Whittaker coefficients of geometric Eisenstein series
Abstract.
Geometric Langlands predicts an isomorphism between Whittaker coefficients of Eisenstein series and functions on the moduli space of -local systems. We prove this formula by interpreting Whittaker coefficients of Eisenstein series as factorization homology and then invoking Beilinson and Drinfeld’s formula for chiral homology of a chiral enveloping algebra.
1. Introduction
1.1. Notation and conventions
Let be a simply connected complex reductive group with Langlands dual group defined over . Choose a maximal torus and a Borel subgroup with unipotent radical . Let be half the sum of the positive coroots. Let be a smooth projective complex genus curve. Choose a square root of the canonical bundle on and form the anticanonical -bundle . All categories and functors are derived.
Let be a -local system on , and let be the derived moduli stack of -local systems on whose underlying -local system is identified with , see (1.3). A -local system is called regular if for every coroot the associated rank 1 local system is nontrivial. If is regular then is a classical affine scheme isomorphic to a vector space.
Let be the Hecke -eigensheaf on whose stalk at twisted by a negative coweight valued divisor is
(1.1) |
Above, means the fiber at of the rank 1 local system obtained from using . Here and is the shift appearing in section 6.4.8 of [Gai13].
Remark.
Let correspond under class field theory to the skyscraper sheaf . The Hecke eigensheaf condition determines up to tensoring by a line. Whittaker normalization says that global sections of equals the costalk at the trivial -bundle of . Thus is only noncanonically isomorphic to a shift of . On the degree connected component , there is a canonical identification . We translated by , which has the effect of tensoring it by a certain line, see section 4.1 of [Gai10].
The Whittaker or Poincaré series sheaf on is the pullback then pushforward of the exponential sheaf along
see 5.4.1 of [FR22]. The function is defined in for example [FGV01]. The character sheaf on is normalized so that its costalks are in degree zero. Up to a shift, its Verdier dual is the inverse character sheaf. Both and corepresent shifted vanishing cycles for conic sheaves on , so the distinction is not so important. In the Betti setting we do not have the exponential D-module. Because is -equivariant for the action on and the weight 2 action on , the sheaf defined in 2.5.2 of [NY19] serves as a substitute. The Whittaker sheaf does not have nilpotent singular support.
The automorphic and spectral Eisenstein series functors, and , are defined by pullback then pushforward along
All of the above functors are left adjoints, in particular is a left adjoint because is proper. In (1.2), the functor is modified according to section 4.1 of [Gai10] or section 6.4.8 [Gai13]. This matches the translation by and shift by built into our definition of in (1.1).
1.2. Main theorem statement
The geometric Langlands conjecture is supposed to be compatible with parabolic induction. Moreover the Whittaker functional is expected to correspond under Langlands to global sections on , up to a shift by . Thus commutativity of conjectural diagram
(1.2) |
applied to the skyscraper , predicts the following isomorphism.
Main theorem.
Let be a -local system on and let be the Hecke eigensheaf on defined in (1.1). Whittaker coefficients of Eisenstein series equals functions on moduli space of -local systems:
The proof uses a combination of [Ras21] and [BG08] to relate twisted cohomology of the Zastava space to the formal completion of .
Both sides of the main theorem are coweight graded vector spaces. On the automorphic side, let be the restriction to the degree connected component . On the spectral side, the adjoint -action on induces an action on .
Our results apply for all three versions of geometric Langlands: de Rham, restricted, and Betti. On the automorphic side is a constructible sheaf, equivalently regular holonomic D-module, with nilpotent singular support, see [Gin01]. On the spectral side there are three versions of the moduli space of local systems, all having the same complex valued points. For a unipotent group
(1.3) |
coincide by proposition 4.3.3 and section 4.8.1 of [AGK+20].
Remark.
Consistency check.
If is a regular then theorem 10.2 of [BG08] says that is perverse. The Whittaker functional is exact by [NT22] or [FR22], so the automorphic side of the main theorem is concentrated in degree 0. This is consistent with being a classical scheme if is regular. Here
(1.4) |
is the dimension of the degree connected component.
