There are genus one curves of every index over every infinite, finitely generated field (Q1741448): Difference between revisions

The authors first prove a generalisation of a result of Shafarevich: if \(K\) is an infinite finitely generated field or more generally a Hilbertian field (in Shafarevich's original result, \(K\) is a number field), and \(A\) is an abelian variety over \(K\) such that \(A(K)/nA(K)\) is finite for some \(n>1\), and \(\operatorname{char}K\) does not divide \(n\), then there are infinitely many classes in the Galois cohomology group \(H^1(K,A)\) of period (i.e.\ order) \(n\). This part is relatively straightforward, though not trivial: it ultimately depends on a well-known result of \textit{M. Ikeda} [Abh. Math. Semin. Univ. Hamb. 24, 126--131 (1960; Zbl 0095.02901)]. Even if \(\operatorname{char}K\) divides \(n\), the result still holds as long as \(A\) has a point of order \(n\) over a separable closure of \(K\). The second result of the paper is the one described by the title. It is more striking, not least because it has no hypothesis on the characteristic. The result is that if \(K\) is an infinite finitely generated field then there is an elliptic curve \(E\) over \(K\) such that for every \(n>1\), there are infinitely many classes \(\eta\in H^1(K,E)\) whose period and index are both equal to \(n\). (This is index in the sense of Galois cohomology: the gcd of the degrees of all splitting fields for \(\eta\). It is a multiple of the degree and divides a power of the degree but is generally hard to control.) The proof is partly an enhanced version of an earlier result of the first author [J. Reine Angew. Math. 594, 201--206 (2006; Zbl 1097.14024)]. To produce the required elliptic curve requires some control over the Tate-Shafarevich group, and thus relies on know special cases of the Birch-Swinnerton-Dyer conjecture for function fields.