|
Karatsuba's divisor problem and related questions
M. R. Gabdullina, S. V. Konyagina, V. V. Iudelevichb a Steklov Mathematical Institute of Russian Academy of Sciences, Moscow, Russia
b Faculty of Mechanics and Mathematics, Lomonosov Moscow State University, Moscow, Russia
Abstract:
We prove that
∑p≤x1τ(p−1)≍x(logx)3/2and∑n≤x1τ(n2+1)≍x(logx)1/2,
where τ(n)=∑d∣n1 is the number of divisors of n, and the first sum is taken over prime numbers.
Bibliography: 14 titles.
Keywords:
divisor function, sums of values of functions, shifted primes and squares.
Received: 25.07.2022 and 31.03.2023
§ 1. Introduction In 2004 Karatsuba posed the following problem at his seminar “Analytic number theory and applications”: find the asymptotic behaviour of the sum as x→∞, where τ(n)=∑d∣n1 is the divisor function and the sum is taken over the prime numbers not exceeding x. This is a natural ‘hybrid’ of the following two classical problems in analytic number theory. The first (so-called Titchmarsh divisor problem) asks for the asymptotic behaviour of the sum It is well known (see [1]–[4]) that where ζ(s) denotes the Riemann zeta function. The other problem is to find the asymptotics for the sum This problem was solved by Ramanujan [5], who showed that
T(x)=c0x√logx(1+O(1√logx)),
where
c0=1√π∏p√p2−plogpp−1=0.5486….
The sum Φ(x) was studied previously. In the recent work [6] it was shown that
Φ(x)⩽4Kx(logx)3/2+O(xloglogx(logx)5/2),
where
K=1√π∏p√pp−1(plogpp−1−1p−1)=0.2532….
Note that the bound Φ(x)≪x(logx)3/2 follows from Corollary 1.2 in [7] and can also be derived by arguing similarly to the proof of the upper bound in Theorem 1.2 below. We make the conjecture that but this is probably hard to show. In this paper we prove that the aforementioned upper bound for Φ(x) is sharp up to a constant. Thus, we find the correct order of magnitude of the sum Φ(x): which answers Karatsuba’s question in the first approximation. In addition to Φ(x), we consider the sum and establish the following. Now we discuss the main ideas of the proofs. In the sum Φ(x), for each prime number p⩽x we write p−1 in the form ab, where a consists of the prime factors not exceeding z, and b has only prime factors greater than z; here z=xε for some small fixed ε>0. Now Φ(x) can be rewritten as
Φ(x)=∑a⩽xp∣a ⇒ p⩽z1τ(a)∑b⩽(x−1)/ap∣b ⇒ p>zab+1 is prime1τ(b),
and, since τ(b)=Oε(1), we have
Φ(x)≫ε∑a⩽xε1τ(a)∑b⩽(x−1)/ap∣b ⇒ p>xεab+1 is prime1.
Using the Brun-Hooley sieve one can estimate the inner sum from below by a quantity of order where the contribution from R(x;a) to the outer sum is negligible. Combining these estimates, we see that
Φ(x)≫εx(logx)2∑a⩽xε1aτ(a)≫εx(logx)3/2,
which gives us the required lower bound. Note that this argument does not yield any upper bound for Φ(x), since the estimation of the contribution of the numbers a>x1−ε to (1.2) is actually equivalent to the initial problem. A lower bound for the sum F(x) is obtained in a similar way. The upper bound for F(x) follows from Theorem 1 in [8]. We also note that the same upper bound can be derived from the inequality τ(n)⩾2ω(n) (here ω(n) denotes the number of distinct prime divisors of n) and a bound for the number of integers n⩽x with a fixed value of ω(n2+1). For completeness, we provide this argument. We observe that the methods used in this work can also be applied to other functions similar to τ(n). For instance, let τk(n) be the generalized divisor function, τk(n)=∑n=d1d2⋯dk1; then one can show that
∑p⩽x1τk(p−1)≍kx(logx)1/k−2.
§ 2. Notation By φ(n)=#{k⩽n:(k,n)=1} we denote the Euler totient function, and we use P+(n) and P−(n) for the least and greatest prime divisors of the integer n>1, respectively; by convention, P+(1)=0 and P−(1)=∞. We denote by π(x) the number of primes not exceeding x, and π(x;q,a) is the number of primes not exceeding x and belonging to the arithmetic progression a (modq); we also set R(x;q,a)=π(x;q,a)−π(x)/φ(q). The notation f(x)≪g(x), as well as f(x)=O(g(x)), means that |f(x)|⩽Cg(x) for some absolute constant C>0 and all possible values of x. We write f(x)≍g(x) if f(x)≪g(x)≪f(x), and we write f(x)≪kg(x) if we want to stress that the implied constant depends on k. Now we recall some notation from the sieve methods. Let A be a finite set of positive integers and P be a finite set of prime numbers. We define P=∏p∈Pp and S(A,P)=#{a∈A:(a,P)=1}. Also let Ad=#{a∈A:a≡0 (modd)}. We assume that, for all d∣P, where g(d) is a multiplicative function such that 0<g(p)<1 for p∈P and g(p)=0 for p∉P. Further, let the set P be divided into disjoint subsets P1,P2,…,Pt, and let Pj=∏p∈Pjp. Finally, let {kj}tr=1 be a sequence of even integers. Then we set
Vj=∏p∈Pr(1−g(p)),Lj=logV−1j,E=t∑j=1eLj(Lj)kj+1(kj+1)!,R=∑dj∣Pjω(dj)⩽kj|rd1⋯dt|andR′=t∑l=1∑dj∣Pjω(dj)⩽kj,j≠lω(dl)=kl+1|rd1⋯dt|.
