arXiv is now an independent nonprofit! Learn more
License: CC BY 4.0
arXiv:2609.16716v1 [math.NT] 15 Sep 2026

Distribution of squarefree integers with double congruence conditions

Tang Wuji
Abstract

Given a positive integer MM with reduced residue system Ξ¦M\Phi_{M}, we partition the set of squarefree integers according to some congruence conditions on the number of prime factors in each class Ο•βˆˆΞ¦M\phi\in\Phi_{M}, and prove two relevant equidsitribution results in this setting. We discover the bias phenomenon in some special cases, and formulate a conjecture with numerical support.

1 Introduction

Let S⁑(x,b,M)S(x;b,M) denote the number of squarefree numbers that are ≀x\leq x and congruent to bb, modulo MM, where (b,M)=1(b,M)=1. In 1885 Gegenbauer[3] proved that squarefree integers have a density of 6Ο€2\frac{6}{\pi^{2}} among all integers, that is

S⁑(x,0,1)∼6​xΟ€2(xβ†’βˆž).S(x;0,1)\sim\frac{6x}{\pi^{2}}\quad(x\rightarrow\infty).

Landau[5] proved in 1909 that

S⁑(x,b,M)∼6​xΟ€2β‹…1Mβ‹…βˆp|k(1βˆ’1p2)βˆ’1(xβ†’βˆž).S(x;b,M)\sim\frac{6x}{\pi^{2}}\cdot\frac{1}{M}\cdot\prod_{p|k}\left(1-\frac{1}{p^{2}}\right)^{-1}\quad(x\rightarrow\infty). (1)

In 1974 Hooley[4] refined this result to

S⁑(x,b,M)=6​xΟ€2β‹…1Mβ‹…βˆp|M(1βˆ’1p2)βˆ’1+O⁑(xM+M12+Ο΅).S(x;b,M)=\frac{6x}{\pi^{2}}\cdot\frac{1}{M}\cdot\prod_{p|M}\left(1-\frac{1}{p^{2}}\right)^{-1}+O\left(\sqrt{\frac{x}{M}}+M^{\frac{1}{2}+\epsilon}\right).

This is the best possible individual error term bound we have obtained so far. On the other hand, the square mean of the error term, i.e.,

M2​(x,M):=βˆ‘b∈(β„€/M​℀)⋆|S⁑(x,b,M)βˆ’6​xΟ€2β‹…1Mβ‹…βˆp|M(1βˆ’1p2)βˆ’1|2,M_{2}(x,M):=\sum_{b\in(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}}\left|S(x;b,M)-\frac{6x}{\pi^{2}}\cdot\frac{1}{M}\cdot\prod_{p|M}\left(1-\frac{1}{p^{2}}\right)^{-1}\right|^{2},

are studied by many people including Valentin Blomer, Ramon M. Nunes, Pierre Le Boudec. The best estimates can be found in [6] and [1].

Here we will consider a new distribution pattern of squarefree integers. Let MM be an integer β‰₯3\geq 3, m:=ϕ⁑(M)m:=\phi(M), where Ο•\phi is the Euler totient function, that is m=|(β„€/M​℀)⋆|m=|(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}|. Let {m1,…,mm}βŠ‚β„€m\{m_{1},\dots,m_{m}\}\subset\mathbb{Z}^{m} be a set of representatives of (β„€/M​℀)⋆(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}. Given NN squarefree and coprime with M, write N=p1​…​pkN=p_{1}\dots p_{k},where p1,…,pkp_{1},\dots,p_{k} are the prime divisors of NN. Define ni=ni(N):=#{pi≑mi(modM)}n_{i}=n_{i}(N):=\#\{p_{i}\equiv m_{i}\pmod{M}\}.

The first main result of this paper is an equidistribution.

Theorem 1.1.

Let {li}1≀i≀m,liβˆˆβ„€β‰₯1\{l_{i}\}_{1\leq i\leq m},l_{i}\in\mathbb{Z}_{\geq 1}, and a=(a1,…,am)a=(a_{1},\dots,a_{m}) where 0≀ai≀liβˆ’10\leq a_{i}\leq l_{i}-1. Let L:=∏i=1mliL:=\prod_{i=1}^{m}l_{i}. Define:

Aa(x):={NΒ is squarefree and coprime withΒ M.ni≑ai(modli),N≀x}.A_{a}(x):=\{N\text{ is squarefree and coprime with }M.\\ n_{i}\equiv a_{i}\pmod{l_{i}},\quad N\leq x\}.

Then we have

|Aa​(x)|∼6​xΟ€2β‹…Lβ‹…βˆp​p​r​i​m​ep|M(1+pβˆ’1),xβ†’βˆž.{|A_{a}(x)|}\sim\frac{6x}{\pi^{2}\cdot L\cdot\prod_{\begin{subarray}{c}p\ prime\\ p|M\end{subarray}}(1+p^{-1})},\ x\rightarrow\infty.

Theorem 1.1 shows that {Aa​(x)}\{A_{a}(x)\} are equal-distributed. We also want to consider bias among {Aa​(x)}\{A_{a}(x)\}, that is, given distinct a,aβ€²a,\ a^{\prime}, does the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| stablize as xβ†’βˆžx\rightarrow\infty? Generally, there is no bias among {Aa​(x)}\{A_{a}(x)\}. However in Β§4.1 we discover that under certain conditions there should be a strict bias among {Aa​(x)}\{A_{a}(x)\}. The computation there leads us to raise the following conjecture.

Conjecture.

When ϕ⁑(M)β‰₯3,li≀2\phi(M)\geq 3,l_{i}\leq 2 for all ii, there exists a strict bias among {Aa​(x)}\{A_{a}(x)\}. That is, given distinct a,aβ€²a,\ a^{\prime}, the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| always stablizes as xβ†’βˆžx\rightarrow\infty.

The second main result of this paper studies a ”finite” version of Aa​(x)A_{a}(x).

Theorem 1.2.

Let a=(a1,…,am)a=(a_{1},\dots,a_{m}), where aiβˆˆβ„€β‰₯0a_{i}\in\mathbb{Z}_{\geq 0}. Write A:=βˆ‘i=1maiA:=\sum_{i=1}^{m}a_{i}
Define:

Aa(x):={NΒ is squarefree and coprime withΒ M.ni=ai,N≀x}.A^{a}(x):=\{N\text{ is squarefree and coprime with }M.\ n_{i}=a_{i},\ N\leq x\}.

Then we have

|Aa​(x)|∼Aa1!​…​am!β‹…mAβ‹…xβ‹…(l​o​g​x)βˆ’1β‹…(l​o​g​l​o​g​x)Aβˆ’1,xβ†’βˆž.{|A^{a}(x)|}\sim\frac{A}{a_{1}!\dots a_{m}!\cdot m^{A}}\cdot x\cdot(log\ x)^{-1}\cdot(log\ log\ x)^{A-1},x\rightarrow\infty.
Remark.

Given distinct a=(a1,…,am),aβ€²=(a1β€²,…,amβ€²)a=(a_{1},\dots,a_{m}),a^{\prime}=(a_{1}^{\prime},\dots,a_{m}^{\prime}), under the condition βˆ‘ai=βˆ‘aiβ€²=A\sum a_{i}=\sum a_{i}^{\prime}=A, we have:

limxβ†’βˆž|Aa​(x)|:|Aa′​(x)|=(Aa1,a2,…,an):(Aa1β€²,a2β€²,…,anβ€²).\lim_{x\rightarrow\infty}|A^{a}(x)|:|A^{a^{\prime}}(x)|=\binom{A}{a_{1},a_{2},\dots,a_{n}}:\binom{A}{a^{\prime}_{1},a^{\prime}_{2},\dots,a^{\prime}_{n}}.

Where (Aa1,a2,…,an)\binom{A}{a_{1},a_{2},\dots,a_{n}} is the multinomial coefficient and equal to A!a1!​…​am!.\frac{A!}{a_{1}!\dots a_{m}!}.

