Étale cohomology and a generalisation of Hilbert's Theorem 132 (Q5957305)

Hilbert's Theorem 132 states that for squarefree \(n\), \(\zeta_n\) generates a normal integral basis for the abelian extension \(L/K\) with \(K={\mathbb Q}\) and \(L={\mathbb Q}(\zeta_n)\). Chinburg's second conjecture is in a sense a generalisation of this; at least it contains as a special case the statement that \(O_L\) is \(O_K[G]\)-free, if \(L/K\) is tame \(G\)-abelian, without however (and this is an important point to make) specifying a generator. In general, Chinburg's conjecture states that for \(L/K\) \(G\)-abelian, a certain invariant concocted by means of projective \({\mathbb Z}[G]\)-modules, which in turn are in some sense obtained from canonical classes, is trivial. This was essentially proved by the reviewer [J. Reine Angew. Math. 479, 1-37 (1996; Zbl 0856.11051)] for \(L\) absolutely abelian. In the present paper, a conjecture is presented which lifts Chinburg's conjecture to a relative \(K\)-group, that is: it postulates that a certain invertible \({\mathbb Z}[G]\)-module, \(J\) say, is generated by a specific element \(\xi_{L/K}\). Now Hilbert's Theorem 132 is indeed a special case of this since for \(L={\mathbb Q}(\zeta_n)\) as above, \(\xi_{L/K}\) does essentially work out as \(\zeta_n\) and \(J=O_L\). The authors prove this ``lifted'' conjecture for \(L\) absolutely abelian, with the same small restriction (2 unramified) as in the reviewer's work, resuming some of the reviewer's techniques but obviously ``getting more out of them''. They then proceed to prove a weaker version of their conjecture (the integral group ring of \(G\) is replaced by its maximal order) for general abelian extensions. The construction of the invertible module \(J\) is still intricate, but not more so than the constructions in the unlifted Chinburg conjecture. \(J\) is defined, in the spirit of Kim's formula, as the tensor product of a global term \(\text{Det}_{\mathbb Z}[G](\mathcal L)\) with local terms \(I(v,{\mathcal L})\), one for each wildly ramified prime \(v\). Here \(\mathcal L\) is a chosen projective \({\mathbb Z}[G]\)-lattice in \(O_L\) with some further properties; but the choice of \(\mathcal L\) cancels out, so that the construction of \(J\) enjoys a naturality which was lacking in the unlifted approach. The local terms involve local canonical classes (these have to come in somewhere), but in a slightly different guise: one obtains four-term exact sequences whose two middle terms are cohomologically trivial from the framework of derived categories, and the cohomology of the sheaf \({\mathbf G_m}\). That this approach gives the right thing was previously proved by \textit{D. Burns} and \textit{M. Flach} [Am. J. Math. 120, 1343-1397 (1998; Zbl 0929.11050)] and ultimately seems to go back to Serre. This nice article is highly recommended reading, since it succeeds in proving substantial results in modern language, without assuming the entire framework of equivariant Tamagawa numbers.