Elliptic curves over real quadratic fields are modular (Q2516379)

This paper proves the beautiful result that every elliptic curve over a real quadratic field \(K\) is modular (i.e., arises from a Hilbert newform over \(K\)). Previously, it was established by Breuil, Conrad, Diamond, and Taylor [\textit{C. Breuil} et al., J. Am. Math. Soc. 14, No. 4, 843--939 (2001; Zbl 0982.11033)], that every elliptic curve over \(\mathbb Q\) is modular, following the pioneering work of Wiles and Taylor-Wiles in the semistable case. A crucial starting point for the paper are modularity lifting theorems, which have become increasingly powerful since the initial work of Wiles and Taylor-Wiles as a result of work by many authors, including Barnet-Lamb, Breuil, Diamond, Gee, Geraghty, and Kisin. In particular, for an elliptic curve \(E\) over a totally real field \(K\), these results imply that \(E\) is modular, provided its residual Galois representation \(\bar \rho = \bar \rho_{E,p}\) is modular and \(\bar \rho \mid_{G_K(\zeta_p)}\) is absolutely irreducible (\(p \not= 2\)). Using a 3-5 and 3-7 switching argument, it is shown that \(E\) is modular unless \({\bar \rho}_{E,p} \mid G_{K(\zeta_p)}\) are simultaneously absolutely reducible for \(p = 3, 5, 7\). Such an elliptic curve \(E\) gives rise to a \(K\)-rational point on one of 27 modular curves, denoted by \(X(u,v,w)\), where \(u \in \left\{b3,s3,ns3 \right\}, v \in \left\{ b5,s5,ns5 \right\}, w \in \left\{ b7, s7, ns7 \right\}\). The symbols \(b, s, ns\) refer to Borel, normalizer of split Cartan, normalizer of non-split Cartan level structures on \(E\) of the indicated level. We remark that the 3-7 switching argument is essential and its proof is nontrivial. A more refined analysis shows that in fact \(E\) gives rise to a \(K\)-rational point on one of seven modular curves \[ X(b5,b7), X(b3,s5), X(s3,s5), \] and \[ X(b3,b5,d7), X(s3,b5,d7), X(b3,b5,e7), X(s3,b5,e7), \] where \(e\) and \(d\) refer to a certain cover of the modular curves with normalizer of split Cartan and non-split Cartan level structures, respectively. For \(X\) being one of the first group of modular curves above, the real quadratic points on the first group of modular curves are determined to be defined over \(\mathbb Q\), have CM or are \(\mathbb Q\)-curves, and hence are modular. This is accomplished by embedding the symmetric square \(X^{(2)}\) of \(X\) into the jacobian \(J\) of \(X\) and determining \(J(\mathbb Q)\). For \(X\) being one of the second group of modular curves, \(X\) is viewed as (the normalization of) the fibered product \(X_1 \times_{X(1)} X_2\) using the \(j\)-map to \(X(1)\), where \[ X_1 = X(b3,b5), X(s3,b5) \text{ and } X_2 = X(d7), X(e7). \] Using the natural map \(X^{(2)} \rightarrow J_1 \times J_2\) and a novel Mordell-Weil sieve that applies in this situation, it is shown that the real quadratic points on \(X\) are non-existent or have \(j\)-invariant in \(\mathbb Q\) (so are modular). Modularity results for elliptic curves \(E\) over a general totally real field are also given, with some additional restrictions.