1.3. Proof outline
It is convenient to take the coweight graded linear dual to avoid topological rings and because Lie algebra homology behaves better than Lie algebra cohomology. Here is the proof of our main theorem in one sentence:
(1.5) | ||||
In section 2.1, we use [NT22] or [FR22] to exchange for a right adjoint, then apply base change and a result of [AG15] to get a calculation on the Zastava space. In section 2.2, we pushforward to the space of positive coweight valued divisors and, by theorem 4.6.1 of [Ras21], obtain a certain factorizable perverse sheaves on .
In section 2.3, we interpret in terms of the chiral enveloping algebra of as in [BG08]. In section 2.4, we explain, following [BG08], how the cohomology of equals factorization homology of . Beilinson and Drinfeld’s formula says factorization homology of is Lie algebra homology of . In section 2.5, we study moduli of -local systems using deformation theory. Since is the shifted tangent complex of , its Lie algebra homology is related the formal completion of at . Using that is the quotient of an affine scheme by a unipotent group and using the contracting -action, we show that is the graded linear dual ring of functions.
The idea of using factorization homology to study the formal completion of is from [BG08]. For regular, propositions 11.3 and 11.4 of [BG08] give an isomorphism between and the completed ring of functions . Sections 2.3 and 2.4 review some of their arguments and do not contain new content apart from filling in some details. Our main contribution is in section 2.5 where we extend the results of [BG08] to the more interesting case of irregular , and obtain a formula for the ring of functions on (not just its formal completion) using the contracting -action.
1.4. Acknowledgements
I thank David Nadler for suggesting Whittaker coefficients of Eisenstein series and for generous discussions. This work was partially supported by NSF grant DMS-1646385.
2. Proof of the main theorem
2.1. Base change to Zastava
In this section we interpret Whittaker coefficients of Eisenstein series as twisted cohomology of the Zastava space .
The fiber product has a stratification indexed by the Weyl group, determined by the generic relative position of two flags. Let be the open inclusion of the locus where the two flags are generically transverse, called the Zastava space.
Consider the compositions
and let and be their restrictions to .
Proposition 1.
There is an isomorphism
(2.1) |
Proof.
We cannot directly apply adjunction to calculate Whittaker coefficients of Eisenstein series because is a left not right adjoint. It is shown in [FR22] and [NT22] that the shifted Whittaker functional on nilpotent sheaves commutes with Verdier duality . This allows us to exchange for . Then apply adjunction and base change to reduce to a calculation on the fiber product .
Finally by equation (3.5) of [AG15], restriction to the open generically transverse locus does not change the calculation. More precisely the map
is an isomorphism. For the shifts use (1.4) and . ∎
2.2. Pushforward to the configuration space
In this section we recall how to factor the projection through the configuration space of positive coweight valued divisors of total degree . Hence a description of the -graded piece of proposition 1 as cohomology of a certain perverse sheaf on .
Let be a point in the connected component of Zastava space, that is a -bundle with generically transverse -reductions , such that has degree and . For each dominant weight the Plucker description gives maps
(2.2) |
Here is a line bundle and is the vector bundle associated to the simple -module of highest weight .
By the generic transversality condition, the composition (2.2) is nonzero map of line bundles, so is a non-negative coweight. For each point in the Zastava space, there is a unique positive coweight valued divisor such that (2.2) factors through an isomorphism . Since is assumed simply connected, we can write as a sum of simple coroots and is a product of symmetric powers of the curve. Therefore factors through a map to the configuration space followed by the Abel-Jacobi map,
Let be a coweight and . Let , an -local system on . The Chevalley complex on the coweight graded Ran space gives a equivariant sheaf on . Symmetrizing by pushing forward along gives a perverse sheaf , see section 4 of [Ras21] and equation (2.6). The definition of involves the Chevalley differential but the associated graded of with respect to the Cousin filtration is easier to describe, see section 3.3 of [BG08]. The stalk of at a positive coweight valued divisor is
(2.3) |
Proposition 2.
There is an isomorphism
(2.4) |
Proof.
