Sifting limits for the \(\varLambda ^{2}\varLambda ^{ - }\) sieve (Q555298)

The Selberg \(\Lambda^2\Lambda_{-}\) sieve is a lower bound sieve method based on the inequality \[ S(\mathcal{A},\mathcal{P},z)\geq\sum_{n\in\mathcal{A}} \left(\sum_{d\mid(n,P(z))}\lambda_d\right)^2\left(1- \sum_{p\mid n,\,p<z}1\right). \] Here \[ P(z)=\prod_{p\in\mathcal{P},\,p<z}p \] as usual. By making an appropriate choice for the coefficients, \(\lambda_d\), \textit{A. Selberg} [``Lectures on sieves'', Collected papers. Volume II. Berlin etc.: Springer-Verlag (1991; Zbl 0729.11001), pp 65--247] showed that the sieving limit satisfies \(\beta_{\kappa}\leq 2\kappa+19/36+o(1)\) as the dimension \(\kappa\) tends to infinity. The purpose of the present paper is to investigate the case in which the dimension \(\kappa\) is an integer in the range \(2\leq\kappa\leq 10\). It is shown that one has \[ \beta_2\leq 4.516,\;\;\beta_3\leq 6.520,\;\;\beta_4\leq 8.522,\;\; \beta_5\leq 10.523,\;\;\beta_6\leq 12.524, \] \[ \beta_7\leq 14.524,\;\;\beta_8\leq 16.524,\;\;\beta_9\leq 18.525, \;\;\beta_{10}\leq 20.525. \] These results suggest that one might have \(\beta_{\kappa}\leq 2\kappa+19/36\) for all \(\kappa\). They may be compared with bounds for the Buchstab-iterated \(\Lambda^2\) sieve, given by \textit{H. G. Diamond, H. Halberstam} and \textit{W. F. Galway} [A higher-dimensional sieve method. With procedures for computing sieve functions. Cambridge Tracts in Mathematics 177. Cambridge: Cambridge University Press (2008; Zbl 1207.11099)] which yield \(\beta_2\leq 4.266\), but are inferior for the other cases given above. It should also be noted that \textit{S. Blight} [Refinements of Selberg's sieve, PhD thesis, Rutgers, 2010] has used a variant of the \(\Lambda^2\Lambda_-\) sieve in which the \(\Lambda_{-}\) factor includes terms with up to 3 prime factors, yielding \(\beta_2\leq 4.450\), \(\beta_3\leq 6.458\) and \(\beta_4\leq 8.470\). The values of \(\lambda_d\) used are given in terms of a linear polynomial \(x+a\), as in Selberg's original work. The author explains that his ``investigations of the use of higher degree polynomials in this problem has not met with much success'' (\textit{sic.}) The constant term \(a\) is chosen optimally, depending on the dimension. The necessary computations are far from easy, and use techniques from the work of \textit{F. Grupp} and \textit{H.-E. Richert} [J. Number Theory 22, 208--239 (1986; Zbl 0578.10048)].