This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

On a t-exactness property of the Harish-Chandra transform

Roman Gonin, Andrei Ionov, Kostiantyn Tolmachov
Abstract

Using hyperbolic localization, we identify the nearby cycles along the Vinberg degeneration with the composition of Radon and Harish-Chandra functors, both considered for the category of character sheaves. This provides a new, simple proof of the exactness of this composition, extending previously known results to arbitrary monodromy and more general sheaf-theoretic set-ups.

1   Introduction.

1.1   Notations.

For a stack 𝒳\mathcal{X} defined over \mathbb{C}, let Db(𝒳)D^{b}(\mathcal{X}) be the bounded algebraically constructible derived category of sheaves on 𝒳\mathcal{X} with analytic topology, with coefficients in a ring 𝕜\mathbb{k}. All the stacks we encounter will be quotient stacks of the form X/HX/H, where XX is a scheme and HH is an algebraic group over \mathbb{C}. The sheaf theory for such stacks is described in [BL06]. We write pt=Spec()\operatorname{pt}=\operatorname{Spec}(\mathbb{C}). We write 𝕜¯𝒳\underline{\mathbb{k}}_{\mathcal{X}} for the constant sheaf on 𝒳\mathcal{X}.

For an algebraic group HH acting on a scheme XX and its subgroup HH^{\prime} let

ForHH:Db(H\X)Db(H\X)\operatorname{For}^{H}_{H^{\prime}}\colon D^{b}(H\backslash X)\to D^{b}(H^{\prime}\backslash X)

stand for the forgetful functor, that is the *-pullback along the projection H\XH\XH^{\prime}\backslash X\to H\backslash X. Let

AvH,!HAvH,H:Db(H\X)Db(H\X)\text{$\operatorname{Av}^{H}_{H^{\prime},!}$, $\operatorname{Av}_{H^{\prime},*}^{H}$}\colon D^{b}(H^{\prime}\backslash X)\to D^{b}(H\backslash X)

stand for the !! and *-pushforward from H\XH^{\prime}\backslash X to H\XH\backslash X, respectively. If HH^{\prime} is clear from the context, we will omit it from the notation, writing ForH,Av!H,AvH\operatorname{For}^{H},\operatorname{Av}_{!}^{H},\operatorname{Av}_{*}^{H}, respectively.

We will adopt the same notation for the right action of the group, writing ForR,AvR\prescript{R}{}{\operatorname{For}},\prescript{R}{}{\operatorname{Av}} for the right forgetful and averaging functors, when the emphasis is needed.

1.2   Background and the main result.

Let GG be a reductive group, BB a fixed Borel subgroup, UBU\subset B its unipotent radical and TBT\subset B a maximal torus. The Harish-Chandra transform 𝔥𝔠!\mathfrak{hc}_{!} for constructible sheaves was introduced by Lusztig [Lus85] in his study of character sheaves. It is given by a pull-push functor along the horocycle correspondence

G/AdGG/AdB(U\G/U)/AdT,G/_{\operatorname{Ad}}G\xleftarrow{}G/_{\operatorname{Ad}}B\xrightarrow{}(U\backslash G/U)/_{\operatorname{Ad}}T,
𝔥𝔠!:=Av!UForG.\mathfrak{hc}_{!}:=\operatorname{Av}^{U}_{!}\circ\operatorname{For}^{G}.

It was observed by Ginzburg [Gin89] that it produces objects in the categorical center of the finite Hecke category attached to a reductive group GG and its Borel subgroup BB. This observation has been elaborated upon in various contexts in (among other works) [BFO12], [BN09], [Lus], [Lus15], [HL23], [BITV23].

Another remarkable property of the Harish-Chandra transform on the character sheaves, playing a key role in [BFO12], is that it becomes tt-exact and Verdier self-dual after composing it with the Radon transform functor RR_{*}. The latter transform is given by the averaging with respect to the unipotent radical UU^{-} of the opposite Borel subgroup BB^{-}, or, equivalently, a pull-push functor along the following diagram:

(U\G/U)/AdT(G/U)/AdT(U\G/U)/AdT,(U\backslash G/U)/_{\operatorname{Ad}}T\xleftarrow{}(G/U)/_{\operatorname{Ad}}T\xrightarrow{}(U^{-}\backslash G/U)/_{\operatorname{Ad}}T,
R:=AvUForU[2dimU]:Db((U\G/U)/AdT)Db((U\G/U)/AdT).R_{*}:=\operatorname{Av}_{*}^{U^{-}}\circ\operatorname{For}^{U}[2\dim U]\colon D^{b}\left((U\backslash G/U)/_{\operatorname{Ad}}T\right)\to D^{b}\left((U^{-}\backslash G/U)/_{\operatorname{Ad}}T\right).

For a local system \mathcal{L} of finite order, let Db(G)D^{b}_{\mathfrak{C}_{\mathcal{L}}}(G) be the derived category of character sheaves with the character \mathcal{L} (see Section 2.2 for the precise definition). Our main result is as follows:

Theorem 1.2.1 (Corollary 2.3.2 and Remark 2.3.3).

The functor R𝔥𝔠!R_{*}\circ\mathfrak{hc}_{!}, when restricted to the category Db(G)D^{b}_{\mathfrak{C}_{\mathcal{L}}}(G), is t-exact with respect to the perverse t-structure and commutes with the Verdier duality.

We prove the Theorem 1.2.1 by identifying R𝔥𝔠!R_{*}\circ\mathfrak{hc}_{!} with the nearby cycles functor along the Vinberg degeneration. The key idea is to use that the nearby cycles functor commutes with the hyperbolic localization, as shown in [Nak16], [Ric19].

All our methods are geometric and can be restated with the same proofs, up to standard modifications, for Db(𝒳)D^{b}(\mathcal{X}) for stacks 𝒳\mathcal{X} over \mathbb{C} being replaced by either of the following:

  • the bounded derived category of holonomic DD-modules,

  • the bounded constructible derived category of ¯\overline{\mathbb{Q}}_{\ell}-sheaves on a stack 𝒳\mathcal{X} defined over an algebraically closed field of characteristic p>0p>0 for primes pp\neq\ell as in [LO08].

We expect that the methods of the present paper can be extended to handle the parabolic case. We intend to treat it in the subsequent publication.

1.3   Relation to the previous works.

The idea to relate the nearby cycles functor along the Vinberg degeneration with the functor R𝔥𝔠!R_{*}\circ\mathfrak{hc}_{!} appears already in [BFO12], where an analogue of Theorem 1.2.1 in the setting of D-modules is also proved. The same idea was used in [CY17] to prove the Theorem 1.2.1 for unipotent character sheaves (that is for a trivial local system \mathcal{L}), with methods applicable to either of the sheaf-theoretic cases, and also in the more general setting associated to a symmetric pair. It is very plausible that the methods of [CY17] can be adapted to treat the general monodromy \mathcal{L}. However, we believe that the systematic use of hyperbolic localization presented in this paper makes for a particularly short and straightforward proof.

