\cdots P_{z_{k,\vec{n}}}}\leavevmode\nobreak\ \sum_{u^{\prime}\in\prod_{i\notin F}P_{z_{i,\vec{n}^{\prime}}}}\text{\small$\dfrac{1}{qu^{\prime}}$}$	$\displaystyle\,\leqslant\,\sum_{q\in P_{z_{1,\vec{n}}}\cdots P_{z_{k,\vec{n}}}}\leavevmode\nobreak\ \sum_{u^{\prime}\in\prod_{i\notin F}P_{z_{i,\vec{n}^{\prime}}}}\left(\prod_{i=1}^{k}\text{\small$\dfrac{1}{8^{n_{i}}}$}\right)\left(\prod_{i\notin F}\text{\small$\dfrac{1}{8^{n_{i}^{\prime}}}$}\right)$
		$\displaystyle\,=\,\left(\prod_{i=1}^{k}\text{\small$\dfrac{\|P_{z_{i,\vec{n}}}\|}{8^{n_{i}}}$}\right)\left(\prod_{i\notin F}\text{\small$\dfrac{\|P_{z_{i,\vec{n}^{\prime}}}\|}{8^{n_{i}^{\prime}}}$}\right).$

Thereafter, using (3.6), we get

\left(\prod_{i=1}^{k}\text{\small$\dfrac{|P_{z_{i,\vec{n}}}|}{8^{n_{i}}}$}\right)\left(\prod_{i\notin F}\text{\small$\dfrac{|P_{z_{i,\vec{n}^{\prime}}}|}{8^{n_{i}^{\prime}}}$}\right)\,\leqslant\,\left(\prod_{i=1}^{k}\text{\small$\dfrac{D}{n_{i}}$}\right)\left(\prod_{i\notin F}\text{\small$\dfrac{D}{n_{i}^{\prime}}$}\right).

All together, this implies that

	$\displaystyle\sum_{q,q^{\prime}\in B_{2}}\text{\small$\dfrac{\Phi(q,q^{\prime})}{qq^{\prime}}$}$	$\displaystyle\leavevmode\nobreak\ \leqslant\leavevmode\nobreak\ \sum_{F\subset\{1,\ldots,k\}\atop F\neq\emptyset}\leavevmode\nobreak\ \leavevmode\nobreak\ \sum_{\vec{n},\vec{n}^{\prime}\in A_{1}\times\ldots\times A_{k}\atop{n_{i}=n_{i}^{\prime},\leavevmode\nobreak\ i\in F}}\leavevmode\nobreak\ \leavevmode\nobreak\ \left(\prod_{i=1}^{k}\text{\small$\dfrac{D}{n_{i}}$}\right)\left(\prod_{i\notin F}\text{\small$\dfrac{D}{n_{i}^{\prime}}$}\right)$
		$\displaystyle\leavevmode\nobreak\ =\leavevmode\nobreak\ \sum_{F\subset\{1,\ldots,k\}\atop F\neq\emptyset}\leavevmode\nobreak\ {D^{2k-\|F\|}}\left(\prod_{i=1}^{k}\sum_{n\in A_{i}}\text{\small$\dfrac{1}{n_{i}}$}\right)\left(\prod_{i\notin F}\sum_{n\in A_{i}}\text{\small$\dfrac{1}{n_{i}^{\prime}}$}\right).$

A calculation similar to (3.10) shows that $\sum_{q\in B_{2}}\frac{1}{q}\geqslant\frac{D^{k}}{8^{3k}}\prod_{i=1}^{k}\sum_{n\in A_{i}}\frac{1}{n}$ . Putting everything together gives

