Local indecomposability of Hilbert modular Galois representations (Q486809): Difference between revisions

Let \(f\) be a weight two (parallel) Hilbert modular form over a totally real field \(F\), and suppose that it is a Hecke eigenform. Let \(K_f\) be its Hecke field, \(\lambda\) be a place of \(K_f\) of residual characteristic \(p\), and \(\rho_f : \mathrm{Gal}\left(\overline{F}/F \right) \to \mathrm {GL}_2\left(K_{f,\lambda} \right)\) be the associated Galois representation. Finally, fix a place \(\mathfrak{p}\) of \(\mathcal{O}_F\) of residual characteristic \(p\) and let \(D_{\mathfrak{p}}\) be the decomposition group of \(\mathrm{Gal}\left(\overline{F}/F \right)\) at \(\mathfrak{p}\). Finally, suppose that \(f\) is nearly \(\mathfrak{p}\)-ordinary. The main theorem of the paper is that if \(f\) does not have complex multiplication, then the restriction of \(\rho_f\) to \(D_{\mathfrak{p}}\) is indecomposable (which partially answers a question of Greenberg). This result comes with a minor technical restriction when \([F:\mathbb{Q}]\) is even, namely that there exists a place \(v\) of \(\mathcal{O}_F\) such that \(\pi_v\) is square integrable, where \(\pi_v\) is the \(v\)-component of the automorphic representation of \(\mathrm{GL}_2(\mathbb{A}_F)\) associated with \(f\). The method of proof is geometric in nature, and relies on a fine study of certain \(\mathrm{GL}(2)\)-type abelian varieties. More precisely, the representation \(\rho_f\) is known to arise (up to twist by a character) from the \(\lambda-\)adic Tate module of an abelian variety \(A_f\), defined over \(F\) and admitting an action of a field \(L\) (a finite extension of \(K_f\)). Assuming \(f\) (hence \(A_f\)) does not have complex multiplication, the author shows that there are two places \(\mathcal{Q}, \mathcal{L}\) of \(F\) (of good reduction for \(A_f\), not lying over \(p\)) such that the endomorphisms algebras of the geometric special fibers of \(A_f\) at \(\mathcal{Q}, \mathcal{L}\) are nonisomorphic. A contradiction is then reached by showing that such primes cannot exist if \(T_\lambda(A_f)\) is decomposable: this is achieved by regarding \(A_f\) as a point on a suitable Shimura variety, whose local properties and endomorphisms are then studied in some detail. It should be pointed out that this technique had already been employed by \textit{H. Hida} in [J. Am. Math. Soc. 26, No. 3, 853--877 (2013; Zbl 1284.14033)] to prove a similar result, which however depended on some unramifiedness assumptions that are completely removed in the work under review. Finally, some consequences of the main theorem are discussed, which include an application to a question of Coleman, asking which modular forms lie in the image of the operator \(\theta\) defined in [\textit{R. F. Coleman}, Invent. Math. 124, No. 1--3, 215--241 (1996; Zbl 0851.11030)]. This improves on a theorem of \textit{M. Emerton} [``A \(p\)-adic variational Hodge conjecture and modular forms with complex multiplication'', preprint, \url{http://www.math.uchicago.edu/~emerton/pdffiles/cm.pdf}].