\AtEveryBibitem\clearfield

issn \clearfieldisbn \clearfielddoi \clearfieldurl \clearfieldeprint \clearlistlocation \clearfieldmonth \clearfieldday \clearfieldseries \clearfieldpages \clearlistlanguage \clearlisttranslator \renewbibmacroin: \addbibresourcereferences.bib

Quantization of the universal centralizer and central D-modules

Tom Gannon and Victor Ginzburg

Abstract.

The group scheme of universal centralizers of a complex reductive group $G$ has a quantization called the spherical nil-DAHA. The category of modules over this ring is equivalent, as a symmetric monoidal category, to the category of bi-Whittaker $\mathcal{D}$ -modules on $G$ . We construct a braided monoidal equivalence, called the Knop-Ngô functor, of this category with a full monoidal subcategory of the abelian category of Ad $G$ -equivariant $\mathcal{D}$ -modules, establishing a $\mathcal{D}$ -module abelian counterpart of an equivalence established by Bezrukavnikov and Deshpande, in a different way.

As an application of our methods, we prove conjectures of Ben-Zvi and Gunningham by relating this equivalence to parabolic induction and prove a conjecture of Braverman and Kazhdan in the $\mathcal{D}$ -module setting.

1. Main results

We work over the field of complex numbers to be denoted $k$ and write ${\mathcal{O}}(X)$ for the algebra of regular functions on a variety $X$ .

1.1. Knop-Ngô morphism

Let $G$ be a connected reductive group with Lie algebra $\mathfrak{g}$ . Let ${\mathfrak{c}}:=\mathfrak{g}^{*}/\!/G=\operatorname{{\mathrm{Spec}}}(% \operatorname{{\mathrm{Sym}}}\mathfrak{g})^{G}$ be the categorical quotient of the coadjoint action of $G$ in $\mathfrak{g}^{*}$ (to be denoted ‘ $\operatorname{{\mathrm{Ad}}}$ ’), and $Z=\{(g,\xi)\in G\times\mathfrak{g}^{*}\mid\operatorname{{\mathrm{Ad}}}g(\xi)=\xi\}$ the commuting variety viewed as a reduced closed subscheme of $G\times\mathfrak{g}^{*}$ . The projection map $pr:Z\to\mathfrak{g}^{*}$ makes $Z$ a group scheme over $\mathfrak{g}^{*}$ such that the fiber $pr^{-1}(\xi)$ for $\xi\in\mathfrak{g}^{*}$ is the stabilizer of $\xi\in\mathfrak{g}^{*}$ in $G$ . Restricting to the set $\mathfrak{g}^{*}_{\text{reg}}\subseteq\mathfrak{g}^{*}$ of regular elements, one obtains the group scheme $Z_{\text{reg}}=pr^{-1}(\mathfrak{g}^{*}_{\text{reg}})\to\mathfrak{g}^{*}_{% \text{reg}}$ of ‘regular centralizers.’

In his proof of the fundamental lemma, Ngô constructed a smooth affine abelian group scheme $J\to{\mathfrak{c}}$ , called the ‘universal centralizer,’ equipped with a canonical isomorphism $\mathfrak{g}^{*}_{\text{reg}}\times_{{\mathfrak{c}}}J{\;\stackrel{{{}_{\sim}}}% {{\to}}\;}Z_{\text{reg}}$ of $G$ -equivariant group schemes on $\mathfrak{g}^{*}_{\text{reg}}$ that fits into commutative diagram

(1.1.1)

of morphisms of $G$ -equivariant group schemes, [NgoLeLemmeFondamentalPourLesAlgebresdeLie, Section 2.1]. Similar constructions were also considered by Knop [KnopAutomorphismsRootSystemsandCompactificationsofHomogeneousVarieties], and so we will refer to the map $j$ in 1.1.1 as the Knop-Ngô morphism.

The group scheme structure makes $\mathcal{O}(Z)$ , resp. $\mathcal{O}(J)$ , a Hopf algebra over $\operatorname{{\mathrm{Sym}}}(\mathfrak{g})=\mathcal{O}(\mathfrak{g}^{*})$ , resp. $\mathcal{O}({\mathfrak{c}})$ , and there is an isomorphism $\mathcal{O}(\mathfrak{g}^{*}\times_{{\mathfrak{c}}}J)=\operatorname{{\mathrm{% Sym}}}(\mathfrak{g})\otimes_{{\mathcal{O}}({\mathfrak{c}})}\mathcal{O}(J)$ , of Hopf algebras over $\operatorname{{\mathrm{Sym}}}(\mathfrak{g})$ . Therefore, it follows from the above diagram that restriction of regular functions from $Z$ to $Z_{\text{reg}}$ can be factored as a composition of Hopf algebra maps

(1.1.2)

\mathcal{O}(Z)\xrightarrow{{\mathsf{j}}}\operatorname{{\mathrm{Sym}}}(% \mathfrak{g})\otimes_{\mathcal{O}({\mathfrak{c}})}\mathcal{O}(J)\xrightarrow{}% \mathcal{O}(Z_{\text{reg}}).

We also refer to the first map in 1.1.2 as the Knop-Ngô morphism.

Let $\mathcal{O}(J)\operatorname{{\text{-}{\mathrm{mod}}}}$ , resp. $\mathcal{O}(Z)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}$ , be the abelian category of $\mathcal{O}(J)$ -modules, resp. $G$ -equivariant $\mathcal{O}(Z)$ -modules. The group scheme structure gives each of these categories a symmetric monoidal structure with respect to convolution. Using Diagram 1.1.1 we obtain a chain of monoidal functors

(1.1.3)

\mathcal{O}(J)\operatorname{{\text{-}{\mathrm{mod}}}}\ \xrightarrow{% \operatorname{{\mathrm{Sym}}}(\mathfrak{g})\otimes_{\mathcal{O}({\mathfrak{c}}% )}({\text{-}})}\ \mathcal{O}(\mathfrak{g}^{*}\times_{\mathfrak{c}}J)% \operatorname{{\text{-}{\mathrm{mod}}}}^{G}\ \xrightarrow{j_{*}}\ \mathcal{O}(% Z)\operatorname{{\text{-}{\mathrm{mod}}}}^{G},

where the second functor $j_{*}$ , which corresponds to push-forward of quasi-coherent sheaves along the imbedding $j$ , makes an $\mathcal{O}(\mathfrak{g}^{*}\times_{\mathfrak{c}}J)$ -module into an $\mathcal{O}(Z)$ -module via the algebra map ${\mathsf{j}}$ .

One of the goals of this paper, inspired to a large extent by the ideas of Ben-Zvi and Gunningham [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction], is to construct a quantization of the algebra morphism ${\mathsf{j}}$ , resp. the composite functor in 1.1.3.

The algebras that appear in (1.1.2) each have a well known quantization, that is, non-commutative deformation. The quantization of $\operatorname{{\mathrm{Sym}}}(\mathfrak{g})$ , resp. $\mathcal{O}({\mathfrak{c}})$ , is the enveloping algebra $U\mathfrak{g}$ , resp. the center $Z\mathfrak{g}$ of $U\mathfrak{g}$ . The quantization of $\mathcal{O}(G\times\mathfrak{g}^{*})$ is the algebra $D(G)$ of differential operators on $G$ , where we have used the identification $G\times\mathfrak{g}^{*}\cong T^{*}G$ . The left ideal $D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}\subseteq D(G)$ quantizes the ideal of definition of the not necessarily reduced subscheme $Z_{\text{scheme}}\subseteq G\times\mathfrak{g}^{*}$ . Therefore, the $D(G)$ -module ${\mathbf{N}}:=D(G)/D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}$ provides a quantization of the functions on this subscheme, viewed as an $\mathcal{O}(T^{*}G)$ -module. We show that the coalgebra structure on ${\mathcal{O}}(Z)$ can be deformed to the structure of a coalgebra object on ${\mathbf{N}}$ , cf. 2.4.1.

The algebra $\mathcal{O}(J)$ is known to have a deformation to a noncommutative algebra ${\mathfrak{J}}$ called spherical nil DAHA. This algebra may be defined as a quantum Hamiltonian reduction (aka bi-Whittaker reduction) of $D(G)$ with respect to $(N\times N_{op},\,\psi\times\psi)$ , where $N\subset G$ is a maximal unipotent subgroup, $\psi$ is a nondegenerate character of $N$ , and the group $N$ , resp. $N_{op}$ , acts on $G$ by left, resp. right, translations, see [BezrukavnikovFinkelbergMirkovicEquivariantHomologyandKTheoryofAffineGrassmanniansandTodaLattices], [BezrukavnikovFinkelbergEquivariantSatakeCategoryandKostantWhittakerReduction], and also Section 3.3 for more details. It was shown in loc cit that $\mathfrak{J}$ is isomorphic, via geometric Satake, to the equivariant Borel-Moore homology of the affine Grassmannian $\mathsf{Gr}$ of the Langlands dual group as an algebra. The coproduct on the equivariant Borel-Moore homology of the affine Grassmannian induced, via the Künneth formula, by the diagonal imbedding $\mathsf{Gr}\,\hookrightarrow\,\mathsf{Gr}\times\mathsf{Gr}$ , therefore equips $\mathfrak{J}$ with a coalgebra structure. In Section 3 we will give an algebraic construction of this coproduct and an associated symmetric monoidal structure on the category ${\mathfrak{J}}\operatorname{{\text{-}{\mathrm{mod}}}}$ .

In view of the above, it is natural to expect that the quantized Knop-Ngô map is a map of the form

(1.1.4)

{\mathbf{N}}\to U\mathfrak{g}\otimes_{Z\mathfrak{g}}{\mathfrak{J}}.

A primary difficulty in constructing such a map is that it is not a priori clear what kind of structure to expect from the map in 1.1.4: the domain of the map has the structure of a $D(G)$ -module while the codomain is a $(U\mathfrak{g},{\mathfrak{J}})$ -bimodule that has no immediately visible $D(G)$ -action.

We construct quantizated Knop-Ngô morphism and quantizated Knop-Ngô functor using the monoidal category $\mathcal{HC}$ of Harish-Chandra $(U\mathfrak{g},U\mathfrak{g})$ -bimodules, a category which serves as a bridge between the categories of ${\mathfrak{J}}$ -modules and $D(G)$ -modules. To this end, given ${\mathfrak{J}}$ -module $E$ we make $U\mathfrak{g}\otimes_{Z\mathfrak{g}}E$ a Harish-Chandra bimodule with ‘diagonal’ action of the algebra $U\mathfrak{g}\otimes U\mathfrak{g}_{op}$ defined the formula $u_{1}\otimes u_{2}:u\otimes e\mapsto u_{1}uu_{2}\otimes e$ . A key idea is that the ${\mathfrak{J}}$ -module structure on $E$ provides the Harish-Chandra bimodule $U\mathfrak{g}\otimes_{Z\mathfrak{g}}E$ with a canonical nontrivial central structure. That is, we show that there is a canonical lift of $U\mathfrak{g}\otimes_{Z\mathfrak{g}}E$ to an object of the Drinfeld center of $\mathcal{HC}$ . The Drinfeld center has been identified in [BezrukavnikovFinkelbergOstrikCharacterDModulesViaDrinfeldCenterofHarishChandraBimodules] with the category $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{\operatorname{{\mathrm{Ad}}}G}$ of $G$ -equivariant $D(G)$ -modules, where $G$ acts by the adjoint action. We will actually use a strengthening of this result that identifies the latter category with the centralizer of the category $\operatorname{{\mathrm{Rep}}}G$ in $\mathcal{HC}$ , see Section 6. We therefore obtain the structure of a $G$ -equivariant $D(G)$ -module on $U\mathfrak{g}\otimes_{Z\mathfrak{g}}E$ .

Let $W$ be the Weyl group and $D(T)^{W}$ the algebra of differential operators on the maximal torus $T$ . In part (ii) of the theorem below we use the algebra isomorphism $(\operatorname{End}_{D(G)}{{\mathbf{N}}})_{\text{op}}\cong D(T)^{W}$ constructed by Harish-Chandra via the ‘radial parts’ map, cf. [GinzburgParabolicInductionandtheHarishChandraDModule, Remark (iv)] for a short exposition. This gives ${{\mathbf{N}}}$ a right $D(T)^{W}$ -action that commutes with the $D(G)$ -action.

Let $1_{{\mathbf{N}}}\in{\mathbf{N}}$ denote the image of $1\in U\mathfrak{g}$ . Our first result, whose first part is proved in Section 6.2 and second part is proved in Section 7.4, reads

Theorem 1.1.5.

${{\enspace}\sf{(i)}}\;$ The functor $\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to{\mathcal{HC}},\,E% \mapsto U\mathfrak{g}\otimes_{Z\mathfrak{g}}E$ , has a canonical lift to a fully faithful, braided monoidal, exact functor $\operatorname{KN}:{\mathfrak{J}}\operatorname{{\text{-}{\mathrm{mod}}}}\to D(G% )\operatorname{{\text{-}{\mathrm{mod}}}}^{\operatorname{{\mathrm{Ad}}}G}$ .

${\;\sf{(ii)}}\;$ For any $\mathfrak{J}$ -module $E$ , the map $D(G)\otimes E\to U\mathfrak{g}\otimes_{Z\mathfrak{g}}E,\,u\otimes e\mapsto u(1% \otimes e)$ , given by the $D(G)$ -action factors yields an isomomorphism of $D(G)$ -modules:

\mathbf{N}\otimes_{D(T)^{W}}E{\;\stackrel{{{}_{\sim}}}{{\to}}\;}\operatorname{% KN}(E).

Applying the above in the special case where $E$ is a rank one free ${\mathfrak{J}}$ -module yields the desired $D(G)$ -module structure on the object $U\mathfrak{g}\otimes_{Z\mathfrak{g}}{\mathfrak{J}}=\operatorname{KN}({% \mathfrak{J}})$ , that appears in (1.1.4). Using this, we prove the following result, which confirms a conjecture of Nadler:

Corollary 1.1.6.

The map

{{\mathbf{N}}}\ \xrightarrow{}\ {U\mathfrak{g}}\otimes_{Z\mathfrak{g}}{% \mathfrak{J}},\enspace u1_{{\mathbf{N}}}\,\mapsto\,u(1\otimes 1),

gives a $D(G)$ -module map of coalgebras in $\mathcal{HC}$ that quantizes the Knop-Ngô morphism.

Remark 1.1.7 (Comparison to work of Ben-Zvi and Gunningham).

The idea to relate the category $\mathfrak{J}$ -mod to the category $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}$ has been implemented in a different way by Ben-Zvi and Gunningham [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction], who have obtained a result resembling our 1.1.5(i). However, their approach is different from ours in that they work with the full triangulated category of $G$ -equivariant $\mathcal{D}$ -modules on $G$ rather than with the abelian heart of a particular $t$ -structure. The categorical center construction does not produce reasonable results when applied to triangulated categories, and the formalism of $\infty$ -categories (or, more precisely, DG categories as defined in [GRI, Chapter 1.10]) has been employed to overcome this difficulty. Using the exactness of Kostant-Whittaker reduction allows us to bypass that difficulty and work with abelian categories; thus the technique of the present paper is more elementary than that of [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction]. Also, the proof of [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction, Conjecture 2.9] discussed below, which was one of our main motivations for this project, is unlikely to be accessible by the methods of [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction] alone.

1.2. Parabolic induction and very central $D$ -modules

Given an algebraic group $H$ and a smooth $H$ -variety $X$ , we write ${\mathscr{D}}^{H}(X)$ for the $H$ -equivariant derived category of $D$ -modules on $X$ , so $\mathcal{D}_{X}\operatorname{{\text{-}{\mathrm{mod}}}}^{H}={\mathscr{D}}^{H}(X% )^{\heartsuit}$ is the abelian heart of ${\mathscr{D}}^{H}(X)$ .

The $W$ -equivariant derived category $\mathscr{D}^{W}(T)$ is known to be equivalent to the derived category of its heart $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}$ , the abelian category of $W$ -equivariant $D(T)$ -modules. Via the Morita equivalence

(1.2.1)

D(T)^{W}\sim D(T)\rtimes W

the latter category is equivalent to the category $D(T)^{W}\operatorname{{\text{-}{\mathrm{mod}}}}$ . Further, in Section 5.3, cf. also [Gin], we will see that there is a natural algebra imbedding $D(T)^{W}\,\hookrightarrow\,\mathfrak{J}$ , so one has the forgetful functor $\operatorname{Obl}^{{\mathfrak{J}}}_{{}_{D(T)^{W}}}:{\mathfrak{J}}% \operatorname{{\text{-}{\mathrm{mod}}}}\to D(T)^{W}\operatorname{{\text{-}{% \mathrm{mod}}}}$ .

Let $\operatorname{{\mathrm{Ind}}}_{T}^{G}:\mathscr{D}^{W}(T)\to\mathscr{D}^{G}(G)$ denote the functor of parabolic induction to the derived category $G$ -equivariant $D(G)$ -modules. The functor $\operatorname{{\mathrm{Ind}}}_{T}^{G}$ is exact by [BezrukavnikovYomDinOnParabolicRestrictionofPerverseSheaves] and so gives the induced functor $\operatorname{{\mathrm{Ind}}}_{T}^{G}:D(T)\operatorname{{\text{-}{\mathrm{mod}% }}}\to D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}$ on the respective abelian categories. Furthermore, for any ${\mathcal{F}}\in\mathscr{D}^{W}(T)$ the object $\operatorname{{\mathrm{Ind}}}_{T}^{G}({\mathcal{F}})$ comes equipped with a canonical $W$ -action, see for example [ChenOnTheConjecturesofBravermanKazhdan, Section 3.2]. Write $\operatorname{{\mathrm{Ind}}}_{T}^{G}({\mathcal{F}})^{W}$ for the direct summand of $W$ -invariants, which is well-defined since the group $W$ is finite.

1.1.5 combined with [GinzburgParabolicInductionandtheHarishChandraDModule, Theorem 1.6] yields the following result, which proves [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction, Conjecture 2.9]:

Corollary 1.2.2.

The functor $\operatorname{KN}$ is isomorphic to the composite functor

(1.2.3)

\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}\xrightarrow{\ % \operatorname{Obl}^{{\mathfrak{J}}}_{D(T)^{W}}\ }D(T)^{W}\operatorname{{\text{% -}{\mathrm{mod}}}}\xrightarrow{\ \text{\em Morita equivalence}\ }D(T)% \operatorname{{\text{-}{\mathrm{mod}}}}^{W}\xrightarrow{\ \operatorname{{% \mathrm{Ind}}}_{T}^{G}\ }D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}.

Let $p:G\to G/N$ be the natural projection. Recall that the Harish-Chandra functor may be defined as the push-forward functor $\int_{p}:\mathscr{D}^{G}(G)\to\mathscr{D}^{N}({G/N})$ for equivariant derived categories of $D$ -modules. In [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction], [ChenAVanishingConjecturetheGLnCase] and [BezrukavnikovDeshpandeVanishingSheavesandtheGeometricWhittakerModel], the authors considered a full monoidal subcategory $\mathscr{D}^{G}_{\operatorname{vc}}(G)^{\heartsuit}$ , whose objects $M$ , sometimes referred to as ‘very central $D$ -modules’, have the property that the support of $\int_{p}M$ is contained in the closed subvariety $B/N\subseteq G/N$ , [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction, Definition 2.12].

Our 1.2.5 below provides, in particular, a remarkably simple interpretation of the heart of the category of very central $D$ -modules.

To explain this, we introduce the full abelian subcategory $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}_{\circ}$ of $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}$ , resp. $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}_{\circ}$ of $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}$ , whose objects $M$ are induced from their invariants in the sense that the following map induced by the action of the subalgebra $\operatorname{{\mathrm{Sym}}}({\mathfrak{t}})\subseteq D(T)$ , resp. $U\mathfrak{g}\subseteq D(G)$ ,

(1.2.4)

\operatorname{{\mathrm{Sym}}}({\mathfrak{t}})\otimes_{\operatorname{{\mathrm{% Sym}}}({\mathfrak{t}})^{W}}M^{W}\to M,{\enspace}\text{resp.}{\enspace}U% \mathfrak{g}\otimes_{Z\mathfrak{g}}M^{G}\to M,

is a bijection. We remark that, the group $G$ being connected and the $D(G)$ -module $M$ being strongly $G$ -equivariant, the space $M^{G}$ above equals $\{m\in M\mid\xi_{l}m=\xi_{r}m=0\ \forall\xi\in\mathfrak{g}\}$ , where $\xi_{l}$ , resp. $\xi_{r}$ , is the right, resp. left, invariant vector field associated to $\xi$ . Therefore, replacing the algebra $U\mathfrak{g}$ of right invariant differential operators in (1.2.4) with the algebra $U\mathfrak{g}_{\text{op}}$ of left invariant differential operators leads to an equivalent property.

In Section 7 we prove the following theorem that, combined with 1.1.5, confirms [BenZviGunninghamSymmetriesofCategoricalRepresentationsandtheQuantumNgoAction, Conjecture 2.14]:

Theorem 1.2.5.

For $M\in D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}$ , the following are equivalent:

${{\enspace}\sf{(i)}}\;$ $M\in D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}_{\circ}$ .

${\;\sf{(ii)}}\;$ The support of the $\mathcal{D}$ -module $\int_{p}M$ is contained in $B/N$ , i.e. $M$ is very central;

${\sf{(iii)}}\;$ $M$ is contained in the essential image of the functor $\operatorname{KN}$ . Moreover, there are braided monoidal equivalences

(1.2.6)

D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}_{\circ}\ \xleftarrow[% \operatorname{Obl}^{{\mathfrak{J}}}_{D(T)^{W}}]{\sim}\ {\mathfrak{J}}% \operatorname{{\text{-}{\mathrm{mod}}}}\ \xrightarrow[\ \operatorname{KN}\ ]{% \sim}\ D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}_{\circ};

in particular, $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}_{\circ}$ is a symmetric monoidal category.

The left equivalence in 1.2.6 has been proved in [Gin, Theorem 1.5.1], which states that $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}_{\circ}$ is exactly the essential image of the functor $\operatorname{Obl}^{{\mathfrak{J}}}_{D(T)^{W}}$ , cf. also [Lo], [LoRemark], [GannonClassificationofNondegenerateGCategories], [GannonUniversalCategoryOandGelfandGraevAction], [GannonDescentToTheCoarseQuotientForPseudoreflectionAndAffineWeylGroups], [BezrukavnikovDeshpandeVanishingSheavesandtheGeometricWhittakerModel] for closely related results. The composed equivalence

D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}_{\circ}\cong\mathscr{D}^{W}_{% \circ}(T)^{\heartsuit}

has been obtained earlier by Bezrukavnikov and Deshpande, see [BezrukavnikovDeshpandeVanishingSheavesandtheGeometricWhittakerModel, Corollary 1.7]. Our proof in loc cit is totally different from the one in loc cit which uses some earlier results from [ChenOnTheConjecturesofBravermanKazhdan, Section 6] and [BezrukavnikovIonovTolmachovVarshavskyEquivariantDerivedCategoryofaReductiveGroupasaCategoricalCenter].

Remark 1.2.7.

Recall that convolution of $D(G)$ -modules is given by the formula, cf. [BezrukavnikovFinkelbergOstrikCharacterDModulesViaDrinfeldCenterofHarishChandraBimodules, Lemma 2.1] and Section 2.3 below, $M_{1}\star M_{2}=M_{1}\otimes_{U\mathfrak{g}}M_{2}$ , so it only involves the $(U\mathfrak{g},U\mathfrak{g})$ -bimodule structure of $D(G)$ -modules. Therefore, by the definition of the functor $\operatorname{KN}$ one has an isomorphism

(1.2.8)

\operatorname{KN}(E)\star M\cong E\otimes_{Z\mathfrak{g}}M,\quad C\in{% \mathfrak{J}}\operatorname{{\text{-}{\mathrm{mod}}}},\ M\in\mathcal{HC},

of Harish-Chandra bimodules. This formula plays an important role in the proof of 1.2.5.

1.3. The exactness and vanishing conjectures

As an application of our methods, we give proofs of conjectures of Braverman and Kazhdan in the $\mathcal{D}$ -module setting. Recall that, motivated by Langlands functoriality, Braverman and Kazhdan [BravermanKazhdanGammaSheavesonReductiveGroups] associated to every finite dimensional representation $G^{\vee}\to GL(\mathbf{V})$ of the Langlands dual group $G^{\vee}$ that satisfies a positivity condition (referred to as good in [BravermanKazhdanGammaSheavesonReductiveGroups] and $\sigma$ -positive in [ChenAVanishingConjecturetheGLnCase]) and a fixed choice of additive character $\mathbb{G}_{a}\to\mathbb{A}^{1}$ a certain $\operatorname{{\mathrm{Ad}}}G$ -equivariant perverse sheaf ${\boldsymbol{\Phi}}_{\mathbf{V}}$ on $G$ , known as the Bessel sheaf or $\gamma$ -sheaf. It was shown by Chen [ChenAVanishingConjecturetheGLnCase] that such $\gamma$ -sheaves lie in the essential image of the composite of functors in 1.2.3.

Braverman and Kazhdan proposed several conjectures involving $\gamma$ -sheaves, in particular, the Vanishing conjecture and the Exactness conjecture. The $D$ -module counterpart (of the generalization given in [ChenAVanishingConjecturetheGLnCase]) of the Vanishing conjecture reads

(1.3.1)

\textsf{Vanishing conjecture}:\qquad\;\mathcal{F}\in\mathscr{D}^{W}_{\circ}(T)% \enspace\Longrightarrow\enspace\operatorname{supp}\Big{(}\int_{p}\operatorname% {{\mathrm{Ind}}}_{T}^{G}(\mathcal{F})^{W}\Big{)}\,\subseteq\,B/N.

The vanishing conjecture (in the original form as well as in the $\mathcal{D}$ -module setting) was proved by Chen [ChenOnTheConjecturesofBravermanKazhdan], cf. also earlier papers [ChenAVanishingConjecturetheGLnCase], [ChenOnTheConjecturesofBravermanKazhdan], Laumon-Letellier [LaumanLetellierNoteonaConjectureofBravermanKazhdan], and Ngô [NgoHankelTransformLanglandsFunctorialityandFunctionalEquationofAutomorphicLFunctions] for partial earlier results and also [BezrukavnikovIonovTolmachovVarshavskyEquivariantDerivedCategoryofaReductiveGroupasaCategoricalCenter], [BezrukavnikovDeshpandeVanishingSheavesandtheGeometricWhittakerModel]. We will use 1.2.5 to derive (1.3.1) as a corollary. Thus, we obtain an alternate proof of the vanishing conjecture in the $D$ -module setting.

As a simple application of 1.2.8, we also prove the following $D$ -module counterpart of [BravermanKazhdanGammaSheavesonReductiveGroups, Conjecture 1.8] in Section 7.2:

Theorem 1.3.2.

[Exactness Conjecture] If $\mathbf{V}$ is good, then convolution functor $\boldsymbol{\Phi}_{\mathbf{V}}(\text{-})$ is a $t$ -exact endofunctor of the derived category of $D(G)$ -modules.

Remark 1.3.3.

${{\enspace}\sf{(i)}}\;$ It might be possible to deduce the Vanishing conjecture, resp. Exactness conjecture, in the $\ell$ -adic setting from the corresponding result in the $\mathcal{D}$ -module setting using the Riemann-Hilbert correspondence and standard comparison theorems (provided the ground field has characteristic zero).

${\;\sf{(ii)}}\;$ The proof of the vanishing conjecture in [ChenOnTheConjecturesofBravermanKazhdan] actually reduces the general case of the conjecture to the special case of a certain $D(G)$ -module which is a character sheaf, or in other words has nilpotent singular support, see [ChenOnTheConjecturesofBravermanKazhdan, Section 6].

1.4. Outline

In Section 2, we recall the definition of the category $\mathcal{HC}$ of Harish-Chandra bimodules, and discuss coalgebra type structures on the ring $D(G)$ of differential operators on $G$ inherited from the convolution monoidal structure on its category of $D(G)$ -modules. In Section 3, we construct the coalgebra structure on $\mathfrak{J}$ mentioned in the introduction and study monoidal structure on the categories of $\mathfrak{J}$ -modules and $\mathfrak{J}$ -comodules from an algebraic standpoint. In Section 4, we use this to give an explicit relationship between the monoidal categories $\mathcal{HC}$ and the category of comodules for $\mathfrak{J}$ . After proving some preliminary results on the Miura bimodule in Section 5, we quantize the Knop-Ngô morphism in Section 6 using a foundational result, 6.2.1, whose proof is deferred to Section 8. In Section 7 we complete the proofs of our remaining theorems.

1.5. Acknowledgements

We thank Paul Balmer, David Ben-Zvi, Roman Bezrukavnikov, Tsao-Hsien Chen, Peter Crooks, Pavel Etingof, Joakim Færgeman, Sam Gunningham, Ivan Losev, David Nadler, Cris Negron, Sam Raskin, Aaron Slipper, and Germán Stefanich for interesting and useful discussions.

2. Monoidal structure on $D(G)$ -modules

2.1. Conventions

By an algebraic representation (aka linear action) of an algebraic group $H$ on a not necessarily finite dimensional vector space $V$ we mean an $\mathcal{O}(H)$ -comodule with coaction $V\to V\otimes\mathcal{O}(H)$ which is coassociative and counital in the natural way. Let $\text{Rep}(H)$ denote the symmetric monoidal category of algebraic representations of $H$ . We will make no distinction between ${\mathbb{Z}}$ -gradings and algebraic actions of the multiplicative group ${\mathbb{G}}_{m}$ .

Recall the following

Definition.

