Partitions into Primes (Q4955699)

Let \(\Lambda(m)\) denote the von Mangoldt function (that is, \(\Lambda(m)=\log p\) when \(m\) is a power of a prime \(p\), and \(\Lambda(m)=0\) otherwise), and define the partition function \(p_{\Lambda}(n)\) by the relation \(\sum_{n=0}^{\infty}p_{\Lambda}(n)x^n= \prod_{m=1}^{\infty}(1-x^m)^{-\Lambda(m)}\). Therefore \(p_{\Lambda}(n)\) represents a weighted count of the number of partitions of \(n\) into prime powers. In 1975, Richmond proved that \(\log p_{\Lambda}(n)\sim 2\sqrt{\zeta(2)n}\) as \(n\rightarrow\infty\), where \(\zeta(s)\) denotes the Riemann zeta function, or \(2\sqrt{\zeta(2)}=\pi\sqrt{2/3}\). Hereafter we write \(E(n)=\log p_{\Lambda}(n)-2\sqrt{\zeta(2)n}\) for short. In the paper under review, the author first establishes that for some positive constant \(c\) and for all large \(n\), one has \[ E(n)\ll \sqrt{n}\exp\bigl(-c(\log n)(\log\log n)^{-2/3} (\log\log\log n)^{-1/3}\bigr), \] improving Richmond's estimate for \(E(n)\). Secondly it is proved that \(E(n)=\Omega_{\pm}(n^{1/4})\) as \(n\rightarrow\infty\). The proof of the latter result gives a bit stronger conclusion in case the Riemann hypothesis is false. On writing \(\theta\) for the least upper bound for the real parts of the zeros of the Riemann zeta function, it is indeed proved that for any \(\varepsilon>0\), \(E(n)=\Omega_{\pm}(n^{\theta/2-\varepsilon})\) as \(n\rightarrow\infty\). Richmond had conversely shown that \(E(n)\ll n^{\theta/2}\), and it is thus revealed that the size of \(E(n)\) is directly concerned with \(\theta\). In particular, \(n^{1/4}\) is the true order of magnitude of \(E(n)\) if the Riemann hypothesis is true, and vice versa. Moreover, it is pointed out that the Riemann hypothesis is equivalent to the slightly weaker statement that \(E(n)\ll_{\varepsilon}n^{1/4+\varepsilon}\) for all \(\varepsilon>0\). This paper also includes a new proof for the bound \(E(n)\ll n^{\theta/2}\), which is simpler than the original one due to Richmond. To mention a rough outline of the argument, put \(P_{\Lambda}(u)=\sum_{n\leq u}p_{\Lambda}(n)\), and let \(F_{\Lambda}(x)\) be the Laplace-Stieltjes transform of \(P_{\Lambda}(u)\), that is, \(F_{\Lambda}(x)=\int_0^1 e^{-xu} dP_{\Lambda}(u)\). The key tools are some general Abelian and Tauberian theorems established in the first stage of the paper. They are applied in order to present a relation between the behaviour of \(P_{\Lambda}(u)\) and that of \(F_{\Lambda}(x)\). Then, on the one hand, \(\log p_{\Lambda}(n)\) may be simply connected with \(\log P_{\Lambda}(u)\) by the observation that \(p_{\Lambda}(n+k)\geq p_{\Lambda}(n)\) for any \(k\geq 15\). On the other hand, analytic properties of \(F_{\Lambda}(x)\) may be understood via those of the corresponding Dirichlet series \(-\zeta'(s)/\zeta(s)\). The behaviour of \(E(n)\) is thus related to the zeros of \(\zeta(s)\).