§ 3. Auxiliary results We need the following version of the Brun-Hooley sieve, which is due to Ford and Halberstam [9]. Theorem 3.1. Let assumption (2.1) be met. Then and
S(A,P)⩾X∏p∈P(1−g(p))(1−E)−R−R′.
For a proof, see [9]. Lemma 3.2. Let a⩽x1/40 be an even positive integer, and let
Fa(x)=#{n⩽x−1a:an+1 is a prime number and P−(n)>x1/40}.
Then for x⩾x0, where c1>0 is an absolute constant and
0⩽R1⩽∑d⩽x13/40|R(x;ad,1)|.
Remark 3.3. Similar upper bounds without the error term R1 can be obtained by sifting the set
{(an+1)(n+P):n⩽x−1a},where P=∏p⩽zp.
Proof of Lemma 3.2. We apply Theorem 3.1 to the sets and P={p⩽z}, where z=x1/40 (the other parameters will be chosen later). Then Fa(x)=S(A,P), and for each d∣P=∏p⩽zp we have
Ad=#{k⩽x−1ad:adk+1 is prime}=π(x;ad,1)=π(x)φ(ad)+R(x;ad,1).
Now we write a number d∣P in the form d=d1d2, where (d1,a)=1 and d2 consists only of primes dividing a. Then, since φ(ad2)=ad2∏p∣a(1−1/p)=d2φ(a), we find that
1φ(ad)=1φ(d1)φ(ad2)=1φ(d1)d2φ(a),
hence (2.1) holds for X=π(x)/φ(a), the multiplicative function g defined by
g(p)={1p−1if (p,a)=1,1potherwise,
and In addition, we have with an absolute implied constant.
Now we choose a partition of P and define the numbers {kj}tj=1. Set zj=z21−j and then t is the unique positive integer such that zt+1<2⩽zt. Let kj=b+2(j−1), where b⩾2 is even and will be chosen later. We also set By [ 10], Theorem 7 (see (3.26) and (3.27) in that paper), for any x>1,
e−γlogx(1−1log2x)<∏p⩽x(1−1p)<e−γlogx(1+12log2x),
where γ is the Euler constant. Hence, for any z⩾√2 we have indeed, this follows from (3.1) for z⩾4 and can be checked manually for √2⩽z<4. Thus,
V−1j=∏zj+1<p⩽zj(1−g(p))−1⩽C∏zj+1<p⩽zj(1−1p)−1⩽3C⩽5.
Hence Lj=logV−1j⩽L=log5 and
E=t∑j=1eLjLkj+1j(kj+1)!⩽eLt∑j=1Lb+2j−1(b+2j−1)!.
Now we estimate R+R′. If an integer d corresponds to a summand in R, then d=d1⋯dt, where dj∣Pj and ω(dj)⩽kj=b+2(j−1). Therefore,
d⩽zk11⋯zktt⩽zb+(b+2)/2+(b+4)/4+⋯=z2b+4.
If a number d corresponds to a summand in R′, then we find similarly that d⩽z2b+4z=z2b+5. Since the numbers d corresponding to the sums R and R′ are distinct and do not exceed z2b+5, we see that Now we take b=4. Then 2b+5=13,
E⩽5∞∑j=1(log5)2j+3(2j+3)!⩽0.48
and
R+R′⩽∑d⩽x13/40|R(x;ad,1)|.
The claim follows. We set
M={a⩾1:a∣n2+1 for some n⩾1}.
It is well known that a∈M if and only if 4∤ and p \nmid a for any p \equiv 3\ (\operatorname{mod}4). Lemma 3.4. Let 2\leqslant a, z\leqslant x^{1/30}, a \in \mathcal{M} and P^{+}(a)\leqslant z. Let
\begin{equation*}
W_a(x,z)=\# \biggl\{ n \leqslant x\colon a\mid (n^2 + 1) \textit{ and } P^-\biggl(\frac{n^2+1}{a}\biggr) > z \biggr\}.
\end{equation*}
\notag
Then
\begin{equation*}
W_a(x,z) \asymp \frac{2^{\omega(a)}}{\varphi(a)}\,\frac{x}{\log z} .