1.4   Acknowledgments.

We would like to thank Roman Bezrukavnikov, Alexander Yom Din, and Yakov Varshavsky for helpful discussions. The work of R.G. was supported by the EPSRC grant EP/V053787/1. R.G. also acknowledges the hospitality of the Hebrew University of Jerusalem during his postdoctoral fellowship. While working on the project, K. T. was at different times funded by the EPSRC programme grant EP/R034826/1 and by the Deutsche Forschungsgemeinschaft SFB 1624 grant, Projektnummer 506632645.

2   Setup and the statement of the main results.

2.1   Limit functors.

For a scheme X~\widetilde{X} and a map f:X~𝔸1f\colon\widetilde{X}\to\mathbb{A}^{1}, let

ψf:Db(X~×𝔸1𝔾m)Db(X~×𝔸1pt),\psi_{f}\colon D^{b}(\widetilde{X}\times_{\mathbb{A}^{1}}\mathbb{G}_{m})\to D^{b}(\widetilde{X}\times_{\mathbb{A}^{1}}\operatorname{pt}),

stand for the nearby cycles functor. Here ×𝔸1\times_{\mathbb{A}^{1}} denotes the fiber product with respect to embeddings pt𝔸1𝔾m\operatorname{pt}\to\mathbb{A}^{1}\leftarrow\mathbb{G}_{m}: the embedding of 0 into 𝔸1\mathbb{A}^{1} and the complementary open embedding, respectively.

Assume that 𝔾m\mathbb{G}_{m} acts on XX and let a:𝔾m×XXa\colon\mathbb{G}_{m}\times X\to X be the action map. Consider X~=X×𝔸1\widetilde{X}=X\times\mathbb{A}^{1}. Let f:X~𝔸1f\colon\widetilde{X}\to\mathbb{A}^{1} be the projection to the second factor. Following [CY17], define the limit functor

𝕃:Db(X)Db(X),𝕃=ψfa[1].\mathbb{L}\colon D^{b}(X)\to D^{b}(X),\quad\mathbb{L}=\psi_{f}\circ a^{*}[1].

Let HH be an algebraic group acting on XX, and let us fix a cocharacter γ:𝔾mH\gamma\colon\mathbb{G}_{m}\to H. This defines the 𝔾m\mathbb{G}_{m}-action on XX, and we write

𝕃γ:Db(X)Db(X)\operatorname{\mathbb{L}_{\gamma}}\colon D^{b}(X)\to D^{b}(X)

for the corresponding limit functor.

Let GG be a reductive group and TT its maximal torus. Choice of a regular cocharacter λ:𝔾mT\lambda\colon\mathbb{G}_{m}\to T fixes a choice of a pair of opposite Borel subgroups B,BTB,B^{-}\supset T for which λ\lambda is dominant and antidominant, respectively. Let UU and UU^{-} be the unipotent radicals of BB and BB^{-}, respectively.

Let W=NG(T)/TW=N_{G}(T)/T be the Weyl group of GG, and let w0Ww_{0}\in W be its longest element. We have two projections qB:BTq_{B}\colon B\to T and qB:BTq_{B^{-}}\colon B^{-}\to T.

Let KGK\simeq G stand for the diagonal subgroup of G×GG\times G, and define three subgroups K0K_{0}, KK_{\infty}, KG×GK^{\prime}\subset G\times G as the following fiber products

(1) K0B×TB,KB×TB,KB×TB,K_{0}\simeq B\times_{T}B^{-},\quad K_{\infty}\simeq B^{-}\times_{T}B,\quad K^{\prime}\simeq B\times_{T}B,

with respect to the homomorphisms qB:BTq_{B}\colon B\to T and qB:BTq_{B^{-}}\colon B^{-}\to T.

Fix a cocharacter of H=G×GH=G\times G given by

γ:𝔾mG×G,γ=(λ1,λ).\gamma:\mathbb{G}_{m}\to G\times G,\quad\gamma=(\lambda^{-1},\lambda).

Let XX be a scheme equipped with a left action of G×GG\times G. As shown in [CY17, Lemma 4.12], the limit functor 𝕃γ\operatorname{\mathbb{L}_{\gamma}} associated with this case, can be lifted to either of the following functors:

(2) 𝕃γ:Db(K\X)Db(K0\X),\displaystyle\operatorname{\mathbb{L}_{\gamma}}\colon D^{b}(K\backslash X)\to D^{b}(K_{0}\backslash X),
(3) 𝕃γ:Db(K\X)Db(K\X).\displaystyle\operatorname{\mathbb{L}_{\gamma}}\colon D^{b}(K^{\prime}\backslash X)\to D^{b}(K^{\prime}\backslash X).
Remark 2.1.1.

Consider the constant group scheme G~=G×1\widetilde{G}=G\times\mathbb{P}^{1} over 1\mathbb{P}^{1}. One can define its subgroup scheme K~\widetilde{K}, with the fibers Kt=γ(t)Kγ(t)1K_{t}=\gamma(t)K\gamma(t)^{-1} for t0,t\neq 0,\infty, and the fibers over 0 and \infty given by K0K_{0} and KK_{\infty}, see (1). One can verify that K~\widetilde{K} is a well-defined smooth group scheme over 1\mathbb{P}^{1}, see [DG14, Proposition 2.5.2].

Let X~=X×1\widetilde{X}=X\times\mathbb{P}^{1} and K~=K×1\widetilde{K}^{\prime}=K^{\prime}\times\mathbb{P}^{1} be the constant scheme/constant group scheme over 1\mathbb{P}^{1}. Informally, the functors (2) and (3) can be thought of as the limit functor for the families K~\X~,\widetilde{K}\backslash\widetilde{X}, K~\X~1\widetilde{K}^{\prime}\backslash\widetilde{X}\rightarrow\mathbb{P}^{1}. We refer to [CY17] for details.

We collect the required properties of this functor in the following lemma. These follow from the standard properties of the nearby cycles functor, see [SGA06, XIII].

Lemma 2.1.2.

Let X1,X2X_{1},X_{2} be schemes equipped with a 𝔾m\mathbb{G}_{m}-action, and let f:X1X2f:X_{1}\to X_{2} be a 𝔾m\mathbb{G}_{m}-equivariant map.

  1. (a)

    There is a natural transformation f!𝕃𝕃f!f_{!}\circ\mathbb{L}\to\mathbb{L}\circ f_{!} of functors Db(X1)Db(X2)D^{b}(X_{1})\to D^{b}(X_{2}). This natural transformation is an isomorphism if ff is proper.

