Professional Documents
Culture Documents
DUALITY
Abstract. Let GR be a connected real reductive group and let X be the corresponding
complex symmetric variety under the Cartan bijection. We construct a canonical equiva-
lence between the relative Satake category of G(O)-equivariant C-constructible complexes
on the loop space X(K) and the real Satake category of GR (OR )-equivariant C-constructible
complexes on the real affine Grassmannian GrR = GR (KR )/GR (OR ). We show that the
equivalence is t-exact with respect to the natural perverse t-structures and is compatible
with the fusion products and Hecke actions. We further show that the relative Satake
category is equivalent to the category of C-constructible complexes on the moduli stack
BunGR (P1 (R)) of GR -bundles on the real projective line P1 (R) and hence provides a con-
nection between the relative Langlands program and the geometric Langlands program for
real groups.
We provide numerous applications of the main theorems to real and relative Langlands
duality including the formality and commutativity conjectures for the real and relative
Satake categories and an identification of the dual groups for GR and X.
Contents
1. Introduction 3
1.1. Main results 4
1.2. Applications 5
1.3. Organization 9
1.4. Acknowledgements 9
2. Orbits on Gr and X(K) 9
2.1. Loop groups 9
2.2. Parametrization of orbits 11
3. The Matsuki flow 12
3.1. Polynomial Loop spaces of Xc 12
3.2. Geometry of orbits 12
3.3. The energy flow on ΩGc 13
3.4. The Matsuki flow on Gr 14
4. Real Beilinson-Drinfeld Grassmannians 17
4.1. Real affine Grassmannians 17
4.2. Beilinson-Drinfeld Grassmannians 18
1
4.3. Real forms of Beilinson-Drinfeld Grassmannians 19
5. Multi-point generalizations 20
5.1. Multi-point version of real loop groups 20
5.2. Generalization of Gram-Schmidt factorization 21
5.3. Multi-point version of ΩXc 22
6. Uniformizations of real bundles 23
6.1. Stack of real bundles 23
6.2. Uniformizations of real bundles 24
6.3. Multi-point uniformization 28
7. Quasi-maps and Quillen’s homeomorphism 28
7.1. Definition of quasi-maps 29
7.2. Real forms of quasi-maps 29
7.3. Morphisms 29
7.4. Uniformizations of quasi-maps 29
7.5. Quillen’s homeomorphism 30
7.6. Trivialization of real quasi-maps 32
7.7. Flows on quasi-maps 32
8. Real-symmetric equivalence 33
8.1. Placid stacks 33
8.2. Stacks admitting gluing of sheaves 35
8.3. Real and relative Satake categories 37
8.4. Perverse t-structures 37
8.5. Real-symmetric equivalence for affine Grassmannians 39
9. Affine Matsuki correspondence for sheaves 41
9.1. The functor Υ 41
9.2. Bijection between local systems 41
9.3. Standard and co-standard sheaves 43
9.4. Fully-faithfulness 45
9.5. Proof of Theorem 9.1 48
10. Nearby cycles functors and Radon transforms 48
10.1. A square of equivalences 48
10.2. The nearby cycles functor Ψ 49
10.3. The Radon transform 49
10.4. The functor ΨR 50
10.5. Proof of Theorem 10.1 and Theorem 10.2 50
11. Compatibility of Hecke actions 50
11.1. Hecke actions 51
2
11.2. From complex to real kernels 51
11.3. Compatibility of actions 52
12. Compatibility with fusion product 53
12.1. Fusion product for GrR 53
12.2. Fusion product for X(K) 54
13. Applications 56
13.1. t-exactness criterion and semi-simplicity of Hecke actions 56
13.2. Formality and commutativity of dg Ext algebras 57
13.3. Identification of dual groups 59
Appendix A. Semi-analytic stacks 66
A.1. Basic definitions 66
A.2. From stacks to semi-analytic stacks 67
A.3. Constructible complexes 68
References 69
1. Introduction
Let GR be a real form of a connected complex reductive group G. Let X = K\G be the
associated symmetric variety under Cartan’s bijection, where K is the complexification of a
maximal compact subgroup Kc ⊂ GR . A fundamental feature of the representation theory of
the real group GR is that many results of an analytic nature have equivalent purely algebraic
geometry formulations in terms of the corresponding symmetric variety X. We will call this
broad phenomenon the real-symmetric correspondence.
In this paper we study the real-symmetric correspondence in the framework of Langlands
duality. We show that there is an equivalence between the relative Satake category of X and
the real Satake category of GR with remarkable properties (Theorem 1.1). We further show
that the relative Satake category is equivalent to the dg derived category of sheaves on the
moduli stack of GR -bundles on the real projective line P1 (R), hence provides a connection
between real and relative Langlands programs (Theorem 1.2). The proof relies on three
geometric results: (1) a multi-point version of Quillen’s homeomorphism between the loop
spaces for compact symmetric varieties and real affine Grassmannians (Section 7) (2) Morse-
theoretic construction of the Matsuki correspondence for the affine Grassmannian (Section
3) (3) uniformization of moduli stack of quasi-maps and real bundles. (Section 6).
We provide numerous applications of the main results to real and relative Langlands dual-
ity including semi-simplicity and t-exactness criteria of the Hecke actions, the formality and
commutativity conjectures for the real and relative Satake categories, and an identification
of the (Tannakian) dual groups for GR and X. The last application provides an explicit
description of the dual group of X answering a basic open question in relative Langlands
duality.
We now describe the paper in more details.
3
1.1. Main results. Let Gr = G(K)/G(O) be the affine Grassmannian for G and let D(G(O)\Gr)
be the Satake category of G(O)-equivariant C-constructible complexes on Gr. One of the
foundational result in Langlands duality is the geometric Satake equivalence [BD, G, Lu, MV]
Perv(G(O)\Gr) ≃ Rep(G∨ )
providing a description of the category of representations of the Langlands dual group G∨
in terms of the abelian Satake category Perv(G(O)\Gr) ⊂ D(G(O)\Gr) of G(O)-equivariant
perverse sheaves on Gr.
Let GrR = GR (KR )/GR (OR ) be the real affine Grassmanian of GR and let X(K) be
the loop space of X. We are interested in the real Satake category D(GR (OR )\GrR ) of
GR (OR )-equivariant C-constructible complexes on GrR and the relative Satake category
D(X(K)/G(O)) of G(O)-equivariant C-constructible complexes on X(K). Those categories
are one of the main players in the geometric Langlands for real groups [BZN] and the relative
Langlands program [BZSV].
In this paper we will assume GR is connected (equivalently, K is connected). Our first
main result is a remarkable equivalence between the real and relative Satake category, called
the real-symmetric equivalence:
Theorem 1.1. There is a natural equivalence
D(X(K)/G(O)) ≃ D(GR (OR )\GrR )
which is t-exact with respect to the perverse t-structures and is compatible with the fusion
products and Hecke actions of Rep(G∨ ).
Theorem 1.1 is the combination of Theorem 8.14, Theorem 11.1, and Theorem 12.2 in the
text; we refer to Section 8, Section 11 and Section 12 for a more detailed explanation of the
statement, including the definition of t-structures, fusion products, and Hecke actions. The
main ingredient in the proof is a multi-point version of Quillen’s homeomorphism between the
loop spaces for compact symmetric varieties and the real affine Grassmannians, see Theorem
7.5.
To state our second main result, let BunGR (P1 (R)) be the real analytic stack of GR -bundles
on the real projective line P1 (R). Let LGR and GR (R[t−1 ]) ⊂ GR (KR ) be the polynomial
loop group and polynomial arc group of GR respectively. Denote by D(BunGR (P1 (R))) the
dg category of C-constructible complexes on the real analytic stack BunGR (P1 (R)), and
by D(LGR \Gr) (resp. D(GR (R[t−1 ])\GrR )) the dg categories of LGR -equivariant (resp.
GR (R[t−1 ])-equivariant) C-constructible complexes on Gr respectively.
Theorem 1.2. There are natural commutative diagram of equivalences
≃
D(X(K)/G(O)) / D(GR (OR )\GrR )
Υ ≃ ΥR ≃
≃
D(LGR \Gr) / D(GR (R[t−1 ])\GrR )
❙❙❙ ❥
❙❙❙
❙❙❙ ❥❥❥❥❥❥❥
≃
❙❙❙ ❥❥❥
❙❙) t ❥❥❥ ≃
❥
D(BunGR (P1 (R)))
4
where Υ is the so called affine Matsuki equivalence, ΥR is the Radon transform, the horizontal
equivalences are nearby cycles functors along quasi-maps family, and the vertical equivalences
in the lower triangle are induced from the complex and real uniformizations of GR -bundles
≃ ≃
LGR \Gr / BunGR (P1 (R)) o GR (R[t−1 ])\GrR .
Moreover, the equivalences above are compatible with the natural Hecke actions of Rep(G∨ ).
Theorem 1.2 is restated in Theorem 10.2. We refer to Section 9 and Section 10 for a more
detailed explanation of the statement. The main step in the proof of Theorem 1.2 is the
affine Matsuki equivalence Υ whose proof relies on a Morse-theoretic construction (Theorem
3.9) of the Matsuki correspondence for the affine Grassmannian in [N1]: an isomorphism
between G(O)-orbits poset on X(K) (or rather K(K)-orbits poset on Gr) and LGR -orbits
poset on Gr.
Theorem 1.1 and Theorem 1.2 provide a connection between relative Langlands program
and geometric Langlands program for real groups and we expect applications of such a
connection to both subjects. For example, Theorem 1.2 has been used in [CMNO] to establish
versions of the relative Langland duality conjecture and geometric Langlands for P1 (R) in the
case (GR , X) = (GLn (H), Sp2n \GL2n ) where GLn (H) is the real quaternionic linear group
(see next section for more applications).
Remark 1.3. In the case when GR ≃ H is a connected complex reductive group viewed as
a real group, Theorem (1.2) recovers the results of V. Lafforgue [La, Proposition 2.1] saying
that there are equivalences
≃ ≃
(1.1) D(H(O)\GrH ) / D(H(C[t−1 ])\GrH ) / D(BunH (P1 ))
where the first equivalence is given by the Radon transform and the second equivalence
comes from the uniformization isomorphism H(C[t−1 ])\GrH ≃ BunH (P1 ).
Remark 1.4. The proof of Theorem 1.2 incorporated technical material in [CN2] on Matsuki
equivalence for affine grassmannians but with many new results, including a proof of a
conjecture on identification of the spherical and real dual groups in [CN2, Section 1.4.2]
(see Section 1.2.5 for more details). On the other hand, an updated version of [CN2] will
include various generalizations of the main results in loc. cit., including the case for affine
flag varieties.
1.2. Applications. Our main results allow one to use powerful algebraic geometry tools on
the symmetric side (e.g., Deligne’s theory of weights) to study questions on the real side, and
conversely, to use the concrete geometry on the real side (e.g., the real affine Grassmannian
or moduli of real bundles) to study questions on the symmetric side. Here are a few notable
examples.
Corollary 1.6 (Theorem 13.1). The Hecke action of Rep(G∨ ) on the relative Satake category
D(X(K)/G(O)) (resp. the real Satake category D(GR (OR )\GrR )) is t-exact with respect to
the perverse t-structure if and only if X is quasi-split (resp. GR is quasi-split).
In the case of real Satake category the corollary follows from the semi-smallness of the
convolution morphisms for real affine Grassmannian and the t-exactness criterion of nearby
cycles functor in [N2, Theorem 1.2.3]. On the other hand, the corresponding t-exactness cri-
terion for the relative Satake category is not obvious due the complicated spherical geometry
of the loop space X(K).
1.2.3. Formality and commutativity of dg Ext algebras. The next corollary confirms the
formality and commutativity conjecture for the real and relative Satake category (see,
e.g.,[BZSV, Conjecture 8.1.8]). Recall the dg extension algebras AR = RHomD(GR (OR )\GrR ) (δR , δR ⋆
ICreg ) and AX = RHomD(X(K)/G(O)) (ωX(O)/G(O) , ωX(O)/G(O) ⋆ ICreg ) for D(GR (OR )\GrR ) and
D(X(K)/G(O)) (see Section 13.2 for the precise definition). Note that both AR and AX
carry natural G∨ -actions induced from the one on ICreg .
The formality of AX is proved in [CY, Theorem 27] using a pointwise purity result for
IC-complexes of spherical orbits in X(K).1 Thus the formality for AR is a bit surprising and
non-obvious since Hodge theory or the theory of weights is not available on the real analytic
setting. The proof of the commutativity of H • (AR ) is similar to the case of complex groups
studied in [ABG, BFN].
Remark 1.8. The formality of AR in the case of the real quaternionic group GLn (H) was
proved in [CMNO] by a different method. It relies on an explicit computation of a morphism
A → AR from the dg Ext algebra A for the Satake category of the complex group GL2n to
AR . The argument in loc. cit. uses some particular properties of the real quaternionic group
which might not hold for other real groups (but it provides more information about AR ).
1In [CY, Theorem 27], we only treat the case of classical symmetric varieties. Thanks to the work of
Drinfeld and Bouthier [D, B], we now know that X(K) is G(O)-ind placid and the argument in loc. cit. can
be generalized to the general case. The details will appear in the revised version of [CY].
6
1.2.4. Hamiltonian duals of X and GR .
Definition 1.9. Introduce the affine schemes MX∨ = Spec(H • (AX )) and MR∨ = Spec(H • (AR )).
Inspired by the work of [BZSV] and [BFN] on relative Langlands duality and Coulomb
branches, we will call MX∨ and MR∨ the Hamiltonian dual of X and GR .
Let Dc (X(K)/G(O)) ⊂ D(X(K)/G(O)) and Dc (GR (O)\GrR ) ⊂ D(GR (O)\GrR ) be the
(non co-complete) full subcategories consisting of constructible complexes that are extensions
by zero off of substacks and let Dc (X(K)/G(O))0 ⊂ Dc (X(K)/G(O)) and Dc (GR (OR )\GrR )0 ⊂
Dc (GR (OR )\GrR ) be the full subcategories generated by the irreducible direct summands of
ωX(O)/G(O) ⋆ ICreg and δR ⋆ ICreg respectively. Denote by Coh(MX∨ /G∨ ) and Coh(MR∨ /G∨ )
the dg derived categories of coherent complexes on the stack MX∨ /G∨ and MR∨ /G∨ . The
following corollary follows from Corollary 1.7:
Corollary 1.10 (Theorem 13.5). (1) There is a G∨ -equivariant isomorphism MX∨ ≃ MR∨
(2) There are equivalences of categories
(1.2) Dc (X(K)/G(O))0 ≃ Coh(MX∨ /G∨ ) Dc (GR (OR )\GrR )0 ≃ Coh(MR∨ /G∨ ).
Remark 1.11. (1) It would be nice if one can find a description of MX∨ or MR∨ in terms of the
combinatoric structures of X or GR . In the recent work [BZSV], the authors proposed such
a description for a certain class of symmetric varieties (in fact, in loc. cit. they consider a
more general setting of spherical varieties) (2) Due to the existence of non-trivial equivariant
local systems on G(O)-orbits in X(K) (resp. GR (OR )-orbits in GrR ), the relative Satake
category D(X(K)/G(O)) (resp. real Satake category D(GR (OR )\GrR )) in general might be
bigger than Dc (X(K)/G(O))0 (resp. Dc (GR (OR )\GrR )0 ) and it is an interesting question
to extend the spectral description in (1.2) to to the entire dg derived category. In view of
Theorem 1.2, such a spectral description would imply a version of geometric Langlands on
the real projective line P1 (R), and vice versa.
1.2.5. Identification of dual groups. The paper [N1] associates to each real form GR ⊂ G a
connected complex reductive subgroup Hreal ∨
⊂ G∨ of the dual group.2 The construction of
∨
Hreal is via Tannakian formalism: the tensor category of finite-dimensional representations
∨
Rep(Hreal ) can be realized as a certain full subcategory QR ⊂ Perv(GR (OR )\GrR ) of perverse
∨
sheaves on GrR . In [N2, Section 10], a concrete description of Hreal is given including the
root datum and the Weyl group. On the other hand, the papers [GN1, GN2] associate
∨
to every spherical subgroup K ⊂ G a reductive subgroup Hsph ⊂ G∨ of the dual group.
∨
Again, the construction of Hsph is via Tannakian formalism: its tensor category of finite-
dimensional representations Rep(Hsph ∨
) can be realized as a certain full subcategory Qglob
K of
generic-Hecke equivariant perverse sheaves on the (global) moduli stack of quasi-maps with
∨
target X = K\G. However, unlike the real group case, a concrete description of Hsph is
∨
not known. For example, the fact that the root systems or the Weyl group of Hsph is the
same as that associated to X in the structure theory of spherical varieties [B, KS] remains
conjectural.
2While ∨
the notation suggests regarding Hreal itself as a dual group, at the moment we do not know of a
concrete role for its dual group.
7
Consider the case when K ⊂ G is the symmetric subgroup of a real form GR ⊂ G. Our
last example provides a tensor equivalence between QR ≃ QglobK and hence an isomorphism
∨ ∨
Hreal ≃ Hsph of reductive subgroups of G∨ . In particular, we obtain a concrete description
∨
of Hsph answering a basic open question in relative Langlands duality:
Corollary 1.12. There are horizontal tensor equivalences in the following commutative di-
agram of tensor functors
Rep(G∨ )
✈ ■■
✈✈ ■■
✈✈ ■■
✈ ■■
✈✈ ■$
{✈
✈ ≃
QR / Qglob
K
where the vertical arrows are given by the Hecke action of Rep(G∨ ) on the monoidal units.