\end{equation*}
\notag
Proof. Setting \mathcal{A}=\{k\colon ak=n^2+1 \text{ for some } n\leqslant x\} and
\begin{equation*}
\mathcal{P}= \begin{cases} \{p \leqslant z\colon p\equiv 1\ (\operatorname{mod}4)\} &\text{if $a$ is even}, \\ \{2\}\cup \{p \leqslant z\colon p\equiv 1\ (\operatorname{mod}4)\} &\text{if $a$ is odd}. \end{cases}
\end{equation*}
\notag
We have W_a(x,z) = S(\mathcal{A}, \mathcal{P}). As before, we write each d dividing P=\prod_{p\in\mathcal{P}}p as d=d_1d_2, where (d_1,a)=1 and d_2 consists only of primes dividing a. Suppose that one of the integers d and a is even. Then by the Chinese remainder theorem and the fact that the congruence x^2+1\equiv0\ (\operatorname{mod}p) has two solutions for p\equiv 1\ (\operatorname{mod}4) and one solution for p=2, we obtain
\begin{equation*}
\begin{aligned} \, \mathcal{A}_d &=\#\{n\leqslant x\colon n^2+1 \equiv 0 \ (\operatorname{mod}ad) \} \\ &= \frac{x2^{\omega(ad)-1}}{ad}+O(2^{\omega(ad)}) =\frac{x2^{\omega(a)+\omega(d_1)-1}}{ad_1d_2}+O(2^{\omega(ad)}). \end{aligned}
\end{equation*}
\notag
If both a and d are odd, then a similar argument shows that
\begin{equation*}
\mathcal{A}_d= \frac{x2^{\omega(a)+\omega(d_1)}}{ad_1d_2}+O(2^{\omega(ad)}).
\end{equation*}
\notag
In both cases we have
\begin{equation*}
\mathcal{A}_d=\frac{x2^{\omega(a)+\omega(d_1)-\mathbb I(2\mid ad)}}{ad_1d_2}+O(2^{\omega(ad)}) =\frac{x2^{\omega(a)-\mathbb I(2\mid a)}}{a}\frac{2^{\omega(d_1)-\mathbb I(2\mid d_1)}}{d_1d_2}+O(\tau(ad)),
\end{equation*}
\notag
where \mathbb I(2\mid l) is equal to one if l is even and to zero otherwise. Thus we see that condition (2.1) holds for X= (x2^{\omega(a)-\mathbb I(2\mid a)})/a, the multiplicative function g defined on the prime numbers in \mathcal{P} by
\begin{equation*}
g(p)=\begin{cases} \dfrac{2}{p} &\text{if } p\nmid a, \\ \dfrac{1}{p} &\text{if } p\mid a \end{cases}
\end{equation*}
\notag
(and also g(2)=1/2 in the case of odd a), and
\begin{equation*}
r_d=O(\tau(ad)).
\end{equation*}
\notag
It is well known that
\begin{equation}
\prod _{\substack{p\leqslant x \\ p\equiv 1\, (\operatorname{mod}4)}}\biggl(1-\frac{1}{p} \biggr)\asymp \frac{1}{\sqrt{\log x}}
\end{equation}
\tag{3.3}
(see [11]). Thus, in both cases in the definition of \mathcal{P} we have
\begin{equation*}
\prod_{p \in \mathcal{P}}(1-g(p))\asymp \prod_{p\in \mathcal{P}, \, p>2}\biggl(1-\frac2p\biggr) \prod_{p\mid a,\,p>2}\frac{1-1/p}{1-2/p} \asymp \frac{a}{\varphi(a)\log z}
\end{equation*}
\notag
for an absolute implied constant.
Now we choose a partition of \mathcal{P} and define the numbers \{k_j\}_{j=1}^t. Again, we set z_j= z^{2^{1-j }} and
\begin{equation*}
\mathcal{P}_j=\mathcal{P} \cap (z_{j+1}, z_j],
\end{equation*}
\notag
where t is the unique positive integer such that z_{t+1} < 2 \leqslant z_t and k_j = b + 2(j-1) for some even b \geqslant 2. Now (3.2) implies that
\begin{equation*}
\begin{aligned} \, V_j^{-1} &=\prod_{z_{j+1} < p \leqslant z_j} (1 - g(p))^{-1} \leqslant 2\prod_{z_{j+1}<p\leqslant z_j, \, p\neq2}\biggl(1-\frac2p\biggr)^{-1} \\ &=2\prod_{z_{j+1}<p\leqslant z_j, \, p\neq2}\biggl(1-\frac1p\biggr)^{-2}\frac{(1-1/p)^2}{1-2/p} \leqslant 18C\leqslant 28. \end{aligned}
\end{equation*}
\notag
Therefore, L_j = \log V_j^{-1}\leqslant L=\log 28 and
\begin{equation*}
E=\sum_{j=1}^t \frac{e^{L_j} L_j^{k_j+1}}{(k_j+1)!} \leqslant e^L\sum_{j=1}^t\frac{L^{b+2j-1}}{(b+2j-1)!}.
