On truncation of irreducible representations of Chevalley groups (Q740861)

Let \(\mathbf{G}\) be a semisimple algebraic group defined over \(\mathbb{Q}\), which will in addition be assumed to be split over \(\mathbb{Q}\). If \(\lambda\) is a weight of \(\mathbf{G}\) denote by \(L_\lambda\) the representation of \(\mathbf{G}\) of highest weight \(\lambda\). If \(\Gamma\) is a subgroup of \(\mathbf{G}(\mathbb{Z})\) of finite index (where the latter is defined with respect to some \(\mathbb{Z}\)-lattice in \(L_\lambda(\mathbb{Q})\)), there are defined cohomology groups \(H^*(\Gamma,L_\lambda(\mathbb{Q}_p))\) for any rational prime \(p\). On these spaces there is also a distinguished operator \(T\), the Hecke operator, and a decomposition into eigenspaces of \(H^*(\Gamma,L_\lambda(\overline{\mathbb{Q}_p}))\). For \(\alpha\in\mathbb{Q}, \alpha\geq 0\) the subspace of slope \(\alpha\) is then defined to be the sum of all eigenspaces with eigenvalues \(a\) of \(p\)-valuation \(v_p(a)=\alpha\) (note that all eigenvalues of \(T\) are integral); this is a \(\mathbb{Q}_p\)-rational subspace, denoted \(H^*(\Gamma,L_\lambda(\mathbb{Q}_p))^\alpha\). The main result of the paper is then the following : for any \(\beta\geq 0\) the sum of dimensions \[ \sum_{\alpha\leq\beta} \dim_{\mathbb{Q}_p} H^i(\Gamma,L_\lambda(\mathbb{Q}_p))^\alpha \leq C \] is bounded independently of \(\lambda\), for any degree \(i\) (the constant depends a priori on the prime \(p\) and the group \(\Gamma\)), under a condition at \(p\) on \(\Gamma\) (that it be contained in an explicit open subgroup of \(\mathbf{G}(\mathbb{Z}_p)\)). The main ideas of the proof are well-explained in the introduction to the present paper, let us try to give an even shorter summary here. It rests on the truncation of \(\Gamma\)-modules from which the paper draws its title: by this it is meant a finite quotient \(L_\lambda^{[r]}(\mathbb{Z}_p)\) of \(L_\lambda(\mathbb{Z}_p)\) on which \(\Gamma\) acts (this is what the congruence condition is needed for), depending on a positive integre \(r>0\). It is roughly defined as follows: for a weight \(\mu\) one kills the integral elements of the \(\mu\)-eigenspace of \(p\)-valuation larger than \(r-\mathrm{ht}(\lambda-\mu)\) (where \(\mathrm{ht}\) is the polyhedral norm on the root lattice). In particular one only sees what is close to \(\lambda\) (where the `close' depends on \(r\) --this last sentence is made precise by Proposition B in the introduction). Then, the author gives an estimate of the size of the subspace of slope \(\alpha\) by the size of the cohomology \(H^*(\Gamma, L_\lambda^{[r]}(\mathbb{Z}_p)\) (Proposition A in the introduction).