Estimating the 2-rank of cubic fields by Selmer groups of elliptic curves (Q676229)

Let \(K\) be a non-normal cubic number field; then there exist \(a,b\in\mathbb{Z}\) such that \(K\) is generated by a root of \(f(x)=x^3+ax+b\). Let \(D\) denote the discriminant of \(f\); then \(k=\mathbb{Q}(\sqrt{D})\) is a quadratic number field, and \(N=kK\) is the normal closure of \(K\). Consider the elliptic curve \(E:y^2=f(x)\). The first explicit relation between ranks of elliptic curves over \(\mathbb{Q}\) and \(2\)-class groups of cubic number fields is due to \textit{G. Billing} [ Nova Acta Soc. Sci. Upsal., IV. Ser. 11, 1-165 (1938; Zbl 0018.05491)]. His rather crude bound was refined by various authors, notably by \textit{H. Eisenbeis, G. Frey} and \textit{B. Ommerborn} [Math. Comput. 32, 559-569 (1978; Zbl 0385.12001)] for the special case of pure cubic fields. In the paper under review, their approach is generalized (without proofs - they are given in the author's dissertation) to general non-Galois cubic fields \(K\); the main result bounds the \(2\)-rank of Cl\((K)\) by the cardinalities of two subgroups of the Selmer group of \(E\) corresponding to the multiplication by \(2\). This is achieved as follows: the \(2\)-rank of the relative class group Cl\((N/k)\) is twice the \(2\)-rank of Cl\((K)\) (this is claimed to hold only under an additional assumption on the coefficients \(a, b\), but actually holds in general). The \(2\)-torsion of Cl\((N/k)\) is then related to unramified extensions \(N'/N\) with Gal\((N'/N) \simeq A_4\) (called `admissible') by class field theory, and these extensions are then shown to correspond to certain subgroups of the Selmer group of \(E\). There are, however, two points that need to be clarified. The first is connected with the splitting of primes in unramified admissible extensions \(N'/N\). By class field theory, a prime ideal in \(N\) splits completely in each \(N'/N\) if and only if it has odd order in the relative class group Cl\((N/k)\). The author's claim that primes dividing \(D\) which split completely in \(N/\mathbb{Q}\) also split completely in \(N'/N\) is false, since we can make \(D\) divisible by any given prime by modifying \(a\) and \(b\) appropriately. The second difficulty concerns the `general facts concerning \(2\)-coverings' which are announced on p. 391; the reviewer could not find the statement that `principal homogeneous spaces can be described by quartics even if they are not locally trivial everywhere' in the relevant literature. \textit{J. W. S. Cassels} [J. Lond. Math. Soc. 41, 193-291 (1966; Zbl 0138.27002)] explicitly mentions local triviality as a necessary criterion on p. 235 and p. 269. The author also observes that cubic fields with large \(2\)-class group have infinite \(2\)-class field towers. The `result of Lamprecht' which she cites is, however, the well known inequality of Golod-Shafarevich as improved by Gaschütz and Vinberg. Moreover, she confuses the \(2\)-rank of the unit group, i.e. the rank of \(E_k/E_k^2\), with the \(\mathbb{Z}_2\)-rank of the unit group as in Leopoldt's conjecture. A correct application of the inequality shows that \(K\) has infinite \(2\)-class field tower if the \(2\)-rank of its class group has rank \(\geq 6\), which holds for two of the three fields given as examples. We remark that \textit{E. F. Schaefer} [J. Number Theory 56, 79-114 (1996; Zbl 0859.11034)] constructed a cubic field with \(2\)-class rank equal to \(13\).