\end{equation*}
\notag
Now we take b=10; then E\leqslant 0.43 and 2b+5=25. As in the proof of Lemma 3.2, we obtain
\begin{equation*}
R+R' \leqslant \sum_{d \leqslant z^{2b+5}} |r_d|\ll \tau(a)\sum_{d\leqslant z^{25}}\tau(d)\ll x^{1/30}z^{25} \ll x^{26/30},
\end{equation*}
\notag
and the main term is at least
\begin{equation*}
X\prod_{p\in \mathcal{P}}(1-g(p)) \asymp\frac{x2^{\omega(a)}}{\varphi(a)\log z} \gg\frac{x^{1-1/30}}{\log x}.
\end{equation*}
\notag
An application of Theorem 3.1 completes the proof. The next lemma is a special case of a result due to Tenenbaum [12]. For completeness, we provide the proof. Lemma 3.5. Let R>0 be fixed. Then there exist positive constants A=A(R) and B=B(R) such that, for all 1\leqslant k \leqslant R\log\log x,
\begin{equation*}
\#\bigl\{n\leqslant x\colon \omega(n^2+1)=k\bigr\} \leqslant \frac{Ax(\log\log x+B)^{k-1}}{(k-1)!\, \log x}.
\end{equation*}
\notag
Proof. For m\geqslant 2 and y\geqslant2 we define the ‘y-smooth part’ of m by
\begin{equation*}
d(m,y)=\max\bigl\{d\mid m\colon P^+(d)\leqslant y\bigr \}.
\end{equation*}
\notag
For each n\leqslant x the number n^2+1 can be written as the product ab, where
\begin{equation*}
a=a(n)=\max\bigl\{d(n^2+1,y)\colon d(n^2+1,y)\leqslant x^{1/30} \bigr\}
\end{equation*}
\notag
and b=b(n)=(n^2+1)/a; note that (a,b)=1. Now we set
\begin{equation*}
\begin{gathered} \, A_1 =\bigl\{ n\leqslant x\colon a\leqslant x^{1/60},\, P^-(b)>x^{1/60},\, b>1\bigr\}, \\ A_2 =\bigl\{ n\leqslant x\colon a\leqslant x^{1/60},\, P^-(b)\leqslant x^{1/60},\, b>1\bigr\} \end{gathered}
\end{equation*}
\notag
and
\begin{equation*}
A_3 =\bigl\{ n\leqslant x\colon x^{1/60}< a \leqslant x^{1/30},\, b>1 \bigr\}.
\end{equation*}
\notag
Since the number of integers n\leqslant x for which b(n)=1 does not exceed x^{1/60}, we conclude that
\begin{equation}
\#\bigl\{n\leqslant x\colon \omega(n^2+1)=k\bigr\}=N_1+N_2+N_3+O(x^{1/60}),
\end{equation}
\tag{3.4}
where
\begin{equation*}
N_i=\#\bigl\{n\in A_i\colon \omega(n^2+1)=k \bigr\}, \qquad i=1, 2, 3.
\end{equation*}
\notag
First we estimate N_1. In this case 1\leqslant \omega(b)\leqslant 120; by Lemma 3.4 we have
\begin{equation}
\begin{aligned} \, \notag N_1 &\leqslant \sum_{l=k-120}^{k-1}\sum_{\substack{a\leqslant x^{1/60}, \, a\in \mathcal{M} \\ \omega(a)=l }} W_a(x, x^{1/60}) \ll \sum_{l=k-120}^{k-1}\sum_{\substack{a\leqslant x^{1/60}, \, a\in \mathcal{M} \\ \omega(a)=l }}\frac{x2^{\omega(a)}}{\varphi(a)\log x} \\ &\leqslant \frac{x}{\log x}\sum_{l=k-120}^{k-1}2^l\sum_{\substack{a\leqslant x^{1/60}, \, a\in \mathcal{M} \\ \omega(a)=l }}\frac{1}{\varphi(a)}. \end{aligned}
\end{equation}
\tag{3.5}
Taking the logarithm of both sides of (3.3) we obtain
\begin{equation*}
\sum _{\substack{p\leqslant x \\ p\equiv 1\, (\operatorname{mod}4)}}\frac{1}{p}= \frac{1}{2}\log\log x+O(1).
\end{equation*}
\notag
Hence for each l\geqslant 0 the inner sum in (3.5) is at most
\begin{equation*}
\frac{1}{l!}\biggl(\,\sum_{p\leqslant x^{1/60},\, p\neq 3\, (\operatorname{mod}4)}\frac{1}{\varphi(p)}+\frac{1}{\varphi(p^2)}+\dotsb \biggr)^l \leqslant \frac{1}{l!}\biggl(\frac{1}{2}\log\log x+B_1\biggr)^l
\end{equation*}
\notag
for some absolute positive constant B_1. Thus,
\begin{equation}
N_1\ll \frac{x(\log\log x+B_1)^{k-1}}{(k-1)!\,\log x}\bigl(1+R+\dots+R^{120} \bigr).