  2. (b)

    There is a natural transformation f𝕃𝕃ff^{*}\circ\mathbb{L}\to\mathbb{L}\circ f^{*} of functors Db(X2)Db(X1)D^{b}(X_{2})\to D^{b}(X_{1}). This natural transformation is an isomorphism if ff is smooth.

Consider a Cartesian diagram

X1×SX2{X_{1}\times_{S}X_{2}}X1{X_{1}}X2{X_{2}}S{S}p1\scriptstyle{p_{1}}p2\scriptstyle{p_{2}}q1\scriptstyle{q_{1}}q2\scriptstyle{q_{2}}

of schemes with a 𝔾m\mathbb{G}_{m} action, where all the arrows are 𝔾m\mathbb{G}_{m}-equivariant.

  1. (c)

    The following diagram, where the horizontal arrows are induced by the natural transformations from (a) and (b) and the vertical ones are base change isomorphisms, is commutative

    (4) q2q1!𝕃{q_{2}^{*}q_{1!}\circ\mathbb{L}}q2𝕃q1!{q_{2}^{*}\circ\mathbb{L}\circ q_{1!}}𝕃q2q1!{\mathbb{L}\circ q_{2}^{*}q_{1!}}p2!p1𝕃{p_{2!}p_{1}^{*}\circ\mathbb{L}}p2!𝕃p1{p_{2!}\circ\mathbb{L}\circ p_{1}^{*}}𝕃p2!p1{\mathbb{L}\circ p_{2!}p_{1}^{*}}
  2. (d)

    When restricted to a subcategory of 𝔾m\mathbb{G}_{m}-monodromic sheaves on X1X_{1}, the functor 𝕃\mathbb{L} is isomorphic to the identity functor.

Remark 2.1.3.

It may be convenient for further arguments to think about the commutativity of the diagram (4) as follows. Consider the cube diagram below. Its front and back faces are identical Cartesian squares, and its arrows coming out of the plane of the paper are limit functors 𝕃\mathbb{L}.

Db(X1){D^{b}(X_{1})}Db(S){D^{b}(S)}Db(X1){D^{b}(X_{1})}Db(S){D^{b}(S)}Db(X1×SX2){D^{b}(X_{1}\times_{S}X_{2})}Db(X2){D^{b}(X_{2})}Db(X1×SX2){D^{b}(X_{1}\times_{S}X_{2})}Db(X2){D^{b}(X_{2})}𝕃\scriptstyle{\mathbb{L}}q1!\scriptstyle{q_{1!}}q2\scriptstyle{q_{2}^{*}}p1\scriptstyle{p_{1}^{*}}𝕃\scriptstyle{\mathbb{L}}p2!\scriptstyle{p_{2!}}

The proposition says that two natural transformations between red and dashed blue paths Db(X1)Db(X2)D^{b}(X_{1})\to D^{b}(X_{2}), one given by composition of natural transformations on the top, bottom and right faces of the cube, and another along left, bottom and front faces, coincide.

2.2   Character sheaves.

Let XX be a scheme with an action of an algebraic torus TT, and let a:T×XXa\colon T\times X\to X, p:T×XXp\colon T\times X\to X be the action and projection maps, respectively. Let \mathcal{L} be a rank 1 local system on TT. We say that Db(X)\mathcal{F}\in D^{b}(X) is weakly \mathcal{L}-equivariant if there is an isomorphism aa^{*}\mathcal{F}\simeq\mathcal{L}\boxtimes\mathcal{F}. We denote by Db(X)D^{b}_{\mathcal{L}}(X) the full idempotent-complete triangulated subcategory of Db(X)D^{b}(X) generated by weakly \mathcal{L}-equivariant complexes. We call the complexes in Db(X)D^{b}_{\mathcal{L}}(X) \mathcal{L}-monodromic complexes or monodromic complexes with generalized monodromy \mathcal{L}.

If T=𝔾mT=\mathbb{G}_{m} and \mathcal{L} is trivial, we will write Dmonb(X)D^{b}_{mon}(X) for Db(X)D^{b}_{\mathcal{L}}(X) and refer to this category as a category of 𝔾m\mathbb{G}_{m}-monodromic complexes.

For a scheme XX with an action of an algebraic group HH which is normalized by the action of TT, we will write Db(H\X)D^{b}_{\mathcal{L}}(H\backslash X) for the full triangulated subcategory of Db(H\X)D^{b}(H\backslash X) consisting of objects \mathcal{F} with ForHDb(X)\operatorname{For}^{H}\mathcal{F}\in D^{b}_{\mathcal{L}}(X).

We call a local system \mathcal{L} on TT rational if α\alpha^{*}\mathcal{L} is trivial for some surjective homomorphism α:TT\alpha\colon T\to T.

Let X=G×GX=G\times G, and consider actions of G×GG\times G on XX by left and right multiplication. Consider a functor

𝔥𝔠!:Db(K\X/K)Db(K\X/K),𝔥𝔠!=AvKK,!KRForKKKR,\mathfrak{hc}_{!}\colon D^{b}(K\backslash X/K)\to D^{b}(K\backslash X/K^{\prime}),\quad\mathfrak{hc}_{!}=\prescript{R}{}{\operatorname{Av}}^{K^{\prime}}_{K\cap K^{\prime},!}\circ\prescript{R}{}{\operatorname{For}}^{K}_{K\cap K^{\prime}},

where the superscript RR on averaging and forgetful functor is used to emphasized that these are with respect to the groups acting on the right. The notation 𝔥𝔠\mathfrak{hc} stands for Harish-Chandra, and the functor thus defined is commonly called the Harish-Chandra functor. Let

χ:Db(K\X/K)Db(K\X/K),χ=AvKK,!KRForKKKR.\chi\colon D^{b}(K\backslash X/K^{\prime})\to D^{b}(K\backslash X/K),\quad\chi=\prescript{R}{}{\operatorname{Av}}^{K}_{K\cap K^{\prime},!}\circ\prescript{R}{}{\operatorname{For}}^{K^{\prime}}_{K\cap K^{\prime}}.

The torus TT acts on K\X/KK\backslash X/K^{\prime} on the right. Let \mathcal{L} be a rational rank 1 local system on TT. The derived category of character sheaves with character \mathcal{L}, denoted by Db(G)D^{b}_{\mathfrak{C}_{\mathcal{L}}}(G), is the full triangulated idempotent-complete subcategory of Db(K\X/K)D^{b}(K\backslash X/K) generated by the essential image of Db(K\X/K)D^{b}_{\mathcal{L}}(K\backslash X/K^{\prime}) under the functor χ\chi.

