Subgroups of \(\mathrm{Spin}(7)\) or \(\mathrm{SO}(7)\) with each element conjugate to some element of \(\mathrm{G}_2\) and applications to automorphic forms (Q1738361): Difference between revisions

Let \(G\) be a group and \(H\) be a subgroup of \(G\). Consider the following group-theoretic property \(\mathcal{P}(G, H)\): for all subgroups \(\Gamma\) of \(G\), if every element of \(\Gamma\) is conjugate to some element of \(H\), then \(\Gamma\) is conjugate to a subgroup of \(H\). A reasonable question is to try to classify the pairs \((G, H)\) such that \(\mathcal{P}(G, H)\) holds, say for \(G\) a connected compact Lie group and \(H\) a closed connected subgroup. When \(\mathcal{P}(G, H)\) fails, one can also try to classify the exceptions \(\Gamma\) violating that property. In the first part of this work, the author proves that \(\mathcal{P}(\mathrm{Spin}(7), \mathrm{G}_2)\) holds (Theorem A). More generally, there is a version (Theorem B) for \(\mathrm{Spin}(E)\), where \(E\) is the \(7\)-dimensional quadratic space of pure quaternions in a quaternion \(k\)-algebra \(C\) (where \(k\) is a field), and \(H\) is \(\mathrm{Aut}_{k\text{-alg}}(C)\) embedded in \(G\), which is a semisimple group of type \(\mathrm{G}_2\). The corresponding statement for \((\mathrm{SO}(7), \mathrm{G}_2)\) fails. The possible exceptions \(\Gamma\) are reasonably rare, and a complete description is given in Theorems C and D. The proofs of these results are group-theoretic or invariant-theoretic in nature. As an application, one proves in Theorem F the following statement about Langlands' functoriality conjecture for automorphic representations. Fix an irreducible algebraic representation \(\rho: \mathrm{G}_2(k) \to \mathrm{GL}_7(k)\) where \(k\) is a suitable algebraically closed field of characteristic zero. Let \(F\) be a totally real number field, \(\pi\) a cuspidal automorphic representation of \(\mathrm{GL}_7(\mathbb{A}_F)\) that is regular algebraic at each real place of \(F\), with a coefficient field \(E\), such that the Satake parameter of \(\pi_v\) is conjugate into \(\rho(\mathrm{G}_2(\mathbb{C}))\) for almost all places \(v\). Let \(\ell\) be a prime number and \(\lambda\) a place of \(E\) lying over \(\ell\). Then there is a continuous semisimple homomorphism \[ \tilde{r}_{\pi, \lambda}: \mathrm{Gal}_F \to \mathrm{G}_2(\overline{E_\lambda}), \] unique up to \(\mathrm{G}_2(\overline{E_\lambda})\)-conjugacy, such that \(\tilde{r}_{\pi, \lambda}\) is unramified whenever \(\pi_v\) is (with \(v\) coprime to \(\ell\)), and the images of these unramified parameters in \(\mathrm{GL}_7\) have matching characteristic polynomials. The proof is based on the existence of the compatible system of \(\ell\)-adic representations attached to \(\pi\) together with the Theorem D alluded to above. The exceptional cases in Theorem D are ruled out in this application, using the knowledge about the Hodge-Tate weights of these representations.