The estimate of Aa​(x)A_{a}(x) refines the estimate of S⁑(x,b,M)S(x;b,M). More precisely, let

B(x;b,M):={N≀x|Nissquarefree,N≑b(modM)}B(x;b,M):=\left\{N\leq x\middle|\ N\ is\ squarefree,\ N\equiv b\pmod{M}\right\}

so that S⁑(x,b,M)=|B⁑(x,b,M)|S(x;b,M)=|B(x;b,M)|. In the setting of Theorem 1.1, we choose li:=|<mi>(β„€/M​℀)⋆|l_{i}:=|<m_{i}>_{(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}}| to be the order of mim_{i} in the group (β„€/M​℀)⋆(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}. Let AA denote the set of all aa. We can give AA an additive group structure by identifying AA with ∏i=1m(β„€/li​℀)\prod_{i=1}^{m}\left(\mathbb{Z}/\penalty l_{i}\mathbb{Z}\right). Then

h⁑(a):=∏i=1mmiai(modM)h(a):=\prod_{i=1}^{m}m_{i}^{a_{i}}\pmod{M}

gives a group homomorphism from AA to (β„€/M​℀)⋆(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}. We claim that

B⁑(x,b,M)=⨆h⁑(a)=b(modM)Aa​(x).B(x;b,M)=\bigsqcup_{h(a)=b\pmod{M}}A_{a}(x). (2)

For βˆ€Nβˆˆβ¨†h⁑(a)=b(modM)Aa​(x)\forall N\in\bigsqcup_{h(a)=b\pmod{M}}A_{a}(x), N∈Aa​(x)N\in A_{a}(x) for some a=(a1,…,am)a=(a_{1},\dots,a_{m}). Write N=p1​…​pkN=p_{1}\dots p_{k}, where p1,…,pkp_{1},\dots,p_{k} are the prime divisors of NN. Recall ni=#{pi≑mi(modM)}n_{i}=\#\{p_{i}\equiv m_{i}\pmod{M}\}, and ni≑ai(modli)n_{i}\equiv a_{i}\pmod{l_{i}}. Let ni=ai+bi​lin_{i}=a_{i}+b_{i}l_{i}. For 1≀i≀m1\leq i\leq m, let

Pi:=∏p|Np≑mimodMp.P_{i}:=\prod_{\begin{subarray}{c}p|N\\ p\equiv m_{i}\mod M\end{subarray}}p.

Then N=∏i=1mPiN=\prod_{i=1}^{m}P_{i}. Pi≑mini(modM)≑miai+bi​li(modM)P_{i}\equiv m_{i}^{n_{i}}\pmod{M}\equiv m_{i}^{a_{i}+b_{i}l_{i}}\pmod{M}. Since lil_{i} is the order of mim_{i} in the group (β„€/M​℀)⋆(\mathbb{Z}/\penalty M\mathbb{Z})^{\star}, mili≑1(modM).m_{i}^{l_{i}}\equiv 1\pmod{M}. This gives Nβ‰‘βˆi=1mmiai(modM)≑b(modM)N\equiv\prod_{i=1}^{m}m_{i}^{a_{i}}\pmod{M}\equiv b\pmod{M}, and N∈B⁑(x,b,M)N\in B(x;b,M).

Conversely, given N∈B⁑(x,b,M)N\in B(x;b,M), decompose NN and introduce nin_{i}, aia_{i}, PiP_{i} as above. Then N∈Aa​(x)N\in A_{a}(x), with a=(a1,…,am)a=(a_{1},\dots,a_{m}), and

h⁑(a)=∏i=1mmiaiβ‰‘βˆi=1mminiβ‰‘βˆi=1mPi=N≑b(modM)h(a)=\prod_{i=1}^{m}m_{i}^{a_{i}}\equiv\prod_{i=1}^{m}m_{i}^{n_{i}}\equiv\prod_{i=1}^{m}P_{i}=N\equiv b\pmod{M}

Assuming Theorem 1.1, we can deduce the main term of S⁑(x,b,M)S(x;b,M) given by (1) as follows. Define

A(mi):={a=(a1,…,am)| 0≀ai≀liβˆ’1,βˆ€N∈Aa,N≑mi(modM)}A(m_{i}):=\{a=(a_{1},\dots,a_{m})|\ 0\leq a_{i}\leq l_{i}-1,\ \forall N\in A_{a},\ N\equiv m_{i}\pmod{M}\}

Each A⁑(mi)=hβˆ’1​(mi)A(m_{i})=h^{-1}(m_{i}) is a coset of A/A⁑(1)A/\penalty A(1), hence independently of ii, we have

|A⁑(mi)|=|A|m=Lϕ⁑(M)=LMβ€‹βˆp|M(1βˆ’pβˆ’1).|A(m_{i})|=\frac{|A|}{m}=\frac{L}{\phi(M)}=\frac{L}{M\prod_{p|M}(1-p^{-1})}.

This gives

S⁑(x,b,M)∼|A⁑(b)|β‹…6​xΟ€2β‹…Lβ‹…βˆp|M(1+pβˆ’1)=6​xΟ€2β‹…1Mβ‹…βˆp|M(1βˆ’1p2)βˆ’1,xβ†’βˆž.S(x;b,M)\sim|A(b)|\cdot\frac{6x}{\pi^{2}\cdot L\cdot\prod_{p|M}(1+p^{-1})}=\frac{6x}{\pi^{2}}\cdot\frac{1}{M}\cdot\prod_{p|M}\left(1-\frac{1}{p^{2}}\right)^{-1},\ x\rightarrow\infty.
Remark.

Theorem 1.1 and Theorem 1.2 can be extended to nn-free numbers.

2 Tauberian Theorems

We need the following lemmas which connect the pole of a Dirichlet series with the order of growth of its partial sum of its coefficients. The first lemma is due to Selberg-Delange [7]:
Let zβˆˆβ„‚,0<δ≀1,M>0.z\in\mathbb{C},0<\delta\leq 1,M>0. We say that a Dirichlet series F⁑(s)F(s) has property P⁑(z,c0,Ξ΄,M)P(z;c_{0},\delta,M) if the Dirichlet series G⁑(s,z):=F⁑(s)​΢​(s)βˆ’zG(s;z):=F(s)\zeta(s)^{-z} may be analytically continued to Οƒβ‰₯1βˆ’c0/(1+l​o​g+​(Ο„))\sigma\geq 1-c_{0}/(1+log^{+}(\tau)), and, in this domain, satifies the bound

|G⁑(s,z)|≀M​(1+|Ο„|)1βˆ’Οƒ.|G(s;z)|\leq M(1+|\tau|)^{1-\sigma}.

We will say F⁑(s)F(s) has a pole of order zz at s=1s=1. Here s=Οƒ+iβ€‹Ο„βˆˆβ„‚,l​o​g+​(x):=M​a​x​(0,l​o​g​(x)).s=\sigma+i\tau\in\mathbb{C},log^{+}(x):=Max(0,log(x)).
If F:=βˆ‘n=0∞an​nβˆ’sF:=\sum_{n=0}^{\infty}a_{n}n^{-s} has property P⁑(z,c0,Ξ΄,M)P(z;c_{0},\delta,M) and there exist real numbers {bn}n=1∞\{b_{n}\}_{n=1}^{\infty} such that |an|≀bn|a_{n}|\leq b_{n} and βˆ‘n=1∞bn​nβˆ’s\sum_{n=1}^{\infty}b_{n}n^{-s} satisfies P⁑(w,c0,Ξ΄,M)P(w,c_{0},\delta,M) for some wβˆˆβ„‚.w\in\mathbb{C}. We say F⁑(s)F(s) has property T⁑(z,w,c0,Ξ΄,M).T(z,w;c_{0},\delta,M).

Lemma 2.1.

