On maximal tori of algebraic groups of type \(G_2\) (Q903157)

The study of maximal tori in semisimple algebraic groups is an interesting classical topic which has gained renewed importance in the last decade or so with various problems like the genus problem and work of Gopal Prasad, Andrei Rapinchuk and others. In the paper under review, the authors consider the exceptional groups of type \(G_2\). Such a group \(G\) over a field \(k\) is the automorphism group of a (uniquely determined) octonion algebra over \(k\). Associated to a \(k\)-embedding \(i\) of a rank \(2\) torus \(T\) in \(G\), there is an action of the absolute Galois group on the root system of \(G_{k_{\mathrm{sep}}}\) with respect to \(i(T_{k_{\mathrm{sep}}})\) and hence, an element of \(H^1(k,W_0)\), where \(W_0 \cong \mathbb{Z}/2 \mathbb{Z} \times S_3\) is the Weyl group of the split group of type \(G_2\). By Galois descent, a cohomology class is essentially a pair \((k',l)\) where \(k'\) is a quadratic étale \(k\)-algebra and \(l\) is a cubic étale \(k\)-algebra. Therefore, the natural question is which pairs correspond to a \(k\)-embedding of a maximal torus as above. The authors answer this embedding problem. They apply (a refined version proved by them of) \textit{R. Steinberg}'s theorem on rational conjugacy classes to Galois cohomology [Publ. Math., Inst. Hautes Étud. Sci. 25, 281--312 (1965; Zbl 0136.30002)]. As the Galois cohomological problem can be reformulated in terms of existence of \(k\)-rational point of an associated homogeneous space, the authors are able to address the following general question formulated by \textit{B. Totaro} [Duke Math. J. 121, No. 3, 425--455 (2004; Zbl 1048.11031)]: For a smooth connected affine \(k\)-group \(G\) over the field \(k\) and a homogeneous \(G\)-variety \(Y\) such that \(Y\) has a zero-cycle of degree \(d>0\), does \(Y\) necessarily have a closed étale point of degree dividing \(d\)? There are cases where this has been answered positively but there are also counter-examples known. In this paper, the authors construct -- when \(k\) is a large enough field like \(\mathbb{Q}(t)\) -- a \(G\)-homogeneous space with a quadratic and a cubic point, but no \(k\)-rational point. Thus, they have a class of counter-examples with \(d=1\) which are simpler than examples given by \textit{M. Florence} [Int. Math. Res. Not. 2004, No. 54, 2897--2914 (2004; Zbl 1092.14061)] and \textit{R. Parimala} [Asian J. Math. 9, No. 2, 251--256 (2005; Zbl 1095.14018)]. The geometric stabilizers of the homogeneous spaces in this paper are maximal \(K\)-tori while they are finite, noncommutative in Florence's work [loc. cit.] and parabolic subgroups in Parimala's work [loc. cit.]. The authors have used their embedding theorems in later work on the genus problem.