Prime geodesic theorem for the Picard manifold (Q2213776)

Let \(\mathbb{H}^3\) be the 3-\(d\) hyperbolic Riemannian space, \(\mathrm{PSL}(2,\mathbb{C})\simeq \mathrm{Isom}(\mathbb{H}^3)\) its isometry group and \(\Gamma\) a discrete cofinite torsion free group in \(\mathrm{Isom}(\mathbb{H}^3)\). Any non trivial closed geodesic is associated uniquely to a conjugacy class \([\gamma]\) of an hyperbolic or loxodromic group element whose norm \(N(\gamma)\) determines its length. The function \(\pi_\Gamma(X)=\{[\gamma]|N(\gamma)\le X\}\) counts the closed geodesic on \(X_\Gamma=\Gamma \backslash\mathbb{H}^3\) according to their length. The counting function \(\pi_\Gamma(X)\) decomposes along \(\pi_\Gamma(X)=\mathrm{Li}(X^2)+E_\Gamma(X)\) as the sum of a principal asymptotic term and a remainder: upper bounds on this remainder \(E_\Gamma(X)\), of great interest in number theory, are called Prime Geodesic Theorem (PGT), giving finer and finer error estimates with different asymptotics. \textit{P. Sarnak} [Acta Math. 151, 253--295 (1983; Zbl 0527.10022)] established the upper bound \(E_\Gamma(X)=\mathcal{O}(X^{5/3+\varepsilon})\) with any small \(\varepsilon>0\). In case of the Picard group \(\Gamma=\mathrm{PSL}(2,\mathbb Z[\mathrm{i}])\) induced by the Gaussian integers \(\mathbb Z[\mathrm{i}]\), different PGT have been proved, either with a \(L\)-function hypothesis by \textit{S. Y. Koyama} [Forum Math. 13, No. 6, 781--793 (2001; Zbl 1061.11024)] or unconditionally by \textit{O. Balkanova} et al. [Trans. Am. Math. Soc. 372, No. 8, 5355--5374 (2019; Zbl 1479.11085)]. The main result proved in this paper is the PGT for the Picard manifold \[ E_\Gamma(X)=\mathcal{O}\left(X^{3/2+\theta/2+\varepsilon}\right). \] Here \(\varepsilon>0\) is a priori given, while \(\theta\) denotes a subconvexity exponent for quadratic Dirichlet \(L\)-function defined over Gaussian integers: \textit{P. Nelson} [``Eisenstein series and the cubic moment for \(\mathrm{PGL}_2\)'', Preprint, \url{arXiv.1911.06310}] proved that \(\theta=1/3\) may be taken. The proof of this new remainder estimate is based on two upper bounds \begin{itemize} \item For \(X\) big enough, \(T\in [X^\varepsilon,X^{1/2}]\) and \(|t|=\mathcal O(T^\varepsilon)\) \[ \sum_{r_j}\frac{r_j}{\sinh(\pi r_j)}\omega_\Gamma(r_j) X^{\mathrm{i}r_j}L(u_j\otimes u_j, 1/2+it)=\mathcal{O}\left(T^{3/2}X^{1/2+\theta+\varepsilon}\right).\tag{1} \] \item For \(T\in[1,X^{1/2}]\), \[ \sum_{0<r_j\le T}X^{\mathrm{i}r_j}=\mathcal{O}\left(X^{(1+\theta)/2}T(TX)^\varepsilon\right).\tag{2} \] \end{itemize} Here the family \((u_j,1+r_j^2)\) is a maximal orthonormal cusp forms \((u_j)\) basis, with corresponding eigenvalues \(1+r_j^2\), the function sums \(\omega_T\) is a smooth characteristic function of the interval \([T,2T]\) and \(L(u\otimes u,s)\) is a Maaß-Rankin-Selberg \(L\)-function. The proof core is to establish the upper bound (1) which implies the estimate (2), the PGS being a consequence of (2). The main idea to prove (1) is to avoid the usual way where bounds are obtained through crude absolute value estimates on terms in the (1) left-hand side. Instead the authors introduce exact formulæ for the first moment of Maaß-Rankin-Selberg which allows to consider oscillations of the exponentials \(X^{\mathrm{i}r_j}\). This method has been used successfully for the 2-d modular surface \(\mathrm{PSL}(2,\mathbb{Z})\backslash\mathbb{H}^2\) by the two authors in [J. Lond. Math. Soc., II. Ser. 99, No. 2, 249--272 (2019; Zbl 1456.11092)] with improvements on already known PGTs. The current proof used Kuznetsov formula to win finer control on Kloosterman sums. Some features in 3-d are new (e.g. geometry of the Picard manifold, special functions in the trace formula) and are solved in the particular case of the Picard manifold.