\end{equation}
\tag{3.6}
Now we estimate N_2. Let p=P^-(b), and let r be the largest integer such that p^r\mid b. Then by the definition of a, we have ap^r>x^{1/30}. Thus, for any n\in A_2 we have p^r> x^{1/60}, and since p\leqslant x^{1/60}, we see that r\geqslant 2. Let \nu = \min\{{u\geqslant 1}: {p^u>x^{1/60}}\}. Then 2\leqslant \nu\leqslant r and p^{\nu-1}\leqslant x^{1/60}. Hence {p^\nu\leqslant x^{1/60}p\leqslant x^{1/30}}. We conclude that for each n\in A_2 the integer n^2+1 is divisible by p^\nu, where p is a prime number such that x^{1/60}<p^\nu\leqslant x^{1/30} and \nu\geqslant 2. Therefore,
\begin{equation*}
N_2\leqslant\sum_{x^{1/60}<p^\nu\leqslant x^{1/30},\,\nu\geqslant 2}\biggl( \frac{2x}{p^\nu}+O(1)\biggr)\ll \sum_{x^{1/60}<p^\nu \leqslant x^{1/30},\,\nu\geqslant 2}\frac{x}{p^\nu}.
\end{equation*}
\notag
Note that p^{\nu} \geqslant \max\{p^2, x^{1/60}\}, and therefore p^{\nu}\geqslant px^{1/120}. It follows that
\begin{equation}
N_2\ll \sum_{p\leqslant x^{1/30}}\frac{x}{px^{1/120}} \ll x^{1-1/120}\log\log x.
\end{equation}
\tag{3.7}
Finally, consider N_3. Setting q=P^+(a), we have P^-(b)\geqslant q+1 and
\begin{equation}
\omega(b)\leqslant \frac{\log (x^2+1)}{\log (q+1)}\leqslant \frac{2\log x}{\log q}=:\eta.
\end{equation}
\tag{3.8}
Lemma 3.4 implies that
\begin{equation}
\begin{aligned} \, N_3 &\leqslant \sum_{\substack{q\leqslant x^{1/30} \\ q\not{\equiv} 3\, (\operatorname{mod}4)}}\sum_{k-\eta\leqslant l\leqslant k-1}\sum_{\substack{x^{1/60}<a\leqslant x^{1/30} \\ P^+(a)=q, \, \omega(a)=l}}W_a(x,q) \nonumber \\ &\ll x\sum_{\substack{q\leqslant x^{1/30} \\ q\not{\equiv} 3\, (\operatorname{mod}4)}}\frac{1}{\log q}\sum_{k-\eta\leqslant l\leqslant k-1}2^l\sum_{\substack{x^{1/60}<a\leqslant x^{1/30} \\ P^+(a)=q, \, \omega(a)=l}}\frac{1}{\varphi(a)}. \end{aligned}
\end{equation}
\tag{3.9}
Let
\begin{equation}
\delta=\frac{C}{\log q}, \quad\text{where } C=120\log(R+2)+60.
\end{equation}
\tag{3.10}
Let us denote by N_3^{(1)} and N_3^{(2)} the contributions from q\leqslant e^{2C} and q>e^{2C}, respectively. Since
\begin{equation*}
\varphi(a)\gg\frac{a}{\log\log a}\gg\frac{x^{1/60}}{\log\log x},
\end{equation*}
\notag
we have
\begin{equation}
N_3^{(1)}\ll x^{59/60}\log\log x\sum_{q\leqslant e^{2C}}\sum_{1\leqslant l\leqslant \pi(q)}2^l\sum_{\substack{a\leqslant x^{1/30} \\ P^+(a)=q}}1 \ll x^{59/60}(\log x)^{c_R},
\end{equation}
\tag{3.11}
where c_R>0 depends only on R. Now we turn to N_3^{(2)}. For fixed q and l we see that
\begin{equation*}
\begin{aligned} \, S_{q,l} &:=\sum_{\substack{x^{1/60}<a\leqslant x^{1/30} \\ P^+(a)=q, \, \omega(a)=l}}\frac{1}{\varphi(a)} \leqslant \sum_{\substack{a\leqslant x^{1/30} \\ P^+(a)=q, \, \omega(a)=l}}\biggl(\frac{a}{x^{1/60}}\biggr)^{\delta}\frac{1}{\varphi(a)} \\ &\leqslant x^{-\delta/60}\frac{1}{(l-1)!}\biggl(\sum_{\substack{p<q \\ p\not{\equiv} 3\, (\operatorname{mod}4)}}\frac{p^{\delta}}{\varphi(p)} +\frac{p^{2\delta}}{\varphi(p^2)}+\dotsb \biggr)^{l-1}\biggl(\frac{q^{\delta}}{\varphi(q)} +\frac{q^{2\delta}}{\varphi(q^2)}+\dotsb\biggr). \end{aligned}
\end{equation*}
\notag
Note that
\begin{equation*}
\frac{q^{\delta}}{\varphi(q)}+\frac{q^{2\delta}}{\varphi(q^2)}+\dotsb \ll_R \frac1q \quad\text{for } q>e^{2C},
