On the mean-square of the error term related to \(\sum_{n\leq x} \lambda^2 (n^j)\) (Q544856)

The authors prove a mean square result for \[ E_j(x) := \sum\limits_{n\leqslant x}\lambda^2(n^j) - c_{(j-1)}x\qquad(j = 2,3,4), \] where \(c_{(j-1)}\) is a suitable positive constant, and the notation is as follows. Let \(S_k(\Gamma)\) be a holomorphic cusp form of even integral weight \(k\) for the full modular group. Let further \(f(z)\) be an eigenfunction of all Hecke operators belonging to \(S_k(\Gamma)\). Then the Hecke eigenform \(f(z)\) has the Fourier expansion \[ f(z) = \sum_{n=1}^\infty a(n){\roman e}^{2\pi inz} \] which is normalized by setting \(a(1)=1\), and \(\lambda(n) := a(n)n^{(1-k)/2}\). The authors' main result is that, for \(j = 2,3,4\), \[ \int_0^x|E_j(y)|^2 \,dy \ll_{f,\varepsilon} x^{2c^*+1+\varepsilon} \] if \(c* \leqslant 1 - 2(j+1)^{-2}\). This is better than the bound which follows from \[ E_j(x) \ll_ \varepsilon x^{C_j+\varepsilon},\quad C_j = 1 - 2/((j+1)^2+2), \] which is a recent result of the authors [Proc. Am. Math. Soc. 137, No. 8, 2557--2565 (2009; Zbl 1225.11059)]. The key ingredient in the proof, which is of independent interest, is a mean value theorem for the corresponding Dirichlet series \(Z(s)\), precisely defined in the text. The condition that the result holds for \(j = 2,3,4\) is clear from the decomposition given in Lemma 3.2 for the associated Rankin-Selberg zeta-function.