Given F⁑(s)=βˆ‘n=1∞an​nβˆ’sF(s)=\sum_{n=1}^{\infty}a_{n}n^{-s} with property T⁑(z,w,c0,Ξ΄,M)T(z,w;c_{0},\delta,M), we have:

βˆ‘n≀xan=x​(l​o​g​x)zβˆ’1​[βˆ‘k=0NΞ»k​(z)(l​o​g​x)k+O⁑(M​RN​(x))],\sum_{n\leq x}a_{n}=x(log\ x)^{z-1}\left[\sum_{k=0}^{N}\frac{\lambda_{k}(z)}{(log\ x)^{k}}+O(MR_{N}(x))\right],

where

RN​(x):=eβˆ’c1​l​o​g​x+(c2​N+1l​o​g​x)N+1,R_{N}(x):=e^{-c_{1}\sqrt{logx}}+(\tfrac{c_{2}N+1}{logx})^{N+1},

and

Ξ»k​(z):=1Γ⁑(zβˆ’k)β€‹βˆ‘h+j=k1h!​j!​G(h)​(1,z)​γj​(z).\lambda_{k}(z):=\frac{1}{\Gamma(z-k)}\sum_{h+j=k}\frac{1}{h!j!}G^{(h)}(1;z)\gamma_{j}(z).

Here Ξ³j​(z)\gamma_{j}(z) is defined by the Taylor series coefficient:

sβˆ’1​((sβˆ’1)​΢​(s))z=βˆ‘j=0∞1j!​γj​(z)​(sβˆ’1)j.s^{-1}((s-1)\zeta(s))^{z}=\sum_{j=0}^{\infty}\frac{1}{j!}\gamma_{j}(z)(s-1)^{j}.

And in the domain where G⁑(s,z)G(s;z) is holomorphic, we set

G(k)​(s,z):=βˆ‚kβˆ‚sk​G​(s,z).G^{(k)}(s;z):=\frac{\partial^{k}}{\partial s^{k}}G(s;z).

The next Lemma is due to Delange[2]:

Lemma 2.2.

Let F⁑(s):=βˆ‘n=1∞an​nβˆ’sF(s):=\sum_{n=1}^{\infty}a_{n}n^{-s}be a Dirichlet series with anβ‰₯0a_{n}\geq 0, and converges for Οƒ>1\sigma>1. Suppose F⁑(s)F(s) is holomorphic at all points of the line Οƒ=1\sigma=1 except at s=1s=1, and, in a neighborhood of s=1s=1 in the half plane Οƒ>1\sigma>1, we have:

F⁑(s)=(sβˆ’1)βˆ’wβˆ’1β€‹βˆ‘j=0qgi​(s)​(l​o​g​(1sβˆ’1))j+g⁑(s),F(s)=(s-1)^{-w-1}\sum_{j=0}^{q}g_{i}(s)\left(log(\tfrac{1}{s-1})\right)^{j}+g(s),

where ww is a negative integer,and gi​(s),g​(s)g_{i}(s),g(s) are holomorphic at s=1s=1, with gq​(1)β‰ 0g_{q}(1)\neq 0. Then as xβ†’βˆžx\rightarrow\infty

A⁑(x):=βˆ‘n≀xan∼(βˆ’1)βˆ’wβˆ’1​(βˆ’wβˆ’1)!​q​gq​(1)​x​(l​o​g​x)w​(l​o​g​l​o​g​x)qβˆ’1.A(x):=\sum_{n\leq x}a_{n}\sim(-1)^{-w-1}(-w-1)!\ q\ g_{q}(1)x(log\ x)^{w}(log\ log\ x)^{q-1}.

3 Proof

3.1 Congruence restriction on the number of prime divisors

Proof of Theorem 1.1.

For each lil_{i}, let ei:=e2​π​ilie_{i}:=e^{\frac{2\pi i}{l_{i}}} denote a lil_{i}-th root of unity. Define J:={(j1,…,jm)| 0≀ji≀liβˆ’1,jiβˆˆβ„€}.J:=\{(j_{1},\dots,j_{m})|\ 0\leq j_{i}\leq l_{i}-1,j_{i}\in\mathbb{Z}\}.

For each j=(j1,…,jm)∈Jj=(j_{1},\dots,j_{m})\in J, define

Lj​(s):=∏1≀i≀m∏p​p​r​i​m​ep≑mi​ mod ​M(1+eiji​pβˆ’s).L_{j}(s):=\prod_{1\leq i\leq m}\prod_{\begin{subarray}{c}p\ prime\\ p\equiv m_{i}\text{ mod }M\end{subarray}}(1+e_{i}^{j_{i}}p^{-s}). (3)

Define

L0​(s):=L(0,…,0)​(s)=΢⁑(s)΢⁑(2​s)β€‹βˆp​p​r​i​m​ep|M(1+pβˆ’s),L_{0}(s):=L_{(0,\dots,0)}(s)=\frac{\zeta(s)}{\zeta(2s)\prod_{\begin{subarray}{c}p\ prime\\ p|M\end{subarray}}(1+p^{-s})},
Aa:=⋃x=1∞Aa​(x),A_{a}:=\bigcup_{x=1}^{\infty}A_{a}(x),
LAa​(s):=βˆ‘n∈Aanβˆ’s.L_{A_{a}}(s):=\sum_{n\in A_{a}}n^{-s}.

LAa​(s)L_{A_{a}}(s) can be written as a linear combination of Lj​(s)L_{j}(s):

LAa​(s)=1Lβ€‹βˆ‘j∈J(∏1≀i≀meiβˆ’ai​ji​Lj​(s)).L_{A_{a}}(s)=\frac{1}{L}\sum_{j\in J}\left(\prod_{1\leq i\leq m}e_{i}^{-a_{i}j_{i}}L_{j}(s)\right).

Especially, the coefficient of L0​(s)L_{0}(s) is 1L\frac{1}{L} for all aa.

Define

Lj⋆​(s):=∏1≀i≀m∏p​p​r​i​m​ep≑mimodM(1+pβˆ’s)eiji.L_{j}^{\star}(s):=\prod_{1\leq i\leq m}\prod_{\begin{subarray}{c}p\ prime\\ p\equiv m_{i}\bmod M\end{subarray}}(1+p^{-s})^{e_{i}^{j_{i}}}.

Note that for a Dirichlet character Ο‡\chi mod MM:

log⁑L⁑(s,Ο‡)=βˆ‘p,kχ⁑(pk)k​pk​s=βˆ‘pχ⁑(p)ps+βˆ‘p,kβ‰₯2χ⁑(pk)k​pk​s.\log L(s,\chi)=\sum_{p,k}\frac{\chi(p^{k})}{kp^{ks}}=\sum_{p}\frac{\chi(p)}{p^{s}}+\sum_{p,k\geq 2}\frac{\chi(p^{k})}{kp^{ks}}.

The second term converges absolutely for R​e​(s)>12Re(s)>\frac{1}{2}, hence in a neighborhood of s=1s=1 we can write

log⁑L⁑(s,Ο‡)=βˆ‘pχ⁑(p)ps+O⁑(1).\log L(s,\chi)=\sum_{p}\frac{\chi(p)}{p^{s}}+O(1).

Therefore we have

βˆ‘p≑mimodM1ps=\displaystyle\sum_{p\equiv m_{i}\bmod M}\frac{1}{p^{s}}= 1mβ€‹βˆ‘pβˆ‘Ο‡Ο‡β‘(p)​χ¯​(mi)ps\displaystyle\frac{1}{m}\sum_{p}\sum_{\chi}\frac{\chi(p)\overline{\chi}(m_{i})}{p^{s}}
=\displaystyle= 1mβ€‹βˆ‘Ο‡Ο‡Β―β€‹(mi)​(βˆ‘pχ⁑(p)ps)\displaystyle\frac{1}{m}\sum_{\chi}\overline{\chi}(m_{i})\left(\sum_{p}\frac{\chi(p)}{p^{s}}\right)
=\displaystyle= 1mβ€‹βˆ‘Ο‡modMχ¯​(mi)​log⁑L⁑(s,Ο‡)+O⁑(1).\displaystyle\frac{1}{m}\sum_{\chi\bmod M}\overline{\chi}(m_{i})\log L(s,\chi)+O(1).