Corollary 1.12 is the combination of Proposition 13.7, Proposition 13.12, and Theorem
13.14 in the text; we refer to Section 13.3 for a more detailed explanation of the statement.
The main ingredient in the proof is a local-global comparison theorem for relative Satake
category in Theorem 13.14.
Corollary 1.13. [GN1, Conjecture 7.3.2] There is an isomorphism of reductive groups
∨ ∨ ∨
Hsph ≃ Hreal . In particular, the root datum and Weyl group of Hsph are the same as that
associated to X in the theory of symmetric varieties.
Corollary 1.13 is restated in Theorem 13.15. In loc. cit. we also obtain a description of
irreducible objects in Qglob
K confirming a conjecture in [GN1].
1.2.6. Braidings. We conclude the introduction with the following conjecture. Note that
Theorem 1.1 implies that
Corollary 1.14. There is an equivalence
≃
(1.3) Perv(X(K)/G(O)) / Perv(GR (OR )\GrR )
of abelian categories.
1.4. Acknowledgements. The authors thank David Ben-Zvi, Alexis Bouthier, Pavel Etingof,
Mark Macerato, John O’Brien, Yiannis Sakellaridis, Jeremy Taylor, Akshay Venkatesh, Ruo-
tao Yang, and Lingfei Yi for many useful discussions. T.-H. Chen also thanks the Max Planck
Institute for Mathematics and D. Nadler the Miller Institute where parts of this work were
done. The research of T.-H. Chen is supported by NSF grant DMS-2143722 and that of
D. Nadler by NSF grant DMS-2101466.
2.1. Loop groups. Let GR be a connected real reductive algebraic group, and G = GR ⊗R C
its complexification. From this starting point, one constructs the following diagram of Lie
9
groups
(2.1) G
⑤⑤> O a❈❈❈
⑤⑤ ❈❈
⑤⑤ ❈❈
⑤ ⑤ ❈
K `❆ GO R G
= c
❆❆ ⑤⑤
❆❆ ⑤⑤
❆❆ ⑤⑤
❆ ⑤⑤
Kc
Here G = G(C) and GR = GR (R) are the Lie groups of complex and real points respectively,
Kc is a maximal compact subgroup of GR , with complexification K, and Gc is the maximal
compact subgroup of G containing Kc .
The real forms GR and Gc of G correspond to anti-holomorphic involutions η and ηc . The
involutions η and ηc commutes with each other and θ := ηηc = ηc η is an involution of G.
We have K = Gθ , GR = Gη , and Gc = Gηc . We fix a maximal split tours AR ⊂ GR and
a maximal torus TR such that AR ⊂ TR . We write A and T for the complexification of AR
and TR . We denote by ΛT the lattice of coweights of T and ΛA the lattice of real coweights.
We write Λ+ T the set of dominant coweight with respect to the Borel subgroup B and define
+
ΛA := ΛA ∩ Λ+ T . For any λ ∈ ΛT we define η(λ) ∈ ΛT as
c λ η
η(λ) : C× → C× → T → T,
where c is the complex conjugation of C× with respect to R× . The assignment λ → η(λ)
defines an involution on ΛT , which we denote by η, and ΛA is the fixed points of η.
We have a natural projection map
(2.2) σ : ΛT → ΛA σ(λ) = η(σ) + σ
whose image we denote by σ(ΛT ) ⊂ ΛA .
Let LG := G(C[t, t−1 ]) be the (polynomial) loop group associated to G. We define the
following involutions on LG: for any (γ : C× → G) ∈ LG we set
τ c γ η
η τ (γ) : C× → C× → C× → G → G
τ c γ ηc
ηcτ (γ) : C× → C× → C× → G → G.
Here τ (x) = x−1 is the the inverse map. Denote by K = C((t)) and O = C[[t]]. We have the
following diagram
G(K)
✈: O c❋❋ τ
❋❋ ηc
θ ✈✈✈ τ ❋❋
✈ η
✈✈ ❋❋
✈✈ ❋
K(K) LGO R ;
LGc
d■■ ✇✇
■■ ✇✇
■■ ✇
■■
■ ✇✇
✇✇
LKc
Here LGR and LGc are the fixed points subgroups of the involutions η τ and ηcτ on LG
respectively. Equivalently, LGR (resp. LGc ) is the subgroup of LG consisting of maps that
10
take the unit circle S 1 ⊂ C to GR (resp. Gc ). We define the based loop group ΩGc to be the
subgroup of LGc consisting of maps that take 1 ∈ S 1 to e ∈ Gc .
We define Gsym ⊂ G (resp. Gc,sym ⊂ Gc ) to be the fixed point subspace of the involution
θ̃ = θ−1 on G (resp. Gc ) and write G0sym (resp. G0c,sym) be the identity component. The
map π : G → G, g → θ̃(g)g induces a G-equivariant isomorphism ι : X ≃ G0sym (resp. Gc -
equivariant isomorphism ιc : Xc ≃ G0c,sym ). We have a natural embedding A ֒→ G0sym ≃ X.
Let q : G → K\G = X be the quotient map.
In this paper, we assume GR (and equivalently K) is connected. Throughout this paper,
we will be concerned exclusively with the topology of loop spaces and related moduli and
ignore their potentially non-reduced structure.
Proof. The isomorphism ι : X ≃ G0sym induces an embedding X(K) ⊂ G(K) and it follows
from Proposition 2.1 (1) that X(K)λ := X(K) ∩ G(O)tλ G(O) is nonempty if and only if
λ ∈ Λ+ λ
A . Since P = ΩX
S c ∩ X(K)
λ
,Sit implies P λ is non-empty if and only if λ ∈ Λ+
A and we
conclude that ΩXc = λ∈Λ+ P λ = λ∈λ+ P λ .
T A
Proposition 3.2. We have the following.
(1) The projection q : ΩGc → ΩXc maps OλK into P λ and the resulting map OλK → P λ
induces a homeomorphism ΩKc \OλK ≃ P λ.
12
(2) The projection q : ΩGc → ΩXc maps OλR into Qλ and the resulting map OλR → Qλ
induces a homeomorphism ΩKc \OλR ≃ Qλ
(3) There is a Kc -equivariant stratified homeomorphism ΩKc \ΩGc ≃ ΩXc .
Proof. Fix λ ∈ Λ+ τ
A . Proposition 2.1 together with the fact that θ̃ = η̃ on ΩGc imply
q(OλK ) ⊂ P λ and q(OλR ) ⊂ Qλ . By [N1, Proposition 6.3], the restriction of the energy function
E to P λ is Bott-Morse and B λ is the only critical manifold. Since Kc is connected and acts
transitively on B λ , it follows that B λ and hence P λ connected. Now [N2, Proposition 6.4]
impliesSP λ ⊂ q(ΩGc ). Thus q(OλK ) = P λ and part (1) follows. For part (2) we observe that
Qλ = λ≤µ,µ∈Λ+ Qλ ∩ P µ . Since B λ = Qλ ∩ P λ is in the closure of Qλ ∩ P µ , Lemma 3.1
SS
implies Qλ = λ≤µ,µ∈Λ+ Qλ ∩ P µ and part (1) implies Qλ ⊂ q(ΩGc ), hence Qλ = q(OλR ). Part
A
(2) follow. Part (3) follows from part (1) and Lemma 3.1.
Corollary 3.3. K(K) and LGR -orbits on Gr are transversal.
Proof. By Proposition 3.2, it suffices to show that the strata P λ and Qµ in ΩXc are transver-
sal. This follows from the fact that the orbits S λ and T λ on ΩGc are transversal and both
S λ , T λ are invariant under the involution θ̃ on ΩGc as θ̃ = η̃ τ on ΩGc and S λ (resp. T λ ) is
θ̃-invariant (resp. η̃ τ -invariant).
3.3. The energy flow on ΩGc . We recall the construction of energy flow on ΩGc following
[PS, Section 8.9]. For any γ ∈ LGc and v ∈ Tγ LGc we denote by γ −1 v ∈ Lgc (resp.
vγ −1 ∈ Lgc ) the image of v ∈ Tγ LGc under the isomorphism Tγ LGc ≃ Te LGc ≃ Lgc induced
by the left action (resp. right action).
Fix a Gc -invariant metric h, i on gc . Observe that the formula
Z
ω(v, w) := h(γ −1 v)′ , γ −1 widθ
S1
defines a left invariant symplectic form on Tγ ΩGc . According to [PS, Theorem 8.6.2], the
composition ΩGc → G(K) → Gr defines a diffeomorphism
ΩGc ≃ Gr.
Let Jγ be the automorphism of Tγ ΩGc which corresponds to multiplication by i in terms
of the complex structure on Gr. The formula g(v, w) = ω(v, Jγ w) defines a positive inner
product on Tγ ΩGc and the Kähler form on Tγ ΩGc is given by g(v, w) + iω(v, w). Finally,
for any smooth function F : ΩGc → R there corresponds so-called Hamiltonian vector field
R(γ) and gradient vector field ∇F (γ) on ΩGc characterized by
ω(R(γ), v) = dF (γ)(u), g(∇F (γ), u) = dF (γ)(u).
Consider the energy function on ΩGc :
Z
′ ′
(3.1) E : ΩGc → R, γ → (γ , γ )γ = hγ −1 γ ′ , γ −1 γ ′ idθ.
S1
Lemma 3.5. The K(K)-orbits and LGR -orbits are stable under the rotation flow γa (t) (see
Proposition 3.4).
Proof. We give a proof for the case of K(K)-orbits. The proof for the LGR -orbits is similar.
Let OλK be a K(K)-orbit and let γ = γ(t) ∈ OλK . By Proposition 2.1, we need to show
that θ̃(γa )γa ∈ G(C[t])tλ G(C[t]). A direct computation shows that θ̃(γa )γa = θ(γ(a))θ̃(γ(t +
a))γ(t + a)γ(a)−1 . Note that θ̃(γ(t + a))γ(t + a) ∈ G(C[t])tλ G(C[t]) as γ(t) ∈ OλK , the desired
claim follows.
3.4. The Matsuki flow on Gr. The Cartan decomposition gR = kR ⊕ pR induces a decom-
position of gc = kc ⊕ ipR , gR = kc ⊕ pR and the corresponding loop algebra Lg = Lk ⊕ Lp,
Lgc = Lkc ⊕ L(ipR ), LgR = Lkc ⊕ LpR .
Recall the non-degenerate bilinear form (, )γ on Tγ LGc
Z
(v1 , v2 )γ := hγ −1 v1 , γ −1 v2 idθ.
S1
Let γ ∈ LGc and Tγ (LKc · γ) ⊂ Tγ LGc be the tangent space of the LKc -orbit LKc · γ through
γ. The bilinear form above induces an orthogonal decomposition
Tγ LGc = Tγ LKc · γ ⊕ (Tγ LKc · γ)⊥
and for any vector v ∈ Tγ LGc we write v = v0 ⊕ v1 where v0 ∈ Tγ LKc · γ, v1 ∈ (Tγ LKc · γ)⊥ .
Note that we have
(3.2) γ −1 v0 ∈ Adγ −1 Lkc , γ −1 v1 ∈ Adγ −1 L(ipR ).
Recall that the loop group ΩGc can be identified with a “co-adjoint” orbit in LGc via the
embedding
ΩGc ֒→ Lgc , γ → γ −1 γ ′ .
Consider the following functions on ΩGc
Z
′ ′
E : ΩGc → R, γ → (γ , γ )γ = hγ −1 γ ′ , γ −1 γ ′ idθ,
S1
Z
′ ′
E0 : ΩGc → R, γ → (γ0 , γ0 )γ = hγ −1 γ0′ , γ −1 γ0′ idθ,
S1
14
Z
E1 : ΩGc → R, γ → (γ1′ , γ1′ )γ = hγ −1 γ1′ , γ −1 γ1′ idθ.
S1
Note that E is the energy function in (3.1).
Lemma 3.6. Recall the map π : ΩGc → ΩGc , γ → θ(γ)−1 γ. We have
(3.3) 4E1 = E ◦ π : ΩGc → R.
In particular, the function E1 is LKc -invariant.
Proof. Write ||v|| = hv, vi for v ∈ gc . For any γ ∈ ΩGc we have
Z Z
−1 ′
E ◦ π(γ) = ||π(γ) π(γ) ||dθ = ||γ −1γ ′ − γ −1 η(γ)′ η(γ)−1 γ||dθ.
S1 S1
Note that γ −1 γ ′ − γ −1 θ(γ)′ θ(γ)−1 γ = 2γ −1 γ1′ , hence we have ||γ −1 γ ′ − γ −1 θ(γ)′ θ(γ)−1 γ|| =
4||γ −1γ1′ ||. The lemma follows.
Lemma 3.7. The Hamiltonian vector field on ΩGc which correspond to E1 (resp. E0 ) is
given by
γ → R1 (γ) = γ1′ − γγ1′ (0) (resp. γ → R0 = γ0′ − γγ0′ (0)).
In particular, we have
γ −1 R1 (γ) ∈ Adγ −1 LipR + ipR (resp. γ −1 R0 (γ) ∈ Adγ −1 LkR + kR ).
Proof. Since R0 (γ) + R1 (γ) = R(γ) = γ ′ − γγ(0)′ , it is enough to show that R1 (γ) =
γ1′ − γγ1′ (0). Let γ ∈ ΩGc , x = π(γ) = θ(γ)−1 γ, and u ∈ Tγ ΩGc . According to Proposition
3.4 and Lemma 3.6, we have
4dE1 (γ)(u) = π ∗ dE(γ)(u) = dE(x)(π∗ u) = ω(x′ , π∗ u) = ω(x−1 x′ , x−1 π∗ u).
Using the equalities x−1 x′ = 2γ −1 γ1′ , x−1 π∗ u = 2γ −1 u1 , and the fact that hγ −1 γ1′ , (γ −1 u0)′ i =
0, we get
Z Z
−1 ′ −1 −1 ′ −1 ′
4dE1 (γ)(u) = 4ω(γ γ1 , γ u1 ) = 4 hγ γ1 , (γ u1 ) idθ = 4 hγ −1 γ1′ , (γ −1 u)′ idθ
S1 S1
Z
=4 hγ −1 γ1′ − γ1′ (0), (γ −1u)′ idθ = 4ω(R1 (γ), u).
S1
The lemma follows.
S S
Let ΩGc = λ∈Λ+ OλK and ΩGc = λ∈Λ+ OλR be the K(K)-orbits and LGR -orbits stratifi-
A A
cations of ΩGc . Let Oλc = OλK ∩ OλR which is a single LKc -orbit.
Proposition 3.8. Let E1 : ΩGc → R be the function above and ∇E1 be the corresponding
gradient vector field.
(1) ∇E1 is tangential
F to both OλK and OλR ,
(2) The union λ∈Λ+ Oλc is the critical manifold of ∇E1 .
A
15
(3) For any γ ∈ Oλc , let Tγ ΩGc = T + ⊕T 0 ⊕T − be the orthogonal direct sum decomposition
into the positive, zero, and negative eigenspaces of the Hessian d2 E1 . We have
Tγ OλK = T + ⊕ T 0 , Tγ OλR = T − ⊕ T 0 .
Proof. Proof of (1). We first show that ∇E1 is tangential to OλR = ΩGc ∩ LGR tλ G(C[t]).
Since the tangent space Tγ OλR at γ ∈ OλR is identified, by left translation, with the space
Ωgc ∩ (Adγ −1 LgR + g(C[t])) ⊂ Ωgc
it suffices to show that γ −1 ∇E1 (γ) ∈ Adγ −1 LgR + g(C[t]). Recall that, by Proposition 3.4,
we have γ −1 ∇E1 (γ) = J(γ −1 R1 (γ)). Note that J(v) + iv ∈ g(C[t]) for v ∈ Lg and by Lemma
3.7 we have
iγ −1 R1 (γ) = i(γ −1 (γ1′ − γγ1′ (0)) ∈ Adγ −1 LpR + pR .
All together, we get
γ −1 ∇E1 (γ) = −iγ −1 R1 (γ) + (J(γ −1 R1 (γ)) + iγ −1 R1 (γ)) ∈ Adγ −1 LpR + g(C[t])
which is contained in Adγ −1 LgR + g(C[t]). We are done. The same argument as above,
replacing LGR by K(K), shows that the gradient field ∇E0 of E0 is tangential to OλK . Since,
by Corollary 3.5, the orbit OλK is a complex submanifold of ΩGc = Gr invariant under the
rotation flow γa (t), it follows from Proposition 3.4 that ∇E is tangential to OλK . Since
∇E1 = ∇E − ∇E0 , we conclude that ∇E1 is also tangential to OλK . This finishes the proof
of (1).
Proof of (2) and (3). By proposition 3.2 and lemma 3.6, the function E1 factors as
π E
E1 : ΩGc → ΩXc ⊂ ΩGc → R.
Thus to prove (2) and (3), it is enough to prove following:
F
(i) The union λ B λ is the critical manifold of the restriction E to ΩXc ,
eigenspaces of the Hessian E. Note that θ̃ induces a linear map on Tγ ΩGc , which we still
denoted by θ̃, and we have Tγ ΩXc = (Tγ ΩGc )θ̃ is the fixed point subspace. So to prove (i)
and (ii) it suffices to show that the subspaces Tγ S λ and Tγ T λ are θ̃-invariant. It is true, since
θ̃ = η̃ τ on ΩGc and S λ (resp. T λ ) is θ̃-invariant (resp. η̃ τ -invariant). This finished the proof
of (2) and (3).
