On the growth of torsion in the cohomology of arithmetic groups (Q2249468)

Let \(G\) be one of the Lie groups \(\mathrm{SL}_3(\mathbb{R})\) or \(\mathrm{SO}(p,q)\) (where \(p\) and \(q\) are odd) and \(\Gamma\) a torsion-free cocompact lattice in \(G\) (which is necessarily arithmetic by Margulis' theorem). In this paper, the authors study the torsion subgroup in the cohomology groups of the simplest such \(\Gamma\) with coefficients in a growing sequence of \(\mathbb{Z}[\Gamma]\)-modules. They prove that such torsion can be found in abundance for some natural sequences by giving a lower bound for the size of the torsion subgroup of the total cohomology, generalizing a previous result of the first author with \textit{S. Marshall} which is applied to some hyperbolic 3-manifolds [Duke Math. J. 162, No. 5, 863--888 (2013; Zbl 1316.11042)]. The restriction to the Lie groups mentioned above is natural since for other Lie groups it is expected that the size of torsion subgroups should be smaller -- for more on this heuristic see the work of \textit{N. Bergeron} and \textit{A. Venkatesh} [J. Inst. Math. Jussieu 12, No. 2, 391--447 (2013; Zbl 1266.22013)]. Let us describe first the results for \(G = \mathrm{SL}_3(\mathbb{R})\). The simplest cocompact lattices in this group come from the unit groups of orders in nine-dimensional central simple algebras over \(\mathbb{Q}\) which split over \(\mathbb{R}\). Denoting by \(\Lambda\) the weight of a representation of \(G\) which is trivial on the center, and by \(V_\Lambda(m)\) the space of the representation of \(G\) with highest weight \(m\Lambda\), it is proven that each \(V_\Lambda(m)\) contains a \(\Gamma\)-invariant lattice \(M_\Lambda(m)\) and that the cohomology groups \(H^*(\Gamma, M_\Lambda(m))\) are finite. The main result is then the following: if \(X\) denotes the compact 5-dimensional Riemannian manifold \(\Gamma\backslash\mathrm{SL}_3(\mathbb{R})/\mathrm{SO(3)}\), the asymptotic expansion: \[ \sum_{p=0}^5 (-1)^p\log|H^p(\Gamma, M_\Lambda(m))| = C\cdot \mathrm{vol}(X)\cdot m\dim(V_\Lambda(m)) + O(\dim(V_\Lambda(m)), \] holds, where \(C\) is a constant, depending only on \(\Lambda\), which is computed explicitly in the paper. In particular this implies that the torsion subgroup in the cohomology groups of odd degree is finite and its size grows exponentially in \(m^3\). The authors conjecture that only \(H^3(\Gamma,M_\Lambda(m))\) should have this behaviour, and also that the even cohomology is sub-exponential (in \(m^3\)). For \(\mathrm{SO}(p,q)\) they consider arithmetic lattices coming from quadratic forms over totally real number fields and prove a result similar to the one above. The proof in both cases are the same: let us explain it in the case mentioned above, keeping the same notation. By a result of the first author the alternate product of orders of cohomology groups \(H^j(\Gamma,M_\Lambda(m))\) on the left-hand side above is equal to the analytic torsion of \(X\) with coefficients in the flat bundle over \(X\) induced by the representation \(\pi_1(X) = \Gamma \to \mathrm{SL}(V_\Lambda(m))\). This analytic torsion was studied in previous work of the authors [J. Differ. Geom. 95, No. 1, 71--119 (2013; Zbl 1281.58022)] and it was proven to have the behaviour described by the right-hand side. Note that this analytical part of the proof completely breaks down for non-uniform lattices, which is the main reason why they are not considered in this paper.