Where Ο‡\chi ranges over Dirichlet characters mod MM. For all nontrivial Dirichlet characters Ο‡\chi, L⁑(s,Ο‡)L(s,\chi) converges at s=1s=1 and L⁑(1,Ο‡)β‰ 1L(1,\chi)\neq 1. For the trivial character Ο‡0\chi_{0}, L⁑(s,Ο‡0)L(s,\chi_{0}) has a simple pole at s=1s=1, that is to say for R​e​(s)>1Re(s)>1:

l​o​g​(L⁑(s,Ο‡0))=l​o​g​(1sβˆ’1)+O⁑(1).log(L(s,\chi_{0}))=log\left(\frac{1}{s-1}\right)+O(1).

Hence

l​o​g​(Lj⋆​(s))=1mβ€‹βˆ‘i=1meiji​l​o​g​(1sβˆ’1)+O⁑(1).log\left(L_{j}^{\star}(s)\right)=\frac{1}{m}\sum_{i=1}^{m}e_{i}^{j_{i}}log\left(\frac{1}{s-1}\right)+O(1).

Therefore, Lj⋆​(s)L_{j}^{\star}(s) has a pole of order 1mβ€‹βˆ‘i=1meiji\frac{1}{m}\sum_{i=1}^{m}e_{i}^{j_{i}} at s=1s=1.

Notice that ∏p​p​r​i​m​ep≑mimodM(1+pβˆ’s)eiji=∏p(1+eiji​pβˆ’s+O⁑(pβˆ’2​s))\prod_{\begin{subarray}{c}p\ prime\\ p\equiv m_{i}\bmod M\end{subarray}}(1+p^{-s})^{e_{i}^{j_{i}}}=\prod_{p}(1+e_{i}^{j_{i}}p^{-s}+O(p^{-2s})), where O⁑(pβˆ’2​s)O(p^{-2s}) is bounded independently with ii, therefore, Lj​(s)/Lj⋆​(s)=∏p(1+O⁑(pβˆ’2​s))L_{j}(s)/L_{j}^{\star}(s)=\prod_{p}(1+O(p^{-2s})) which converges at s=1s=1, hence LjL_{j} has a pole of order 1mβ€‹βˆ‘i=1meiji\frac{1}{m}\sum_{i=1}^{m}e_{i}^{j_{i}} at s=1s=1 as well. In addition, the real part of the order of the pole is always <1<1 unless j=(0,…,0)j=(0,\dots,0).

Let Lj​(s)=βˆ‘n=1∞aj,n​nβˆ’sL_{j}(s)=\sum_{n=1}^{\infty}a_{j,n}n^{-s} be the Dirichlet series expansion, and let A~j​(x):=βˆ‘n≀xaj,n\tilde{A}_{j}(x):=\sum_{n\leq x}a_{j,n}. Then

|Aa​(x)|=1Lβ€‹βˆ‘j∈J∏1≀i≀meiβˆ’ai​ji​A~j​(x).|A_{a}(x)|=\frac{1}{L}\sum_{j\in J}\prod_{1\leq i\leq m}e_{i}^{-a_{i}j_{i}}\tilde{A}_{j}(x).

We can now apply Lemma 2.1, to Lj​(s)L_{j}(s), and we have

A~j​(x)=O⁑(x​(l​o​g​x)1mβ€‹βˆ‘i=1meijiβˆ’1),\tilde{A}_{j}(x)=O\left(x(logx)^{\frac{1}{m}\sum_{i=1}^{m}e_{i}^{j_{i}}-1}\right),

which =o⁑(x)=o(x) when jβ‰ (0,…,0)j\neq(0,\dots,0).

L0​(s)L_{0}(s) has a regular pole at s=1s=1 with residue 1΢⁑(2)β€‹βˆp​p​r​i​m​ep|M(1+pβˆ’1)\frac{1}{\zeta(2)\prod_{\begin{subarray}{c}p\ prime\\ p|M\end{subarray}}(1+p^{-1})} at s=1s=1, hence we have:

|A0​(x)|=1΢⁑(2)β‹…βˆp​p​r​i​m​ep|M(1+pβˆ’1)β‹…x+o⁑(x).|A_{0}(x)|=\frac{1}{\zeta(2)\cdot\prod_{\begin{subarray}{c}p\ prime\\ p|M\end{subarray}}(1+p^{-1})}\cdot x+o(x).

This gives

|Aa​(x)|=6Ο€2β‹…Lβ‹…βˆp​p​r​i​m​ep|M(1+pβˆ’1)β‹…x+o⁑(x).|A_{a}(x)|=\frac{6}{\pi^{2}\cdot L\cdot\prod_{\begin{subarray}{c}p\ prime\\ p|M\end{subarray}}(1+p^{-1})}\cdot x+o(x).
Remark.

If we apply Lemma 2.1 to all of Lj​(s)L_{j}(s) in LAa​(s)L_{A_{a}}(s), we can refine the error term to O⁑(xβ‹…(l​o​g​x)c​o​s​(2​πm​a​x​(li)i)βˆ’1)O\left(x\cdot(log\ x)^{cos\left(\frac{2\pi}{max(l_{i})_{i}}\right)-1}\right). However the power of xx in the error term can never be reduced below 11 with this method alone. Another result is that if we consider |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a}^{\prime}(x)|, the L0L_{0} term cancels out and the resulting difference has to be of the form x​(l​o​g​x)1mβ€‹βˆ‘i=1meijiβˆ’1​(C+O⁑(1l​o​g​(sβˆ’1)))x(logx)^{\frac{1}{m}\sum_{i=1}^{m}e_{i}^{j_{i}}-1}(C+O(\frac{1}{log(s-1)})). If given specific a,aβ€²a,a^{\prime}, we may directly calculate CC, and Cβ‰ 0C\neq 0 would imply the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a}^{\prime}(x)| changes infinitely many times as xβ†’βˆžx\rightarrow\infty. We conjecture that this is true for any distinct a,aβ€²a,a^{\prime}.

∎

3.2 Restriction of the number of prime divisors

Proof of Theorem 1.2.

Let a^i:=(0,…,0,ai,0,…,0)\hat{a}_{i}:=(0,\ldots,0,a_{i},0,\dots,0), where aia_{i} is in the ii-th component.

Define:

LAa​(s):=\displaystyle L_{A^{a}}(s):= βˆ‘n∈Aanβˆ’s,\displaystyle\sum_{n\in A^{a}}n^{-s},
LAa^i​(s):=\displaystyle L_{A^{\hat{a}_{i}}}(s):= βˆ‘n∈Aa^inβˆ’s.\displaystyle\sum_{n\in A^{\hat{a}_{i}}}n^{-s}.

Then we have:

LAa​(s)=∏i=1m(βˆ‘n∈Aa^inβˆ’s)=∏i=1mLAa^i​(s).L_{A^{a}}(s)=\prod_{i=1}^{m}\left(\sum_{n\in A^{\hat{a}_{i}}}n^{-s}\right)=\prod_{i=1}^{m}L_{A^{\hat{a}_{i}}}(s).