${{\enspace}\sf{(i)}}\;$ A lax monoidal functor, resp. left lax monoidal functor, between monoidal categories $\mathcal{C}\to\mathcal{C}^{\prime}$ is the data a functor $F:\mathcal{C}\to\mathcal{C}^{\prime}$ , a morphism $\epsilon:\mathbf{1}_{\mathcal{C}^{\prime}}\to F(\mathbf{1}_{\mathcal{C}})$ , resp. $\epsilon^{\prime}:F(\mathbf{1}_{\mathcal{C}})\to\mathbf{1}_{\mathcal{C}^{% \prime}}$ where $\mathbf{1}_{\mathcal{C}},\mathbf{1}_{\mathcal{C}^{\prime}}$ are the respective monoidal units, and morphisms $\mu_{X,Y}:F(X)\otimes F(Y)\to F(X\otimes Y)$ , resp. $\mu^{\prime}_{X,Y}:F(X\otimes Y)\to F(X)\otimes F(Y)$ natural in $X,Y\in\mathcal{C}$ which satisfy an associativity and unit constraint given explicitly by the diagrams of [MonoidalFunctorDefinition, Definition 2.1];

${\;\sf{(ii)}}\;$ If $\epsilon$ , resp. $\epsilon^{\prime}$ , is an isomorphism and $\mu_{X,Y}$ , resp. $\mu^{\prime}_{X,Y}$ , is an isomorphism for all $X,Y\in\mathcal{C}$ we say that this lax monoidal functor is monoidal.

Remark 2.1.1.

What we refer to as a monoidal functor is sometimes referred to as a strongly monoidal functor, see [MonoidalFunctorDefinition, Remark 2.2].

We let $k[\hbar]$ denote the graded polynomial algebra with $\hbar$ in weight 2. Given a graded $k[{\hbar}]$ -algebra $\mathscr{A}$ we will use the notation $\mathscr{A}\operatorname{{\text{-}{\mathrm{mod}}}}$ , resp. $\mathscr{A}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ , for the category of graded left $\mathscr{A}$ -modules, resp. ${\mathscr{A}}\otimes_{k[{\hbar}]}{\mathscr{A}}_{\text{op}}$ -module.

For an algebra $A$ equipped with a ${\mathbb{Z}}$ -filtration $F_{\>\raisebox{2.0pt}{\text{\circle*{1.5}}}}A$ , let $A_{\hbar}=\sum_{i\in{\mathbb{Z}}}\,{\hbar}^{i}F_{i}A$ be the corresponding Rees algebra viewed as an even graded ${k[{\hbar}]}$ -algebra such that $F_{i}A$ is placed in degree $2i$ . In particular, given a Lie algebra ${\mathfrak{h}}=\operatorname{Lie}H$ one has the asymptotic enveloping algebra, ${\mathcal{U}}_{\hbar}{\mathfrak{h}}$ associated with the PBW filtration on the enveloping algebra $U{\mathfrak{h}}$ of the Lie algebra ${\mathfrak{h}}=\operatorname{Lie}H$ .

We will use the notations of [HTT] for morphisms of (left) $\mathcal{D}$ -modules. In particular, for any morphism of smooth algebraic varieties $f$ , we will let $\int_{f}$ denote the induced pushforward on the derived category of $\mathcal{D}$ -modules and let $f^{\dagger}$ denote the shifted inverse image functor as in [HTT, Section 1.5].

We write $\otimes=\otimes_{k}$ .

Throughout, we fix a connected reductive group $G$ with Lie algebra $\mathfrak{g}$ . Let $G_{\text{op}}$ , resp. $\mathfrak{g}_{\text{op}}$ , be an opposite group, resp. Lie algebra. We have an isomorphism $G{\;\stackrel{{{}_{\sim}}}{{\to}}\;}G_{\text{op}}$ , resp. $\mathfrak{g}{\;\stackrel{{{}_{\sim}}}{{\to}}\;}\mathfrak{g}_{\text{op}}$ , given by the map $g\mapsto g^{-1}$ , resp. $\xi\mapsto-\xi$ . We use simplified notation $D$ for the algebra $D(G)$ of differential operators on $G$ .

2.2. Asymptotic Harish-Chandra bimodules

Let ${\mathscr{A}}$ be an associative graded ${k[{\hbar}]}$ -algebra. A Hamiltonian $G$ -action is the data of a $G$ -action $\operatorname{{\mathrm{act}}}_{\mathscr{A}}:G\times{\mathscr{A}}\to{\mathscr{A% }},\,(g,a)\mapsto\operatorname{{\mathrm{act}}}_{\mathscr{A}}g(a)$ , by graded algebra automorphisms and a graded ${k[{\hbar}]}$ -algebra (comoment) map $\mu:U_{\hbar}\mathfrak{g}\to{\mathscr{A}}$ such that one has ${\hbar}(d\operatorname{{\mathrm{act}}}_{\mathscr{A}})(\xi)(a)=[\mu(x),\,a]$ for all $\xi\in\mathfrak{g},a\in{\mathscr{A}}$ , where $d\operatorname{{\mathrm{act}}}_{\mathscr{A}}$ is the differential of the $G$ -action. A graded left ${\mathscr{A}}$ -module $M$ equipped with a $G$ -action $\operatorname{{\mathrm{act}}}_{M}:G\times M\to M,\,(g,m)\mapsto\operatorname{{% \mathrm{act}}}_{M}g(m)$ is called a Harish-Chandra ${\mathscr{A}}$ -module if the map $\operatorname{{\mathrm{act}}}_{M}$ is $G$ -equivariant and for all $\xi\in\mathfrak{g},m\in M$ , one has $:{\hbar}d\operatorname{{\mathrm{act}}}_{M}(\xi)(m)=\xi m$ , where $d\operatorname{{\mathrm{act}}}_{M}$ is the differential of $\operatorname{{\mathrm{act}}}_{M}$ . Harish-Chandra ${\mathscr{A}}$ -modules form an abelian category $\mathcal{HC}_{{\mathscr{A}}}$ .

Given a Hamiltonian $G$ -action on $\mathscr{A}$ , one can form a Hamiltonian $G$ -action on the tensor product $\mathscr{A}\otimes_{k[\hbar]}\mathscr{A}_{\text{op}}$ from the ring map $U_{\hbar}\mathfrak{g}\to\mathscr{A}$ : the $G$ -action is given by restriction along the group homomorphism $G\to G\times G_{\text{op}}$ , $g\mapsto(g,g^{-1})$ , and the ring map $U_{\hbar}\mathfrak{g}\to\mathscr{A}\otimes_{k[\hbar]}\mathscr{A}_{\text{op}}$ is the map induced from the ‘diagonal’ Lie algebra imbedding $\mathfrak{g}\,\hookrightarrow\,\mathfrak{g}\oplus\mathfrak{g}_{\text{op}},\,% \xi\mapsto(\xi,-\xi)$ and the inclusion of $\mathfrak{g}$ , resp. $\mathfrak{g}_{\text{op}}$ , into $\mathscr{A}$ , resp. $\mathscr{A}_{\text{op}}$ . Harish-Chandra $\mathscr{A}\otimes_{k[\hbar]}\mathscr{A}_{\text{op}}$ -modules are in particular $(\mathscr{A},\mathscr{A})$ -bimodules, to be referred to as Harish-Chandra $\mathscr{A}$ -bimodules. Let $\mathcal{HC}_{{\mathscr{A}}}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ denote the category of Harish-Chandra $\mathscr{A}$ -bimodules.

We will be mostly interested in the case $\mathcal{A}=U_{\hbar}\mathfrak{g}$ (with Hamiltonian $G$ -action induced by the adjoint action of $G$ and the identity comoment map) and so we will use simplified notation $\mathcal{HC}_{\hbar}:=\mathcal{HC}_{U_{\hbar}\mathfrak{g}\otimes_{{k[{\hbar}]}% }U_{\hbar}\mathfrak{g}_{\text{op}}}$ . The tensor product of $U_{\hbar}\mathfrak{g}$ -bimodules equips $\mathcal{HC}_{\hbar}$ with monoidal structure $(\text{-})\otimes_{U_{\hbar}\mathfrak{g}}(\text{-})$ . One has a monoidal functor $\text{Fr}_{\hbar}:\text{Rep}(G\times\mathbb{G}_{m})\to\mathcal{HC}_{\hbar}$ given by tensor product $V\mapsto U_{\hbar}\mathfrak{g}\otimes V$ of graded vector spaces.

Specializing $\hbar$ at $1\in k$ gives rise to similar definitions; in particular one similarly obtains the monoidal category $\mathcal{HC}$ of (nonasymptotic) Harish-Chandra $(U\mathfrak{g},U\mathfrak{g})$ -bimodules.

2.3.

Let $\mathcal{D}_{X}$ denote the sheaf of differential operators on a smooth variety $X$ and $D(X)=\Gamma(X,\mathcal{D}_{X})$ the algebra of differential operators on $X$ . We let $G\times G_{\text{op}}$ act on $G$ by the formula $(g_{1},g_{2})g=g_{1}gg_{2}$ and identify $U\mathfrak{g}$ , resp. $U\mathfrak{g}_{\text{op}}$ , with the subalgebra of $D(G)$ of $G$ -invariant, resp. $G_{\text{op}}$ -invariant, differential operators. The algebra $U\mathfrak{g}\cap U\mathfrak{g}_{\text{op}}$ of bi-invariant differential operators equals $Z\mathfrak{g}$ , the center of $U\mathfrak{g}$ . Furthermore, the resulting algebra map

(2.3.1)

U\mathfrak{g}\otimes_{Z\mathfrak{g}}U\mathfrak{g}_{\text{op}}\to{D}(G)

is known to be injective. Using the algebra map 2.3.1 we may (and will) view a ${D}(G)$ -module as a $({U}\mathfrak{g},U\mathfrak{g})$ -bimodule.

Given a $G$ -action $\alpha:G\times X\to X$ one defines a convolution operation $\mathcal{M},\mathcal{M}^{\prime}\mapsto\mathcal{M}\star_{{}_{\mathcal{D}}}% \mathcal{M}^{\prime}:=\int_{\alpha}\mathcal{M}\boxtimes\mathcal{M}^{\prime}$ , on the derived category $\mathscr{D}(G)$ of $D$ -modules for $G$ . It is a straightforward consequence of the definition of $\int_{\alpha}$ and the fact that the morphism $\alpha$ is smooth and affine that there is a canonical isomorphism of $U\mathfrak{g}$ -modules, cf. [BezrukavnikovFinkelbergOstrikCharacterDModulesViaDrinfeldCenterofHarishChandraBimodules, Lemma 2.1],

		$\displaystyle\phi:\,R\Gamma(X,\mathcal{M}\star_{{}_{\mathcal{D}}}\mathcal{M}^{% \prime})\xrightarrow{\ \sim\ }R\Gamma(G,\mathcal{M})\,\stackrel{{{}_{L}}}{{% \otimes}}_{U\mathfrak{g}}\,R\Gamma(X,\mathcal{M}^{\prime}),{\enspace}\text{% such that}$
(2.3.2)			$\displaystyle\phi(fm)=f_{(0)}m_{(0)}\otimes f_{(1)}m_{(1)},\quad\forall f\in% \mathcal{O}(X),\ m\in R\Gamma(X,\mathcal{M}\star_{{}_{\mathcal{D}}}\mathcal{M}% ^{\prime}).$

Here, we have used Sweedler notation to write the map $\alpha^{*}:\mathcal{O}(X)\to\mathcal{O}(G\times X)=\mathcal{O}(G)\otimes% \mathcal{O}(X)$ as $f\mapsto f_{(0)}\otimes f_{(1)}$ , resp. $\phi$ as $m\mapsto m_{(0)}\otimes m_{(1)}$ , and we use the ring morphism 2.3.1 to view $R\Gamma(G,\mathcal{M})$ as a $U\mathfrak{g}$ -bimodule (so that in particular $R\Gamma(G,\mathcal{M})$ is a right $U\mathfrak{g}$ -module by left multiplication by the subring $U\mathfrak{g}_{\text{op}}\subseteq D(G)$ ) and we use the moment map $U\mathfrak{g}\to D(X)$ induced by the $G$ action on $X$ to view $R\Gamma(X,\mathcal{M}^{\prime})$ as a left $U\mathfrak{g}$ -module. For the ease of exposition, hereafter we often will not distinguish between a $\mathcal{D}$ -module on an affine scheme and its global sections.

In the special case where $G$ acts on itself by left translations, so $\alpha=m$ is the multiplication map, the convolution operation gives the derived category of $\mathcal{D}_{G}$ -modules a monoidal structure. Since $G$ is affine, the latter category is equivalent to the derived category of $D(G)$ -modules via the functor $\Gamma(G,\text{-})$ and formula (2.3) shows that convolution of $\mathcal{D}_{G}$ -modules goes via the equivalence to the derived tensor product of $U\mathfrak{g}$ -bimodules. The tensor product being left exact, the abelian category ${D}(G)\operatorname{{\text{-}{\mathrm{mod}}}}$ acquires a monoidal structure $(\text{-})\star(\text{-})=(\text{-})\otimes_{U\mathfrak{g}}(\text{-})$ . We will use simplified notation $D=D(G)$ and define the transfer $D$ -module as the convolution

(2.3.3)

{D_{\lhd}}:={D}\star{D}=D\otimes_{U\mathfrak{g}}D,

of two copies of the rank one free left ${D}$ -module $D$ . Thus, ${D_{\lhd}}$ is a left ${D}$ -module and the action of $D_{\text{op}}$ on each of the two copies of ${D}$ by right multiplicaton induces, by functoriality, a $(D\otimes D)_{\text{op}}$ -action on ${D_{\lhd}}$ by ${D}$ -module endomorphisms. Thus, ${D_{\lhd}}$ acquires the structure of a $({D},{D}\otimes{D})$ -bimodule. The convolution in ${D}\operatorname{{\text{-}{\mathrm{mod}}}}$ is given in terms of this object by the formula

M\star M^{\prime}={D_{\lhd}}\otimes_{{}_{{D}\otimes{D}}}\,(M\otimes M^{\prime}% ),\quad M,M^{\prime}\in{D}\operatorname{{\text{-}{\mathrm{mod}}}},

where $M\otimes M^{\prime}$ is viewed as a left ${D}\otimes{D}$ -module.

Remark 2.3.4.

The definition of ${D_{\lhd}}$ agrees with the corresponding notion in the theory of $D$ -modules, cf. [HTT]. Specifically, let $K_{X}$ denote the dualizing sheaf of a smooth variety $X$ . Then, using bi-invariant trivializations $K_{G}\cong{\mathcal{O}}_{G}$ , resp. $K_{G\times G}\cong{\mathcal{O}}_{G\times G}$ , one can check that there is a natural isomorphism of $(D,D\otimes D)$ -bimodules

{D_{\lhd}}\cong\Gamma\big{(}G\times G,\,K_{G\times G}\otimes_{\mathcal{O}_{G% \times G}}\,m^{*}(K^{\otimes-1}_{G}\otimes_{\mathcal{O}_{G}}\mathcal{D}_{G})% \big{)}.

The algebra ${D_{\hbar}}$ of asymptotic differential operators on $G$ is defined as the Rees algebra associated with the filtration on on $D$ by the order of the differential operator. The order filtration induces the tensor product filtration on ${D_{\lhd}}=D\otimes_{U\mathfrak{g}}D$ , and we let ${D_{\lhd,{\hbar}}}={D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}{D_{\hbar}}$ denote the corresponding Rees object. We define convolution of (graded) ${D_{\hbar}}$ -modules (to be denoted by $\star$ again) by the formula $M\star M^{\prime}={D_{\lhd,{\hbar}}}\bigotimes_{{{D_{\hbar}}}\otimes{{D_{\hbar% }}}}(M\otimes M^{\prime})$ .

The left ${{D_{\hbar}}}$ -action on ${D_{\lhd,{\hbar}}}$ gives an algebra map

(2.3.5)

{{D_{\hbar}}}\to\operatorname{End}_{({D_{\hbar}}\otimes{D_{\hbar}})_{\text{op}% }}({D_{\lhd,{\hbar}}})=\operatorname{End}_{({D_{\hbar}}\otimes{D_{\hbar}})_{% \text{op}}}({{D_{\hbar}}}\otimes_{U_{\hbar}\mathfrak{g}}{{D_{\hbar}}}).

In particular, the left action on the element $1\otimes 1\in{{D_{\lhd,{\hbar}}}}$ gives a map $\Delta:{{D_{\hbar}}}\to{D_{\lhd,{\hbar}}}={{D_{\hbar}}}\otimes_{{U_{\hbar}% \mathfrak{g}}}{{D_{\hbar}}}$ . It is easy to check that $\Delta$ is the unique left $U_{\hbar}\mathfrak{g}$ -module map that restricts to the standard comultiplication map on the subalgebra $\mathcal{O}(G)\subset{D_{\hbar}}$ . Also, the associativity of the convolution implies that $\Delta$ is coassociative in the sense that one has an equality $({\operatorname{Id}}\star\Delta)\Delta=(\Delta\star{\operatorname{Id}})\Delta$ of maps ${D_{\hbar}}\to{D_{\hbar}}\star{D_{\hbar}}\star{D_{\hbar}}$ .

Let $\boldsymbol{1}_{{D}\operatorname{{\text{-}{\mathrm{mod}}}}}$ be the simple skyscraper ${D}$ -module supported at the unit element of $G$ . The restriction of the ${D}$ -action to the subalgebra $U\mathfrak{g}\otimes_{Z\mathfrak{g}}U\mathfrak{g}_{\text{op}}$ identifies $\boldsymbol{1}_{{D}\operatorname{{\text{-}{\mathrm{mod}}}}}$ with the diagonal $(U\mathfrak{g},U\mathfrak{g})$ -bimodule $U\mathfrak{g}$ . The ${D}$ -module $\boldsymbol{1}_{{D}\operatorname{{\text{-}{\mathrm{mod}}}}}$ comes equipped with a natural filtration. The Rees object $\boldsymbol{1}_{{{D_{\hbar}}}\operatorname{{\text{-}{\mathrm{mod}}}}}$ associated with this filtration is a graded ${D_{\hbar}}$ -module. As a $(U_{\hbar}\mathfrak{g},U_{\hbar}\mathfrak{g})$ -bimodule, $\boldsymbol{1}_{{{D_{\hbar}}}\operatorname{{\text{-}{\mathrm{mod}}}}}$ may be identified with the diagonal bimodule ${U_{\hbar}\mathfrak{g}}$ and we write $\delta$ for the element $1\in{U_{\hbar}\mathfrak{g}}$ viewed as an element of $\boldsymbol{1}_{{{D_{\hbar}}}\operatorname{{\text{-}{\mathrm{mod}}}}}$ . We see that $\boldsymbol{1}_{{D}\operatorname{{\text{-}{\mathrm{mod}}}}}$ is the monoidal unit of ${D_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}}$ with respect to convolution. Therefore, the forgetful functor ${D_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}}\to(U_{\hbar}\mathfrak{g},U_% {\hbar}\mathfrak{g})\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ is monoidal.

2.4.

We will use the following terminology

Definition 2.4.1.

Let $R$ be an associative ring.

${{\enspace}\sf{(i)}}\;$ An $R$ -ring object is a ring $A$ equipped with a ring morphism $R\to A$ whose image is not necessarily central.

${\;\sf{(ii)}}\;$ An $R$ -coring object is a coalgebra object $A$ in the monoidal category of $R$ -bimodules; in particular, $A$ is an $R$ -bimodule.

${\sf{(iii)}}\;$ If $R$ is commutative then an $R$ -bimodule $M$ is said to be symmetric if $rm=mr$ for all $r\in R,m\in M$ . A ring, resp. coring, object $A$ over such an $R$ is called an $R$ -algebra, resp. $R$ -coalgebra, if it is symmetric as an $R$ -bimodule.

${\sf{(iv)}}\;$ A left comodule over an $R$ -coring $A$ is the data of a left $R$ -module $C$ along with a coaction map $\operatorname{coact}:C\to A\otimes_{R}C$ of $R$ -modules for which $(\Delta\otimes\text{id}_{M})\circ\operatorname{coact}=(\text{id}_{A}\otimes% \operatorname{coact})\circ\operatorname{coact}$ and $(\text{id}_{A}\otimes\epsilon)\circ\operatorname{coact}=\text{id}_{C}$ , where $\Delta$ , resp. $\epsilon$ , is the comultiplication, resp. counit of $A$ .

We use Sweedler notation and write the coaction map as $\operatorname{coact}(c)=c_{(0)}\otimes c_{(1)}$ .

If $R$ in 2.4.1 is graded, we will also say a graded $R$ -module $A$ is a graded ring object, resp. graded coring object, to indicate that $A$ is $R$ -ring object, resp. $R$ -coring object, the unit and multiplication map, resp. counit and comultiplication map, are graded. Analogously to our conventions in Section 2.1, for any graded coring object $A$ we let $A\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ denote the category of graded comodules, i.e. graded $R$ -modules $C$ such that the coaction map respects the grading.

We now take $R=U_{\hbar}\mathfrak{g}$ and view $A={D_{\hbar}}$ as an $(U_{\hbar}\mathfrak{g}\otimes U_{\hbar}\mathfrak{g}_{\text{op}})$ -algebra object via the imbedding (2.3.1).

Proposition 2.4.2.

${{\enspace}\sf{(i)}}\;$ The map $\Delta$ and the map $\epsilon:{{D_{\hbar}}}\to U_{\hbar}\mathfrak{g}$ given by the action of ${{D_{\hbar}}}$ on the generator $\delta\in\boldsymbol{1}_{{{D_{\hbar}}}\operatorname{{\text{-}{\mathrm{mod}}}}}$ equip ${{D_{\hbar}}}$ with the structure of a (graded) coalgebra object in the monoidal category $U_{\hbar}\mathfrak{g}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ .

${\;\sf{(ii)}}\;$ For any ${{D_{\hbar}}}$ -comodule $C$ , the map $C\otimes{U_{\hbar}\mathfrak{g}}\to C,\,(c,u)\mapsto cu:=\epsilon(c_{(0)}u)c_{(% 1)}$ , gives a right $U_{\hbar}\mathfrak{g}$ -action on $C$ that commutes with the left ${U_{\hbar}\mathfrak{g}}$ -action on $C$ ; thus, $C$ acquires the structure of an $(U_{\hbar}\mathfrak{g},U_{\hbar}\mathfrak{g})$ -bimodule.

${\sf{(iii)}}\;$ For any ${{D_{\hbar}}}$ -comodules $C,C^{\prime}$ , the map

C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime}\to{D_{\hbar}}\,\otimes_{U_{\hbar}% \mathfrak{g}}\,(C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime}),\quad c\otimes c^{% \prime}\mapsto c_{(0)}c^{\prime}_{(0)}\otimes c_{(1)}\otimes c^{\prime}_{(1)},

gives the tensor product $C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime}$ of $U_{\hbar}\mathfrak{g}$ -bimodules the structure of a graded ${{D_{\hbar}}}$ -comodule. The operation $(\text{-})\otimes_{U_{\hbar}\mathfrak{g}}(\text{-})$ so defined equips the category ${{D_{\hbar}}}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ with a monoidal structure.

${\sf{(iv)}}\;$ The forgetful functor upgrades to a monoidal equivalence of categories ${D_{\hbar}}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}\xrightarrow{\sim}% \mathcal{HC}_{\hbar}$ .

Remark 2.4.3.

One has a diagram $\mathfrak{g}^{*}\xleftarrow{\mu}T^{*}G\xrightarrow{\mu_{\text{op}}}\mathfrak{g% }^{*}_{\text{op}}$ , of moment maps for the Hamiltonian actions on $T^{*}G$ of $G$ and $G_{\text{op}}$ , respectively. Using this diagram and the identification $\mathfrak{g}^{*}_{\text{op}}\cong\mathfrak{g}^{*}$ of vector spaces, one can give $T^{*}G$ the structure of a groupoid scheme in the sense of [StacksProject, Definition 39.13.1]. 2.4.2(i) can be viewed as a quantization of this structure.

Proof of 2.4.2.

The first statement (i) is a straighforward computation. Moreover, one can compute that, for any $\partial\in{D_{\hbar}},\,u\in{U_{\hbar}\mathfrak{g}},\,u^{\prime}\in{U_{\hbar}% \mathfrak{g}}_{\text{op}}$ , we have $\epsilon(uu^{\prime}\partial)=(uu^{\prime}\partial)1=uu^{\prime}\epsilon(% \partial)1=u\epsilon(\partial)u^{\prime}$ . This computation, along with the fact that $\Delta$ factors through the Takeuchi product¹¹1We first learned of this interpretation of the Takeuchi product in the MathOverflow answer [GunninghamWhyShouldTensorofDxModulesOverOXBeDxModule], which we believe is a useful introduction to the structure of bialgebroids for representation theorists. (see (2.3.5)) implies that the algebra and the coalgebra object structure on ${D_{\hbar}}$ satisfy the axioms of a left ${U_{\hbar}\mathfrak{g}}$ -bialgebroid in the sense of [BohmHopfAlgebroids, Definition 3.3] with coproduct $\Delta$ and counit $\epsilon$ ; this fact also follows from [BohmHopfAlgebroids, Theorem 3.13]. Statements (ii) and (iii) then follow from [BohmHopfAlgebroids, Theorem 3.18]. Moreover, using the isomorphism $U_{\hbar}\mathfrak{g}_{\text{op}}\otimes\mathcal{O}(G)\xrightarrow{\sim}{D_{% \hbar}}$ induced by multiplication, one readily checks that any ${D_{\hbar}}$ -comodule in particular has an $\mathcal{O}(G)$ -comodule structure which is compatible with the $U_{\hbar}\mathfrak{g}$ -module structure and, from this, it is not difficult to check (iv). ∎

3. Kostant-Whittaker Reduction

From now on, we fix a Borel subgroup $B\subseteq G$ with Lie algebra ${\mathfrak{b}}$ and a nondegenerate character $\psi:\mathfrak{n}\to k$ , where $\mathfrak{n}$ is the Lie algebra of $N$ , the unipotent radical of $B$ . We also choose an opposite Borel subgroup $\bar{B}$ , so $T:=\bar{B}\cap B$ is a maximal torus, and let $\overline{\mathfrak{b}},\mathfrak{t}$ denote their Lie algebras.

3.1. Classical Kostant-Whittaker reduction

We identify $\mathfrak{g}^{*}$ with $\mathfrak{g}$ via the Killing form $(\text{-},\text{-})$ . Choose a $\mathfrak{sl}_{2}$ -triple $e,h,f\in\mathfrak{g}$ such that $h$ lies in $\mathfrak{t}$ and $(e,\text{-})|_{{\mathfrak{n}}}=\psi$ . Let ${\mathfrak{s}}\subseteq\mathfrak{g}^{*}$ be the image of the Kostant slice through $e$ associated with this triple under the isomorphism $\mathfrak{g}{\;\stackrel{{{}_{\sim}}}{{\to}}\;}\mathfrak{g}^{*},\,\xi\mapsto(% \xi,\text{-})$ . Then, ${\mathfrak{s}}$ is contained $(\psi+\mathfrak{n}^{\perp})\cap\mathfrak{g}^{*}_{\text{reg}}$ . Let $Z_{\mathfrak{s}}\to{\mathfrak{s}}$ be the restriction to ${\mathfrak{s}}$ of the group scheme $Z_{\text{reg}}\to\mathfrak{g}_{\text{reg}}$ of regular centralizers.

There is a unique cocharacter $\gamma:\mathbb{G}_{m}\to T$ such that $\gamma(z)(\psi)=z^{-2}\cdot\psi,\,\forall z\in\mathbb{G}_{m}$ . We define a ‘twisted’ action of ${{\mathbb{G}}_{m}}$ on $G$ , resp. $\mathfrak{g}^{*}$ and ${\mathfrak{n}}^{*}$ , by $z:g\mapsto\gamma(z)g\gamma(z)^{-1}$ , resp. $z:\xi\mapsto z^{2}\cdot\operatorname{{\mathrm{Ad}}}\gamma(z)(\xi)$ . Then, $\psi\in{\mathfrak{n}}^{*}$ is a ${{\mathbb{G}}_{m}}$ -fixed point, the slice ${\mathfrak{s}}$ is ${{\mathbb{G}}_{m}}$ -stable and $Z_{\mathfrak{s}}\subseteq G\times\mathfrak{g}^{*}$ is stable under the diagonal ${{\mathbb{G}}_{m}}$ -action. We equip ${\mathfrak{c}}$ with an action of ${{\mathbb{G}}_{m}}\ltimes Z_{\mathfrak{s}}$ where $Z_{\mathfrak{s}}$ acts trivially on ${\mathfrak{c}}$ .

Let ${\text{QCoh}}^{{{\mathbb{G}}_{m}}\times G}(\mathfrak{g}^{*})$ , resp. ${\text{QCoh}}^{{{\mathbb{G}}_{m}}\ltimes Z_{\mathfrak{s}}}({\mathfrak{s}})$ , be the abelian category of equivariant quasi-coherent sheaves where the factor ${{\mathbb{G}}_{m}}$ acts on $\mathfrak{g}^{*}$ by dilations, resp. ${{\mathbb{G}}_{m}}\ltimes Z_{\mathfrak{s}}$ acts on ${\mathfrak{c}}$ as defined above. Tensor product of $\mathcal{O}$ -modules gives these categories a monoidal structure. One has a well defined restriction functor $i^{*}:{\text{QCoh}}^{{{\mathbb{G}}_{m}}\times G}(\mathfrak{g}^{*})\to{\text{% QCoh}}^{\mathbb{G}_{m}\ltimes Z_{\mathfrak{s}}}({\mathfrak{s}})$ , even though the closed imbedding is not ${{\mathbb{G}}_{m}}$ -equivariant. Furthermore, this functor is exact and monoidal.

