betaFeedback

WorksMathematics

Sums of Integers with Small Prime Factors

Abstract

Let P+(n)P^+(n) denote the largest prime factor of nn, with P+(1)=1P^+(1)=1. We prove that, for every β>103/(256e)β>103/(256\sqrt e), there is a constant C=C(β)>0C=C(β)>0 such that every sufficiently large integer NN can be written as N=a+bN=a+b, where a,b∈Z>0a,b\in\mathbb{Z}_{>0} and P+(a)≤Nβ,P+(b)≤(log⁡N)C. P^+(a)\leq N^β,\qquad P^+(b)\leq(\log N)^C. This strengthens Balog's result, which gives two NβN^β-smooth summands for β>4/(9e)β>4/(9\sqrt e). The proof combines Balog's argument with convolution estimates that are uniform in the residue class.

Introduction

Goldbach’s conjecture states that every even integer greater than 2 is a sum of two primes. A similar question is whether every sufficiently large integer can be written as a sum of two integers with small prime factors. Let P+(n)P^{+}(n) denote the largest prime factor of a positive integer nn, with P+(1)=1P^{+}(1)=1. We call a positive integer nn yy-smooth if P+(n)≤yP^{+}(n)\le y.

Balog [3] proved that, for every β>4/(9e)\beta>4/(9\sqrt{e}), every sufficiently large integer NN is a sum of two NβN^{\beta}-smooth integers. If Ψ(x;y)\Psi(x;y) counts the yy-smooth integers up to xx, then Ψ(x;xβ)≫βx\Psi(x;x^{\beta})\gg_{\beta}x for every fixed β>0\beta>0.

There are far fewer integers whose prime factors are at most a fixed power of log⁡N\log N. More precisely, for every fixed C>1C>1,

Ψ(x;(log⁡x)C)=x1−1/C+o(1)=o(x)(x→∞).\Psi(x;(\log x)^{C})=x^{1-1/C+o(1)}=o(x)\qquad(x\to\infty).

Our main result improves Balog’s exponent from 4/(9e)4/(9\sqrt{e}) to 103/(256e)103/(256\sqrt{e}) and ensures that all prime factors of the second summand are at most (log⁡N)K(\log N)^{K}, for some fixed K>0K>0.

Theorem 1. For any β>103/(256e)\beta>103/(256\sqrt{e}), there exist constants N0=N0(β)N_{0}=N_{0}(\beta) and K=K(β)>0K=K(\beta)>0 such that, for every integer N>N0N>N_{0}, there exist a,b∈Z>0a,b\in\mathbb{Z}_{>0} with N=a+bN=a+b and

P+(a)≤NβandP+(b)≤(log⁡N)K.P^{+}(a)\le N^{\beta}\qquad\text{and}\qquad P^{+}(b)\le(\log N)^{K}.

AI Usage

AI Usage. ChatGPT/Codex was used to assist with mathematical auditing, filling in details and revision, and improvements to exposition and \LaTeX formatting. Initial drafts of the proofs of Lemma 5 and Lemmas 6 and 7 were generated using ChatGPT and Gemini Pro, respectively. The author reviewed and edited the resulting material and takes full responsibility for all arguments and results.

Conversations with LLMs seem to suggest that the level of distribution in Theorem 2 can be improved further, thereby also improving the exponent 103/(256e)103/(256\sqrt{e}) in Theorem 1. An initial draft is available on Hexagon.

Acknowledgements

The author would like to thank his adviser, Kaisa Matomäki, for introducing the author to the paper of Balog [3], for many insightful suggestions and encouragement throughout this project, and for carefully reading the manuscript. The author also thanks Alexandru Pascadi for valuable email exchanges concerning [12] and for providing Lemma 99. The author is grateful to Sary Drappeau and Jared Lichtman for helpful clarifications regarding [5] and [10], respectively. The author was supported by the Research Council of Finland (grant numbers 346307, 364214, and 370133) and by the University of Turku Graduate School’s Exactus fellowship while working on this project.