2.3   Main results.

Consider the diagram

(5) Db(K\X/K){D^{b}(K\backslash X/K)}Db(K0\X/K){D^{b}(K_{0}\backslash X/K)}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K)}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K)}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}AvB,!KForBK0\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}}α\scriptstyle{\alpha}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}

Here α:AvB,!KForBK0𝕃γ𝕃γAvB,!KForBK\alpha\colon\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\to\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B} is the natural transformation given by Lemma 2.1.2.

We will prove the following

Theorem 2.3.1.

When restricted to the category of character sheaves Db(G)D^{b}_{\mathfrak{C}_{\mathcal{L}}}(G), the transformation α\alpha is an isomorphism.

Corollary 2.3.2.

After the restriction to the category of character sheaves Db(G)D^{b}_{\mathfrak{C}_{\mathcal{L}}}(G), there is an isomorphism of functors

𝕃γAvB,K0ForBKAvB,!KForBK.\mathbb{L}_{\gamma}\simeq\operatorname{Av}^{K_{0}}_{B,*}\circ\operatorname{For}^{K^{\prime}}_{B}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}.
Proof.

By Lemma 2.1.2 (d), the lower horizontal arrow in the diagram (5) is isomorphic to the identity functor when restricted to the monodromic subcategory. We obtain an isomorphism

AvB,!KForBK0𝕃γAvB,!KForBK.\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\mathbb{L}_{\gamma}\simeq\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}.

It is well-known that, restricted to the monodromic subcategory, the functor AvB,!KForBK0\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B} is an equivalence, with inverse given by AvB,K0ForBK\operatorname{Av}^{K_{0}}_{B,*}\circ\operatorname{For}^{K^{\prime}}_{B}, and the Corollary follows. ∎

Remark 2.3.3.

To match the notations of the introduction, note that

K\X/K=G\G×G/GG/AdG,K\backslash X/K=G\backslash G\times G/G\simeq G/_{\operatorname{Ad}}G,
K\X/K=(B×TB)\G×G/G(U\G/U)/AdT,K^{\prime}\backslash X/K=(B\times_{T}B)\backslash G\times G/G\simeq(U\backslash G/U)/_{\operatorname{Ad}}T,
K0\X/K=(B×TB)\G×G/G(U\G/U)/AdT.K_{0}\backslash X/K=(B\times_{T}B^{-})\backslash G\times G/G\simeq(U^{-}\backslash G/U)/_{\operatorname{Ad}}T.

We will reduce Theorem 2.3.1 to a similar statement of the form “nearby cycles functor commutes with the averaging functor” with X/KX/K replaced first by X/KX/K^{\prime} and then by ×\mathcal{B}\times\mathcal{B}, where =G/B\mathcal{B}=G/B is the flag variety of GG. Consider the following diagram, which is similar to diagram (5), with X/KX/K replaced by X/KX/K^{\prime}.

(6) Db(K\X/K){D^{b}(K\backslash X/K^{\prime})}Db(K0\X/K){D^{b}(K_{0}\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K^{\prime})}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}AvB,!KForBK0\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}}α\scriptstyle{\alpha^{\prime}}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}

Here α:AvB,!KForBK0𝕃γ𝕃γAvB,!KForBK\alpha^{\prime}\colon\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\to\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B} is again the natural transformation given by Lemma 2.1.2.

We have the following

Theorem 2.3.4.

When restricted to the monodromic category Db(X/K)D^{b}_{\mathcal{L}}(X/K^{\prime}), the transformation α\alpha^{\prime} is an isomorphism.

Let =G/B\mathcal{B}=G/B be the flag variety of GG. Consider the following diagram, which is similar to diagram (5), with X/KX/K^{\prime} replaced by ×\mathcal{B}\times\mathcal{B}.

(7) Db(K\×){D^{b}(K\backslash\mathcal{B}\times\mathcal{B})}Db(K0\×){D^{b}(K_{0}\backslash\mathcal{B}\times\mathcal{B})}Db(K\×){D^{b}(K^{\prime}\backslash\mathcal{B}\times\mathcal{B})}Db(K\×){D^{b}(K^{\prime}\backslash\mathcal{B}\times\mathcal{B})}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}AvB,!KForBK0\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}}α′′\scriptstyle{\alpha^{\prime\prime}}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}

Here α′′:AvB,!KForBK0𝕃γ𝕃γAvB,!KForBK\alpha^{\prime\prime}\colon\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\to\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B} is again the natural transformation given by Lemma 2.1.2.

Theorem 2.3.5.

The transformation α′′\alpha^{\prime\prime} is an isomorphism.

The implication Theorem 2.3.4 \Rightarrow Theorem 2.3.1 will be deduced from the diagram chase using the standard properties of the nearby cycles functor.

For unipotent monodromy, Theorem 2.3.5 implies the Theorem 2.3.4 in a similar way.

Proof of Theorem 2.3.4 and a simpler Theorem 2.3.5 will use properties of the hyperbolic localization functor, which we proceed to explain.

3   Hyperbolic localization.

3.1   Braden’s theorem.

We recall some notions from [DG14]. Let XX be a scheme and a:𝔾m×XXa\colon\mathbb{G}_{m}\times X\rightarrow X be a 𝔾m\mathbb{G}_{m}-action. Consider the scheme of fixed points X𝔾mX^{\mathbb{G}_{m}}. Let by X+X^{+} and XX^{-} the attracting and the repelling schemes respectively. Consider the diagram

X±{X^{\pm}}X𝔾m{X^{\mathbb{G}_{m}}}X{X}p±\scriptstyle{p_{\pm}}i±\scriptstyle{i_{\pm}}

where i±i_{\pm} is the inclusion and p±p_{\pm} is taking the limit. Let us define the functors of hyperbolic restriction

(8) +=\displaystyle\operatorname{\mathbb{H}}^{+}= (p+)!i+\displaystyle(p_{+})_{!}\circ i_{+}^{*} =\displaystyle\operatorname{\mathbb{H}}^{-}= (p)i!\displaystyle(p_{-})_{*}\circ i_{-}^{!}
Theorem 3.1.1 ([Bra03], [DG14]).

There is a natural transformation +\operatorname{\mathbb{H}}^{-}\rightarrow\operatorname{\mathbb{H}}^{+} which is an isomorphism on 𝔾m\mathbb{G}_{m}-monodromic complexes.

Denote by \operatorname{\mathbb{H}} the restriction of ±\operatorname{\mathbb{H}}^{\pm} to the 𝔾m\mathbb{G}_{m}-monodromic subcategory.