There is an alternate, somewhat more canonical construction of the functor $i^{*}$ in terms of ‘classical Kostant-Whittaker reduction’ that we are going to quantize in subsequent subsections. To explain the construction, let $T^{*}_{\psi}(G/N)=G\times_{N}(e+{\mathfrak{n}}^{\perp})$ be the $\psi$ -twisted cotangent bundle on $G/N$ . We define a ${{\mathbb{G}}_{m}}\times G$ -action on $T^{*}_{\psi}(G/N)$ by letting ${{\mathbb{G}}_{m}}$ , resp. $G$ , act by $z:(g,\xi)\mapsto(g\gamma(z),\,z^{2}\cdot\operatorname{{\mathrm{Ad}}}\gamma(z)^% {-1}(\xi))$ , resp. $g^{\prime}:(g,\xi)\mapsto(g^{\prime}g,\xi)$ . We have a commutative diagram

Here, $\varpi$ is the coadjoint quotient morphism, $\jmath:\mathfrak{g}_{\text{reg}}\,\hookrightarrow\,\mathfrak{g}$ is the open imbedding, $q$ sends $(g,\xi)$ to $\operatorname{{\mathrm{Ad}}}g(\xi)$ , and $p=\varpi{{}_{{}^{\,{}^{\circ}}}}\jmath{{}_{{}^{\,{}^{\circ}}}}q$ . Observe that the map $p$ is a $G$ -torsor (see also [GinzburgKazhdanDifferentialOperatorsOnBasicAffineSpaceandtheGelfandGraevAction, Lemma 3.2.3(iii)]) and $q$ is a $\pi^{*}J$ -torsor, where $J\to{\mathfrak{c}}$ is the universal centralizer.

For any ${{\mathcal{F}}}\in{\text{QCoh}}^{{{\mathbb{G}}_{m}}\times G}(\mathfrak{g}^{*}_% {\text{reg}})$ and $\xi\in\mathfrak{g}^{*}_{\text{reg}}$ the stabilizer of $\xi$ in $G$ acts in the fiber of ${\mathcal{F}}$ at $\xi$ . This upgrades to an action $p^{*}J\otimes q^{*}{\mathcal{F}}\to q^{*}{\mathcal{F}}$ , giving the sheaf $q^{*}{{\mathcal{F}}}$ a canonical $G\times({{\mathbb{G}}_{m}}\ltimes p^{*}J)$ -equivariant structure, where ${{\mathbb{G}}_{m}}$ acts by the twisted action. It is clear that one has the following mutually inverse monoidal equivalences:

Corollary 3.1.1.

The functor $\kappa_{0}:=p_{*}(\text{-})^{G}{{}_{{}^{\,{}^{\circ}}}}q^{*}{{}_{{}^{\,{}^{% \circ}}}}\jmath^{*}$ is an exact monoidal functor and $\jmath_{*}{{}_{{}^{\,{}^{\circ}}}}(q_{*})^{Z_{\text{reg}}}{{}_{{}^{\,{}^{\circ% }}}}p^{*}$ is a right adjoint of $\kappa_{0}$ . Furthermore, the functor ${\text{QCoh}}^{{{\mathbb{G}}_{m}}\times G}(\mathfrak{g}^{*})/\ker(\kappa_{0}){% \;\stackrel{{{}_{\sim}}}{{\to}}\;}{\text{QCoh}}^{{{\mathbb{G}}_{m}}\ltimes J}(% {\mathfrak{c}})$ induced by $\kappa_{0}$ is an equivalence and the functor $i^{*}$ corresponds to $\kappa_{0}$ under the isomorphism $\varpi{{}_{{}^{\,{}^{\circ}}}}i:{\mathfrak{s}}{\;\stackrel{{{}_{\sim}}}{{\to}}% \;}{\mathfrak{c}}$ .∎

3.2. Kostant-Whittaker reduction

Let ${\mathfrak{n}}^{\psi}$ be the image of the map ${\mathfrak{n}}\to{U_{\hbar}\mathfrak{g}},\,\xi\mapsto\xi-\psi(\xi)$ . Thus, ${\mathfrak{n}}^{\psi}$ is a Lie subalgebra of ${U_{\hbar}\mathfrak{g}}$ .

For any graded $k[{\hbar}]$ -algebra ${\mathscr{A}}$ equipped with a Hamiltonian $G$ -action with comoment map $\mu$ , let ${\mathscr{A}}{\mathfrak{n}}^{\psi}$ , resp. ${\mathfrak{n}}^{\psi}{\mathscr{A}}$ , denote the left, resp. right, ideal of ${\mathscr{A}}$ generated by the image of the composite ${\mathfrak{n}}^{\psi}\,\hookrightarrow\,{U_{\hbar}\mathfrak{g}}\xrightarrow{% \mu}{\mathscr{A}}$ . These ideals are stable under the action of the subgroup $N\subseteq G$ . Thus, $E^{\psi}_{\ell}={\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi}$ , resp. $E^{\psi}_{r}={\mathscr{A}}/{\mathfrak{n}}^{\psi}{\mathscr{A}}$ , acquires the structure of a left, resp. right, Harish-Chandra ${\mathscr{A}}$ -module (for the group $N$ ) and there are natural isomorphisms

(3.2.1)

(\operatorname{End}_{\mathcal{HC}_{{\mathscr{A}}}}E^{\psi}_{\ell})_{\text{op}}% \cong({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N},\enspace\text{resp% . }\enspace\operatorname{End}_{{\mathcal{HC}_{{{\mathscr{A}}}_{op}}}}E^{\psi}_% {r}\cong({\mathscr{A}}/{\mathfrak{n}}^{\psi}{\mathscr{A}})^{N}.

Composition of endomorphisms provides, via the above isomorphisms, algebra structures on the objects appearing on the right hand side of the isomorphisms; moreover, for any left, resp. right, Harish-Chandra ${\mathscr{A}}$ -module $M$ , it gives the coinvariants

(3.2.2)

M/M{\mathfrak{n}}^{\psi}\cong M\otimes_{{\mathscr{A}}}E^{\psi}_{\ell},\enspace% \text{resp. }\enspace M/{\mathfrak{n}}^{\psi}M\cong E^{\psi}_{r}\otimes_{% \mathscr{A}}M

the natural structure of a right module over $({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}$ , respectively left module for $({\mathscr{A}}/{\mathfrak{n}}^{\psi}{\mathscr{A}})^{N}$ .

Similar considerations also show that if $M$ is a Harish-Chandra ${\mathscr{A}}$ -bimodule, see Section 2.2, then the space

\varkappa_{\hbar}(M):=(M/M{\mathfrak{n}}^{\psi})^{N}=\operatorname{\mathrm{Hom% }}_{\mathcal{HC}_{{\mathscr{A}}}}(E^{\psi}_{\ell},M\otimes_{\mathscr{A}}E^{% \psi}_{\ell})

has the structure of a left module for the ring $({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}\otimes_{{k[{\hbar}]}}({% \mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}_{\text{op}}$ . Observe that if $M={\mathscr{A}}$ is the diagonal bimodule for ${\mathscr{A}}$ then $\varkappa_{\hbar}({\mathscr{A}})$ is the diagonal bimodule for $({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}$ , so we will identify $\varkappa_{\hbar}({\mathscr{A}})$ with the algebra $({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}$ .

The basic case we will be interested in is the case ${\mathscr{A}}={U_{\hbar}\mathfrak{g}}$ with Hamiltonian $G$ -action induced by the adjoint action, see Section 2.2. For $k\in{\mathbb{Z}}$ and an asymptotic Harish-Chandra ${U_{\hbar}\mathfrak{g}}$ -bimodule $M$ let $M^{(k)}=\{m\in M\mid\gamma(z)(m)=z^{k}\cdot m\ \forall z\in\mathbb{G}_{m}\}$ be the corresponding weight space with respect to $\gamma$ . Write $M=\oplus_{m}\,M_{m}$ for the grading on $M$ . The ‘Kazhdan grading’ on $M$ is the grading:

M=\oplus_{n\in{\mathbb{Z}}}\ M_{[n]},\enspace\text{where}\enspace M_{[n]}:=% \oplus_{\{m,k\,\mid\,n=m+k\}}\ M^{(k)}\cap M_{m}.

In particular, one has the Kazhdan grading on $U_{\hbar}\mathfrak{g}$ and ${\mathfrak{n}}^{\psi}$ is a graded Lie subalgebra of $U_{\hbar}\mathfrak{g}$ .

More generally, for an arbitrary ${\mathscr{A}}$ as above, any Harish-Chandra ${\mathscr{A}}$ -bimodule $M$ may be viewed as a Harish-Chandra ${U_{\hbar}\mathfrak{g}}$ -bimodule via the comoment map. Therefore, one has the Kazhdan grading on $M$ and $M{\mathfrak{n}}^{\psi}$ is a graded subspace of $M$ . Thus, the Kazhdan grading on $M$ induces a ${\mathbb{Z}}$ -grading on $\varkappa_{\hbar}(M)$ .

The Kazhdan grading and the standard (doubled) grading on ${U_{\hbar}\mathfrak{g}}$ induce the same grading on ${Z_{\hbar}\mathfrak{g}}$ and the composite ${Z_{\hbar}\mathfrak{g}}\,\hookrightarrow\,{U_{\hbar}\mathfrak{g}}\to{\mathscr{% A}}$ induces an algebra map ${Z_{\hbar}\mathfrak{g}}\to\varkappa_{\hbar}({\mathscr{A}})$ that respects each of these gradings. In the special case ${\mathscr{A}}={U_{\hbar}\mathfrak{g}}$ this map is an algebra isomorphism

(3.2.3)

{Z_{\hbar}\mathfrak{g}}\xrightarrow{\sim}\varkappa_{\hbar}({U_{\hbar}\mathfrak% {g}})=(U_{\hbar}\mathfrak{g}/U_{\hbar}\mathfrak{g}\cdot{\mathfrak{n}}^{\psi})^% {N}

by a well known result of Kostant [KostantonWhittakerVectorsandRepresentationTheory].

The following result is essentially known, cf. [GinzburgHarishChandraBimodulesforQuantizedSlodowySlices, Theorem 4.14], [BezrukavnikovFinkelbergEquivariantSatakeCategoryandKostantWhittakerReduction, Section 4]. We provide a proof of (i) since we have been unable to find the explicit computation in the proof below, which is relevant to the present paper, in the literature.

Proposition 3.2.4.

${{\enspace}\sf{(i)}}\;$ The assignment $M\mapsto\varkappa_{\hbar}(M)$ upgrades to a monoidal functor $\varkappa_{{\hbar}}:\,\mathcal{HC}_{{\mathscr{A}}}\operatorname{{\mathrm{\text% {-}\mathrm{bimod}}}}\to\varkappa_{\hbar}({\mathscr{A}})\operatorname{{\mathrm{% \text{-}\mathrm{bimod}}}}$ , where the monoidal structure:

\varkappa_{\hbar}(M)\otimes_{\varkappa_{{\hbar}}({\mathscr{A}})}\varkappa_{% \hbar}(M^{\prime})\xrightarrow{\sim}\varkappa_{\hbar}(M\otimes_{{\mathscr{A}}}% M^{\prime}),\quad M,M^{\prime}\in\mathcal{HC}_{{\mathscr{A}}}\operatorname{{% \mathrm{\text{-}\mathrm{bimod}}}},

is induced by the identity map $M\otimes M^{\prime}\to M\otimes M^{\prime}$ .

${\;\sf{(ii)}}\;$ In the special case ${\mathscr{A}}={U_{\hbar}\mathfrak{g}}$ (where $\varkappa_{\hbar}({U_{\hbar}\mathfrak{g}})={Z_{\hbar}\mathfrak{g}}$ ), the functor $\varkappa_{\hbar}:\mathcal{HC}_{\hbar}\to{Z_{\hbar}\mathfrak{g}}\operatorname{% {\mathrm{\text{-}\mathrm{bimod}}}}$ is exact.

Proof.

To prove (i) put $\bar{\mathscr{A}}:=({\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi})^{N}$ and $E:=E^{\psi}_{\ell}={\mathscr{A}}/{\mathscr{A}}{\mathfrak{n}}^{\psi}$ . Observe that for $M\in\mathcal{HC}_{{\mathscr{A}}}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ the left action of ${\mathfrak{n}}^{\psi}$ on $M/M{\mathfrak{n}}^{\psi}$ is locally nilpotent. It follows that the natural map $E\otimes_{\bar{\mathscr{A}}}\varkappa_{\hbar}(M)\to M\otimes_{{\mathscr{A}}}E$ is an isomorphism, by a version of Skryabin’s isomorphism, [Gin, Proposition 3.1.4(i)]. We deduce a chain of isomorphisms

	$\displaystyle\big{(}E\otimes_{\bar{\mathscr{A}}}\varkappa_{\hbar}(M)\big{)}% \otimes_{\bar{\mathscr{A}}}\varkappa_{\hbar}(M^{\prime})$	$\displaystyle\xrightarrow{\sim}\ (M\otimes_{{\mathscr{A}}}E)\ \otimes_{\bar{% \mathscr{A}}}\varkappa_{\hbar}(M^{\prime})$
		$\displaystyle\xrightarrow{\sim}M\otimes_{{\mathscr{A}}}\big{(}E\otimes_{\bar{% \mathscr{A}}}\varkappa_{\hbar}(M^{\prime})\big{)}\xrightarrow{\sim}\ (M\otimes% _{{\mathscr{A}}}M^{\prime})\otimes_{{\mathscr{A}}}E.$

The isomorphism of the proposition is given by applying $(\text{-})^{N}$ to the composite of the above isomorphisms using that the functor $(E\otimes_{\bar{\mathscr{A}}}\text{-})^{N}$ , resp. $(\text{-}\otimes_{\mathscr{A}}E)^{N}$ , is the identity.

Part (ii) is well known, cf. eg. [GinzburgHarishChandraBimodulesforQuantizedSlodowySlices, Theorem 4.1.4(iii)]. ∎

Observe next that since $\psi([\mathfrak{n},\mathfrak{n}])=0$ , the assignment $x\mapsto\psi(x)$ also gives a character of an opposite Lie algebra $\mathfrak{n}_{\text{op}}$ , to be denoted by $\psi$ as well. Hence, the image of the map ${\mathfrak{n}}_{\text{op}}\to U_{\hbar}\mathfrak{g}_{\text{op}}$ is a Lie subalgebra which will be denoted by ${\mathfrak{n}}^{\psi}_{r}$ and the algebra $({U_{\hbar}\mathfrak{g}}_{\text{op}}/{\mathfrak{n}}^{\psi}_{r}{U_{\hbar}% \mathfrak{g}}_{\text{op}})^{N}$ is the Kostant-Whittaker reduction of ${U_{\hbar}\mathfrak{g}}_{\text{op}}$ .

We also consider the case of ${\mathscr{A}}={D_{\hbar}}$ . The action of the group $G\times G_{\text{op}}$ on ${D_{\hbar}}$ is Hamiltonian and we have ${\mathfrak{n}}^{\psi}_{\ell}:={\mathfrak{n}}^{\psi}\subset U_{\hbar}\mathfrak{% g}\subset{D_{\hbar}}$ , resp. ${\mathfrak{n}}^{\psi}_{r}\subset U_{\hbar}\mathfrak{g}_{\text{op}}\subset{D_{% \hbar}}$ . We introduce the following Kostant-Whittaker reduction of ${D_{\hbar}}$ :

D^{\psi_{\ell}}_{\hbar}:=({D_{\hbar}}/{D_{\hbar}}\mathfrak{n}^{\psi}_{\ell})^{% N},\enspace\text{resp.}\enspace D^{\psi_{r}}_{\hbar}:=({D_{\hbar}}/\mathfrak{n% }^{\psi}_{r}{D_{\hbar}})^{N_{\text{op}}}.

The residual action of the group $G_{\text{op}}$ , resp. $G$ , on $D^{\psi_{\ell}}$ , resp. $D^{\psi_{r}}$ , is Hamiltonian.

We also use completely analogous definitions and (bi)module structures in the non-asymptotic setting.

3.3. Monoidal structure on Whittaker reduction

Let $R$ be a ring and $A$ a coalgebra object in the category of $R$ -bimodules. We will use the following

Definition.

Given a left comodule $C_{\ell}$ for $A$ with coaction map $\text{coact}_{\ell}$ and $C_{r}$ is a right comodule for $R$ with coaction map $\text{coact}_{r}$ , define the cotensor product

C_{r}\otimes_{R}^{A}C_{\ell}:=\big{\{}\sum\nolimits_{i}\,c^{r}_{i}\otimes c^{% \ell}_{i}\in C_{r}\otimes_{R}C_{\ell}\mid\sum\nolimits_{i}\,\text{coact}_{r}(c% ^{r}_{i})\otimes c^{\ell}_{i}=\sum\nolimits_{i}\,c^{r}_{i}\otimes\text{coact}_% {\ell}(c^{\ell}_{i})\big{\}}.

Example.

If $H$ is an algebraic group then $A:=R\otimes{\mathcal{O}}(H)$ is a coalgebra object in the category of $R$ -bimodules. In this case, we use simplified notation

C_{r}\otimes_{R}^{H}C_{\ell}:=C_{r}\otimes_{R}^{A}C_{\ell}

since, if we use the inverse map to view $C_{r}$ as a representation of $H$ , this space equals the invariants of the diagonal $H$ -action on $C_{r}\otimes_{R}C_{\ell}$ .

One can apply Kostant-Whittaker reduction to the algebra $D^{\psi_{r}}_{\hbar}$ , resp. $D^{\psi_{\ell}}_{\hbar}$ , using the Lie subalgebra ${\mathfrak{n}}^{\psi}_{\ell}\subseteq U_{\hbar}\mathfrak{g}\subseteq D^{\psi_{% r}}_{\hbar}$ , resp. ${\mathfrak{n}}^{\psi}_{r}\subseteq U_{\hbar}\mathfrak{g}_{\text{op}}\subseteq D% ^{\psi_{\ell}}_{\hbar}$ . This way one obtains the following algebras and algebra maps:

(3.3.1)

(D^{\psi_{r}}_{\hbar}/D^{\psi_{r}}_{\hbar}\mathfrak{n}^{\psi}_{\ell})^{N}% \xrightarrow{}({D_{\hbar}}/({\mathfrak{n}}^{\psi}_{r}{D_{\hbar}}+{D_{\hbar}}{% \mathfrak{n}}^{\psi}_{\ell}))^{N\times N_{\text{op}}}\xleftarrow{}(D^{\psi_{% \ell}}_{\hbar}/\mathfrak{n}^{\psi}_{r}D^{\psi_{\ell}}_{\hbar})^{N_{\text{op}}}

The functor $(\text{-})^{N}$ is exact on the subcategory of $U_{\hbar}\mathfrak{g}$ -modules for which $\mathfrak{n}^{\psi}$ acts locally nilpotently by Skryabin’s equivalence. Since any exact functor on an abelian category commutes with all finite limits and colimits (see for example [StacksProject, Lemma 12.7.2]) we deduce that the both maps in (3.3.1) are algebra isomorphisms. We let $\mathfrak{J}_{\hbar}$ denote the resulting algebra, referred to as the algebra of (asymptotic) bi-Whittaker differential operators.

It is clear that the maps ${Z_{\hbar}\mathfrak{g}}\to(D^{\psi_{r}}_{\hbar}/D^{\psi_{r}}_{\hbar}\mathfrak{% n}^{\psi}_{\ell})^{N}$ and ${Z_{\hbar}\mathfrak{g}}\to(D^{\psi_{\ell}}_{\hbar}/\mathfrak{n}^{\psi}_{r}D^{% \psi_{\ell}}_{\hbar})^{N_{\text{op}}}$ correspond to each other via the identification given by (3.3.1), so one obtains an algebra homomorphism $u_{\mathfrak{J}}:{Z_{\hbar}\mathfrak{g}}\to{{\mathfrak{J}}_{\hbar}}$ .

The first two parts of the proposition below can be viewed as a quantization of the fact that the centralizer $J\to{\mathfrak{c}}$ admits a group scheme structure (so that the category of modules and comodules for $\mathcal{O}(J)$ acquires a monoidal structure) and the other two parts can be viewed as quantizations of the action of $G\times J_{\text{op}}$ on $T^{*}_{\psi}(G/N)$ .

Proposition 3.3.2.

${{\enspace}\sf{(i)}}\;$ There is a left $\mathfrak{J}_{\hbar}$ -module structure on the tensor product

{{\mathfrak{J}}_{\lhd,{\hbar}}}:=\mathfrak{J}_{\hbar}\otimes_{Z_{\hbar}% \mathfrak{g}}\mathfrak{J}_{\hbar}

of left $Z_{\hbar}\mathfrak{g}$ -modules which commutes with the obvious right module structure for the ring $\mathfrak{J}_{\hbar}\otimes\mathfrak{J}_{\hbar}$ . Acting on $1\otimes 1$ gives rise to a map $\Delta:\mathfrak{J}_{\hbar}\to{{\mathfrak{J}}_{\lhd,{\hbar}}}$ which upgrades to a graded $Z_{\hbar}\mathfrak{g}$ -coalgebra structure on ${{\mathfrak{J}}_{\hbar}}$ . Furthermore, the operation

E_{1},E_{2}\mapsto E_{1}\star E_{2}:={{\mathfrak{J}}_{\lhd,{\hbar}}}\otimes_{{% {\mathfrak{J}}_{\hbar}}\otimes{{\mathfrak{J}}_{\hbar}}}(E_{1}\otimes E_{2})\ (% \cong E_{1}\otimes_{Z_{\hbar}\mathfrak{g}}E_{2})

gives the category ${{\mathfrak{J}}_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}}$ of graded left ${{\mathfrak{J}}_{\hbar}}$ -modules a monoidal structure such that the forgetful functor ${{\mathfrak{J}}_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}}\to{Z_{\hbar}% \mathfrak{g}}\operatorname{{\text{-}{\mathrm{mod}}}}$ is monoidal.

${\;\sf{(ii)}}\;$ The ${Z_{\hbar}\mathfrak{g}}$ -module structure on any left comodule for the coalgebra ${{\mathfrak{J}}_{\hbar}}$ can be extended to a ${Z_{\hbar}\mathfrak{g}}$ -bimodule structure. Moreover, the category of $\mathfrak{J}_{\hbar}$ -comodules can be equipped with the canonical monoidal structure such that the forgetful functor ${{\mathfrak{J}}_{\hbar}}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}\to{Z_% {\hbar}\mathfrak{g}}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ is monoidal.

${\sf{(iii)}}\;$ Let $M$ , resp. $M^{\prime},\,M^{\prime\prime}$ , and $E$ , be a left module over ${D_{\hbar}}$ , resp. $D^{\psi_{r}}_{\hbar},\,D^{\psi_{\ell}}_{\hbar}$ , and ${{\mathfrak{J}}_{\hbar}}$ . Then, the tensor products

M\otimes_{Z_{\hbar}\mathfrak{g}}M^{\prime}\enspace\text{\em and}\enspace M^{% \prime}\otimes_{{Z_{\hbar}\mathfrak{g}}}E,\quad\text{\em resp.}\quad M^{\prime% \prime}\otimes_{Z\mathfrak{g}}M\enspace\text{\em and}\enspace E\otimes_{Z_{% \hbar}\mathfrak{g}}M^{\prime\prime},

have the canonical structure of a (left) module for $D^{\psi_{r}}$ , respectively $D^{\psi_{\ell}}$ . This equips $D^{\psi_{r}}_{\hbar}\operatorname{{\text{-}{\mathrm{mod}}}}$ , resp. $D^{\psi_{\ell}}_{\hbar}\operatorname{{\text{-}{\mathrm{mod}}}}$ , with the structure of a $({D_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}},\,{{\mathfrak{J}}_{\hbar}}% \operatorname{{\text{-}{\mathrm{mod}}}})$ -bimodule, resp. $({{\mathfrak{J}}_{\hbar}}\operatorname{{\text{-}{\mathrm{mod}}}},\,{D_{\hbar}}% \operatorname{{\text{-}{\mathrm{mod}}}})$ -bimodule, category.

${\sf{(iv)}}\;$ One can equip $D^{\psi_{r}}_{\hbar}$ , viewed as a ${U_{\hbar}\mathfrak{g}}$ -module, resp. ${Z_{\hbar}\mathfrak{g}}$ -module, by left multiplication by the subalgebra ${U_{\hbar}\mathfrak{g}}$ , resp. ${Z_{\hbar}\mathfrak{g}}$ of $D^{\psi_{r}}_{\hbar}$ , with the structure of a left ${D_{\hbar}}$ -comodule, resp. right ${{\mathfrak{J}}_{\hbar}}$ -comodule. Combined together, this gives $D^{\psi_{r}}_{\hbar}$ the structure of a $({D_{\hbar}},{{\mathfrak{J}}_{\hbar}})$ -bicomodule.

Proof.

The $({D_{\hbar}},{D_{\hbar}}\otimes_{k}{D_{\hbar}})$ -bimodule structure on ${D_{\lhd,{\hbar}}}$ induces, according to Section 3.2, a $({D_{\hbar}},{D_{\hbar}}\otimes_{k}D^{\psi_{\ell}}_{\hbar})$ -bimodule structure on ${D_{\hbar}}\otimes_{U\mathfrak{g}}({D_{\hbar}}/{D_{\hbar}}\mathfrak{n}^{\psi}_% {\ell})$ . Therefore, the invariants ${D_{\hbar}}\otimes_{U\mathfrak{g}}^{N}({D_{\hbar}}/{D_{\hbar}}\mathfrak{n}^{% \psi}_{\ell})$ for the induced diagonal $N$ -action acquires the structure of a $({D_{\hbar}},D^{\psi_{r}}_{\hbar}\otimes D^{\psi_{\ell}}_{\hbar})$ -bimodule. The monoidality of Kostant-Whittaker reduction for Harish-Chandra bimodules (3.2.4) gives a canonical isomorphism

D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}D^{\psi_{\ell}}_{\hbar}% \xrightarrow{\sim}{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}^{N}({D_{\hbar}}/{% D_{\hbar}}\mathfrak{n}^{\psi}_{\ell})

and, using this, we equip $D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}D^{\psi_{\ell}}_{\hbar}$ with the structure of a $({D_{\hbar}},D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}D^{\psi_{\ell}% }_{\hbar})$ -bimodule. Tensoring with this bimodule, we see that for $M^{\prime}$ and $M^{\prime\prime}$ , the tensor product $M^{\prime}\otimes_{Z_{\hbar}\mathfrak{g}}M^{\prime\prime}$ has the natural structure of a left ${D_{\hbar}}$ -module. Similarly, the exactness of Kostant-Whittaker reduction (for the action of $G_{\text{op}}$ on $D^{\psi_{\ell}}_{\hbar}$ ) gives isomorphisms

(3.3.3)

{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}D^{\psi_{r}}_{\hbar}\xrightarrow{% \sim}\varkappa_{\hbar}({D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}{D_{\hbar}})% \text{, resp. }D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}\mathfrak{J}% _{\hbar}\xrightarrow{\sim}(D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}% D^{\psi_{\ell}}_{\hbar}/D^{\psi_{\ell}}_{\hbar}\mathfrak{n}^{\psi}_{r})^{N_{% \text{op}}}

and so by our above analysis ${D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}D^{\psi_{r}}$ acquires a $D^{\psi_{r}}$ -module structure commuting with the obvious right ${D_{\hbar}}\otimes_{k}D^{\psi_{r}}$ -module structure, resp. a commuting $\mathfrak{J}_{\hbar}$ -module structure on $D^{\psi_{r}}\otimes_{Z\mathfrak{g}}\mathfrak{J}$ . Tensoring with these bimodules and using the obvious identifications, we obtain the first two claims of (iii); symmetric arguments give the second two claims of (iii). Similarly, the map $\partial\mapsto\partial\cdot(1\otimes 1)$ for $\partial\in D^{\psi_{r}}$ , resp. $\nu\mapsto\nu\cdot(1\otimes 1)$ for $\nu\in\mathfrak{J}_{\hbar}$ , gives the left coaction of ${D_{\hbar}}$ , resp. right coaction map of $\mathfrak{J}_{\hbar}$ . Applying similar exactness arguments to the associativity isomorphism

({D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}{D_{\hbar}})\otimes_{U_{\hbar}% \mathfrak{g}}{D_{\hbar}}\cong{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}({D_{% \hbar}}\otimes_{U_{\hbar}\mathfrak{g}}{D_{\hbar}})

shows that these two coactions are coassociative and thus form a bicomodule. Finally, applying Kostant-Whittaker reduction to the $D^{\psi_{r}}$ -module $D^{\psi_{r}}\otimes_{Z\mathfrak{g}}\mathfrak{J}$ and again using exactness we obtain a graded $\mathfrak{J}_{\hbar}$ -module structure on ${{\mathfrak{J}}_{\lhd,{\hbar}}}$ which commtues with the natural right module structure for the ring $\mathfrak{J}_{\hbar}\otimes\mathfrak{J}_{\hbar}$ . Now proceeding essentially identically as in the proof of 2.4.2, we obtain (i)-(ii); the counit of the graded coalgebra structure is given by the induced $\mathfrak{J}_{\hbar}$ -module structure on $Z_{\hbar}\mathfrak{g}\xrightarrow{\sim}\varkappa_{\hbar}(U_{\hbar}\mathfrak{g})$ from the ${D_{\hbar}}$ -module structure on $U_{\hbar}\mathfrak{g}$ . ∎

There is also a ‘nonasymptotic’ counterpart of 3.3.2 where ${\hbar}$ is specialized to $1$ . It is proved in a similar way.