Pushing forward to the configuration space , the left of (2.4) becomes
We used that the configuration space is smooth so the dualizing sheaf is a rank 1 local system. And we used theorem 4.6.1 of [Ras21], which says that
Here and is a perverse sheaf on with stalks
Under class field theory (1.1), the stalks of are
and its -pullback to is the equivariant rank 1 local system . By the projection formula, tensoring with has the effect of twisting by . ∎
2.3. The chiral enveloping algebra as a Chevalley complex
The local system determines a Lie* algebra on the Ran space. Its Lie algebra homology is a factorization algebra, related to by partial symmetrization in (2.6).
A sheaf on the Ran space of is a collection of sheaves on each power of the curve compatible under -restriction along all partial diagonal maps, see 4.2.1 of [BD04] or 2.1 of [FG12]. Recall from section 1.2.1 of [FG12] that the category of sheaves on the Ran space admits two tensor products with a map between them.
Pushing forward along the main diagonal , we can regard as a Lie algebra for the -tensor product. Restricting to , the Lie* bracket comes by adjunction from the Lie bracket.
Let be Lie algebra homology of with respect to the -tensor product, viewed by the forgetful functor as a cocommutative coalgebra with respect to the -tensor product. Proposition 6.1.2 of [FG12] says that corresponds to the chiral enveloping algebra of under the equivalence between factorization and chiral algebras.
The Chevalley complex is coweight graded because is coweight graded, and because the Chevalley differential preserves the grading. Choose a coweight and let . The sheaf on is -equivariant and perverse. Symmetrize it along to get a perverse sheaf on the th symmetric power. (In other words we pushed forward from the stack quotient to the coarse quotient .)
Now we describe a certain perverse subsheaf defined in section 3 of [BG08]. Let be the space of effective degree divisors supported at exactly points. The -restriction of to is a local system whose stalk at a divisor is given by
The -restriction of to is the summand whose stalks are
(2.5) |
2.4. Factorization homology
In this section we review, following [BG08], how factorization homology of can be computed as cohomology on the symmetric power , where .
Let be the category whose objects are finite nonempty sets and whose morphisms are surjective maps. For each surjection there is a partial diagonal map . By definition of a sheaf on the Ran space so adjunction gives maps . Factorization homology is defined in section 6.3.3 of [FG12] or section 4.2.2 of [BD04] as the colimit over these maps
The following proposition is stated in 11.6 of [BG08] and below we fill in the proof using the Cousin filtration and ideas from section 4.2 of [BD04].
Proposition 3.
The cohomology of , see (2.3), is the factorization homology of the Chevalley complex,
(2.7) |
Proof.
Equation (2.6) relates to the symmetrization of . Thus it suffices to show that
(2.8) |
is an isomorphism for . Indeed we will prove that (2.8) is compatible with the Cousin filtration and that it induces an isomorphism on the associated graded pieces.
Consider the filtration on (2.8) whose th filtered piece consists of sections supported on the partial diagonals of dimensions . The th graded piece is
(2.9) |
Here is the !-restriction of to the space of effective degree divisors supported at exactly points. Similarly is the !-restriction of to the space of -tuples supported at exactly points.
The symmetric group acts freely on the space of distinct -tuples of points. By section 4.2.3 of [BD04], the th graded piece of the factorization homology of is .
The connected components of are indexed by partitions . Also the local system splits as a direct sum indexed by such partitions. Restricting (2.9) to the connected component indexed by a certain partition,
is an isomorphism onto the corresponding summand of by (2.5). Summing over partitions shows that the th graded piece of (2.8) is an isomorphism. ∎
2.5. Deformation theory
In this section we show that
(2.11) |
Lie algebra homology of the shifted tangent complex equals the graded dual ring of functions on . Deformation theory says that is global sections of the dualizing sheaf on the formal completion at . Using the structure of described in proposition 4, we can recover the graded dual ring of functions on , not just its completion, from .
First we show that is the quotient by a unipotent group of an affine derived scheme with a contracting -action. Let (respectively ) be the Betti moduli of (respectively ) local systems trivialized at a point . Let be the moduli of -local systems with underlying -local system identified with , plus a -reduction at .
Since is abelian, it acts by automorphisms on so there is a canonical lift . We also sometimes regard as a point in via the inclusion .
