On Linnik's almost Goldbach theorem (Q1974194): Difference between revisions

The author refines the theorem, known as Linnik's ``almost Goldbach'' theorem, that every large even integer may be written as a sum of two primes and a bounded number of powers of \(2\). Write \(r_k''(N)\) for the number of representations of \(N\) in the form \[ N=p_1+p_2+2^{\nu_1}+\cdots+2^{\nu_k}, \] with prime numbers \(p_1\), \(p_2\) and natural numbers \(\nu_1,\ldots,\nu_k\). Then it is proved that for any integer \(k\geq 2250\), there exists an \(N_k\) depending only on \(k\) such that whenever \(N\) is an even number \(\geq N_k\), one has \(r_k''(N)\gg N(\log N)^{k-2}\). This theorem considerably improves the recent result of \textit{J. Y. Liu, M. C. Liu} and the author [J. Theor. Nombres Bordx. 11, 133-147 (1999; Zbl 0979.11051)] that assured the above conclusion for \(k\geq 54000\). On assuming the generalised Riemann hypothesis for Dirichlet \(L\)-functions, in the meantime, the author also establishes the above conclusion for \(k\geq 160\). Further, let \(\varepsilon_0\) be a fixed sufficiently small positive number, and write \(r_k'(n)\) for the number of representations of \(n\) in the form \[ n=p+2^{\nu_1}+\cdots+2^{\nu_k}, \] with natural numbers \(\nu_1,\ldots,\nu_k\) and primes \(p\) satisfying \(\varepsilon_0 N\leq p\leq N\). It is also shown that for any integer \(k\geq 1125\), there exists an \(N_{k,\varepsilon_0}\) depending only on \(k\) and \(\varepsilon_0\) such that for any \(N\geq N_{k,\varepsilon_0}\), one has \[ \sum_{n\leq N, 2\nmid n} (r'_k(n)-2(\log_2 N)^k(\log N)^{-1})^2 \leq 2\times 0.998N(\log_2 N)^{2k}(\log N)^{-2}. \] This theorem is an explicit version of the corresponding result due to Gallagher, and is utilized to establish the above unconditional result on \(r_k''(N)\). Roughly speaking, the proof philosophically follows the strategy of previous work of Linnik, and of Gallagher. The crucial points of this paper are explicit zero-free regions and zero density estimates for Dirichlet \(L\)-functions near the line \(\sigma=1\). Some of the important results in these directions rely upon the forthcoming paper of the author [``On zeros of Dirichlet's \(L\)-functions'', to appear in Acta Math. Sin.]. The argument also requires lots of numerical computation, and the author remarks that such burdens were carried out by the software package Mathematica.