Some refinements for error terms estimates for certain additive problems with primes (Q1970623)

Let \(\Lambda(n)\) denote the von Mangoldt function. For natural numbers \(k\) and \(n\), define \(R_k(n)= \sum\Lambda(m_1)\ldots\Lambda(m_k)\) where the summation is taken over the natural numbers \(m_j\) satisfying \(m_1+\cdots+m_k=n\). Also write \[ {\mathfrak S}_k(n)=\prod_{p|n} (1-(1-p)^{1-k}) \prod_{p\nmid n}(1-(1-p)^{-k}), \] and put \(E_k(n)=R_k(n)-{\mathfrak S}_k(n) n^{k-1}/(k-1)!\). Then, for \(k=3\) and \(4\), the estimates \[ E_k(n)\ll n^{(3k-2)/4} (\log n)^{k-1} \quad\text{and}\quad \sum_{n\leq N}E_k(n)^2\ll N^{2k-2} (\log N)^{14-3k} \] are established in this paper under the generalized Riemann hypothesis (for Dirichlet \(L\)-functions). These results refine the previous ones due to Friedlander and Goldston by some logarithmic factors. The key ingredient of this paper is the use of the weighted exponential sum \(\sum_{m=1}^{\infty}\Lambda(m) e^{-m/N}e(m\alpha)\), in contrast with the truncated sum \(\sum_{m\leq N}\Lambda(m)e(m\alpha)\) employed in the work of Friedlander and Goldston, where we write \(e(\alpha)=e^{2\pi i\alpha}\). This idea may be applied to related problems as well, and the author mentions a couple of conditional results obtained by this method, in addition. One of them concerns prime twins, and another is about a problem on the values taken by \(\lambda_1p_1+\lambda_2p_2\) with primes \(p_j\), where real numbers \(\lambda_j\) are given.