Theorem 3.9. The gradient ∇E1 and gradient-flow φt associated to the LKc -invariant func-
tion E1 : Gr → R and the LGc -invariant metric g(, ) satisfy the following:
F
(1) The critical locus ∇E1 = 0 is the disjoint union of LKc -orbits λ∈Λ+ Oλc
A
(2) The gradient-flow φt preserves the K(K)-and LGR -orbits.
16
(3) The limits lim φt (γ) of the gradient-flow exist for any γ ∈ Gr. For each LKc -orbit
t→±∞
Oλc in the critical locus, the stable and unstable sets
(3.4) OλK = {γ ∈ Gr| lim φt (γ) ∈ Oλc } OλR = {γ ∈ Gr| lim φt (γ) ∈ Oλc }
t→∞ t→−∞
Proof. Part (1) and (2) follows from Proposition 3.8. The LKc -invariant function E1 , respec-
tivley the LGc -invariant metric g(, ), and the flow φt , descends to a Kc -invariant Morse-Bott
function E 1 : ΩKc \Gr → R, respectivley a Kc -invariant metric g(, ) on ΩKc \Gr, and a flow
φt . Since the function E 1 is bounded below and the quotient ΩKc \OλK is finite dimensional
S
with ΩKc \OλK = µ≤λ ΩKc \OµK , Proposition 3.8 and standard results for gradient flows (see,
e.g., [AB, Proposition 1.19] or [P, Theorem 1]) imply that the limit lim φt (γ) exists for any
t→±∞
γ ∈ ΩKc \Gr and ΩKc \OλK is the stable manifold for ΩKc \Oλc and ΩKc \OλR is the unstable
manifold for ΩKc \Oλc . Part (3) and (4) follows.
4.1. Real affine Grassmannians. We recall results from [N1] about the real affine Grass-
mannian. Let GrR := GR (KR )/GR (OR ) be the real affine Grassmannian. For any λ ∈ Λ+ T we
denote by S λ and T λ the G(O) and G(C[t−1 ])-orbit of tλ ∈ Gr. The orbits S λ and T µ on Gr
are transversal and the intersection C λ = S λ ∩ T λ is isomorphic to the flag manifold G/P λ
where the parabolic subgroup P λ is the stabilizer of λ. The affine Grassmannian Gr is the
disjoint union of the orbits S λ (resp. T λ ) for λ ∈ Λ+
T
G G
Gr = S λ (resp. Gr = T λ)
λ∈Λ+
T λ∈Λ+
T
and we have
λ G G
S = Sµ (resp. T λ = T µ ).
µ≤λ λ≤µ
The intersection of S λ (resp. T λ ) with GrR is nonempty if and only if λ ∈ Λ+ A and we write
+
λ λ
SR (resp. TR ), λ ∈ ΛA for the intersection. We define CR to be the intersection of SRλ and
λ
TRλ . SRλ (resp. TRλ ) is equal to the GR (OR )-orbit (resp. GR (R[t−1 ])-orbit) of tλ and CRλ is
isomorphic to the real flag manifold GR /PRλ where the parabolic subgroup PRλ ⊂ GR is the
17
stabilizer of λ. The real affine Grassmannian Gr is the disjoint union of the orbits SRλ (resp.
TRλ ) for λ ∈ Λ+
T
G G
(4.1) GrR = SRλ (resp. GrR = TRλ )
λ∈Λ+
A λ∈Λ+
A
and we have G G
λ
SR = SRµ (resp. TRλ = TRµ ).
µ≤λ λ≤µ
(n)
Y
k
(1)
(4.3) GrR |P1(R)p ≃( GrR )|P1 (R)k0
i=1
Consider the case when n = 2m is even. Let σ2m : (P1 )2m → (P1 )2m be the complex con-
jugation given by σ2m (z1 , ..., z2m ) = (c(zm+1 ), ..., c(z2m ), c(z1 ), ..., c(zm )). Then σ2m together
with η defines anti-holomorphic involutions of G(K)(2m) , G(O)(2m) , Gr(2m) and we write
(σ ) (σ ) (σ ) (σ ) (σ )
G(K)R 2m , G(O)R 2m , GrR 2m = G(K)R 2m /G(O)R 2m for the corresponding real forms. The
projection map (P1 )2m → (P1 )m sending (z1 , ..., z2m ) → (z1 , ..., zm ) restricts to a real analytic
isomorphism ((P1 )2m )(σ2m ) ≃ (P1 )m and we have a natural projection map
(σ )
(4.4) GrR 2m → (P1 )m .
(σ )
Y
k
(σ )
(4.6) GrR 2m |(P1 )p ≃ ( GrR 2 )|(P1 )k0
i=1
(σ )
Example 4.1. Consider the case when m = 1. Then the fibers of the family GrR 2 → P1
(σ ) (σ )
over a point z ∈ P1 are given by GrR 2 |z=z̄ ≃ Grz,R and GrR 2 |z6=z̄ ≃ Grz .
19
5. Multi-point generalizations
We study multi-point version of loop groups and loop spaces and we establish a multi-
point generalization of Gram-Schmidt factorization for loop groups. The main reference is
[CN1].
5.1. Multi-point version of real loop groups. Consider the functor LG(n) that assigns
to an affine scheme S the set of sections
Assume n = 2m. The conjugations σ2m and η together define an anti-holomorphic involution
on LG(n) and we write
(σ )
LGR 2m → (P1 )m
for the corresponding semi-analytic space. We will write
(m) (σ ) (m) (σ )
LGR := LGR 2m |Hm → Hm L− GR := LGR 2m |P1 (R)m → P1 (R)m
(σ )
for the restriction of LGR 2m to Hm and P1 (R)m respectively. Concretely, we have
(m)
LGR = {(x, γ)|x ∈ Hm , γ : P1 \ |x| ∪ |x̄| → G satisfying γ(P1 (R)) ⊂ GR }.
(m)
L− GR = {(x, γ)|x ∈ P1 (R)m , γ : P1 \ |x| → G satisfying γ(P1 (R) \ |x|) ⊂ GR }.
We consider the subgroup indscheme ΩG(n) ⊂ LG(n) |Cn consisting of (x, γ) ∈ LG(n) |Cn
such that γ(∞) = e. Since the conjugations cn and σ2m fixed ∞ ∈ P1 , they induces an
involution of ΩG(2m) and let
(σ )
ΩGR 2m → Cm
be the associated semi-analytic space and the subspaces
(m) (σ ) (m) (σ )
ΩGR := ΩGR 2m |Hm → Hm L−
∞ GR := ΩGR 2m |Rm → Rm .
(1)
Example 5.1. Assume m = 1. Then the fiber of LGR over x ∈ H is equal to
(1)
LGR |x = Lx GR := {γ : P1 \ x ∪ x̄ → G satisfying γ(P1 (R)) ⊂ GR }
z−x (1)
The change of coordinate t = z−x̄
induces isomorphism Lx GR ≃ LGR . The fiber of L− GR
over x ∈ P1 (R) is equal to
(1)
L− GR |x = L− 1 1
x GR := {γ : P \ x → G satisfying γ(P (R) \ x) ⊂ GR }
(m) (m)
Lemma 5.2. There is an isomorphism L− Gc ≃ Gc × P1 (R)m (resp. L−
∞ Gc ≃ {e} × Rm )
of group schemes.
Proof. Indeed the lemma above (in the case m = 1) implies Gc (R[t−1 ]) = Gc . Since the
Gc (R[t−1 ])-orbit in Grc = Gc (KR )/Gc (OR ) through the based point is open dense (see Section
4.1), it implies Grc is a point and hence Gc (KR ) = Gc (OR ).
Proposition 5.4. (1) For any partition p ∈ p(m), there is a natural homeomorphism
Y
k
(5.1) ΩG(σ
c
2m )
|(P1 )p ≃( ΩG(σ
c
2)
)|(P1 )k0
i=1
Y
k
(5.2) ΩGc(m) |Hp ≃( ΩG(1)
c )|H0k
i=1
Proof. Proof of (1) and (2). [CN1, Theorem 3.3 (3)] implies that the natural multiplication
Q (σ ) (σ )
map ( ki=1 ΩGc 2 )|(P1 )k0 → ΩGc 2m |(P1 )p is a homeomorphism.
(m)
There is a natural map ΩGc → G(K)(m) |Hm sending (x, γ) to (x, γ|Γ̂0x ), where γ|Γ̂0x is the
restriction of the section γ : P1 \ |x| ∪ |x̄| → G to Γ̂0x . It induces a map
and [?, Theorem 3.3 (2)] implies that it is a homeomorphism. Part (3) follows. Since
Gr(m) ≃ G(K)(m) /G(O)(m) , part (3) implies part (4).
Example 5.5. Assume m = 1. Then Proposition 5.4 (4) over a point z ∈ H specializes to
the well-known Gram-Schmidt factorization for loop groups Ωz Gc × G(Oz ) ≃ G(Kz ).
21
5.3. Multi-point version of ΩXc . For any non-negative integer m, we define
ΩXc(m) = {(x, γ)|x ∈ Hm , γ : P1 \ |x| ∪ |x̄| → X satisfying γ(P1 (R)) ⊂ Xc , γ(∞) = e}.
We have natural projection map
ΩXc(m) → Hm
(m)
and we denote by Ωz Xc the fiber over z ∈ Hm . The involution θ̃ on G induces an
(m)
involution on ΩGc , still denoted by θ̃, and the isomorphism ι : X ≃ G0sym = (Gθ̃ )0 induces
an isomorphism
ι(m) : ΩXc(m) ≃ (ΩGc(m) )θ̃ .
We define
P (m),λp = (ι(m) )−1 (S (m),λp )θ̃ )
(m)
The collection {P (m),λp }λp :p→Λ+ forms a stratifiaction of ΩXc . We have the following multi-
A
point version of Proposition 3.2.
Proposition 5.6. The quotient map q : G → X induces a Kc -equivariant stratified homeo-
morphisms
ΩKc(m) \ΩGc(m) ≃ ΩXc(m)
which restricts to Kc -equivariant homeomorphisms on stratum
(m),λp
ΩKc(m) \OK ≃ P (m),λp
(m) (m)
Proof. Consider the natural map q (m) : ΩGc → ΩXc . We shall show that, for any
(m) (m)
z ∈ Hm , the map q (m) : Ωz Gc → Ωz Xc on the corresponding fiber is surjective. Since
(m) (m) (m)
ΩGc is ind-proper, we conclude that the natural map q (m) : ΩGc → ΩXc is surjective
(m) (m (m)
and closed, and it follows that the induced map ΩKc \ΩGc → ΩXc is a continuous,
(m),λ
closed, bijective map, and hence a homeomorphism. Since q (m) (OK p ) ⊂ P (m),λp , it implies
(m) (m),λ
the induced map ΩKc \OK p → P (m),λp is also a homeomorphism. The Kc -equivariance
is clear.
(m),λp (m) (m),λp
Proof of the claim. Let Pz = P (m),λp ∩ Ωz Xc . It suffices to show Pz ⊂
(m) (m) (m)
(m) (m) ˜ (m) G(O)z (m)
q (Ωz Gc ). Let Grz ×Grz = G(K)z × Grz . The multi-point Gram-Schmidt
factorization
Ωz Gc(m) × G(O)z(m) ≃ G(K)z(m) , Ωz Gc(m) ≃ Grz(m)
(m) (m) (m) (m)
in Proposition 5.4 implies that the natural inclusion Ωz Gc × Ωz Gc → G(K)z × G(K)z
induces a natural homeomorphism
Ωz Gc(m) × Ωz Gc(m) ≃ Grz(m) ×Gr
˜ z(m) .
Moreover, we have the following commutative diagram
(m) (m) ≃
Ωz Gc × Ωz Gc / Grz(m) ×Gr
˜ z(m)
mult p
(m) ≃
Ωz Gc / Grz(m)
22
(m)
where mult is the multiplication map and p is the convolution map. Pick a s ∈ Ωz Gc
(m),λp θ̃ (m),λp (m)
such that γ = θ̃(s)s ∈ (Sz ) ≃ Pz . Consider the G(O)z -action on Grz(m) ×Gr ˜ z(m)
(m) (m) (m)
given by g+ (γ1 , γ2 ) = (g+ γ1 , γ2 ) and S be the G(O)z -orbit of (θ̃(s), s) ∈ Ωz Gc ×Ωz Gc ≃
˜ z(m) . Let Z be the intersection of the closure S of S with the pre-image p−1 (Sz(m),λp ).
Grz(m) ×Gr
(m) (m),λp
Note that Z is a disjoint union of G(O)z -orbits and p : Z → Sz is proper and restricts
to a submersion on each orbit.
(m) (m) (m)
Consider the involution sw ◦θ̃ on Ωz Gc × Ωz Gc ≃ Ωz Gc ≃ Grz(m) ×Gr ˜ z(m) given by
sw ◦θ̃(γ1 , γ2 ) = (θ̃(γ2 ), θ̃(γ1 )). The lemma below implies that Z is stable under the involution
(m),λp
sw ◦θ̃ and the map p : Z → Sz is compatible with the involutions sw ◦θ̃ on Z and θ̃
(m),λp
on Sz . Now we can apply the general lemma [N1, Proposition 6.4] to conclude that
(m),λp ,◦ (m),λp ,◦ (m),λp (m),λp θ̃
Pz ⊂ p(Z sw ◦θ̃ ) where Pz ⊂ Pz ≃ (Sz ) is a connected component of
(m),λp
containing γ = θ̃(s)s. By Theorem 7.5, there is a homeomorphism between Pz and a fiber
(m),λp (m)
of the strata SR ⊂ GrR in the real Beilinson-Drinfeld grassmannian. Since the latter
fiber is a vector bundle over the real flag manifold GR /PR (see [N2, Proposition 3.6.1]) which
(m),λp (m),λp ,◦
is connected (we assume GR is connected), we conclude that Pz = Pz ⊂ p(Z sw ◦θ̃ ).
Now we can conclude the proof by noting there is natural identification of the sw ◦θ̃-fixed
˜ z(m) with Ωz Gc(m) so that p restricts to Ωz Gc(m) coinsides with q (m) .
points of Grz(m) ×Gr
(m) (m) (m)
Lemma 5.7. The involution sw ◦θ̃ on Ωz Gc × Ωz Gc ˜ z(m) sends the G(O)z -
≃ Grz(m) ×Gr
orbit of (γ1 , γ2 ) to the (θ̃(γ2 ), θ̃(γ1 )).
(m)
Proof. Let (a, b) = g(γ1, γ2 ) ∈ Grz(m) ×Gr
˜ z(m) with g ∈ G(O)z . By the factorization home-
(m)
omorphism in Proposition 5.4, it implies (gγ1, γ2 ) = (au, u−1bg ′ ) in G(K)z for some
(m) (m)
g ′ , u ∈ G(O)z . Let h = θ(g ′ ) ∈ G(O)z . Then a direct computation show show that
we have
(hθ̃(γ2 ), θ̃(γ1 )) = (θ̃(b)θ̃(u−1 ), θ̃(u)θ̃(a)θ̃(g −1 ))
(m) (m)
in G(K)z × G(K)z and it implies
h(θ̃(γ2 ), θ̃(γ1 )) = (θ̃(b), θ̃(a))
in Grz(m) ×Gr
˜ z(m) . The lemma follows.
6.1. Stack of real bundles. Let BunG (P1 ) be the moduli stack of G-bundles on the complex
projective line P1 . The standard complex conjugation z → z̄ on P1 together with the
involution η of G defines a real structure c : BunG (P1 ) → BunG (P1 ) on BunG (P1 ) with real
form BunGR (P1R ), the real algebraic stack of GR -bundles on the projective real line P1R . We
write BunG (P1 )R for the real analytic stack of real points of BunGR (P1R ). By definition, we
23
have BunG (P1 )R ≃ ΓR \YR where Y → BunGR (P1R ) is a R-surjective presentation3 of the real
algebraic stack BunGR (P1R ), Γ = Y ×BunG (P1R ) Y is the corresponding groupoid, and XR , ΓR
R
are the real analytic spaces of real points of X, Γ (see Appendix A).
A point of BunG (P1 )R is a GR -bundle ER on P1R and, by descent, corresponds to a pair
(E, γ) where E is a G-bundle on P1 and γ : E ≃ c(E) is an isomorphism such that the induced
composition is the identity
γ c(γ)
E → c(E) → c(c(E)) = E.
We call such pair (E, γ) a real bundle on P1 and BunG (P1 )R the stack of real bundles on P1 .
For any GR -bundle ER , the restriction of ER to the (real) point ∞ is a GR -bundle on
Spec(R) and the assignment ER → ER |∞ defines a morphism
BunGR (P1R ) −→ BGR .
For each α ∈ H 1 (Gal(C/R), G), let Tα be a GR -torsor on Spec(R) in the isomorphism class of
α and we define GR,α = AutGR (Tα ). The collection {GR,α , α ∈ H 1 (Gal(C/R), G)} is the set
of pure inner forms of GR . Let GR,α = GR,α (R) be the real analytic group associated to GR,α .
We denote by α0 the isomorphism class of trivial GR -torsor with real group GR,α0 = GR . By
Example A.4, the morphism above induces a morphism
G
cl∞ : BunG (P1 )R −→ 1
BGR,α
α∈H (Gal(C/R),G)
We will call BunGR,α (P1 (R)) the stack of GR,α -bundles on P1 (R).
Example 6.1. Consider G = C× . In the case η is the split conjugation, the cohomology
group H 1 (Gal(C/R), G) is trivial and we have
BunG (P1 )R ≃ Z × BR× .