3.2   Hyperbolic localization commutes with nearby cycles.

For a scheme X~\widetilde{X} and a map f:X~𝔸1f\colon\widetilde{X}\to\mathbb{A}^{1} one has the nearby cycles functor ψf\psi_{f} as in Subsection 2.1. Let a:𝔾m×X~X~a^{\prime}\colon\mathbb{G}_{m}\times\widetilde{X}\rightarrow\widetilde{X} be an action which commutes with the map ff for the trivial action on 𝔸1\mathbb{A}^{1}. Consider the fixed points of the action and the map f0:X~𝔾m𝔸1f_{0}\colon\tilde{X}^{\mathbb{G}_{m}}\rightarrow\mathbb{A}^{1}. The natural transformations from Lemma 2.1.2 yield natural transformations

(9) +ψfψf0+\displaystyle\operatorname{\mathbb{H}}^{+}\circ\psi_{f}\rightarrow\psi_{f_{0}}\circ\operatorname{\mathbb{H}}^{+} ψf0ψf\displaystyle\psi_{f_{0}}\circ\operatorname{\mathbb{H}}^{-}\rightarrow\operatorname{\mathbb{H}}^{-}\circ\psi_{f}
Theorem 3.2.1.

([Nak16, Proposition 5.4.1 (2)], [Ric19, Theorem 3.3] Restricted to the category of 𝔾m\mathbb{G}_{m}-monodromic sheaves Dmonb(X~𝔾m×𝔸1𝔾m)D^{b}_{mon}(\tilde{X}^{\mathbb{G}_{m}}\times_{\mathbb{A}^{1}}\mathbb{G}_{m}) the natural transformations (9) is an isomorphism ψfψf0\operatorname{\mathbb{H}}\circ\psi_{f}\simeq\psi_{f_{0}}\circ\operatorname{\mathbb{H}}.

3.3   Hyperbolic localization commutes with Radon transform. Case of the flag variety.

Consider the 𝔾m\mathbb{G}_{m}-action on ×\mathcal{B}\times\mathcal{B} given by a cocharacter (λ,λ)(\lambda,\lambda) for a dominant λ\lambda. This gives us the functor

:Dmonb(×)Db(𝔾m×𝔾m)\operatorname{\mathbb{H}}\colon D^{b}_{mon}(\mathcal{B}\times\mathcal{B})\rightarrow D^{b}(\mathcal{B}^{\mathbb{G}_{m}}\times\mathcal{B}^{\mathbb{G}_{m}})

Moreover, consider UU acting on the right factor of ×\mathcal{B}\times\mathcal{B}. This gives us functors

ForU:Dmonb(×U\)Dmonb(×)\displaystyle For^{U}\colon D^{b}_{mon}(\mathcal{B}\times U\backslash\mathcal{B})\rightarrow D^{b}_{mon}(\mathcal{B}\times\mathcal{B})
Av!U:Dmonb(×)Dmonb(×U\)\displaystyle Av_{!}^{U}\colon D^{b}_{mon}(\mathcal{B}\times\mathcal{B})\rightarrow D^{b}_{mon}(\mathcal{B}\times U\backslash\mathcal{B})

The following lemma is straightforward, see [CGY19, Sect. 1.4.5].

Lemma 3.3.1.

Counit map induces a natural isomorphism

ForUAv!U\operatorname{\mathbb{H}}\simeq\operatorname{\mathbb{H}}\circ For^{U}\circ Av_{!}^{U}

As a corollary we obtain the following proposition

Proposition 3.3.2.
  1. (a)

    The unit map IdForBKAvB,!K\operatorname{Id}\rightarrow\operatorname{For}^{K^{\prime}}_{B}\circ\operatorname{Av}^{K^{\prime}}_{B,!} induces an isomorphism of functors making the following diagram commutative:

    Db(K\×){D^{b}(K\backslash\mathcal{B}\times\mathcal{B})}Db(K\×){D^{b}(K^{\prime}\backslash\mathcal{B}\times\mathcal{B})}Db(𝔾m×𝔾m){D^{b}(\mathcal{B}^{\mathbb{G}_{m}}\times\mathcal{B}^{\mathbb{G}_{m}})}ForK\scriptstyle{\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}ForK\scriptstyle{\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}}
  2. (b)

    The same holds for KK replaced by K0K_{0}.

Proof.

Using basechange for the following Cartesian diagram

×{\mathcal{B}\times\mathcal{B}}B\×{B\backslash\mathcal{B}\times\mathcal{B}}×U\{\mathcal{B}\times U\backslash\mathcal{B}}K\×{K^{\prime}\backslash\mathcal{B}\times\mathcal{B}}

we get

ForKAvB,!KForBKForUForUKAvB,!KForBKForUAv!UForBForBKForUAv!UForK\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{For}^{U}\circ\operatorname{For}^{K^{\prime}}_{U}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\\ \operatorname{For}^{U}\circ\operatorname{Av}^{U}_{!}\circ\operatorname{For}^{B}\circ\operatorname{For}^{K}_{B}\cong\operatorname{For}^{U}\circ\operatorname{Av}^{U}_{!}\circ\operatorname{For}^{K}

The sheaves in the essential image of the ForK\operatorname{For}^{K} are 𝔾m\mathbb{G}_{m}-monodromic, which allows us to use Lemma 3.3.1. This finishes the proof of (a). The case of K0K_{0} is similar. ∎

3.4   Hyperbolic localization commutes with Radon transform. Case of the base affine space.

Consider the 𝔾m\mathbb{G}_{m}-action on X/KX/K^{\prime} given by a cocharacter

𝚊w1,w2:𝔾m×X/KX/K\displaystyle\mathtt{a}_{w_{1},w_{2}}\colon\mathbb{G}_{m}\times X/K^{\prime}\rightarrow X/K^{\prime}
𝚊w1,w2(t,(g1,g2))=(λ(t)g1w11(λ(t1)),λ(t)g2w21(λ(t1)))\displaystyle\mathtt{a}_{w_{1},w_{2}}\left(t,(g_{1},g_{2})\right)=\Big{(}\lambda(t)\cdot g_{1}\cdot w_{1}^{-1}\left(\lambda(t^{-1})\right),\lambda(t)\cdot g_{2}\cdot w_{2}^{-1}\left(\lambda\left(t^{-1}\right)\right)\Big{)}

for a dominant λ\lambda. This gives us the functors of hyperbolic localization

w1,w2:Dmonb(X/K)Db(X/K)𝔾m.\operatorname{\mathbb{H}}_{w_{1},w_{2}}\colon D^{b}_{mon}(X/K^{\prime})\rightarrow D^{b}(X/K^{\prime})^{\mathbb{G}_{m}}.

For the rest of this subsection we fix w1,w2w_{1},w_{2} and shorten :=w1,w2\operatorname{\mathbb{H}}:=\operatorname{\mathbb{H}}_{w_{1},w_{2}}.

