Improved Averaged Distribution of d 3 ( n ) d_{3}(n) in Prime Arithmetic Progressions
Metin Can Aydemir, Muhammet Boran · 2026
Read Improved Averaged Distribution of d 3 ( n ) d_{3}(n) in Prime Arithmetic Progressions by Metin Can Aydemir, Muhammet Boran (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-12601.txt.
Introduction
Metin Can Aydemir and Muhammet Boran improve a result in analytic number theory about a divisor function. The function d3 counts ways to write a number as a product of three factors. They study how its values spread across prime arithmetic progressions on average. The key yardstick is the exponent of distribution, a measure of how far the average stays well behaved.
Their title claims an improved exponent for this function. A higher exponent means the good average holds for larger moduli. Results of this type feed sieve methods, a core engine for prime number results. Even small gains in an exponent can unlock new theorems downstream. The work states its claims in the uniform form that sieve users need.
Catalog record
| CREATOR | Metin Can Aydemir, Muhammet Boran |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-12601
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Improved Averaged Distribution of d 3 ( n ) d_{3}(n) in Prime Arithmetic Progressions Metin Can Aydemir, Muhammet Boran Abstract . We say that d 3 ( n ) d_{3}(n) has exponent of distribution θ \theta if, for all ε > 0 \varepsilon>0 , the expected asymptotic holds uniformly for all moduli q ≤ x θ − ε q\leq x^{\theta-\varepsilon} . Nguyen proved that, after averaging over reduced residue classes a mod q a\bmod q , the function d 3 ( n ) d_{3}(n) has exponent of distribution 2 / 3 2/3 , following earlier work of Banks et al. Using the Petrow–Young subconvexity bound for Dirichlet L L -functions, we improve this to an exponent of distribution 8 / 11 8/11 when averaging over residue classes modulo a prime q q . 1991 Mathematics Subject Classification Primary 11N37, Secondary 11M06. 1. Introduction 1.1. Historical Context The study of the distribution of divisor functions in arithmetic progressions has a rich history, with significant breakthroughs occurring in recent decades. Definition 1.1 . For any n,\ k\in\mathbb{Z}^{+} , the k -fold divisor function d_{k}(n) counts the number of ways to write n as a product of k positive integers, d_{k}(n)\ :=\ \sum_{n_{1}n_{2}\dots n_{k}=n}1, where each n_{i}\in\mathbb{Z}^{+} . Definition 1.2 . Let n be a fixed positive integer and consider the sum of the n th powers of the primitive q th roots of unity. This sum is known as Ramanujan’s sum and is denoted by c_{q}(n)\ :=\ \sideset{}{{}^{\prime}}{\sum}_{m\ (q)}e\Big(\frac{mn}{q}\Big), where e(z):=e^{2\pi iz} for any z\in\mathbb{R} , and \sideset{}{{}^{\prime}}{\sum}\limits_{m\ (q)} indicates that m runs over residues coprime to q . It is known that this sum reduces to the Möbius function when n is coprime to q , and c_{q}(n)=\mu(q) . Classical results by Selberg, Linnik, and Hooley (independently, 