Epipelagic representations and rigid local systems (Q739814)

This paper considers the epipelagic representations of [\textit{M. Reeder} and \textit{J. K. Yu}, J. Am. Math. Soc. 27, No. 2, 437--477 (2014; Zbl 1284.22011)] from a global geometric perspective. It is a sequel to [\textit{J. Heinloth} et al., Ann. Math. (2) 177, No. 1, 241--310 (2013; Zbl 1272.14012)] and [\textit{Z. Yun}, Invent. Math. 196, No. 2, 267--337 (2014; Zbl 1374.14013)]. Let \(k\) be a finite field and \(\mathbb{G}\) be a split reductive \(k\)-group. Let \(X = \mathbb{P}^1_k\). Fix \(e\) and consider the \(\tilde{X} = \mathbb{P}^1_k \to X\) given by mapping the coordinate function \(t\) to \(t^e\). It is totally ramified at \(0, \infty \in X(k)\) and restricts to a \(\mu_e\)-torsor \(\tilde{X}^\circ \to X^\circ\) off \(0, \infty\). Fix an embedding of \(\mu_e\) into the group of pinned automorphisms of \(\mathbb{G}\), and denote the image of \(1\) by \(\sigma\). This datum gives rise to a pinned quasisplit reductive group scheme \(G\) over \(X^\circ\) that splits over \(\tilde{X}^\circ\). The \(L\)-group \({}^L G = \hat{G} \rtimes \mu_e\) is also defined. Write \(F = k(t)\) and denote by \(K = F_\infty\) the completion at \(\infty\). Let \(\mathbf{P} = \mathbf{P}_\infty \subset G(K)\) be a standard parahoric. The first two terms of the Moy--Prasad filtration are \(\mathbf{P} \supset \mathbf{P}^+ \supset \mathbf{P}^{++}\), with subquotients \(\mathbf{L}_P\) and \(V_{\mathbf{P}}\) respectively; note that \(V_{\mathbf{P}}\) is actually a \(k\)-vector space. The group \(\mathbb{G}_m^{\mathrm{rot}} = \mathbb{G}_m\) acts on \(\tilde{X}^\circ\) and on its completion at \(\infty\) by scaling \(t^{-1/e}\). It induces an action of \(L_{\mathbf{P}} \rtimes \mathbb{G}^{\mathrm{rot}}_m\) on \(V_{\mathbf{P}}\). Given an \(\ell\)-adic additive character \(\psi\) of \(k\), together with a stable linear functional \(\phi \in V_{\mathbf{P}}^{*, \mathrm{st}}(k)\), Reeder--Yu constructs the compactly induced representation \(\mathrm{c-Ind}_{\mathbf{P}^+}^{G(K)}(\psi \circ \phi)\): it is a finite direct sum of supercuspidals, which are called the epipelagic supercuspidals associated to \(\phi\) and \(\psi\). This inscribes in a global framework by taking \(\mathbf{P}_0 \subset G(F_0)\) to be the parahoric at \(0\) of the same type but opposite to \(\mathbf{P}_\infty\). Let \(\tilde{\mathbf{P}}_0\) be its normalizer and let \(\tilde{L}_{\mathbf{P}}\) be \(\tilde{\mathbf{P}}/\mathbf{P}_0^+\). Besides the data \(\phi, \psi\) at \(\infty\), we also choose a character \(\chi\) of \(\tilde{L}_{\mathbf{P}}(k)/L^{\mathrm{sc}}_{\mathbf{P}}(k)\) at \(0\). It is shown in Proposition 2.11 that there is a unique automorphic representation \(\pi = \pi(\chi, \phi)\) of \(G(\mathbb{A}_F)\) such that \(\pi\) is unramified over \(X^\circ\), has a nonzero \((\tilde{\mathbf{P}}_0, \chi)\)-eigenvector