Notation

The letter pp, with or without subscripts, is reserved for primes; sums and products indexed by pp are therefore over primes. All logarithms are natural. We write Z>0\mathbb{Z}_{>0} for the positive integers and (a,b)(a,b) for the greatest common divisor of aa and bb. Divisors in sums and products are positive unless otherwise specified. For a summation variable, n∼Nn\sim N means N<n≤2NN<n\le2N. We write 1E\mathbf{1}_E for the indicator of a condition EE, and ∣A∣|A| for the cardinality of a finite set AA.

We use Landau’s notation F=O(G)F=O(G) and Vinogradov’s notation F≪GF\ll G or G≫FG\gg F to mean ∣F∣≤CG|F|\le CG for some constant C>0C>0, with G≥0G\ge0. Subscripts indicate permitted dependence of the implied constant, as in OεO_\varepsilon or ≪ε\ll_\varepsilon, and constants may change from line to line. For positive quantities, F≍GF\asymp G means F≪G≪FF\ll G\ll F. In a specified limit, F=o(G)F=o(G) means F/G→0F/G\to0, and F∼GF\sim G means F/G→1F/G\to1. In particular, F(x)≪xo(1)F(x)\ll x^{o(1)} means F(x)≪εxεF(x)\ll_\varepsilon x^\varepsilon for every ε>0\varepsilon>0. The bound A≥xΩ(1)A\ge x^{\Omega(1)} means A≥xcA\ge x^c for some fixed constant c>0c>0.

For an integer nn with ∣n∣≥2|n|\ge2, let P+(n)P^+(n) and P−(n)P^-(n) denote the largest and smallest prime divisors of ∣n∣|n|, respectively, with P+(±1)=1P^+(\pm1)=1 and P−(±1)=∞P^-(\pm1)=\infty. A positive integer nn is yy-smooth if P+(n)≤yP^+(n)\le y. For real x≥0x\ge0, y≥2y\ge2, q∈Z>0q\in\mathbb{Z}_{>0}, and a∈Za\in\mathbb{Z}, put

Ψ(x;y):=∣{n∈Z>0:n≤x, P+(n)≤y}∣,\Psi(x;y):=\left|\{n\in\mathbb{Z}_{>0}:n\le x,\ P^+(n)\le y\}\right|,
Ψq(x;y):=∣{n∈Z>0:n≤x, P+(n)≤y, (n,q)=1}∣,\Psi_q(x;y):=\left|\{n\in\mathbb{Z}_{>0}:n\le x,\ P^+(n)\le y,\ (n,q)=1\}\right|,
Ψ(x,y;a,q):=∣{n∈Z>0:n≤x, P+(n)≤y, n≡a(modq)}∣.\Psi(x,y;a,q):=\left|\{n\in\mathbb{Z}_{>0}:n\le x,\ P^+(n)\le y,\ n\equiv a\pmod q\}\right|.

Thus Ψq\Psi_q counts smooth integers coprime to qq, while Ψ(x,y;a,q)\Psi(x,y;a,q) counts those in the residue class a(modq)a\pmod q. We use φ(n)\varphi(n), μ(n)\mu(n), and ω(n)\omega(n) for Euler’s totient function, the Möbius function, and the number of distinct prime divisors of nn, respectively; π(x)\pi(x) counts the primes at most xx. For an integer k≥1k\ge1, the divisor function τk(n)\tau_k(n) counts ordered factorisations n=n1⋯nkn=n_1\cdots n_k into positive integers, and τ=τ2\tau=\tau_2. For arithmetic functions f,gf,g, their Dirichlet convolution is (f∗g)(n):=∑d∣nf(d)g(n/d)(f*g)(n):=\sum_{d\mid n}f(d)g(n/d).

For an integer aa coprime to the modulus, a‾\overline{a} denotes its multiplicative inverse modulo that modulus, which is understood from the expression; overbars on complex coefficients denote complex conjugation.

Notation used only in § 5 is introduced there as needed.