4. Bi-Whittaker reduction and $\mathfrak{J}_{\hbar}$ -comodules

4.1.

The main result of this section is as follows, which informally constructs a functor $\kappa_{\hbar}$ on $\mathcal{HC}_{\hbar}$ with many similar properties to the pullback functor $\jmath^{*}$ for the open embedding $\jmath$ as in Section 3.1:

Theorem 4.1.1.

${{\enspace}\sf{(i)}}\;$ The functor $\varkappa_{\hbar}$ lifts to an exact monoidal functor $\kappa_{\hbar}:\mathcal{HC}_{\hbar}\to\mathfrak{J}_{\hbar}\operatorname{{% \mathrm{\text{-}\mathrm{comod}}}}$ to the category of graded $\mathfrak{J}_{\hbar}$ -comodules which induces a monoidal equivalence of categories

(4.1.2)

\mathcal{HC}_{\hbar}/\operatorname{ker}(\kappa_{\hbar}){\;\stackrel{{{}_{\sim}% }}{{\to}}\;}{{\mathfrak{J}}_{\hbar}}\operatorname{{\mathrm{\text{-}\mathrm{% comod}}}}.

${\;\sf{(ii)}}\;$ The assignment $C\mapsto\kappa_{\hbar}^{R}(C):=D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak% {g}}^{{{\mathfrak{J}}_{\hbar}}}C$ upgrades to a fully faithful functor

\kappa_{\hbar}^{R}:{{\mathfrak{J}}_{\hbar}}\operatorname{{\mathrm{\text{-}% \mathrm{comod}}}}\to\mathcal{HC}_{\hbar}

which is right adjoint to $\kappa_{\hbar}$ and commutes with filtered colimits.

${\sf{(iii)}}\;$ There is a natural isomorphism of functors $\kappa_{\hbar}(\text{-})\cong D^{\psi_{\ell}}_{\hbar}\otimes_{U_{\hbar}% \mathfrak{g}}^{G}(\text{-})$ .

${\sf{(iv)}}\;$ The unit $\text{Fr}_{\hbar}(V)\to\kappa_{\hbar}^{R}\kappa_{\hbar}(\text{Fr}_{\hbar}(V))$ for the adjunction $(\kappa_{\hbar},\kappa_{\hbar}^{R})$ for any (not necessarily finite dimensional) $G$ -representation $V$ is an isomorphism.

To prove 4.1.1 we first need the following

Lemma 4.1.3.

The composite functor $\varkappa_{\hbar,\ell}:\mathcal{HC}_{\hbar}\xrightarrow{\,\varkappa_{\hbar}\,}% Z_{\hbar}\mathfrak{g}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}% \xrightarrow{\,\operatorname{Obl}_{r}\,}Z_{\hbar}\mathfrak{g}\operatorname{{% \text{-}{\mathrm{mod}}}}$ admits a right adjoint $\varkappa_{\hbar,\ell}^{R}(\text{-}):=D^{\psi_{r}}_{\hbar}\otimes_{Z\mathfrak{% g}}(\text{-})$ . Here, the functor $\operatorname{Obl}_{r}$ forgets the right $Z_{\hbar}\mathfrak{g}$ -action and $D^{\psi_{r}}_{\hbar}$ in the tensor product is viewed as a $Z_{\hbar}\mathfrak{g}$ -module via left multiplication.

Proof of 4.1.3.

We first construct the counit $c(M)$ of the desired adjunction. In Section 3.3 we equipped $Z_{\hbar}\mathfrak{g}$ with a left $\mathfrak{J}_{\hbar}$ -module structure. Define the map $\overline{\epsilon}:\mathfrak{J}_{\hbar}\to Z_{\hbar}\mathfrak{g}$ by the formula $\overline{\epsilon}(w)=w\cdot 1$ . For any $M\in Z_{\hbar}\mathfrak{g}\operatorname{{\text{-}{\mathrm{mod}}}}$ , define $c(M):\varkappa_{\hbar,\ell}\varkappa_{\hbar,\ell}^{R}(M)\to M$ to be the composite of the following maps

\varkappa_{\hbar,\ell}(\varkappa_{\hbar,\ell}^{R}(M))=\varkappa_{\hbar,\ell}(D% ^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}M)\xleftarrow{\sim}\mathfrak% {J}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}M\xrightarrow{\overline{\epsilon}% \otimes\text{id}}Z_{\hbar}\mathfrak{g}\otimes_{Z_{\hbar}\mathfrak{g}}M% \xrightarrow{\sim}M

where the leftward pointing arrow is the isomorphism induced by the exactness of $\varkappa_{\hbar,\ell}$ and the final arrow is the isomorphism induced by the module structure.

We now construct the unit map of our desired adjunction. Recall that any asymptotic Harish-Chandra bimodule $M$ is equivalently a comodule for ${D_{\hbar}}$ by 2.4.2(iv). Letting $\rho_{M}$ denote we will show that the map $u(M)$ given by the composite, read left to right, of the maps

M\xrightarrow{\rho_{M}}{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}^{G}M% \xrightarrow{}{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}^{N}M/M\mathfrak{n}^{% \psi}\xleftarrow{\sim}D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}% \varkappa_{\hbar,\ell}(M)

where the unlabelled arrow is induced by the quotient map and the final arrow is the (graded) monoidality isomorphism of 3.2.4, gives the unit.

We now show that $\varkappa_{\hbar,\ell}^{R}(c(M))\circ u(\varkappa_{\hbar,\ell}^{R}(M))$ is the identity for any $Z_{\hbar}\mathfrak{g}$ -module $M$ . Observe that it suffices to prove this equality when $M=Z_{\hbar}\mathfrak{g}$ or, in other words, that the composite of maps (read left to right)

D^{\psi_{r}}_{\hbar}\xrightarrow{p^{\psi_{r}}}{D_{\hbar}}\otimes_{U_{\hbar}% \mathfrak{g}}^{G}D^{\psi_{r}}_{\hbar}\xrightarrow{}{D_{\hbar}}\otimes_{U_{% \hbar}\mathfrak{g}}^{\overline{N}}D^{\psi_{r}}_{\hbar}/D^{\psi_{r}}_{\hbar}% \mathfrak{n}^{\psi}_{\ell}\xleftarrow{\sim}D^{\psi_{r}}_{\hbar}\otimes_{Z_{% \hbar}\mathfrak{g}}\mathfrak{J}_{\hbar}\xrightarrow{\text{id}\otimes\overline{% \epsilon}}D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}Z_{\hbar}% \mathfrak{g}\xrightarrow{\overline{\partial}\otimes z\mapsto z\overline{% \partial}}D^{\psi_{r}}_{\hbar}

is the identity, where $p^{\psi_{r}}$ denotes the coaction on $D^{\psi_{r}}_{\hbar}$ as in 3.3.2. Indeed, this follows from the fact that

\varkappa_{\hbar,\ell}^{R}(c(M))\circ u(\varkappa_{\hbar,\ell}^{R}(M))=% \varkappa_{\hbar,\ell}^{R}(c(Z_{\hbar}\mathfrak{g}))\circ u(\varkappa_{\hbar,% \ell}^{R}(Z_{\hbar}\mathfrak{g}))\otimes_{Z_{\hbar}\mathfrak{g}}M

which in turn follows from the exactness of $\bar{N}$ -invariants on $U_{\hbar}\mathfrak{g}\operatorname{{\text{-}{\mathrm{mod}}}}^{N,\psi}$ and exactness of $(\text{-})^{G}$ . If we choose $\overline{\partial}\in D^{\psi_{r}}_{\hbar}$ and write $p^{\psi_{r}}(\overline{\partial})=\partial_{(0)}\otimes\overline{\partial_{(1)}}$ for $\partial_{(0)}\in{D_{\hbar}}$ and $\overline{\partial_{(1)}}\in D^{\psi_{r}}_{\hbar}$ and choose some $x_{i},y_{i}\in{D_{\hbar}}$ so that $\overline{x_{i}}\in D^{\psi_{r}}_{\hbar},\overline{y_{i}}\in D^{\psi_{r}}_{\hbar}$ , and $\overline{\overline{y_{i}}}\in\mathfrak{J}_{\hbar}$ for which $\sum_{i}x_{i}\otimes\overline{y_{i}}=\partial_{(0)}\otimes\overline{\partial_{% (1)}}$ in ${D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}^{\overline{N}}D^{\psi_{r}}_{\hbar}/% D^{\psi_{r}}_{\hbar}\mathfrak{n}^{\psi}_{\ell}$ , then this map is given explicitly by the formula

\overline{\partial}\mapsto\sum_{i}\overline{\epsilon}(\overline{\overline{y_{i% }}})\overline{x_{i}}=\sum_{i}\epsilon^{\psi_{r}}(\overline{y_{i}})\overline{x_% {i}}=\sum_{i}\overline{\epsilon(y_{i})x_{i}}

in $D^{\psi_{r}}_{\hbar}$ , where $\epsilon^{\psi_{r}}$ denotes the $D^{\psi_{r}}_{\hbar}$ -module structure on $U_{\hbar}\mathfrak{g}/U_{\hbar}\mathfrak{g}\cdot\mathfrak{n}^{\psi}$ described explicitly in Section 3.3. By assumption, then, for such $x_{i},y_{i}$ there exists some $\sum_{j}p_{j}\otimes\overline{p_{j}^{\prime}}\in{D_{\hbar}}\otimes_{U_{\hbar}% \mathfrak{g}}D^{\psi_{r}}_{\hbar}$ and some $\xi\in\mathfrak{n}^{\psi}_{\ell}$ such that $\sum_{i}x_{i}\otimes\overline{y_{i}}=\partial_{(0)}\otimes\overline{\partial_{% (1)}}+\sum_{j}p_{j}\otimes\overline{p_{j}^{\prime}}\xi$ in ${D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}D^{\psi_{r}}_{\hbar}$ and so we see that

\sum_{i}\overline{\epsilon(y_{i})x_{i}}=\overline{\epsilon(\partial_{(1)})% \partial_{(0)}}+\sum_{j}\overline{\epsilon(p_{j}^{\prime}\xi)p_{j}}=\overline{% \epsilon(\partial_{(1)})\partial_{(0)}}=\overline{\partial}

as desired. The verification that $u(\varkappa_{\hbar,\ell}(B))\circ\varkappa_{\hbar,\ell}(u(B))$ is the identity for any $B\in\mathcal{HC}$ is similar. ∎

Corollary 4.1.4.

The $Z_{\hbar}\mathfrak{g}$ -modules $D^{\psi_{r}}_{\hbar}$ and $\mathfrak{J}_{\hbar}$ are flat.

Proof.

Since the functor $D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}(\text{-})$ is a right adjoint, it is left exact and thus $D^{\psi_{r}}_{\hbar}$ is flat. Hence, the composite functor $\varkappa_{\hbar,\ell}(D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}(% \text{-}))$ is exact. Using the isomorphism $\mathfrak{J}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}(\text{-})\cong\varkappa_{% \hbar,\ell}(D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}(\text{-}))$ obtained from the fact that $\varkappa_{\hbar,\ell}$ is a left adjoint and therefore commutes with all colimits, we deduce that $\mathfrak{J}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}(\text{-})$ is exact, proving the flatness statement for $\mathfrak{J}_{\hbar}$ . ∎

Since ${D_{\hbar}}\cong U_{\hbar}\mathfrak{g}\otimes\mathcal{O}(G)$ is flat as a $U_{\hbar}\mathfrak{g}$ -module and $\varkappa_{\hbar,\ell}$ is exact by 3.2.4, 4.1.1(iii) follows immediately from the following more general result:

Lemma 4.1.5.

Assume $D$ is some $R$ -coring object which is flat as an $R$ -module and

K:D\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}(R\operatorname{{\text{-}{% \mathrm{mod}}}})\to\mathcal{A}

is some exact functor. Then there is a canonical isomorphism $K(C)\to K(D)\otimes_{R}^{D}C$ which is natural in $C$ .

Proof.

For any comodule $C$ , the coaction map naturally factors through a map $C\xrightarrow{}D\otimes_{R}^{D}C$ of $D$ -comodules by definition of a comodule. This map is easily shown to be an isomorphism, see for example [BrzezinskiWisbauerCoringsandComodules, Section 21.1]. Therefore applying $K$ to this isomorphism we obtain an isomorphism

K(C)\xrightarrow{\sim}K(D\otimes_{R}^{D}C)\xrightarrow{\sim}K(D\otimes_{R}C)^{% D}\xleftarrow{\sim}K(D)\otimes_{R}^{D}C

where the middle arrow, respectively rightmost arrow, is an isomorphism by the left exactness, respectively right exactness, of $K$ . These maps are evidently natural in $C$ . ∎

Before proving the remainder of 4.1.1, we will state the following standard lemma (see, for example, [AguiarMahajanMonoidalFunctors, Proposition 3.84, Proposition 3.85]) which we will use many times in what follows:

Lemma 4.1.6.

The right adjoint, resp. left adjoint, to any monoidal functor $F$ is naturally lax monoidal, left lax monoidal, in such a way that the unit and counit maps are compatible with the monoidal structure. Moreover, if $F$ is braided monoidal then the lax monoidality, resp. left lax monoidality, morphisms are compatible with the braiding.

Proof of 4.1.1(i)-(ii).

From the adjunction of 4.1.3, we formally obtain a comonad structure on the endofunctor $\varkappa_{\hbar,\ell}\varkappa_{\hbar,\ell}^{R}$ . Moreover, since $\varkappa_{\hbar,\ell}$ is exact, we see that this comonad is naturally isomorphic to tensoring with the $Z_{\hbar}\mathfrak{g}$ -module $\mathfrak{J}_{\hbar}:=\varkappa_{\hbar,\ell}\varkappa_{\hbar,\ell}^{R}(Z_{% \hbar}\mathfrak{g})$ , and thus $\varkappa_{\hbar,\ell}\varkappa_{\hbar,\ell}^{R}(Z_{\hbar}\mathfrak{g})$ acquires the structure of a graded $Z_{\hbar}\mathfrak{g}$ -coalgebra. Tracing through the constructions of the coalgebra structure on $\mathfrak{J}_{\hbar}$ in 3.3.2, we see that these two a priori distinct coalgebra structures agree. We therefore may obtain our desired functor $\kappa_{\hbar}$ by following the procedure as in [MacLaneCategoriesForTheWorkingMathematician, Chapter VI], with the direction of the arrows reversed. Explicitly, to any ${D_{\hbar}}$ -comodule $C$ , the coaction on $\varkappa_{\hbar,\ell}(C)$ is given by the composite, read left to right, of the maps:

\varkappa_{\hbar,\ell}(C)\xrightarrow{\varkappa_{\hbar,\ell}(u(C))}\varkappa_{% \hbar,\ell}\varkappa_{\hbar,\ell}^{R}(\varkappa_{\hbar,\ell}(C))\xleftarrow{% \sim}\mathfrak{J}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g}}\varkappa_{\hbar,\ell}% (C)

where the final arrow is an isomorphism by the exactness of $\kappa_{\hbar}$ . This comodule structure can also be described by the following formula: for any $\overline{b}\in(C/C\mathfrak{n}^{\psi})^{N}$ , there exists $\overline{\overline{\partial_{i}}}\in\mathfrak{J}_{\hbar}$ and $\overline{y_{i}}\in(C/C\mathfrak{n}^{\psi})^{N}$ such that $\overline{b_{(0)}}\otimes\overline{b_{(1)}}=\sum_{i}\overline{\partial_{i}}% \otimes\overline{y_{i}}$ in ${D_{\hbar}}/{D_{\hbar}}\mathfrak{n}^{\psi}_{\ell}\otimes_{U_{\hbar}\mathfrak{g% }}C/C\mathfrak{n}^{\psi}$ , and this coaction map is defined by $\overline{b}\mapsto\sum_{i}\overline{\overline{\partial_{i}}}\otimes\overline{% y_{i}}$ . With respect to the monoidal functor $\mathfrak{J}_{\hbar}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}\to Z_{% \hbar}\mathfrak{g}\text{-bimod}$ constructed in 3.3.2 we can compute that $\kappa_{\hbar}$ lifts $\varkappa$ .

Since $D^{\psi_{r}}_{\hbar}$ is flat as a $Z_{\hbar}\mathfrak{g}$ -module by 4.1.4, the left ${D_{\hbar}}$ -comodule structure on $D^{\psi_{r}}_{\hbar}$ induces a left ${D_{\hbar}}$ -comodule structure on $\kappa_{\hbar}^{R}(C)$ . The fact $\kappa_{\hbar}^{R}:\mathfrak{J}_{\hbar}\operatorname{{\mathrm{\text{-}\mathrm{% comod}}}}\to{D_{\hbar}}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ is a right adjoint follows formally from the fact that the unit and counit map of the adjunction constructed in 4.1.3 factors through the $\mathfrak{J}_{\hbar}$ -cotensor product. The induced counit map for this adjunction is readily checked to be an isomorphism, and so $\kappa_{\hbar}^{R}$ is fully faithful.

We now equip $\kappa_{\hbar}$ with a monoidal structure. Observe that we may equivalently describe the induced coaction $\rho_{\kappa_{\hbar}(C)}$ as follows: it is the unique map of $Z_{\hbar}\mathfrak{g}\otimes_{k[\hbar]}Z_{\hbar}\mathfrak{g}_{{\text{op}}}$ -modules such that the induced diagram of $Z_{\hbar}\mathfrak{g}\otimes Z_{\hbar}\mathfrak{g}_{{\text{op}}}\otimes% \mathfrak{J}_{\hbar}^{{\text{op}}}$ -modules

(4.1.7)

commutes, where $\varkappa^{2}$ denotes the Kostant-Whittaker reduction for the group $G\times G_{\text{op}}$ , $\widetilde{\rho}_{C}$ is the right ${D_{\hbar}}$ -linear map induced from the coaction $\rho_{C}$ , and the unlabelled arrows are the isomorphisms are the isomorphisms induced by exactness and monoidality; for example: the left vertical arrow is given by the formula $\overline{b}\otimes\overline{\overline{\partial}}\mapsto\overline{\overline{b% \otimes\partial}}$ . Similarly observe that the monoidal structure on the category ${D_{\hbar}}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ (which we have shown is equivalent as a monoidal category to $\mathcal{HC}$ in 2.4.2(iv)) in 2.4.2 has the property and is characterized by the fact that

(4.1.8)

\widetilde{\rho}_{C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime}}=(\widetilde{\rho% }_{C}\otimes\text{id}_{C^{\prime}})\circ(\text{id}_{C}\otimes\widetilde{\rho}_% {C^{\prime}})

and, moreover, that this map is isomorphism of $U_{\hbar}\mathfrak{g}\otimes U_{\hbar}\mathfrak{g}_{{\text{op}}}\otimes{D_{% \hbar}}^{{\text{op}}}$ -modules.

Now, using (4.1.7) we obtain commutative diagrams

and

where as above all unlabelled arrows are induced by exactness and monoidality. Applying $\kappa_{\hbar}^{2}$ to the equation (4.1.8) for the ${D_{\hbar}}$ -comodule $C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime}$ , we obtain the equality

\kappa_{\hbar}^{2}(\widetilde{\rho}_{C\otimes_{U_{\hbar}\mathfrak{g}}C^{\prime% }})=\kappa_{\hbar}^{2}(\widetilde{\rho}_{C}\otimes\text{id}_{C^{\prime}})\circ% \kappa_{\hbar}^{2}(\text{id}_{C}\otimes\widetilde{\rho}_{C^{\prime}})

of morphisms. Combining this equation with these two commutative diagrams gives that the monoidality isomorphism of 3.2.4 respects the $\mathfrak{J}_{\hbar}$ -comodule structure, and completes our proof that $\kappa_{\hbar}$ is monoidal. Since $\kappa_{\hbar}^{R}$ is right adjoint to a monoidal functor, one can equip it with a lax monoidal structure using 4.1.6.

We now verify that $\kappa_{\hbar}^{R}$ preserves filtered colimits. The compact objects in $\mathcal{HC}$ are precisely those Harish-Chandra bimodules which are finitely generated as left, equivalently right, $U_{\hbar}\mathfrak{g}$ -modules. The functor $\kappa_{\hbar}$ sends finitely generated $U_{\hbar}\mathfrak{g}$ -modules to finitely generated $Z_{\hbar}\mathfrak{g}$ -modules. In particular, since the forgetful functor $\mathfrak{J}_{\hbar}\text{-comod}\to Z_{\hbar}\mathfrak{g}\operatorname{{\text% {-}{\mathrm{mod}}}}$ reflects compact objects we see that $\kappa_{\hbar}$ preserves compact objects. Since any object of $\mathcal{HC}$ is a filtered colimit of finitely generated ones, a Yoneda lemma argument gives that $\kappa_{\hbar}^{R}$ preserves filtered colimits.

Finally, since $\kappa_{\hbar}$ is monoidal, the universal property of quotient categories induces the desired monoidal functor in 4.1.2, see for example [DayNoteonMonoidalLocalization]. Moreover, [GabrielDesCategoriesAbeliannes, Proposition III.5] implies that this induced functor is an equivalence. ∎

Proof of 4.1.1(iv).

We have isomorphisms

{D_{\hbar}}\otimes_{U_{\hbar}\mathfrak{g}}U_{\hbar}\mathfrak{g}/U_{\hbar}% \mathfrak{g}\cdot\mathfrak{n}^{\psi}\cong\mathcal{O}(G)\otimes U_{\hbar}% \mathfrak{g}/U_{\hbar}\mathfrak{g}\cdot\mathfrak{n}^{\psi}\cong\oplus_{\lambda% }(V_{\lambda}^{\text{dim}(V_{\lambda})}\otimes U_{\hbar}\mathfrak{g}/U_{\hbar}% \mathfrak{g}\cdot\mathfrak{n}^{\psi}).

Since the quotient $U_{\hbar}\mathfrak{g}/U_{\hbar}\mathfrak{g}\cdot\mathfrak{n}^{\psi}$ has nonnegative grading, it follows that the grading on $D^{\psi_{\ell}}_{\hbar}$ , respectively $D^{\psi_{r}}_{\hbar}$ induced by the actions of $G_{\text{op}}$ , respectively $G$ , have isotypic components bounded below.

Note that $\kappa_{\hbar}^{R}$ commutes with filtered colimits and $\kappa_{\hbar}$ commutes with all colimits, and in particular filtered colimits, as $\kappa_{\hbar}$ is a left adjoint; both of these claims are justified by 4.1.1(ii). Therefore it suffices to prove the claim when $V$ is a finite dimensional representation. In this case, the unit map $u(\text{Fr}_{\hbar}(V))$ from the adjunction of 4.1.1 is a map in $\mathcal{HC}_{\hbar}$ from a filtered vector space with separated filtration whose isotypic components are bounded below, since $\kappa_{\hbar}^{R}\kappa_{\hbar}(\text{Fr}_{\hbar}(V))\subseteq D^{\psi_{r}}_{% \hbar}\otimes_{Z_{\hbar}\mathfrak{g}}\kappa_{\hbar}(\text{Fr}_{\hbar}(V))$ and the latter is the tensor product of a vector space whose grading is bounded from below with $D^{\psi_{r}}_{\hbar}$ , whose isotypic components are bounded below by the above analysis. By graded Nakayama and separatedness, then, to show $u(\text{Fr}_{\hbar}(V))$ is an isomorphism it therefore suffices to check that $k\otimes_{k[\hbar]}u(\text{Fr}_{\hbar}(V))$ is an isomorphism. Moreover, since $\kappa_{\hbar}(\text{Fr}_{\hbar}(V))$ is torsion free as a $k[\hbar]$ -module, the natural map

k\otimes_{k[\hbar]}\kappa_{\hbar}^{R}\kappa_{\hbar}(\text{Fr}_{\hbar}(V))% \xrightarrow{}\kappa_{0}^{R}\kappa_{0}(k\otimes_{k[\hbar]}\text{Fr}_{\hbar}(V)% )\cong\mathcal{O}(T^{*}_{\psi}(G/N))\otimes_{Z\mathfrak{g}}^{J}\kappa_{0}(% \operatorname{{\mathrm{Sym}}}(\mathfrak{g})\otimes V)

is injective, and so it suffices to show that the composite of $k\otimes_{k[\hbar]}u(\text{Fr}_{\hbar}(V))$ with this map is an isomorphism. However, this composite is the unit map for $\kappa_{0}$ . Therefore by 3.1.1 this unit map is an equivalence if and only if the unit map $\operatorname{{\mathrm{Sym}}}(\mathfrak{g})\otimes V\to\jmath_{*}\jmath^{*}(% \operatorname{{\mathrm{Sym}}}(\mathfrak{g})\otimes V)$ for $(\jmath^{*},\jmath_{*})$ is an equivalence. But this map is an equivalence since the complement of $\mathfrak{g}^{*}_{\text{reg}}$ in $\mathfrak{g}^{*}$ has codimension at least two and $\operatorname{{\mathrm{Sym}}}(\mathfrak{g})\otimes V$ is a vector bundle on $\mathfrak{g}^{*}$ . ∎

Various versions of the following corollary have appeared earlier (at least in the classical setting ${\hbar}=0$ ) in the context of Sicilian theories and Coulomb branches, see [MooreTachikawaOn2dTQFTsWhoseValuesareHolomorphicSymplecticVarieties], [BravermanFinkelbergNakajimaRingObjectsIntheEquivariantDerivedSatakeCategoryArisingFromCoulombBranches], [GinzburgKazhdanDifferentialOperatorsOnBasicAffineSpaceandtheGelfandGraevAction], [BielawskiOnTheMooreTachikawaVarieties], [GannonCotangentBundleofParabolicBaseAffineSpaceandKostantWhittakerDescent], [CrooksMayrandMooreTachikawaViaShiftedSymplecticGeometry].

Corollary 4.1.9.

The unit, respectively counit, for the adjoint pair $(\kappa_{\hbar},\kappa_{\hbar}^{R})$ constructed above gives a algebra isomorphisms ${D_{\hbar}}\xrightarrow{\sim}D^{\psi_{r}}_{\hbar}\otimes_{Z_{\hbar}\mathfrak{g% }}^{\mathfrak{J}_{\hbar}}D^{\psi_{\ell}}_{\hbar}$ and $\mathfrak{J}_{\hbar}\xrightarrow{\sim}D^{\psi_{\ell}}_{\hbar}\otimes_{U_{\hbar% }\mathfrak{g}}^{G}D^{\psi_{r}}_{\hbar}$ . Similar algebra isomorphisms exist in the non-filtered and associated graded settings.

Proof.

The first claim is given by taking $V=\mathcal{O}(G)$ in 4.1.1(iv). The second claim follows from the fact that $\kappa_{\hbar}^{R}$ is fully faithful by 4.1.1(iii) as well as the fact that a right adjoint is fully faithful if and only if the counit map is an equivalence. ∎

5. Kostant-Whittaker reduction and the Miura bimodule

In the rest of the paper we work in the ‘quantum setting’, that is, we specialize ${\hbar}=1$ unless explicitly stated otherwise. Many of the results below have neither asymptotic nor quasi-classical counterparts. The constructions and results of the previous section apply in the ${\hbar}=1$ setting, with essentially identical proofs. In particular, the functor $\varkappa$ lifts to a functor $\kappa$ giving the following adjoint pair

such that $\kappa^{R}$ is fully faithful and commutes with filtered colimits. One can also run analogous arguments (using the exactness of $\kappa_{0}$ ) and give an alternate proof of 3.1.1.

5.1. The Miura bimodule and the Gelfand-Graev action

Taking the Hamiltonian reduction for the (left) $\bar{N}$ -action, we obtain the algebra ${D}(\bar{N}\backslash G)$ of global differential operators on $\bar{N}\backslash G$ . The action of $G$ resp. $T=\bar{B}/\bar{N}$ , on $\bar{N}\backslash G$ by right, resp. left, translations induces an action of $T\times G_{\text{op}}$ on ${D}(\bar{N}\backslash G)$ . One can show that the natual morphisms

(5.1.1)

\varkappa({D}(\bar{N}\backslash G))\xrightarrow{}\big{(}D(G)/(D(G)\bar{% \mathfrak{n}}+\mathfrak{n}^{\psi}_{r}D(G))\big{)}^{\bar{N}\times N}\xleftarrow% {}(D^{\psi_{r}}/D^{\psi_{r}}\bar{\mathfrak{n}})^{\bar{N}}

are algebra isomorphisms [Gin, Lemma 5.4.1]; moreover, restriction of differential operators on $\bar{N}\backslash G$ to the to the open $B_{\text{op}}$ -orbit $\bar{N}\backslash\bar{N}B\subseteq\bar{N}\backslash G$ identifies the algebras in 5.1.1 with the algebra $D(T)$ , see [Gin, Proposition 5.5.2]. Therefore, completely analogously to our arguments in Section 3.2, there is a $(D(T),D^{\psi_{r}})$ -bimodule structure on $\bar{\mathfrak{n}}D^{\psi_{r}}\backslash D^{\psi_{r}}$ , resp. $(D^{\psi_{r}},D(T))$ -bimodule structure on $D^{\psi_{r}}/D^{\psi_{r}}\bar{\mathfrak{n}}$ . This gives the space

{\mathbb{M}}:=\bar{\mathfrak{n}}D^{\psi_{r}}\backslash D^{\psi_{r}}/D^{\psi_{r% }}\mathfrak{n}^{\psi}_{\ell},\enspace\text{resp.}\enspace\check{\mathbb{M}}:=% \mathfrak{n}^{\psi}_{\ell}D^{\psi_{r}}\backslash D^{\psi_{r}}/D^{\psi_{r}}\bar% {\mathfrak{n}},

the structure of a $(D(T),\mathfrak{J})$ -bimodule, resp. $(\mathfrak{J},D(T))$ -bimodule, which we will refer to as the Miura bimodule, resp. ‘transposed’ Miura bimodule.

