Geometric constant term functor(s) (Q730318)

When studying the space of automorphic functions on an adéle group \(G\), one defines a pair of adjoint functors, suggestively called the ``Eisenstein series'' \(\text{Eis}\) and the ``constant term'' \(\text{CT}\), which connect it to similar spaces on \(G\)'s Levi subgroups. The paper introduces analogs of these functors in algebraic geometry. Consider a smooth connected projective curve \(X\) and the moduli stack \(\text{Bun}_G\) of principal \(G\)-bundles over \(X\) for a reductive \(G\). The space of automorphic functions corresponds to the DG category of (not necessarily holonomic) \(D\)-modules on \(\text{Bun}_G\). If \(P\) is a parabolic subgroup of \(G\), and \(M\) is the Levi quotient of \(G\) by \(P\), then we have a diagram of stacks \(\text{Bun}_M\overleftarrow{q}\text{Bun}_G \overrightarrow{p}\text{Bun}_P\), which restricts to each connected component of \(\text{Bun}_M\), these are labeled by \(\mu\in\pi_1(M)\). This defines a pair of functors \(\text{Eis}_*^\mu:=p_*\circ q^!\) and \(\text{CT}_*^\mu:= q_*\circ p^!\). It is not obvious that their left adjoints, \(\text{Eis}_!^\mu\) and \(\text{CT}_!^\mu\), respectively, are well-defined because \(p\), while quasi-compact and representable, is neither smooth nor proper, and \(q\), while smooth and safe, is not representable. The authors first show that \(p_!\) is well-defined on the essential image of \(q^*\), so \(\text{Eis}_!^\mu:= p_!\circ q^*\) is well-defined (it turns out to be closely related to the geometric Langlands conjecture). However, the same does not hold for \(p^*\), and one can not define \(\text{CT}_!^\mu\) analogously. The main result of the paper is that nonetheless the naïve composition can be made sense of, and moreover \(\text{CT}_!^\mu\) is isomorphic to \(\text{CT}_*^{\mu,-}\), where the \(-\) indicates taking the \(\text{CT}_*^\mu\) as above, but for the opposite parabolic subgroup \(P^-\). The result can be interpreted as deriving a non-standard functional equation. The authors give two proofs of the main result. The first one imitates the authors' earlier reproof of Braden's theorem on hyperbolic restrictions, and directly establishes the \(\text{CT}_*^{\mu,-}\), \(\text{Eis}_*^\mu\) adjunction by specifying the unit and the counit morphisms. The second proof deduces the isomorphism from the Braden's theorem by using schemes acted on by \(\mathbb{G}_m\). The result is similar to the Second Adjointness theorem in the theory of \(\mathfrak{p}\)-adic groups, and to a theorem of Lusztig on restrictions of character sheaves. The latter and the main result of the paper are particular cases of a general statement, which replaces \(\text{Bun}_G\) by the moduli stack of \(G\)-bundles with level structure at a finite collection of points on \(X\) (the Lusztig's theorem corresponds to \(X=\mathbb{P}^1\) with level \(1\) structure at \((0,\infty)\)).