Assume aiβ‰₯2a_{i}\geq 2. Consider the Dirichlet series expansion for (βˆ‘p≑mimodM1ps)ai\left(\sum_{p\equiv m_{i}\bmod M}\frac{1}{p^{s}}\right)^{a_{i}}. βˆ€n∈Aa^i\forall n\in A^{\hat{a}_{i}}, nβˆ’sn^{-s} has coefficient ai!a_{i}!. On the other hand, the terms not in Aa^iA^{\hat{a}_{i}} all contain pβˆ’2​sp^{-2s} for some p≑mimodMp\equiv m_{i}\bmod M. This gives

|1ai!​(βˆ‘p1ps)aiβˆ’LAa^i​(s)|=O⁑((βˆ‘p1p2​s)​(βˆ‘p1ps)aiβˆ’2).\left|\frac{1}{a_{i}!}\left(\sum_{p}\frac{1}{p^{s}}\right)^{a_{i}}-L_{A^{\hat{a}_{i}}}(s)\right|=O\left(\left(\sum_{p}\frac{1}{p^{2s}}\right)\left(\sum_{p}\frac{1}{p^{s}}\right)^{a_{i}-2}\right).

Here all summations are over pp prime and p≑mimodMp\equiv m_{i}\bmod M. We have

LAa^i​(s)=1Mai​ai!​(l​o​g​1sβˆ’1)ai+O⁑((l​o​g​1sβˆ’1)aiβˆ’2).L_{A^{\hat{a}_{i}}}(s)=\frac{1}{M^{a_{i}}a_{i}!}\left(log\ \frac{1}{s-1}\right)^{a_{i}}+O\left(\left(log\ \frac{1}{s-1}\right)^{a_{i}-2}\right).

If ai=1a_{i}=1, then we have

LAa^i​(s)=βˆ‘p≑mimodM1ps=1M​l​o​g​1sβˆ’1+O⁑(1).L_{A^{\hat{a}_{i}}}(s)=\sum_{p\equiv m_{i}\bmod M}\frac{1}{p^{s}}=\frac{1}{M}log\frac{1}{s-1}+O(1).

In any case, we have

LAa^i​(s)=1Mai​ai!​(l​o​g​1sβˆ’1)ai+O⁑((l​o​g​1sβˆ’1)aiβˆ’1).L_{A^{\hat{a}_{i}}}(s)=\frac{1}{M^{a_{i}}a_{i}!}\left(log\ \frac{1}{s-1}\right)^{a_{i}}+O\left(\left(log\ \frac{1}{s-1}\right)^{a_{i}-1}\right).

Hence

LAa​(s)=∏i=1mLAa^i​(s)=1mAβ‹…βˆi=1mai!​(l​o​g​1sβˆ’1)A+O⁑((l​o​g​1sβˆ’1)Aβˆ’1).L_{A^{a}}(s)=\prod_{i=1}^{m}L_{A^{\hat{a}_{i}}}(s)=\frac{1}{m^{A}\cdot\prod_{i=1}^{m}a_{i}!}\left(log\ \frac{1}{s-1}\right)^{A}+O\left(\left(log\ \frac{1}{s-1}\right)^{A-1}\right).

Apply Lemma 2.2 to LAa​(a)L_{A^{a}}(a) and we have:

|Aa​(x)|∼Aa1!​…​am!β‹…mAβ‹…xβ‹…(l​o​g​x)βˆ’1β‹…(l​o​g​l​o​g​x)Aβˆ’1,xβ†’βˆž.{|A^{a}(x)|}\sim\frac{A}{a_{1}!\dots a_{m}!\cdot m^{A}}\cdot x\cdot(log\ x)^{-1}\cdot(log\ log\ x)^{A-1},\ x\rightarrow\infty.

∎

4 Some possible Bias Among Aa​(X)A_{a}(X)

Given MM and lil_{i}, we consider the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)|, where aa and aβ€²a^{\prime} are distinct. We’ll show that if some liβ‰₯3l_{i}\geq 3, this sign changes infinitely many times, and if li≀2l_{i}\leq 2 for all ii, we may expect the sign to become stable as xβ†’βˆžx\rightarrow\infty. We shall justify this phenomenon for certain specific M,liM,l_{i}.

4.1 ϕ⁑(M)β‰₯3,li≀2\phi(M)\geq 3,l_{i}\leq 2 for All ii

After re-indexing, we may assume l1=β‹―=lh=2l_{1}=\dots=l_{h}=2, and lh+1=β‹―=lm=1l_{h+1}=\dots=l_{m}=1. If h=0h=0, there would be only one element aa in AA and it makes no sense to talk about bias. So we may assume hβ‰₯1h\geq 1. We’ll use (β„€/2​℀)h(\mathbb{Z}/\penalty 2\mathbb{Z})^{h} to denote the elements of AA, and let A1A_{1} denote the subset of A with exactly one index being 11, and the rest being 00. For Lj​(s)L_{j}(s) defined in (3), let |j||j| denote the number of indices that are equal to 11, then Lj​(s)L_{j}(s) has a pole of order 1βˆ’2​|j|m1-\frac{2|j|}{m} at s=1s=1. Hence, among the collection of Lj​(s)L_{j}(s), {Lj​(s)}j∈A1\{L_{j}(s)\}_{j\in A_{1}} has the second highest order pole of order 1βˆ’2m>01-\frac{2}{m}>0 after L0​(s)L_{0}(s), which has a pole of order 1 at s=1s=1.

Given 1≀c≀h1\leq c\leq h, let jc∈A1j_{c}\in A_{1} be the element whose cc-th component equals 11. Then

Ljc​(s):=(∏1≀i≀miβ‰ c∏p​p​r​i​m​ep≑mi​ mod ​M(1+pβˆ’s))β€‹βˆp​p​r​i​m​ep≑mc​ mod ​M(1βˆ’pβˆ’s)L_{j_{c}}(s):=\left(\prod_{\begin{subarray}{c}1\leq i\leq m\\ i\neq c\end{subarray}}\prod_{\begin{subarray}{c}p\ prime\\ p\equiv m_{i}\text{ mod }M\end{subarray}}(1+p^{-s})\right)\prod_{\begin{subarray}{c}p\ prime\\ p\equiv m_{c}\text{ mod }M\end{subarray}}(1-p^{-s})

Let βˆ‘n=1∞aj,n​nβˆ’s\sum_{n=1}^{\infty}a_{j,n}n^{-s} be the Dirichlet series expansion of Ljc​(s)L_{j_{c}}(s) and let Ajc​(x):=βˆ‘n≀xaj,nA_{j_{c}}(x):=\sum_{n\leq x}a_{j,n}. By Lemma 2.1 we have:

Ajc​(x)∼ccβ‹…x​(l​o​g​x)βˆ’2m,cc=1Γ⁑(1βˆ’2m)​limsβ†’1Ljc​(s)​΢​(s)βˆ’1+2mβ‰ 0A_{j_{c}}(x)\sim c_{c}\cdot x(log\ x)^{-\frac{2}{m}},\ c_{c}=\frac{1}{\Gamma(1-\frac{2}{m})}\lim_{s\rightarrow 1}L_{j_{c}}(s)\zeta(s)^{-1+\frac{2}{m}}\neq 0
cc\displaystyle c_{c} =1Γ⁑(1βˆ’2m)​limsβ†’1Ljc​(s)​΢​(s)βˆ’1+2m\displaystyle=\frac{1}{\Gamma(1-\frac{2}{m})}\lim_{s\rightarrow 1}L_{j_{c}}(s)\zeta(s)^{-1+\frac{2}{m}}
=1Γ⁑(1βˆ’2m)β€‹βˆp|M(1βˆ’pβˆ’1)1βˆ’2m\displaystyle=\frac{1}{\Gamma(1-\frac{2}{m})}\prod_{p|M}(1-p^{-1})^{1-\frac{2}{m}}
limsβ†’1∏iβ‰ c∏p≑mi​(M)((1+pβˆ’s)​(1βˆ’pβˆ’s)1βˆ’2m)β‹…βˆp≑mc​(M)(1βˆ’pβˆ’s)2βˆ’2m.\displaystyle\lim_{s\rightarrow 1}\prod_{i\neq c}\prod_{p\equiv m_{i}(M)}\left((1+p^{-s})(1-p^{-s})^{1-\frac{2}{m}}\right)\cdot\prod_{p\equiv m_{c}(M)}(1-p^{-s})^{2-\frac{2}{m}}.