Note that (X/K)𝔾m=(w1B×w2B)/K(X/K^{\prime})^{\mathbb{G}_{m}}=(w_{1}B\times w_{2}B)/K^{\prime}. Moreover, consider UU acting on the right factor of X/K=G×G/KX/K^{\prime}=G\times G/K^{\prime}. This gives us functors

ForU:Dmonb(U\X/K)Dmonb(X/K)\displaystyle For^{U}\colon D^{b}_{mon}(U\backslash X/K^{\prime})\rightarrow D^{b}_{mon}(X/K^{\prime})
Av!U:Dmonb(X/K)Dmonb(U\X/K)\displaystyle Av_{!}^{U}\colon D^{b}_{mon}(X/K^{\prime})\rightarrow D^{b}_{mon}(U\backslash X/K^{\prime})

The following lemma is straightforward generalization of [CGY19, Sect. 1.4.5].

Lemma 3.4.1.

The unit map induces a functor isomorphism

ForUAv!U\operatorname{\mathbb{H}}\simeq\operatorname{\mathbb{H}}\circ For^{U}\circ Av_{!}^{U}
Proof.

The map p+p_{+} factors

(X/K)+U\(X/K)+p~(X/K)𝔾m\left(X/K^{\prime}\right)^{+}\rightarrow U\backslash\left(X/K^{\prime}\right)^{+}\xrightarrow{\tilde{p}}\left(X/K^{\prime}\right)^{\mathbb{G}_{m}}

Hence

(p+)!ForUAv!U=(p~+)!Av!UForUAv!U(p~+)!Av!U=(p+)!.(p_{+})_{!}\circ For^{U}\circ Av_{!}^{U}=(\tilde{p}_{+})_{!}\circ Av_{!}^{U}\circ For^{U}\circ Av_{!}^{U}\simeq(\tilde{p}_{+})_{!}\circ Av_{!}^{U}=(p_{+})_{!}.

Then we have the following chain of isomorphisms

(10) ForUAv!U=(p+)!i+ForUAv!U(p+)!ForUAv!Ui+(p+)!i+=,\operatorname{\mathbb{H}}\circ For^{U}\circ Av_{!}^{U}=(p_{+})_{!}\circ i_{+}^{*}\circ For^{U}\circ Av_{!}^{U}\simeq\\ (p_{+})_{!}\circ For^{U}\circ Av_{!}^{U}\circ i_{+}^{*}\simeq(p_{+})_{!}\circ i_{+}^{*}=\operatorname{\mathbb{H}},

where the second isomorphism is the base change and the third isomorphism is explained above. ∎

As a corollary we obtain the following proposition

Proposition 3.4.2.
  1. (a)

    The unit map IdForBKAvB,!K\operatorname{Id}\rightarrow\operatorname{For}^{K^{\prime}}_{B}\circ\operatorname{Av}^{K^{\prime}}_{B,!} induces an isomorphism of functors for the following diagram

    Db(K\X/K){D^{b}(K\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K^{\prime})}Db(X/K)𝔾m{D^{b}(X/K^{\prime})^{\mathbb{G}_{m}}}ForK\scriptstyle{\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}ForK\scriptstyle{\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}}
  2. (b)

    The same holds for KK replaced by K0K_{0}.

Proof.

Using basechange for the following Cartesian diagram

X/K{X/K^{\prime}}B\X/K{B\backslash X/K^{\prime}}U\X/K{U\backslash X/K^{\prime}}K\X/K{K^{\prime}\backslash X/K^{\prime}}

we get

ForKAvB,!KForBKForUForUKAvB,!KForBKForUAv!UForBForBKForUAv!UForK\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{For}^{U}\circ\operatorname{For}^{K^{\prime}}_{U}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\\ \operatorname{For}^{U}\circ\operatorname{Av}^{U}_{!}\circ\operatorname{For}^{B}\circ\operatorname{For}^{K}_{B}\cong\operatorname{For}^{U}\circ\operatorname{Av}^{U}_{!}\circ\operatorname{For}^{K}

The sheaves in the essential image of the ForK\operatorname{For}^{K} are 𝔾m\mathbb{G}_{m}-monodromic on the attracting locus, which allows us to use Lemma 3.4.1. This finishes the proof of (a). The case of K0K_{0} is identical. ∎

Informally, one can refer to the proposition above as hyperbolic restriction commutes with Radon transform R!=AvB,!KForBK0R_{!}=\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}.

4   Proof of the main results.

4.1   Proof for the flag variety.

In this subsection we give a proof of Theorem 2.3.5. The following Lemma is clear.

Lemma 4.1.1.

The functor

ForK:Db(K\×)Db(𝔾m×𝔾m)\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\colon D^{b}(K^{\prime}\backslash\mathcal{B}\times\mathcal{B})\rightarrow D^{b}(\mathcal{B}^{\mathbb{G}_{m}}\times\mathcal{B}^{\mathbb{G}_{m}})

is conservative.

Using the lemma above it is enough to check that

(11) ForK(α′′):ForKAvB,!KForBK0𝕃γForK𝕃γAvB,!KForBK\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}(\alpha^{\prime\prime})\colon\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\to\\ \operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}

is an isomorphism. We have the following chain of isomorphisms

(12) ForKAvB,!KForBK0𝕃γForK0𝕃γ𝕃γForK𝕃γForKForK.\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\simeq\operatorname{\mathbb{H}}\circ\operatorname{For}^{K_{0}}\circ\operatorname{\mathbb{L}_{\gamma}}\simeq\\ \operatorname{\mathbb{H}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{For}^{K}\simeq\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}\simeq\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}.

The first map is inverse of the unit transformation, see Proposition 3.3.2(b). The second isomorphism is given by Lemma 2.1.2. The third isomorphism is Theorem 3.2.1. The last isomorphism comes from the fact that family is 𝔾m×𝔾m\mathcal{B}^{\mathbb{G}_{m}}\times\mathcal{B}^{\mathbb{G}_{m}} is trivial.

Analogously, we have the following chain of isomorphism

(13) ForK𝕃γAvB,!KForBK𝕃γForKAvB,!KForBK𝕃γForKAvB,!KForBK𝕃γForKForK.\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{\mathbb{H}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\\ \operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}\simeq\operatorname{\mathbb{H}}\circ\operatorname{For}^{K}.

The first isomorphism comes from Lemma 2.1.2. The second isomorphism is Theorem 3.2.1. The third map is inverse of the unit transformation, see Proposition 3.3.2(a). The last isomorphism comes from the fact that family is 𝔾m×𝔾m\mathcal{B}^{\mathbb{G}_{m}}\times\mathcal{B}^{\mathbb{G}_{m}} is trivial.

A diagram chase along the base change transformations implies that the natural transformations (11), (12), and (13) form a commutative triangle.

Remark 4.1.2.