\end{equation*}
\notag
and the sum over the primes p is (since e^u\leqslant 1+O_K(u) for 0\leqslant u\leqslant K) equal to
\begin{equation*}
\begin{aligned} \, &\sum_{\substack{p<q \\ p\not{\equiv} 3\, (\operatorname{mod}4)}}\frac{p^\delta}{p}+O_R(1) \\ &\qquad=\sum_{\substack{p<q \\ p\not{\equiv} 3\, (\operatorname{mod}4)}}\frac{1}{p}+ O_R\biggl(\delta\sum_{\substack{p<q \\ p\not{\equiv} 3\, (\operatorname{mod}4)}}\frac{\log p}{p}+1\biggr)\leqslant \frac{1}{2}\log\log x+B, \end{aligned}
\end{equation*}
\notag
where B=B(R)>0. Thus,
\begin{equation*}
S_{q,l}\ll_R \frac{x^{-\delta/60}}{q(l-1)!}\biggl(\frac{1}{2}\log\log x+B\biggr)^{l-1}
\end{equation*}
\notag
and
\begin{equation}
\begin{aligned} \, N_3^{(2)} &\ll_R x\sum_{\substack{e^{2C}<q\leqslant x^{1/30} \\ q\not\equiv 3\, (\operatorname{mod}4)}}\frac{x^{-\delta/60}}{q\log q}\sum_{k-\eta\leqslant l\leqslant k-1}\frac{2^l(1/2\log\log x+B)^{l-1}}{(l-1)!}\notag \\ &\ll \frac{x(\log\log x+2B)^{k-1}}{(k-1)!}\sum_{q\leqslant x^{1/30}}\frac{x^{-\delta/60}}{q\log q}(1+R+\dots +R^{[\eta]}). \end{aligned}
\end{equation}
\tag{3.12}
Using the inequality
\begin{equation*}
1+R+\dots+R^{[\eta]}\leqslant (R+2)^{\eta+1}
\end{equation*}
\notag
and the fact that
\begin{equation*}
(R+2)^\eta x^{-\delta/60}=x^{-1/\log q}
\end{equation*}
\notag
(which follows from the definitions (3.8) and (3.10) of \eta and \delta, respectively), we see that the sum over primes q in (3.12) is at most
\begin{equation*}
\begin{aligned} \, &(R+2) \sum_{q\leqslant x^{1/30}}\frac{(R+2)^{\eta}x^{-\delta/60}}{q\log q} \leqslant (R+2)\sum_{q\leqslant x^{1/2}}\frac{x^{-1/\log q}}{q\log q} \\ &\qquad\leqslant \frac{R+2}{\log x}\sum_{j\geqslant 1}\sum_{q\in (x^{2^{-(j+1)}}, \,x^{-2^j}]}\frac1q \,2^{j+1}\exp(-2^j) \ll_R \frac{1}{\log x}\,. \end{aligned}
\end{equation*}
\notag
Substituting this into (3.12) and taking (3.11) into account we find that
\begin{equation}
N_3 \ll_R \frac{x(\log\log x+2B)^{k-1}}{(k-1)!\,\log x}.
\end{equation}
\tag{3.13}
Combining (3.4), (3.6), (3.7) and (3.13) we complete the proof.
§ 4. Proof of Theorem 1.1 Writing p-1=ab, where P^+(a)\leqslant x^{1/40} and P^-(b)>x^{1/40}, we can rewrite the sum \Phi(x) as
\begin{equation*}
\Phi(x)=\sum_{\substack{a \leqslant x \\ P^+(a) \leqslant x^{1/40}}} \frac{1}{\tau(a)} \sum_{\substack{b \leqslant (x-1)/a \\ P^-(b) > x^{{1}/{40}} \\ ab +1\text{ is prime}}} \frac{1}{\tau(b)}.
\end{equation*}
\notag
Let b = p_1^{\alpha_1} p_2^{\alpha_2} \dotsb p_s^{\alpha_s} be the prime factorization of b. Then
\begin{equation*}
x^{ (\alpha_1 + \alpha_2 + \dots + \alpha_s)/40} < b \leqslant x,
\end{equation*}
\notag
so that \alpha_1+\dots+\alpha_s < 40 and
\begin{equation*}
\tau(b)=(\alpha_1+1) \dotsb (\alpha_s+1) \leqslant 2^{\alpha_1+\dots+\alpha_s} < 2^{40}.
\end{equation*}
\notag
Therefore,
\begin{equation*}
\Phi(x) \geqslant 2^{-40} \sum_{\substack{a \leqslant x \\ P^+(a) \leqslant x^{1/40}}} \frac{1}{\tau(a)} \sum_{\substack{b \leqslant (x-1)/a \\ P^-(b) > x^{1/40} \\ ab +1\text{ is prime}}} 1 \geqslant 2^{-40} \sum_{\substack{a \leqslant x^{1/40} \\ a \text{ is even}}} \frac{1}{\tau(a)}F_a(x) .