Recall next that the algebra $D(\bar{N}\backslash G)$ has an action of the Weyl group known as the Gelfand-Graev action, see [GinzburgRicheDifferentialOperatorsOnBasicAffineSpaceandtheAffineGrassmannian, Section 3.2], [GinzburgKazhdanDifferentialOperatorsOnBasicAffineSpaceandtheGelfandGraevAction]. The subalgebra $U\mathfrak{g}_{\text{op}}\subseteq D(\bar{N}\backslash G)$ is fixed by this action. Furthermore, the actions of the groups $G_{\text{op}},T$ , and $W$ , combined together give an action of the group $(T\rtimes W)\times G_{\text{op}}$ on $D(\bar{N}\backslash G)$ by algebra automorphisms. The action of $T\rtimes W$ survives the Whittaker reduction, so the $(D(T),\mathfrak{J})$ -bimodule, resp. $(\mathfrak{J},D(T))$ -bimodule, structure on ${\mathbb{M}}$ , resp. $\check{\mathbb{M}}$ , can be upgraded to a $(D(T)\rtimes W,\,\mathfrak{J})$ -bimodule, resp. $(\mathfrak{J},\,D(T)\rtimes W)$ -bimodule, structure.

Proposition 5.1.2.

${{\enspace}\sf{(i)}}\;$ The category $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}$ of modules for $D(T)$ in $\text{Rep}(W)$ acquires the structure of a $(D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W},\mathfrak{J}\operatorname{{% \text{-}{\mathrm{mod}}}})$ -bimodule category. ${\;\sf{(ii)}}\;$ The category ${D}(\bar{N}\backslash G)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}$ of modules for global differential operators on $\bar{N}\backslash G$ equipped with a compatible $W$ -representation is a $((D(T)\rtimes W)\operatorname{{\text{-}{\mathrm{mod}}}},D(G)\operatorname{{% \text{-}{\mathrm{mod}}}})$ -bimodule category.

The proof of the proposition is entirely analogous to the proof of LABEL:Asymptotic_Enhancement_of_All_PreliminaryResults.

5.2. Parabolic induction and the Miura bimodule

The functor of parabolic induction is defined as a pull-push functor $\int_{p}q^{*}(\text{-}):{\mathscr{D}}^{W}(T)\to{\mathscr{D}}^{\operatorname{{% \mathrm{Ad}}}G}(G)$ with respect to the diagram $G\,\xleftarrow{p}\,\tilde{G}=G\times^{B}B\,\xrightarrow{q}\,T$ . Here, $B$ acts on itself by conjugation, the map $q$ sends $(g,b)$ to $b\text{ mod N}\in B/N=T$ , and $p$ is the Grothendieck-Springer morphism. It is known that parabolic induction is an exact functor and we let $\operatorname{Ind}_{T}^{G}:D(T)\rtimes W\operatorname{{\text{-}{\mathrm{mod}}}% }\xrightarrow{}D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G_{\text{ad}}}$ denote the induced functor between the corresponding abelian hearts.

The functor $\operatorname{Ind}_{T}^{G}$ has a simple algebraic description in terms of the $(D(G),D(T)^{W})$ -bimodule ${\mathbf{N}}=D(G)/D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}$ , where $D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}$ is the left ideal generated by the elements $\operatorname{{\mathrm{ad}}}\xi:=\xi-\xi_{r},\,\xi\in\mathfrak{g}$ , and the right $D(T)^{W}$ -action comes from the radial parts isomorphism, see Section 1.1. Specifically, let ${\mathbf{M}}:={\mathbf{N}}\otimes_{D(T)^{W}}D(T)$ . This tensor product has the structure of a $(D(G),\,D(T)\rtimes W)$ -bimodule where the action of $W$ comes from the natural $W$ -action on the second tensor factor. Then, it was shown in [GinzburgParabolicInductionandtheHarishChandraDModule] that there are isomorphisms of functors

(5.2.1)

\operatorname{Ind}_{T}^{G}(\text{-})\cong\mathbf{M}\otimes_{D(T)}(\text{-})% \cong{\mathbf{N}}\otimes_{D(T)^{W}}D(T)\otimes_{D(T)}(\text{-}).

Taking $W$ -invariants one deduces an isomorphism

(5.2.2)

\mathbf{N}\otimes_{D(T)^{W}}(\text{-})\cong\operatorname{Ind}_{T}^{G}(\text{-}% )^{W}.

he right $D(T)^{W}$ -action on $\mathbf{N}$ commutes with the left $D(G)$ -action, hence it induces, by functoriality, a $D(T)^{W}$ -action on $\kappa(\mathbf{N})$ .

Proposition 5.2.3.

There is an isomorphism of $(D(T)^{W},\mathfrak{J})$ -bimodules $\kappa({\mathbf{N}})\xrightarrow{\sim}\check{\mathbb{M}}^{W}$ .

Proof.

We claim that the following diagram

commutes. Here, each term in the diagram is labelled by a ring and corresponds to its category of modules, and superscripts refer to strong equivariance. In this diagram, $\nu^{*}$ denotes the (exact) pullback functor by the quotient map $\nu:\bar{B}\to\bar{B}/\bar{N}=T$ , $\iota:\bar{B}\xhookrightarrow{}G$ is the closed embedding, the functors labelled $\text{Av}_{*}$ are the trunctions of the $t$ -exact averaging functors on the corresponding derived categories²²2Standard arguments reviewed in [GannonUniversalCategoryOandGelfandGraevAction, Corollary 2.44], [GannonUniversalCategoryOandGelfandGraevAction, Proposition 1.3] give both of the corresponding derived functors are $t$ -exact, so they agree with their truncations. as in [AcharPerverseSheavesandApplicationstoRepresentationTheory, Chapter 6.6] cohomologically shifted so as to give exact functors on the corresponding abelian categories, and the arrow labelled ‘ $\sim$ ’ is the functor ${(D(G)/D(G)\mathfrak{n}^{\psi}_{r})\otimes_{D^{\psi_{r}}}(\text{-})}$ which is an equivalence of categories by Skryabin’s equivalence. By the definition of parabolic induction and 5.2.1, the upper left triangle commutes. The upper right triangle commutes by [Gin, Theorem 1.6.4(ii)]. To see that the lower right box commutes, first observe that, completely analogously to the untwisted case in [ChenOnTheConjecturesofBravermanKazhdan, Example 6.1], one has an isomorphism $\text{Av}_{*}^{\psi_{\ell}}(\mathcal{F})^{\mathfrak{n}^{\psi}_{\ell}}\cong% \mathfrak{n}^{\psi}_{\ell}\mathcal{F}\backslash\mathcal{F}$ for any $D(G)$ -module $\mathcal{F}$ . Moreover, if $\mathcal{F}$ is $(N_{\text{op}},\psi)$ -equivariant the canonical map $\mathfrak{n}^{\psi}_{\ell}\mathcal{F}^{\mathfrak{n}^{\psi}_{r}}\backslash% \mathfrak{n}^{\psi}_{\ell}\mathcal{F}^{\mathfrak{n}^{\psi}_{r}}\xrightarrow{}(% \mathfrak{n}^{\psi}_{\ell}\mathcal{F}\backslash\mathcal{F})^{\mathfrak{n}^{% \psi}_{r}}$ is an isomorphism by the exactness of $(-)^{\mathfrak{n}^{\psi}_{r}}$ . Thus we obtain that $\text{Av}_{*}^{\psi_{\ell}}(\mathcal{F})^{\mathfrak{n}^{\psi}_{\ell}\times% \mathfrak{n}^{\psi}_{r}}\cong(\mathfrak{n}^{\psi}_{\ell}\mathcal{F}\backslash% \mathcal{F})^{\mathfrak{n}^{\psi}_{r}}$ and since the inverse of the vertical map is given by $(-)^{\mathfrak{n}^{\psi}_{r}}$ we see that the right lower box comutes. By a standard argument the lower left box commutes, see [GannonUniversalCategoryOandGelfandGraevAction, Proposition 1.3]. Finally, the argument of [RaskinAffineBBLocalization, Section 4.6, Step 3] shows that the middle square commutes.

Since the Kostant-Whittaker reduction functor is exact, the functor given by traversing the top of the diagram is naturally isomorphic to tensoring with the bimodule $\varkappa(\mathbf{M})$ . The bottom functor is obviously given by tensoring with $\check{\mathbb{M}}$ , and so $\check{\mathbb{M}}$ is isomorphic to $\varkappa(\mathbf{M})$ . Since we have a canonical isomorphism $\check{{\mathbb{M}}}\cong{\mathbf{N}}\otimes_{D(T)^{W}}D(T)$ , taking $W$ -invariants yields our claim. ∎

5.3. The Miura bimodule and comodules

Let $\operatorname{Q}$ denote the field of fractions of $\operatorname{Sym}(\mathfrak{t})$ . Given a $\operatorname{Sym}(\mathfrak{t})$ -module $M$ , we write $M_{\operatorname{loc}}:=\operatorname{Q}\otimes_{\operatorname{Sym}(\mathfrak{% t})}M$ . When clear from context, we also use the same notation $M_{\operatorname{loc}}:=\operatorname{Q}\otimes_{Z\mathfrak{g}}M$ if $M$ is a $Z\mathfrak{g}$ -module, and use the notation $f_{\operatorname{loc}}$ for the map induced by the tensor product of a map $f$ of modules for either $\operatorname{Sym}(\mathfrak{t})$ or $Z\mathfrak{g}$ .

By Skryabin’s equivalence, the $\operatorname{Sym}(\mathfrak{t})$ -module map induces an isomorphism

(5.3.1)

\operatorname{Sym}(\mathfrak{t})\otimes_{Z\mathfrak{g}}\mathfrak{J}% \xrightarrow{\sim}{\mathbb{M}}

and so taking the $W$ -invariant summand, we see that ${\mathbb{M}}^{W}$ is free rank one as a $\mathfrak{J}$ -module, cf. [Gin, Section 6.2]. Therefore, acting by $D(T)^{W}$ induces a ring morphism $D(T)^{W}\to\mathfrak{J}$ . Since $1\in{\mathbb{M}}$ is fixed by $W$ , we also obtain an induced $W$ -equivariant map

\varphi:D(T)\to{\mathbb{M}}

where $W$ acts on $D(T)$ by the usual conjugation action.

Proposition 5.3.2.

The $\operatorname{Sym}(\mathfrak{t})$ -module structure on ${\mathbb{M}}$ upgrades to the structure of a cocommutative coalgebra object in $(D(T)\rtimes W)\operatorname{{\text{-}{\mathrm{mod}}}}$ such that $\varphi$ is a map of coalgebras. Moreover, the induced map $\operatorname{Q}\otimes_{\operatorname{Sym}(\mathfrak{t})}\varphi$ is an isomorphism of coalgebras.

Proof.

The distinguished $D(T)$ -module $\delta\cong\operatorname{Sym}(\mathfrak{t})$ supported at the unit element of $T$ is a monoidal unit of the monoidal category $D(T)\operatorname{{\text{-}{\mathrm{mod}}}}^{W}$ , where the monoidal structure is given by convolution. Thus, one may view this category as a free rank one module category over itself with generator $\delta$ , in other words, the functor $\operatorname{act}_{\delta}:E\mapsto E\star\delta$ given by acting on $\delta$ is an equivalence.

Observe next that any object $M\in\mathfrak{J}$ -mod gives rise to a $D(T)\rtimes W$ -module equivariant endofunctor on the category of $W$ -equivariant $D(T)$ -modules which commutes with colimits. It follows that one can equip the functor

F:\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to D(T)\rtimes W% \operatorname{{\text{-}{\mathrm{mod}}}}

defined by tensoring with ${\mathbb{M}}$ with a monoidal structure, which follows directly from [KalSaf, Proposition 2.23]. Indeed, following [KalSaf, Section 2.4], if $\operatorname{act}^{R}_{\delta}$ denotes the right adjoint to the functor $\operatorname{act}_{\delta}$ , for any $M,M^{\prime}\in\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ , it is not difficult to construct a canonical map, sometimes known as the lax linearity morphism, of the form

(5.3.3)

\operatorname{act}^{R}_{\delta}(\delta\otimes M)\otimes\operatorname{act}^{R}_% {\delta}(\delta\otimes M^{\prime})\to\operatorname{act}^{R}_{\delta}(% \operatorname{act}^{R}_{\delta}(\delta\otimes M)\otimes\delta\otimes M^{\prime})

using adjunction and the fact $\operatorname{act}_{\delta}^{R}$ is $D(T)\rtimes W\operatorname{{\text{-}{\mathrm{mod}}}}$ linear; composing 5.3.3 with the unit

\operatorname{act}^{R}_{\delta}(\operatorname{act}^{R}_{\delta}(\delta\otimes M% )\otimes\delta\otimes M^{\prime})\xrightarrow{}\operatorname{act}^{R}_{\delta}% (\delta\otimes M\otimes M^{\prime})

of the adjoint pair $(\operatorname{act}_{\delta},\operatorname{act}_{\delta}^{R})$ gives our lax monoidality morphisms for $F=\operatorname{act}^{R}_{\delta}(\delta\otimes-)$ , which are isomorphisms since $\operatorname{act}^{R}_{\delta}$ is an equivalence. Unitality is similar.

Since $F$ is a monoidal functor, $F$ preserves coalgebra objects. Since $\mathfrak{J}$ is a coalgebra object in $\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ and ${\mathbb{M}}\cong F(\mathfrak{J})$ we see that ${\mathbb{M}}$ is a coalgebra object of the category of modules for $D(T)\rtimes W$ . We temporarily let $\Delta_{D(T)}$ denote the comultiplication in $D(T)$ and $\Delta_{{\mathbb{M}}}$ denote the comultiplication for the Miura bimodule. To show $\varphi$ is a coalgebra map, we must show

(5.3.4)

\Delta_{{\mathbb{M}}}\circ\varphi=(\varphi\otimes\varphi)\circ\Delta_{D(T)}

and that $\varphi$ is compatible with counits. However, by construction, $\Delta_{D(T)}$ and $\Delta_{{\mathbb{M}}}$ are $D(T)\rtimes W$ -module maps, and $\varphi$ is a $(D(T)\rtimes W)$ -module map since it is given by acting on some element. Moreover, for $\nu,\nu^{\prime}\in D(T)$ and $\partial\in D(T)\rtimes W$ we have

\displaystyle(\varphi\otimes\varphi)(\partial\cdot(\nu\otimes\nu^{\prime}))

\displaystyle:=(\varphi\otimes\varphi)(\partial_{(0)}\nu\otimes\partial_{(1)}% \nu^{\prime})=\varphi(\partial_{(0)}\nu)\otimes\varphi(\partial_{(1)}\nu^{% \prime})=\partial_{(0)}\varphi(\nu)\otimes\partial_{(1)}\varphi(\nu^{\prime})

and by the definition of the $(D(T)\rtimes W)$ -module structure on the tensor product we have

\partial_{(0)}\varphi(\nu)\otimes\partial_{(1)}\varphi(\nu^{\prime})=\partial% \cdot(\varphi(\nu)\otimes\partial_{(1)}\varphi(\nu^{\prime}))=\partial\cdot(% \varphi\otimes\varphi)(\nu\otimes\nu^{\prime})

and so combining these we see that $\varphi\otimes\varphi$ is also a $(D(T)\rtimes W)$ -module map. Thus both terms in 5.3.4 are $(D(T)\rtimes W)$ -module maps which send $1$ to $1\otimes 1$ and therefore are the same map. Compatibility with counits is similar.

Using [Gin, Proposition 8.1.1], one can show that the left $(D(T)\rtimes W)$ -module structure upgrades to a module for the nil DAHA $\mathbb{H}$ , a certain $\operatorname{Sym}(\mathfrak{t})$ -ring object which contains $D(T)\rtimes W$ as a subring for which the injective map

(5.3.5)

(D(T)\rtimes W)_{\operatorname{loc}}\to\mathbb{H}_{\operatorname{loc}}

induced by inclusion is an isomorphism by construction. Moreover, one can use the arguments of the proof of [Gin, Theorem 8.1.2(i)] to show that the left $\mathbb{H}$ -module ${\mathbb{M}}$ is isomorphic to $\mathbb{H}e$ for a certain idempotent $e\in\mathbb{H}$ . From this, using the isomorphism 5.3.5 we obtain that $\varphi_{\operatorname{loc}}$ is an isomorphism.

It remains to prove that the coalgebra structure on $\mathbb{M}$ is cocommutative. First observe that ${\mathbb{M}}$ is flat as a $\operatorname{Sym}(\mathfrak{t})$ -module: indeed, $\operatorname{Sym}(\mathfrak{t})\otimes_{Z\mathfrak{g}}\mathfrak{J}$ is flat as a $\operatorname{Sym}(\mathfrak{t})$ -module since $\mathfrak{J}$ is flat as a $Z\mathfrak{g}$ -module by 4.1.4, so the isomorphism 5.3.1 gives our desired flatness. Thus to prove cocommutativity of ${\mathbb{M}}$ it suffices to prove the cocommutativity of ${\mathbb{M}}_{\operatorname{loc}}$ . However, the isomorphism $\varphi_{\operatorname{loc}}$ identifies ${\mathbb{M}}_{\operatorname{loc}}$ with a cocommutative coalgebra, and so we obtain our desired cocommutativity. ∎

5.4. Trivial comodules

Let $\mathfrak{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}_{\text{triv}}$ denote the full subcategory of graded $\mathfrak{J}$ -comodules which are isomorphic to comodules ${C}$ whose coaction is given by the map $y\mapsto 1\otimes y$ ; this subcategory is obviously a monoidal subcategory. We similarly define $\mathfrak{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}_{\text{triv}}$ .

Lemma 5.4.1.

There is a natural isomorphism $\kappa^{R}|_{\mathfrak{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}_{% \text{triv}}}\cong U\mathfrak{g}\otimes_{Z\mathfrak{g}}(\text{-})$ of lax monoidal functors. In particular, the lax monoidal morphisms for these functors are isomorphisms on the full subcategory of trivial comodules.

Proof.

For ${C}\in\mathfrak{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}_{\text{triv}}$ we have

\kappa^{R}({C})\ =\ D^{\psi_{r}}\otimes_{Z\mathfrak{g}}^{\mathfrak{J}}{C}\ =\ % (\mathcal{D}^{\psi_{r}})^{\mathfrak{J}}\otimes_{Z\mathfrak{g}}{C}\ \xleftarrow% {\sim}\ U\mathfrak{g}\otimes_{Z\mathfrak{g}}{C}.

Here, the second equality holds by the triviality assumption on $C$ and the map on the right is the isomorphism of 4.1.1(iv) in the special case of the trivial representation $V=k$ ; furthermore, it is a map of algebras since the unit and counit for the adjoint pair $(\kappa,\kappa^{R})$ are monoidal natural transformations. Thus, we obtain the desired natural isomorphism and, tracing through the constructions, one readily checks that this natural isomorphism respects the monoidal structure. ∎

Remark 5.4.2.

There is an asymptotic analogue of the above lemma with identical proof.

Corollary 5.4.3.

If $C$ is a left $\mathfrak{J}$ -comodule which is flat as a $Z\mathfrak{g}$ -module for which the induced $Z\mathfrak{g}$ -bimodule structure of 3.3.2 is symmetric in the sense of 2.4.1, then the $\mathfrak{J}$ -comodule structure is trivial.

Before proving 5.4.3, we prove a preliminary lemma on the induced $Z\mathfrak{g}\subseteq\operatorname{Q}$ bimodule structure on a $D(T)_{\operatorname{loc}}$ -comodule:

Lemma 5.4.4.

Assume $\tilde{C}$ is a comodule for $D(T)_{\operatorname{loc}}$ for which the induced right $Z\mathfrak{g}$ -module structure on $\tilde{C}$ agrees with the left $Z\mathfrak{g}$ -module structure obtained by restriction of scalars from ${\operatorname{Q}}$ . Then the comodule structure on $\tilde{C}$ is trivial.

Proof.

The category of comodules for $D(T)$ is equivalent to the category of Harish-Chandra bimodules for $T$ by 3.3.2(iv). Thus we may write $\tilde{C}\cong\oplus_{\lambda\in\mathfrak{t}^{\ast}}\tilde{C}_{\lambda}$ where $c\in C_{\lambda}$ implies $c_{(0)}\otimes c_{(1)}=\lambda\otimes c$ . Thus the induced $\operatorname{Sym}(\mathfrak{t})$ -bimodule structure has support contained in the ind-closed subscheme given by the set theoretic image of the map $\text{act},\text{proj}:X^{\bullet}(T)\times\mathfrak{t}^{\ast}\xrightarrow{}% \mathfrak{t}^{\ast}\times\mathfrak{t}^{\ast}$ and it suffices to show that the support of this $\operatorname{Sym}(\mathfrak{t})$ -bimodule is contained in the image of the diagonal map $\Delta_{\mathfrak{t}^{\ast}}$ . Since the left $\operatorname{Sym}(\mathfrak{t})$ is obtained by restriction of scalars from an ${\operatorname{Q}}$ -module, the support of $\tilde{C}$ is in fact contained in the disjoint union of the ${\operatorname{Q}}$ -points $\Gamma_{\lambda}^{\text{gen}}\in(\mathfrak{t}^{\ast}\times\mathfrak{t}^{\ast})% ({\operatorname{Q}})$ where, by definition, $\Gamma_{\lambda}^{\text{gen}}$ is the $k$ -point given by restricting the map (act, proj) to $\{\lambda\}\times\operatorname{{\mathrm{Spec}}}({\operatorname{Q}})$ where we identify $\operatorname{{\mathrm{Spec}}}({\operatorname{Q}})$ with the generic point of $\mathfrak{t}^{\ast}$ .

Recall that the associated map of affine schemes given by the Harish-Chandra map is isomorphic to the quotient $q:\mathfrak{t}^{\ast}\to\mathfrak{t}^{\ast}\!\sslash W$ of the Weyl group by the $\rho$ -shifted action (aka dot-action). We claim that restriction of the map $(q\times\text{id})$ to $X^{\bullet}(T)\times\operatorname{{\mathrm{Spec}}}({\operatorname{Q}})$ is an injective map. Indeed, since $(q,\text{id})$ is a finite flat map, the preimage of $(q,\text{id})(\Gamma_{\lambda}^{\text{gen}})$ has at most $|W|$ elements in it. However, the preimage of $(q,\text{id})(\Gamma_{\lambda}^{\text{gen}})$ is precisely the points $\Gamma_{w\lambda}^{\text{gen}}$ for $w\in W$ . Since $\Gamma_{w\lambda}^{\text{gen}}=\Gamma_{\nu}^{\text{gen}}$ for some $\nu\in X^{\bullet}(T)$ if and only if $w=1$ and $\nu=\lambda$ , we obtain our desired injectivity.

Our assumption implies that, as an object of $\operatorname{QCoh}(\mathfrak{t}^{\ast}\!\sslash W\times\mathfrak{t}^{\ast})$ , $\tilde{C}$ has support on $(q\times\text{id})(\Gamma_{0}^{\text{gen}})$ . The injectivity implies therefore that $\tilde{C}$ has support on $\Gamma_{0}^{\text{gen}}$ , and thus the left and right $\operatorname{Sym}(\mathfrak{t})$ -bimodule structures agree, as desired. ∎

Proof of 5.4.3.

One has a $Z\mathfrak{g}$ -module direct sum decomposition $\mathfrak{J}=Z\mathfrak{g}\oplus J$ where $J$ is the kernel of the counit map ${\mathfrak{J}}\to Z\mathfrak{g}$ . It suffices to show that the coaction map $p$ factors through the natural map $Z\mathfrak{g}\otimes_{Z\mathfrak{g}}C\to\mathfrak{J}\otimes_{Z\mathfrak{g}}C$ since if $p(c)=1\otimes c_{0}$ for some $c_{0}$ in $C$ then the identity axiom of comodules forces $c_{0}=c$ . Since $C$ is flat, one readily checks that the natural maps

(5.4.5)

\alpha:C\to\operatorname{Q}\otimes_{Z\mathfrak{g}}C\text{, and }\beta:% \mathfrak{J}\otimes_{Z\mathfrak{g}}C\to\mathfrak{J}_{\operatorname{loc}}% \otimes_{\operatorname{Q}}C_{\operatorname{loc}}

are injective. Letting $p_{\operatorname{loc}}:C_{\operatorname{loc}}\to\mathfrak{J}_{\operatorname{% loc}}\otimes_{\operatorname{Q}}C_{\operatorname{loc}}$ denote the induced $\mathfrak{J}_{\operatorname{loc}}$ -comodule structure on $C_{\operatorname{loc}}$ , we see that the maps of 5.4.5 intertwine $p$ and $p_{\operatorname{loc}}$ in the sense that $p_{\operatorname{loc}}\circ\alpha=\beta\circ p$ . Therefore by the injectivity of $\beta$ , it suffices to show that the image of $p_{\operatorname{loc}}$ factors through the corresponding summand of $\mathfrak{J}_{\operatorname{loc}}$ .

Since for any $c\in C$ we have by assumption that $\epsilon(c_{(0)}z)c_{(1)}=zc$ , by ${\operatorname{Q}}$ -linearity we have that $\epsilon_{\operatorname{loc}}(\tilde{c}_{(0)}z)\tilde{c}_{(1)}$ for any $\tilde{c}\in C_{\operatorname{loc}}$ . Thus using the isomorphism of 5.3.2, we see that $C_{\operatorname{loc}}$ is a $D(T)_{\operatorname{loc}}$ -comodule whose induced right $Z\mathfrak{g}$ -module agrees with the restriction of scalars from ${\operatorname{Q}}$ . Therefore by 5.4.4, $p_{\operatorname{loc}}(c)=1\otimes c$ and so we see in particular that $p_{\operatorname{loc}}$ factors through ${\operatorname{Q}}\cdot 1\otimes_{\operatorname{Q}}C_{\operatorname{loc}}$ , as desired. ∎

Recall the $(D(G),D(T)^{W})$ -bimodule $\mathbf{N}$ and the induced $D(T)^{W}$ -action on $\kappa({\mathbf{N}})$ , see Section 5.2.

Proposition 5.4.6.

${{\enspace}\sf{(i)}}\;$ $\mathbf{N}$ is a coalgebra object in $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}$ .

${\;\sf{(ii)}}\;$ The induced $\mathfrak{J}$ -comodule structure on $\kappa({\mathbf{N}})$ is trivial.

Proof.

To prove (i) consider the functor $D(G)\otimes_{{}_{U\mathfrak{g}\otimes U\mathfrak{g}_{\text{op}}}}(\text{-}):\,% U\mathfrak{g}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}\to D(G)% \operatorname{{\text{-}{\mathrm{mod}}}}$ . This functor is a left adjoint to the monoidal forgetful functor $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}\to U\mathfrak{g}\operatorname{{% \mathrm{\text{-}\mathrm{bimod}}}}$ , see Section 2.3, hence this functor is left lax monoidal by 4.1.6. Since the diagonal bimodule $U\mathfrak{g}$ is a coalgebra object in $U\mathfrak{g}\operatorname{{\mathrm{\text{-}\mathrm{bimod}}}}$ , it follows that $D(G)/D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}\cong D(G)\otimes_{{}_{U% \mathfrak{g}\otimes U\mathfrak{g}_{\text{op}}}}U\mathfrak{g}$ is an coalgebra object in $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}$ . This implies (i) thanks to an obvious isomorphism

\mathbf{N}=D(G)/D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}\,\cong\,D(G)% \otimes_{{}_{U\mathfrak{g}\otimes U\mathfrak{g}_{\text{op}}}}U\mathfrak{g}.

The resulting comultiplication $\mathbf{N}\to\mathbf{N}\otimes_{U\mathfrak{g}}\mathbf{N}$ can also be described explicitly in terms of the comultiplication $\Delta$ for $D(G)$ considered in Section 2.3. Specifically, it is straightforward to check that the composite

D(G)\xrightarrow{\Delta}D(G)\otimes_{U\mathfrak{g}}D(G)\xrightarrow{}{\mathbf{% N}}\otimes_{U\mathfrak{g}}{\mathbf{N}}

sends the left ideal $D(G)\operatorname{{\mathrm{ad}}}\mathfrak{g}$ to zero. Hence, this composite factors through a map $\mathbf{N}\to\mathbf{N}\otimes_{U\mathfrak{g}}\mathbf{N}$ , which is the desired comultiplication for $\mathbf{N}$ .

