Hasse principle for pencils of curves of genus one whose Jacobians have rational 2-division points (Q1283461)

Let \(k\) be a number field and \(X\) a smooth, geometrically integral \(k\)-variety. Denoting by \({\mathbb A}_k\) the ring of adeles of \(k\), the Hasse principle is said to hold for \(X\) if \(X({\mathbb A}_k)\neq\emptyset\) implies \(X(k)\neq\emptyset\). In his 1970 ICM address, \textit{Yu. I. Manin} [Actes Congr. internat. Math. 1970, Vol. 1, 401-411 (1971; Zbl 0239.14010)] introduced the following idea to account for the failure of the Hasse principle: Define a pairing \(X({\mathbb A}_k)\times \text{Br} (X)\to {\mathbb Q}/{\mathbb Z}\) (where \(\text{Br} (X)\) is the cohomological Brauer group of \(X\)) by evaluating elements of \(\text{Br}(X)\) at each component and then taking the sum of local invariants (which is known to be finite). By global class field theory, the diagonal image of \(X(k)\) in \(X({\mathbb A}_k)\) is then contained in the subset \(X({\mathbb A}_k)^{\text{Br}}\) of adeles annihilated by the above pairing and thus the emptyness of \(X({\mathbb A}_k)^{\text{Br}}\) is an obstruction to the Hasse principle if \(X({\mathbb A}_k)\) itself is nonempty. One says that the Brauer-Manin obstruction is the only obstruction to the Hasse principle if \(X(k)\neq 0\) is actually equivalent to \(X({\mathbb A}_k)^{\text{Br}}\neq 0\). Manin himself pointed out several examples where this is the case; one of them is that of curves of genus one -- but to show this one has to assume the famous (and partially known) conjecture that the Tate-Shafarevich group of an elliptic curve is finite. During the nearly three decades which elapsed since Manin's address, this idea has been the object of intensive study in which the authors of the present paper have a lion's share. In particular, in recent years the Hasse principle has been much investigated for fibrations over the projective line \({\mathbb P}^1\). Here for the variety to possess a \(k\)-point it is necessary to have a fibre \(X_P\) over some \(k\)-point \(P\) of \({\mathbb P}^1\) with \(X_P({\mathbb A}_k)\neq 0\). To control the existence of such a fibre it is natural to introduce the subgroup \(\text{Br}_{\text{vert}}(X)\subset \text{Br} (X)\) consisting of those elements whose restriction to the generic fibre \(X_{\eta}\) of the fibration is in the image of the natural map \(\text{Br} (k({\mathbb P}^1))\to \text{Br}(X_{\eta})\); there is an associated obstruction of the Brauer-Manin type called the vertical obstruction. In preceding works [see \textit{J.-L. Colliot-Thélène's} recent survey: ``The Hasse principle in a family of algebraic varieties'', in: Number Theory, Proc. int. Conf. on Discrete Math. and Number Theory, Tiruchirapalli 1996, Contemp. Math. 210, 19-39 (1998; Zbl 0910.14008)] this obstruction has been mainly analysed for fibrations whose smooth fibres satisfy the Hasse principle (hence finding a smooth \(X_P\) as above suffices). But until \textit{P. Swinnerton-Dyer}'s paper [``Rational points on certain intersections of two quadrics'', in: Abelian varieties, Proc. int. Conf., Egloffstein 1993, 273-292 (1995; Zbl 0849.14009)] nothing was known, for instance, for pencils of curves of genus at least one. The present paper generalises the results of Swinnerton-Dyer and puts them into a more conceptual perspective. The authors study fibrations in curves of genus one over the projective line under several restrictions of which the most crucial are that the jacobian \(E_{\eta}\) of the generic fibre \(X_{\eta}\) should have 2-division points rational over \(k({\mathbb P}^1)\) and that the class of \(X_{\eta}\) as an \(E_{\eta}\)-torsor in the appropriate WC-group should be of order 2 but not divisible by 2. They then impose a technical condition generalising condition (D) of Swinnerton-Dyer (loc. cit.) and assume the finiteness of the Tate-Shafarevich group for elliptic curves as well as the so-called Schinzel hypothesis (a broad generalisation of the twin prime number conjecture). Under these assumptions the authors prove that the vertical Brauer-Manin obstruction is the only one to the Hasse principle for such surfaces and if moreover there is no such obstruction then the \(k\)-points are actually Zariski dense on \(X\). The authors point out that (apart from the conjectures admitted) all of their conditions are effectively computable and include many examples of surfaces (even of K3 type) for which they can be checked ``by hand''. Furthermore, they conjecture that their theorem should hold unconditionally and that the conclusion should also hold with the vertical Brauer group replaced by the whole of \(\text{Br}(X)\) -- as they remark, this would weaken the assumptions but impair their effective computability. The proof uses techniques from previous papers to show that when there is no vertical obstruction one can find ``good'' fibres with points everywhere locally and then continues with a very delicate analysis of the 2-Selmer groups attached to these fibres to show that the assumptions made ensure the existence of rational points as well. In a last section, the authors relate the technical condition (D) to the Brauer group via a theory of ``geometric Selmer groups'' (which may be of independent interest) and show that it is in fact slightly stronger than the assumption that the 2-primary torsion of \(\text{Br}(X)\) lies entirely in \(\text{Br}_{\text{vert}}(X)\) -- this explains why ``horizontal'' elements of the Brauer group do not obstruct the existence of the \(k\)-points exhibited in the paper. One more remark about the difficult conjectures admitted. As already Manin's result demonstrates, the finiteness of the Tate-Shafarevich group is a natural assumption in these matters (and its general proof is hopefully not totally out of sight). As for Schinzel's hypothesis, it was first applied in this area 20 years ago by Colliot-Thélène and Sansuc and it has turned out to be a fairly general rule since that given a result on rational points conditional on Schinzel's hypothesis, an unconditional proof for the analogous statement concerning zero-cycles of degree one could be obtained by using a trick of P. Salberger. The present paper contains no zero-cycle version but hopefully the question will be settled before long.