In the case η = ηc is the compact conjugation, we have H 1 (Gal(C/R), G) = {α0 , α1 } ≃ Z/2Z
and
BunG (P1 )R ≃ BunGR,α1 (P1 (R)) ∪ BunGR,α2 (P1 (R)),
where BunGR,αi (P1 (R) ≃ BS 1 .
6.2. Uniformizations of real bundles. We shall introduce and study two kinds of uni-
formization of real bundles: one uses a real point of P1 called the real uniformization the
other uses a complex point of P1 called the complex uniformization.
3A presentation of a real algebraic stack is R-surjective if it induces a surjective map on the isomorphism
classes of R-points.
24
6.2.1. Real uniformizations. The unifomization morphism
u : Gr → BunG (P1 )
for BunG (P1 ) exhibits Gr as a G(C[t−1 ])-torsor over BunG (P1 ), in particular, we have an
isomorphism
(6.2) G(C[t−1 ])\Gr ≃ BunG (P1 ).
The map u is compatible with the real structures on Gr and BunG (P1 ) and we denote by
(6.3) uR : GrR → BunG (P1 )R
the associated map between the corresponding semi-analytic stacks of real points. We call
the morphism uR the real uniformization. It follows from (6.2) that uR factors through an
embedding
(6.4) GR (R[t−1 ])\GrR → BunG (P1 )R .
We shall describe the image of uR .
Proposition 6.2. The map uR factors through
uR : GrR → BunGR (P1 (R)) ⊂ BunG (P1 )R
and induces an isomorphism of semi-analytic stacks
∼
GR (R[t−1 ])\GrR −→ BunGR (P1 (R)).
Proof. Since every GR -bundle ER in the image of uR is trivial over P1R −{0}, in particular at ∞,
we have ER ∈ BunGR (P1 (R)). Thus the map uR factors through BunGR (P1 (R)). We show that
the resulting morphism uR : GrR → BunGR (P1 (R)) is surjective. Let f : S → BunGR (P1 (R))
be a smooth presentation (note that S is smooth as BunGR (P1 (R)) is smooth). It suffices
to show that, étale locally on S, f admits a lifting to GrR . Consider the fiber product
Y := S ×BunGR (P1 (R)) GrR and we denote by h : Y → S the natural projection map. It
suffices to show that h is surjective and admits a section étale locally on S. By Theorem
1.1 in [MS], every GR -bundle ER on P1R which is trivial at ∞ admits a trivialization on
P1R − {0}. It implies h is surjective. To show that h admits a section, we observe that Y is
a real analytic ind-space smooth over GrR and, as uR is formally smooth, for any y ∈ Y and
s = h(y) ∈ S, the tangent map dhy : Ty Y → Ts S is surjective. Choose a finite dimensional
subspace W ⊂ Ty Y such that dhy (W ) = Ts S. We claim that there exists a smooth real
analytic space U ⊂ Y such that y ∈ U and Ty U = W . This implies h|U : U → S is smooth
around y, thus f admits a section étale locally around s = h(y). Finally, by (6.4), we obtain
an isomorphism GR (R[t−1 ])\GrR ≃ BunGR (P1 (R)).
To prove the claim, we observe that Y is locally isomorphic to GrR times a smooth real
analytic space. So it suffices to show for any finite dimensional subspace W ⊂ Te GrR , there
exists a smooth real analytic space U such that Te U = W . This follows from the fact that
the exponential map exp : Te GrR → GrR associated to the metric g(, )|GrR (here g(, ) is the
metric on Gr in Section 3.3) is a local diffeomorphism.
25
We can also consider the uniformization map Grz → BunG (P1 ) for any real point x ∈ P1 (R)
and the discussion above applies to this setting and hence we have obtain real uniformization
at x
(6.5) ux,R : L− 1
x GR \Grx,R ≃ BunGR (P (R)).
where L−
x GR is real analytic group in Example 5.1.
6.2.2. Complex uniformizations. We now discuss complex uniformizations. The natural map
u(2) : Gr(2) → BunG (P1 ) (x, E, φ) → E
is compatible with the complex conjugation σ2 on Gr(2) and the natural one on BunG (P1 )
and we denote by
(σ ) (σ )
(6.6) uR 2 : GrR 2 −→ BunG (P1 )R
the map between the corresponding real analytic stacks.
(σ ) (σ )
Let us study the restriction of uR 2 to the fiber GrR 2 |x over some x ∈ P1 . If x ∈ P1 (R),
(σ )
then we have GrR 2 |x ≃ Grx,R (see Example 4.1) and the restriction
(σ ) (σ )
ux,R := uR 2 |x : Grx,R ≃ GrR 2 |x → BunG (P1 )R
of (6.6) is isomorphic to the real unifomization map in (6.3). If x ∈ H, then we have
(σ )
GrR 2 |x ≃ Grx and we will call the map
(σ ) (σ )
(6.7) ux,C := uR 2 |x : Grx ≃ GrR 2 |x −→ BunG (P1 )R
the complex uniformization associated to x.
We shall give a description of ux,C . Let (E, φ) ∈ Grx where E is a G-bundle on P1 and
φ : E|P1−{x} ≃ G × (P1 − {x}) is a trivialization of E over P1 − {x}. Let c(E) be complex
conjugation of E (see Sect. 6.1) and let F be the G-bundle on P1 obtained from gluing
of E|P1−{x̄} and c(E)|P1−{x} using the isomorphism c(φ)−1 ◦ φ : E|P1 −{x,x̄} ≃ c(E)|P1−{x,x̄} .
By construction, there is a canonical isomorphism γ : F ≃ c(F) and the resulting real
bundle (F, γ) ∈ BunG (P1 )R is the image ux,C ((E, φ)). Note that the cohomology class in
H 1 (Gal(C/R), G) given by the restriction of the real bundle F to ∞ is represented by the co-
boundary c(φ(v))−1 (φ)(v) (here v ∈ E|∞ ), hence is trivial. Thus the complex uniformization
ux,C factors as
ux,C : Grx → BunGR (P1 (R)).
We shall describe the image of ux,C . For each z ∈ C× let az : P1 → P1 be the multiplication
map by z. Consider the flows on Gr(2) and BunG (P1 ):
(6.8) ψz : Gr(2) → Gr(2) , (x, E, φ) → (az (x), (az −1 )∗ E, (az −1 )∗ φ).
ψz : BunG (P1 ) → BunG (P1 ), E → (az −1 )∗ E
For z ∈ R>0 the flows above restrict to flows
(σ ) (σ )
(6.9) ψz1 : GrR 2 → GrR 2 ψz2 : BunG (P1 )R → BunG (P1 )R
26
and we have the following commutative diagram
(σ ) ψz1 (σ )
(6.10) GrR 2 / GrR 2 .
(σ2 ) (σ2 )
uR uR
ψz2
BunG (P1 )R / BunG (P1 )R
Lemma 6.3. We have the following properties of the flows:
(σ )
(1) The critical manifold of the flow ψz1 are the cores CRλ ⊂ GrR ≃ GrR 2 |0 and the stable
manifold for CRλ is the strata SRλ ⊂ GrR .
(2) For each λ ∈ Λ+A , we denote by
(σ )
T̃Rλ = {γ ∈ GrR 2 |lim ψz1 (γ) ∈ CRλ }
z→0
Proof. Let γ ∈ Grx and let E = ux,C (γ) ∈ BunGR (P1 (R)) be the image of the complex
uniformization map. By Lemma 6.4(2) we have
(σ )
(6.13) |E| = |aE (0)| = |q(aγ (0)| = |uR 2 (lim ψz1 (γ))|,
z→0
(2)
[
(6.14) lim ψz1 (Grx ≃ GrR |x ) = CRλ .
z→0
λ∈Λ+
A
S (σ )
As the image λ∈Λ+ |q(CRλ )| of the critical manifolds under uR 2 is equal to | BunGR (P1 (R))|,
A
equations (6.13) and (6.14) imply that ux,C factors through ux,C : Gr → BunGR (P1 (R))
and induces a surjection between the sets of isomorphism classes of objects. Now a similar
argument as in the proof Proposition 6.2 shows that ux,C : Grx → BunGR (P1 (R)) is surjective
and hence an isomorphism LGx,R \Grx ≃ BunGR (P1 (R)).
7.2. Real forms of quasi-maps. Let n = 2m. The twisted conjugation σ2m on (P1 )2m
together with the involution η on G defines an involution on QM (2m) (P1 , X) and we denote
by QM (σ2m ) (P1 , X)R the corresponding semi-analytic stack. Note there are natural maps
QM (σ2m ) (P1 , X)R → (P1 )m QM (σ2m ) (P1 , X)R → BunG (P1 )R .
Since c(∞) = ∞, we have a real form of QM (σ2m ) (P1 , X, ∞)R of QM (n) (P1 , X, ∞). We
have natural maps
QM (σ2m ) (P1 , X, ∞)R → Cm QM (σ2m ) (P1 , X, ∞)R → BunG (P1 )R .
7.3. Morphisms. Let G1 and G2 be two reductive groups with complex conjugations η1
and η2 and Cartan involutions θ1 and θ2 respectively. Then the constructions of quasi-maps,
rigidified quasi-maps, uniformization morphisms, and real forms of those are functorial with
respect to homomorphism f : G1 → G2 that intertwine η1 , η2 and θ1 , θ2 .
7.5. Quillen’s homeomorphism. We fist recall some results in [CN1] on quasi-maps for
complex groups.
Consider the complex group G ≃ G\(G × G) viewed as a symmetric variety for G × G.
It corresponds to the swap involution sw(g1 , g2 ) = (g2 , g1 ) of G × G and the corresponding
conjugation on G × G is given by swc (g1 , g2 ) = (ηc (g2 ), ηc (g1 )). Consider the real quasi-maps
space QM (σ2m ) (P1 , G, ∞)R → Cm associated to G\(G × G).
The natural projections p1 , p2 : G × G → G induces natural isomorphism
(2m)
GrG×G ≃ Gr(2m) ×(P1 )2m Gr(2m) .
(2m)
Consider the base changes Gr(2m) |(P1 )m and GrG×G |(P1 )m along the embedding (P1 )m ֒→
(P1 )2m , (z1 , ..., zm ) → (z1 , ..., zm , z̄1 , ..., z̄m ).
We have a natural embedding
(7.5) Gr(m) → Gr(2m) |(P1 )m
(σ ) (2m)
sending (z ∈ (P1 )m , E, φ) → ((z, z̄) ∈ (P1 )2m , E, φ|P1\z∪z̄ ). Note that GrG×G,R
2m
⊂ GrG×G |(P1 )m
and it was proved in [?, Corollary 5.2 and Corollary 5.4] that the natural projection map
(2m) p1
GrG×G |(P1 )m ≃ Gr(2m) ×(P1 )2m Gr(2m) |(P1 )m −→ Gr(2m) |(P1 )m
restricts to an isomorphism
(σ )
(7.6) 2m
GrG×G,R ≃ Gr(2m) |(P1 )m
and the composed map
7.5 (7.6) (σ2m )
Gr(m) → Gr(2m) |(P1 )m ≃ GrG×G,R
induces a real analytic isomorphism
(σ )
(7.7) Gr(m) |Cm ≃ ΩG(σ
c
2m ) 2m
\GrG×G,R |Cm ≃ QM (σ2m ) (P1 , G, ∞)R .
Let S := Rm × iR and viewed as a subset of Cm via the embedding
(7.8) S −→ Cm , (b1 , ..., bm , ia) → (b1 + ia, ..., bm + ia)
We now consider the base change of the isomorphism (7.7) to S:
(7.9) v : Gr(m) |S ≃ QM (σ2m ) (P1 , G, ∞)R |S .
30
The conjugation η × η on G × G commutes with swc , together with the complex conjugation
on P1 , it defines an involution (η ×η)(σ2m ) on QM (σ2m ) (P1 , G, ∞)R |S and, via the isomorphism
vS , it induces an involution η (σ2m ) on Gr(m) |S .
The following description of η (σ2m ) is proved in [CN1, Proposition 5.12].
Proposition 7.3. The involution η (σ2m ) on Gr(m) |S is given by the following formula:
η (σ2m ) = η (m) on Gr(m) |Rm ×{0}
Denote by
(σ2m )
(7.10) Q(m) := (Gr(m) |S )η −→ S = Rm × iR
the fixed-points. Then the isomorphism (7.9) restricts to natural isomorphisms
∼ (m) (m)
(7.11) v0 : Q(m) |Rm ×{0} / (Gr(m) )η |Rm ×{0} = GrR |Rm ×{0}
∼ (m) (m)
va6=0 : Q(m) |Rm ×{ia} / (ΩGc )θ̃ |Rm ×{ia} ≃ ΩXc |Rm ×{ia}
∼ (1) ∼
Q(m),λp |Rp ×ia / P (m),λp |Rp ×{ia} (resp. Qλ |R×ia / Q(1),λ |R×{ia} . )
(m),λ (1),λ
As SR (resp. TR ) is non-empty if and only if λp : p → Λ+ + +
A ⊂ ΛT (resp. λ ∈ ΛA ), the
(1)
same applies to the stratum Q(m),λp (resp. Qλ ). We have the following:
Proposition 7.4. Equip Q(m) (resp. Q(1) ) and S = Rm ×iR with the stratifications {Q(m),pλ }
(1)
(resp. {Q(1),λ } or {Qλ }) and {Rp ×R}p∈p(m) respectively. There is a Kc -equivariant stratified
trivilization of the family
Q(m) → S = Rm × iR
over iR.
Now the homeomorphisms in (7.11) and (7.12) together with Proposition 5.6 and Propo-
sition 7.4 imply the following
31
Theorem 7.5. There are Kc -equivariant strata-preserving homeomorphisms
(m)
ΩKc(m) \Gr(m) |Rm ×{i} ≃ ΩXc(m) |Rm ×{i} ≃ GrR |Rm ×{0}
which restrict to Kc -equivariant homeomorphisms
(n),λp (m),λp
ΩKc(m) \OK ≃ P (m),λp ≃ SR
(1),λ (1) (1),λ (1),λ
ΩKc(1) \OK ≃ P (1),λ ≃ SR ΩKc(1) \OR ≃ Q(1),λ ≃ TR if m = 1
Proof. According to Example 7.2 and (7.11), there are natural identifications
(m) (m)
QM (σ2m ) (P1 , X, ∞)R,L ≃ GrR , Q(m) ≃ GrR over R × {0}
az
C / C
Lemma 7.7. We have the following properties of the flows:
(1) The flow ψz3 on QM (σ2 ) (P1 , X, ∞)R is Kc -equivariant.
(σ )
(2) Recall the flow ψz1 on GrR 2 (6.9). We have the following commutative diagram
(σ ) ψz1 (σ )
(7.17) GrR 2 |C / GrR 2 |C .
ψz3
(σ2 ) 1 (σ2 )
QM (P , X, ∞)R / QM (P1 , X, ∞)R
the corresponding unstable manifold. We have T̃Rλ |0 ≃ TRλ ⊂ GrR for λ ∈ Λ+ A . The
open embedding ΩKc \Gr → QM (σ2 ) (P1 , X, ∞)R |i restricts to an isomorphism
ΩKc \OλR ≃ T̃Rλ |i
for λ ∈ Λ+
A.
Proof. Part (1) and (2) follows from the construction of the flows. Part (3) and (4) follows
from Lemma 6.3 and diagram (7.17).
8. Real-symmetric equivalence
In this section we construct the real-symmetric equivalence between the real and relative
Satake categories. The theory of sheaves on infinite dimensional stacks developed in [BKV]
plays an important role here.
8.1. Placid stacks. We first review the notion of placid (ind)-schemes and (ind)-stacks.
Let Y be a scheme acted on by an affine group scheme H. we say that Y is H-placid if
• Y can be written as filtered limit Y = limj Yj , where each Yj is a H-scheme of finite
type and the transition maps Yj ′ → Yj for j → j ′ are affine, smooth, surjective, and
H-equivariant.
33
• the action H × Yj → Yj factors through a group scheme Hj of finite type. Moreover
the Hj can be chosen so that {Hj }j forms a projective system with H = limj Hj and
the transition maps Hj ′ → Hj are smooth surjective with unipotent kernel.
Let Z ⊂ Y a H-invariant subscheme. We shall say that the inclusion Z → Y is placid if
there is a presentation of Y = lim Yj as above and and index j and a H-invariant subscheme
Zj ⊂ Yj such that Z ∼ = Zj ×Yj Y .
Let Y be an ind-scheme acted on by an affine group scheme H. We say that Y is H-ind
placid if can be written as filtered colimit
(8.1) Y ∼
= colimi Y i
′
where each Y i is H-placid and the transition maps Y i → Y i are placid closed embeddings.
A presentation as in (8.1) is called a placid presentation of Y .
We call a stack Y a placid stack (resp. ind-placid stack) if it is isomorphic to a stack of
the form Y ≃ Y /H where Y is H-placid scheme (resp. H-ind placid scheme).
We recall the following basic result due to Drinfeld and Bouthier:
Proposition 8.1. (1) [D, Theorem 6.3] and [B, Proposition 2.0.1] For any smooth affine
C-scheme Y of finite type, the loop space Y (K) is ind-placid.
(2) [D, Proposition 3.8] and [B, Proposition 1.1.4] Let G be a complex connected reductive
group and let H ⊂ G be a connected reductive subgroup. The natural map G(K) →
(G/H)(K) is a H(K)-torsor in the h-topology. In particular, there is an isomorphism
of stacks
G(K)/H(K) ≃ (G/H)(K)
Remark 8.2. Part (2) above is equivalent to the following fact: Let A be a C-algebra. Any
G-torsor on Spec(A((t))) is trivial over Spec(B((t))) for a h-cover Spec(B) → Spec(A). This
fact is claimed in [D, Proposition 3.8, Remark (b)] with sketch of proof. In [B], the author
provided a different proof.
