On the Hasse principle for Shimura curves (Q836121)

An algebraic curve \(C\) defined over a number field \(K\) may or may not have \(K\)-rational points. One of our most powerful tools for determining if \(C(K) \neq \emptyset\) is to work locally. If \(C(K_v) = \emptyset\) for some place \(v\) of \(K\) (where \(K_v\) denotes the completion at \(v\)), then \(C(K)\) must be empty as well. However, it is well known that there exist curves violating the Hasse principle: \(C(K_v) \neq \emptyset\) for all places \(v\) of \(K\), and yet \(C(K) = \emptyset\). By staring at a set of defining equations for \(C_{/K}\), one infers that there is a finite extension \(L / K\) such that \(C\) has \(L\)-rational points. One can similarly play this game at each place of \(K\) to ensure the existence of points everywhere locally. But can one go half way between these two constructions? That is, starting with a curve \(C_{/K}\) with no \(K\)-rational point, does there exist a finite extension \(L / K\) such that \(C_{/L}\) has points everywhere locally, but \(C(L) = \emptyset\)? If so, one says that \(C_{/K}\) is a \textit{potential violation of the Hasse principle} (\(C_{/K}\) is PVHP). Clark makes the following guess on the subject: \textbf{Main Conjecture.} Let \(C\) be a curve defined over a number field \(K\). At least one of the following holds: (i) \(C\) has genus~0; (ii) \(C\) has \(K\)-rational points; (iii) \(C_{/K}\) is a potential Hasse principle violation. The author provides a gonality criterion for determining if a curve is PVHP, which is an easy consequence of theorems of Faltings and Frey (Theorem~6). Then he uses it to exhibit several convincing pieces of evidence for the Main Conjecture by constructing infinitely many Shimura curves that violate the Hasse principle. Let \(D > 1\) and \(N \geq 1\) be squarefree coprime integers, and let \(X_0^D(N)_{/\mathbb Q}\) and \(X_1^D(N)_{/\mathbb Q}\) be the usual Shimura curves. \textbf{Theorem 1.} If \(D > 546\), there is an integer \(m\) such that \(X_0^D(1)/_{\mathbb Q(\sqrt m)}\) violates the Hasse principle. \textbf{Theorem 2.} There exists a constant \(C\) such that if \(D \cdot N > C\), then there exist number fields \(K = K(D,N)\) and \(L = L(D,N)\) such that \(X_0^D(N)_{/K}\) and \(X_1^D(N)_{/L}\) violate the Hasse principle. \textbf{Theorem 3.} Assume \(D \cdot N > C\) in the setting of Theorem~2. (a) One may choose \(K\) such that \([K:\mathbb Q] \mid 4\). (b) Let \(\{N_i\}\) be a sequence of squarefree positive integers tending to infinity, and for each \(i\), choose any squarefree positive integer \(D_i > 1\) that is prime to \(N_i\) and such that \(D_i \cdot N_i > C\). For all \(i\), choose any number field \(L_i\) such that \(X_1^{D_i}(N_i)_{/L_i}\) violates the Hasse principle. Then \([L_i: \mathbb Q] \to \infty\) as \(i \to \infty\). In the final section of the paper, the author indicates that the set of counterexamples to the Main Conjecture among all semistable Shimura curves over all totally real fields is finite (so possibly empty). He deduces it from a more general result on sequences of semistable curves whose \(K\)-gonality tends to infinity much faster than the log of the genus.