Consider the diagonal embedding Δ:K\K\×\Delta:K\backslash\mathcal{B}\to K\backslash\mathcal{B}\times\mathcal{B}. The localization of the object 𝕃γ(Δ𝕜¯K\)\mathbb{L}_{\gamma}(\Delta_{*}\underline{\mathbb{k}}_{K\backslash\mathcal{B}}) in the setting of DD-modules is known as a semiregular module, see [Don08]. For wWw\in W let 𝒪w=UwB,𝒪w=UwB\mathcal{O}_{w}=UwB\subset\mathcal{B},\mathcal{O}_{w}^{-}=U^{-}wB\subset\mathcal{B} be the Bruhat cells, and let jw:𝒪w,jw:𝒪wj_{w}:\mathcal{O}_{w}\to\mathcal{B},j^{\prime}_{w}:\mathcal{O}^{-}_{w}\to\mathcal{B} be the corresponding locally-closed embeddings. Let Δw=jw!𝕜¯𝒪w[dim𝒪w],w=jw𝕜¯𝒪w[dim𝒪w]\Delta_{w}=j_{w!}\underline{\mathbb{k}}_{\mathcal{O}_{w}}[\dim\mathcal{O}_{w}],\nabla_{w}^{-}=j^{\prime}_{w*}\underline{\mathbb{k}}_{\mathcal{O}^{-}_{w}}[\dim\mathcal{O}^{-}_{w}]. Using Lemma 3.2.1 it is easy to see that ForK0𝕃γ(Δ𝕜¯K\)[dimK0]\operatorname{For}^{K_{0}}\mathbb{L}_{\gamma}(\Delta_{*}\underline{\mathbb{k}}_{K\backslash\mathcal{B}})[\dim K_{0}] admits a filtration with associated graded pieces given by Δw0w1w,wW\Delta_{w_{0}w^{-1}}\boxtimes\nabla_{w},w\in W. After localization, this recovers [Don08, Theorem 11.14]. See also [Che23] for a related discussion in the affine setting.

4.2   Proof for monodromic sheaves.

In this subsection we give a proof of Theorem 2.3.4.

Denote =w1,w2w1,w2\operatorname{\mathbb{H}_{\bullet}}=\oplus_{w_{1},w_{2}}\operatorname{\mathbb{H}}_{w_{1},w_{2}}.

Lemma 4.2.1.

The functor

ForK:Db(K\X/K)w1,w2Db(w1B×w2B/K)\operatorname{\mathbb{H}}_{\bullet}\circ\operatorname{For}^{K^{\prime}}\colon D^{b}(K^{\prime}\backslash X/K^{\prime})\rightarrow\bigoplus_{w_{1},w_{2}}D^{b}\big{(}w_{1}B\times w_{2}B/K^{\prime}\big{)}

is conservative.

Proof.

Let ≄0\mathcal{F}^{\prime}\not\simeq 0 be a complex in Db(K\X/K)D^{b}(K^{\prime}\backslash X/K^{\prime}), and let =ForK\mathcal{F}=\operatorname{For}^{K^{\prime}}\mathcal{F}^{\prime}. Assume that =0\mathbb{H}_{\bullet}\mathcal{F}=0.

For (w1,w2)W2(w_{1},w_{2})\in W^{2}, write 𝒪w1,w2=w1B×w2B/K,\mathcal{O}_{w_{1},w_{2}}=w_{1}B\times w_{2}B/K^{\prime}, and let jw1,w2:𝒪w1,w2X/Kj_{w_{1},w_{2}}:\mathcal{O}_{w_{1},w_{2}}\to X/K^{\prime} be the embedding. Pick a pair (w1,w2)W(w_{1},w_{2})\in W such that 𝒪w1,w2\mathcal{O}_{w_{1},w_{2}} is open in the support of \mathcal{F}. Let

(14) jw1,w2!jw1,w2𝒵j_{w_{1},w_{2}!}j^{*}_{w_{1},w_{2}}\mathcal{F}\to\mathcal{F}\to\mathcal{Z}

be the standard triangle associated to the natural transformation

jw1,w2!jw1,w2IdDb(X/K).j_{w_{1},w_{2}!}j^{*}_{w_{1},w_{2}}\to\operatorname{Id}_{D^{b}(X/K^{\prime})}.

It is easy to see that v1,v2(jw1,w2!𝒢)0\mathbb{H}_{v_{1},v_{2}}(j_{w_{1},w_{2}!}\mathcal{G})\simeq 0 unless (v1,v2)=(w1,w2)(v_{1},v_{2})=(w_{1},w_{2}) for any 𝒢Db(𝒪w1,w2)\mathcal{G}\in D^{b}(\mathcal{O}_{w_{1},w_{2}}). Applying \mathbb{H}_{\bullet} to the triangle (14) and using this consideration, we get that

w1,w2jw1,w2!jw1,w2=0,\mathbb{H}_{w_{1},w_{2}}j_{w_{1},w_{2}!}j^{*}_{w_{1},w_{2}}\mathcal{F}=0,

which is easily seen to imply jw1,w20j^{*}_{w_{1},w_{2}}\mathcal{F}\simeq 0, since this complex is smooth along the left KK^{\prime}-orbits. This contradicts our assumptions on the support of \mathcal{F}. ∎

Using the lemma above it is enough to check that

(15) ForK(α):ForKAvB,!KForBK0𝕃γForK𝕃γAvB,!KForBK\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}(\alpha^{\prime})\colon\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\to\\ \operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}

is an isomorphism. We have the following chain of isomorphisms

(16) ForKAvB,!KForBK0𝕃γForK0𝕃γ𝕃γForK𝕃γForK.\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\simeq\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K_{0}}\circ\operatorname{\mathbb{L}_{\gamma}}\simeq\\ \operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{For}^{K}\simeq\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K}.

The first map is inverse of the unit transformation, see Proposition 3.4.2(b). The second isomorphism is given by Lemma 2.1.2. The third isomorphism is Theorem 3.2.1.

Analogously, we have the following chain of isomorphism

(17) ForK𝕃γAvB,!KForBK𝕃γForKAvB,!KForBK𝕃γForKAvB,!KForBK𝕃γForK.\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\\ \operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K^{\prime}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\simeq\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{\mathbb{H}_{\bullet}}\circ\operatorname{For}^{K}.

The first isomorphism comes from Lemma 2.1.2. The second isomorphism is Theorem 3.2.1. Note that the sheaves in the essential image of the ForK\operatorname{For}^{K^{\prime}} are 𝔾m\mathbb{G}_{m}-monodromic on the attracting locus, which makes the theorem applicable. The third map is inverse of the unit transformation, see Proposition 3.4.2(a).

Diagram chase along the base change transformations implies that the natural transformations (15), (16), and (17) form a commutative triangle.

4.3   From monodromic sheaves to character sheaves.

Here we prove the implication Theorem 2.3.4 \Rightarrow Theorem 2.3.1. The following Lemma is clear.

