Quantitative spectral gap for thin groups of hyperbolic isometries (Q2018709)

One of the most important problems in number theory is the Selberg conjecture [\textit{A. Selberg}, Proc. Sympos. Pure Math. 8, 1--15 (1965; Zbl 0142.33903)] claiming that the first non-zero eigenvalue \(\lambda_1(X'(N))\) of the Laplacian operator on the quotient \(X'(N)=\mathbb{H}^2\slash\Gamma'\) of the hyperbolic plane \(\mathbb{H}^2\) by an arithmetic congruence subgroup \(\Gamma'\) of \(\mathrm{SL}_2(\mathbb{Z})\) of level \(N\) is at least \(1/4\). There are several known lower bounds for \(\lambda_1(X'(N))\) towards the Selberg conjecture, currently the best one being that of H. H. Kim and P. Sarnak [\textit{H. H. Kim}, J. Am. Math. Soc. 16, No. 1, 139--183 (2003; Zbl 1018.11024)]. Similar results, giving lower bounds for eigenvalues of the Laplacian operator, for arithmetic quotients of higher-dimensional hyperbolic space by an arithmetic subgroup, are obtained in [\textit{P. Sarnak}, Acta Math. 151, 253--295 (1983; Zbl 0527.10022); \textit{J. Elstrodt} et al., Invent. Math. 101, No. 3, 641--685 (1990; Zbl 0737.11013); \textit{J. Cogdell} et al., Acta Math. 167, No. 3--4, 229--285 (1991; Zbl 0743.11028); \textit{N. Bergeron} and \textit{L. Clozel}, Invent. Math. 192, No. 3, 505-532 (2013; Zbl 1309.22014)]. This kind of result regarding the spectrum of the Laplacian operator is referred to as spectral gap. The paper under review considers the spectral gap problem for the following family of manifolds. Let \(\Lambda\) be a subgroup of an arithmetic lattice in \(\mathrm{SO}(n+1,1)\), the special orthogonal group of signature \((n+1,1)\). The quotient \(X=\mathbb{H}^{n+1} / \Lambda\) of the hyperbolic manifold of dimension \(n+1\) by the action of \(\Lambda\) exhibits a natural family of covers \(X(\mathcal{I})=\mathbb{H}^{n+1} / \Lambda(\mathcal{I})\) indexed by the ideals \(\mathcal{I}\) in the ring of integers of the underlying number field. Under several technical assumptions on \(\Lambda\), to be specified below, the main result of the paper is a quantitative spectral gap for the eigenvalues of the Laplacian operator on these manifolds, generalizing the results of \textit{P. Sarnak} and \textit{X. X. Xue} [Duke Math. J. 64, No.1, 207--227 (1991; Zbl 0741.22010)] and \textit{A. Gamburd} [Isr. J. Math. 127, 157--200 (2002; Zbl 1009.22018)]. This is obtained by passing (away from some primes) to the setting of associated family of Cayley graphs \(\mathcal{H}_{\mathcal{I}}\) and proving a certain ``super-strong'' approximation property for this family of graphs. It is related to the graph Laplacian and the strong connectedness properties of graphs, in particular, the behavior of expander coefficients. More precisely, let \(\mathcal{G}\) be the special orthogonal group of a quadratic form on an \((n+2)\)-dimensional vector space over a totally real number field \(F\), such that \(\mathcal{G}\) is \(\mathrm{SO}(n+1,1)\) at one Archimedean place, while at the remaining Archimedean places it is compact. Define the arithmetic subgroup \[ \Gamma =\mathcal{G}(F)\cap \mathrm{GL}_{n+2}(\mathcal{O}_F), \] where \(\mathcal{O}_F\) is the ring of integers of \(F\). Let \(\Lambda\) be a subgroup of \(\Gamma\), and given an ideal \(\mathcal{I}\) of \(\mathcal{O}_F\), let \[ \Lambda(\mathcal{I})=\{\gamma\in\Lambda \,:\, \gamma\equiv I \mod\mathcal{I} \}, \] where \(I\) is the identity. Let \(X=\mathbb{H}^{n+1}/\Lambda\) be the quotient manifold. Then \(X(\mathcal{I})=\mathbb{H}^{n+1}/\Lambda(\mathcal{I})\) is a family of congruence covers of \(X\) considered in the paper. The assumptions on \(\Lambda\) under which the main result of the paper is proved are the following: {\parindent=6mm \begin{itemize} \item[-] Algebraic fullness: \(\Lambda\) is Zariski-dense in \(\mathcal{G}(\mathbb{C})\) and the traces of its adjoint representation generate \(\mathcal{O}_F\). This can be thought of as thickness of \(\Lambda\), that is, although \(\Lambda\) is possibly of infinite index in \(\Gamma\), it is still thick enough. \item [-] Geometric regularity: the image of \(\Lambda\) at the Archimedean place at which \(\mathcal{G}\) is not compact is geometrically finite, orientation preserving and torsion free as an isometry group of \(\mathbb{H}^{n+1}\). Geometric finiteness means that any Dirichlet fundamental domain for \(\Lambda\) in \(\mathbb{H}^{n+1}\) is finitely faced. It implies that \(\Lambda\) is finitely generated. The remaining conditions can be satisfied by passing to a subgroup of finite index. \item [-] Fractal fullness: the Hausdorff dimension \(\delta\) of the limit set of \(\Lambda\) is greater than \(s_n^0=n-\frac{2(n-1)}{(n+1)(n+2)}\). The limit set of \(\Lambda\) is a Cantor-like set in the boundary of \(\mathbb{H}^{n+1}\), on which the orbit \(\Lambda o\) of a point \(o\in\mathbb{H}^{n+1}\) accumulates. The bound on its Hausdorff dimension makes sure that \(\Lambda\) is ``fractally'' thick enough, so that the ``super-strong'' approximation alluded to above holds. \end{itemize}} Under these assumptions on \(\Lambda\), passing to a finite index subgroup if necessary, the main result says that, for \(|\mathcal{O}_F / \mathcal{I}|\) large enough and \(\mathcal{I}\) coprime to a finite set of primes, \[ \Omega(X(\mathcal{I}))\cap \big[ \delta (n-\delta),s_n^0(n-s_n^0) ) = \Omega(X)\cap \big[ \delta (n-\delta),s_n^0(n-s_n^0) ), \] where \(\Omega(\cdot)\) denotes the spectrum of the Laplacian operator on the corresponding manifold. The point is that, in the specified interval, there are no additional eigenvalues when passing to the cover \(X(\mathcal{I})\). As a consequence, under the same assumptions, a lower bound \[ \lambda_1(X(\mathcal{I}))\geq \min\big(\lambda_1(X),s_n^0(n-s_n^0)\big) \] for the second smallest eigenvalue \(\lambda_1(X(\mathcal{I}))\) of the Laplacian operator on \(X(\mathcal{I})\) is obtained.