Note on 2-rational fields (Q999728)

For a prime number \(p,\) a number field \(K\) is called \(p\)-rational if the Galois group \(G_K =\text{Gal} (M_K/K)\) of the maximal \(p\)-ramified, \(\infty\)-split pro-\(p\)-extension \(M_K\) of \(K,\) is pro-\(p\)-free [\textit{A. Movahhedi} and the reviewer, Sémin. Théor. Nombres, Paris/Fr. 1987-88, Prog. Math. 81, 155--200 (1990; Zbl 0703.11059)]. The field \(K\) is called \(p\)-regular if the \(p\)-regular kernel of \(K,\) i.e. the kernel, in the \(p\)-part of \(K_2 (K),\) of all the Hilbert symbols attached to non-complex places not dividing \(p,\) is trivial \textit{G. Gras} and \textit{J.-F. Jaulent} [Math. Z. 202, No.3, 343--365 (1989; Zbl 0704.11040)]. These two notions coincide if \(K\) contains the maximal real subfield of \({\mathbb Q}(\xi_p),\) thus especially for \(p = 2\). More recently, \textit{J. Jossey} [J. Number Theory 124, No. 1, 42--56 (2007; Zbl 1126.11063)] introduced a notion of \(p\)-rationalty which is incompatible with the above when \(p = 2,\) as soon as \(K\) contains real embeddings (hence, unfortunately, the rational field \({\mathbb Q}\) is not 2-rational in Jossey's sense!). To avoid confusion, let us say that \(K\) is super-2-rational if the Galois group \(G'_K =\text{Gal} (M'_K/K)\) of the maximal 2-ramified pro-2-extension \(M'_K\) of \(K,\) is pro-2-free. If \(K\) is 2-rational, then \(G'_K\) is a priori virtually pro-2-free, thanks to a going up theorem of Movahhedi-Nguyen, Gras-Jaulent [op. cit.], and one of Jossey's main results [thm 2, op. cit.] gives the precise structure of \(G'_K\) as a free pro-2-product. But this was actually already known by Movahhedi-Nguyen [thm. 2--8, op. cit.], using Galois cohomology. Here the authors give an alternative proof (in the spirit of Jossey's, but much simpler) relying on functorial properties of \(p\)-ramification, in particular concerning the generation of the torsion module \(tor_{{\mathbb Z}_2}(G_K^{'ab}).\) In fact, their main purpose is to clarify the notion of 2-rationality and to recall (a useful task) classically known results on 2-ramification.