To prove (ii) it suffices to show, similarly to the proof of 5.4.3, that the coaction map $\kappa({\mathbf{N}})$ is trivial it suffices to show that the coaction map factors through the image of $Z\mathfrak{g}\otimes_{Z\mathfrak{g}}\kappa({\mathbf{N}})\subseteq\mathfrak{J}% \otimes_{Z\mathfrak{g}}\kappa({\mathbf{N}})$ . As in [Gin, Section 6.2], Skryabin’s equivalence identifies $\check{\mathbb{M}}^{W}$ with a free rank one $\mathfrak{J}$ -module. Therefore by 5.2.3 we see that $\kappa({\mathbf{N}})$ is a free rank one $\mathfrak{J}$ -module and so is flat as a $Z\mathfrak{g}$ -module by 4.1.4. Moreover, the $U\mathfrak{g}$ -bimodule structure on ${\mathbf{N}}$ naturally upgrades to a $U\mathfrak{g}\otimes_{Z\mathfrak{g}}U\mathfrak{g}_{\text{op}}$ -module structure since ${\mathbf{N}}$ is in fact a $\mathcal{D}$ -module. Therefore the left and right $Z\mathfrak{g}$ -module structures on the $\mathfrak{J}$ -comodule $\kappa({\mathbf{N}})$ agree. Therefore, by 5.4.3, the $\mathfrak{J}$ -comodule structure is trivial, as desired. ∎

6. Construction of Knop-Ngô functor

6.1. Relative Drinfeld center

Given a monoidal functor $F:\mathcal{R}\to\mathcal{H}$ (which we will later take to be the free functor $U\mathfrak{g}\otimes_{k}(\text{-}):\text{Rep}(G)\to\mathcal{HC}$ ) we define its relative Drinfeld center as follows. It is the category $\mathcal{Z}_{F}(\mathcal{H})$ whose objects are pairs $(B,Z_{(\text{-})})$ where $Z_{(\text{-})}$ is a natural isomorphism $B\otimes F(\text{-})\xrightarrow{\sim}F(\text{-})\otimes B$ and such that the following diagram commutes for all $R_{1},R_{2}\in\mathcal{R}$ :

(6.1.1)

where $m$ is the monoidality isomorphism for $F$ , and morphisms $(B^{1},Z_{(\text{-})}^{1})\xrightarrow{}(B^{2},Z_{(\text{-})}^{2})$ are given by maps $f:B^{1}\to B^{2}$ in $\mathcal{H}$ which intertwine the maps $Z^{i}_{(\text{-})}$ in the sense that $Z^{2}_{(\text{-})}(f\otimes\text{id}_{F(\text{-})})=(\text{id}_{F(\text{-})}% \otimes f)Z^{1}_{(\text{-})}$ as natural transformations $B^{1}\otimes F(\text{-})\xRightarrow{}F(\text{-})\otimes B^{2}$ . It is not difficult to check that $\mathcal{Z}_{\mathcal{R}}(\mathcal{H})$ admits a monoidal structure, where monoidality is given by the formula

(B^{1},Z^{1}_{\bullet})\otimes(B^{2},Z^{2}_{\bullet}):=(B^{1}\otimes B^{2},Z^{% 1,2}_{\bullet})

with $Z^{1,2}_{R}:=(Z^{1}_{R}\otimes\text{id})\circ(\text{id}_{B^{1}}\otimes Z^{2}_{% R})$ , and unit ( $\mathbf{1}_{\mathcal{H}},Z^{\mathbf{1}})$ defined from the unit isomorphisms as in [EtingofGelakiNikshychOstrikTensorCategories, Chapter 7.13].

In the special case where $F$ is the identity functor on $\mathcal{H}$ the this recovers the usual Drinfeld center construction, with braiding given by the formula

Z^{1}_{B_{2}}:(B^{1},Z^{1}_{\bullet})\otimes(B^{2},Z^{2}_{\bullet})% \xrightarrow{\sim}(B^{2},Z^{2}_{\bullet})\otimes(B^{1},Z^{1}_{\bullet}).

Unlike the Drinfeld center, the relative Drinfeld center is not braided in general.

We record the following observation, which is a routine check:

Proposition 6.1.2.

Assume $K:\mathcal{H}\to\mathcal{C}$ is a monoidal functor. There is an induced monoidal functor

\underline{K}:\mathcal{Z}_{F}(\mathcal{H})\to\mathcal{Z}_{KF}(\mathcal{C})

by the formula $(B,Z_{(\text{-})})\mapsto(K(B),Z_{K(\text{-})})$ where by definition $Z_{K(\text{-})}$ is the natural isomorphism $K(B)\otimes KF(\text{-})\xRightarrow{\sim}KF(\text{-})\otimes K(B)$ making the following diagram commute for every $R\in\mathcal{R}$ :

(6.1.3)

where the vertical arrows are the monoidality isomorphisms.

Remark 6.1.4.

We give two remarks on this construction which will not be used in what follows but provides motivation and context for our constructions:

${{\enspace}\sf{(i)}}\;$ One can also construct an equivalence of categories $\mathcal{Z}_{F}(\mathcal{H})\cong\text{Hom}_{\mathcal{R}\boxtimes\mathcal{R}}^% {\otimes}(\mathcal{R},\mathcal{H})$ in settings where $\mathcal{R}\boxtimes\mathcal{R}$ is defined. This gives a natural interpretation of the functor in 6.1.2: namely, the functor $\underline{K}$ is given by pullback.

${\;\sf{(ii)}}\;$ If $\mathcal{A}$ is a tensor category and $\mathcal{N},\mathcal{M}$ are module categories given by modules for some algebra $A_{\mathcal{N}},A_{\mathcal{M}}\in\mathcal{A}$ (respectively) then one can show $\text{Hom}_{\mathcal{A}}(\mathcal{N},\mathcal{M})$ is the category of $(A_{\mathcal{M}},A_{\mathcal{N}})$ -modules in $\mathcal{A}$ . Since $\mathcal{HC}=U\mathfrak{g}\operatorname{{\text{-}{\mathrm{mod}}}}(\text{Rep}(G))$ , one can use this to show that the relative Drinfeld center $\mathcal{Z}_{\text{Fr}}(\mathcal{HC})$ is given by the category of strongly $G$ -equivariant $D(G)$ -modules, which, combined with 6.4.4 below with $F:=\text{Fr}$ , gives an alternate proof of [BezrukavnikovFinkelbergOstrikCharacterDModulesViaDrinfeldCenterofHarishChandraBimodules, Lemma 3.8].

We thank Pavel Etingof for explaining these facts to us.

6.2. Construction of Knop-Ngô functor

Our construction is based on the following two theorems.

Theorem 6.2.1.

Any $M\in\mathfrak{J}$ -mod, viewed as a trivial $\mathfrak{J}$ -comodule, gives an element of the Drinfeld center of $\mathfrak{J}$ -comod whose central structure $M\otimes C\xrightarrow{\sim}C\otimes M$ has inverse given by the formula $c\otimes m\mapsto c_{(0)}m\otimes c_{(1)}$ . This assignment upgrades to a fully faithful, braided monoidal functor $\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to\mathcal{Z}(\mathfrak{J}% \operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ .

Theorem 6.2.1 will be proved in Section 8. The second theorem is the following general result on functoriality of Drinfeld center:

Theorem 6.2.2.

Let $\mathcal{R},\mathcal{H}$ and $\mathcal{C}$ be cocomplete abelian categories equipped with monoidal structures which commute with colimits separately in each entry and let

(6.2.3)

\mathcal{R}\xrightarrow{F}\mathcal{H}\xrightarrow{K}\mathcal{C}

a pair of monoidal functors between them such that the following holds:

${{\enspace}\sf{(i)}}\;$ Every object in $\mathcal{R}$ is a filtered colimit of projective dualizable objects.

${\;\sf{(ii)}}\;$ The functor $F$ commutes with colimits and its essential image forms a projective generating set.

${\sf{(iii)}}\;$ The functor $K$ admits a fully faithful right adjoint $K^{R}$ which commutes with filtered colimits. Then, there is a braided monoidal functor $\underline{K}:\mathcal{Z}(\mathcal{H})\to\mathcal{Z}(\mathcal{C})$ which lifts the monoidal functor $K$ , and this braided monoidal functor admits a fully faithful, braided lax monoidal right adjoint $\underline{K}^{R}$ which lifts the lax monoidal functor $K^{R}$ . Moreover, the unit and counit map for the adjoint pair $(\underline{K},\underline{K}^{R})$ are induced by the unit and counit map for $K$ .

Using 6.2.1, we define the Knop-Ngô functor $\operatorname{KN}$ to be the composite of the braided lax monoidal functors

\mathfrak{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to\mathcal{Z}(\mathfrak{J}% \operatorname{{\mathrm{\text{-}\mathrm{comod}}}})\xrightarrow{\underline{% \kappa}^{R}}\mathcal{Z}(\mathcal{HC})

where the first arrow is given by 6.2.1 and the lax braided monoidal functor $\underline{\kappa}^{R}$ is defined in 6.2.2, which satisfies the hypotheses of 6.2.2 by 4.1.1. Equipping a $\mathfrak{J}$ -module $M$ with a trivial $\mathfrak{J}$ -comodule structure via the counit map for $\mathfrak{J}$ , we see that the underlying object of $\operatorname{KN}(M)$ is

\kappa^{R}(M)\cong D^{\psi_{r}}\otimes_{Z\mathfrak{g}}^{\mathfrak{J}}B=D^{\psi% _{r},\mathfrak{J}}\otimes_{Z\mathfrak{g}}B\xleftarrow{\sim}U\mathfrak{g}% \otimes_{Z\mathfrak{g}}B

where the final isomorphism is given by 4.1.1(iv). Thus this functor is given on objects by tensor product $U\mathfrak{g}\otimes_{Z\mathfrak{g}}(\text{-})$ and so the braided lax monoidal structure on $\operatorname{KN}$ in fact has the property of being braided monoidal, and is fully faithful as it is the composite of functors which are fully faithful by 6.2.1 and 6.2.2. By light abuse of notation, we identify the Knop-Ngô functor with its preimage under the global sections isomorphism $\Gamma:D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G_{\text{ad}}}\xrightarrow% {\sim}\mathcal{Z}(\mathcal{HC})$ of [BezrukavnikovFinkelbergOstrikCharacterDModulesViaDrinfeldCenterofHarishChandraBimodules, Lemma 3.8]. This completes the proof of 1.1.5(i). ∎

Proof of 1.1.6.

Consider the composite, read left to right, of the maps

(6.2.4)

{\mathbf{N}}\xrightarrow{u({\mathbf{N}})}\kappa^{R}\kappa({\mathbf{N}})\cong% \kappa^{R}(\mathfrak{J})=\kappa^{R}|_{\mathfrak{J}\operatorname{{\mathrm{\text% {-}\mathrm{comod}}}}_{\text{triv}}}(\mathfrak{J})\cong U\mathfrak{g}\otimes_{Z% \mathfrak{g}}\mathfrak{J}

where the first map of coalgebra objects is the unit of the adjunction given by 4.1.1, the second map is given by using the fact that $\kappa({\mathbf{N}})$ is canonically a coalgebra in $\mathfrak{J}$ -modules which is free rank one $\mathfrak{J}$ -module, using 5.2.3, the equality of coalgebras is given by the fact that the comodule structure on $\kappa({\mathbf{N}})$ is trivial by 5.4.6, and the final isomorphism is given by 5.4.1. Since each usage of a natural transformation is a monoidal transformation of strongly monoidal functors, this map is a map of coalgebra objects. Now, setting $F:=\text{Fr}$ and $K:=\kappa$ , we see that by 6.2.2 this unit map can be upgraded to a map in $\mathcal{Z}(\mathcal{HC})$ , as required. This morphism is a map of a quotient of a free $D(G)$ -module which sends $1$ to $1\otimes 1$ , and therefore is given by acting on $1\otimes 1$ . ∎

Remark 6.2.5.

The composite of morphisms 6.2.4 giving rise to the quantization of the Knop-Ngô morphism as a map of Harish-Chandra bimodules does not require any of our above results on the relative Drinfeld center. In particular, using our arguments, one can construct the morphism 1.1.4, as a map of Harish-Chandra bimodules, without defining the functor $\operatorname{KN}$ .

6.3.

In this subsection we prove some results required for the proof of 6.2.2. The proof of the theorem will be completed in Section 6.4.

In Section 6.3, we work under the assumption that conditions (i)-(ii) of 6.2.2 hold.

Observe first that, for any $C\in\mathcal{C}$ and $B\in\mathcal{H}$ , we have maps

(6.3.1)

K^{R}(C)\otimes_{\mathcal{H}}B\xrightarrow{\text{id}\otimes u(B)}K^{R}(C)% \otimes_{\mathcal{H}}K^{R}K(B)\xrightarrow{}K^{R}(C\otimes_{\mathcal{C}}K(B))

where the second map is given by the lax monoidality morphism for $K^{R}$ . Similarly, we have maps

(6.3.2)

B\otimes_{\mathcal{H}}K^{R}(C)\xrightarrow{u(B)\otimes\text{id}}K^{R}K(B)% \otimes_{\mathcal{H}}K^{R}(C)\xrightarrow{}K^{R}(K(B)\otimes C).

We now show that the arguments of [FlakeLaugwitzPosurProjectionFormulasandInducedFunctorsonCentersofMonoidalCategories] imply a relative verison of [FlakeLaugwitzPosurProjectionFormulasandInducedFunctorsonCentersofMonoidalCategories, Theorem B]:

Proposition 6.3.3.

The functor

\underline{K}:\mathcal{Z}_{F}(\mathcal{H})\to\mathcal{Z}_{KF}(\mathcal{C})

admits a lax monoidal right adjoint $\underline{K}^{R}$ which can be defined on objects by the formula

\underline{K}^{R}(C,Z^{1}_{KF(\text{-})}):=(K^{R}(C),\tilde{Z}^{1}_{F(\text{-}% )})

where $\tilde{Z}^{1}_{F(\text{-})}$ is the unique natural transformation such that for all $V\in\mathcal{R}$ , the following diagram commutes:

(6.3.4)

where the vertical arrows are given by the projection formulas. The functor $\underline{K}^{R}$ is fully faithful if $K^{R}$ is fully faithful.

Proof.

The projection formulas of (6.3.1) and (6.3.2) are isomorphisms for dualizable objects by [FlakeLaugwitzPosurProjectionFormulasandInducedFunctorsonCentersofMonoidalCategories, Corllary 3.19]. Since by assumption $K^{R}$ commutes with filtered colimits and $K$ commutes with colimits as it is a left adjoint, the projection formula holds for filtered colimits of dualizable objects. Therefore, $\tilde{Z}^{1}_{F(V)}$ is defined.

We claim that, for any $(B,Z_{F(\text{-})})\in\mathcal{Z}_{F}(\mathcal{H})$ , respectively $(C,Z^{\prime}_{KF(\text{-})})\in\mathcal{Z}_{KF}(\mathcal{C})$ , the unit map $u(B)$ , respectively the counit map $c(C)$ , for the adjunction $(K,K^{R})$ is a map in $\mathcal{Z}_{F}(\mathcal{H})$ , respectively $\mathcal{Z}_{KF}(\mathcal{C})$ . We show the claim for the unit map; the counit map is similar. Letting $B=F(B)$ , we observe that the following diagram commutes:

where the unlabelled maps are the monoidality isomorphisms. Indeed, the leftmost box commutes by naturality, the middle two boxes commute by the respective definitions in (6.1.3) and (6.3.4), and the top and bottom squares commute because the unit and counit for the adjoint pair $(K,K^{R})$ are monoidal natural transfomations. Reading the exterior of this diagram shows that $u(B)$ is a map in $\mathcal{Z}_{F}(\mathcal{H})$ as desired.

Applying 4.1.6 to our functor $K$ yields a lax monoidal structure on $\underline{K}^{R}$ . Finally, the fully faithfulness claim follows since a right adjoint is fully faithful if and only if the counit is an isomorphism. ∎

6.4. Proof of 6.2.2

We now assume all three conditions (i)-(iii) of 6.2.2 hold.

As $K$ is monoidal, there is a monoidal functor $\mathcal{P}_{K}:\mathcal{Z}(\mathcal{C})\to\mathcal{Z}_{K}(\mathcal{C})$ given on objects by the formula $(C,Z_{(\text{-})})\mapsto(C,Z_{K(\text{-})})$ .

Proposition 6.4.1.

The functor $\mathcal{P}_{K}$ is a monoidal equivalence of categories.

Proof.

This is given by standard arguments: we give a proof for the convenience of the reader. This functor is evidently faithful; we now argue it is full: let $f:C_{1}\to C_{2}$ be a map in $\mathcal{C}$ inducing a map in $\mathcal{Z}_{K}(\mathcal{C})$ . We wish to show that

(6.4.2)

commutes. However, observe that this diagram naturally fits as the front face of a cube whose ‘back face’ commutes by assumption that $f$ induces a map in $\mathcal{Z}_{K}(\mathcal{C})$ and whose four other non-front faces commute by naturality of $Z^{1},Z^{2}$ , and the counit of our adjunction. Therefore (6.4.2) commutes and so $\mathcal{P}_{K}$ is fully faithful.

Finally, we argue $\mathcal{P}_{K}$ is essentially surjective: if $(C_{1},Z^{1}_{K(B)})\in\mathcal{Z}_{K}(\mathcal{C})$ , we define an extension $Z^{1}_{C}$ as the unique map which makes the following diagram commute:

where the horizontal arrows induced by the counit, which are isomorphisms as $K$ is fully faithful by assumption. Using this definition along with the fact that the unit and counit of this adjunction are monoidal natural transformations, one can indeed check that $(C,Z_{C}^{1})\in\mathcal{Z}(\mathcal{C})$ . Now observe that if $C=K(B)$ the diagram

where $u(B)$ is the unit map, by construction of $Z^{1}_{C}$ and naturality of $Z^{1}_{K(\text{-})}$ . The composite of the horizontal arrows in both rows of the diagram give the identity functor by the unit counit criterion of the adjunction, and so we see that $\mathcal{P}_{K}(C,Z_{C}^{1})=(C,Z_{K(B)}^{1})$ as required. ∎

Using that $\mathcal{P}_{K}$ is a monoidal equivalence of categories by 6.4.1, we may use $\mathcal{P}_{K}$ to equip $\mathcal{Z}_{K}(\mathcal{C})$ with a braiding.

Corollary 6.4.3.

The functor $\underline{K}:\mathcal{Z}(\mathcal{H})\to\mathcal{Z}_{K}(\mathcal{C})$ defined in 6.1.2 is braided monoidal.

Proof.

Let $(B_{1},Z^{1}),(B_{2},Z^{2})\in\mathcal{Z}(\mathcal{H})$ . Applying the substitutions $B:=B_{1},R:=B_{2}$ , and $F=\text{id}$ in the diagram (6.1.3), we see that the upper horizontal arrow is $K(Z^{1}_{B_{2}})$ and the lower horizontal arrow is the braiding morphism in $\mathcal{Z}_{K}(\mathcal{C})$ by construction of the braiding. The fact that this diagram commutes thus proves that $\underline{K}$ is braided. ∎

Finally, we prove the following Proposition which informally states that, in the above notation, the categories $\mathcal{Z}(\mathcal{H})$ and $\mathcal{Z}_{K}(\mathcal{C})$ are determined by the restriction of the central structure to free objects:

Proposition 6.4.4.

The functors

\mathcal{Z}(\mathcal{H})\xrightarrow{\mathcal{P}_{F}}\mathcal{Z}_{F}(\mathcal{% H})\text{, and }\mathcal{Z}_{K}(\mathcal{C})\xrightarrow{\mathcal{P}_{F}}% \mathcal{Z}_{KF}(\mathcal{C})

are monoidal equivalences of abelian categories.

We first recall the following standard categorical lemma:

Lemma 6.4.5.

Assume $\mathcal{A}$ is a cocomplete abelian category and $\mathcal{P}$ is a full subcategory of projective generators, and assume $F,G:\mathcal{A}\to\mathcal{B}$ are functors to some abelian category $\mathcal{B}$ which commute with colimits. Then any natural transformation of restricted functors $\eta:F|_{\mathcal{P}}\xRightarrow{}G|_{\mathcal{P}}$ extends uniquely to a natural transformation $\widetilde{\eta}:F\xRightarrow{}G$ which is a natural isomorphism if $\eta$ is a natural isomorphism.

Proof of 6.4.4.

We show the latter functor is a monoidal equivalence, the former functor is an essentially identical argument. The monoidality morphisms are given by the identity maps. The functor $\mathcal{P}_{F}$ is faithful since the forgetful functor to $\mathcal{C}$ is faithful; we now show it is full. Assume $(C_{1},Z^{1}_{K(\text{-})}),(C_{2},Z^{2}_{K(\text{-})})\in\mathcal{Z}_{K}(% \mathcal{C})$ and $f:C^{1}\to C^{2}$ is a map for which we have the equality

(\text{id}\otimes f)\circ Z^{1}_{KF(\text{-})}=Z^{2}_{KF(\text{-})}\circ(f% \otimes\text{id})

of natural transformations. Then the two natural transformations $(\text{id}\otimes f)\circ Z^{1}_{K(\text{-})},Z^{2}_{K(\text{-})}\circ(f% \otimes\text{id})$ have the property that they agree upon restriction to a projective generating set, so they agree by the uniqueness statement of 6.4.5. (Note that our assumptions that $F,K$ , and the monoidal structure on $\mathcal{C}$ in the first argument and the second argument all commute with colimits, and so these functors commute with colimits.) Similarly, to show that this functor is essentially surjective, observe that if we are given some $(C_{1},Z^{1}_{KF(\text{-})})\in\mathcal{Z}_{KF}(\mathcal{C})$ we may use the existence statement in 6.4.5 to extend $Z^{1}_{KF(\text{-})}$ to a natural transformation $Z^{1}_{K(\text{-})}$ of functors $C_{1}\otimes K(\text{-})\xRightarrow{}K(\text{-})\otimes C_{1}$ . ∎

As above, we use the braidings on the domain categories in 6.4.4 to equip the respective codomains with braidings for which the functors appearing in the statement of 6.4.4 are braided monoidal.

Proof of 6.2.2.

We have shown in 6.4.4 and 6.4.1 that the horizontal arrows in the commutative diagram

give monoidal equivalences of categories which, by construction, are braided. Moreover, the left vertical arrow is braided by 6.4.3. Therefore by 6.3.3 we obtain our desired fully faithful right adjoint; the fact that it is braided lax monoidal immediately follows from 4.1.6. ∎

7. Knop-Ngô functor and parabolic restriction

In this section, we first prove 1.1.5(ii) and then use it to prove 1.3.2. After that we complete the proof of 1.2.5. Recall that we set ${\hbar}=1$ in Section 5; we continue to work in this setting in Section 7.

7.1. Proof of 1.1.5(ii)

The functor $\operatorname{{\mathrm{Res}}}^{G}_{T}:D(G)\operatorname{{\text{-}{\mathrm{mod}% }}}^{G_{\text{ad}}}\to D(T)\rtimes W\operatorname{{\text{-}{\mathrm{mod}}}}$ of parabolic restriction is a right adjoint of $\operatorname{{\mathrm{Ind}}}_{T}^{G}$ . Hence, it follows from the isomorphism 5.2.1 and standard tensor- $\operatorname{\mathrm{Hom}}$ adjunctions, that one has isomorphisms of functors

(7.1.1)

\operatorname{{\mathrm{Res}}}^{G}_{T}(\text{-})^{W}\cong\operatorname{\mathrm{% Hom}}_{D(G)}(\mathbf{N},\text{-})\cong(\text{-})^{G}.

which also directly follows from [GinzburgParabolicInductionandtheHarishChandraDModule, Theorem 4.4]. Thus, for any $M\in D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G_{\text{ad}}}$ , the counit of the adjunction gives a morphism of $D(G)$ -modules $\alpha:\operatorname{{\mathrm{Ind}}}^{G}_{T}(M^{G})^{W}\to M$ . More explicitly, under the identification $\operatorname{{\mathrm{Ind}}}^{G}_{T}(M^{G})^{W}\cong{\mathbf{N}}\otimes_{% \mathcal{D}(T)^{W}}M^{G}$ of 5.2.1, the map $\alpha$ corresponds to the map ${\mathbf{N}}\otimes_{\mathcal{D}(T)^{W}}M^{G}\to M$ induced from the fact that the action map $D(G)\otimes M^{G}\to M,\,u\otimes m\mapsto um$ factors through the quotient

\displaystyle D(G)\otimes M^{G}\,\twoheadrightarrow\,D(G)/D(G)\operatorname{{% \mathrm{ad}}}\mathfrak{g}\otimes M^{G}

\displaystyle\,\twoheadrightarrow\,(D(G)/D(G)\operatorname{{\mathrm{ad}}}% \mathfrak{g})\otimes_{D(T)^{W}}M^{G}

\displaystyle=\mathbf{N}\otimes_{D(T)^{W}}M^{G}.

Lemma 7.1.2.

The map $\alpha$ is injective.

Proof.

The kernel of $\alpha$ is a $D(G)$ -submodule $K\subseteq\mathbf{N}\otimes_{D(T)^{W}}M^{G}$ . Since $(\mathbf{N}\otimes_{D(T)^{W}}M^{G})^{G}=M^{G}$ , the exactness of the functor $(\text{-})^{G}$ on the category of strongly $G$ -equivariant $D(G)$ -modules implies that $\operatorname{{\mathrm{Res}}}^{G}_{T}(K)^{W}=K^{G}=0$ . On the other hand, the functor $\operatorname{{\mathrm{Res}}}^{G}_{T}(-)^{W}$ is isomorphic to a left adjoint of $\operatorname{{\mathrm{Ind}}}^{G}_{T}(-)^{W}$ , by the Second Adjointness theorem. Therefore, we obtain $\operatorname{\mathrm{Hom}}(K,\operatorname{{\mathrm{Ind}}}(M^{G})^{W})=% \operatorname{\mathrm{Hom}}(\operatorname{{\mathrm{Res}}}_{T}^{G}(K)^{W},M^{G}% )=0$ , which would contradict the existence of an injective morphism $K\,\hookrightarrow\,\mathbf{N}\otimes_{D(T)^{W}}M^{G}$ if $K\neq 0$ . ∎

Proof of 1.1.5(ii).

Using the isomorphisms

U\mathfrak{g}\otimes_{Z\mathfrak{g}}(\text{-})\,\cong\,(U\mathfrak{g}\otimes_{% Z\mathfrak{g}}{\mathfrak{J}})\otimes_{{\mathfrak{J}}}(\text{-}),\enspace\text{% resp.}\enspace\mathbf{N}\otimes_{D(T)^{W}}(\text{-})\,\cong\,(\mathbf{N}% \otimes_{D(T)^{W}}{\mathfrak{J}})\otimes_{{\mathfrak{J}}}(\text{-}),

we see that proving 1.1.5(ii) reduces to proving that the map

(7.1.3)

{\mathbf{N}}\otimes_{D(T)^{W}}{\mathfrak{J}}\ \xrightarrow{}\ {U\mathfrak{g}}% \otimes_{Z\mathfrak{g}}{\mathfrak{J}},\enspace u1_{{\mathbf{N}}}\otimes j\,% \mapsto\,u(1\otimes j),

is an isomorphism of $(D(G),{\mathfrak{J}})$ -bimodules. The surjectivity of the map is clear since the element $1\otimes 1$ generates $U\mathfrak{g}\otimes_{Z\mathfrak{g}}\mathfrak{J}$ as a $(U\mathfrak{g}\otimes{\mathfrak{J}}_{\text{op}})$ -module. To prove injectivity, we observe that

(\mathbf{N}\otimes_{D(T)^{W}}{\mathfrak{J}})^{G}\,=\,1\otimes{\mathfrak{J}}\,=% \,({U\mathfrak{g}}\otimes_{Z\mathfrak{g}}{\mathfrak{J}})^{G}.

Therefore, the restriction of the map 7.1.3 to $G$ -invariants is the identity map. Hence, this map is an isomorphism and the desired injectivity now follows from 7.1.2 ∎

7.2. Proof of 1.3.2

We now prove 1.3.2. In fact, we prove 1.3.2 in the following more general setting:

Theorem 7.2.1.

Assume that $\mathbf{V}$ is a representation of $G^{\vee}$ such that $\mathbf{V}^{T^{\vee}}=\{0\}$ . Then $\mathbf{\Phi}_{G,\mathbf{V}}$ is convolution exact.

Proof.

Using 2.3, we see that suffices to prove that $\mathbf{\Phi}_{G,\mathbf{V}}$ is flat as a right $U\mathfrak{g}$ -module. For a $G^{\vee}$ -representation $V$ , we let $\mathbf{\Phi}_{T,\mathbf{V}}$ denote the $W$ -equivariant $\gamma$ -sheaf associated to the restricted representation for the Langlands dual torus. By combining [ChenAVanishingConjecturetheGLnCase, Theorem 1.4, Section 1.6] and the non-holonomic variant of [Gin, Theorem 1.5.1] (which holds with the same proof) the $D(T)\rtimes W$ -module structure on $\mathbf{\Phi}_{T,\mathbf{V}}$ upgrades to a module for $\mathfrak{J}$ and we have an isomorphism

(7.2.2)

\operatorname{Sym}(\mathfrak{t})\otimes_{\operatorname{Sym}(\mathfrak{t})^{W}}% \mathbf{\Phi}_{T,\mathbf{V}}^{W}\xrightarrow{\sim}\mathbf{\Phi}_{T,\mathbf{V}}

induced by the $\operatorname{Sym}(\mathfrak{t})$ -module structure. Thus for any $D(G)$ -module $M$ we have isomorphisms

	$\displaystyle\mathbf{\Phi}_{G,\mathbf{V}}\star M\cong\mathbf{\Phi}_{G,\mathbf{% V}}\otimes_{U\mathfrak{g}}M$	$\displaystyle\cong\operatorname{Ind}_{T}^{G}(\mathbf{\Phi}_{T,\mathbf{V}})^{W}% \otimes_{U\mathfrak{g}}M$
		$\displaystyle\cong({\mathbf{N}}\otimes_{D(T)^{W}}\mathbf{\Phi}_{T,\mathbf{V}}^% {W})\otimes_{U\mathfrak{g}}M\cong\operatorname{KN}(\mathbf{\Phi}_{T,\mathbf{V}% }^{W})\otimes_{U\mathfrak{g}}M\cong\mathbf{\Phi}_{T,\mathbf{V}}^{W}\otimes_{Z% \mathfrak{g}}M$

where the first isomorphism is given by 2.3, the second follows by the definition of $\mathbf{\Phi}_{G,\mathbf{V}}$ , the third is obtained from 5.2.2, the fourth follows from using 1.1.5(ii), and the final follows from the definition of the functor $\operatorname{KN}$ . Therefore it suffices to prove that $\mathbf{\Phi}_{T,\mathbf{V}}^{W}$ is flat as a $Z\mathfrak{g}$ -module.

