Level of distribution of quadratic polynomials and an upper bound sieve for friable integers (Q2178260): Difference between revisions

This paper is largely devoted to the investigation of the multiplicative structure of the polynomial values \(n^2-D\), for fixed non-square \(D\). The first theorem gives a level of distribution estimate. Let \(V\) be a suitable compactly supported weight function, and let \(\lambda(q)\) be well-factorable weights (as in the Rosser-Iwaniec linear sieve, for example). One then has \[\sum_{q\le Q}\left(\sum_{q\mid n^2-D}V(n/x)-x\widehat{V}(0)\nu(q)/q\right) \ll_{\eta,V,D}x^{1-\eta/3},\] for all \(Q\) up to \(x^{1+25/178-\eta}\). (Here \(\nu(q)\) is the number of \(n\) modulo \(q\) for which \(q\mid n^2-D\).) The range \(Q\le x^{1+25/178-\eta}\) improves a result of \textit{H. Iwaniec} [Invent. Math. 47, 171--188 (1978; Zbl 0389.10031)], in which \(25/178\) is replaced by \(1/15\). The proof uses large sieve estimates for Fourier coefficients of cusps forms, following \textit{Á. Tóth} [Int. Math. Res. Not. 2000, No. 14, 719--739 (2000; Zbl 1134.11339)] and \textit{S. Drappeau} [Proc. Lond. Math. Soc. (3) 114, No. 4, 684--732 (2017; Zbl 1392.11059)]. Related ideas produce the second theorem, which shows in particular that if \(x\) is large enough in terms of \(D\), there will be an integer \(n\in(x,2x]\) for which \(P^+(n^2-D)\ge x^{1.2182}\). (Here \(P^+(m)\) is the largest prime factor of \(m\).) This gives a slight improvement of the result given by \textit{J. M. Deshouillers} and \textit{H. Iwaniec} [Ann. Inst. Fourier 32, No. 4, 1--11 (1982; Zbl 0489.10038)] in which the exponent is 1.2024. The paper goes on to consider smooth/friable values of polynomials. For any square-free integral polynomial \(F\) write \(\Psi_F(x,y)\) for the number of integers \(n\le x\) with \(P^+(F(n))\le y\). Very little is known about \(\Psi_F(x,y)\) when \(F\) has degree \(d> 1\), but one would conjecture that \[\Psi_F(x,y)=x\rho(u)^{d+o(1)}, \;\;\; u=\frac{\log x}{\log y}\] when \(u\to\infty\), and \(y\) is not too small in terms of \(x\). Here it is shown that \(\Psi_F(x,y)\ll x\rho(u)^{c+o(1)}\) for \(y\le x\) as soon as \((\log y)/(\log\log x)\to\infty\), with various exponents \(c=c(F)\). One may take \(c=1\) in all cases, recovering a result of \textit{N. Hmyrova} [Sov. Math., Dokl. 5, 576--579 (1964; Zbl 0127.27005); translation from Dokl. Akad. Nauk SSSR 155, 1268--1271 (1964)]. When \(F\) has an irreducible quadratic factor one can take \(c=1+25/178\). When \(F\) has two or more linear factors \(c=8/5\) is admissible, and when there is just one linear factor one may take \(c=3/2\).