Let act on by changing the trivialization at , equivalently by the adjoint action on . Restricting the adjoint action along gives a -action that contracts to . Thus we expect a -action that contracts to , as is made precise below.
Proposition 4.
The moduli space is a finite type affine scheme with a -action. Restricting the action along gives a non-negative grading such that is cut out by the ideal of strictly positively graded functions.
Proof.
We argue in the Betti setting, but the restricted and de Rham versions also follow by (1.3). First rewrite
(2.12) |
The contracting -action induces non-negative gradings on the classical rings and . Since is smooth, the augmentation module admits a finite graded resolution by free -modules, with all but one term shifted into strictly positive -gradings. Therefore is a finite type non-negatively graded ring and . ∎
Now we review some derived deformation theory. Let be the formal completion of a derived stack at a point . The shifted tangent bundle is a dg Lie algebra whose enveloping algebra is endomorphisms of the skyscraper at . By chapter 7 of [GR17b] or remark 2.4.2 of [Lur11], there is an equivalence
between Lie algebra modules for the shifted tangent complex and indcoherent sheaves on the formal completion. Let be the map to a point. By chapter 7 section 5.2 of [GR17b], the trivial -module corresponds to the dualizing sheaf . Moreover Lie algebra homology corresponds to global sections
(2.13) |
Suppose is the spectrum of an Artinian local ring . By properness, is right adjoint to . Therefore the dualizing complex is the linear dual of viewed as an -module.
Suppose where and let . Since is proper, is the colimit under -pushforward of , see chapter 1 proposition 2.5.7 of [GR17a]. The dualizing sheaf can be written as a colimit, , see chapter 7 corollary 5.3.3 of [GR17b]. Since is continuous it follows that
(2.14) |
is the topological dual of the completed local ring . In this case, equation (2.13) is corollary 5.2 of [Hin96].
Proposition 5.
Let be a non-negatively graded finite type derived ring with . Let be the formal completion with respect to the ideal of positively graded functions. Then the graded dual equals the topological dual of the completion .
Proof.
First suppose is classical and choose homogeneous generators . Let be the maximum of their degrees, so . Therefore the graded dual (linear functionals that vanish on some ) equals the topologogical dual (linear functionals that vanish on some ).
Now suppose that is derived. The finite type assumption means that after taking cohomology is a finitely generated module over , a finitely generated graded classical ring. Choose a finite collection of homogeneous elements whose images generate .
The formal completion is the topological ring
For the first equality, see section 6.7 of [GR14]. The second equality uses that fiber products commute with filtered colimits and that . (The formal completion of a classical positively graded polynomial algebra can be computed using the grading filtration.)
Since is smooth, has finite dimensional cohomology and therefore is concentrated in bounded degrees. Hence for sufficiently large the quotient map factors through
Therefore the formal completion of can be computed using the grading filtration
Taking the topological dual proves . ∎
The following proposition shows (2.11), completing the final step of (1.5) and the proof of the main theorem.
Proposition 6.
Lie algebra homology of the shifted tangent complex of equals the graded dual of the ring of functions,
Proof.
Write as in proposition 4. Let act by changing the -reduction at . Since normalizes , the quotient retains the -action. The formal completion of at is the inf-scheme , the quotient by the formal group .
Deformation theory says
The first equality is equation (2.13). For the second equality we pushed forward the dualizing sheaf in two steps,
The pushforward of to is an -module. By proper base change and (2.14), the underlying vector space is and the -module structure comes from the -action. Further pushing forward along corresponds to taking -coinvariants so .
Now we show that -coinvariants of the topological dual of equals the graded dual ring of functions on ,
The ideal is an -module because the -action increases -weights. For the first equality, proposition 5 says that , and coinvariants commutes with colimits. For the second equality, because has finite dimensional cohomology. For the third equality, the image of is concentrated in degrees so we get a map . Moreover since is concentrated in bounded degrees, for sufficiently large the quotient map factors through
For the fourth equality, we used the van Est isomorphism, see theorem 5.1 of [Hoc61]. Since is unipotent, Lie algebra cohomology coincides with group cohomology . ∎
Example.