Observe by the Chevalley-Shephard-Todd theorem [ChevalleyInvariantsofFiniteGroupsGeneratedByReflections], $\operatorname{Sym}(\mathfrak{t})$ is a faithfully flat $Z\mathfrak{g}$ -module. Therefore it suffices to prove that the $\operatorname{Sym}(\mathfrak{t})$ -module $\operatorname{Sym}(\mathfrak{t})\otimes_{\operatorname{Sym}(\mathfrak{t})^{W}}% \mathbf{\Phi}_{T,\mathbf{V}}^{W}$ is flat and, thus, using the fact that (7.2.2) is an isomorphism, it suffices to show $\mathbf{\Phi}_{T,\mathbf{V}}$ is flat as a $\operatorname{Sym}(\mathfrak{t})$ -module or, equivalently by 2.3, that $\mathbf{\Phi}_{T,\mathbf{V}}$ is convolution exact. We may identify the representation $\rho|_{T^{\vee}}$ with a product of characters $\lambda_{i}$ of $T^{\vee}$ for $1\leq i\leq n$ ; our assumption on $\rho$ implies that $\lambda_{i}\neq 0$ for all $i$ . Viewing the $\lambda_{i}$ as maps $\mathbb{G}_{m}\to T$ , it is not difficult to exhibit an isomorphism

\mathbf{\Phi}_{T,\mathbf{V}}\cong\int_{\lambda_{1}}\mathcal{E}\star...\star% \int_{\lambda_{n}}\mathcal{E}

where $\mathcal{E}$ is the restriction of the exponential $\mathcal{D}$ -module on $\mathbb{A}^{1}$ to the open subset $\mathbb{G}_{m}$ , see also [BravermanKazhdanGammaSheavesonReductiveGroups, (4.6)]. Therefore it suffices to prove that $\int_{\lambda_{i}}\mathcal{E}$ is convolution exact for all $i$ . In fact we claim that $\int_{\lambda}\mathcal{F}$ is convolution exact for any nonzero character $\lambda:\mathbb{G}_{m}\to T$ and any $D(\mathbb{G}_{m})$ -module $\mathcal{F}$ which is torsion free as a module for the ring $k[\partial]=D(\mathbb{G}_{m})^{\mathbb{G}_{m}}$ , where $\partial$ is the differential operator $t\frac{d}{dt}$ on $\mathbb{G}_{m}=\text{Spec}(k[t^{\pm 1}])$ ; since $\mathcal{E}$ is readily checked to be torsion free as a $k[\partial]$ -module, this obviously implies that $\int_{\lambda_{i}}\mathcal{E}$ is convolution exact.

To prove this claim, fix a nonzero map $\lambda$ and such a $D(\mathbb{G}_{m})$ -module $\mathcal{F}$ . We may factor $\lambda$ as a composition of a quotient map $q$ and a closed embedding. Observe that, as $\lambda$ is nonzero, the codomain of $q$ is isomorphic to $\mathbb{G}_{m}$ in such a way that, under this isomorphism, $q$ corresponds to a map $x\mapsto x^{n}$ for $n\neq 0$ . One can directly compute that de Rham pushforward by such a map preserves the property of being $k[\partial]$ -torsion free, and so we are reduced to the case where $\lambda$ is a closed embedding. In this case, we may choose a splitting of the short exact sequence induced by $\lambda$ and identify $\lambda$ with the map $(\text{id},e):\mathbb{G}_{m}\to\mathbb{G}_{m}\times T/\mathbb{G}_{m}$ . Since $\int_{(\text{id},e)}\mathcal{F}$ is readily checked to have global sections $\mathcal{F}\otimes_{k}\operatorname{{\mathrm{Sym}}}(\mathfrak{t}/\text{Lie}(% \mathbb{G}_{m}))$ as a module for $\operatorname{Sym}(\mathfrak{t})\cong k[\partial]\otimes\operatorname{{\mathrm% {Sym}}}(\mathfrak{t}/\text{Lie}(\mathbb{G}_{m}))$ , we see that the global sections of $\int_{(\text{id},e)}\mathcal{F}$ is tensor exact for $\operatorname{Sym}(\mathfrak{t})$ . We obtain our desired convolution exactness again using 2.3. ∎

7.3. Vanishing for central $\mathcal{D}$ -modules induced from their invariants

We now prove a key vanishing result in the more general parabolic setting. To this end, let $P$ be a parabolic subgroup which contains $\bar{B}$ . Let $U_{P}$ denote the unipotent radical of $P$ , and let $\pi^{P}:G\to G/U_{P}$ denote the quotient map. Let $L$ denote the group $P/U_{P}$ . Let $\mathfrak{l}$ denote its Lie algebra and $\mathfrak{u}_{L}$ denote the Lie algebra of the the unipotent radical $U_{L}$ of $L$ . Let $Z_{P}$ denote the inverse image of the center $Z(L)$ under the map $\pi^{P}$ , and let $\ell:P/U_{P}\xhookrightarrow{}G/U_{P}$ denote the closed embedding; to avoid clutter we omit the dependence of $\ell$ on $P$ . The scheme $G$ in particular acquires an adjoint action of $P$ by the formula $p\cdot g=pgp^{-1}$ . This induces a $P$ -action on $G/U_{P}$ for which $\pi^{P}$ is $P$ -equivariant. In what follows we consider $G/U_{P}$ as a $P$ -scheme for this ‘adjoint’ action of $P$ . Let $W_{L}:=N_{L}(T)/T$ denote the Weyl group for $L$ , and let $\operatorname{KN}_{L}$ denote the Knop-Ngô functor for the group $L$ .

Theorem 7.3.1.

If $M$ is a $D(G)$ -module which is induced from its invariants, then:

${{\enspace}\sf{(i)}}\;$ The canonical morphism $c:\int_{\ell}\ell^{\dagger}(\int_{\pi^{P}}M)\xrightarrow{}\int_{\pi^{P}}M$ , a priori a morphism in the derived category of $\mathcal{D}_{G/U_{P}}$ -modules, is an isomorphism of sheaves concentrated in degree zero. In particular, $\int_{\pi^{P}}M$ is supported on the closed subscheme $P/U_{P}$ .

${\;\sf{(ii)}}\;$ We have isomorphisms

\operatorname{Res}_{L}^{G}(M)=\ell^{\dagger}(\int_{\pi^{P}}M)\cong U\mathfrak{% l}\otimes_{Z\mathfrak{g}}M^{G}=\operatorname{KN}_{L}(M^{G})

of left $D(P/U_{P})$ -modules.

${\sf{(iii)}}\;$ For any $W$ -equivariant $D(T)$ -module $Y$ for which the $D(T)^{W}$ -module structure on $Y^{W}$ upgrades to a $\mathfrak{J}$ -module structure, there is an isomorphism

\int_{\pi^{P}}(M_{0}\star_{G}\operatorname{Ind}_{T}^{G}(Y)^{W})\cong\int_{\pi^% {P}}(M_{0})\star_{L}\operatorname{Ind}_{T}^{L}(Y)^{W_{L}}

for any $D(G)$ -module $M_{0}$ .

We prove 7.3.1 after proving the preliminary 7.3.2 and 7.3.3.

Proposition 7.3.2.

The $Z\mathfrak{g}$ -module $\Delta_{P}:=U\mathfrak{g}/U\mathfrak{g}\cdot\mathfrak{u}_{P}$ is flat.

Proof.

Observe that $\Delta_{P}$ naturally acquires a representation of $L$ and contains an $L$ -subrepresentation of the form $U\mathfrak{p}/U\mathfrak{p}\cdot\mathfrak{u}_{P}\cong U(\mathfrak{p}/\mathfrak% {u}_{P})$ . The left action of a given $z\in Z\mathfrak{g}$ on 1 naturally lies in $U(\mathfrak{p}/\mathfrak{u}_{P})^{L}$ ; denote the image by $\chi_{L}^{G}(z)\in U(\mathfrak{p}/\mathfrak{u}_{P})^{L}$ . It is not difficult to check that $\chi_{L}^{G}:Z\mathfrak{g}\to U(\mathfrak{p}/\mathfrak{u}_{P})^{L}$ is a ring homomorphism. Moreover, since $\Delta_{P}\otimes_{U\mathfrak{l}}U\mathfrak{l}/U\mathfrak{u}_{L}\cong\Delta_{T}$ , one can easily check from the construction of the usual Harish-Chandra homomorphism $\chi_{T}^{G}$ (which identifies $Z\mathfrak{g}$ with the $(W,\cdot)$ -invariants of $\operatorname{Sym}(\mathfrak{t})$ ) that $\chi_{T}^{L}\chi_{L}^{G}=\chi_{L}^{G}$ . Thus $\chi_{L}^{G}$ is a finite morphism between smooth varieties and therefore by miracle flatness $\chi_{L}^{G}$ is finite flat. Since pushforward by a finite flat map of affine schemes is finite flat (by the projection formula, see for example [StacksProject, Lemma 36.22.1]) we see that the flatness of $\Delta_{P}$ as a $Z\mathfrak{l}$ -module (which follows since we can check flatness for the $\operatorname{{\mathrm{Sym}}}(\mathfrak{l})$ -module structure on the associated graded of the filtration induced by the PBW filtration and the fact that the Chevalley map is flat) implies the flatness as a $Z\mathfrak{g}$ -module. ∎

Lemma 7.3.3.

If $\mathcal{F}\in\operatorname{QCoh}(G/U_{P})$ is equipped with a $Z_{P}^{\circ}$ -equivariant structure such that $\Gamma(\mathcal{F})^{U_{P}}$ is trivial as a representation of $Z(L)^{\circ}$ , then any $m\in\Gamma(\mathcal{F})^{U_{P}}$ is supported on $P/U_{P}$ .

Proof.

It is known that there exists a finite dimensional representation $V\in\operatorname{{\mathrm{Rep}}}(G)$ and a $G$ -equivariant locally closed embedding $\eta:G/U_{P}\to V$ given by acting on some $v\in V$ such that, if we write $V\cong\oplus_{i}V_{\lambda_{i}}$ as a direct sum of its irreducible subrepresentations $V_{\lambda_{i}}$ of highest weight $\lambda_{i}\in X^{\bullet}(T)$ and write $v=\sum_{i}v_{\lambda_{i}}$ , then each $v_{\lambda_{i}}$ is in the irreducible $P$ -subrepresentation $S_{1}^{i}$ of $V_{\lambda_{i}}$ containing the highest weight line of $V_{\lambda_{i}}$ . Fix some irreducible summand $V_{\lambda_{i}}$ of $V$ . We now argue that the support of $\Gamma(\mathcal{F})^{U_{P}}$ , as a $\operatorname{{\mathrm{Sym}}}(V_{\lambda_{i}}^{*})$ -module, is contained in the $P$ -subrepresentation $S_{1}^{i}$ . We first observe that, by the Lie-Kolchin theorem, we may write $V_{\lambda_{i}}$ as an increasing union of $\bar{B}$ -representations

\{0\}\subsetneq S_{1}^{i}\subsetneq S_{2}^{i}\subsetneq...\subsetneq S^{i}_{n_% {i}}=V_{\lambda_{i}}

such that for all $j\geq 2$ , $S^{i}_{j}/S^{i}_{j-1}$ is a one dimensional $\bar{B}$ -representation. Choose a basis for $S_{1}^{i}$ , and complete it via some vectors $\{u_{2},...,u_{n_{i}}\}$ to a basis which is compatible with this filtration, and let $u_{j}^{*}$ be the corresponding elements of the dual basis. To prove that the support of $\Gamma(\mathcal{F})^{U_{P}}$ is contained in $S_{1}^{i}$ , it obviously suffices to recursively prove the following claim for $j\in\{n_{i},n_{i}-1,...,2\}$ :

The vector $u_{j}^{*}$ acts on the $\operatorname{{\mathrm{Sym}}}(S_{j}^{*})$ -module $\Gamma(\mathcal{F})^{U_{P}}$ by zero, and so $\Gamma(\mathcal{F})^{U_{P}}$ is naturally a $\operatorname{{\mathrm{Sym}}}(S_{j-1}^{*})$ -module, and, moreover, $V_{\lambda_{i}}^{Z_{P}^{\circ}}\subseteq S_{j-1}^{i}$ , where we take invariants for the ‘adjoint’ $Z_{P}^{\circ}$ -action.

To prove this claim, first observe that, since $S_{j}^{i}/S_{j}^{i-1}$ is $\bar{B}$ -invariant, $u_{j}^{*}\in\operatorname{{\mathrm{Sym}}}(S_{j}^{i,*})^{\bar{N}}$ , so that multiplication by $u_{j}^{*}$ on $\Gamma(\mathcal{F})$ preserves $\Gamma(\mathcal{F})^{U_{P}}$ , and $u_{j}^{*}$ is homogeneous with respect to the action of $T$ . Since $u_{j}\in V_{\lambda_{i}}$ , $T$ acts on the line spanned by $u_{j}^{*}$ by some character of the form $\lambda_{i}-\sum_{r}p_{r}\alpha_{r}-\sum_{j}q_{j}\beta_{j}$ where $\{\beta_{j}\}$ is the complement of $\{\alpha_{r}\}$ in the set of simple roots and $p_{r},q_{j}$ are some nonnegative integers. Moreover, since $u_{j}$ is homogeneous for the $T$ -action and is not in the $P$ -subrepresentation $S_{1}^{i}$ , $\sum_{j}q_{j}\beta_{j}\neq 0$ . Thus, since $Z(L)^{\circ}=(\cap_{r}\text{ker}(\alpha_{r}))^{\circ}$ , $u_{j}^{*}$ is homogeneous with respect to the left $Z(L)^{\circ}$ -action on $S_{j}^{i}$ for the character $\lambda_{i}-\sum_{j}q_{j}\beta_{j}$ of $Z(L)^{\circ}$ , and this character is not $\lambda_{i}$ . Therefore, $u_{j}^{*}$ is homogeneous with respect to the ‘adjoint’ $Z(L)^{\circ}$ -action on $S_{j}^{i}$ of degree $-\sum_{j}q_{j}\beta_{j}\neq 0$ . Therefore this function must vanish on the fixed point subscheme $V_{\lambda_{i}}^{Z_{P}^{\circ}}$ and so $V_{\lambda_{i}}^{Z_{P}^{\circ}}$ lies in the closed subscheme $S_{j-1}^{i}$ of $S_{j}^{i}$ cut out by the vanishing of $u_{j}^{*}$ . Moreover, the fact that $u_{j}^{*}$ is homogeneous of nonzero degree implies that if $m\in\Gamma(\mathcal{F})^{U_{P}}=\Gamma(\mathcal{F})^{Z_{P}^{\circ}}$ , the ( $\leq 1$ dimensional) vector subspace spanned by $u_{j}^{*}m\in\Gamma(\mathcal{F})^{U_{P}}$ is homogeneous for the $Z(L)^{\circ}$ with respect to a nontrivial character of $Z(L)^{\circ}$ . Thus, since the action of $Z(L)^{\circ}$ is trivial on $\Gamma(\mathcal{F})^{U_{P}}$ by assumption, $u_{j}^{*}m=0$ , as desired, which proves our claim.

Our claim implies that the support of any $m\in\Gamma(\mathcal{F})^{U_{P}}$ , viewing $\mathcal{F}$ as a quasicoherent sheaf on $V$ , has support contained in $\oplus_{i}(V_{\lambda_{i}})^{Z_{P}^{\circ}}\cong V^{Z_{P}^{\circ}}$ . Thus the support of any $U_{P}$ -invariant global section of $\mathcal{F}$ is contained in $(G/U_{P})^{Z_{P}^{\circ}}$ . By the parabolic Bruhat decomposition, the points of $(G/U_{P})^{Z_{P}^{\circ}}$ are precisely the points of $P/U_{P}$ , and so 7.3.3 is proved. ∎

Proof of 7.3.1.

Since $G/U_{P}$ is quasi-affine (see for example [MilneAlgebraicGroupsBook, Proposition 14.26]), the functor $R\Gamma$ is conservative, and so it suffices to prove that $R\Gamma(c)$ is an isomorphism. Applying the same argument as in [ChenOnTheConjecturesofBravermanKazhdan, Example 6.1] gives an isomorphism

R\Gamma\int_{\pi^{P}}M\cong M\mathop{\otimes}\limits^{L}_{U\mathfrak{g}}\Delta% _{P}

of $D(G/U_{P})$ -modules, where the $\mathcal{O}(G/U_{P})$ -module structure on the $\mathfrak{u}_{P}$ -coinvariants $M\otimes_{U\mathfrak{g}}\Delta_{P}$ is given through the inclusion of the ring $\mathcal{O}(G/U_{P})\xrightarrow{\sim}\mathcal{O}(G)^{U_{P}}$ into the ring $D(G/U_{P})=(D(G)/\mathfrak{u}_{P}D(G))^{U_{P}}$ of differential operators on $G/U_{P}$ . Therefore we see that, since $M$ is induced from its invariants and $U\mathfrak{g}$ is flat as a $Z\mathfrak{g}$ -module, we have isomorphisms

(7.3.4)

R\Gamma\int_{\pi^{P}}M\cong M\mathop{\otimes}\limits^{L}_{U\mathfrak{g}}\Delta% _{P}\cong(U\mathfrak{g}\mathop{\otimes}\limits_{Z\mathfrak{g}}M^{G})\mathop{% \otimes}\limits^{L}_{U\mathfrak{g}}\Delta_{P}\cong(U\mathfrak{g}\mathop{% \otimes}\limits_{Z\mathfrak{g}}^{L}M^{G})\mathop{\otimes}\limits^{L}_{U% \mathfrak{g}}\Delta_{P}\cong M^{G}\mathop{\otimes}\limits^{L}_{Z\mathfrak{g}}% \Delta_{P}

and so $R\Gamma\int_{\pi^{P}}M$ is concentrated in degree zero by 7.3.2. A base change argument identifies $\ell^{\dagger}\int_{\pi^{P}}$ with the parabolic restriction functor (see [BezrukavnikovYomDinOnParabolicRestrictionofPerverseSheaves, Remark 4.1]) which is $t$ -exact by [BezrukavnikovYomDinOnParabolicRestrictionofPerverseSheaves, Theorem 5.4]. Therefore since de Rham pushforward by a closed embedding is $t$ -exact, we see that both terms the counit morphism of 7.3.1(i) are concentrated in degree zero.

Since the counit map $c$ is the inclusion of the sections of $\mathcal{F}$ supported on $P/U_{P}$ (see for example [HTT, Proposition 1.7.1(iii)]) it suffices to prove that any section of $\int_{\pi^{P}}M$ is supported on $P/U_{P}$ . Since $G/U_{P}$ is quasi-affine, any quasicoherent sheaf on $G/U_{P}$ is globally generated [EGAII, Proposition 5.1.2], so it suffices to show the support of any object of $\Gamma(\int_{\pi^{P}}M)$ is contained in $P/U_{P}$ .

It of course further suffices to prove that the global sections $\Gamma(\int_{\pi^{P}}M\otimes_{k}k_{2\rho_{G}-2\rho_{L}})$ of the identical sheaf with the right $Z_{P}^{\circ}$ -action twisted by the representation $2\rho_{G}-2\rho_{L}$ is supported on $P/U_{P}$ . However, by the isomorphism (7.3.4), we see that all global sections of $\Gamma(M)$ are generated as a $U\mathfrak{g}$ -module by the $U_{P}$ -invariant sections, and moreover the $Z(L)$ -representation on $\Gamma(\int_{\pi^{P}}M\otimes_{k}k_{2\rho_{G}-2\rho_{L}})^{U_{P}}$ is trivial. Therefore by 7.3.3 all sections are supported on $P/U_{P}$ .

Observe that, taking $U_{P}$ -invariants of (7.3.4), we obtain (ii). We now show (iii). If $Y$ denotes some $W$ -equivariant $D(T)$ -module such that $\operatorname{Sym}(\mathfrak{t})\otimes_{\operatorname{Sym}(\mathfrak{t})^{W}}% Y^{W}\xrightarrow{}Y$ is an isomorphism, then by 1.1.5(ii) we have $\operatorname{KN}(Y^{W})=\operatorname{Ind}_{T}^{G}(Y)^{W}$ . Observe that

\int_{\pi^{P}}(M_{0}\star_{G}\operatorname{Ind}_{T}^{G}(Y)^{W})=\int_{\pi^{P}}% (M_{0}\star_{G}\operatorname{KN}(Y^{W}))\simeq M_{0}\star_{G}\operatorname{KN}% (Y^{W})\star_{G}\delta_{1U_{P}}\simeq M_{0}\star_{G}\int_{\pi^{P}}% \operatorname{KN}(Y^{W})

which by 7.3.1(i) identifies with

M_{0}\star_{G}\int_{\ell}(\ell^{\dagger}\int_{\pi^{P}}\operatorname{KN}(Y^{W})% )\simeq M_{0}\star_{G}\int_{\ell}(\delta_{1U_{P}})\star_{L}\ell^{\dagger}\int_% {\pi^{P}}\operatorname{KN}(Y^{W})\simeq\int_{\pi^{P}}(M_{0})\star_{L}\ell^{% \dagger}\int_{\pi^{P}}\operatorname{KN}(Y^{W})

using the counit map of the adjunction and the $L_{\text{op}}$ -equivariance of $\ell$ . By 7.3.1(ii), we may continue this chain of isomorphisms to obtain

\simeq\int_{\pi^{P}}(M_{0})\otimes_{U\mathfrak{l}}U\mathfrak{l}\otimes_{Z% \mathfrak{g}}Y^{W}\cong\int_{\pi^{P}}(M_{0})\otimes_{Z\mathfrak{l}}(Z\mathfrak% {l}\otimes_{Z\mathfrak{g}}Y^{W})\cong\int_{\pi^{P}}(M_{0})\otimes_{Z\mathfrak{% l}}Y^{W_{L}}\cong\int_{\pi^{P}}(M_{0})\otimes_{U\mathfrak{l}}U\mathfrak{l}% \otimes_{Z\mathfrak{l}}Y^{W_{L}}

\int_{\pi^{P}}(M_{0})\otimes_{U\mathfrak{l}}\operatorname{KN}_{L}(Y^{W_{L}})=% \int_{\pi^{P}}(M_{0})\otimes_{U\mathfrak{l}}\operatorname{Ind}_{T}^{L}(Y)^{W_{% L}}=\int_{\pi^{P}}(M_{0})\star_{L}\operatorname{Ind}_{T}^{L}(Y)^{W_{L}}

again using 1.1.5(ii). ∎

7.4. Proof of 1.2.5

We have constructed the Knop-Ngô functor, as a braided monoidal functor, and verified that it is fully faithful in Section 6.2. It remains to determine its essential image. Observe that the underlying Harish-Chandra bimodule of $\operatorname{KN}(M_{0})$ is $U\mathfrak{g}\otimes_{Z\mathfrak{g}}M_{0}$ , which is clearly induced from its invariants. Conversely, assume we are given a $D(G)$ -module $M$ induced from its invariants. By 7.3.1(ii), the canonical map $\operatorname{Sym}(\mathfrak{t})\otimes_{Z\mathfrak{g}}M^{G}\xrightarrow{}% \operatorname{Res}_{T}^{G}(M)$ is an isomorphism. Since there are isomorphisms

\operatorname{Res}_{T}^{G}(M)\xleftarrow{\sim}D(T)\otimes_{D(T)^{W}}% \operatorname{Res}_{T}^{G}(M)^{W}\cong D(T)\otimes_{D(T)^{W}}M^{G}

by applying [GinzburgParabolicInductionandtheHarishChandraDModule, Proposition 4.1] and [GinzburgParabolicInductionandtheHarishChandraDModule, Theorem 4.4(i)], respectively, we may trace through the constructions of these isomorphisms and obtain that the canonical map $\operatorname{Sym}(\mathfrak{t})\otimes_{Z\mathfrak{g}}M^{G}\xrightarrow{}D(T)% \otimes_{D(T)^{W}}M^{G}$ is an isomorphism. Therefore, by [Gin, Theorem 1.5.1] (which also holds, with the same proof, after removing the condition of being holonomic) the $D(T)^{W}$ -module structure on $\operatorname{Res}_{T}^{G}(M)^{W}=M^{G}$ upgrades to a module for $\mathfrak{J}$ . We therefore obtain a map

\operatorname{KN}(M^{G})={\mathbf{N}}\otimes_{D(T)^{W}}\mathfrak{o}(M^{G})% \xrightarrow{}M

of central Harish-Chandra bimodules given by the module structure. Since we can check that this map is an isomorphism at the level of underlying Harish-Chandra bimodules, we obtain an isomorphism $\operatorname{KN}(M^{G})\cong M$ and so $M$ is in the essential image of the Knop-Ngo functor.

It remains to show that a central Harish-Chandra bimodule is induced from its invariants if and only if it is very central. However, observe by 7.3.1(i), any $G$ -equivariant $D(G)$ -module induced from its invariants is very central. Conversely, assume $M\in D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G_{\text{ad}}}$ is very central. First, we claim that the $D(T)^{W}$ -module structure on $M^{G}$ upgrades to a $\mathfrak{J}$ -module structure. To see this, observe that, since $M$ is very central, the canonical map

\int_{i}\operatorname{{\mathrm{Res}}}^{G}_{T}(M)=\int_{i}i^{\dagger}\int_{\pi^% {\bar{B}}}M\xrightarrow{}\int_{\pi^{\bar{B}}}M

is an isomorphism where $i$ is the inclusion of $\bar{B}/\bar{N}$ into $G/\bar{N}$ . In particular, using the exactness of parabolic restriction [BezrukavnikovYomDinOnParabolicRestrictionofPerverseSheaves], we see that $\int_{\pi^{\bar{B}}}M$ is a complex concentrated in degree zero and moreover supported on an affine open subset. Thus applying the global sections functor and using 2.3 (see also [ChenOnTheConjecturesofBravermanKazhdan, Example 6.1]) we obtain isomorphisms of complexes

	$\displaystyle U\mathfrak{g}/U\mathfrak{g}\cdot\bar{\mathfrak{n}}\otimes_{% \operatorname{Sym}(\mathfrak{t})}D(T)\otimes_{D(T)^{W}}M^{G}$	$\displaystyle\cong U\mathfrak{g}/U\mathfrak{g}\cdot\bar{\mathfrak{n}}\otimes_{% \operatorname{Sym}(\mathfrak{t})}\text{res}(M)$
		$\displaystyle\xrightarrow{\sim}\Gamma\Big{(}\int_{\pi^{\bar{B}}}M\Big{)}=R% \Gamma\Big{(}\int_{\pi^{\bar{B}}}M\Big{)}=M\otimes_{U\mathfrak{g}}^{L}U% \mathfrak{g}/U\mathfrak{g}\cdot\bar{\mathfrak{n}}$

where the first isomorphism uses [GinzburgParabolicInductionandtheHarishChandraDModule]. Applying the functor $\mathfrak{n}^{\psi}\cdot U\mathfrak{g}\backslash U\mathfrak{g}\otimes_{U% \mathfrak{g}}(\text{-})$ of left $\mathfrak{n}^{\psi}$ -coinvariants, we obtain an isomorphism

D(T)\otimes_{D(T)^{W}}M^{G}\cong\mathfrak{n}^{\psi}\cdot U\mathfrak{g}% \backslash U\mathfrak{g}\otimes_{U\mathfrak{g}}M\otimes_{U\mathfrak{g}}U% \mathfrak{g}/U\mathfrak{g}\cdot\bar{\mathfrak{n}}

and so applying Skryabin’s equivalence to this isomorphism we obtain an isomorphism

D(T)\otimes_{D(T)^{W}}M^{G}\cong D^{\psi_{\ell}}_{\hbar}/D^{\psi_{\ell}}_{% \hbar}\mathfrak{n}^{\psi}_{r}\otimes_{\mathfrak{J}}\varkappa(M)\otimes_{U% \mathfrak{g}}U\mathfrak{g}/U\mathfrak{g}\cdot\bar{\mathfrak{n}}\cong\bar{% \mathfrak{n}}^{r}D^{\psi_{\ell}}_{\hbar}\backslash D^{\psi_{\ell}}_{\hbar}/D^{% \psi_{\ell}}_{\hbar}\mathfrak{n}^{\psi}_{r}\otimes_{\mathfrak{J}}\varkappa(M)=% {\mathbb{M}}\otimes_{\mathfrak{J}}\varkappa(M)

where ${\mathbb{M}}$ is the Miura bimodule defined in Section 5.3. Tracing through the constructions, we see that this isomorphism is $W$ -equivariant, and so taking invariants we obtain that $M^{G}\cong{\mathbb{M}}^{W}\otimes_{\mathfrak{J}}\varkappa_{\hbar}(M)$ lies in the essential image of the fully faithful functor ${\mathbb{M}}^{W}\otimes_{\mathfrak{J}}(\text{-})$ and so we see that the $D(T)^{W}$ -module structure on $M^{G}$ upgrades to a $\mathfrak{J}$ -module, as desired.