at \(0\), and is an epipelagic supercuspidal associated to \(\psi \circ \phi\) at \(\infty\). Moreover, \(\pi\) is cuspidal and occurs with multiplicity one. Next, consider the stack \(\mathrm{Bun} := \mathrm{Bun}_G(\mathbf{P}^+_0, \mathbf{P}^{++}_\infty)\) classifying \(G\)-torsors over \(X\) with level structures at \(0, \infty\). Then \(\mathrm{Bun} \to \mathrm{Bun}_G(\mathbf{P}_0, \mathbf{P}_\infty^+)\) is an \(L_{\mathbf{P}} \ltimes V_{\mathbf{P}}\)-torsor, and there is an open immersion \(j: \tilde{L}_{\mathbf{P}} \times V_{\mathbf{P}} \hookrightarrow \mathrm{Bun}\). We geometrize \(\chi\) to a rank one character sheaf \(\mathcal{K}\) on \(\tilde{L}_{\mathbf{P}}\) and \(\psi\) to an Artin--Schreier sheaf \(\mathrm{AS}_S\) on \(V_{\mathbf{P}} \times S\), where \(S\) is any scheme over \(V_{\mathbf{P}}^*\). One uses \(j\), \(\mathcal{K}\) and \(\mathrm{AS}_S\) to construct \((\mathcal{K}, \mathrm{AS}_S)\)-equivariant sheaves \(A(\mathcal{K}, S)\) on \(\mathrm{Bun} \times S\). Assume \(S \to V_{\mathbf{P}}^{*, \mathrm{st}}\) is of finite type. In \S 3, it is shown that \(A(K, S)\) admits the structure of an \(S\)-family of Hecke eigensheaves; the corresponding eigen-\({}^L G\)-local system over \(X^\circ\) is denoted by \(\mathrm{Kl}_{{}^L G, \mathbf{P}}(\mathcal{K}, S)\). Moreover, there exists a \(\hat{G}\)-local system \(\mathrm{Kl}_{\hat{G}, \mathbf{P}}(\mathcal{K})\) over \(V_{\mathbf{P}}^{*, \mathrm{st}}\) which pulls back to \(\mathrm{Kl}_{{}^L G, \mathbf{P}}(\mathcal{K}, \phi)\), via \(\tilde{X}^\circ \simeq \mathbb{G}_m^{\mathrm{rot}} \to V_{\mathbf{P}}^{*, \mathrm{st}}\) for any given \(\phi \in V_{\mathbf{P}}^{*,\mathrm{st}}(k)\), where \(\mathbb{G}_m^{\mathrm{rot}} \to V_{\mathbf{P}}^{*, \mathrm{st}}\) is simply the action on \(\phi\). Furthermore, this realizes \(\text{Kl}_{{}^L G, \mathbf{P}}(\mathcal{K}, \phi)\) as the global \(L\)-parameter of a cuspidal representation of \(G(\mathbb{A}_F)\) characterized as before, with a geometric origin. The author proceeds to describe the local monodromy of \(\text{Kl}_{{}^L G, \mathbf{P}}(\mathcal{K}, \phi)\) at the tame point \(0\). When \(G\) splits and \(\mathcal{K}\) is trivial, this is described in Theorem 4.5 in terms of Lusztig's theory of twos-sided cells in \(W_{\mathrm{aff}}\), using the unipotent class in \(\hat{G}^{\sigma, 0}\) corresponding to the longest element in \(W^{L_{\mathbf{P}}}\). The cohomological rigidity of the local systems is also obtained in Proposition 5.2, under the same assumptions plus that \(\text{char}(k) \gg_{\mathbb{G}} 0\) and that we have the expected monodromy at \(\infty\). The last three sections contain the computations of \(\text{Kl}_{{}^L G, \mathbf{P}}(\mathcal{K}, \phi)\) for classical groups. This is mainly based on Proposition 3.12 which expresses those local systems via Fourier--Deligne transform on \(D^b_c(V_{\mathbf{P}})\).