Cohomological invariants for \(G\)-Galois algebras and self-dual normal bases (Q507689): Difference between revisions

Let \(k\) be a field of characteristic not \(2\), let \(G\) be a finite group and let \(L\) be an étale \(k\)-algebra, i.e., \(L\) is a finite direct product of finite separable extensions of \(k\). Then \(L\) is called a \(G\)-Galois algebra if \(G\) acts on \(L\) such that \(L\cong k[G]\) as \(k[G]\)-modules. In this case, the trace form \(q_L(x,y)=\mathrm{Tr}_{L/k}(xy)\) is \(G\)-invariant, i.e., \(q_L(gx,gy)=q_L(xy)\) for any \(g\in G\). Let \((gx)_{g\in G}\) be a normal basis for a suitable \(x\in L\) (such bases always exist). Then it is called self-dual if the Gram matrix of \(q_L\) with respect to this basis is the identity matrix. There is a considerable literature on the question when a \(G\)-Galois algebra has a self-dual normal basis. If \(k\) is a global field, then the Hasse principle holds in the sense that there is a self-dual normal basis over \(k\) if and only if there is a self-dual normal basis over every localization of \(k\) as was shown by the authors et al. [Izv. Math. 77, No. 3, 437--460 (2013; Zbl 1368.11030); translation from Izv. Ross. Akad. Nauk, Ser. Mat. 77, No. 3, 5--28 (2013)]. In the present paper, the authors complete the picture by providing necessary and sufficient criteria for the existence of self-dual normal bases over local fields. The first author and \textit{J.-P. Serre} [Am. J. Math. 116, No. 1, 1--64 (1994; Zbl 0804.12004)] have constructed cohomological invariants in \(H^1(k,\mathbb{Z}/2\mathbb{Z})\) attached to \(L\) and they have shown that a necessary condition for the existence of self-dual normal bases is the vanishing of these invariants. For local fields, this is not sufficent in general. Intuitively, this is not surprising as the local non-Archimedean fields one has to deal with have \(2\)-cohomological dimension \(2\). Thus, a crucial ingredient is the construction of a new set of invariants living in \(H^2(k,\mathbb{Z}/2\mathbb{Z})\). It turns out that the vanishing of certain of these new \(H^2\)-invariants is necessary for the existence of self-dual normal bases. For local fields, one can say more. The main theorem of the paper states that if \(k\) is a local non-archimedean field, then a \(G\)-Galois algebra \(L\) over \(k\) possesses a self-dual normal basis if and only if all the \(H^1\)-conditions from above hold and certain of these newly constructed \(H^2\)-invariants vanish (Theorem 7.1). The construction of these invariants is roughly as follows. Consider the semisimple \(k\)-algebra \(k[G]/J\), where \(J\) denotes the Jacobson radical of \(k[G]\). This is a product of simple algebras, and we consider those simple factors \(A\) that are stable under the involution induced by \(g\mapsto g^{-1}\) on \(k[G]/J\). The restriction of this involution to \(A\) will be denoted by \(\sigma_A\). Now the authors define a cohomology class in \(H^2(k,\mathbb{Z}/2\mathbb{Z})\) attached to each such \(A\). These are the \(H^2\)-invariants one has to consider in the main theorem, but only for those those \(A\) where the fixed field under \(\sigma_A\) inside the center of \(A\) is of odd degree over \(k\), and in addition, \(A\) is split if \(\sigma_A\) is an orthogonal involution.