betaFeedback

Reading v1View latest version

WorksMathematics

On an Erdős–Kac-type conjecture of Elliott

Abstract

Elliott and Halberstam proved that ∑p<n2ω(n−p)\sum_{p<n} 2^{ω(n-p)} is asymptotic to ϕ(n)ϕ(n). In analogy to the Erdős–Kac Theorem, Elliott conjectured that if one restricts the summation to primes pp such that ω(n−p)≤2log⁡log⁡n+λ(2log⁡log⁡n)1/2ω(n-p)\le 2 \log \log n+λ(2\log \log n)^{1/2} then the sum will be asymptotic to ϕ(n)∫−∞λe−t2/2dt/2πϕ(n)\int_{-\infty}^λ e^{-t^2/2}dt/\sqrt{2π}. We show that this conjecture follows from the Bombieri–Vinogradov Theorem. We further prove a related result involving Poisson–Dirichlet distribution, employing deeper lying level of distribution results of the primes.

Introduction

Let ω(m)=∑p∣m1\omega(m)=\sum_{p\mid m}1 be the prime divisor function. Elliott and Halberstam proved, using the Bombieri–Vinogradov Theorem, that [5]

∑p<n2ω(n−p)=ϕ(n)+O(nlog⁡n(log⁡log⁡n)2).(1)\sum_{p<n}2^{\omega(n-p)}=\phi(n)+O\left(\frac{n}{\log n}(\log\log n)^2\right). \tag*{(1)}

Here and later pp denotes a prime and ϕ\phi is Euler’s totient function. Elliott gave a talk at the meeting held in Urbana–Champaign, June 5–7, 2014, in memory of Paul and Felice Bateman, and Heini Halberstam, where he revisited his works with Halberstam. A telegraphic representation of the talk was published, which contains the following conjecture [6]: for each real λ\lambda,

∑p<nω(n−p)≤2log⁡log⁡n+λ(2log⁡log⁡n)1/22ω(n−p)⟶ϕ(n)2π∫−∞λe−u2/2 du,n→∞.(2)\sum_{\substack{p<n\\ \omega(n-p)\le2\log\log n+\lambda(2\log\log n)^{1/2}}}2^{\omega(n-p)} \longrightarrow\frac{\phi(n)}{\sqrt{2\pi}}\int_{-\infty}^{\lambda}e^{-u^2/2}\,du,\qquad n\to\infty. \tag*{(2)}

In this note we establish the conjecture.

Theorem 1. The asymptotic relation (1.2) holds for each real λ\lambda.

Without the 2ω(n−p)2^{\omega(n-p)} weight, variants of (1.2) were known for a long time [8, 10].

The strategy of proof of Theorem 1 is as follows. By work of Granville and Soundararajan [7], Theorem 1 can be reduced to understanding the distribution of 2ω(n−p)2^{\omega(n-p)} in arithmetic progressions. More precisely, for aa in a certain range it is required to consider

∑p<na∣n−p2ω(n−p).\sum_{\substack{p<n\\ a\mid n-p}}2^{\omega(n-p)}.

Similar as in [5], we obtain the asymptotics for this by using the distribution of primes in arithmetic progressions. However, as the required range of aa is large, larger than any bounded power of log⁡n\log n, the Bombieri–Vinogradov Theorem is no longer enough for this. The requirements for the replacement of the Bombieri–Vinogradov Theorem in the proof of Theorem 1 are strong:

  • We need to handle arithmetic progressions with modulus larger than n1/2n^{1/2}.

  • The weight on the moduli contains a component of non-negligible range to be handled with absolute value, arising from aa.

  • The progression to be considered is of the form p≡n(d)p\equiv n(d), where nn is of the same order of magnitude as the range of summation of pp.

We succeed by showing that these conditions can be handled by an application of a recent result due to Blomer, the second-named author, Li and Myerson [2].

Acknowledgements

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 851318).

Initial reduction

Theorem 1 can be restated as follows. Let XnX_n be a random variable taking the value m∈{1,2,…,n}m \in\{1,2,\ldots,n\} with probability proportional to 2ω(m)1n−m a prime2^{\omega(m)}1_{n-m\text{ a prime}}. Then

ω(Xn)−2log⁡log⁡n(2log⁡log⁡n)1/2⟶dN(0,1),n→∞,(3)\frac{\omega(X_n)-2\log\log n}{(2\log\log n)^{1/2}} \mathrel{\mathop{\longrightarrow}^{d}} N(0,1), \qquad n\to\infty, \tag*{(3)}

where N(0,1)N(0,1) is the standard gaussian distribution and the arrow indicates convergence in distribution. Let

Ck:=EXk,Ck′:=E∣X∣k,(4)\begin{aligned} C_k&:=\mathbb{E}X^k,\\ C'_k&:=\mathbb{E}|X|^k, \tag*{(4)} \end{aligned}

where X∼N(0,1)X\sim N(0,1). It is known that for (2.1) to hold it suffices to show that

E(ω(Xn)−2log⁡log⁡n(2log⁡log⁡n)1/2)k→Ck,n→∞,(5)\mathbb{E}\left(\frac{\omega(X_n)-2\log\log n}{(2\log\log n)^{1/2}}\right)^k\to C_k,\qquad n\to\infty, \tag*{(5)}

holds for every integer k≥1k\ge1 [4] 1. To approach (2.3) it is convenient to use a general result of Granville and Soundararajan [7].

Proposition 1 (Granville–Soundararajan). Let f:N→R≥0f:\mathbb{N}\to\mathbb{R}_{\geq0} be an arithmetic function and P\mathcal{P} be a squarefree integer. Let h:N→R≥0h:\mathbb{N}\to\mathbb{R}_{\geq0} be a multiplicative function with h(p)≤ph(p)\leq p for p∣Pp\mid\mathcal{P}. For a∣Pa\mid\mathcal{P} define rar_a via

∑m≤na∣mf(m)=h(a)a∑m≤nf(m)+ra.\sum_{\substack{m\leq n\\a\mid m}}f(m)=\frac{h(a)}{a}\sum_{m\leq n}f(m)+r_a.

Let

μP=∑p∣Ph(p)p,σP2=∑p∣Ph(p)p(1−h(p)p).\mu_{\mathcal{P}}=\sum_{p\mid\mathcal{P}}\frac{h(p)}{p},\qquad\sigma_{\mathcal{P}}^2=\sum_{p\mid\mathcal{P}}\frac{h(p)}{p}\left(1-\frac{h(p)}{p}\right).

Uniformly for 1≤k≤σP2/31\leq k\leq\sigma_{\mathcal{P}}^{2/3} we have

∑m≤nf(m)(∑p∣P1p∣m−μP)k=σPk∑m≤nf(m)(Ck+O((k3/2σP−1)1+2∣k∣Ck′))+O(μPk∑a∣Pω(a)≤k∣ra∣),\sum_{m\leq n}f(m)\left(\sum_{p\mid\mathcal{P}}1_{p\mid m}-\mu_{\mathcal{P}}\right)^k =\sigma_{\mathcal{P}}^k\sum_{m\leq n}f(m)\left(C_k+O\left((k^{3/2}\sigma_{\mathcal{P}}^{-1})^{1+2|k|}C'_k\right)\right)+O\left(\mu_{\mathcal{P}}^k\sum_{\substack{a\mid\mathcal{P}\\\omega(a)\leq k}}|r_a|\right),

where CkC_k, Ck′C'_k are given by (2.2).