We now claim that the counit map

(7.4.1)

{\mathbf{N}}\otimes_{D(T)^{W}}M^{G}\xrightarrow{}M

for parabolic restriction is an isomorphism, which we will prove using arguments similar to those appearing in the proofs of [BravermanKazhdanGammaSheavesonReductiveGroups, Proposition 2.9] and [ChenOnTheConjecturesofBravermanKazhdan, Proposition 3.3]. To this end, first observe that, since $M$ is very central, one can identify $\operatorname{Ind}_{T}^{G}(\operatorname{Res}_{T}^{G}(M))$ with $\text{hc}^{L}(\text{hc}(M))$ where hc is the Harish-Chandra functor $D(G)\operatorname{{\text{-}{\mathrm{mod}}}}^{G}\to\mathcal{D}_{G/N}% \operatorname{{\text{-}{\mathrm{mod}}}}^{B_{\text{ad}}}$ induced by pushforward, and $\text{hc}^{L}$ is its left adjoint. Direct computation proves the (well known) fact that $\text{hc}^{L}\text{hc}$ is naturally isomorphic to convolution with the Springer sheaf, and so we see that $\operatorname{Ind}_{T}^{G}(\operatorname{{\mathrm{Res}}}^{G}_{T}(M))$

I replaced res by $\operatorname{{\mathrm{Res}}}^{G}_{T}$ is also given by convolution with the Springer sheaf. Therefore we see that the counit map

\operatorname{Ind}_{T}^{G}(\operatorname{Res}_{T}^{G}(M))^{W}\to M

is an isomorphism, and thus using 5.2.2 we see that (7.4.1) is an isomorphism as well, since it is the counit for a naturally isomorphic adjoint pair. Now, since $M^{G}$ has a $\mathfrak{J}$ -module structure, we therefore obtain an isomorphism $\operatorname{KN}(M^{G})\xrightarrow{\sim}M$ and so $M$ lies in the essential image of the Knop-Ngô functor, as required.

Remark 7.4.2.

Since we have shown that a $G$ -equivariant $D(G)$ -module is very central if and only if it is induced from its $G$ -invariants, 7.3.1(i) can be used to re-derive a Levi upgrade of the vanishing conjecture–a devissage argument immediately shows the claim at the level of derived categories follows from the vanishing conjecture for abelian categories. This Levi upgrade was originally proved by Chen [ChenTowardTheDepthZeroStableBernsteinCenterConjecture, Theorem 6.2].

8. Proof of 6.2.1

8.1.

Recall that we have constructed a left $U\mathfrak{g}$ -bialgebroid structure (in the sense of [BohmHopfAlgebroids, Definition 3.3]) on $D(G)$ in the proof of 2.4.2, and used this to construct a left $Z_{\hbar}\mathfrak{g}$ -bialgebroid structure on $\mathfrak{J}_{\hbar}$ . We have also seen that the coalgebra structure on $\mathfrak{J}$ cocommutative in 5.3.2, and that $\mathfrak{J}$ is flat as a left $Z\mathfrak{g}$ -module in 4.1.4. Using the monoidal structure on the category of right $D(G)$ , one can completely analogously equip $D(G)$ with a right $U\mathfrak{g}_{\text{op}}$ -structure, and run completely analogous arguments to prove that this induces a right $Z\mathfrak{g}$ -bialgebroid structure on $\mathfrak{J}$ in the sense of [BohmHopfAlgebroids, Definition 3.1] with source and target map the inclusion of $Z_{\hbar}\mathfrak{g}$ into $\mathfrak{J}_{\hbar}$ .

Observe, moreover, that the ring $D(G)$ admits an algebra anti-automorphism $S$ uniquely determined by the conditions that it swaps $U_{\hbar}\mathfrak{g}$ and $U_{\hbar}\mathfrak{g}_{{\text{op}}}$ and extended the antipode on $\mathcal{O}(G)$ . It is not difficult to check that this equips $D(G)$ with the structure of a Hopf algebroid with base algebras $U\mathfrak{g}$ and $U\mathfrak{g}_{\text{op}}$ in the sense of [BohmHopfAlgebroids, Definition 4.1]; since all of the maps in question are $U\mathfrak{g}$ -linear or antilinear and restrict to $\mathcal{O}(G)\subseteq D(G)$ to the usual coproduct, counit, or antipode, it is not difficult to check the Hopf algebroid axioms hold for $D(G)$ . Now, using the algebra isomorphism of (3.3.1), we see that $S$ induces an algebra involution $\overline{S}$ on $\mathfrak{J}_{\hbar}$ . The Hopf algebroid compatibilities are readily checked. We summarize this discussion in the following Proposition:

Proposition 8.1.1.

The ring structure on $\mathfrak{J}$ upgrades to a structure of a Hopf algebroid with both base algebras $Z\mathfrak{g}$ with a bijective antipode. Moreover, in either of the $Z\mathfrak{g}$ -coalgebra structures on $\mathfrak{J}$ , the comultiplication is cocommutative, and as a left $Z\mathfrak{g}$ -module $\mathfrak{J}$ is flat.

Using 8.1.1, we now prove 6.2.1. In fact, 6.2.1 follows formally from the properties of $\mathfrak{J}$ listed in 8.1.1. To exhibit this, we hereafter adapt the following notation: we assume we are given a commutative $K$ -algebra $Z$ for $K$ some commutative ring and a Hopf algebroid $\mathcal{J}$ with bijective antipode whose base algebras both are $Z$ for which the source map $s:Z\to\mathcal{J}$ for the left bialgebroid structure agrees with the map

(8.1.2)

Z=Z_{\text{op}}\xrightarrow{t}\mathcal{J}

induced by the target of the left bialgebroid structure such that the following conditions hold:

${{\enspace}\sf{(i)}}\;$ Both left and right multiplication by the image of the source map exhibit $\mathcal{J}$ as a flat $Z$ -module.

${\;\sf{(ii)}}\;$ The left (equivalently right) coring structure is cocommutative. If $\mathcal{J}$ is a graded algebra, we will say that $\mathcal{J}$ is a graded Hopf algebroid if $\mathcal{J}$ is equipped with a grading for which the multiplication map, unit map for the algebra structure, the comultiplication and counit maps, and the antipode all respect the grading.

As $\mathcal{J}$ is a Hopf algebroid, its category of modules $\mathcal{J}$ -mod and comodules $\mathcal{J}\text{-comod}$ for its left bialgebroid of $\mathcal{J}$ both acquire a monoidal structure and, moreover, the monoidal structure on $\mathcal{J}$ -modules is symmetric, as we review in Section 8.2. As $\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ is monoidal, we may consider its Drinfeld center $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ (defined in for example [EtingofGelakiNikshychOstrikTensorCategories, Section 7.13]) whose objects are given by pairs $(C^{\prime},Z^{C^{\prime}}_{C})$ where $C^{\prime}\in\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ and $Z^{C^{\prime}}_{C}$ is a collection of isomorphisms

Z^{C^{\prime}}_{C}:C\otimes_{Z}C^{\prime}\xrightarrow{\sim}C^{\prime}\otimes_{% Z}C

natural in $C$ . The collection of such objects naturally form a braided monoidal category: indeed, as we also recall in Section 6.1, the Drinfeld center as defined in [EtingofGelakiNikshychOstrikTensorCategories] is evidently equivalent to the braided monoidal category obtained from the relative Drinfeld center construction $\mathcal{Z}_{F}(\mathcal{H})$ defined in Section 6.1 in the special case $F:=\text{id}$ .

Observe that our assumptions imply that the comultiplication and counit maps for the left bialgebroid $\mathcal{J}$ form a $Z$ -coalgebra. Therefore, to any $X\in Z\operatorname{{\text{-}{\mathrm{mod}}}}$ , we may view $X$ as a trivial comodule for this coalgebra. In particular, restricting scalars along the map $s$ , we may view any $\mathcal{J}$ -module as a trivial $\mathcal{J}$ -comodule. With this observation, we may now state our generalization of 6.2.1, which says that this construction lifts to an embedding of the category of $\mathcal{J}$ -modules into the Drinfeld center of the category of $\mathcal{J}$ -comodules:

Theorem 8.1.3.

There is a braided monoidal, fully faithful functor

\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to\mathcal{Z}(\mathcal{J}% \text{-comod})

to the Drinfeld center of the category of $\mathcal{J}$ -comodules such that the underlying comodule of $M$ is the $Z$ -module $M$ (viewed as a trivial $\mathcal{J}$ -comodule) and the central structure $Z_{C}$ has inverse given by the formula $c\otimes m\mapsto c_{(0)}m\otimes c_{(1)}$ for any $c\in C\in\mathcal{J}\text{-comod}$ and $m\in M\in\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ . If $\mathcal{J}$ is a graded Hopf algebroid, then there moreover is a braided monoidal, fully faithful functor

\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}_{\text{gr}}\to\mathcal{Z}(% \mathcal{J}\text{-comod}_{\text{gr}})

with central structure $Z_{C}^{-1}$ given by the same formula.

The remainder of Section 8 will be devoted to the proof of 8.1.3. In proving 8.1.3, we will also give an explicit formula (8.4.1) for the map $Z_{C}$ , using the structure of comodules for Hopf algebroids discussed in Section 8.3.

Remark 8.1.4.

The statement of 8.1.3 can be viewed as a variant of the known result that, for any finite abelian group $A$ , there is a fully faithful functor, symmetric monoidal functor $Z_{A}:\mathcal{O}(A)\operatorname{{\text{-}{\mathrm{mod}}}}\to\mathcal{Z}(% \text{Rep}(A))$ . While the functor in 8.1.3 is constructed completely analogously to $Z_{A}$ , new technical considerations occur in the proof that this functor is defined and has the desired properties in the Hopf algebroid setting. Therefore, we provide a complete proof of 8.1.3 below.

8.2. Notation for monoidal structure on bialgebroid

Recall that, for any $C\in\mathcal{J}\text{-comod}$ , in addition to the underlying (left) $Z$ -module structure there is also a right $Z$ -module structure on $C$ , given by the formula $cz=\epsilon_{L}(c_{(0)}z)c_{(1)}$ , where we use Sweedler’s notation and where $\epsilon_{L}$ is the counit for the left bialgebroid of $\mathcal{J}$ such that the induced functor $\mathcal{J}\to Z\text{-bimod}$ is monoidal [BohmHopfAlgebroids, Lemma 3.17]. Explicitly, in Sweedler’s notation, the $\mathcal{J}$ -comodule structure on the bimodule tensor product $B\otimes_{Z}C$ is given by the formula

(8.2.1)

b\otimes c\mapsto b_{(0)}c_{(0)}\otimes b_{(1)}\otimes c_{(1)}

for any $b\in B,c\in C$ , see [BohmHopfAlgebroids, Theorem 3.18]. If $B,C$ are equipped with compatible gradings, the tensor product also upgrades to a graded $\mathcal{J}$ -comodule.

As $\mathcal{J}$ is a Hopf algebroid, the category of $\mathcal{J}$ -modules also acquires a monoidal structure such that restriction of scalars along the ring map

Z\otimes_{K}Z_{\text{op}}\to\mathcal{J}

gives a monoidal functor to the category of $Z$ -bimodules [BohmHopfAlgebroids, Theorem 3.13]. Explicitly, given $\mathcal{J}$ -modules $M$ , $N$ we may define a $\mathcal{J}$ -module structure on the tensor product $M\otimes_{Z}N$ by the formula

(8.2.2)

u(m\otimes n):=u_{(0)}m\otimes u_{(1)}n

where $m\in M,n\in N$ , and $u\in\mathcal{J}$ . Our assumption that $s$ is the composite of the maps in (8.1.2) implies that this restriction of scalars naturally factors through the monoidal subcategory of $Z$ -modules in $Z$ -bimodules. Moreover, $\mathcal{J}$ is cocommutative by assumption, and so from the formula (8.2.2) it immediately follows that restriction of scalars along $s$ upgrades to a symmetric monoidal functor to the category of $Z$ -modules. As in the comodule case, if $M,N$ are equipped with compatible gradings, then $M\otimes_{Z}N$ is a graded $\mathcal{J}$ -module.

This discussion in particular shows that we may view any $\mathcal{J}$ -module and any $\mathcal{J}$ -comodule both as $Z$ -bimodules, and any unadorned tensor product hereafter is taken with respect to the bimodule tensor product. Additionally, if $P$ and $Q$ are any $Z$ -bimodules, we use the symbol $P\otimes_{\ell,\ell}Q$ to denote the tensor product of the corresponding left $Z$ -module structures. We similarly use the notation $P\otimes_{r,r}Q$ , $P\otimes_{\ell,r}Q$ , and $P\otimes_{r,\ell}Q=P\otimes Q$ .

8.3. Comodules of bialgebroid vs. comodules of Hopf algebroid

By our flatness assumption, [BohmHopfAlgebroids, Theorem 4.8] immediately implies the following:

Proposition 8.3.1.

For any comodule $(C,\rho_{L})$ for the left bialgebroid $\mathcal{J}$ , there exists left comodule structure $\rho_{R}$ for the right bialgebroid for $\mathcal{J}$ such that $\rho_{R}$ is a comodule map for the left bialgebroid for $\mathcal{J}$ map and $\rho_{L}$ is a comodule map for the right bialgebroid of $\mathcal{J}$ , i.e.

(8.3.2)

(\text{id}_{\mathcal{J}}\otimes\rho_{R})\circ\rho_{L}=(\Delta_{L}\otimes\text{% id}_{\mathcal{J}})\circ\rho_{R}\text{, }(\text{id}_{\mathcal{J}}\otimes\rho_{L% })\circ\rho_{R}=(\Delta_{R}\otimes\text{id}_{\mathcal{J}})\circ\rho_{L}

and moreover any morphism of $\mathcal{J}$ -comodules is automatically compatible with $\Delta_{R}$ .

For any $C\in\mathcal{J}\text{-comod}$ , we use superscripts to refer to the coaction of $\rho_{R}$ , i.e. we set $\rho_{R}(c)=c^{(0)}\otimes c^{(1)}$ and, as above, we use subscripts to refer to the coaction $\rho_{L}$ .

8.4. Construction of central structure $Z_{C}$

For a fixed $C\in\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}}$ and $M\in\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ , we define a map of $Z$ -bimodules via the formula $Z_{C}:M\otimes C\xrightarrow{}C\otimes M$ via the formula

(8.4.1)

Z_{C}(m\otimes c):=c^{(1)}\otimes A(c^{(0)})m

which, when we wish to make the dependence on the $\mathcal{J}$ -module $M$ clear, we also denote by $Z_{C}^{M}$ .

Similarly we may define a map of $Z$ -bimodules $C\otimes M\xrightarrow{}M\otimes C$ via the formula

(8.4.2)

c\otimes m\mapsto c_{(0)}m\otimes c_{(1)}

which we denote by $Z_{C}^{-1}$ or, when we wish to make the dependence on $M$ clear, we denote it by $Z_{C}^{M,-1}$ . This notation is justified by the following claim:

Proposition 8.4.3.

The map $Z_{C}^{M}$ is graded if $\mathcal{J}$ is graded, and the inverse to $Z_{C}^{M}$ is given by the formula (8.4.2).

Before proving 8.4.3, we make one (non-circular) observation which will also be used later. Observe that the map given by the formula (8.4.2) fits into the following commutative diagram:

(8.4.4)

where the downward arrows are the maps induced by the $\mathcal{J}$ -module structure,³³3In particular, the rightmost arrow is given by the formula $b\otimes c\otimes m\mapsto bm\otimes c$ . and a similar diagram using the maps $Z_{C}^{M}$ and $Z_{C}^{\mathcal{J}}$ commutes.

Proof of 8.4.3.

Using the commutative diagram (8.4.4) and the analogous diagram for $Z_{C}^{M}$ , it suffices to show this claim when $M=\mathcal{J}$ itself. Letting $\mu$ denote the multiplication morphism in $\mathcal{J}$ , one readily checks that $Z^{\mathcal{J}}_{C}$ identifies with the composite

(8.4.5)

\mathcal{J}\otimes_{\ell,\ell}C\xrightarrow{A\otimes\text{id}}\mathcal{J}% \otimes_{r,\ell}C\xrightarrow{\text{id}\otimes\rho_{R}}\mathcal{J}\otimes_{r,% \ell}\mathcal{J}\otimes_{r,r}C\xrightarrow{\mu\otimes\text{id}}\mathcal{J}% \otimes_{r,r}C\xrightarrow{A\otimes\text{id}}\mathcal{J}\otimes_{\ell,r}C% \xrightarrow{\text{swap}}C\otimes_{r,\ell}\mathcal{J}

which is in particular right $\mathcal{J}$ -linear. When $M=\mathcal{J}$ , the map (8.4.2) is right $\mathcal{J}$ -linear as well, as it is induced by the comultiplication map for the left bialgebroid. Therefore it suffices to show that, for any $c\in C$ , the formula (8.4.2) provides a right inverse to $Z_{C}^{\mathcal{J}}$ on objects of the form $1\otimes c$ , or in other words

(8.4.6)

Z^{\mathcal{J}}_{C}(c_{(0)}\otimes c_{(1)})=c\otimes 1,

and that the formula (8.4.2) also gives a left inverse to $Z_{C}^{\mathcal{J}}$ on objects of the form $1\otimes c$ for $c\in C$ , i.e. if we apply (8.4.2) to $Z^{\mathcal{J}}_{C}(1\otimes c)$ we obtain $1\otimes c$ . We have equalities

(\text{swap}\circ Z_{C}^{\mathcal{J}}(c_{(0)}\otimes c_{(1)}))=(A\otimes\text{% id})(\mu\otimes\text{id})(\text{id}\otimes\rho_{R})(A\otimes\text{id})\rho_{L}% (c)=(A\otimes\text{id})(\mu\otimes\text{id})(A\otimes\rho_{R})\rho_{L}(c)

=(A\otimes\text{id})(\mu\otimes\text{id})(A\otimes\text{id}\otimes\text{id})(% \text{id}\otimes\rho_{R})\rho_{L}(c)=(A\otimes\text{id})(\mu\otimes\text{id})(% A\otimes\text{id}\otimes\text{id})(\Delta_{t}\otimes\text{id}_{\mathcal{J}})% \circ\rho_{R}(c)

=(A\otimes\text{id})\circ(\epsilon_{s}\otimes\text{id})\circ\rho_{R}(c)=(A% \otimes\textit{id})(1\otimes c)=1\otimes c

where the first step uses the description of $Z_{C}$ as in (8.4.5) and the definition of $\rho_{L}(c)$ , the second and third steps follow from the definition of the tensor product of maps, the fourth step uses (8.3.2), the fifth step is by the definition of the inverse in a Hopf algebroid, see for example [BohmHopfAlgebroids, Definition 4.1(iv)], the sixth step uses that $C$ is a comodule and the seventh step uses the fact the antipode is an algebra morphism. Thus applying the swap map to both sides of this equality in (8.4.6). The fact that the formula (8.4.2) provides a left inverse to $Z_{C}$ is completely analogous. The fact that $Z_{C}^{M}$ is graded follows either by direct computation or by the observation that $Z_{C}^{M,-1}$ is obviously graded. ∎

Now, using cocommutativity, we may verify that $Z_{C}$ actually gives a map of $\mathcal{J}$ -comodules:

Proposition 8.4.7.

Equipping $\mathcal{J}$ with a trivial left $\mathcal{J}$ -comodule structure, the map $Z_{C}$ is a $\mathcal{J}$ -comodule map.

Proof.

It suffices to show $Z_{C}^{-1}$ is a $\mathcal{J}$ -comodule map. As above using (8.4.2) it suffices to prove this in the case $M=\mathcal{J}$ . Let $P_{L}$ denote the induced comodule structure on $\mathcal{J}\otimes C$ . We wish to show that

(8.4.8)

P_{L}\circ Z_{C}=(\text{id}_{\mathcal{J}}\otimes Z_{C})\circ P_{L}

for any comodule $C$ . Since both maps are right $\mathcal{J}$ -linear, it suffices to show this on objects of the form $c\otimes 1$ . In this case, we see that

P_{L}\circ Z_{C}(c\otimes 1)=P_{L}(c_{(0)}\otimes c_{(1)})=c_{(1,0)}\otimes c_% {(0)}\otimes c_{(1)}=c_{(0,1)}\otimes c_{(0,0)}\otimes c_{(1)}

where the second equality uses the formula (8.2.1) for the coaction on the tensor product and the final equality is obtained by swapping the first two factors in the equality

c_{(0)}\otimes c_{(1,0)}\otimes c_{(1,1)}=c_{(0,0)}\otimes c_{(0,1)}\otimes c_% {(1)}

which in turn holds since $C$ is a comodule. Similarly using (8.2.1) we have that

(\text{id}_{\mathcal{J}}\otimes Z_{C})\circ P_{L}(c\otimes 1)=(\text{id}_{% \mathcal{J}}\otimes Z_{C})(c_{(0)}\otimes 1\otimes c_{(1)})=c_{(0)}\otimes c_{% (1,0)}\otimes c_{(1,1)}=c_{(0,0)}\otimes c_{(0,1)}\otimes c_{(1)}

and since $\mathcal{J}$ is cocommutative assumption we obtain the equality in (8.4.8) as desired. ∎

8.5. Construction of functor

Using the isomorphism in Section 8.4, we construct the functor in 8.1.3:

Corollary 8.5.1.

The assignment $M\mapsto(M,Z_{C}^{M})$ upgrades to a functor $\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}\to\mathcal{Z}(\mathcal{J}% \text{-comod})$ , where we view $M$ as a trivial $\mathcal{J}$ -comodule as above. If $\mathcal{J}$ has a grading compatible with the algebra structure, coalgebra structures, and antipode, then there is a graded analogue $\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}_{\text{gr}}\to\mathcal{Z}(% \mathcal{J}\text{-comod}_{\text{gr}})$ .

Proof.

Observe that the map $Z_{C}^{\mathcal{J}}$ is induced by coaction map of $C$ , which is a map of right $Z$ -modules. Therefore the maps $Z_{C}^{\mathcal{J}}$ are natural in $C$ . Using the diagram (8.4.4) we therefore see that $Z_{C}^{M}$ is natural in $C$ for any $\mathcal{J}$ -module $M$ .

We wish to show that $(M,Z_{C}^{M})$ gives an object in $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ ; to do this, we must show that, for any $B,C\in\mathcal{J}\text{-comod}$ that

Z_{B\otimes C}=(\text{id}_{B}\otimes Z_{C})\circ(Z_{B}\otimes\text{id}_{C}).

Of course, we may show that their inverses agree.

As above, using (8.4.2) it suffices to prove this equality in the case $M=\mathcal{J}$ . Since all of these maps are right $\mathcal{J}$ -module maps, we may show this equality on objects of the form $b\otimes c\otimes 1$ . We see that

Z_{B\otimes C}^{-1}(b\otimes c\otimes 1)=(b\otimes c)_{(0)}\otimes(b\otimes c)% _{(1)}=b_{(0)}c_{(0)}\otimes b_{(1)}\otimes c_{(1)}

by the definition of the comodule structure on the tensor product as in (8.2.1). On the other hand, we have

(Z_{B}^{-1}\otimes\text{id}_{C})\circ(\text{id}_{B}\otimes Z_{C}^{-1})(b% \otimes c\otimes 1)=(Z_{B}^{-1}\otimes\text{id}_{C})(b\otimes c_{(0)}\otimes c% _{(1)})=b_{(0)}c_{(0)}\otimes b_{(1)}\otimes c_{(1)}

obtained from twice applying 8.4.3, and so we see these maps agree and thus $M$ indeed gives rise to an object of $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ . Using the commutative diagram (8.4.4), one also shows that any map $M_{1}\to M_{2}$ of $\mathcal{J}$ -modules induces a map in $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ . ∎

Proof of 8.1.3.

We have defined the functor in 8.5.1; it remains to show it is fully faithful and braided monoidal. Since the forgetful functor from the Drinfeld center of a monoidal category to the monoidal category is faithful, the functor of 8.5.1 is faithful. To see that it is full, assume that $M,M^{\prime}\in\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ and assume $f:M\to M^{\prime}$ is a map in $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ . We wish to show that $f$ is a map of $\mathcal{J}$ -modules. However, observe that if we view $\mathcal{J}$ as a left $\mathcal{J}$ -comodule via the left comultiplication of the left bialgebroid, the diagram

commutes: the leftmost square commutes since $f$ is a map in $\mathcal{Z}(\mathcal{J}\operatorname{{\mathrm{\text{-}\mathrm{comod}}}})$ , the middle and final square commute since the counit $\epsilon_{L}$ for the left bialgebroid is $Z$ -linear, and the unlabelled maps are induced by the module structure and in particular commute. Traversing the three arrows on the top of the diagram, we obtain the map $b\otimes m\mapsto\epsilon_{L}(b_{(1)})b_{(0)}m=bm$ where this equality holds since $\mathcal{J}$ is a coalgebra; similarly the bottom composition of the three arrows on the bottom of the diagram give the map $b\otimes m^{\prime}\mapsto bm^{\prime}$ , and so this diagram shows that $f$ must be a map of $\mathcal{J}$ -modules, as required.

We now show braided monoidality. Denote the monoidal structure in $\mathcal{Z}(\mathcal{J}\text{-comod})$ by $\star$ . We claim that the monoidality isomorphism

(M,Z_{C}^{M})\star(N,Z_{C}^{N})\cong(M\otimes N,Z_{C}^{M\otimes N})

is simply an equality, which of course is natural in $M$ and $N$ . Using an analogous diagram to (8.4.2) it suffices to prove this when $M=N=\mathcal{J}$ .

Of course, the underlying (trivial) $\mathcal{J}$ -comodules agree, and so it remains to verify the central structures are the same or, in other words, we wish to show that

(8.5.2)

(Z^{\mathcal{J}}_{C}\otimes\text{id}_{\mathcal{J}})\circ(\text{id}_{\mathcal{J% }}\otimes Z^{\mathcal{J}}_{C})=Z^{\mathcal{J}\otimes_{Z}\mathcal{J}}_{C}

for any $C\in\mathcal{J}$ -comod. As above, we may check that the inverse of both maps agree and, since both maps are right $\mathcal{J}\otimes_{k}\mathcal{J}$ -module maps, we may check that these inverses agree on objects of the form $c\otimes 1\otimes 1$ . Now one directly checks that the inverse of the left map in (8.5.2) takes $c\otimes 1\otimes 1$ to $c_{(0)}\otimes c_{(1,0)}\otimes c_{(1,1)}$ and that

Z^{\mathcal{J}\otimes_{Z}\mathcal{J},-1}_{C}(c\otimes 1\otimes 1)=c_{(0)}(1% \otimes 1)\otimes c_{(1)}=c_{(0,0)}\otimes c_{(0,1)}\otimes c_{(1)}

using the formula (8.2.2). Both of these sums agree by the definition of a comodule.

Finally, since our monoidality morphisms are equalities, to show that this functor is braided it suffices to prove that the braiding $Z_{N}^{M}:M\otimes N\to N\mathop{\otimes}M$ is the swap morphism for any $M,N\in\mathcal{J}\operatorname{{\text{-}{\mathrm{mod}}}}$ . However, this follows by direct computation:

Z^{M}_{N}(m\otimes n)=n^{(1)}\otimes A(n^{(0)})m=n\otimes m

since the comodule structure on $N$ is trivial. ∎

\printbibliography

Quantization of the universal centralizer and central D-modules

Abstract.

1. Main results

1.1. Knop-Ngô morphism

Theorem 1.1.5.

Corollary 1.1.6.

Remark 1.1.7 (Comparison to work of Ben-Zvi and Gunningham).

1.2. Parabolic induction and very central D𝐷Ditalic_D-modules

Corollary 1.2.2.

Theorem 1.2.5.

Remark 1.2.7.

1.3. The exactness and vanishing conjectures

Theorem 1.3.2.

Remark 1.3.3.

1.4. Outline

1.5. Acknowledgements

2. Monoidal structure on D⁢(G)𝐷𝐺D(G)italic_D ( italic_G )-modules

2.1. Conventions

Definition.

Remark 2.1.1.

2.2. Asymptotic Harish-Chandra bimodules

2.3.

Remark 2.3.4.

2.4.

Definition 2.4.1.

Proposition 2.4.2.

Remark 2.4.3.

Proof of 2.4.2.

3. Kostant-Whittaker Reduction

3.1. Classical Kostant-Whittaker reduction

Corollary 3.1.1.

3.2. Kostant-Whittaker reduction

Proposition 3.2.4.

Proof.

3.3. Monoidal structure on Whittaker reduction

Definition.

Example.

Proposition 3.3.2.

Proof.

4. Bi-Whittaker reduction and 𝔍ℏsubscript𝔍Planck-constant-over-2-pi\mathfrak{J}_{\hbar}fraktur_J start_POSTSUBSCRIPT roman_ℏ end_POSTSUBSCRIPT-comodules

4.1.

Theorem 4.1.1.

Lemma 4.1.3.

Proof of 4.1.3.

Corollary 4.1.4.

Proof.

Lemma 4.1.5.

Proof.

Lemma 4.1.6.

Proof of 4.1.1(i)-(ii).

Proof of 4.1.1(iv).

Corollary 4.1.9.

Proof.

5. Kostant-Whittaker reduction and the Miura bimodule

5.1. The Miura bimodule and the Gelfand-Graev action

Proposition 5.1.2.

5.2. Parabolic induction and the Miura bimodule

Proposition 5.2.3.

Proof.

5.3. The Miura bimodule and comodules

Proposition 5.3.2.

Proof.

5.4. Trivial comodules

Lemma 5.4.1.

Proof.

Remark 5.4.2.

Corollary 5.4.3.

Lemma 5.4.4.

Proof.

Proof of 5.4.3.

Proposition 5.4.6.

Proof.

6. Construction of Knop-Ngô functor

6.1. Relative Drinfeld center

Proposition 6.1.2.

Remark 6.1.4.

6.2. Construction of Knop-Ngô functor

Theorem 6.2.1.

Theorem 6.2.2.

Proof of 1.1.6.

1.2. Parabolic induction and very central $D$ -modules

2. Monoidal structure on $D(G)$ -modules

4. Bi-Whittaker reduction and $\mathfrak{J}_{\hbar}$ -comodules

7.3. Vanishing for central $\mathcal{D}$ -modules induced from their invariants

8.4. Construction of central structure $Z_{C}$