Disproving Hooley's conjecture (Q6064885)

The central quantity considered in this paper is the variance \[ G(x;q):=\sum_{\substack{a\bmod q\\ (a,q)=1}} \Biggl\lvert\sum_{\substack{p\leq x\\ p\equiv a\bmod q}} \log p-\frac x{\phi(q)}\Biggr\rvert^2. \] \textit{C. Hooley} [Proc. Lond. Math. Soc. (3) 33, 535--548 (1976; Zbl 0338.10043)] proved, assuming GRH, that \[ \frac1{\log T} \int_2^T G(t;q)\frac{\mathrm dt}{t^2}\ll\log q \] uniformly for \(q\leq T\); he writes (in different notation) ``[\(\ldots\)] a result that is at least consistent with the truth of the natural conjecture \(G(x;q)\ll x\log q\)''. In the present paper, the authors refute Hooley's conjecture. The first part of their main theorem (the second being an analogous statement for the related quantity \(V_\Lambda(x_j;q_j)\)) is as follows. Theorem. Fix any sufficiently large positive real number \(M\). There exists an infinite sequence of pairs \((q_j,x_j)\), with \(q_j\asymp \tfrac1M\log\log x_j\) (with an absolute implied constant) both tending to infinity, for which \[ G(x_j;q_j)\ge Mx_j\log q_j. \] The proof relies in an essential way on the fact that Hooley's conjecture (in certain ranges) implies GRH, which greatly helps in obtaining a contradiction. Refining arguments previously employed by Littlewood, see Theorem~15.11 in [\textit{H. L Montgomery} and \textit{R. C. Vaughan}, Multiplicative number theory. I. Classical theory. Reprint of the 2007 hardback edition. Cambridge: Cambridge University Press (2012; Zbl 1245.11002)], and distinguishing between the case that GRH holds true, and the other case, the authors are able to deduce their main theorems.