For f≡1f\equiv1 this is [GS07, Prop. 3] and the proof works the same way for general f≥0f\geq0. Proposition 1 allows us to reduce the proof of Theorem 1 to a level-of-distribution result. We now state this reduction in somewhat more general terms, for the proof of Theorem 1 only the case θ=2\theta=2 will be used.

Corollary 1. Fix θ>0\theta>0 and k∈Nk\in\mathbb{N}. For every n≫1n\gg1 let YnY_n be a random variable taking values in {1,2,…,n}\{1,2,\ldots,n\} and set

In:=(exp⁡(exp⁡(log⁡1/3log⁡n)),nexp⁡(−log⁡1/3log⁡n)),T:=n1/log⁡1/3log⁡n,P=∏p∈Inp.I_n:=\left(\exp\left(\exp\left(\log^{1/3}\log n\right)\right),n^{\exp\left(-\log^{1/3}\log n\right)}\right),\qquad T:=n^{1/\log^{1/3}\log n},\qquad\mathcal{P}=\prod_{p\in I_n}p.

Suppose that, as n→∞n \to\infty

∑ω(a)≤kp∣a ⇒ p∈[2,T)∖Inμ2(a)=1P(a divides Yn)=o((log⁡log⁡n)k/2),(6)\sum_{\substack{\omega(a)\le k\\ p\mid a\ \Rightarrow\ p\in[2,T)\setminus\mathcal{I}_n\\ \mu^2(a)=1}} \mathbb{P}(a\text{ divides }Y_n)=o\left((\log\log n)^{k/2}\right), \tag*{(6)}
∑ω(a)≤ka∣P∣P(a divides Yn)−θω(a)a∣=o((log⁡log⁡n)−k/2).(7)\sum_{\substack{\omega(a)\le k\\ a\mid\mathcal{P}}}\left|\mathbb{P}(a\text{ divides }Y_n)-\frac{\theta^{\omega(a)}}{a}\right|=o\left((\log\log n)^{-k/2}\right). \tag*{(7)}

Then we have

E(ω(Yn)−θlog⁡log⁡n(θlog⁡log⁡n)1/2)k→Ck,n→∞.\mathbb{E}\left(\frac{\omega(Y_n)-\theta\log\log n}{(\theta\log\log n)^{1/2}}\right)^k\to C_k,\qquad n\to\infty.

Proof. We consider a given nn and apply Proposition 1 with f(m)=P(Yn=m)f(m)=\mathbb{P}(Y_n=m), h(m)=θω(m)h(m)=\theta^{\omega(m)}. Define

ωP(n)=∑p∣P1p∣n,μ=∑p∣Pθp,σ2=∑p∣Pθp(1−θp),ra=P(a divides Yn)−θω(a)a.\omega_{\mathcal{P}}(n)=\sum_{p\mid\mathcal{P}}1_{p\mid n},\qquad\mu=\sum_{p\mid\mathcal{P}}\frac{\theta}{p},\qquad\sigma^2=\sum_{p\mid\mathcal{P}}\frac{\theta}{p}\left(1-\frac{\theta}{p}\right),\qquad r_a=\mathbb{P}(a\text{ divides }Y_n)-\frac{\theta^{\omega(a)}}{a}.

We find that

∑m=1nP(Yn=m)(ωP(m)−μσ)k=∑m=1nP(Yn=m)(Ck+O(σ−1−2∣k∣))+O(μkσk∑ω(a)≤ka∣P∣ra∣)(8)\sum_{m=1}^{n}\mathbb{P}(Y_n=m)\left(\frac{\omega_{\mathcal{P}}(m)-\mu}{\sigma}\right)^k = \sum_{m=1}^{n}\mathbb{P}(Y_n=m)\left(C_k+O(\sigma^{-1-2\lvert k\rvert})\right) +O\left(\frac{\mu^k}{\sigma^k}\sum_{\substack{\omega(a)\le k\\a\mid\mathcal{P}}}|r_a|\right) \tag*{(8)}

holds for every fixed k≥1k\ge1. Since ∑p≤x1/p=log⁡log⁡x+O(1)\sum_{p\le x}1/p=\log\log x+O(1) we have

μ,σ2=θlog⁡log⁡n+O(log⁡1/3log⁡n).(9)\mu,\sigma^2=\theta\log\log n+O(\log^{1/3}\log n). \tag*{(9)}

We express (8) as

E(ωP(Yn)−μσ)k−Ck≪(log⁡log⁡n)−1/2+(log⁡log⁡n)k/2∑ω(a)≤ka∣P∣ra∣.(10)\mathbb{E}\left(\frac{\omega_{\mathcal{P}}(Y_n)-\mu}{\sigma}\right)^k-C_k \ll(\log\log n)^{-1/2}+(\log\log n)^{k/2}\sum_{\substack{\omega(a)\le k\\a\mid\mathcal{P}}}|r_a|. \tag*{(10)}

By (7), the right-hand side of (10) goes to 00 as n→∞n\to\infty. Let T=n1/log⁡1/3log⁡nT=n^{1/\log^{1/3}\log n}. We have the identity

ω(Yn)−θlog⁡log⁡n(θlog⁡log⁡n)1/2=A1A2+A3+A4+A5\frac{\omega(Y_n)-\theta\log\log n}{(\theta\log\log n)^{1/2}}=A_1A_2+A_3+A_4+A_5

for

A1=σ(θlog⁡log⁡n)1/2,A2=ωP(Yn)−μσ,A3=μ−θlog⁡log⁡n(θlog⁡log⁡n)1/2,A_1=\frac{\sigma}{(\theta\log\log n)^{1/2}},\qquad A_2=\frac{\omega_{\mathcal{P}}(Y_n)-\mu}{\sigma},\qquad A_3=\frac{\mu-\theta\log\log n}{(\theta\log\log n)^{1/2}},
A4=∑p∈[2,T)∖In1p∣Yn(θlog⁡log⁡n)1/2,A5=∑p∈[T,n]1p∣Yn(θlog⁡log⁡n)1/2.A_4=\frac{\sum_{p\in[2,T)\setminus\mathcal{I}_n}1_{p\mid Y_n}}{(\theta\log\log n)^{1/2}},\qquad A_5=\frac{\sum_{p\in[T,n]}1_{p\mid Y_n}}{(\theta\log\log n)^{1/2}}.