In [CY, Proposition 26], we deduce the following results from Proposition 8.1 (1).4
Proposition 8.3. (1) X(K) is G(O)-ind placid. (2) The orbits closure X(K)λ is G(O)-
placid. (3) The inclusion X(K)λ → X(K)λ is placid.
the general case follows from the fact that the functors in (8.3) commutes with colimits.
The proof of the base change isomorphism (ηS )! fS! → f ! η! is similar. The only non-trivial
part is the the Beck-Chevalley condition in (3) and it follows from the same argument above
using [BKV, Proposition 5.3.9].
Proposition 8.11. (1) The quotient LK (n) \Gr(n) is an ind-stack ind-locally of finite type
(2) the natural projection map f : LK (n) \Gr(n) → K(K)(n) \Gr(n) is strongly pro-smooth.
In particular, both stacks LK (n) \Gr(n) and K(K)(n) \Gr(n) admit gluing of sheaves.
Proof. Since the stack of quasi maps QM (n) (P1 , X) is an ind-stack ind-locally of finite type,
the uniformization isomorphism LK (n) \Gr(n) ≃ QM (n) (P1 , X) in (7.2) implies part (1). On
36
the other hand, the pullback of f along Gr(n → K(K)(n) \Gr(n) is isomorphic to the natural
projection map
Gr(n) ×(P1 )n K(K)(n) /LK (n) −→ Gr(n) .
Note that the quotient K(K)(n) /LK (n) is the K(O)(n) -torsor over (P1 )n ×BunK (P1 ) classifying
a K-bundle EK ∈ BunK (P1 ) on P1 , a point z ∈ (P1 )n , and a trivialization of EK on the formal
neighborhood of the points z in P1 . Since K(O)(n) is a projective limit of smooth schemes
with smooth affine transition maps (see, e.g., [R, Lemma 2.5.1])) and BunK (P1 ) is a smooth
stack locally of finite type it implies that f is strongly pro-smooth. The proposition follows.
8.3. Real and relative Satake categories. The loop group K(K), arc group G(O), and
real arc group GR (OR )) act natural on Gr, X(K), and GrR and we can form the quotient
stacks K(K)\Gr and X(K)/G(O) and the quotient semi-analytic stack GR (OR )\GrR (see Ap-
pendix A.2). Let D(K(K)\Gr), D(X(K)/G(O)), and D(GR (OR )\GrR )) be the dg-categories
of C-constructible complexes on K(K)\Gr, X(K)/G(O), and GR (OR )\GrR respectively (see
Appnedix A.3).
By Proposition 8.4, there is an isomorphism of stacks K(K)\Gr ≃ X(K)/G(O) and hence
a canonical equivalence
(8.4) D(K(K)\Gr) ≃ D(X(K)/G(O))
We will call D(X(K)/G(O)) the relative Satake category and D(GR (OR )\GrR ) the real
Satake category.
8.4. Perverse t-structures. Let D(GR \GrR ) be the dg category of constructible C-complexes
on GR \GrR and let DSR (GR \GrR ) to be the full subcategory of D(GR \Gr) of complexes con-
structible with respect to the GR (OR )-orbits stratification SR = {SRλ }λ∈Λ+ . We have a natural
A
equivalence
(8.5) DSR (GR \GrR ) ≃ D(GR (OR )\GrR )
Since GR is connected, it follows from [N2, Lemma 3.9.1] that the dimension dimR SRλ =
2hλ, ρi of the real spherical orbits SRλ are even numbers. Thus there is a classical perverse
t-structure (p D ≤0 (GR (OR )\GrR ), p D ≥0 (GR (OR )\GrR )) on D(GR (OR )\GrR ) ≃ DSR (GR \GrR )
dim Grλ
given by the middle perversity function pR : SR → Z, pR (SRλ ) = − R2 R = −hλ, ρi. We will
write
Perv(GR (OR )\GrR ) = pcl D ≤0 (GR (OR )\GrR ) ∩ pcl D ≥0 (GR (OR )\GrR )
for the heart of the perverse t-structure.
We shall introduce a t-structure on D(X(K)/G(O)) following [BKV, Section 6]. For any
finite type stack Y we denote by (pcl D ≤0 (Y), pcl D ≥0 (Y)) the classical perverse t-structure on
D(Y). According to [BKV, Section 6.3.2], for every placid stack Y ≃ Y /H there is a unique
!-adapted t-structure on (p D ≤0 (Y), p D ≥0 (Y)) on D(Y) characterized by
p
(8.6) D ≥0 (Y) ≃ colimi∈I pcl D ≥0 (Yi )[dim Yi ]
where Y ≃ limi∈I Yj = Yj /H
37
Following [BKV, Section 2.4.8], we say that an ind-stack Y is placid stratified if there is a
stratification {Sα }α∈I of Y such that (1) each stratum Sα is a placid stack and the inclusion
j α : Sα → Y is a fp-locally closed embedding (2) there is a presentation Y = colimj∈J Yj of Y
as a filtered colimit of closed substacks such that each Y is a finite union of the strata Sα .
Recall the following construction of t-structures functor for placid stratified stack, admit-
ting gluing of sheaves in [BKV, Proposition 6.4.2].
Proposition 8.12. Let (Y, {Sα }α∈I ) be a placid stratified stack, admitting gluing of sheaves,
and equipped with a perversity p : I → Z. There is a unique t-structure (p D ≤0 (Y), p D ≥0 (Y))
on D(Y) satisfying
p
(8.7) D ≥0 (Y) = {F ∈ D(Y)|(j α )! F ∈ p D ≥0 (Sα )[p(α)].}
The heart of the t-structure is denoted by
Pervp (Y) = p D ≤0 (Y) ∩ p D ≥0 (Y).
Note that every ind-placid stack Y is a placid stratified stack and [BKV, Lemma 5.5.6]
implies that it admits glueing of sheaves. Thus Proposition 8.3, Corollary 8.4 and Lemma 8.7
imply that K(K)\Gr ≃ X(K)/G(O) together with the orbits stratification {X(K)λ /G(O) ≃
K(K)\OλK } is a placid stratified stack, admitting glueing of sheaves. We are mainly interested
in the following perversity function
(8.8) pX : Λ +
A → Z, pX (λ) = −hλ, ρi
Proof. Consider the quotient K(K)/LKc of K(K) by the subgroup LKc and the natural
embedding Gr → Gr × K(K)/LKc sending γ to (γ, eLKc ). It induces an isomorphism of
stacks LKc \Gr ≃ K(K)\(Gr×K(K)/LKc ) (where K(K) acts diagonally on Gr×K(K)/LKc )
and we have
pr
q : LKc \Gr ≃ K(K)\(Gr × K(K)/LKc ) → K(K)\Gr
where pr is induced by the natural projection map Gr × K(K)/LKc → Gr. Thus it suffices
to show pr! : D(K(K)\Gr) → D(K(K)\(Gr × K(K)/LKc )) is fully-faithful. Note that we
have a commutative diagram
∼
=
D(K(K)\Gr) / lim[n] D(K(K)n × Gr)
pr! lim[n] (pr[n] )!
∼
=
D(K(K)\(Gr × K(K)/LKc )) / lim[n] D(K(K)n × (Gr × K(K)/LKc ))
where the horizontal equivalences come from the Čech complexes for the coverings Gr →
K(K)\Gr and Gr×K(K)/LKc → K(K)\(Gr×K(K)/LKc ), and the right vertical arrow are
induced by the pull-back functors along the projections pr[n] = idK(K)n × pr : K(K)n × (Gr ×
K(K)/LKc ) → K(K)n × Gr. Since the space K(K)/LKc ≃ K(O)/Kc ≃ K(O)+ × K/Kc is
contractible (here K(O)+ ⊂ K(O) is the first congruence subgroup) the functor (pr[n] )! is
fully-faithful and it follows that pr! is fully-faithful.
We show that the resulting functor q ! : D(K(K)\Gr) → DSK (LKc \Gr) is essentially
surjective. Since D(K(K)\Gr) admits gluing of sheaves, Lemma 8.6 implies that the category
D(K(K)\Gr) is generated by jK,∗ λ
(Lλ ), λ ∈ Λ+ λ
A , where Lλ ∈ D(K(K)\OK ). Consider the
following Cartesian diagram
iλ
+
LKc \OλK / LKc \Gr
qλ q
λ
jK
K(K)\OλK / K(K)\Gr
The desired claim follows from the facts that (1) DS (LKc \Gr) is generated by (iλ+ )∗ (qλ! )(Lλ )
under colimit and (2) the base change isomorphism for fp-locally closed embedding q ! (jK,∗
λ
(Lλ )) ≃
λ λ !
(i+ )∗ (q ) (Lλ ) (see, e.g., [BKV, Lemma 5.4.5]).
39
Theorem 8.14. There are natural equivalences
D(X(K)/G(O)) ≃ D(K(K)\Gr) ≃ D(GR (OR )\GrR )
which are t-exact with respect to the perverse t-structures.
Proof. Theorem 7.5 (the case when m = 1) implies that there is a Kc -equivariant stratified
homeomorphism ΩKc \Gr ≃ GrR which induces a natural equivalence
(8.9) DSK (LKc \Gr) ≃ DSR (GR \GrR ).
Now combining (8.4) and Lemma 8.13 we obtain the desired equivalences
(8.4) Lem (8.13) (8.9)
D(X(K)/G(O)) ≃ D(K(K)\Gr) ≃ DSK (LKc \Gr) ≃ DSR (GR \GrR ) ≃
(8.5)
≃ D(GR (OR )\GrR ).
To check that the equivalences are t-exact it suffices to check that it restricts to an equivalence
p ≥0
D (K(K)\Gr) ≃ p D ≥0 (GR (OR )\GrR ). Note that the commutative diagram
iλ
+
Kc \SRλ ≃ LKc \OλK / Kc \GrR ≃ LKc \Gr
qλ q
λ
jK
K(K)\OλK / K(K)\Gr
implies that there is a commutative square of functors
q!
D(K(K)\Gr) ≃
/ D(GR (OR )\GrR )
λ )! (iλ !
(jK +)
(q λ )!
D(K(K)\OλK ) / D(GR (OR )\SRλ )
In view of the characterization of
p
D ≥0 (K(K)\Gr) = {F ∈ D(K(K)\Gr)|(jK
λ !
) F ∈ p D ≥0 (K(K)\OλK )[pX (λ)]}
in (8.7), we need to check (q λ )! induces an equivalence
(q λ )! : p D ≥0 (K(K)\OλK )[pX (λ)] ≃ p D ≥0 (GR (OR )\SRλ ).
Let K(K)\OλK ≃ limi∈I Yi be a placid presentation with natural evaluation map evi :
K(K)\OλK → Yi . According to (8.6), we have
p
D ≥0 (K(K)\OλK ) ≃ colimi∈I p D ≤0 (Yi )[dim Yi ]
Since the fiber of the composition evi ◦ q λ : Kc \SRλ → K(K)\OλK → Yi are contractible and
dimR SRλ /2 = hλ, ρi = −pX (λ), we have
(evi ◦ q λ )! : p D ≤0 (Yi )[dim Yi ] → p D ≤0 (GR (OR )\SRλ )[dimR SRλ /2] ⊂ D ≤0 (Kc \SRλ )[dimR SRλ ].
and the resulting map
(q λ )! ≃ colimi∈I (evi ◦q λ )! : p D ≥0 (K(K)\OλK )[pX (λ)] ≃ colimi∈I p D ≥0 (Yi )[dim Yi −dimR SRλ /2] →
40
→ D ≤0 (Kc \SRλ )[dimR SRλ /2]
factors through an equivalence
(q λ )! : p D ≥0 (K(K)\OλK )[pX (λ)] ≃ p D ≥0 (GR (OR )\SRλ ) ⊂ D ≥0 (Kc \SRλ )[dimR SRλ /2]
The desired claim follows.
9.1. The functor Υ. Consider the dg category D(LGR \Gr) of C-constructible complexes
on the semi-analytic stacks LGR \Gr. Consider the following correspondence
q u
K(K)\Gr o LKc \Gr / LGR \Gr
Define
(9.1) Υ = u! q ! : D(K(K)\Gr) → D(LKc \Gr) → D(LGR \Gr)
Theorem 9.1 (Affine Matsuki correspondence for sheaves). The functor Υ defines an equiv-
alence of categories
∼
Υ : D(K(K)\Gr) −→ D(LGR \Gr).
9.2. Bijection between local systems. Write [OλK ] = LKc \OλK , [OλR ] = LKc \OλR , [Oλc ] =
LKc \Oλc , and [Eλ ] = LGR \OλR ∈ LGR \Gr. Recall the Matsuki flow φt : Gr → Gr in Theorem
3.9. As φt is LKc -equivariant, it descends to a flow φ̃t : LKc \Gr → LKc \Gr and we define
G
φ± : LKc \Gr → [Oλc ] ⊂ LKc \Gr, γ → lim φ̃t (γ).
t→±∞
λ∈Λ+
A
Here iλ± and j±λ are the natural embeddings and φλ± (resp. uλ ) is the restriction of φ± (resp.
u) along j+λ (resp. j−λ ).
Lemma 9.2. We have the following:
(1) There is a bijection between isomorphism classes of local systems τ + on [OλK ], local
systems τ − on [OλR ], local systems τ on [Oλc ], and local systems τR on [Eλ ] = [LGR \OλR ],
characterizing by the property that τ ± ≃ (φλ± )∗ τ and τ − ≃ (uλ )∗ τR .
41
(2) The map uλ factors as
φλ pλ
uλ : [OλR ] → [Oλc ] → [Eλ ]
−
(9.2)
where pλ is smooth of relative dimension dim[Eλ ] − dim[Oλc ]. Moreover, we have
(pλ )∗ τR ≃ τ .
Proof. Since the fibers of φ± are contractible, pull-back along φλ+ (resp. φλ− ) defines an
equivalence between LKc -equivariant local systems on Oλc and LKc -equivariant local systems
on OλK (resp. OλR ). We show that the fiber of uλ is contractible, hence pull back along uλ
defines an equivalence between local systems on [Eλ ] and LKc -equivariant local systems on
OλR . Pick y ∈ Oλc and let LKc (y), LGR (y) be the stabilizers of y in LKc and LGR respectively.
The group LKc (y) acts on the fiber ly := (φλ− )−1 (y) and we have OλR ≃ LKc ×LKc (y) ly .
Moreover, under the isomorphism [OλR ] ≃ LKc \OλR ≃ LKc (y)\ly , [Oλc ] ≃ LKc (y)\y, and
[Eλ ] ≃ LGR (y)\y, the map uλ takes the form
φλ pλ
uλ : [OλR ] ≃ LKc (y)\ly → [Oλc ] ≃ LKc (y)\y → [Eλ ] ≃ LGR (y)\y,
−
where the first map is induced by the projection ly → y and the second map is induced by
the inclusion LKc (y) → LGR (y). We claim that the quotient LKc (y)\LGR (y) is contractible,
hence uλ has contractible fibers and pλ is smooth of relative dimension dim[Eλ ] − dim[Oλc ].
Part (1) and (2) follows.
Proof of the claim. Pick y ′ ∈ CRλ ⊂ GrR and let Kc (y ′) and GR (R[t−1 ])(y ′) be the sta-
bilizers of y ′ in Kc and GR (R[t−1 ]) respectively. The composition of the complex and real
uniformizations of BunGR (P1 (R))
Prop. 6.5 Prop. 6.2
LGR \Gr ≃ BunGR (P1 (R)) ≃ GR (R[t−1 ])\GrR
identifies
LGR (y)\y ≃ [Eλ ] ≃ GR (R[t−1 ])(y ′)\y ′ .
Hence we obtain a natural isomorphism
LGR (y) ≃ Aut([Eλ ]) ≃ GR (R[t−1 ])(y ′)
sending LKc (y) = Kc (y) ⊂ LGR (y) to Kc (y ′) ⊂ GR (R[t−1 ])(y ′). Thus we reduce to show that
the quotient Kc (y ′)\GR (R[t−1 ])(y ′) is contractible. This follows from the fact that evaluation
map Kc (y ′ )\GR (R[t−1 ])(y ′ ) → Kc (y ′ )\GR (y ′ ), γ(t−1 ) → γ(0) has contractible fibers and the
quotient Kc (y ′)\GR (y ′ ) is contractible as Kc (y ′ ) is a maximal compact subgroup of the Levi
subgroup of GR (y ′ ).
Recall the flow
ψz3 : QM (σ2 ) (P1 , X, ∞)R → QM (σ2 ) (P1 , X, ∞)R
in §7.7. For λ ∈ Λ+ λ λ
A , we have the critical manifold CR , the stable manifold SR , and the
λ
unstable manifold T̃R . We write
s+ λ
λ : SR → QM
(σ2 )
(P1 , G, K, ∞)R, t̃λ : T̃Rλ → QM (σ2 ) (P1 , G, K, ∞)R
42
for the inclusion maps and we write
c+ λ λ
λ : SR → C R , d˜λ : T̃Rλ → CRλ
for the contraction maps. Note that all the maps above are Kc -equivalent with respect to
natural Kc -actions. Recall that, by Lemma 7.7, we have isomorphisms T̃Rλ |i ≃ ΩKc \OλR ,
T̃Rλ |0 ≃ TRλ , for λ ∈ Λ+
A and we write
s− λ
λ : TR → QM
(σ2 )
(P1 , G, K, ∞)R, tλ : ΩKc \OλR → QM (σ2 ) (P1 , G, K, ∞)R
for the restriction of t̃λ and
c− λ λ
λ : TR → CR , dλ : ΩKc \OλR → CRλ
for the restriction of the contractions d˜λ .