Outline of the Method

We outline the proof of Theorem 1. Take y2=(log⁡N)Ky_2=(\log N)^K, where KK is a sufficiently large fixed constant, and let y1≥y2y_1\ge y_2 be chosen below. We seek a representation N=a+bN=a+b with P+(a)≤y1P^+(a)\le y_1 and P+(b)≤y2P^+(b)\le y_2. The reduction at the start of the proof of Theorem 1 allows us to assume that P−(N)>y2P^{-}(N)>y_{2}. Every tuple in the following set gives the required representation, so it suffices to prove that the set is nonempty:

N′:={(s,ℓ,m,n):N=s+ℓmn, 1≤s<N, P+(s)≤y2, P+(ℓmn)≤y1, ℓ∼L, m,n∼M}.\mathcal{N}' := \{(s,\ell,m,n): N=s+\ell mn,\ 1\le s<N,\ P^{+}(s)\le y_{2},\ P^{+}(\ell mn)\le y_{1},\ \ell\sim L,\ m,n\sim M\}.

where L,M≥NΩ(1)L,M\ge N^{\Omega(1)} satisfy LM2=N/ψ(N)LM^{2}=N/\psi(N), with ψ(N)=log⁡log⁡N\psi(N)=\log\log N. We further assume that 2L<y1≤M2L<y_{1}\le M. Let

N:={(s,ℓ,m,n):N=s+ℓmn, 1≤s<N, P+(s)≤y2, ℓ∼L, m,n∼M}.\mathcal{N}:=\{(s,\ell,m,n):N=s+\ell mn,\ 1\le s<N,\ P^{+}(s)\le y_{2},\ \ell\sim L,\ m,n\sim M\}.

Let

N1:={(s,ℓ,pm,n)∈N:p prime, p>y1},N2:={(s,ℓ,m,pn)∈N:p prime, p>y1}.\mathcal{N}_{1}:=\{(s,\ell,pm,n)\in\mathcal{N}:p\text{ prime},\ p>y_{1}\}, \qquad \mathcal{N}_{2}:=\{(s,\ell,m,pn)\in\mathcal{N}:p\text{ prime},\ p>y_{1}\}.

References

  1. [1]E. Assing, V. Blomer, and J. Li. Uniform Titchmarsh divisor problems. Adv. Math. 393 (2021), 108076. doi:10.1016/j.aim.2021.108076.DOI
  2. [2]R. C. Baker and G. Harman. Shifted primes without large prime factors. Acta Arith. 83 (1998), 331–361. doi:10.4064/aa-83-4-331-361.DOI
  3. [3]A. Balog. An additive representation of integers. Acta Math. Hungar. 54 (1989), 297–301.DOI
  4. [4]R. de la Bretèche and G. Tenenbaum. Propriétés statistiques des entiers friables. Ramanujan J. 9 (2005), 139–202.DOI
  5. [5]S. Drappeau. Théorèmes de type Fouvry-Iwaniec pour les entiers friables. Compos. Math. 151 (2015), no. 5, 828–862.DOI
  6. [6]É. Fouvry. Autour du théorème de Bombieri-Vinogradov. II. Ann. Sci. École Norm. Sup. 20 (1987), no. 4, 617–640.DOI
  7. [7]A. J. Harper. Bombieri–Vinogradov and Barban–Davenport–Halberstam type theorems for smooth numbers. Preprint (2012), arXiv:1208.5992v1 [math.NT]. https://arxiv.org/abs/1208.5992v1.
  8. [8]A. Hildebrand and G. Tenenbaum. On integers free of large prime factors. Trans. Amer. Math. Soc. 296 (1986), no. 1, 265–290.DOI
  9. [10]J. D. Lichtman. Primes in arithmetic progressions to large moduli, and Goldbach beyond the square-root barrier. Preprint (2023), arXiv:2309.08522v1 [math.NT]. https://arxiv.org/abs/2309.08522v1.
  10. [11]J. D. Lichtman. Primes in arithmetic progressions to large moduli and shifted primes without large prime factors. Preprint (2022), arXiv:2211.09641v1 [math.NT]. https://arxiv.org/abs/2211.09641v1.
  11. [12]A. Pascadi. Large sieve inequalities for exceptional Maass forms and the greatest prime factor of n² + 1. Forum Math. Pi 14 (2026), e8. doi:10.1017/fmp.2026.10025.DOI
  12. [13]A. Pascadi. On the exponents of distribution of primes and smooth numbers. Preprint (2025), arXiv:2505.00653v2 [math.NT]. https://arxiv.org/abs/2505.00653v2.

Paper details

vlnc:W_tpetfb2y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

ChatGPT/Codex, ChatGPT, Gemini Pro, LLMs

View statement
Looking for related papers…
Contents