By (9), the constants A1A_1 and A3A_3 tend to $1andand0, respectively. We have observed\text{, respectively. We have observed}\mathbb{E}A_2^k\to C_k.Aninteger. An integermcanhaveatmostcan have at most\log m/\log Tprimefactorsofsizeatleastprime factors of size at leastT,implying, implyingA_5\ll (log⁡log⁡n)−1/6(\log\log n)^{-1/6} and so EA5k→0\mathbb{E}A_5^k \to0 as n→∞n \to\infty. To handle A4A_4 we use (6) which yields

EA4k=∑p1,…,pk∈[2,T)∖InP([p1,…,pk] divides Yn)(θlog⁡log⁡n)k/2≪(log⁡log⁡n)−k/2∑ω(a)<kp∣a⇒p∈[2,T)∖Inμ2(a)=1P(a divides Yn)=o(1).\begin{aligned} \mathbb{E}A_4^k=\frac{\sum_{p_1,\ldots,p_k\in[2,T)\setminus\mathcal{I}_n}\mathbb{P}([p_1,\ldots,p_k]\text{ divides }Y_n)}{(\theta\log\log n)^{k/2}} \\ &\ll(\log\log n)^{-k/2} \sum_{\substack{\omega(a)<k\\ p\mid a\Rightarrow p\in[2,T)\setminus\mathcal{I}_n\\ \mu^2(a)=1}}\mathbb{P}(a\text{ divides }Y_n)=o(1). \end{aligned}

We have shown E∣A3+A4+A5∣k→0\mathbb{E}|A_3+A_4+A_5|^k\to0 and E(A1A2)k→Ck\mathbb{E}(A_1A_2)^k\to C_k. By Hölder, (5) follows.

Corollary 1 may also be proved by generalizing an argument of Billingsley [3]. To deduce Theorem 1 from Corollary 1 we apply it with Yn=XnY_n=X_n and θ=2\theta=2. This reduces matters to the following lemma, whose proof we postpone to the next section.

Lemma 1. Let η>0\eta>0 be sufficiently small. Fix k≥1k\ge1. Then

∑1≤a≤nηω(a)≤k(a,n)=1μ2(a)∣P(a divides Xn)−2ω(a)a+2Daϕ(a)log⁡n∣≪k(log⁡log⁡n)Ok(1)log⁡n.\sum_{\substack{1\le a\le n^\eta\\ \omega(a)\le k\\ (a,n)=1}}\mu^2(a)\left|\mathbb{P}(a\text{ divides }X_n)-\frac{2^{\omega(a)}}{a}+\frac{2D_a}{\phi(a)\log n}\right| \ll_k\frac{(\log\log n)^{O_k(1)}}{\log n}.

where

Da:=∑d∣a∞g(d)log⁡d,g(d)=∏pk∥d(−1)k−1(p−2)(p−1)−k.D_a:=\sum_{d\mid a^\infty}g(d)\log d,\qquad g(d)=\prod_{p^k\parallel d}(-1)^{k-1}(p-2)(p-1)^{-k}.

We remark that the occurrence of the somewhat unexpected lower order term involving DaD_a may merit further study.

Proof of Theorem 1, assuming Lemma 1. By Corollary 1 it suffices to establish (6) and (7) (with θ=2\theta=2 and Yn=XnY_n=X_n). To deduce this from Lemma 1, we must explain how to deal with two issues: the condition (a,n)=1(a,n)=1 being present in the lemma, and the existence of the lower order term involving DaD_a. The contribution of aa-s with (a,n)>1(a,n)>1 to (6) is negligible since

P(a divides Xn)≪n−1+o(1)(11)\mathbb{P}(a\text{ divides }X_n)\ll n^{-1+o(1)} \tag*{(11)}

uniformly for aa with (a,n)>1(a,n)>1. The contribution of aa-s with (a,n)>1(a,n)>1 to (7) is negligible by (11) and by

∑ω(a)≤ka∣P(a,n)>12ω(a)a≪k(1+∑p∣P1p)k−1∑p∣n1p≪kexp⁡(−exp⁡(log⁡1/4log⁡n)).\sum_{\substack{\omega(a)\le k\\ a\mid\mathcal{P}\\ (a,n)>1}}\frac{2^{\omega(a)}}{a} \ll_k\left(1+\sum_{p\mid\mathcal{P}}\frac1p\right)^{k-1}\sum_{p\mid n}\frac1p \ll_k\exp\left(-\exp\left(\log^{1/4}\log n\right)\right).

where we used ∑p≤x1/p=log⁡log⁡x+O(1)\sum_{p\le x}1/p=\log\log x+O(1) and the fact that nn has ≪log⁡n\ll\log n prime factors. Next, we use ∑d∣a∞∣g(d)∣=2ω(a)−12∣a\sum_{d\mid a^\infty}|g(d)|=2^{\omega(a)-1}2^{|a} to bound DaD_a as

Da≪∑p∣ai≥1log⁡(pi)∑pi∣d∣a∞∣g(d)∣≪2ω(a)∑p∣ai≥1log⁡(pi)(p−2)(p−1)−i≪2ω(a)log⁡aD_a\ll\sum_{\substack{p\mid a\\ i\ge1}}\log(p^i)\sum_{p^i\mid d\mid a^\infty}|g(d)| \ll2^{\omega(a)}\sum_{\substack{p\mid a\\ i\ge1}}\log(p^i)(p-2)(p-1)^{-i} \ll2^{\omega(a)}\log a

and it follows that for any squarefree Q\mathcal{Q},

∑ω(a)≤ka∣Q∣Da∣ϕ(a)log⁡n≪k1log⁡n(1+∑p∣Q1p)k−1∑p∣Qlog⁡pp.\sum_{\substack{\omega(a)\le k\\ a\mid\mathcal{Q}}}\frac{|D_a|}{\phi(a)\log n} \ll_k\frac1{\log n}\left(1+\sum_{p\mid\mathcal{Q}}\frac1p\right)^{k-1}\sum_{p\mid\mathcal{Q}}\frac{\log p}{p}.

In the notation of Corollary 1, we specialize to Q=P\mathcal{Q}=\mathcal{P} and Q=∏p∈[2,T)∖Inp\mathcal{Q}=\prod_{p\in[2,T)\setminus\mathcal{I}_n}p and find DaD_a contributes negligibly to (6) and (7) by invoking ∑p≤xlog⁡pp=log⁡x+O(1)\sum_{p\le x}\frac{\log p}{p}=\log x+O(1) and ∑p≤x1p=log⁡log⁡x+O(1)\sum_{p\le x}\frac1p=\log\log x+O(1). ▫

Proof of Lemma 1

We shall work with arithmetic sums instead of probabilities. Given a≥1a \ge1 let

Aa(n):=∑p<n:a∣n−p2ω(n−p)A_a(n) := \sum_{\substack{p<n:\\a\mid n-p}} 2^{\omega(n-p)}

so that P(a divides Xn)=Aa(n)/A1(n)\mathbb{P}(a\text{ divides }X_n)=A_a(n)/A_1(n). Lemma 1 is equivalent to