We write kλ : ΩKc \Oλc → CRλ for the restriction of dλ and pλ : TRλ → GR (R[t−1 ])\TRλ for
the natural quotient map.
Lemma 9.3. The map kλ : ΩKc \Oλc → CRλ is a Kc -equivariant isomorphism. There is
a bijection between isomorphism classes of Kc -equivariant local systems ω + on SRλ , Kc -
equivariant local systems ω − on TRλ , Kc -equivariant local systems ω on CRλ , Kc -equivariant
local systems τ on ΩKc \Oλc , and local system ωR on GR (R[t−1 ])\TRλ , characterizing by the
property that ω ± ≃ (c± ∗ ∗ ∗ − ∗
λ ) ω, τ ≃ (kλ ) ω , and (pλ ) ωR ≃ (cλ ) ω
Proof. The first claim follows from the fact that ΩKc \Oλc ≃ CRλ ≃ Kc (λ)\Kc , where Kc (λ)
is the stabilizer of λ in Kc , and the Kc -equivariant property of kλ . The second claim follows
from the facts that the contraction maps c± ±
λ are Kc -equivariant and the fibers of cλ and the
quotient Kc \GR (R[t−1 ]) are contractible.
Here τ + and τR are local system on [OλK ] and [Eλ ] corresponding to τ as in Lemma 9.2. Let
dλ := dim BunG (P1 )R − dim[OλK ].
Write
(9.5) ιλ+ : [Oµc ] → [OµK ], ιλ− : [Oµc ] → [OµR ]
for the natural embeddings. We recall the following fact, see [MUV, Lemma 5.4].
ιµ φµ
(1) Consider [Oµc ] → [OµK ] → [Oµc ]. Let F ∈ Dc ([OµK ]). If F is smooth (=
+ +
Lemma 9.4.
locally constant) on the trajectories of the flow φ̃t , then we have canonical isomor-
phisms (ιµ+ )! F ≃ (φµ+ )! F and (ιµ+ )∗ F ≃ (φµ+ )∗ F.
43
ιµ φµ
(2) Consider [Oµc ] → [OµR ] → [Oµc ] where ιµ− is the natural embedding. Let F ∈ Dc ([OµR ]).
− −
If F is smooth (= locally constant) on the trajectories of the flow φ̃t and is supported
on a finite dimensional substack Y ⊂ [OµR ], then we have canonical isomorphisms
(ιµ− )! F ≃ (φµ− )! F and (ιµ− )∗ F ≃ (φµ− )∗ F.
We shall show that the functor Υ sends standard sheaves to co-standard sheaves. Introduce
the following local system on [Oλc ]
(9.6) Lλ := (ιλ− )∗ L′λ ⊗ L′′λ ⊗ or∨pλ
where
(9.7) (L′µ )∨ := (iµ− )! (C)[codim[OµR ]] and L′′λ := (ιλ+ )! C[codim[OλK ] [Oλc ]]
are local systems on [OλR ] and [Oλc ] respectively and orpλ := (pλ )! C[− dim[Eλ ] + dim[Oλc ]] is
the orientation sheaf for the smooth map pλ : [Oλc ] → [Eλ ] in (9.2).
Lemma 9.5. For any local system τ on [Oλc ] we have
Υ(S∗+ (λ, τ )) ≃ S!− (λ, τ ⊗ Lλ )[dλ ].
Proof. Let λ, µ ∈ Λ+
A . Consider the following diagram
s uµ
(9.8) [OλK ∩ OµR ] / [OµR ] / [Eµ ] = LGR \OµR .
ι iµ
−
µ
j−
iλ
+ u
[OλK ] / LKc \Gr / LGR \Gr
Let G = (j−µ )∗ Υ(S∗ (λ, τ )) ≃ (j−µ )∗ u! (iλ+ )∗ (τ + ) ≃ (uµ )! (iµ− )∗ (iλ+ )∗ (τ + ). It suffices to show that
G ≃ 0 if λ 6= µ and G ≃ (τ ⊗ Lλ )R if λ = µ.
By Corollary 3.3, the orbits OµR and OλK are trasversal to each other, hence we have
(9.9) (iµ− )∗ (iλ+ )∗ (τ + ) ≃ (iµ− )! (iλ+ )∗ (τ + ) ⊗ L′µ [codim[OµR ]].
where
(9.10) (L′µ )∨ = (iµ− )! (C)[codim[OµR ]]
is a local system on [OµR ]. Thus
(9.9)
G ≃ (uµ )! (iµ− )∗ (iλ+ )∗ (τ + ) ≃ (uµ )! ((iµ− )! (iλ+ )∗ (τ + ) ⊗ L′µ )[codim[OµR ]] ≃
(uµ )! (s∗ ι! (τ + ) ⊗ L′µ )[codim[OµR ]].
According to Lemma 9.2 the map uµ factors as
φµ pµ
uµ : [OµR ] → [Oµc ] → [Eµ ]
−
where
(9.18) L′′′ λ !
λ = (u ◦ i+ ) C[dλ ]
10.1. A square of equivalences. Recall the quasi-map family QM (σ2 ) (P1 , X)R → P1 in
Section 7.2. Consider the base change QM (σ2 ) (P1 , X)R |iR≥0 along the natural inclusion
iR≥0 → P1 . By Proposition 6.6 and Proposition 7.1, we have the following cartesian di-
agram
j i
(LKc \Gr) × iR>0 / QM (σ2 ) (P1 , X)R |iR≥0 o Kc \GrR .
f0 f f0
j̄
ī
LGR \Gr × iR>0 / BunGR (P1 (R)) × iR≥0 o GR (R[t−1 ])\GrR
iR>0 / iR≥0 o {0}
Define the following nearby cycles functors
(10.1) Ψ : D(K(K)\Gr) ≃ DS (LKc \Gr) → D(Kc \GrR ), F → Ψ(F) := i∗ j∗ (F ⊠ CiR>0 ),
(10.2) ΨR : D(LGR \Gr) → D(GR (R[t−1 ])\GrR ), F → ΨR (F) = (ī)∗ (j̄)∗ (F ⊠ CiR>0 ).
We also have the Radon transform
(10.3) ΥR : D(GR (OR )\GrR ) → D(GR (R[t−1 ])\GrR )
given by the restriction to D(GR (OR )\GrR ) ⊂ D(GR \GrR ) of the push-forward p! : D(GR \GrR ) →
D(GR (R[t−1 ])\GrR ) along the quotient map p : GR \Gr → GR (R[t−1 ])\GrR . Denote by
D(BunGR (P1 (R))) be the dg category of C-constructible complexes on BunGR (P1 (R)).
Here are the main results of this section.
48
Theorem 10.1. The nearby cycles functors and the Radon transform induce equivalences
of categories:
∼
Ψ : D(K(K)\Gr) −→ D(GR (OR )\GrR ),
∼
ΨR : D(LGR \Gr) −→ D(GR (R[t−1 ])\GrR ),
∼
ΥR : D(GR (OR )\GrR ) −→ D(GR (R[t−1 ])\GrR ).
Theorem 10.2. We have a commutative square of equivalences
Ψ
D(K(K)\Gr) / D(GR (OR )\GrR )
Υ ΥR
ΨR
D(LGR \Gr) / D(GR (R[t−1 ])\GrR )
❘❘❘ ❥❥❥
❘❘❘
❘❘❘
❘ ❥❥❥❥❥❥❥
≃ ❘❘❘❘
t ❥❥❥ ≃
❥
) ❥
D(BunGR (P1 (R)))
where the vertical equivalences in the lower triangle come from the real and complex uni-
formization isomorphism
≃ ≃
LGR \Gr / BunGR (P1 (R)) o GR (R[t−1 ])\GrR
The rest of the section is devoted to the proof of Theorem 10.1 and Theorem 10.2.
10.2. The nearby cycles functor Ψ. For any λ ∈ L and a Kc -equivaraint local system ω
on CRλ , one has the standard and co-standard sheaves
T∗+ (λ, ω) := (s+ + + + +
λ )∗ (ω ) and T! (λ, ω) := (sλ )! (ω )
in D(GR (OR )\GrR ) (see Lemma 9.3). Recall the standard sheaf S+
∗ (λ, τ ) in D(K(K)\Gr)
(see (9.3)).
Proposition 10.3. The nearby cycles functor induces an equivalence Ψ : D(K(K)\Gr) ≃
D(GR (OR )\GrR ), which is the real-symmetric equivalence in Theorem 8.14. Moreover, we
have Ψ(S+ +
∗ (λ, τ )) ≃ T∗ (λ, ω)
10.5. Proof of Theorem 10.1 and Theorem 10.2. We already proved Theorem 10.1.
We have Ψ(F) ≃ i∗ j∗ (F ⊠ CiR>0 ) ≃ i! j! (F ⊠ CiR>0 [1]) and the natural arrow (f0 )! i! → (ī)! f!
gives rise to a natural transformation
(10.4)
ΥR ◦ Ψ(F) ≃ (f0 )! i! j! (F ⊠ CiR>0 [1]) → (ī)! f! j! (F ⊠ CiR>0 [1]) ≃ (ī)! (j̄)! (f 0 )! (F ⊠ CiR>0 [1]) ≃
≃ ΨR ◦ Υ(F).
Moreover, it follows from Lemma 9.5, Proposition 10.3, Proposition 10.4, and Proposition
10.5 that (10.4) is an isomorphism for the standard sheaf S+ ∗ (λ, τ ). Since the category
+
Dc (K(K)\Gr) is generated by S∗ (λ, τ ), it implies (10.4) is an isomorphism. Theorem 10.2
follows.
u
LKc \Gr / LGR \Gr
and its natural iterations.
Similarly, the Radon equivalence
∼
ΥR : D(GR (OR )\GrR ) −→ D(GR (R[t−1 ])\GrR )
is naturally an equivalence of Dc (GR (OR )\GrR )-modules by convolution on the right. To see
this, recall ΥR is the restriction to D(GR (OR )\GrR ) ⊂ D(GR \GrR ) of the push-forward p! :
D(GR \GrR ) → D(GR (R[t−1 ])\GrR ) along the quotient map p : GR \Gr → GR (R[t−1 ])\GrR .
We can equip this construction with compatibility with convolution on the right by using
the commutative action diagram
p×id
GR \G(KR ) ×G(OR ) GrR / GR (R[t−1 ])\G(KR ) ×G(OR ) GrR
p
−1
GR \GrR / GR (R[t ])\GrR
and its natural iterations.
11.2. From complex to real kernels. Following [N1], nearby cycles in the real Beilinson-
(σ )
Drinfeld Grassmannian GrR 2 over iR≥0 gives a functor
(11.1) ψ : D(G(O)\Gr) −→ D(GR (OR )\GrR )
Namely, there is a canonical diagram of GR -equivariant maps
π (σ ) j (σ ) i ? _ Gr(σ2 ) |0
(11.2) Gr o Gr × iR>0 ≃ GrR 2 |iR>0 / GrR 2 |iR≥0 o R ≃ GrR
(2)
where we view GR ⊂ LGR as the constant group-scheme. One defines ψ = i∗ j∗ π ∗ fR where
we write fR : D(G(O)\Gr) → D(GR \Gr) for the forgetful functor.
Note the domain and codomain of ψ both have natural convolution monoidal structures.
To equip ψ with a monoidal structure, we proceed as follows.
Let Gr(2) ×Gr
˜ (2) be the moduli of x1 , x2 ∈ P1 , E1 , E2 G-torsors on P1 , φ a trivialization of E1
(σ )
over P1 \ {x1 , x2 }, and α an isomorphism from E1 to E2 over P1 \ {x1 , x2 }. Let GrR 2 ×Gr ˜ (σ
R
2)
51
be the real form of Gr(2) ×Gr
˜ (2) with respect to the twisted conjugation that exchanges x1
and x2 .
Then there is a canonical diagram of GR -equivariant maps
π (σ ) j
˜ (σ2)
(11.3) G(K) ×G(O) Gr o G(K) ×G(O) Gr × iR>0 ≃ GrR 2 ×Gr R |iR>0
/
(σ ) i
˜ (σ
GrR 2 ×Gr 2)
R |iR≥0
o ? _ Gr(σ2 ) ×Gr
R
˜ (σ2 ) |0 R ≃ GR (KR ) ×GR (OR ) GrR
Moreover, the convolution maps on the end terms naturally extend to the entire diagram.
By standard identities, we arrive at a canonical isomorphism ψ(F1 ⋆ F2 ) ≃ ψ(F1 ) ⋆ ψ(F2 ).
By using iterated versions of the above moduli spaces, we may likewise equip ψ with the
associativity constraints of a monoidal structure.
11.3. Compatibility of actions. Note we can view the Radon equivalence ΥR as an equiv-
alence of D(G(O)\Gr)-modules via the monoidal functor
ψ : D(G(O)\Gr) −→ D(GR (OR )\GrR )
Now we have the following further compatibility of our constructions.
Theorem 11.1. Via the monoidal functor
ψ : D(G(O)\Gr) −→ D(GR (OR )\GrR )
the equivalences
∼
Ψ : D(K(K)\Gr) −→ D(GR (OR )\GrR ),
∼
ΨR : D(LGR \Gr) −→ D(GR (R[t−1 ])\GrR )
of Theorem 10.1 and commutative square
Ψ
D(K(K)\Gr) / D(GR (OR )\GrR )
Υ ΥR
ΨR
D(LGR \Gr) / D(GR (R[t−1 ])\GrR ).
of Theorem 10.2 are naturally of D(G(O)\Gr)-modules.
Proof. We will focus on the compatibility for the top row and indicate the moduli spaces
needed. We leave it to the reader to pass to sheaves and apply standard identities. The
compatibility for the bottom row and entire square can be argued similarly.
˜ (2) be the moduli of x1 , x2 ∈ P1 , E1 , E2 G-torsors on P1 , σ a section
Let QM (2) (P1 , X)×Gr
of E1 ×G X over P1 \ {x1 , x2 }, and α an isomorphism from E1 to E2 over P1 \ {x1 , x2 }. Let
˜ (σ
QM (σ2 ) (P1 , X)R ×Gr R
2)
˜ (2) with respect to the twisted
be the real form of QM (2) (P1 , X)×Gr
conjugation that exchanges x1 and x2 .
Then there is a canonical diagram of Kc -equivariant maps
(11.4)
π (σ ) j
˜ R 2 |iR>0
LKc \G(K) ×G(O) Gr o LKc \G(K) ×G(O) Gr × iR>0 ≃ QM (σ2 ) (P1 , X)R ×Gr /
52
i
˜ (σ
QM (σ2 ) (P1 , X)R ×Gr 2)
R |iR≥0
o ˜ (σ2 ) |0
? _ QM (σ2 ) (P1 , X)R ×Gr
R ≃ Kc \GR (KR ) ×GR (OR ) GrR
Note we could equivalently obtain diagram (11.4) by taking diagram (11.3) and quotienting
(σ )
by the left action of the group-scheme LKR 2 .
As with the convolution maps in diagram (11.3), the actions maps on the end terms of
diagram (11.4) naturally extend to the entire diagram. By standard identities, we arrive
at a canonical isomorphism Ψ(M ⋆ F) ≃ Ψ(M) ⋆ ψ(F). By using iterated versions of the
above moduli spaces, we may likewise equip Ψ with the associativity constraints of a module
map.
12.1. Fusion product for GrR . We first define fusion product for the real Satake category.
Consider the family
(2)
GrR |R≥0 −→ R≥0
(2)
obtained by the restriction of the family GrR → R2 along embedding R≥0 ֒→ R2 sending
(2) (2)
t → (t, −t). The natural action of G(O)R on GrR is compatible with the factorization
isomorphisms in Section 4.3 and it follows that there are Cartesian diagrams
jR (2) (2) iR
(GR (OR )\GrR )2 × R>0 / (G(O)R \GrR )|R≥0 o GR (OR )\GrR
R>0 / R≥0 o {0}
For any F, F ′ ∈ D(GR (OR )\GrR ), we define the fusion product F ⋆R F ′ as the following nearby
cycles:
F ⋆f F ′ := (iR )∗ (jR )∗ (F ⊠ F ′ ⊠ CR>0 )
where CR>0 is the constant sheaf on R>0 . Like in the case of complex reductive groups,
there is a natural monoidal structure on D(GR (OR )\GrR ) given by the convolution product:
consider the convolution diagram
p q m
(12.1) ˜ R := GR (KR ) ×GR (OR ) GrR → GrR
GrR × GrR ← GR (KR ) × GrR → GrR ×Gr
where p and q are the natural quotient maps and m(x, y mod GR (OR )) = xy mod GR (OR ).
For any F1 , F2 ∈ D(GR (OR )\GrR ), the convolution is defined as
˜ 2)
F1 ⋆ F2 = m! (F1 ⊠F
˜ 2 ∈ D(GR (OR )\GrR ) is the unique complex such that q ∗ (F1 ⊠F
where F1 ⊠F ˜ 2 ) ≃ p∗ (F1 ⊠ F2 ).
C× / Co {0}
where s is the natural closed embedding.
Since j is a fp-open embedding and i is an fp-closed embedding, [BKV, Lemma 5.4.1] and
Proposition 8.11 implies that the functors j ! and i! admits a (continuous) right adjoint j∗
and a left adjoint i∗ respectively.
For any F, F ′ ∈ D(X(K)/G(O)) (resp. F, F ′ ∈ D(K(K)\Gr)) we define the fusion product
F ⋆f F ′ as the following nearby cycles
F ⋆f F ′ := i∗ j∗ (F ⊠ F ′ ⊠ CR>0 )
here we view CR>0 as constant sheaf supported on R>0 ⊂ C× .