\end{equation*}
\notag
Lemma 3.2 implies that
\begin{equation*}
\Phi(x) \geqslant \frac{c_2 \pi(x)}{\log x}\sum_{\substack{a \leqslant x^{1/40} \\ a \text{ is even}}} \frac{1}{\tau(a)\varphi(a)}-R_2,
\end{equation*}
\notag
where c_2>0 and
\begin{equation*}
0\leqslant R_2 \leqslant \sum_{a\leqslant x^{1/40}}\sum_{d\leqslant x^{13/40}}|R(x;ad,1)| \leqslant \sum_{q\leqslant x^{7/20}}\tau(q)|R(x;q,1)|.
\end{equation*}
\notag
Using the trivial bound |R(x;q,1)|\ll x/q and the Cauchy-Schwarz inequality we obtain
\begin{equation*}
\begin{aligned} \, R_2 &\ll x^{1/2} \sum_{q \leqslant x^{0.35}} \frac{\tau(q)}{q^{1/2}} (|R(x; q, 1)|)^{1/2} \\ &\leqslant x^{1/2} \biggl(\sum_{q \leqslant x^{0.35}} \frac{\tau^2(q)}{q}\biggr)^{1/2} \biggl(\sum_{q \leqslant x^{0.35}} |R(x; q, 1)|\biggr)^{1/2}. \end{aligned}
\end{equation*}
\notag
Applying the bound
\begin{equation*}
\sum_{n\leqslant x}\frac{\tau^2(n)}{n} \asymp (\log x)^4,
\end{equation*}
\notag
which follows from \sum_{n\leqslant x}\tau^2(n)\asymp x(\log x)^3 (see [13], Ch. III, Exercise 7) by partial summation, and the Bombieri-Vinogradov theorem we find that, for any fixed {A>0},
\begin{equation*}
R_2 \ll_A x^{1/2}(\log x)^2 \frac{x^{1/2}}{(\log x)^{A/2}}= \frac{x}{(\log x)^{A/2-2}}.
\end{equation*}
\notag
Taking A = 24, say, we obtain
\begin{equation*}
R_2 \ll \frac{x}{(\log x)^{10}}.
\end{equation*}
\notag
Now we turn to the sum
\begin{equation*}
T_1=\sum_{\substack{a \leqslant x^{1/40} \\ a \text{ is even}}} \frac{1}{\tau(a) \varphi(a)}=\sum_{l \leqslant 0.5x^{1/40}} \frac{1}{\tau(2l)\varphi(2l)}.
\end{equation*}
\notag
Since \tau(mn) \leqslant \tau(m)\tau(n) and \varphi(n) \leqslant n, we see that
\begin{equation*}
T_1\gg \sum_{l \leqslant 0.5x^{1/40}} \frac{1}{\tau(l)l}.
\end{equation*}
\notag
By (1.1), using partial summation we have
\begin{equation*}
T_1\gg (\log x)^{1/2}.
\end{equation*}
\notag
Hence
\begin{equation*}
\Phi(x)\gg \frac{\pi(x)}{(\log x)^{1/2}}-O\biggl( \frac{x}{(\log x)^{10}}\biggr)\gg \frac{x}{(\log x)^{3/2}}.
\end{equation*}
\notag
Theorem 1.1 is proved.
§ 5. Proof of Theorem 1.2 First we prove the lower bound. For any z\geqslant2,
\begin{equation*}
F(x)=\sum_{n \leqslant x} \frac{1}{\tau(n^2+1)}=\sum_{\substack{ab=n^2 + 1,\,n \leqslant x\\ P^+(a) \leqslant z,\, P^-(b) > z}} \frac{1}{\tau(ab)}.
\end{equation*}
\notag
Setting z = x^{1/30} we have \tau(b) \leqslant 2^{60}, and now by Lemma 3.4,
\begin{equation*}
\begin{aligned} \, F(x) &\geqslant 2^{-60} \sum_{\substack{a \leqslant x^{1/30} \\ a \in \mathcal{M}}} \frac{1}{\tau(a)} W_a(x, x^{1/30})\gg \frac{x}{\log x} \sum_{\substack{a \leqslant x^{1/30} \\ a \in \mathcal{M}}} \frac{2^{\omega(a)}}{\tau(a)\varphi(a)} \\ &\geqslant \frac{x}{\log x}\sum_{\substack{a \leqslant x^{1/30} \\ a \in \mathcal{M} \\ a\text{ is square-free}}}\frac{1}{a}= \frac{x}{\log x}\sum_{\substack{a \leqslant x^{1/30} \\ a \in \mathcal{M}}} \frac{1}{a}\,\sum_{\delta^2\mid a}\mu(\delta). \end{aligned}
\end{equation*}
\notag
Changing the order of summation and using the fact that
\begin{equation*}
\sum_{n=1}^{\infty}\frac{1}{n^2}=\frac{\pi^2}{6},
\end{equation*}
\notag
we have
\begin{equation}
F(x)\gg \frac{x}{\log x}\sum_{\substack{\delta \leqslant x^{1/60} \\ \delta \in \mathcal{M}}} \frac{\mu(\delta)}{\delta^2} \sum_{\substack{a \leqslant x^{1/30}\delta^{-2} \\ a \in \mathcal{M} }} \frac{1}{a} \geqslant \biggl( 2-\frac{\pi^2}{6}\biggr)\frac{x}{\log x}\sum_{\substack{a \leqslant x^{1/30} \\ a \in \mathcal{M}}} \frac{1}{a}.