Lemma 4.3.1.

Let Φ,Ψ:𝒞𝒟\Phi,\Psi\colon\mathcal{C}\to\mathcal{D} be two exact functors between triangulated categories 𝒞\mathcal{C}, 𝒟\mathcal{D} and let α:ΦΨ\alpha\colon\Phi\to\Psi be a natural transformation. Assume that for any object c𝒞c\in\mathcal{C}, there is an object cc^{\prime} containing cc as a direct summand, such that αc:Φ(c)Ψ(c)\alpha_{c^{\prime}}\colon\Phi(c^{\prime})\to\Psi(c^{\prime}) is an isomorphism. Then α\alpha is an isomorphism.

By Lemma 4.3.1 it is enough to show that the natural transformation

αχ:AvB,!KForBK𝕃γχ𝕃γAvB,!KForBKχ\alpha\circ\chi\colon\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\circ\operatorname{\mathbb{L}_{\gamma}}\circ\chi\to\operatorname{\mathbb{L}_{\gamma}}\circ\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}\circ\chi

is an isomorphism. To this end, consider the following cube diagram, where in each cardinal plane parallel arrows have the same name.

Db(K\X/K){D^{b}(K\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K\backslash X/K)}Db(K0\X/K){D^{b}(K_{0}\backslash X/K^{\prime})}Db(K0\X/K){D^{b}(K_{0}\backslash X/K)}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K)}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K^{\prime})}Db(K\X/K){D^{b}(K^{\prime}\backslash X/K)}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}χ=AvB,!KRForBKR\scriptstyle{\chi=\prescript{R}{}{\operatorname{Av}}^{K}_{B,!}\circ\prescript{R}{}{\operatorname{For}}^{K^{\prime}}_{B}}AvB,!KForBK\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K}_{B}}AvB,!KForBK0\scriptstyle{\operatorname{Av}^{K^{\prime}}_{B,!}\circ\operatorname{For}^{K_{0}}_{B}}𝕃γ\scriptstyle{\operatorname{\mathbb{L}_{\gamma}}}AvB,!KRForBKR\scriptstyle{\prescript{R}{}{\operatorname{Av}}^{K}_{B,!}\circ\prescript{R}{}{\operatorname{For}}^{K^{\prime}}_{B}}

Iterating Lemma 2.1.2 (c) we get that the two natural transformations between red and dashed blue paths Db(K\X/K)Db(K\X/K)D^{b}(K\backslash X/K^{\prime})\to D^{b}(K^{\prime}\backslash X/K), one given by composition of natural transformations on the top, back and right faces of the cube, and another along left, bottom and front faces, coincide. Note that the natural transformation corresponding to the right face is exactly the natural transformation αχ\alpha\circ\chi. Thus, it is enough to show that natural transformations corresponding to the other five faces are isomorphisms.

Natural transformation corresponding to the left face is an isomorphism by Theorem 2.3.4. The back and front faces are compositions of base change isomorphisms. The top and bottom face are isomorphisms by Lemma 2.1.2 (a) and (b).

References

  • [BFO12] Roman Bezrukavnikov, Michael Finkelberg, and Victor Ostrik. Character D-modules via Drinfeld center of Harish-Chandra bimodules. Inventiones mathematicae, 188(3):589–620, June 2012.
  • [BITV23] Roman Bezrukavnikov, Andrei Ionov, Kostiantyn Tolmachov, and Yakov Varshavsky. Equivariant derived category of a reductive group as a categorical center. arXiv preprint arXiv:2305.02980, 2023.
  • [BL06] Joseph Bernstein and Valery Lunts. Equivariant Sheaves and Functors. Springer, 2006.
  • [BN09] David Ben-Zvi and David Nadler. The character theory of a complex group. arXiv preprint arXiv:0904.1247, 2009.
  • [Bra03] Tom Braden. Hyperbolic localization of intersection cohomology. Transform. Groups, 8(3):209–216, 2003.
  • [CGY19] T.-H. Chen, D. Gaitsgory, and A. Yom Din. On the Casselman-Jacquet functor. 101:73–112, 2019.
  • [Che23] Lin Chen. Nearby cycles on Drinfeld-Gaitsgory-Vinberg interpolation grassmannian and long intertwining functor. Duke Math. J., 172(3):447–543, 2023.
  • [CY17] Tsao-Hsien Chen and Alexander Yom Din. A formula for the geometric Jacquet functor and its character sheaf analogue. Geometric and Functional Analysis, 27(4):772–797, 2017.
  • [DG14] Vladimir Drinfeld and Denis Gaitsgory. On a theorem of Braden. Transformation Groups, 19(2):313–358, June 2014.
  • [Don08] Dmitry Donin. Lie algebras of differential operators and D-modules. PhD thesis, University of Toronto, 2008.
  • [Gin89] Victor Ginzburg. Admissible modules on a symmetric space. Astérisque, (173-74):199–255, 1989.
  • [HL23] Quoc P. Ho and Penghui Li. Graded character sheaves, HOMFLY-PT homology, and Hilbert schemes of points on 2\mathbb{C}^{2}. arXiv preprint arXiv:2305.01306, 2023.
  • [LO08] Yves Laszlo and Martin Olsson. The six operations for sheaves on Artin stacks II: Adic coefficients. Publications Mathématiques de l’IHÉS, 107:169–210, 2008.
  • [Lus] George Lusztig. Truncated convolution of character sheaves. Bulletin of the Institute of Mathematics Academia Sinica, 10(1):1–72.
  • [Lus85] George Lusztig. Character sheaves I. Advances in Mathematics, 56(3):193–237, 1985.
  • [Lus15] George Lusztig. Non-unipotent character sheaves as a categorical centre. arXiv preprint arXiv:1506.04598, 2015.
  • [Nak16] Hiraku Nakajima. Lectures on perverse sheaves on instanton moduli spaces. arXiv preprint arXiv:1604.06316, 2016.
  • [Ric19] Timo Richarz. Spaces with 𝔾m\mathbb{G}_{m}-action, hyperbolic localization and nearby cycles. Journal of Algebraic Geometry, 28:251–289, 2019.
  • [SGA06] Groupes de Monodromie en Geometrie Algebrique: Seminaire de Geometrie Algebrique du Bois-Marie 1967-1969 (SGA 7 II), Pierre Deligne, Nicholas Katz, volume 340. Springer, 2006.

R. Gonin, School of Mathematics, University of Cardiff, Cardiff, United Kingdom

E-mail address: [email protected]

A. Ionov, Department of Mathematics, Boston College, Chestnut Hill, Massachusetts, USA

E-mail address: [email protected]

K. Tolmachov, Department of Mathematics, University of Hamburg, Hamburg, Germany

E-mail address: [email protected]