1950s, unpublished) established that the usual divisor function d(n) has an exponent of distribution 2/3 . For the ternary divisor function d_{3}(n) , early progress was much harder: Friedlander and Iwaniec [ FI ] achieved the first development beyond 1/2 , attaining an exponent of distribution 1/2+1/230 . This was subsequently improved to 1/2+1/82 by Heath-Brown [ HB ] , and to 1/2+1/46 by Fouvry, Kowalski, and Michel [ FKM ] (in the case of prime moduli). A different line of development was the use of averaging over arithmetic progressions to extend the range of distribution. In 2005, Banks, Heath-Brown, and Shparlinski [ BHS ] showed that by averaging over all coprime residue classes modulo q , one can obtain a power-saving error term for divisor sums for the full range of moduli q . Notably, Nguyen [ Ng ] applied the Banks, Heath-Brown, Shparlinski method to the 3-fold divisor function and obtained an averaged level of distribution of 2/3 for d_{3}(n) . In concrete terms, Nguyen showed that when one averages over all coprime residue classes a\ (\bmod\ q) , the error term in the sum of d_{3}(n) up to X in the progression n\equiv a\ (\bmod\ q) exhibits cancellation for all moduli q\ll X^{2/3-\varepsilon} .This result matched the 2/3 distribution level on average, corresponding to the classical Selberg–Hooley bound in an averaged setting. Blomer [ Bl ] introduced a simplified, more analytic approach using the Voronoi summation formula, which streamlined the analysis of divisor sums in progressions. Parry [ Pa1 ] revisited the distribution of d_{3}(n) in arithmetic progressions and gave a proof same as Blomer that d_{3} has an exponent of distribution 2/3 on average. Recently, for k=4 , Parry [ Pa2 ] used the Petrow-Young [ PY ] subconvexity bound for Dirichlet L -functions to show that d_{4}(n) has an exponent of distribution 4/7 when allowing an average over a\ (\bmod q) , for prime q . 1.2. Statement of Main Results In this paper, we push the averaged level of distribution for the ternary divisor function d_{3}(n) to 8/11 by using the Petrow–Young [ PY ] subconvexity bound. This work was motivated by a problem originally suggested to us by Parry. We extend the result of Parry [ Pa1 ] on d_{3}(n) , which has an exponent of distribution 2/3 , to 8/11 for primes by following the method of Parry in [ Pa2 ] and we obtain the following theorem. Theorem 1.1 . Let \displaystyle E_{a/q}(s)\ \displaystyle=\ \sum_{n=1}^{\infty}\frac{d_{3}(n)}{n^{s}}e\left(\frac{na}{q}\right), \displaystyle f_{a/q}(x)\ \displaystyle=\ \underset{s=1}{\operatorname{Res}}\left\{\frac{E_{a/q}(s)\left((2x)^{s}-x^{s}\right)}{s}\right\}, \displaystyle\Delta(a/q)\ \displaystyle=\ \sum_{x\leq n\leq 2x}d_{3}(n)e\left(\frac{na}{q}\right)\ -\ f_{a/q}(x). Then for a prime q\leq x^{8/11} and any \varepsilon>0 , \sum_{a=1}^{q}|\Delta(a/q)|^{2}\ \ll_{\varepsilon}\ x^{3/2+\varepsilon}q^{11/16}. Let \theta_{n}(q)=\frac{\mu(q/(q,n))}{\phi(q/(q,n))}. It is straightforward to establish that \sum_{\begin{subarray}{c}n=1\\ d|n\end{subarray}}^{\infty}\frac{d_{3}(n)\chi(n/d)}{n^{s}} is holomorphic past s=1 for non-principal Dirichlet characters so, after breaking the sum according to the residue r\bmod q , then according to d=(r,q) , and then using Dirichlet characters, we get f_{a/q}(x)=\mathrm{Res}_{s=1}\left\{\frac{(2x)^{s}-x^{s}}{s}\sum_{n=1}^{\infty}\frac{d_{3}(n)\theta_{n}(q/(q,a))}{n^{s}}\right\}=:F_{x}\left(\frac{q}{(q,a)}\right). Then letting \mathcal{M}_{x}(q,a)=\frac{1}{q}\sum_{d|q}c_{d}(a)F_{x}(d)\qquad\quad E_{x}(q,a)=\sum_{\begin{subarray}{c}x<n\leq 2x\\ n\equiv a(q)\end{subarray}}d_{3}(n)-\mathcal{M}_{x}(q,a) we get \sum_{N=1}^{q}\mathcal{M}_{x}(q,N)e\left(\frac{Nr}{q}\right)=F_{x}\left(\frac{q}{(q,r)}\right) so E_{x}(q,a)=\frac{1}{q}\sum_{r=1}^{q}e\left(-\frac{ar}{q}\right)\left(\sum_{x<n\leq 2x}d_{3}(n)e\left(\frac{nr}{q}\right)-\sum_{N=1}^{q}\mathcal{M}_{x}(q,N)e\left(\frac{Nr}{q}\right)\right)=\frac{1}{q}\sum_{b=1}^{q}e\left(-\frac{ar}{q}\right)\Delta(r/q) and therefore (1.1) \sum_{a=1}^{q}|E_{x}(q,a)|^{2}=\frac{1}{q}\sum_{r=1}^{q}|\Delta(r/q)|^{2}. Applying Cauchy-Schwarz to the above equality, which leads to the following theorem, Theorem 1.2 . For prime q\leq x^{8/11} (1.2) \sum_{a=1}^{q}|E_{x}(q,a)|\ \ll_{\varepsilon}\ x^{3/4+\varepsilon}q^{11/32}, which shows equidistribution on average up to q\leq x^{8/11} . In Section 2 we collect the auxiliary results needed for the proof of Theorem 1.1 . A key point is Lemma 6 , where we establish bounds for the third moment of Dirichlet L -functions using the Petrow–Young [ PY ] subconvexity result. These bounds are then applied in Lemmas 8 and 9 to obtain cancellation in sums involving Ramanujan sums, which are crucial in our analysis. Finally, in Section 3 , we combine these ingredients to complete the proof of Theorem 1.1 . 2. Lemmas Lemma 1 . Let E_{h/q}(s) be as in Theorem 1.1 , let w:[0,\infty)\rightarrow\mathbb{R} be smooth and of compact support, and let \displaystyle\tilde{\Delta}(h/q)\ \displaystyle=\ \sum_{n=1}^{\infty}d_{3}(n)e\left(\frac{nh}{q}\right)w(n)-\underset{s=1}{\operatorname{Res}}\left\{E_{h/q}(s)\int_{0}^{\infty}w(t)\,t^{s-1}\,dt\right\}, \displaystyle U(x)\ \displaystyle=\ \frac{1}{2\pi i}\int_{(c)}\left(\frac{\Gamma(s/2)}{\Gamma((1-s)/2)}\right)^{3}\frac{ds}{x^{s}},\quad\text{for }x>0\text{ and }0<c<\frac{1}{6}, \displaystyle\hat{w}_{q}(n)\ \displaystyle=\ \int_{0}^{\infty}w(t)\,U(Nt)\,dt,\quad\text{with}\ \ \ N\ =\ \frac{\pi^{3}n}{q}, \displaystyle R_{a,b,c}(h/q)\ \displaystyle=\ \sum_{1\leq x,y,z\leq q}e\left(\frac{ax+by+cz-hxyz}{q}\right), \displaystyle A_{h/q}(n)\ \displaystyle=\ \frac{1}{2}\sum_{abc=n}R_{a,b,c}(h/q). Then for (h,q)=1 , \tilde{\Delta}(h/q)=\frac{\pi^{3/2}}{q^{3}}\sum_{n=1}^{\infty}A_{h/q}(n)\,\tilde{w}_{q}(n)+\text{3 analogous terms}. Proof. This is Theorem 2 of Ivić’s paper [ Iv ] . ∎ The following Lemma concerns the bound for A_{h/q}(n) from Ivić-Voronoi formula. Lemma 2 . Let A_{h/q}(n) be as in Lemma 1 with prime q and (h,q)=1 . If q\mid n , then A_{h/q}(n)\;\ll_{\varepsilon}\;n^{\varepsilon}\,(q^{2},n). for