Essential dimension of the spin groups in characteristic 2 (Q2416396)

Let \(G\) be an affine group scheme of finite type over a field \(k\). The isomorphism classes of \(G\)-torsors over \(k\) can be identified with the cohomology set \(H^1(k,G)\) in the fppf topology (resp. étale topology if \(G\) is smooth over \(k\)). \textit{Z. Reichstein} [in: Proceedings of the international congress of mathematicians (ICM 2010), Hyderabad, India, August 19--27, 2010. Vol. II: Invited lectures. Hackensack, NJ: World Scientific; New Delhi: Hindustan Book Agency. 162--188 (2011; Zbl 1232.14030)] defined the essential dimension \(\hbox{ed}(G)\) of \(G\) as the least nonnegative integer \(r\) such that for every \(G\)-torsor \(\xi\) over an extension field \(E\) of \(k\), there exists an intermediate field \(k\subseteq F\subseteq E\) such that \(\xi\) is in fact isomorphic to some \(G\)-torsor defined over \(F\) and extended to \(E\). In a certain sense, the essential dimension measures how many parameters are needed to describe all \(G\)-torsors over all fields. There is a vast literature on computing or bounding essential dimensions for various \(G\). The present paper focuses on the spin group \(\hbox{Spin}(n)\). In a series of papers by various authors, the essential dimension has been determined for all \(n\) in the case where the characteristic of \(k\) is different from \(2\). In the present paper, the essential dimension of \(\hbox{Spin}(n)\) in characteristic \(2\) is computed for all \(n\leq 10\) and all \(n\geq 15\). It turns out that for all these \(n\), the essential dimension in characteristic \(2\) is equal to that in characteristic not \(2\). This is in sharp contrast to the case of (special) orthogonal groups, where the essential dimension in characteristic \(2\) is smaller than in characteristic not \(2\). The determination of the upper bound for \(\hbox{Spin}(n)\) in characteristic \(2\) for \(n\geq 15\) invokes results by \textit{S. Garibaldi} and \textit{R. M. Guralnick} [Compos. Math. 153, No. 3, 535--556 (2017; Zbl 1498.11113)] in the case where \(4\) does not divide \(n\) or if \(n=16\), and the remaining cases use methods developed by \textit{V. Chernousov} and \textit{A. Merkurjev} [Algebra Number Theory 8, No. 2, 457--472 (2014; Zbl 1312.11024)]. For the lower bound, in characteristic not \(2\) one uses the fact that inside \(\hbox{Spin}(2r+1)\), there is a natural finite subgroup that is a central extension of \((\mathbb{Z}/2)^{2r}\) by \(\mathbb{Z}/2\). The main modification in the characteristic \(2\) case is that this finite subgroup has to be replaced by a finite group scheme that is a central extension of \((\mathbb{Z}/2)^r\times \mu_2^r\) by \(\mu_2\) (the group scheme of square roots of unity), which eventually leads to the same lower bounds in all characteristics. The case by case determination of \(\hbox{Spin}(n)\) for \(n\leq 10\) uses cohomological invariants and explicit computations involving Pfister forms. The results thus read as follows: if \(n\geq 15\) and if \(v_2(n)\) denotes the \(2\)-adic value of \(n\), then the essential dimension of \(\hbox{Spin}(n)\) is \(2^{n-1}-n(n-1)/2\) if \(v_2(n)=0\), \(2^{(n-2)/2}-n(n-1)/2\) if \(v_2(n)=1\), and \(2^{(n-2)/2}+2^{v_2(n)}-n(n-1)/2\) if \(v_2(n)\geq 2\). For \(n\leq 10\), the essential dimension is \(0\) if \(n\leq 6\), \(4\) if \(n\in\{ 7,10\}\), and \(5\) if \(n\in\{ 8,9\}\).