An elementary approach to the Serre-Rost invariant of Albert algebras (Q2365502)

In the theory of Jordan algebras, a key role is played by the so called Albert algebras, which are 27-dimensional simple and exceptional (i.e., not special) algebras. They constitute the only class of finite-dimensional simple Jordan algebras which are not totally classified. \textit{J. P. Serre} [Résumé des cours de l'année 1990-91, Collège de France] suggested the existence for any Albert algebra \(J\), over a field \(k\) of characteristic \(\neq 2,3\), of three cohomological invariants \(f_3(J)\in H^3(k,\mathbb{Z}/2\mathbb{Z})\), \(f_5(J)\in H^5(k,\mathbb{Z}/2\mathbb{Z})\) and \(g_3(J)\in H^3(k,\mathbb{Z}/3\mathbb{Z})\), which hopefully would classify Albert algebras. The existence of these invariants was proved by \textit{M. Rost} [C. R. Acad. Sci., Paris, Sér. I 313, 823-827 (1991; Zbl 0756.17014)] and [``A descent property for Pfister forms'', Preprint (1991)]. The paper under review deals with the invariant modulo 3, \(g_3(J)\), which characterizes Albert division algebras. The existence of \(g_3(J)\) is proven over arbitrary fields of characteristic \(\neq 3\) in an elementary way. The approach used relies on showing that it is enough to define \(g_3(J)\) in case \(J\) is obtained by the first Tits construction, which uses a central simple associative algebra of degree 3 and a nonzero element \(a\in k^\times\). The crucial point is to show that the definition given of \(g_3(J)\) in terms of \(D\) and \(a\) only depends on the Albert algebra. The paper is very nicely written and motivated, with the first two sections explaining the results on Galois cohomology and Albert algebras needed in the sequel and a last section dealing with open questions and hints to their solutions. It must be noticed that the extension of the results of this paper to fields of characteristic three has already been achieved by the same authors (unpublished).