On the Arf invariant in historical perspective. II. (Q656933)

This is a follow-up paper to [Math. Semesterber. 57, No. 1, 73--102 (2010; Zbl 1201.01015)] in which the authors gave an account in a historical context of \textit{C. Arf}'s seminal work on quadratic forms over fields of characteristic \(2\) [J. Reine Angew. Math. 183, 148--167 (1941; Zbl 0025.01403)]. In that paper, Arf claimed to have proved that if a field \(F\) of characteristic \(2\) has the property that any two nonsplit quaternion algebras over \(F\) share a common quadratic extension of \(F\) (in which case one says that \(F\) is linked), then nonsingular quadratic forms over \(F\) of dimension \(> 4\) are always isotropic. It turned out that this statement is incorrect. For example, Aravire and Jacob showed that the iterated power series field \({\mathbb F}_2((X))((Y))\) is linked but contains nonsingular anisotropic \(8\)-dimensional forms. The correct version of Arf's result is due to \textit{R. Baeza} [Bol. Soc. Bras. Mat. 13, No. 1, 105--114 (1982; Zbl 0573.10014)] who proved that any two nonsplit quaternion algebras over \(F\) share a common inseparable quadratic extension of \(F\) if and only if every nonsingular quadratic form over \(F\) of dimension \(> 4\) is isotropic. The result depends on a subtle distinction between common separable and inseparable quadratic extensions inside two nonsplit quaternion algebras. It was shown by Draxl that a common inseparable quadratic extension implies the existence of a common separable quadratic extension, but the converse is false as first shown by Baeza (a counterexample can also be found in the previous paper by the authors [loc. cit.]). The present paper also gives a self-contained proof of Baeza's theorem.