∑1≤a≤nηω(a)≤k(a,n)=1μ2(a)∣Aa(n)−ϕ(n)(2ω(a)a−2Daϕ(a)log⁡n)∣≪knlog⁡n(log⁡log⁡n)Ok(1)(12)\sum_{\substack{1\le a\le n^\eta\\ \omega(a)\le k\\ (a,n)=1}} \mu^2(a)\left|A_a(n)-\phi(n)\left(\frac{2^{\omega(a)}}{a}-\frac{2D_a}{\phi(a)\log n}\right)\right| \ll_k \frac{n}{\log n}(\log\log n)^{O_k(1)} \tag*{(12)}

where we utilized (1.1), ϕ(n)≫n/log⁡log⁡n\phi(n)\gg n/\log\log n and ∑p≤x1/p=log⁡log⁡x+O(1)\sum_{p\le x}1/p=\log\log x+O(1) to replace A1(n)A_1(n) by ϕ(n)\phi(n). We write 2ω(n−p)2^{\omega(n-p)} as a divisor sum using 2ω(m)=∑d∣mμ2(d)2^{\omega(m)}=\sum_{d\mid m}\mu^2(d) and μ2(d)=∑e2∣dμ(e)\mu^2(d)=\sum_{e^2\mid d}\mu(e), separate the contribution of large square divisors, and apply divisor switching. We have

2ω(n)=∑n=dsμ2(d)=∑n=e2dsμ(e)=∑n=e2dse>Vμ(e)+∑n=e2dse≤Vd≤Dμ(e)+∑n=e2dse≤Vs<n/(e2D)μ(e)2^{\omega(n)}=\sum_{n=ds}\mu^2(d)=\sum_{n=e^2ds}\mu(e)=\sum_{\substack{n=e^2ds\\e>V}}\mu(e)+\sum_{\substack{n=e^2ds\\e\le V\\d\le D}}\mu(e)+\sum_{\substack{n=e^2ds\\e\le V\\s<n/(e^2D)}}\mu(e)

with the choices

V=n2η,D=n1/2+2η.V=n^{2\eta},\qquad D=n^{1/2+2\eta}.

Then, given a≥1a\ge1, we split Aa(n)A_a(n) accordingly:

Aa(n)=∑p<na∣n−p(∑n−p=e2dse>Vμ(e)+∑n−p=e2dse≤Vd≤Dμ(e)+∑n−p=e2dse≤Vs<(n−p)/(e2D)μ(e))=:S0(a)+S1(a)+S2(a).A_a(n)=\sum_{\substack{p<n\\a\mid n-p}}\left(\sum_{\substack{n-p=e^2ds\\e>V}}\mu(e)+\sum_{\substack{n-p=e^2ds\\e\le V\\d\le D}}\mu(e)+\sum_{\substack{n-p=e^2ds\\e\le V\\s<(n-p)/(e^2D)}}\mu(e)\right)=:S_0(a)+S_1(a)+S_2(a).

We write π(x;q,b)\pi(x;q,b) for the number of primes in [1,x][1,x] that are congruent to bb modulo qq. With this notation we have

S0(a)=∑V<e≤nd≤nμ(e)π(n−1;[de2,a],n),S_0(a)=\sum_{\substack{V<e\le\sqrt n\\d\le n}}\mu(e)\pi(n-1;[de^2,a],n),
S1(a)=∑e≤Vd≤Dμ(e)π(n−1;[de2,a],n),S_1(a)=\sum_{\substack{e\le V\\d\le D}}\mu(e)\pi(n-1;[de^2,a],n),
S2(a)=∑e≤Vs<(n−2)/(e2D)μ(e)π(n−se2D−1;[se2,a],n).S_2(a)=\sum_{\substack{e\le V\\s<(n-2)/(e^2D)}}\mu(e)\pi(n-se^2D-1;[se^2,a],n).

Write E(m;q,a)=π(m;q,a)−Li⁡(m)1(q,a)=1φ(a)E(m;q,a)=\pi(m;q,a)-\dfrac{\operatorname{Li}(m)1_{(q,a)=1}}{\varphi(a)} and accordingly

Si(a)=Mi(a)+Ei(a)(i=1,2).S_i(a)=M_i(a)+E_i(a)\qquad(i=1,2).

for

M1(a)=Li⁡(n−1)∑e≤Vd≤D(ed,n)=1μ(e)ϕ([de2,a]),E1(a)=∑e≤Vd≤D(ed,n)=1μ(e)E(n−1;[de2,a],n),M_1(a)=\operatorname{Li}(n-1)\sum_{\substack{e\le V\\ d\le D\\ (ed,n)=1}}\frac{\mu(e)}{\phi([de^2,a])},\qquad E_1(a)=\sum_{\substack{e\le V\\ d\le D\\ (ed,n)=1}}\mu(e)E(n-1;[de^2,a],n),
M2(a)=∑e≤Vd<(n−2)/(e2D)(de,n)=1μ(e)Li⁡(n−de2D−1)ϕ([de2,a]),E2(a)=∑e≤Vd<(n−2)/(e2D)(de,n)=1μ(e)E(n−de2D−1;[de2,a],n).M_2(a)=\sum_{\substack{e\le V\\ d<(n-2)/(e^2D)\\ (de,n)=1}}\frac{\mu(e)\operatorname{Li}(n-de^2D-1)}{\phi([de^2,a])},\qquad E_2(a)=\sum_{\substack{e\le V\\ d<(n-2)/(e^2D)\\ (de,n)=1}}\mu(e)E(n-de^2D-1;[de^2,a],n).

The rest of the proof is separated into four parts which together establish (3.1).

Lemma 2. We have ∑a≤nη∣S0(a)∣≪n1−η2\sum_{a\le n^\eta}|S_0(a)|\ll n^{1-\frac{\eta}{2}}.

Lemma 3. For every B>0B>0, ∑a≤nη∣E1(a)∣≪Bn(log⁡n)−B\sum_{a\le n^\eta}|E_1(a)|\ll_B n(\log n)^{-B}.

Lemma 4. For every B>0B>0, ∑a≤nη∣E2(a)∣≪Bn(log⁡n)−B\sum_{a\le n^\eta}|E_2(a)|\ll_B n(\log n)^{-B}.

Lemma 5. Fix k≥1k\ge1. Suppose a≤nηa\le n^\eta, ω(a)≤k\omega(a)\le k and μ2(a)=1\mu^2(a)=1. We have

M1(a)+M2(a)=ϕ(n)(2ω(a)a−2Daϕ(a)log⁡n)+Ok(n(log⁡log⁡n)2alog⁡n).M_1(a)+M_2(a)=\phi(n)\left(\frac{2^{\omega(a)}}{a}-\frac{2D_a}{\phi(a)\log n}\right)+O_k\left(\frac{n(\log\log n)^2}{a\log n}\right).

Proof of Lemma 2

The upper bound π(n−1;q,n)≤n/q\pi(n-1;q,n)\le n/q gives

∣S0(a)∣≤∑V<e≤nd≤n∑p<n[de2,a]∣n−p1≤∑e>Vd≤nn[de2,a]≤∑e>Vd≤nnde2≪nlog⁡nV,|S_0(a)|\le\sum_{\substack{V<e\le\sqrt n\\ d\le n}}\sum_{\substack{p<n\\ [de^2,a]\mid n-p}}1\le\sum_{\substack{e>V\\ d\le n}}\frac{n}{[de^2,a]}\le\sum_{\substack{e>V\\ d\le n}}\frac{n}{de^2}\ll\frac{n\log n}{V},

which implies the result.