Note that the left adjoint (iR )∗ of (iR )∗ and right adjoint (jR )∗ of (jR )! exist and are
isomorphic to (iR )∗ ≃ i∗ s∗ and (jR )∗ ≃ s! j∗ (s|C× )∗ . Indeed, we have
Hom(i∗ s∗ F, F ′) ≃ Hom(s∗ F, i∗ F ′ ) ≃ Hom(s∗ F, s∗(iR )∗ F ′ ) ≃ Hom(F, (iR )∗ F ′ )
Hom(M, s! j∗ (s|C× )∗ M′ ) ≃ Hom(s∗ M, j∗ (s|C× )∗ M′ ) ≃ Hom(j ! s∗ M, (s|C× )∗ M′ ) ≃
≃ Hom((s|C× )∗ (jR )! M, (s|C× )∗ M′ ) ≃ Hom((jR )! M, M′ )
where the last isomorphisms follow from the fact that the functors s∗ and (s|C× )∗ are fully-
faithful. There is an isomorphism
F ⋆f F ′ ≃ (iR )∗ (jR )∗ (F ⊠ F ′ ⊠ CR>0 )
Theorem 12.2.
(1) The equivalences D(X(K)/G(O)) ≃ D(K(K)\Gr) ≃ D(GR (OR )\GrR ) are compatible
with the fusion products
54
(2) The fusion product for D(X(K)/G(O)) (resp. D(K(K)\Gr)) is t-exact with respect
to the perverse t-structure.
Proof. Proof of (1). The compatibility for D(X(K)/G(O)) ≃ D(K(K)\Gr) follows from the
definition. Consider the family of embeddings
ιa : C → (P1 )2 t → (t + ai, −t + ai)
parametrized by a ∈ R. There is a natural isomorphism
(12.2) (K(K)(2) \Gr(2) )|ιa (C) ≃ (K(K)(2) \Gr(2) )|ι0 (C) = (K(K)(2) \Gr(2) )|C
induced by the translation isomorphism P1 ≃ P1 , x → x + ai and the stratified homeomor-
phism in Theorem 7.5 restricts to a stratified homeomorphism
(2)
(12.3) GrR |R≥0 ≃ (ΩKc(2) \Gr(2) )|ι1 (R≥0 )
Let us consider the following Cartesian diagram
jR (2) iR
(GrR )2 × R>0 / GrR |R≥0 o GrR
≃ (12.3) ≃ ≃
(2) jR (2) iR
(ΩKc \Gr(2) )|R>0 / (ΩKc \Gr(2) )|ι1 (R≥0 ) o ΩKc \Gr
h(2) |R>0 h(2) h
jR iR
(K(K)\Gr)2 × R>0 / (K(K)(2) \Gr(2) )|R≥0 o K(K)\Gr
R>0 / R≥0 o {0}
where
pr (12.2)
h(2) : (ΩKc(2) \Gr(2) )|ι1 (R≥0 ) −→ (K(K)(2) \Gr(2) )|ι1 (R≥0 ) ≃ (K(K)(2) \Gr(2) )|R≥0
and
pr : (ΩKc(2) \Gr(2) ) → (K(K)(2) \Gr(2) )
(2)
is the natural map induced by the natural inclusion ΩKc → K(K)(2) . Note that there is
an isomorphism
(ΩKc(2) \Gr(2) ) ≃ K(K)(2) \(Gr(2) ×(P1 )2 K(K)(2) /ΩKc(2) )
where K(K)(2) -acts diagonally, and the functor
pr! ≃ (h(2) )! : D(K(K)(2) \Gr(2) |R≥0 ) ≃ colim D((Gr(2) )[n] |R≥0 ) −→ D((ΩKc(2) \Gr(2) )|R≥0 ) ≃
≃ colim D((Gr(2) ×(P1 )2 K(K)(2) /ΩKc(2) )[n] )
is induced by the pullback functors (pr[n] )! along the projection maps between the terms of
the Čech complexes:
pr[n] : (Gr(2) ×(P1 )2 K(K)(2) /ΩKc(2) )[n] = (Gr(2) ×(P1 )2 (K(K)(2) /ΩKc(2) )) ×(P1 )2 (K(K)(2) )n →
55
→ (Gr(2) )[n] ≃ Gr(2) ×(P1 )2 (K(K)(2) )n
(2)
Note that the Gram-Schmidt factorization in Proposition 5.4 implies that quotient K(K)(2) /ΩKc ≃
K(O)(2) is strongly pro-smooth, that is, K(O)(2) is a projective limit of smooth schemes with
smooth affine transition maps. Thus the map pr[n] is also strongly pro-smooth and it follows
from [BKV, Proposition 5.2.7 (d) and Lemma 5.4.5] that (pr[n] )! and hence pr! ≃ (h(2) )!
commutes with nearby cycles, that is, we have
h! (iR )∗ (jR )∗ (−) ≃ (iR )∗ (jR )∗ (h(2) |R× )! (−).
Since h! : D(K(K)\Gr) → D(ΩKc \Gr) ≃ D(GrR ) gives rise to the equivalence D(K(K)\Gr) ≃
D(GR (OR )\GrR ) in Theorem 8.14, for any F, F ′ ∈ D(K(K)\Gr), there is a natural isomor-
phism
h! (F ⋆f F ′) ≃ h! ((iR )∗ (jR )∗ (F ⊠ F ′ ⊠ CR>0 )) ≃ (iR )∗ (jR )∗ (h(2) |R× )! (F ⊠ F ′ ⊠ CR>0 ) ≃
13. Applications
We provide numerous applications of the main results to real and relative Langlands
duality.
Proof. It was shown in [N2, Theorem 1.2.3] that the nearby cycles functor ψ : D(G(O)\Gr) →
D(GR (OR )\GrR ) is t-exact if and only if GR is quasi-split. Since the Hecke action on the real
Satake category is given by the convolution F ⋆ ψ(F), the t-exactness for the convolution in
Lemma 12.1 implies the desired claim for the case D(GR (OR )\GrR ). We deduce the case of
D(X(K)/G(O)) from Theorem 8.14 and Theorem 11.1.
Remark 13.2. It would be nice is one can find a direct geometric proof for the t-exactness
criterion for the relative Satake category.
Theorem 13.3. The Hecke action on the real and relative Satake categories satisfies the con-
clusion of decomposition theorem. That is, for any semi-simple complexes M ∈ D(G(O)\Gr)
and F ∈ D(GR (OR )\GrR ) or D(K(K)\Gr) or D(X(K)/G(O)), the convolutions F ⋆ M is
again semi-simple.
In particular, the nearby cycles functor ψ : D(G(O)\Gr) → D(GR (OR )\GrR ) preserves
semi-simplicity.
56
Proof. By the real-symmetric equivalence Theorem 8.14 and Theorem 11.1, it suffices to
prove the case of relative Satake categories. Pick a placid presentation X(K) = colimi∈I limj∈J Yji .
We can assume the twisted product F ⊠M ˜ ∈ D(X(K) ×G(O) Gr) is supported on Y i ×G(O)
G(O)tλ G(O)/G(O) for some i ∈ I and λ ∈ Λ+ ′
T . One can find an (large enough) index i ∈ I
′
such that the image of the action map a : Y i ×G(O) G(O)tλ G(O) → X(K) lands in Y i . Note
that Y i ×G(O) G(O)tλ G(O) is G(O)-placid and the induced map
′
a : Y i ×G(O) G(O)tλ G(O) → Y i .
is a proper morphism. Thus by [EGA IV, Theorem 8.10.5], one can find a placid presentation
′
Y i ×G(O) G(O)tλ G(O) ≃ limj∈J Zj , an index j ∈ J, and a proper morphism ā : Zj → Yji
such that there is a Cartesian diagram
a ′
Y i ×G(O) G(O)tλ G(O)/G(O) / Y i /G(O)
h p
ā ′
Zj /G(O) / Yji /G(O)
Moreover, we can assume F ⊠M˜ ≃ h! F ′ for some semisimple object F ′ ∈ D(Zj /G(O)). Since
ā is proper, the Decomposition Theorem implies ā∗ (F) is semi-simple and the proper base
change theorem implies that
F ⋆ M ≃ a∗ (F ⊠ M) ≃ a∗ h! (F ′ ) ≃ p! (ā)∗ (F ′ )
is semi-simple.
13.2. Formality and commutativity of dg Ext algebras. We have the monoidal abelian
Satake equivalence
Rep(G∨ ) ≃ Perv(G(O)\GrG ) : V → ICV .
By restricting the Hecke action to the subcagegory Rep(G∨ ) ≃ Perv(G(O)\GrG ), we obtain
a monoidal action of Rep(G∨ ) on D(X(K)/G(O)) and D(GR (OR )\GrR ). Let ωX(O)/G(O) ∈
Perv(X(K)/G(O)) be the dualizing complex on the closed G(O)-orbit X(K)0 ≃ X(O) and
δR ∈ Perv(GR (OR )\GrR ) be the IC-complex of the closed GR (OR )-orbit SR0 .
Let ICreg := ICO(G∨ ) (an ind-object in Perv(G(O)\GrG )) be the image of the regular
representation O(G∨ ) under the abelian Satake equivalence. Since O(G∨ ) is a ring object
in Rep(G∨ ), ICreg is naturally a ring object in D(G(O)\Gr), that is, there is a natural
homomorphism
(13.1) m : ICreg ⋆ ICreg → ICreg
satisfying the unit and associativity properties. It follows that the RHom spaces
AX := RHomD(X(K)/G(O)) (ωX(O)/G(O) , ωX(O))/G(O) ⋆ ICreg )
AR := RHomD(GR (OR )\GrR ) (δR , δR ⋆ ICreg )
are naturally dg-algebras with natural G∨ -actions, known as the the de-equivariantized Ext
algebras for the symmetric and real Satake categories respectively. We denote by H • (AX ) =
57
Ext•D(X(K)/G(O)) (ωX(O)/G(O) , ωX(O))/G(O) ⋆ ICreg )) and H • (AR ) = Ext•D(GR (OR )\GrR ) (δR , δR ⋆ ICreg )
the corresponding cohomology algebras with trivial differentials.
Theorem 13.4. (1) There is a G∨ -equivariant isomorphism of dg algebras AX ≃ AR induc-
ing a G∨ -equivariant isomorphism of algebras H • (AX ) ≃ H • (AR ).
(2) The dg algebra AX (resp. AR ) is formal, that is, they are quasi-isomorphic to the
cohomology algebras H • (AX ) (resp. H • (AR )) with trivial differential.
(3) The algebra H • (AX ) (resp. H • (AR )) is commutative.
Proof. Part (1) follows from Theorem 8.14 and Theorem 11.1.
For part (2) and (3), using (1), it suffices to prove that AX is formal and H • (AR ) is
commutative. The formality of AX is proved in [CY] using a pointwise purity result for
IC-complexes of G(O)-obits in X(K). The proof of the commutativity of H • (AR ) is similar
to the case of complex groups. Note that δR ⋆ ICreg ≃ ψ(ICreg ) where ψ is the monoidal
functor in (11.1). Recall that ICreg is a commutative ring object in D(G(O)\Gr), that is,
there is a natural isomorphism
m ◦ σ ≃ m : ICreg ⋆ ICreg → ICreg
where
(13.2) σ : ICreg ⋆ ICreg ≃ ICreg ⋆ ICreg
is the commutativity constraint of the convolution product. The monoidal structure of ψ
gives rise to a multiplication morphism of ψ(ICreg ):
ψ(m)
mR : ψ(ICreg ) ⋆ ψ(ICreg ) ≃ ψ(ICreg ⋆ ICreg ) −→ ψ(ICreg )
satisfying the unit and associativity properties and there is a natural isomorphism
(13.3) mR ◦ σR ≃ mR : ψ(ICreg ) ⋆ ψ(ICreg ) → ψ(ICreg )
where
ψ(σ)
σR : ψ(ICreg ) ⋆ ψ(ICreg ) ≃ ψ(ICreg ⋆ ICreg ) ≃ ψ(ICreg ⋆ ICreg ) ≃ ψ(ICreg ) ⋆ ψ(ICreg ).
Note that the multiplication of
H •(AR ) = Ext•D(GR (OR )\GrR ) (δR , δR ⋆ ICreg ) ≃ Ext•D(GR (OR )\GrR ) (δR , ψ(ICreg ))
is induced from the multiplication morphism mR of ψ(ICreg ): for x : δR → ψ(ICreg )[i] ∈
ExtiD(GR (OR )\GrR ) (δR , ψ(ICreg )), y : δR → ψ(ICreg )[j] ∈ ExtjD(GR (OR )\GrR ) (δR , ψ(ICreg )) there
exists an unique map
˜ : δR ⊠δ
x⊠y ˜ R → ψ(ICreg )[i]⊠ψ(IC ˜ reg )[j] ∈ Ext
i+j ˜ R , ψ(ICreg )⊠ψ(IC
(δR ⊠δ ˜ reg ))
˜ R)
D(GR (OR )\GrR ×Gr
such that
˜ = p∗ (x ⊠ y) ∈ Exti+j
q ∗ (x⊠y) D(GR (OR )\GR (KR )×GrR ) (δR ⊠ δR , ψ(ICreg ) ⊠ ψ(ICreg ))
(where p and q are the maps in the convolution diagram (12.1)) and the product x ⋆ y ∈
Exti+j
D(GR (OR )\GrR ) (δR , ψ(ICreg )) is given by
Theorem 13.4 implies the following spectral descriptions of the full subcategories Dc (X(K)/G(O))0 ⊂
D(X(K)/G(O)) and Dc (GR (OR )\GrR )0 ⊂ D(GR (OR )\GrR ) in Section 1.2.4. Consider the
Hamiltonian duals MX∨ = Spec(H • (AX )) and MR∨ = Spec(H • (AR )) of X and GR .
Theorem 13.5. (1) There is a G∨ -equivariant isomorphism MX∨ ≃ MR∨ . (2)There are
equivalences of categories
(13.4) Dc (X(K)/G(O))0 ≃ Coh(MX∨ /G∨ ) Dc (GR (OR )\GrR )0 ≃ Coh(MR∨ /G∨ ).
Proof. Part (1) follows from Theorem 13.4 and part (2) follows from the formality of dg Ext
algebras AX and AR and the general Barr-Beck-Lurie theorem, see the details in [CMNO,
Theorem 5.5].
13.3. Identification of dual groups. In this section we show that there is an isomorphism
∨
between the dual group Hreal ⊂ G∨ of GR introduced in [N2] and the dual group Hsph∨
⊂ G∨
of X introduced in [GN1].
∨
13.3.1. Construction of Hreal .
Definition 13.6. Let QR ⊂ Perv(GR (OR )\GrR ) be be the full subcategory whose objects
are isomorphic to direct sum of perverse sheaves that appears in the summand of δR ⋆ ICV
for some V ∈ Rep(G∨ ). Let QX ⊂ Perv(X(K)/G(O)) be be the full subcategory whose
objects are isomorphic to direct sum of perverse sheaves that appears in the summand of
ωX(O)/G(O) ⋆ICV for some V ∈ Rep(G∨ ). Let QK ⊂ Perv(K(K)\Gr) be be the full subcategory
whose objects are isomorphic to direct sum of perverse sheaves that appears in the summand
of ωK(K)\O0K ⋆ ICV for some V ∈ Rep(G∨ ).
where Q = QR , QX or QK .
Proposition 13.7. (1) The abelian category QR is semi-simple and irreducible objects are
intersection cohomology sheaves on the orbits closures of strata SRλ , λ ∈ σ(ΛT ) (see (2.2)),
with coefficients in trivial local systems. There exists a unique associativity and commutativ-
ity constraints for the category QR equipped with the convolution product ⋆ such that (QR , ⋆)
is a neutral Tannakian category with fiber functor H∗ : QR → Vect given by cohomology. The
∨
Tannakian group of QR is isomorphic to the connected complex reductive subgroup Hreal ⊂ G∨
of the dual group associated to GR in [N2] and there is a horizontal tensor equivalence in the
59
following diagram of tensor functors
Rep(G∨ )
✈ ◆◆◆
✈✈ ◆◆◆
✈✈✈ ◆◆◆
✈ ◆◆&
z✈
✈ ≃
QR / Rep(H ∨
real )
(13.6) Rep(G∨ ) .
✉ ■■
✉✉ ■■
✉✉✉ ■■
■■
✉
z✉ ≃
✉
≃
■$
QX / QK / QR
where vertical arrows are given by the perverse Heck-actions ⋆p of Rep G∨ (13.5) on ωX(O)/G(O) ,
ωK(K)\O0K , and δR respectively.
Proof. The semi-simplicity of the Hecke action in Theorem 13.3 implies that the category
QR is the same as the full subcategory Q(GrR ) ⊂ Perv(GR (OR )\GrR ) introduced [N2] whose
objects are isomorphic to subquotients of perverse sheaves that appears in the summand of
δR ⋆ ICV for some V ∈ Rep(G∨ ). Now part (1) follows from the main results in [N2].
Proposition 12.2 implies that there are equivalence QR ≃ QX ≃ QK compatible with the
Hecke action, and the convolution product on QR and the fusion products on QX and QK .
Thus we can transport the associativity and commutativity constraints of (QR , ⋆) through
the above equivalence to (QX , ⋆f ) and (QK , ⋆f ). Now the commutativity of (13.6) follows
again from part (1) and Theorem 11.1.
