The Kaplansky radical of a quadratic field extension (Q2443269)

Let \(K\) be a field of characteristic not \(2\), and let \(\sum K^{\times 2}\) denote the set of nonzero sums of squares in \(K\). For any \(a\in K^{\times}\), let \(D_K(\langle 1,a\rangle)\) denote the multiplicative group of nonzero elements in \(K\) represented by the quadratic form \(\langle 1,a\rangle=X^2+aY^2\). The radical (or Kaplansky radical) \(R(K)\) is defined to be \(R(K)=\bigcap_{a\in K^{\times}}D_K(\langle 1,a\rangle)\). One has \(K^{\times 2}\subseteq R(K) \subseteq D_K(\langle 1,1\rangle)\subseteq \sum K^{\times 2}\). \(K\) is said to be radical free if \(K^{\times 2}= R(K)\) and quasi-Pythagorean if \(R(K)= D_K(\langle 1,1\rangle)\). The authors provide criteria for radical freeness as well as for \(R(K)=D_K(\langle 1,1\rangle)\), which turns out to be equivalent to \(R(K)=\sum K^{\times 2}\). They show that strict inclusion between \(K^{\times 2}\) and \(\sum K^{\times 2}\) requires at least \(8\) square classes, and that such examples with \(8\) square classes do exist. They also study \(R(K)\) in the presence of a set of \(\mathbb{Z}\)-valuations on \(K\) with nondyadic residue fields that are not quadratically closed. They show that \(K\) is radical free provided \(K\) satisfies the local-global principle for being a square. As a consequence, they show that if \(K/k\) is purely transcendental of transcendence degree \(1\) and \(k(\sqrt{-1})\) is not quadratically closed, or if \(K=k(\!(X_1,\ldots,X_n)\!)\) with \(n\geq 2\) or \(k\) not quadratically closed, then \(K\) is radical free. Finally, the authors study the radical \(R(L)\) of a quadratic extension \(L/K\). One obtains a complex \(K^{\times}/R(K)\to L^{\times}/R(L)\to K^{\times}/R(K)\) induced by injection and the norm map. \textit{D. Kijima} and \textit{M. Nishi} [Hiroshima Math. J. 11, 443--456 (1981; Zbl 0467.10016)] asked if this complex is exact and they proved it in [\textit{T. Iwakami} et al., Hiroshima Math. J. 15, 81--88 (1985; Zbl 0571.10018)], in the case where \(K\) or, equivalently, \(L\) is quasi-Pythagorean. The authors give a new proof of the latter result and they provide examples that show that the answer to Kijima and Nishi's question is negative in general.