Weak subconvexity without a Ramanujan hypothesis (Q2001643)

Loosely speaking one expects a tight convexity bound of order \(C^{1/4}\) for \(L\)-functions of analytic conductor \(C\). In the present paper one aims for a weak sub-convexity bound of order \(C^{1/4}(\log C)^{-\delta}\) for all \(L\)-functions in a suitable class, with \(\delta>0\) being a constant depending on the class. In [Ann. Math. (2) 172, No. 2, 1469--1498 (2010; Zbl 1234.11066)], the first author proved such a result (with \(\delta\) almost 1) for a wide class of \(L\)-functions satisfying a suitable Ramanujan conjecture. The aim of the present paper is to remove this last assumption. The paper axiomatizes a class \(\mathcal{S}(m)\) of \(L\)-functions of ``weight'' \(m\), somewhat in the spirit of the Selberg class. The Ramanujan conjecture would give pointwise bounds for the coefficients \(b(n)\) in the series \(L'(s)/L(s)=\sum b(n)n^{-s}\). Instead it is assumed only that \[ \sum_{n=1}^{\infty}|b(n)|n^{-1-\eta}\le m\eta^{-1}+m\log C+O(m^2) \] uniformly for \(\eta>0\) and for all \(L\)-functions under consideration, and that \[ \sum_{x<n\le x+\delta x}|b(n)|\ll_m \delta x \text{ for } x\gg_m (\delta^{-1}C)^{144m^3} \] uniformly for \(\delta\in(0,1]\). (Here \(C\), which features in the axiom describing the functional equation for \(L(s)\), is the conductor.) Under these assumptions it is shown that \[ L(\tfrac12)\ll_m |L(\tfrac32)|^2\frac{C^{1/4}}{(\log C)^\delta} \] with \(\delta=10^{-17}m^{-3}\). The paper goes on to deduce that \[ L(\tfrac12,\pi)\ll_m \frac{C(\pi)^{1/4}}{(\log C(\pi))^\delta} \] for any cuspidal automorphic representation \(\pi\in \mathrm{GL}_m\) over \(\mathbb{Q}\) with unitary central character. A key tool in the proof is a log-free zero-density estimate \[ N_L(\sigma,T)\ll_m (CT)^{10^7m^3(1-\sigma)} \] valid for all \(T\ge 1\) and all \(\sigma\in[\frac12,1]\), where \(N_L(\sigma,T)\) is the number of zeros \(\rho=\beta+i\gamma\) of \(L(s)\) in the rectangle \(\sigma\le\beta\le 1\), \(|\gamma|\le T\). This is proved using Turán's power sum method, following the argument of \textit{P. X. Gallagher} [Invent. Math. 11, 329--339 (1970; Zbl 0219.10048)]. The zero-density estimate is then fed into the reviewer's analysis [\textit{D. R. Heath-Brown}, Acta Arith. 136, No. 4, 391--395 (2009; Zbl 1169.11038)], in which convexity estimates for an \(L\)-function are explicitly related to the distribution of its zeros.