\end{equation}
\tag{5.1}
It is well known (see [14], § 183) that for y\geqslant 2 one has
\begin{equation*}
\# \{a\leqslant y\colon p\mid a \ \Rightarrow\ p\equiv 1\ (\operatorname{mod} 4)\} \asymp \frac{y}{(\log y)^{1/2}}.
\end{equation*}
\notag
Using partial summation we obtain
\begin{equation*}
\sum_{\substack{a\leqslant y \\ p\mid a \ \Rightarrow\ p\equiv 1\, (\operatorname{mod} 4) }}\frac{1}{a}\asymp (\log y)^{1/2}.
\end{equation*}
\notag
Applying this to (5.1) we find that
\begin{equation*}
F(x)\gg \frac{x}{(\log x)^{1/2}},
\end{equation*}
\notag
as required. Now we prove the upper bound. By Lemma 3.5,
\begin{equation*}
\begin{aligned} \, F(x)&\leqslant \sum_{n\leqslant x}\frac{1}{2^{\omega(n^2+1)}} \\ &=\sum_{k\leqslant 2\log\log x}\frac{\#\{n\leqslant x\colon \omega(n^2+1)=k\}}{2^k} + O\biggl(\sum_{k>2\log\log x}\frac{x}{2^k}\biggr) \\ &\ll \frac{x}{\log x}\sum_{k=1}^{+\infty}\frac{((1/2)\log\log x+(1/2)B)^{k-1}}{(k-1)!} + O\biggl(\frac{x}{\log x}\biggr)\ll \frac{x}{(\log x)^{1/2}}. \end{aligned}
\end{equation*}
\notag
Theorem 1.2 is proved.
|
|
|
Bibliography
|
|
|
1. |
E. C. Titchmarsh, “A divisor problem”, Rend. Circ. Mat. Palermo, 54 (1930), 414–429 |
2. |
Yu. V. Linnik, “New versions and new uses of the dispersion method in binary additive problems”, Soviet Math. Dokl., 2 (1961), 468–471 |
3. |
H. Halberstam, “Footnote to the Titchmarsh-Linnik divisor problem”, Proc. Amer. Math. Soc., 18 (1967), 187–188 |
4. |
E. Bombieri, J. B. Friedlander and H. Iwaniec, “Primes in arithmetic progressions to large moduli”, Acta Math., 156:3-4 (1986), 203–251 |
5. |
S. Ramanujan, “Some formulae in the analytic theory of numbers”, Messenger Math., 45 (1916), 81–84 |
6. |
V. V. Iudelevich, “On the Karatsuba divisor problem”, Izv. Math., 86:5 (2022), 992–1019 |
7. |
P. Pollack, “Nonnegative multiplicative functions on sifted sets, and the square roots of -1 modulo shifted primes”, Glasg. Math. J., 62:1 (2020), 187–199 |
8. |
M. B. Barban and P. P. Vekhov, “Summation of multiplicative functions of polynomials”, Math. Notes, 5:6 (1969), 400–407 |
9. |
K. Ford and H. Halberstam, “The Brun-Hooley sieve”, J. Number Theory, 81:2 (2000), 335–350 |
10. |
J. B. Rosser and L. Schoenfeld, “Approximate formulas for some functions of prime numbers”, Illinois J. Math., 6:1 (1962), 64–94 |
11. |
S. Uchiyama, “On some products involving primes”, Proc. Amer. Math. Soc., 28:2 (1971), 629–630 |
12. |
G. Tenenbaum, “Note sur les lois locales conjointes de la fonction nombre de facteurs premiers”, J. Number Theory, 188 (2018), 88–95 |
13. |
K. Prachar, Primzahlverteilung, Springer-Verlag, Berlin–Göttingen–Heidelberg, 1957, x+415 pp. |
14. |
E. Landau, Handbuch der Lehre von der Verteilung der Primzahlen, 2 Bände, B. G. Teubner, Leipzig–Berlin, 1909, x+564 pp., ix+567–961 pp. |
Citation:
M. R. Gabdullin, S. V. Konyagin, V. V. Iudelevich, “Karatsuba's divisor problem and related questions”, Sb. Math., 214:7 (2023), 919–933
Linking options:
https://www.mathnet.ru/eng/sm9815https://doi.org/10.4213/sm9815e https://www.mathnet.ru/eng/sm/v214/i7/p27
|
Statistics & downloads: |
Abstract page: | 530 | Russian version PDF: | 40 | English version PDF: | 46 | Russian version HTML: | 266 | English version HTML: | 135 | References: | 45 | First page: | 13 |
|