Proof of Lemma 4

Isolating the contribution of (se2a,n)≠1(se^2a,n)\ne1, we have

∑a≤nη∣E2(a)∣=∑a≤nη∣∑e≤Vs<(n−2)/(e2D)μ(e)E(n−se2D−1;[se2,a],n)∣≤∑d≤n1/2−η(d,n)=1τ2(d)max⁡y≤n∣E(y;d,n)∣+no(1)\begin{aligned} \sum_{a\le n^\eta}|E_2(a)|=\sum_{a\le n^\eta}\left|\sum_{\substack{e\le V\\ s<(n-2)/(e^2D)}}\mu(e)E(n-se^2D-1;[se^2,a],n)\right| \\ &\le\sum_{\substack{d\le n^{1/2-\eta}\\ (d,n)=1}}\tau^2(d)\max_{y\le n}|E(y;d,n)|+n^{o(1)} \end{aligned}

by the divisor bound, where dd stands for [se2,a][se^2,a]. The conclusion follows by applying Cauchy–Schwarz and then the Bombieri–Vinogradov Theorem [5].

Proof of Lemma 3

As E1(a)E_1(a) requires us to break the 1/21/2-barrier in the level of distribution, it is the more delicate error term. We have

∑a≤nη∣E1(a)∣=∑a≤nη∣∑e≤Vμ(e)∑d≤DE(n−1;[de2,a],n)∣=∑a≤nη∣∑e≤Vb∣(ae)∞μ(e)∑d≤Db∣d(d/b,ae)=1E(n−1;[de2,a],n)∣=∑a≤nη∣∑e≤Vb∣(ae)∞μ(e)∑d′≤D/b(d′,ae)=1E(n−1;d′[be2,a],n)∣,\begin{aligned} \sum_{a\le n^\eta} |E_1(a)| = \sum_{a\le n^\eta} \left|\sum_{e\le V} \mu(e) \sum_{d\le D} E(n-1; [de^2,a],n)\right| \\ &= \sum_{a\le n^\eta} \left|\sum_{\substack{e\le V\\ b\mid(ae)^\infty}} \mu(e) \sum_{\substack{d\le D\\ b\mid d\\ (d/b,ae)=1}} E(n-1;[de^2,a],n)\right| \\ &= \sum_{a\le n^\eta} \left|\sum_{\substack{e\le V\\ b\mid(ae)^\infty}} \mu(e) \sum_{\substack{d'\le D/b\\ (d',ae)=1}} E(n-1;d'[be^2,a],n)\right|, \end{aligned}

where we substituted d′=dbd'=db. Let E1,1E_{1,1} denote the contribution to the above with b≤J=n2ηb\le J=n^{2\eta} and E1,2E_{1,2} the remainder. For E1,1E_{1,1} we apply the triangle inequality to arrive at

E1,1≤∑a≤nη∑e≤V∑b≤Jb∣(ae)∞μ2(e)∣∑d′≤D/b(d′,ae)=1E(n−1;d′[be2,a],n)∣.E_{1,1}\le\sum_{a\le n^\eta}\sum_{e\le V}\sum_{\substack{b\le J\\ b\mid(ae)^\infty}}\mu^2(e)\left|\sum_{\substack{d'\le D/b\\ (d',ae)=1}}E(n-1;d'[be^2,a],n)\right|.

To make the summation range of d′d' independent of the other variables, we apply a standard finer-than-dyadic decomposition on the summation over d′d' (see e.g. the proof of [1]). It thus suffices to estimate

E1,1,D′=∑a≤nη∑e≤V∑b≤Jb∣(ae)∞μ2(e)∣∑d′≤D′(d′,ae)=1E(n−1;d′[be2,a],n)∣E_{1,1,D'}=\sum_{a\le n^\eta}\sum_{e\le V}\sum_{\substack{b\le J\\ b\mid(ae)^\infty}}\mu^2(e)\left|\sum_{\substack{d'\le D'\\ (d',ae)=1}}E(n-1;d'[be^2,a],n)\right|

for any D′≤DD'\le D. If [be2,a]=g[be^2,a]=g then aa, bb and ee divide gg and the condition (d′,ae)=1(d',ae)=1 is equivalent to (d′,g)=1(d',g)=1. Thus

E1,1,D′≤∑g<n8ητ(g)3∣∑d′≤D′(d′,g)=1E(n−1;d′g,n)∣.E_{1,1,D'}\le\sum_{g<n^{8\eta}}\tau(g)^3\left|\sum_{\substack{d'\le D'\\ (d',g)=1}}E(n-1;d'g,n)\right|.

We quote [2] specialised to a2=c=c0=1a_2=c=c_0=1.

Proposition 2. There exists some absolute constant ϖ>0\varpi>0 such that the following holds. Let x≥2x\ge2, d,C∈Nd,C\in\mathbb{N}, a∈Z∖{0}a\in\mathbb{Z}\setminus\{0\} such that

Q≤x1/2+ϖ,∣a∣≤x1+ϖ,∣λd∣≪τ(d)CQ\le x^{1/2+\varpi},\qquad|a|\le x^{1+\varpi},\qquad|\lambda_d|\ll\tau(d)^C

hold. Then for any B>0B>0 we have

∑d≤xϖλdsup⁡wmodd(w,d)=1∣∑q≤Q(q,ad)=1(∑n≤xn≡a mod qn≡w mod dΛ(n)−1ϕ(qd)∑n≤x(n,qd)=1Λ(n))∣≪C,Bx(log⁡x)−B.\sum_{d\le x^\varpi}\lambda_d\sup_{\substack{w\bmod d\\ (w,d)=1}}\left|\sum_{\substack{q\le Q\\ (q,ad)=1}}\left(\sum_{\substack{n\le x\\ n\equiv a\ \mathrm{mod}\ q\\ n\equiv w\ \mathrm{mod}\ d}}\Lambda(n)-\frac{1}{\phi(qd)}\sum_{\substack{n\le x\\ (n,qd)=1}}\Lambda(n)\right)\right|\ll_{C,B}x(\log x)^{-B}.

As commented after [2], this holds for the characteristic function of the primes replacing the von Mangoldt function and so implies E1,1,D′≪Bn(log⁡n)−BE_{1,1,D'}\ll_B n(\log n)^{-B}. To prove Lemma 3, it remains to estimate E1,2E_{1,2} and we claim

E1,2≪n1−η/2.(13)E_{1,2}\ll n^{1-\eta/2}. \tag*{(13)}

By the trivial bound E(n−1;q,a)≪1+n/qE(n-1;q,a) \ll1+n/q we have

E1,2≪n∑a<nηe≤Vd′<D/J1d′∑J<b≤D/d′b∣(ae)∞1[be2,a].E_{1,2} \ll n \sum_{\substack{a<n^\eta\\ e\le V\\ d'<D/J}} \frac{1}{d'} \sum_{\substack{J<b\le D/d'\\ b\mid(ae)^\infty}} \frac{1}{[be^2,a]}.