	$\displaystyle\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\,\underset{q^{\prime}\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\Phi(q,q^{\prime})$	$\displaystyle\,=\,\frac{\sum_{q,q^{\prime}\in B_{2}}\frac{\Phi(q,q^{\prime})}{qq^{\prime}}}{\sum_{q,q^{\prime}\in B_{2}}\frac{1}{qq^{\prime}}}$
		$\displaystyle\,\leqslant\,\frac{\sum_{F\subset\{1,\ldots,k\},\,F\neq\emptyset}\leavevmode\nobreak\ {D^{2k-\|F\|}}\left(\prod_{i=1}^{k}\sum_{n\in A_{i}}\frac{1}{n}\right)\left(\prod_{i\notin F}\sum_{n\in A_{i}}\frac{1}{n}\right)}{\left(\frac{D^{k}}{8^{3k}}\prod_{i=1}^{k}\sum_{n\in A_{i}}\frac{1}{n}\right)\left(\frac{D^{k}}{8^{3k}}\prod_{i=1}^{k}\sum_{n\in A_{i}}\frac{1}{n}\right)}$
		$\displaystyle\,\leqslant\,\sum_{F\subset\{1,\ldots,k\}\atop F\neq\emptyset}\frac{{8^{6k}D^{-\|F\|}}}{\left(\prod_{i\in F}\sum_{n\in A_{i}}\frac{1}{n}\right)}.$

Finally, since $\sum_{n\in A_{i}}1/n\geqslant N$ , we see that $\mathbb{E}^{\log}_{q\in B_{2}}\mathbb{E}^{\log}_{q^{\prime}\in B_{2}}\Phi(q,q^{\prime})\leqslant\eta$ as long as $N$ was chosen sufficiently large. ∎

This completes the proof of 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 Foundation 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 actions, 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.
[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 Science] 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 historical 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.
[Lan99] E. Landau, Neuer Beweis der Gleichung $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}$ , Ph.D. thesis, 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 Zahlentheorie, Sitzungsberichte der Kaiserlichen Akadamie der Wissenschaften in Wien, Mathematisch-Naturwissenschaftliche Klasse 120 no. 2a (1911), 1–16.
[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.
[vM97] H. von Mangoldt, Beweis der Gleichung $\sum_{k=1}^{\infty}\frac{\mu(k)}{k}=0$ , Sitzungsberichte 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 number 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, Cambridge Studies in Advanced Mathematics 46, Cambridge University Press, Cambridge, 1995, Translated from the second French edition (1995) by C. B. Thomas.
[TMF00] G. Tenenbaum and M. Mendès France, The prime numbers and their distribution, 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
fkr@northwestern.edu

	$\displaystyle\left\|\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(n))-\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\underset{n\in[\nicefrac{{N}}{{q}}]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(qn))\right\|^{2}$	$\displaystyle=\left\|\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,g(\Omega(n))-\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,q1_{q\mid n}g(\Omega(n))\right\|^{2}+{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)}$
		$\displaystyle\,\leqslant\,\underset{n\in[N]}{{{\mathbb{E}}}^{\leavevmode\nobreak\ }}\,\left\|\,\underset{q\in B_{2}}{{{\mathbb{E}}}^{\text{\footnotesize log}}}\,\big{(}1-q1_{q\mid n}\big{)}\,\right\|^{2}+\,{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)}$
		$\displaystyle\,\leqslant\,\eta\,+\,{\rm O}\Big{(}\text{\small$\dfrac{1}{N}$}\Big{)},$

A new elementary proof of the Prime Number Theorem

Abstract

1. Introduction

Theorem 1.1.

2. The proof

Proposition 2.1.

Proof of 2.1.

Proposition 2.2.

Proof of 1.1 assuming 2.2.

3. Proof of 2.2

Proposition 3.1.

Lemma 3.2.

Proof.

Lemma 3.3.

Proof.

Proof of 3.1.

Lemma 3.4.

Proof.

Lemma 3.5.

Proof.

Proof of 2.2.

Proof that B1subscript𝐵1B_{1} and B2subscript𝐵2B_{2} satisfy (a).

Proof that B1subscript𝐵1B_{1} and B2subscript𝐵2B_{2} satisfy (b).

Proof that B1subscript𝐵1B_{1} and B2subscript𝐵2B_{2} satisfy (c).

Acknowledgments:

References

Proof that $B_{1}$ and $B_{2}$ satisfy (a).

Proof that $B_{1}$ and $B_{2}$ satisfy (b).

Proof that $B_{1}$ and $B_{2}$ satisfy (c).