any \varepsilon>0 . Moreover, if q\nmid m , then \sum_{h=1}^{q-1}A_{h/q}(n)\,\overline{A_{h/q}(m)}\;\ll_{\varepsilon}\;(mn)^{\varepsilon}q(q^{2},n), for any \varepsilon>0 . Proof. By definition, A_{h/q}(n)=\frac{1}{2}\sum_{abc=n}R_{a,b,c}(h/q)=\frac{1}{2}\sum_{abc=n}\sum_{x,y,z=1}^{q}e\left(\frac{ax+by+cz-hxyz}{q}\right). Since q\mid n , at least one of a,b,c must be divisible by q . First, let’s assume that q\mid c . We have R_{a,b,c}(h/q)=\sum_{x,y=1}^{q}e\left(\frac{ax+by}{q}\right)\sum_{z=1}^{q}e\left(\frac{(c-hxy)z}{q}\right). If c\not\equiv hxy\pmod{q} , then it is well-known that the inner sum is 0 . So, the only case that matters is xy\equiv c\bar{h}\equiv 0\pmod{q} . Therefore, \displaystyle R_{a,b,c}(h/q)=q\sum_{\underset{q\mid xy}{x,y=1}}^{q}e\left(\frac{ax+by}{q}\right) \displaystyle=q\left(1+\sum_{x=1}^{q-1}e\left(\frac{ax}{q}\right)+\sum_{y=1}^{q-1}e\left(\frac{by}{q}\right)\right) \displaystyle=q(1+c_{q}(a)+c_{q}(b)) \displaystyle=q(2+c_{q}(a)+c_{q}(b)+c_{q}(c)-q). By the symmetry, the same formula holds whenever q\mid a or q\mid b , hence for all triples with q\mid abc . We have R_{a,b,c}(h/q)=q(2+c_{q}(a)+c_{q}(b)+c_{q}(c)-q)=k_{a,b,c}q^{2}-(q^{2}+q) where q^{k_{a,b,c}}=(q,a)(q,b)(q,c) , or equivalently, k_{a,b,c} equals the number of elements among a,b,c divisible by q . Therefore, \displaystyle A_{h/q}(n) \displaystyle=\frac{1}{2}\sum_{abc=n}R_{a,b,c}(h/q) \displaystyle=\frac{q^{2}}{2}\sum_{abc=n}k_{a,b,c}-\frac{q^{2}+q}{2}\sum_{abc=n}1 \displaystyle=\frac{q^{2}}{2}\log_{q}\left[\prod_{abc=n}(q,a)(q,b)(q,c)\right]-\frac{q^{2}+q}{2}d_{3}(n). Let v_{q}(n)=\alpha\geq 1 and n=q^{\alpha}\ell . Then, we can change the variables as (a,b,c)=(q^{x}a^{\prime},q^{y}b^{\prime},q^{z}c^{\prime}) where (q,a^{\prime})=(q,b^{\prime})=(q,c^{\prime})=1 . \displaystyle\prod_{abc=n}(q,a)(q,b)(q,c) \displaystyle=\prod_{\underset{a^{\prime}b^{\prime}c^{\prime}=\ell}{x+y+z=\alpha}}(q,q^{x}a^{\prime})(q,q^{y}b^{\prime})(q,q^{z}c^{\prime}) \displaystyle=\prod_{\underset{a^{\prime}b^{\prime}c^{\prime}=\ell}{x+y+z=\alpha}}(q,q^{x})(q,q^{y})(q,q^{z}) \displaystyle=\left[\prod_{x+y+z=\alpha}(q,q^{x})(q,q^{y})(q,q^{z})\right]^{d_{3}(\ell)} \displaystyle=q^{\frac{3\alpha(\alpha+1)d_{3}(\ell)}{2}}. Therefore, \displaystyle A_{h/q}(n) \displaystyle=\frac{q^{2}}{2}\log_{q}\left[\prod_{abc=n}(q,a)(q,b)(q,c)\right]-\frac{q^{2}+q}{2}d_{3}(n) \displaystyle=\frac{3q^{2}\alpha(\alpha+1)d_{3}(\ell)}{4}-\frac{(q^{2}+q)d_{3}(n)}{2}. Since d_{3} is multiplicative, d_{3}(n)=d_{3}(\ell)d_{3}(q^{\alpha})=\frac{(\alpha+2)(\alpha+1)}{2}d_{3}(\ell) . \displaystyle A_{h/q}(n) \displaystyle=\frac{3q^{2}\alpha d_{3}(n)}{2(\alpha+2)}-\frac{(q^{2}+q)d_{3}(n)}{2} \displaystyle=d_{3}(n)\left[q^{2}\left(1-\frac{3}{\alpha+2}\right)-\frac{q}{2}\right] \displaystyle=d_{3}(n)\cdot\begin{cases}-\frac{q}{2}&\text{if }\alpha=1,\\ C_{\alpha}q^{2}-\frac{q}{2}&\text{if }\alpha\geq 2,\end{cases}\quad\text{where }C_{\alpha}:=1-\frac{3}{\alpha+2}\neq 0\text{ and bounded}. Notice that A_{h/q}(n) does not depend on h .