Since [be2,a]=be2a/(be2,a)[be^2,a]=be^2a/(be^2,a), by dyadic decomposition to get (13) it suffices to show

∑a∼Ae∼V∑b∼Jb∣(ae)∞(be2,a)≪(2+log⁡AV)CεJεA2V(14)\sum_{\substack{a\sim A\\ e\sim V}} \sum_{\substack{b\sim J\\ b\mid(ae)^\infty}} (be^2,a) \ll(2+\log AV)^{C_\varepsilon} J^\varepsilon A^2V \tag*{(14)}

for all ε>0\varepsilon>0. Here Cε>0C_\varepsilon>0 depends only on ε>0\varepsilon>0 (later occurrences of CεC_\varepsilon might have a different value), a∼Aa\sim A means a∈[A,2A)a\in[A,2A) and e∼Ve\sim V, b∼Jb\sim J should be interpreted in the same way. Estimate (14) implies E1,2≪n1+η(log⁡n)Cε/J1−ε≪n1−η/2E_{1,2}\ll n^{1+\eta}(\log n)^{C_\varepsilon}/J^{1-\varepsilon}\ll n^{1-\eta/2} and so (13) if ε\varepsilon is sufficiently small.

To demonstrate (14), let h=(be2,a)h=(be^2,a). If h∣be2h\mid be^2 then h=h1h2h=h_1h_2 for some (not necessarily unique) h1∣bh_1\mid b and h2∣e2h_2\mid e^2. This implies h2′∣eh'_2\mid e for h2′=∏pk∥h2p⌈k/2⌉h'_2=\prod_{p^k\parallel h_2}p^{\lceil k/2\rceil}. We can bound (14) by

≤∑h1h2∣a∼Ah1h2∑h2′∣e∼Ve∼V/h2′b∼J/h1b∣(ae)∞1≤∑h1h2∣a∼Ah1h2∑e′∼V/h2′b′∼J/h1b′∣(ae′)∞1.\le\sum_{h_1h_2\mid a\sim A}h_1h_2\sum_{\substack{h'_2\mid e\sim V\\ e\sim V/h'_2\\ b\sim J/h_1\\ b\mid(ae)^\infty}}1 \le\sum_{h_1h_2\mid a\sim A}h_1h_2\sum_{\substack{e'\sim V/h'_2\\ b'\sim J/h_1\\ b'\mid(ae')^\infty}}1.

By Rankin’s trick,

∑b≤Bb∣m∞1≤Bε∏p∣m(1−p−ε)−1≤Bεexp⁡(Cε∑p∣mp−ε)≤Bεexp⁡(Cεω(m))\sum_{\substack{b\le B\\ b\mid m^\infty}}1\le B^\varepsilon\prod_{p\mid m}(1-p^{-\varepsilon})^{-1}\le B^\varepsilon\exp\left(C_\varepsilon\sum_{p\mid m}p^{-\varepsilon}\right)\le B^\varepsilon\exp(C_\varepsilon\omega(m))

for every ε>0\varepsilon>0. Hence

∑e′∼V/h2′b′∼J/h1b′∣(ae′)∞1≪∑e′∼V/h2′(J/h1)εexp⁡(Cε(ω(a)+ω(e′)))≪(J/h1)εV(2+log⁡V)Cεexp⁡(Cεω(a))\sum_{\substack{e'\sim V/h'_2\\ b'\sim J/h_1\\ b'\mid(ae')^\infty}}1 \ll\sum_{e'\sim V/h'_2}(J/h_1)^\varepsilon\exp(C_\varepsilon(\omega(a)+\omega(e'))) \ll(J/h_1)^\varepsilon V(2+\log V)^{C_\varepsilon}\exp(C_\varepsilon\omega(a))

and so, since h1h2≪Ah_1h_2\ll A,

∑a∼Ae∼V∑b∼Jb∣(ae)∞(be2,a)≪(2+log⁡V)CεJεV∑h1h2∣a∼Ah11−εh2exp⁡(Cεω(a))≪(2+log⁡V)CεJεAV∑a∼Aexp⁡(Cεω(a))τ(a)2≪(2+log⁡AV)CεJεA2V.\begin{aligned} \sum_{\substack{a\sim A\\ e\sim V}}\sum_{\substack{b\sim J\\ b\mid(ae)^\infty}}(be^2,a) \ll(2+\log V)^{C_\varepsilon}J^\varepsilon V\sum_{h_1h_2\mid a\sim A}h_1^{1-\varepsilon}h_2\exp(C_\varepsilon\omega(a)) \\ &\ll(2+\log V)^{C_\varepsilon}J^\varepsilon AV\sum_{a\sim A}\exp(C_\varepsilon\omega(a))\tau(a)^2 \ll(2+\log AV)^{C_\varepsilon}J^\varepsilon A^2V. \end{aligned}

Proof of Lemma 5

Let 1≤a≤nη1\le a\le n^\eta be squarefree. By completing the ee range we have

M1(a)=Li⁡(n−1)∑e≥1d≤D(de,n)=1μ(e)ϕ([de2,a])+O(nV),M2(a)=∑e≥1d<(n−2)/(e2D)(de,n)=1μ(e)Li⁡(n−de2D−1)ϕ([de2,a])+O(nV).M_1(a)=\operatorname{Li}(n-1)\sum_{\substack{e\ge1\\ d\le D\\ (de,n)=1}}\frac{\mu(e)}{\phi([de^2,a])}+O\left(\frac{n}{V}\right), \qquad M_2(a)=\sum_{\substack{e\ge1\\ d<(n-2)/(e^2D)\\ (de,n)=1}}\frac{\mu(e)\operatorname{Li}(n-de^2D-1)}{\phi([de^2,a])}+O\left(\frac{n}{V}\right).

The n/Vn/V error is more than sufficient for Lemma 5. We now decompose, up to admissible error terms both M1(a)M_1(a) and M2(a)M_2(a) as follows. Let

M1(a)=M1,1(a)+M1,2(a)+O(n/V),M_1(a)=M_{1,1}(a)+M_{1,2}(a)+O(n/V),
M2(a)=M2,1(a)+M2,2(a)+O(n/V),M_2(a)=M_{2,1}(a)+M_{2,2}(a)+O(n/V),

where we split according to de2≤Dde^{2} \le D or not

M1,1(a)=Li⁡(n−1)∑de2≤D(de,n)=1μ(e)ϕ([de2,a]).M_{1,1}(a)=\operatorname{Li}(n-1)\sum_{\substack{de^{2}\le D\\(de,n)=1}}\frac{\mu(e)}{\phi([de^{2},a])}.
M1,2(a)=Li⁡(n−1)∑de2>Dd≤D(de,n)=1μ(e)ϕ([de2,a]).M_{1,2}(a)=\operatorname{Li}(n-1)\sum_{\substack{de^{2}>D\\d\le D\\(de,n)=1}}\frac{\mu(e)}{\phi([de^{2},a])}.

and where for M2M_{2} we use Li⁡(n−de2D−1)=Li⁡(n−1)+O(de2D/log⁡n)\operatorname{Li}(n-de^{2}D-1)=\operatorname{Li}(n-1)+O(de^{2}D/\log n) so that

M2,1(a)=Li⁡(n−1)∑de2<(n−2)/D(de,n)=1μ(e)ϕ([de2,a]).M_{2,1}(a)=\operatorname{Li}(n-1)\sum_{\substack{de^{2}<(n-2)/D\\(de,n)=1}}\frac{\mu(e)}{\phi([de^{2},a])}.
M2,2(a)≪Dlog⁡n∑de2<(n−2)/Dde2μ2(e)ϕ([de2,a]).M_{2,2}(a)\ll\frac{D}{\log n}\sum_{de^{2}<(n-2)/D}\frac{de^{2}\mu^{2}(e)}{\phi([de^{2},a])}.