We will need the following lemma for the purpose of numerically computing ccc_{c}.

Lemma 4.1.

We have the equality of the two following limits:

limsβ†’1∏iβ‰ c∏p≑mi​(M)((1+pβˆ’s)​(1βˆ’pβˆ’s)1βˆ’2m)β‹…βˆp≑mc​(M)(1βˆ’pβˆ’s)2βˆ’2m\displaystyle\lim_{s\rightarrow 1}\prod_{i\neq c}\prod_{p\equiv m_{i}(M)}\left((1+p^{-s})(1-p^{-s})^{1-\frac{2}{m}}\right)\cdot\prod_{p\equiv m_{c}(M)}(1-p^{-s})^{2-\frac{2}{m}}
=\displaystyle= limNβ†’βˆžβˆiβ‰ c∏p≑mi​(M)p≀N((1+pβˆ’1)​(1βˆ’pβˆ’1)1βˆ’2m)β‹…βˆp≑mc​(M)p≀N(1βˆ’pβˆ’1)2βˆ’2m.\displaystyle\lim_{N\rightarrow\infty}\prod_{i\neq c}\prod_{\begin{subarray}{c}p\equiv m_{i}(M)\\ p\leq N\end{subarray}}\left((1+p^{-1})(1-p^{-1})^{1-\frac{2}{m}}\right)\cdot\prod_{\begin{subarray}{c}p\equiv m_{c}(M)\\ p\leq N\end{subarray}}(1-p^{-1})^{2-\frac{2}{m}}.
Proof.

Define

F⁑(s):=∏iβ‰ c∏p≑mi​(M)((1+pβˆ’s)​(1βˆ’pβˆ’s)1βˆ’2m)β‹…βˆp≑mc​(M)(1βˆ’pβˆ’s)2βˆ’2m.F(s):=\prod_{i\neq c}\prod_{p\equiv m_{i}(M)}\left((1+p^{-s})(1-p^{-s})^{1-\frac{2}{m}}\right)\cdot\prod_{p\equiv m_{c}(M)}(1-p^{-s})^{2-\frac{2}{m}}.

Then we have

l​o​g​F​(s)=2mβ€‹βˆ‘iβ‰ cp≑mi​(M)pβˆ’s+(2mβˆ’2)β€‹βˆ‘p≑mc​(M)pβˆ’s+βˆ‘O⁑(pβˆ’2​s).log\ F(s)=\frac{2}{m}\sum_{\begin{subarray}{c}i\neq c\\ p\equiv m_{i}(M)\end{subarray}}p^{-s}+\left(\frac{2}{m}-2\right)\sum_{p\equiv m_{c}(M)}p^{-s}+\sum O(p^{-2s}).

Here βˆ‘O⁑(pβˆ’2​s)\sum O(p^{-2s}) converges absolutely on R​e​(s)>12Re(s)>\frac{1}{2}, hence we can exchange limNβ†’βˆžβˆ‘n≀N\lim_{N\rightarrow\infty}\sum_{n\leq N} with limsβ†’1\lim_{s\rightarrow 1}.

Let an={2mif ​n​prime and​n≑mi​(M)​ for some ​iβ‰ c2mβˆ’2if ​n​prime and​n≑mc​(M)0otherwisea_{n}=\begin{cases}\frac{2}{m}&\text{if }n\ \text{prime and}\ n\equiv m_{i}(M)\text{ for some }i\neq c\\ \frac{2}{m}-2&\text{if }n\ \text{prime and}\ n\equiv m_{c}(M)\\ 0&\text{otherwise}\end{cases}

Let

L⁑(s):=βˆ‘n=1∞anns=2mβ€‹βˆ‘iβ‰ cp≑mi​(M)pβˆ’s+(2mβˆ’2)β€‹βˆ‘p≑mc​(M)pβˆ’s.L(s):=\sum_{n=1}^{\infty}\frac{a_{n}}{n^{s}}=\frac{2}{m}\sum_{\begin{subarray}{c}i\neq c\\ p\equiv m_{i}(M)\end{subarray}}p^{-s}+\left(\frac{2}{m}-2\right)\sum_{p\equiv m_{c}(M)}p^{-s}.

We need to prove: L⁑(s)L(s) can be extended to s=1s=1, and

limsβ†’1+L⁑(s)=limNβ†’βˆž(2mβ€‹βˆ‘p≑mi​(M),iβ‰ cp≀Npβˆ’1+(2mβˆ’2)β€‹βˆ‘p≑mc​(M)p≀Npβˆ’1).\lim_{s\rightarrow 1^{+}}L(s)=\lim_{N\rightarrow\infty}\left(\frac{2}{m}\sum_{\begin{subarray}{c}p\equiv m_{i}(M),\ i\neq c\\ p\leq N\end{subarray}}p^{-1}+\left(\frac{2}{m}-2\right)\sum_{\begin{subarray}{c}p\equiv m_{c}(M)\\ p\leq N\end{subarray}}p^{-1}\right).

Define

sk\displaystyle s_{k} :=βˆ‘n=1kan\displaystyle:=\sum_{n=1}^{k}a_{n}
=2mβˆ‘iβ‰ c#{p≀k|pprime,p≑mi(modM)}+\displaystyle=\frac{2}{m}\sum_{i\neq c}\#\left\{p\leq k|\ p\ prime,\ p\equiv m_{i}\ (\bmod{M})\right\}+
(2mβˆ’2)#{p≀k|pprime,p≑mc(modM)}\displaystyle\left(\frac{2}{m}-2\right)\#\left\{p\leq k|\ p\ prime,\ p\equiv m_{c}\ (\bmod{M})\right\}

By the Siegel-Walfisz theorem[8], we have

#{p≀k|pprime,p≑mi(modM)}=1ϕ⁑(M)Li(k)+O(kβ‹…(logk)βˆ’2).\#\left\{p\leq k|\ p\ prime,\ p\equiv m_{i}\ (\bmod{M})\right\}=\frac{1}{\phi(M)}Li(k)+O(k\cdot\left(log\ k\right)^{-2}).

This gives sk=O⁑(kβ‹…(l​o​g​k)βˆ’2)s_{k}=O(k\cdot\left(log\ k\right)^{-2}).

βˆ‘k=1Nakks\displaystyle\sum_{k=1}^{N}\frac{a_{k}}{k^{s}} =βˆ‘k=1Nskβˆ’sk+1ks\displaystyle=\sum_{k=1}^{N}\frac{s_{k}-s_{k+1}}{k^{s}}
=βˆ‘k=1Nβˆ’1sk​{1ksβˆ’1(k+1)s}+sNNs\displaystyle=\sum_{k=1}^{N-1}s_{k}\left\{\frac{1}{k^{s}}-\frac{1}{(k+1)^{s}}\right\}+\frac{s_{N}}{N^{s}}
=βˆ‘k=1Nβˆ’1fk​(s)+sNNs.\displaystyle=\sum_{k=1}^{N-1}f_{k}(s)+\frac{s_{N}}{N^{s}}.

Where fk​(s):=sk​{1ksβˆ’1(k+1)s}f_{k}(s):=s_{k}\left\{\frac{1}{k^{s}}-\frac{1}{(k+1)^{s}}\right\}, and by the mean value theorem we have:

|fk​(s)|≀cβ‹…sβ‹…kβˆ’sβ‹…(l​o​g​k)βˆ’2,\left|f_{k}(s)\right|\leq c\cdot s\cdot k^{-s}\cdot\left(log\ k\right)^{-2},

where cc is a positive constant. This means βˆ‘fk​(s)\sum f_{k}(s) converges uniformally for s∈[1,2]s\in[1,2]. Hence we can exchange the limit and valuation at s=1s=1. ∎

In particular, we have cc>0c_{c}>0, since both l​o​g​(Ljc​(s))log(L_{j_{c}}(s)) and l​o​g​(΢⁑(s))log(\zeta(s)) are real when s>1s>1.

