The distribution of \(\ell^{\infty }\)-Selmer groups in degree \(\ell\) twist families. I (Q6918366)

This paper, together with its sequel (titled ``The distribution of \(\ell^{\infty }\)-Selmer groups in degree \(\ell\) twist families. II'', also accepted at JAMS), introduces a new technique for obtaining the distribution of class groups and Selmer groups. A necessary prerequisite to apply this technique is that the relevant class groups and Selmer groups share a common Galois module as ``base layer''. For example, in the case of class groups of quadratic fields, one has a common description of their \(2\)-torsion through Gauss genus theory. More generally, the technique applies in the so-called \(n = p\) setting in arithmetic statistics: here one is studying \(p\)-Selmer groups (or even \(p^\infty\)-Selmer groups) in a family of Galois modules that become isomorphic over an extension of degree \(n\).\N\NAll previous results in this area could only obtain the distribution of either the \(p\)-Selmer group or sometimes the \(p^2\)-Selmer group. In spectacular fashion, the author is able to control the full \(p^\infty\)-Selmer group. The most general result of this type is proven in Section 2 of the sequel paper. Here we sample some of its consequences.\N\NLet \(E/\mathbb{Q}\) be an elliptic curve with either \(E(\mathbb{Q})[2] = 0\) or \(E(\mathbb{Q})[2] \cong \mathbb{F}_2^2\), but \(E\) has no cyclic \(4\)-isogeny defined over \(\mathbb{Q}\) (the author also makes some partial progress in the case that \(E(\mathbb{Q})[2] \cong \mathbb{F}_2\)). Then 50\% of the quadratic twists of \(E\) have \(2^\infty\)-Selmer rank \(0\), and 50\% have \(2^\infty\)-Selmer rank \(1\). In particular, \(0\%\) of the quadratic twists have rank \(\geq 2\). The author was able to extend this to all elliptic curves over \(\mathbb{Q}\) in a recent arXiv preprint.\N\NThere are also direct consequences to the statistics of class groups. In this setting the author proves the Cohen-Lenstra-Gerth conjecture (and in fact, an extension of that conjecture to number fields provided that the number fields do not have additional roots of unity). In the case of the rational numbers, the theorem reads as follows. Let \(A\) be a finite, abelian \(2\)-group. Then the proportion of imaginary quadratic fields \(K\), ordered by their discriminant, with \(2\mathrm{Cl}(K)[2^\infty] \cong A\) is proportional to \(1/\#\mathrm{Aut}(A)\). Note that \(2\mathrm{Cl}(K)[2^\infty]\) ``removes'' precisely the \(2\)-torsion from \(\mathrm{Cl}(K)[2^\infty]\): this is a very natural thing to do as the \(2\)-torsion has a predictable structure by Gauss genus theory.\N\NThe key idea underlying the proofs is to exploit, as far as possible, the isomorphism between the base-layer Galois modules. The main insight of the author is that this isomorphism persists in a much subtler way to give structural relations between different Cassels-Tate pairings. This leads to a so-called reflection principle, of which many examples can be found in the literature, although the author's reflection principle is much more general than any previously known result. Analytical arguments (in this case a bilinear sieve) prove that the output of the reflection principle is equidistributed, and an ingenious combinatorial argument is then used to deduce equidistribution of the Cassels-Tate pairing.