Deformations of local systems and Eisenstein series (Q2427029)

Let \(G\) be a reductive group and \(X\) a (smooth and complete) curve. Let \(\text{Bun}_G\) denote the moduli stack of \(G\)-torsors on \(X\); let \(D^b(\text{Bun}_G\)) be the appropriately defined derived category of constructible sheaves. Let \(\check G\) denote the Langlands dual group of \(G\). The nature of \(\check G\) depends on the sheaf-theoretic context one works in. Over any ground field, one can work with \(\ell\)-adic sheaves (where \(\ell\) is different from the ground field); in this case \(\check G\) is an algebraic group over \(\overline{\mathbb Q_\ell}\). Let \(E_{\check G}\) be a \(\check G\)-local system on \(X\), thought of as a tensor functor \(V\mapsto V_{E_{\check G}}\) from the category \(\text{Rep}(\check G)\) of finite-dimensional \(\check G\)-representations to that of local systems (= lisse sheaves) on \(X\). In this case one introduces the notion of Hecke eigensheaf, which is an object \(\mathcal S(E_{\check G}) \in D^b(\text{Bun}_G)\), satisfying \(H^V(\mathcal S(E_{\check G} ))\simeq\mathcal S(E_{\check G} )\boxtimes V_{E_{\check G}} ,\) where \(H^V : D^b(\text{Bun}_G) \to D^b(\text{Bun}_G \times X)\) are the Hecke functors, defined for each \(V \in \text{Rep}(\check G)\). A basic (but in general unconfirmed, and perhaps even imprecise) expectation is that for every \(E_{\check G}\) there corresponds a non-zero Hecke eigensheaf \(\mathcal S(E_{\check G})\). This is a weak form of the geometric Langlands conjecture. A stronger form of the conjecture, which only makes sense in the context of D-modules, says that the assignment \(E_{\check G}\mapsto\mathcal S(E_{\check G})\) should work in families. In other words, if \(E_{\check G,Y}\) is a \(Y\)-family of \(\check G\)-local systems, where \(Y\) is a scheme over a ground field \(k\), then to it there should correspond a \(Y\)-family \(\mathcal S(E_{\check G,Y})\). The necessity to use D-modules here, as opposed to any other sheaf-theoretic context, is that it is only in this case that we have a reasonable notion of \(Y\)-families of objects of \(D^b(\mathcal X)\) on a scheme (or stack) \(\mathcal X\). The strongest (and wildest) form of the geometric Langlands conjecture says that the above assignment should give rise to an equivalence between the category \(D^b(\text{Bun}_G)\) and the appropriately defined derived category of quasi-coherent sheaves on the stack \(\text{LocSys}_{\check G}\), classifying \(\check G\)-local systems on \(X\). Let us consider the following intermediate case. Let \(E_{\check G}\) be a fixed local system, and let \(E_{\check G,Y}\) be its formal deformation. Suppose we have found \(\mathcal S(E_{\check G})\) which is a Hecke eigensheaf with respect to \(E_{\check G}\). Can we extend \(\mathcal S(E_{\check G})\) to a \(Y\)-family \(\mathcal S(E_{\check G,Y} )\) of eigensheaves? This is the question that V. Drinfeld asked on several occasions. Let us take \(Y\) to be \(\text{Def}(E_{\check G})\) -- the base of the universal deformation of \(E_{\check G}\) as a \(\check G\)-local system. In this case, \(Y\) is indeed quasi-isomorphic to the standard complex of a DG-algebra canonically attached to \(E_{\check G}\). Unfortunately, even the assignment \(E_{\check G}\to \mathcal S(E_{\check G})\) has been constructed only in few cases. One such case is when \(G = \text{GL}_n\) and \(E_{\check G} = E_n\) is an \(n\)-dimensional irreducible local system. The case that the authors study, is, in some sense, the opposite one. The authors take \(\check G\) to be arbitrary, but \(E_{\check G}\) is assumed ``maximally reducible'', i.e. \(E_{\check G}\) is induced from a local system \(E_{\check T}\) with respect to the Cartan group \({\check T} \subset {\check G}\). In this case the corresponding Hecke eigensheaf was earlier constructed by the authors, under the name ``geometric Eisenstein series''. Denote it by \(\overline{\text{Eis}}(E_{\check T})\). Along with the geometric Eisenstein series \(\overline{\text{Eis}}(E_{\check T})\) there exists a more naive object called ``classical'' Eisenstein series and denoted by \({\text{Eis}}_!(E_{\check T})\). When we work over the finite ground field and \(\ell\)-adic sheaves, \({\text{Eis}}_!(E_{\check T})\) goes over under the \textit{faisceaux-fonctions} correspondence to the usual Eisenstein series as defined in the theory of automorphic forms. Drinfeld's conjecture is that the family \(\mathcal S(E_{\check G,\text{Def}_{\check B} (E_{\check T})})\) is nothing but (a certain completion of) \({\text{Eis}}_!(E_{\check T})\). This statement has an ideological significance also for the classical (i.e. function theoretic vs. sheaf-theoretic) Langlands correspondence: \textit{The classical Eisenstein series correspond not to homomorphisms \(\mathrm{Galois} \longrightarrow \check G\) that factor through \(\check T\), but rather to the universal family of homomorphisms \(\mathrm{Galois}\longrightarrow {\check B}\) with a fixed composition \(\mathrm{Galois}\longrightarrow {\check B} \longrightarrow\!\!\!\to \check T\).} The present paper is devoted to the proof of Drinfeld's conjecture, under a certain simplifying hypothesis on \(E_{\check T}\). Namely, the authors assume that \(E_{\check T}\) is regular, i.e. that it is as non-degenerate as possible. This means that for every co-root \(\check\alpha\) of \(G\), which is the same as a root of \(\check G\), the induced 1-dimensional local system \(\check\alpha(E_{\check T})\) is non-trivial. This regularity assumption is equivalent to requiring that the DG formal scheme Def\(_ {\check B} (E_{\check T})\) be an ``honest'' scheme. Moreover, the authors show that in this case both versions of Eisenstein series, namely, \(\overline{\text{Eis}}(E_{\check T})\) and \({\text{Eis}}_!(E_{\check T})\) are perverse sheaves. This fact and the simplified nature of Def\(_{\check B} (E_{\check T})\) makes life significantly easier, since one can avoid a lot of complications of homotopy-theoretic nature. Proving the above conjecture amounts to the following: (i) Exhibiting the action of the commutative algebra \(\mathcal O_{\text{Def}_ {\check B} (E_{\check T} )}\) of functions on Def\(_{\check B} (E_{\check T})\) on \({\text{Eis}}_!(E_{\check T})\). (ii) Establishing an isomorphism \(\mathbb C\overset {L} {\underset \mathcal O_{\text{Def}_ {\check B}(E_{\check T})} \otimes} {\text{Eis}}_!(E_{\check T})\simeq \overline{\text{Eis}}(E_{\check T})\). (iii) Verifying the Hecke property \(\text{H}^V (\text{Eis}_!(E_{\check T}))\simeq \text{Eis}_!(\check T) \underset{\mathcal O_{\text{Def}_ {\check B}(E_{\check T})}}\boxtimes V_{E_{\check G,\text{Def}_ {\check B}(E_{\check T})}}\), where \(V_{E_{\check G,\text{Def}_ {\check B}(E_{\check T})}}\) is the canonical family of \(\check G\)-local systems over \(\text{Def}_{\check B} (E_{\check T})\). As is to be expected, the verification of these properties is a nice simple exercise when \(G = \text{GL}_2\), that the authors perform, but not altogether trivial for other groups.