Given a=(a1,…,ah)∈Aa=(a_{1},\dots,a_{h})\in A, we have

LAa​(s)=12h​(L0​(s)+βˆ‘i=1h(βˆ’1)ai​Lji​(s)+o⁑((sβˆ’1)βˆ’1+2m)).L_{A_{a}}(s)=\frac{1}{2^{h}}\left(L_{0}(s)+\sum_{i=1}^{h}(-1)^{a_{i}}L_{j_{i}}(s)+o((s-1)^{-1+\frac{2}{m}})\right). (4)

Hence for distinct a,aβ€²a,a^{\prime}, LAa​(s)βˆ’LAa′​(s)L_{A_{a}}(s)-L_{A_{a}^{\prime}}(s)’s L0​(s)L_{0}(s) term cancels out, but Lj​(s)L_{j}(s), j∈A1j\in A_{1} terms need not. If they don’t, the sign of the leading term in the asymptotic expansion of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| is fixed.

Note that, for a=(0,…,0)a=(0,\dots,0), the coefficient of Lji​(s)L_{j_{i}}(s) in (4) is maximal for every ii, while for a=(1,…,1)a=(1,\dots,1), this coefficient is minimal. We deduce the following result.

Proposition 4.2.

For all aβ‰ (0,…,0)a\neq(0,\dots,0), |A(0,…,0)​(x)|βˆ’|Aa​(x)||A_{(0,\dots,0)}(x)|-|A_{a}(x)| is positive when xx is sufficiently large. For all aβ‰ (1,…,1)a\neq(1,\dots,1), |A(1,…,1)​(x)|βˆ’|Aa​(x)||A_{(1,\dots,1)}(x)|-|A_{a}(x)| is negative when xx is sufficiently large.

For given M,liM,l_{i}, we can always justify whether or not there is a bias by numeric method. We describe the method with the following example:

Let M=12M=12. Then m=4m=4. Let m1=1,m2=5,m3=7,m4=11m_{1}=1,\ m_{2}=5,\ m_{3}=7,\ m_{4}=11. Let l1=…=l4=2l_{1}=...=l_{4}=2. Let a=(1,0,0,0)a=(1,0,0,0) and aβ€²=(0,1,0,0)a^{\prime}=(0,1,0,0). Then LAa​(s)βˆ’LAa′​(s)=116​(βˆ’L(1,0,0,0)​(s)+L(0,1,0,0)​(s)+o⁑(x12))L_{A_{a}}(s)-L_{A_{a}^{\prime}}(s)=\frac{1}{16}(-L_{(1,0,0,0)}(s)+L_{(0,1,0,0)}(s)+o(x^{\frac{1}{2}})). We have:

Aji​(x)∼ciβ‹…x​(l​o​g​x)βˆ’12, 1≀i≀4A_{j_{i}}(x)\sim c_{i}\cdot x(log\ x)^{-\frac{1}{2}},\ 1\leq i\leq 4

Where:

c1=1Γ⁑(12)β€‹βˆp|12(1βˆ’pβˆ’1)12​limNβ†’βˆžβˆpβ‰’1​(12)(p,12)=1p≀N((1+pβˆ’1)​(1βˆ’pβˆ’1)12)β‹…βˆp≑1​(12)p≀N(1βˆ’pβˆ’1)32c_{1}=\frac{1}{\Gamma(\frac{1}{2})}\prod_{p|12}(1-p^{-1})^{\frac{1}{2}}\lim_{N\rightarrow\infty}\prod_{\begin{subarray}{c}p\not\equiv 1(12)\\ (p,12)=1\\ p\leq N\end{subarray}}\left((1+p^{-1})(1-p^{-1})^{\frac{1}{2}}\right)\cdot\prod_{\begin{subarray}{c}p\equiv 1(12)\\ p\leq N\end{subarray}}(1-p^{-1})^{\frac{3}{2}}
c2=1Γ⁑(12)β€‹βˆp|12(1βˆ’pβˆ’1)12​limNβ†’βˆžβˆpβ‰’5​(12)(p,12)=1p≀N((1+pβˆ’1)​(1βˆ’pβˆ’1)12)β‹…βˆp≑5​(12)p≀N(1βˆ’pβˆ’1)32c_{2}=\frac{1}{\Gamma(\frac{1}{2})}\prod_{p|12}(1-p^{-1})^{\frac{1}{2}}\lim_{N\rightarrow\infty}\prod_{\begin{subarray}{c}p\not\equiv 5(12)\\ (p,12)=1\\ p\leq N\end{subarray}}\left((1+p^{-1})(1-p^{-1})^{\frac{1}{2}}\right)\cdot\prod_{\begin{subarray}{c}p\equiv 5(12)\\ p\leq N\end{subarray}}(1-p^{-1})^{\frac{3}{2}}

Denote c=1Γ⁑(12)β€‹βˆp|12(1βˆ’pβˆ’1)12c=\frac{1}{\Gamma(\frac{1}{2})}\prod_{p|12}(1-p^{-1})^{\frac{1}{2}}, and cΒ―i:=cic\bar{c}_{i}:=\frac{c_{i}}{c}, the result for the computation is as follows (where we choose N=pmN=p_{m}, and pmp_{m} is the mm-th prime number):

mm 100 1,000 10,000 100,000 1,000,000
cΒ―1\bar{c}_{1} 1.16546 1.17505 1.1763 1.1766 1.17678
cΒ―2\bar{c}_{2} 0.744375 0.740081 0.740074 0.740005 0.739976

This shows that |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| is always negative when xx is sufficiently large. We will write a<aβ€²a<a^{\prime} in this case.

By computing to the one million-th prime, we have calculated that cΒ―3=0.854448,cΒ―4=1.01913\bar{c}_{3}=0.854448,\ \bar{c}_{4}=1.01913, and we have the following result:

(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,1,1,0)<\displaystyle(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,1,1,0)<
(0,1,1,1)<(1,0,0,1)<(1,0,1,0)<(1,1,0,0)<\displaystyle(0,1,1,1)<(1,0,0,1)<(1,0,1,0)<(1,1,0,0)<
(0,0,1,1)<(0,1,0,1)<(0,1,1,0)<(1,0,0,0)<\displaystyle(0,0,1,1)<(0,1,0,1)<(0,1,1,0)<(1,0,0,0)<
(0,0,0,1)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).\displaystyle(0,0,0,1)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).

We have also computated the case for M=5, 8, 10M=5,\ 8,\ 10, and li=2l_{i}=2 for all ii. Here cΒ―i\bar{c}_{i} is defined similarly in the above case, where

Aji​(x)∼1Γ⁑(12)β€‹βˆp|M(1βˆ’pβˆ’1)12​cΒ―iβ‹…x​(l​o​g​x)βˆ’12, 1≀i≀4A_{j_{i}}(x)\sim\frac{1}{\Gamma(\frac{1}{2})}\prod_{p|M}(1-p^{-1})^{\frac{1}{2}}\bar{c}_{i}\cdot x(log\ x)^{-\frac{1}{2}},\ 1\leq i\leq 4

and all calculations are done to the one million-th prime.

When M=5M=5, let m1=1,m2=2,m3=3,m4=4m_{1}=1,\ m_{2}=2,\ m_{3}=3,\ m_{4}=4. We have the following result:

cΒ―1=1.14715,cΒ―2=0.30951,cΒ―3=0.557665,cΒ―4=1.28255,\bar{c}_{1}=1.14715,\ \bar{c}_{2}=0.30951,\ \bar{c}_{3}=0.557665,\ \bar{c}_{4}=1.28255,
(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,0,0,1)<\displaystyle(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,0,0,1)<
(0,1,1,1)<(1,1,1,0)<(0,0,1,1)<(1,0,1,0)<\displaystyle(0,1,1,1)<(1,1,1,0)<(0,0,1,1)<(1,0,1,0)<
(0,1,0,1)<(1,1,0,0)<(0,0,0,1)<(1,0,0,0)<\displaystyle(0,1,0,1)<(1,1,0,0)<(0,0,0,1)<(1,0,0,0)<
(0,1,1,0)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).\displaystyle(0,1,1,0)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).