Let and let be a -local system, viewed as a rank 1 local system using the positive coroot. Then is regular if and only if it is nontrivial.
If is regular, then is a classical affine scheme because the other cohomologies vanish. The shifted tangent complex is an abelian Lie algebra with enveloping algebra . Lie algebra homology of the shifted tangent complex is
If is trivial then is the graded dual ring of functions on .
References
- [AG15] Dima Arinkin and Dennis Gaitsgory. Asymptotics of geometric Whittaker coefficients. Preprint. https://people.mpim-bonn.mpg.de/gaitsgde/GL/WhitAsympt.pdf, 2015.
- [AGK+20] Dima Arinkin, Dennis Gaitsgory, David Kazhdan, Sam Raskin, Nick Rozenblyum, and Yasha Varshavsky. The stack of local systems with restricted variation and geometric Langlands theory with nilpotent singular support. arXiv preprint arXiv:2010.01906, 2020.
- [BD04] Alexander Beilinson and Vladimir Drinfeld. Chiral algebras, volume 51 of American mathematical society colloquium publications. American Mathematical Society, Providence, RI, 1:25–26, 2004.
- [BG02] Alexander Braverman and Dennis Gaitsgory. Geometric Eisenstein series. Invent. math, 150:287–384, 2002.
- [BG08] Alexander Braverman and Dennis Gaitsgory. Deformations of local systems and Eisenstein series. GAFA Geom. funct. anal, 17:1788–1850, 2008.
- [BHKT19] Gebhard Böckle, Michael Harris, Chandrashekhar Khare, and Jack A Thorne. G-local systems on smooth projective curves are potentially automorphic. Acta Mathematica, 223(1):1–111, 2019.
- [FG12] John Francis and Dennis Gaitsgory. Chiral Koszul duality. Selecta Mathematica, 18(1):27–87, 2012.
- [FGV01] Edward Frenkel, Dennis Gaitsgory, and Kari Vilonen. Whittaker patterns in the geometry of moduli spaces of bundles on curves. Annals of Mathematics, 153(3):699–748, 2001.
- [FR22] Joakim Færgeman and Sam Raskin. Non-vanishing of geometric Whittaker coefficients for reductive groups. 2022.
- [Gai10] Dennis Gaitsgory. Notes on geometric Langlands: The extended Whittaker category, 2010.
- [Gai13] Dennis Gaitsgory. Outline of the proof of the geometric Langlands conjecture for GL(2). arXiv preprint arXiv:1302.2506, 2013.
- [Gin01] Victor Ginzburg. The global nilpotent variety is Lagrangian. Duke Mathematical Journal, 109(3):511–519, 2001.
- [GR14] Dennis Gaitsgory and Nick Rozenblyum. DG indschemes. Perspectives in representation theory, 610:139–251, 2014.
- [GR17a] Dennis Gaitsgory and Nick Rozenblyum. A Study in Derived Algebraic Geometry: Volume I: Correspondences and Duality, volume 221. 2017.
- [GR17b] Dennis Gaitsgory and Nick Rozenblyum. A Study in Derived Algebraic Geometry: Volume II: Deformations, Lie Theory and Formal Geometry, volume 221. 2017.
- [Hin96] Vladimir Hinich. Descent of Deligne groupoids. arXiv preprint alg-geom/9606010, 1996.
- [Hoc61] Gerhard Hochschild. Cohomology of algebraic linear groups. Illinois Journal of Mathematics, 5(3):492–519, 1961.
- [Lur11] Jacob Lurie. DAG X: Formal moduli problems. http://www.math.harvard.edu/ lurie, 2011.
- [NT22] David Nadler and Jeremy Taylor. The Whittaker functional is a shifted microstalk. arXiv preprint arXiv:2206.09216, 2022.
- [NY19] David Nadler and Zhiwei Yun. Geometric Langlands correspondence for SL(2), PGL(2) over the pair of pants. Compositio Mathematica, 155(2):324–371, 2019.
- [Ras21] Sam Raskin. Chiral principal series categories I: finite dimensional calculations. Advances in Mathematics, 388:107856, 2021.