∨
Corollary 13.8. There are tensor equivalences QX ≃ QK ≃ Rep(Hreal ). The abelian cate-
gory QX (resp. QK ) is semi-simple with irreducible objects ICX (resp. ICλK ), λ ∈ σ(ΛT ).
λ
∨
13.3.2. Construction of Hsph . Fix a pole point 0 ∈ P1 and consider the stack of quasi-maps
Z := QM (1) (P1 , 0, X) classifying a G-bundle E on P1 and a section φ : P1 \ {0} → E ×G X, or
equivalently, a K-reduction EK of E on P1 \ {0}. We have the uniformization isomorphism
(13.7) Z ≃ LK (1) \Gr(1) |0 ≃ K(C[t−1 ])\Gr
in Section (7.4), here t is the local coordinate at 0. The stratum OλK ⊂ Gr, λ ∈ Λ+
S in Section
4.2 decends to a stratum Z λ = K(C[t−1 ])\OλK ⊂ Z, which is smooth, locally closed sub-stack
of Z and the collection Z = {Z λ } forms a stratification of Z (see [GN1, Section 3.4]). For
example, the closed stratum Z 0 is isomorphic to
Z 0 ≃ K(C[t−1 ])\K(K)/K(O) ≃ BunK (P1 ).
60
Following [GN1], we define the ind-stack HeckeZ of generic Hecke modifications to the
ind-stack classifying data
(E1 , E2 , φ1 , φ2, z, τ )
where (Ei , φi ) ∈ Z, z ∈ Sym (P ) is a divisor on P1 with support contained in P1 \ {0}, and
1
˜
We also have the ind-stack Z ×Gr = Z̃ ×G(O) Gr of Hecke modifications at {0} where
Z̃ → Z is the G(O)-torsor classiyfing the data
(E, φ, σ)
where (E, φ) ∈ Z and σ is a trivialization E|Spec(O) ≃ G × Spec(O) of E on the formal
neighborhood of {0}. We have natural projection maps
h1 h2
Z ←− ˜
Z ×Gr −→ Z
Let Perv(Z) be the category of perverse sheaves on Z. Let PervZ (Z) ⊂ Perv(Z) the full
subcategory of perverse sheaves which are locally constant with respect to the stratification
Z = {Z λ }. We define a generic Hecke-equivariant perverse sheaf on Z to a perverse sheaf
F ∈ Perv(Z) on Z equipped with isomorphisms
φY : h!1,Y F ≃ h!2,Y F
for every smooth generic Hecke correspondence Y , satisfying some uatural conditions, see
[GN1, Section 3.2]. We denote by Perv(Z)Hecke the category of generic Hecke-equivariant
perverse sheaves on Z. Since we assume K is connected, the condition of generic-Hecke
equivariance is a property, not addition structure of a perverse sheaf on Z by [GN1, Propo-
sition 3.5.2], we see that the natural forgetful map Perv(Z)Hecke → Perv(Z) is fully-faithful
and induces an equivalence Perv(Z)Hecke ≃ PervZ (Z).
Following [GN1, Section 4.2], consider the perverse Hecke action
M
⋆p : Rep(G∨ ) × Perv(Z) → Perv(Z) ICV ⋆p F := p i ˜ V ))
H ((h2 )! (F ⊠IC
i
61
˜ V ∈ Perv(Z ×Gr)
where F ⊠IC ˜ is the twisted product of F⊠ICV with respect to the projections
h1 and h2 . Since the generic Hecke modifications commute with Hecke modification at the
pole point {0}, the perverse Hecke action descends to a well defined functor
(13.8) ⋆p : Rep(G∨ ) × Perv(Z)Hecke → Perv(Z)Hecke
For any λ ∈ Λ+
A , let ICZ λ ∈ Perv(Z) be the intersection cohomology complex of the stratum
Z (with constant coefficient). By [GN1, Proposition 3.5.1], we have ICZ λ ∈ Perv(Z)Hecke .
λ
Definition 13.9. [GN1, Definition 4.2.3] Let QglobK ⊂ Perv(Z)Hecke be the full subcategory
Hecke
of Perv(Z) whose objects are isomorphic to direct summands of perverse sheaves appear
in ICV ⋆p ICZ 0 for some V ∈ Rep(G∨ ).
Proposition 13.10. [GN1, Theorem 1.2.1] Qglob K is a semi-simple abelian category and every
irreducible object is isomorphic to ICZ λ for some λ ∈ Λ+A.
Remark 13.11. The proof of the proposition in loc. cit. is quite involved. We will give
an another proof of it using the results of the paper, see Corollary 13.15. In particular, we
will show that in fact the irreducible objects of Qglob
K are isomorphic to ICZ λ , λ ∈ σ(ΛT ),
confirming a conjecture in loc. cit..
We now recall the construction of fusion product on Qglob K following [GN1, Section 6.3].
Consider the base change
Z (2) = QM (2) (P1 , X) ×(P1 )2 C
of QM (2) (P1 , X) → (P1 )2 along the diagonal embedding C −→ (P1 )2 , z → (z, −z) We have
Z (2) |0 ≃ Z
and there is a Cartesian diagram
j i
Z (2) |C× / Z (2) o Z
f
C× / Co {0}
(2),λ (2),λ
The stratification {OK p } of Gr(2) in Section 4.2 descends to a stratification {LK (2) \OK p }
of QM (2) (P1 , X) ≃ LK (2) \Gr(2) which restricts to a stratification {Z (2),λp } of the base change
Z, where
(2),λp (2),λp
(13.9) Z (2),λp = (LK (2) \OK ) ∩ Z = (LK (2) \OK ) ×(P1 )2 C
We denote by ICZ (2),λp ∈ Perv(Z (2) ) the IC-complex of the stratum Z (2),λp .
For any λ1 , λ2 ∈ Λ+ +
A let λ1∪2 : p = {1} ∪ {2} → ΛA be the the map λ1∪2 (i) = λi . Note that
(2),λ1∪2 (2)
Z ⊂ Z |C× . For any ICZ λ1 , ICZ λ2 ∈ Perv(Z) we define the (global) fusion product of
as the following nearby cycles
(13.10) ICZ λ1 ⋆f ICZ λ2 = ψf (ICZ (2),λ1∪2 ) ∈ Perv(Z)
62
along the projection map f : Z (2) → C
Proposition 13.12. (1) For any ICZ λ1 , ICZ λ2 ∈ QglobK , we have ICZ λ1 ⋆f ICZ λ2 ∈ QK .
glob
(2) There exists unique associativity and commutativity constraints for the category QglobK
equipped with the fusion product ⋆f such that (Qglob
K , ⋆f ) is a neutral Tannakian category with
∨
Tannakian group isomorphic to the reductive subgroup Hsph ⊂ G∨ associated to X in [GN1].
Moreover, we have the following commutative diagram of tensor functors
Rep(G∨ )
▼▼▼
✉✉ ▼▼▼
✉✉✉ ▼▼▼
✉✉ ▼▼&
z✉
✉
≃
Qglob
K
/ Rep(H ∨
sph )
where the left vertical arrow is given by the perverse Hecke action on ICZ 0 ∈ Qglob
K .
Proof. [GN1, Corollary 4.2.6] and [GN1, Lemma 6.3.1] imply that ICZ (2),λ1∪2 ≃ j!∗ j ∗ (ICZ (2),λ1∪2 )
is ULA with respect to the projection f : Z (2) → C. Thus by [Z1, Theorem A.2.6] there is
an isomorphism
ψf (ICZ (2),λ1∪2 ) ≃ i∗ j!∗ (j ∗ ICZ (2),λ1∪2 )[−1].
It follows that the fusion product in (13.10) is the same as the one defined in [GN1, Section
6.3] and the proposition follows from the main results of [GN1].
∨ ∨
13.3.3. The identification Hreal = Hsph . The uniformization map (13.7) induces a map
r : Z ≃ K(C[t−1 ])\Gr → K(K)\Gr.
Proposition 13.13. The functor r ! [dim K] : D(K(K)\Gr) −→ D(Z) is t-exact
Proof. The proof is similar to the one in Theorem 8.14. For any λ ∈ Λ+
A , consider the
following Cartesian diagram
λ
jZ
Zλ / Z
rλ r
λ
jK
K(K)\OλK / K(K)\Gr
We need to check that for any F ∈ Perv(K(K)\Gr), we have
(13.11) (jZλ )∗ r ! [dim K](F) ∈ pcl D ≤0 (Z λ ) (jZλ )! r ! [dim K](F) ∈ pcl D ≥0 (Z λ ).
By Proposition 8.11 (in the case n = 1), the map r is strongly pro-smooth and the functor
r ! satisfies base change along ∗-pullback along any fp-locally closed map S → K(K)\Gr. It
follows that
(13.12) dim Z λ = dim OλK − dim O0K + dim Z 0 = hλ, ρi + dim Z 0 = hλ, ρi − dim K
(note that Z 0 ≃ BunK (P1 ) and dim BunK (P1 ) = − dim K) and
(jZλ )! r ! (F)[dim K] ≃ (r λ )! (jK
λ !
) (F)[dim K]
63
(jZλ )∗ r ! (F)[dim K] ≃ (r λ )! (jK
λ ∗
) (F)[dim K].
Since F is perverse sheaf on K(K)\Gr ≃ X(K)/G(O) with respect to the perversity function
in (8.8), we have
λ !
(jK ) (F) ∈ p D ≥0 (K(K)\OλK )[−hλ, ρi] λ ∗
(jK ) (F) ∈ p D ≤0 (K(K)\OλK )[−hλ, ρi].
On the other hand, since r λ is strongly pro-smooth, the characterization of !-adapted t-
structures in (8.6) implies that (r λ )! is t-exact with respect to the !-adapted t-structures
[BKV, Proposition 6.3.3 (c)] and hence
(13.12) p
(r λ )! ((jK
λ !
) (F))[dim K] ∈ p D ≥0 (Z λ )[−hλ, ρi+dim K] = D ≥0 (Z λ )[− dim Z λ ] = pcl D ≥0 (Z λ )
(13.12) p
(r λ )! ((jK
λ ∗
) (F))[dim K] ∈ p D ≤0 (Z λ )[−hλ, ρi+dim K] = D ≤0 (Z λ )[− dim Z λ ] = pcl D ≤0 (Z λ )
The desired claim (13.11) follows.
It follows from the proposition above that r ! [dim K] restricts to a functor
r ! [dim K] : Perv(K(K)\Gr) → Perv(Z)
on the category of perverse sheaves.
Theorem 13.14. The functor r ! [dim K] restricts to the horizontal tensor equivalence in a
commutative diagram of tensor functors
Rep G❍∨
✇✇ ❍❍
✇✇ ❍❍
✇ ❍❍
✇✇ ❍❍
$
{✇
✇
≃ / Qglob
QK K
Proof. Since the functor r ! [dim K] is t-exact and commutes with the convolution action of the
Hecke category D(G(O)\Gr) from the right, it restricts to a functor r ! [dim K] : QK → Qglob
K .
!
We first show that r [dim K] induces an equivalence of semi-simple abelian categories
(13.13) Φ : QK ≃ Qglob
K
C× / Co {0}
Proposition 8.11 implies that the map r (2) is strongly pro-smooth and Lemma 8.10 implies
that
(13.16) Φ(ICλK1 ⋆f ICλK2 ) = r ! [dim K]i∗ j∗ (ICλK1 ⊠ ICλK2 ⊠ CR>0 )) ≃
(2)
≃ i∗ j∗ (r>0 )! (ICλK1 ⊠ ICλK2 ⊠ CR>0 ))[dim K].
On the other hand, the same argument of proving (13.14) show that
(r (2) |C× )! (ICλK1 ⊠ ICλK2 ⊠ CR>0 [1]))[dim K] ≃ ICZ (2),λ1∪2 |R>0
where ICZ (2),λ1∪2 is the IC-complex on the stratum Z (2),λ1∪2 ⊂ Z (2) in (13.9). Thus we have
(2)
(13.17) i∗ j∗ (r>0 )! (F ⊠ F ′ ⊠ CR>0 ))[dim K] ≃ i∗ j∗ (ICZ (2),λ1∪2 |R>0 )[−1] ≃ ψf (ICZ (2),λ1∪2 )
≃ ICZ λ1 ⋆f ICZ λ2
Combining (13.16) and (13.17), we obtain the desired monoidal structure
c : Φ(ICλK1 ⋆f ICλK2 ) ≃ i∗ j∗ (r (2) |C× )! (ICλK1 ⊠ ICλK2 ⊠ CR>0 ))[dim K] ≃ ICZ λ1 ⋆f ICZ λ2
in (13.15).
Finally, we show that Φ is compatible with associativity and commutativity constraints
of QK and Qglob
K . Consider the following diagram of functors
where the vertical arrows are surjective morphisms of Hopf algebras and the horizontal arrow
Φ∗ is an isomorphism of co-algebras. By [N2, Lemma 9.2.1], the monoidal structure for Φ
implies that Φ∗ respects multiplication and the surjectivity of the vertical arrows implies
that Φ∗ is an isomorphism of Hopf algebras. The theorem follows.
The following corollary follows immediately from Proposition 13.7, Proposition 13.12,
confirming [GN1, Conjecture 7.3.2.] in the case of symmetric varieties.
∨ ∨
Corollary 13.15. (1) There is an isomorphism Hsph ≃ Hreal . In particular, the Weyl group
of Hsph is isomorphic to the small Weyl group of X. (2) The irreducible objects in Qglob
∨
K
are intersection cohomology sheaves ICZ λ on the closures of strata Z λ , λ ∈ σ(ΛT ), with
coefficients in trivial local systems.
A.2. From stacks to semi-analytic stacks. Let F = R or C. For any scheme Y over F
of finite type, its F -points Y (F ) is naturally a semi analytic set, denoted by YF . For any
strict ind-scheme Y ≃ colimi∈I Y i over F , its F -points is naturally a semi-analytic space,
denoted by YF .
In the paper, we will mainly consider F -stacks of the form Y ≃ Y /G where Y is a strict
ind-scheme acted on by a group ind-scheme G. A surjective morphism f : Y → Y from
a strict ind-scheme Y to a F -stack Y is called a F -surjective presentation if it induces a
surjective map Y (F ) → |Y(F )| on the set of isomorphism classes of objects. Note that when
F = C any surjective morphism is a C-surjective presentation.
Lemma A.2. Let f1 : Y1 → Y and f2 : Y2 → Y be two F -surjective presentations of Y. Let
Γi = Yi ×Y Yi ⇒ Yi be the corresponding groupoid. Then there is a canonical isomorphism of
semi-analytic stacks
Γ1,F \Yi,F ≃ Γ2,F \Y2,F .
Proof. Let Y = Y1 ×Y Y2 and Γ = Y ×Y Y be the corresponding groupoid. Then the natural
map ΓF \YF → Γi,F \Yi,F is an isomorphism. The lemma follows.
Definition A.3. Given a stack Y over F which admits a F -surjective presentation, we define
the associated semi-analytic stack to be
YF := ΓF \YF
67
where Y → Y is a F -presentation of Y.
By the lemma above YF is well-defined and the assignment Y → YF defines a functor from
the category of stacks over F which admit F -presentations to the category of semi-analytic
stacks.
Example A.4. Consider the case F = R. Let Y be a R-scheme and G be an algebraic
group over R acting on Y . Consider the algebraic stack Y = G\Y . Let T1 , ..., Ts ∈
H 1 (Gal(C/R), G(C)) be the isomorphism classes of G-torsors. Define Gi := AutG (Ti ) and the
R-scheme Yi := HomG (Ti , Y ). Note that Gi acts on Yi and the collection {G1 , ..., Gs } gives
all the pure-inner
Fs forms of G. Consider the real algebraic stack Gi \Yi . We have Gi \Yi ≃ Y
and the map i=1 Yi → Y is a R-surjective presentation. In addition,
F the R-surjective pre-
sentation above induces an isomorphism of semi-analytic stacks si=1 Gi,R \Yi,R ≃ YR .
One can regard real stacks as complex stacks with real structures and the discussion above
has an obvious generalization to this setting. Let Y be a complex stack and let σ be a real
structure on Y, that is, a complex conjugation (or a semi-linear involution) σ : Y → Y. Then
a surjective morphism f : Y → Y of Y is called a R-surjective presentation if it satisfies the
following properties. (1) There is a real structure σ on X such that f is compatible with
the real structures on Y and Y. (2) The map f induces a surjective map Y (C)σ → |Y(C)σ |.
One can check that Lemma A.2 still holds in this setting, thus for a pair (Y, σ) as above
which admits a R-surjective presentation, there is a well-defined semi-analytic stack YR given
by YR := Γ(C)σ \Y (C)σ , where Y → Y is a R-surjective presentation, Γ = Y ×Y Y is the
corresponding groupoid (Note that Γ has a canonical real structure σ coming from Y and
Y).
69
[N2] D. Nadler. Perverse sheaves on real loop Grassmannians, Inventiones Math. 159 (2005), 1-73.
[P] A. Pressley, The energy flow on the loop space of a compact Lie group, Journal of the London
Mathematical Society, 2 (1982), 557-566.
[PS] A. Pressley, G. Segal. Loop groups. Oxford Univ. Press, New York, 1986,.
[R] S. Raskin, Chiral principal series categories II: the factorizable Whittaker category, preprint, avail-
able at https://gauss.math.yale.edu/ sr2532/cpsii.pdf
[Z1] X. Zhu, An introduction to affine Grassmannians and the geometric Satake equivalence,
axXiv:1603.05593
[Z2] X. Zhu, The geometric Satake correspondence for ramified groups, Annales Scientifiques-cole Nor-
male Superieure Paris 48 (2).
70