Connected components of closed affine Deligne-Lusztig varieties (Q5920439)

Let \(k\) be a finite field, \(\overline{k}\) an algebraic closure of \(k\), \(L = \overline{k}((t))\) the field of Laurent series, and \(\sigma\) the Frobenius on \(\overline{k}\) and \(L\). Let \(G\) be a split connected reductive group over \(k\), let \(A\) be a split maximal torus, and \(B\) a Borel subgroup containing \(A\). The affine Deligne-Lusztig variety associated with an element \(b\in G(L)\) and a dominant coweight \(\mu \in X_*(A)\) is by definition \[ X^G_\mu(b) = X_\mu(b) = \{ g \in G(L)/K;\;g^{-1}b\sigma(g) \in K t^\mu K \}. \] Here \(t^\mu\) denotes the image of \(t\) under the homomorphism \(L^\times = \mathbb G_m(L) \rightarrow A(L) \subset G(L)\). This is a locally closed subset of the affine Grassmannian \(G(L)/K\). The closed affine Deligne-Lusztig variety is defined as the union \[ X_{\leq \mu}(b) = \bigcup_{\lambda \leq \mu} X_\lambda(b). \] It is a closed subset of \(G(L)/K\). We consider both of these sets as subschemes (with the reduced scheme structure). They are schemes locally of finite type over \(\overline{k}\). These constructions are obviously analogous to the construction of (usual) Deligne-Lusztig varieties. In the current situation, the root system is replaced by the affine root system, and the (partial) flag variety is replaced by a (partial) affine flag variety, specifically by the affine Grassmannian. We can identify the set of connected components of \(G(L)/K\) with the algebraic fundamental group \(\pi_1(G)\) of \(G\), and denote by \(\kappa_G \colon G(L)/K \rightarrow \pi_1(G)\) the induced map. In the paper under review, the sets of connected components of these schemes are studied. To simplify the statements, let us assume throughout this review that the group \(G\) is simple. First, there is the Hodge-Newton decomposition (which was essentially already obtained by \textit{R. E. Kottwitz} [Int. Math. Res. Not. 2003, No. 26, 1433--1447 (2003; Zbl 1074.14016)]): Take a standard parabolic subgroup \(P \subseteq G\) with Levi subgroup \(A\subset M \subset P\) which contains the centralizer of the Newton vector of \(b\). We may then assume that \(b\in M\), and in this situation, if \(\kappa_M(b)=\mu\), then the natural inclusion \(X_\mu^M(b) \rightarrow X_\mu^G(b)\) is an isomorphism. Now assume that \(G\) does not contain a proper standard parabolic subgroup to which we can apply the Hodge-Newton decomposition (the pair \((b,\mu)\) being fixed such that \(X_\mu(b)\neq\emptyset\)). Then either \(b\) is \(\sigma\)-conjugate to the translation element \(t^\mu\) and \(t\mu\) is central, in which case \(X_\mu(b)\cong X_{\leq \mu}(b) \cong G(k((t)))/G(k[[t]])\) are discrete, or \(\kappa_G\) induces a bijection \(\pi_0(X_{\leq \mu}(b)) \cong \pi_1(G)\). Finally, those cases where \(X_\mu(b)\) has dimension \(0\) are characterized. It is also shown that the set of connected components of \(X_\mu(b)\) and \(X_{\leq\mu}(b)\) are different in general.