We claim that for square-free 1≤a≤nη1\le a\le n^{\eta} with ω(a)≤k\omega(a)\le k,

M1,2(a),M2,2(a)≪knlog⁡log⁡nalog⁡n.(15)M_{1,2}(a),M_{2,2}(a)\ll_{k}\frac{n\log\log n}{a\log n}. \tag*{(15)}
M1,1(a)+M1,2(a)=Li⁡(n−1)2ω(a)a(ϕ(n)nlog⁡n+Ok((log⁡log⁡n)2))−2Li⁡(n−1)ϕ(n)nDaϕ(a).(16)M_{1,1}(a)+M_{1,2}(a)=\operatorname{Li}(n-1)\frac{2^{\omega(a)}}{a}\left(\frac{\phi(n)}{n}\log n+O_{k}((\log\log n)^{2})\right)-\frac{2\operatorname{Li}(n-1)\phi(n)}{n}\frac{D_{a}}{\phi(a)}. \tag*{(16)}

which implies Lemma 5, as we can replace Li⁡(n−1)\operatorname{Li}(n-1) by n/log⁡n+O(n/log⁡2n)n/\log n+O(n/\log^{2}n), incurring an acceptable error.

We now bound M1,2(a)M_{1,2}(a). We have

M1,2(a)≪nlog⁡n∑D<de2≤De2μ2(e)ϕ([de2,a])≪nlog⁡log⁡nlog⁡n∑D<de2≤De21[de2,a].M_{1,2}(a)\ll\frac{n}{\log n}\sum_{D<de^{2}\le De^{2}}\frac{\mu^{2}(e)}{\phi([de^{2},a])}\ll\frac{n\log\log n}{\log n}\sum_{D<de^{2}\le De^{2}}\frac{1}{[de^{2},a]}.

Since aa is squarefree, letting (de2,a)=g(de^{2},a)=g there are g1,g2g_{1},g_{2} (not necessarily unique) with g=g1g2g=g_{1}g_{2}, g1∣eg_{1}\mid e and g2∣dg_{2}\mid d. Thus,

∑D<de2≤De21[de2,a]=a−1∑e≥1e−2∑D/e2<d≤D(de2,a)(de2,a)d−1≤a−1∑g1g2∣ag1g2∑g1∣ee−2∑D/e2<d≤Dg2∣dd−1≪a−1∑g1g2∣ag1−1(1+log⁡g1)≪a−13ω(a).\begin{aligned} \sum_{D<de^{2}\le De^{2}}\frac{1}{[de^{2},a]}=a^{-1}\sum_{e\ge1}e^{-2}\sum_{\substack{D/e^{2}<d\le D\\(de^{2},a)}}(de^{2},a)d^{-1} \\ &\le a^{-1}\sum_{g_{1}g_{2}\mid a}g_{1}g_{2}\sum_{g_{1}\mid e}e^{-2}\sum_{\substack{D/e^{2}<d\le D\\g_{2}\mid d}}d^{-1} \\ &\ll a^{-1}\sum_{g_{1}g_{2}\mid a}g_{1}^{-1}(1+\log g_{1})\ll a^{-1}3^{\omega(a)}. \end{aligned}

Since ω(a)≤k\omega(a)\le k in Lemma 5, this is sufficient for (3.4).

We now bound M2,2(a)M_{2,2}(a). Writing m=de2m=de^{2}, we have

M2,2(a)≪Dlog⁡n∑m<n/Dm2#{p2∣m}ϕ([m,a]).(17)M_{2,2}(a)\ll\frac{D}{\log n}\sum_{m<n/D}\frac{m2^{\#\{p^{2}\mid m\}}}{\phi([m,a])}. \tag*{(17)}

We let g=(a,m)g=(a,m) in the right-hand side of (17) and observe 1/φ([m,a])≤1/φ(a)φ(m/g)1/\varphi([m,a])\leq1/\varphi(a)\varphi(m/g) to find

M2,2(a)≪Dlog⁡nφ(a)∑g∣ag∑g∣m<n/D(m/g)2#{p2∣m}φ(m/g).M_{2,2}(a)\ll\frac{D}{\log n\varphi(a)}\sum_{g\mid a}g\sum_{g\mid m<n/D}\frac{(m/g)2^{\#\{p^2\mid m\}}}{\varphi(m/g)}.

Let S(x):=∑m≤X(m/φ(m))2#{p2∣m}S(x):=\sum_{m\leq X}(m/\varphi(m))2^{\#\{p^2\mid m\}}. This sum is ≪X\ll X [9]. Hence

M2,2(a)≪Dlog⁡nφ(a)∑g∣ag2ω(g)S(nDg)≪nlog⁡nφ(a)3ω(a),M_{2,2}(a)\ll\frac{D}{\log n\varphi(a)}\sum_{g\mid a}g2^{\omega(g)}S\left(\frac{n}{Dg}\right)\ll\frac{n}{\log n\varphi(a)}3^{\omega(a)},

which suffices for (15).

We now prove (16). We have

M1,1(a)=Li⁡(n−1)∑m≤D(m,n)=1μ2(m)φ([m,a]).M_{1,1}(a)=\operatorname{Li}(n-1)\sum_{\substack{m\leq D\\(m,n)=1}}\frac{\mu^2(m)}{\varphi([m,a])}.

and

M2,1(a)=Li⁡(n−1)∑m<(n−2)/D(m,n)=1μ2(m)φ([m,a]).M_{2,1}(a)=\operatorname{Li}(n-1)\sum_{\substack{m<(n-2)/D\\(m,n)=1}}\frac{\mu^2(m)}{\varphi([m,a])}.