When M=8M=8, let m1=1,m2=3,m3=5,m4=7m_{1}=1,\ m_{2}=3,\ m_{3}=5,\ m_{4}=7. We have the following result:

cΒ―1=1.43848,cΒ―2=0.570731,cΒ―3=0.807147,cΒ―4=1.01717,\bar{c}_{1}=1.43848,\ \bar{c}_{2}=0.570731,\ \bar{c}_{3}=0.807147,\ \bar{c}_{4}=1.01717,
(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,1,1,0)<\displaystyle(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(1,1,1,0)<
(1,0,0,1)<(0,1,1,1)<(1,0,1,0)<(1,1,0,0)<\displaystyle(1,0,0,1)<(0,1,1,1)<(1,0,1,0)<(1,1,0,0)<
(0,0,1,1)<(0,1,0,1)<(1,0,0,0)<(0,1,1,0)<\displaystyle(0,0,1,1)<(0,1,0,1)<(1,0,0,0)<(0,1,1,0)<
(0,0,0,1)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).\displaystyle(0,0,0,1)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).

When M=10M=10, let m1=1,m2=3,m3=7,m4=9m_{1}=1,\ m_{2}=3,\ m_{3}=7,\ m_{4}=9. We have the following result:

cΒ―1=1.23125,cΒ―2=0.598548,cΒ―3=0.996604,cΒ―4=1.37658,\bar{c}_{1}=1.23125,\ \bar{c}_{2}=0.598548,\ \bar{c}_{3}=0.996604,\ \bar{c}_{4}=1.37658,
(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(0,1,1,1)<\displaystyle(1,1,1,1)<(1,0,1,1)<(1,1,0,1)<(0,1,1,1)<
(1,1,1,0)<(1,0,0,1)<(0,0,1,1)<(1,0,1,0)<\displaystyle(1,1,1,0)<(1,0,0,1)<(0,0,1,1)<(1,0,1,0)<
(0,1,0,1)<(1,1,0,0)<(0,1,1,0)<(0,0,0,1)<\displaystyle(0,1,0,1)<(1,1,0,0)<(0,1,1,0)<(0,0,0,1)<
(1,0,0,0)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).\displaystyle(1,0,0,0)<(0,0,1,0)<(0,1,0,0)<(0,0,0,0).

We conjecture that there is always a strict bias between |Aa​(x)||A_{a}(x)| and |Aa′​(x)||A_{a^{\prime}}(x)| when aa and aβ€²a^{\prime} are distinct, that is, there is a strict total ordering on A=(β„€/2​℀)hA=(\mathbb{Z}/\penalty 2\mathbb{Z})^{h}. Obviously, (0,…,0)(0,\dots,0) is the largest element in this ordering and (1,…,1)(1,\dots,1) is the smallest element in this ordering.

4.2 liβ‰₯3l_{i}\geq 3 for Some ii

Let ll be the largest value among li{l_{i}}. Then the second highest order pole among Lj​(s)L_{j}(s) would be 1m​((mβˆ’1)+ej)\frac{1}{m}((m-1)+e_{j}) and 1m​((mβˆ’1)+eΒ―j)\frac{1}{m}((m-1)+\bar{e}_{j}) (by highest we mean the value of its real part being largest).

Hence by 2.1 we know that Aj∼cjβ‹…x​R​e​((l​o​g​x)ejβˆ’1m)A_{j}\sim c_{j}\cdot xRe((log\ x)^{\frac{e_{j}-1}{m}}). Just like 4.1, this implies unless the coefficients happen to cancel out, we can expect the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| to change infinitely many times.

We’ll consider the following example:

Let M=4M=4, m1=1m_{1}=1, m2=3m_{2}=3, let l1=4l_{1}=4, and l2=1l_{2}=1. Let a=(1,0)a=(1,0) and aβ€²=(3,0)a^{\prime}=(3,0). Then LAa​(s)βˆ’LAa′​(s)=i2​(L(3,0)​(s)βˆ’L(1,0)​(s))=I​m​(L1,0​(s))L_{A_{a}}(s)-L_{A_{a}^{\prime}}(s)=\frac{i}{2}(L_{(3,0)}(s)-L_{(1,0)}(s))=Im(L_{1,0}(s)).

|Aa​(x)|βˆ’|Aa′​(x)|∼I​m​(1Γ⁑(1+i2)β‹…cβ‹…x​(l​o​g​x)βˆ’1+i2)|A_{a}(x)|-|A_{a^{\prime}}(x)|\sim Im\left(\frac{1}{\Gamma(\frac{1+i}{2})}\cdot c\cdot x(log\ x)^{\frac{-1+i}{2}}\right)

where

c\displaystyle c =(12)1+i2​limsβ†’1∏p≑1​(4)((1+i​pβˆ’s)​(1βˆ’pβˆ’s)1+i2)β€‹βˆp≑3​(4)((1+pβˆ’s)​(1βˆ’pβˆ’s)1+i2)\displaystyle=(\frac{1}{2})^{\frac{1+i}{2}}\lim_{s\rightarrow 1}\prod_{p\equiv 1(4)}((1+ip^{-s})(1-p^{-s})^{\frac{1+i}{2}})\prod_{p\equiv 3(4)}((1+p^{-s})(1-p^{-s})^{\frac{1+i}{2}})

As cβ‰ 0c\neq 0, the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| changes infinitely many times as xβ†’βˆžx\rightarrow\infty.

We conjecture that the sign of |Aa​(x)|βˆ’|Aa′​(x)||A_{a}(x)|-|A_{a^{\prime}}(x)| changes infinitely many times for all distinct aa, aβ€²a^{\prime}.

Acknowledgments

The author would like to thank Xi Ping for sharing the knowledge on Selberg-Delange method for this problem, and his advisor Wu Han for his guidance and advice on the writing of this paper.

References

  • [1] P. L. Boudec (2014) On the distribution of squarefree integers in arithmetic progressions. External Links: 1411.2360, Link Cited by: Β§1.
  • [2] H. Delange (1954) GΓ©nΓ©ralisation du thΓ©orΓ¨me de ikehara. Annales scientifiques de l’École Normale SupΓ©rieure 71 (3), pp.Β 213–242 (fre). External Links: Link Cited by: Β§2.
  • [3] G. H. Hardy and E. M. Wright (1938) An introduction to the theory of numbers.. (English). Note: Oxford: Clarendon Press. xvi, 403 p. (1938). Cited by: Β§1.
  • [4] C. Hooley (1975) A note on square-free numbers in arithmetic progressions. Bulletin of the London Mathematical Society 7 (2), pp.Β 133–138. External Links: Document, Link, https://londmathsoc.onlinelibrary.wiley.com/doi/pdf/10.1112/blms/7.2.133 Cited by: Β§1.
  • [5] E. Landau (1911) Handbuch der lehre von der verteilung der primzahlen. Monatshefte fΓΌr Mathematik und Physik 22 (1). External Links: ISSN 1436-5081, Document, Link Cited by: Β§1.
  • [6] R. M. Nunes (2014) Square-free numbers in arithmetic progressions. External Links: 1402.0684, Link Cited by: Β§1.
  • [7] G. Tenenbaum (2015) Introduction to analytic and probabilistic number theory. Vol. 163, American Mathematical Soc.. Cited by: Β§2.
  • [8] A. Walfisz (1936) Zur additiven zahlentheorie. ii.. Mathematische Zeitschrift 40 (1), pp.Β 592–607. External Links: ISSN 1432-1823, Document, Link Cited by: Β§4.1.