The fourth moment of Dirichlet \(L\)-functions (Q640735): Difference between revisions

It is shown that there is a quartic polynomial \(F(x)\) such that if \(p\) is prime then \[ \sum_{\chi\,\text{mod }p}|L(\tfrac{1}{2},\chi)|^4= pF(\log p)+O_{\theta}(p^{\theta}) \] for any exponent \(\theta>507/512\). Previously the reviewer [Analysis 1, 25--32 (1981; Zbl 0479.10027)] had obtained an asymptotic formula saving only a factor \(\log p\). To obtain a power saving was a very considerable challenge. \textit{J. B. Conrey} et al. [Proc. Lond. Math. Soc. (3) 91, No. 1, 33--104 (2005; Zbl 1075.11058)] have made a general conjecture which gives an asymptotic formula for \[ \sum_{\chi\mod{p}}L(\tfrac{1}{2}+\alpha,\chi)L(\tfrac{1}{2}+\beta,\chi) L(\tfrac{1}{2}+\gamma,\overline{\chi})L(\tfrac{1}{2}+\delta,\overline{\chi}) \] in particular. The present paper verifies this, up to an error term \(O(p^{\theta})\). The analysis depends essentially on estimates for a divisor sum \[ \sum_{m\equiv n\,\text{mod }p}\frac{d(m)d(n)}{\sqrt{mn}}W(\frac{mn}{p^2}), \] in which \(W\) is a smooth weight. Classical methods using Weil's bound for the Kloosterman sum enable one to obtain a power saving for terms in which \(n/m\geq p^{1+\delta}\) say, but other ranges of \(m\) and \(n\) are distinctly more troublesome. The proof employs a number of estimates, some of which use Weil's bound more efficiently to handle the averaging implicit in the above divisor sum. However the case in which \(m\) and \(n\) are roughly of equal size requires more sophisticated tools, inspired by the work of \textit{Y. Motohashi} [Spectral theory of the Riemann zeta-function. Cambridge: Cambridge University Press (1997; Zbl 0878.11001)]. At a key point one uses the bound of Kim and Sarnak [\textit{H. H. Kim}, J. Am. Math. Soc. 16, No. 1, 139--183 (2003; Zbl 1018.11024)] for Hecke eigenvalues of Maass cusp forms, in which one has an exponent \(\phi=7/64\). The exponent \(507/512\) mentioned above arises as \(1-1/80+\phi/40\).