On a conjecture of Montgomery and Soundararajan (Q2235226)

Let \(\psi(x)\) be the partial sums up to \(x\) of the classical Von Mangoldt function \(\Lambda(n)\). The Prime Number Theorem is equivalent to the fact that \(\psi(x)=(1+o(1))x\), and the Riemann hypothesis is equivalent to the squareroot bound (up to terms that are \(O(x^\varepsilon)\) for any \(\varepsilon>0\)) of the error term \(\psi(x)-x\). A conjecture of Montgomery and Soundararajan quantifies this squareroot bound in short intervals and can be stated as follows. For fixed \(\varepsilon>0\), for each positive integer \(n\) and uniformly for \(\frac{(\log X)^{1+\varepsilon}}{X}\leq \delta \leq \frac{1}{X^\varepsilon}\), the \(n\)-th moment \[\frac{1}{X}\int_1^X\left(\frac{\psi(x+\delta X)-\psi(x)-\delta X}{\sqrt{X}} \right)^ndx=(\mu_n+o(1))(\delta\log(1/\delta))^{n/2},\] where \(\mu_n\) is the \(n\)-th moment of a standard Gaussian. In the paper under review, the authors study a weighted version of \(\frac{\psi(x+\delta X)-\psi(x)-\delta X}{\sqrt{X}} \). For a certain class of test functions \(\eta:\mathbb{R}\to\mathbb{R}\) such that the function itself and its derivative have some exponential decay, the Fourier transform at \(0\) is positive, they define \[\psi_\eta(x,\delta):=\sum_{n=1}^\infty \frac{\Lambda(n)}{\sqrt{n}}\eta\left(\frac{1}{\delta}\log(n/x) \right),\] where now \(0<\delta<1\). As the authors says, ``morally, this function counts prime powers in the interval \([x(1-O(\delta)),x(1+O(\delta))]\), and the weight \(n^{-1/2}\) is equal to \(x^{-1/2}(1+O(\delta))\)''. The expected main term for \(\psi_\eta(x,\delta)\) is given by \(x^{1/2}\delta L_\eta(\delta/2)\), and for small enough \(\delta\), \(L_\eta(\delta/2)=\hat{\eta}(0)+O(\delta)\). The main Theorem of the paper under review assumes the Riemann hypothesis and provides a lower bound for \[M_n(X,\delta;\eta,\Phi):=\frac{1}{(\log X)\int_{0}^\infty\Phi}\int_1^\infty\Phi\left(\frac{\log x}{\log X}\right )(\psi_\eta(x,\delta)-x^{1/2}\delta L_\eta(\delta/2))^n\frac{dx}{x},\] where \(\Phi\) is a certain test function. This lower bound for \(M_n(X,\delta;\eta,\Phi)\) is in agreement with the conjecture by Montgomery and Soundararajan when \(n\) is a positive even integer. They also obtain unconditional results, one of them is an unconditional \(\Omega\) bound for the prime counting function in short intervals \(\psi(x+\delta x)-\psi(x)-\delta x\). The key idea of the proof circumvent the need to understand spacing statistics and Diophantine properties (for higher moments) of zeros of the zeta function. Also, the authors claim that the novelty of the proof is to avoid the need for Parseval's identity. There are other interesting previous works related to this paper, and the paper under review seems to present a complete list of the literature.