The mean square of Dirichlet L-functions. (A generalization of Balasubramanian's method) (Q2277008)

Let \(\chi\) be a primitive Dirichlet character mod \(q\), and \(L(s,\chi)\) the corresponding Dirichlet \(L\)-function. Let \[ E(T,\chi)=\int^T_0 | L(1/2+it,\chi)|^2 \,dt - \frac{\phi (q)}{q}(\log \frac{qT}{2\pi}+2\gamma -1+2\sum_{p| q}\frac{\log p}{p-1})T \] denote the error term in the mean square formula for \(L(1/2+it,\chi)\), where \(\gamma\) is Euler's constant and \(\varphi\) is the Euler function. \textit{Y. Motohashi} proved [Proc. Japan Acad., Ser. A 61, 313--316 (1985; Zbl 0583.10024)], among other things, that \[ E(T,\chi)\ll ((qT)^{1/3}+q^{1/2})(\log qT)^4 \] uniformly for \(T\geq 2\) and any \(q\). His argument was a natural extension of the work of \textit{F. V. Atkinson} [Acta Math. 81, 353--376 (1949; Zbl 0036.18603)] on the mean square of the Riemann zeta-function. Prior to Motohashi's work, the author [Proc. Japan Acad., Ser. A 58, 443--446 (1982; Zbl 0514.10030) and ibid. 65, 344 (1989; Zbl 0699.10059)], investigated the same problem by a different method. His basic idea was to combine the method of \textit{R. Balasubramanian} [Proc. Lond. Math. Soc. (3) 36, 540--576 (1978; Zbl 0375.10025)] with Weil's estimate for Kloosterman sums. Unfortunately his argument contained a serious error, which led to an erroneous result. In the article under review the author sketches the corrected argument, which gives \[ E(T,\chi)\ll (qT)^{1/3+\varepsilon} \] when \(q\) is odd and \(T\gg q^{20}\). This is the right generalization of Balasubramanian's theorem, although it is still weaker than Motohashi's result. A lengthy manuscript exists, in which all the details of the proof are worked out.