No label defined (Q4786358)

Let \(k\) be a number field, \(S\) a finite set of places of \(k\), and \(p\) a prime number. A fundamental object of interest is the maximal \(S\)-ramified \(p\)-extension \(k_S(p)\) of \(k\), its Galois group \(G_S(p)\) over \(k\), and the cohomology theory of the latter group. It is always assumed in this paper that \(S\) contains \(S_p\), the set of places dividing \(p\). If \(p\) is odd, or if \(k\) is totally imaginary, then it is well known that \(G_S(p)\) has cohomological dimension at most 2. The case ``\(p=2\), \(k\) has at least one real place'' was open so far and is of obvious interest. If in this case \(S\) contains all real places, then \(k_S(2)\) is totally complex (e.g. since it contains \(\sqrt{-1}\)), and hence every decomposition group of a real prime in \(k\) provides an involution in \(G_S(2)\), which therefore has infinite cohomological dimension. The first main result of the paper under review is the following (Theorem 1): This bad behaviour persists, provided \(S\) contains at least one real place. If \(S\) contains no real place, then it is shown that we are back in the nice situation that \(G_S(2)\) has coh.dim. at most 2. Moreover, the ``truncated'' Euler characteristic of \(G_S(2)\) is determined (it is defined almost as the usual Euler characteristic, one just stops at the second cohomology group). The author's approach to this result is the following: Let \(T\) be the superset of \(S\) which arises by adjoining all real places to \(S\). One wants to show that \(G_S(2)\) inherits the (bad) cohomological behaviour from \(G_T(2)\). More specifically, Theorem 2 shows that the kernel \(K\) of the canonical epimorphism \(G_T(2) \to G_S(2)\) is isomorphic to a free pro-2-product of inertia groups. (One needs the general machinery of free products of ``bundles of groups'' over a profinite index set. Moreover the theorem is stated and proved for all primes \(p\).) Theorem 1 is then deduced by a spectral sequence argument; indeed one has full control over \(K\), and the \(E_2\) page of the relevant spectral sequence can only have nonzero terms on the boundary of the first quadrant. Theorem 3 considers the case where \(S\) contains no real place and gives a list of criteria for the cohomological dimension of \(G_S(2)\) to be 1. (Recall from above that it is at most 2.) More precisely, the conjunction of the following three conditions is equivalent to coh.dim.\((G_S(p))=1\): (1) The prime 2 does not split in \(k\); (2) \(S\) contains only the dyadic place, and some complex places; (3) the narrow \(S\)-class group of \(k\) has odd order. (The latter class group somehow combines ramification at 2 and at infinity, which may explain the title of the paper). The paper contains several other interesting results, including a connection with modified-etale cohomology in the sense of Zink. It concludes with a series of remarks, one of which presents a rather precise Kummer duality statement in Iwasawa theory for \(p=2\), a notoriously delicate issue. This nice paper fills some annoying gaps in our knowledge of Galois cohomology.