We claim that

∑m<X(m,n)=1μ2(m)φ([m,a])=2ω(a)a(φ(n)nlog⁡X+Ok((log⁡log⁡n)2))−φ(n)nDaφ(a)(18)\sum_{\substack{m<X\\(m,n)=1}}\frac{\mu^2(m)}{\varphi([m,a])}=\frac{2^{\omega(a)}}{a}\left(\frac{\varphi(n)}{n}\log X+O_k\left((\log\log n)^2\right)\right)-\frac{\varphi(n)}{n}\frac{D_a}{\varphi(a)} \tag*{(18)}

holds uniformly for a≤nηa\leq n^\eta and X≥n1/3X\geq n^{1/3} with (a,n)=1(a,n)=1, μ2(a)=1\mu^2(a)=1 and ω(a)≤k\omega(a)\leq k. With this we can evaluate M1,1(a)+M2,1(a)M_{1,1}(a)+M_{2,1}(a) by applying (18) with X=DX=D and X=n/DX=n/D to obtain (16).

To show (18), we introduce the multiplicative function

fn,a(m):=1(m,n)=1μ2(m)φ(a)/φ([m,a]).f_{n,a}(m):=\mathbf{1}_{(m,n)=1}\mu^2(m)\varphi(a)/\varphi([m,a]).

Elliott and Halberstam established [5]

∑m<Xfn,1(m)=φ(n)nlog⁡X+O((log⁡log⁡n)2)(19)\sum_{m<X}f_{n,1}(m)=\frac{\varphi(n)}{n}\log X+O((\log\log n)^2) \tag*{(19)}

for n≤X5n\leq X^5. We have fn,a=fn,1∗gn,af_{n,a}=f_{n,1}*g_{n,a} where gn,ag_{n,a} is multiplicative, supported on odd divisors of a∞a^\infty and defined on pkp^k via gn,a(pk)=(−1)k−1(p−2)/(p−1)kg_{n,a}(p^k)=(-1)^{k-1}(p-2)/(p-1)^k. Hence, by (19),

∑m<X(m,n)=1μ2(m)φ([m,a])=∑m<Xfn,a(m)φ(a)=φ(n)nφ(a)∑d≤X/n1/5gn,a(d)log⁡(X/d)+O(Eφ(a))(20)\sum_{\substack{m<X\\(m,n)=1}}\frac{\mu^2(m)}{\varphi([m,a])}=\sum_{m<X}\frac{f_{n,a}(m)}{\varphi(a)}=\frac{\varphi(n)}{n\varphi(a)}\sum_{d\leq X/n^{1/5}}g_{n,a}(d)\log(X/d)+O\left(\frac{E}{\varphi(a)}\right) \tag*{(20)}

for

E:=∑d<X∣gn,a(d)∣(log⁡log⁡n)2+∑d>X/n1/5∣gn,a(d)∣log⁡d.(21)E:=\sum_{d<X}|g_{n,a}(d)|(\log\log n)^2+\sum_{d>X/n^{1/5}}|g_{n,a}(d)|\log d. \tag*{(21)}

To bound the first sum in EE we use ∑d≥1∣gn,a(d)∣=2ω(a)−1212∣a≪k1\sum_{d\geq1}|g_{n,a}(d)|=2^{\omega(a)-1}2^{\mathbf{1}_{2\mid a}}\ll_k1. We turn to the second sum in EE. If dd is in the support of gn,ag_{n,a} then

∣gn,a(d)∣≤∏p∣d∏pi∣dp−i∏pi∣d(1+1p−1)i−1≤ad∏pi∣d(1+1p−1)i−1≤ad(3/2)log⁡3d.|g_{n,a}(d)|\leq\prod_{p\mid d}\prod_{p^i\mid d}p^{-i}\prod_{p^i\mid d}\left(1+\frac{1}{p-1}\right)^{i-1}\leq\frac{a}{d}\prod_{p^i\mid d}\left(1+\frac{1}{p-1}\right)^{i-1}\leq\frac{a}{d}(3/2)^{\log_3d}.

If ω(a)≤k\omega(a) \le k then there are ≪k(log⁡T)k\ll_k (\log T)^k integers d∈[T,2T)d \in[T,2T) with d∣a∞d \mid a^\infty, so

∑d∈[T,2T)∣gn,a(d)∣≪ka(log⁡T)kTTlog⁡(3/2)/log⁡3.\sum_{d\in[T,2T)} \lvert g_{n,a}(d)\rvert\ll_k \frac{a(\log T)^k}{T}T^{\log(3/2)/\log3}.

This implies that under our assumptions, E/ϕ(a)≪k(log⁡log⁡n)2/aE/\phi(a) \ll_k (\log\log n)^2/a. With the same EE as in (3.10), we express the main term in (3.9) as

∑d≤X/n1/5gn,a(d)log⁡(X/d)=log⁡X∏p∣a(∑i=0∞gn,a(pi))−∑d≥1gn,a(d)log⁡d+O(E).\sum_{d\le X/n^{1/5}}g_{n,a}(d)\log(X/d)=\log X\prod_{p\mid a}\left(\sum_{i=0}^{\infty}g_{n,a}(p^i)\right)-\sum_{d\ge1}g_{n,a}(d)\log d+O(E).

times ϕ(n)/(nϕ(a))\phi(n)/(n\phi(a)). The product is 2ω(a)ϕ(a)/a2^{\omega(a)}\phi(a)/a and the dd-sum is DaD_a.

References

Mathematical Institute, University of Oxford, Oxford, OX2 6GG, UK

E-mail addresses: [email protected] [email protected]

References

  1. [1]Edgar Assing, Valentin Blomer, and Junxian Li. Uniform Titchmarsh divisor problems. Adv. Math., 393:51, 2021. Id/No 108076.
  2. [2]Valentin Blomer, Lasse Grimmelt, Junxian Li, and Simon L. Rydin Myerson. Additive problems with almost prime squares. arXiv preprint arXiv:2111.01601, 2021. To appear in GAFA.arxiv.org/abs/2111.01601
  3. [3]P. Billingsley. On the central limit theorem for the prime divisor function. Am. Math. Mon., 76:132–139, 1969.
  4. [4]Rick Durrett. Probability. Theory and examples, volume 49 of Camb. Ser. Stat. Probab. Math. Cambridge: Cambridge University Press, 5th edition edition, 2019.
  5. [5]P. D. T. A. Elliott and H. Halberstam. Some applications of Bombieri’s theorem. Mathematika, 13:196–203, 1966.DOI
  6. [6]P. D. T. A. Elliott. Additive and multiplicative functions from Nottingham to Urbana-Champaign. Int. J. Number Theory, 11(5):1357–1366, 2015.DOI
  7. [7]Andrew Granville and K. Soundararajan. Sieving and the Erdős-Kac theorem. In Equidistribution in number theory, an introduction, volume 237 of NATO Sci. Ser. II Math. Phys. Chem., pages 15–27. Springer, Dordrecht, 2007.DOI
  8. [8]H. Halberstam. On the distribution of additive number-theoretic functions. III. J. London Math. Soc., 31:14–27, 1956.DOI
  9. [9]Hugh L. Montgomery and Robert C. Vaughan. Multiplicative number theory. I. Classical theory, volume 97 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 2007.
  10. [10]Minoru Tanaka. Some results on additive number theory. IV. Tokyo J. Math., 2:183–191, 1979.DOI

Paper details

vlnc:W_jscc7fx5
Published
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents