D-module Final Report I
Hung Chiang
June 13, 2017
0.1 Borel-Weil-Bott Theorem
Let G/k be a connected semi-simple algebraic group, T be a maximal torus of G, B be a Borel subgroup of G containing T , N be the unipotent part of B, and X be the flag variety G/B. We thus has a choice of positive roots ∆+, simple roots Π ={α1,· · · , αl}, and Weyl vector ρ = α1+· · · + αl
2 . Let P be the weight lattice.
For each λ∈ L = Homk(B/N, k∗), we have a equivariant G-line bundle L(λ) on X ([HTT] p.255). Set
Psing ={λ ∈ P | ∃α ∈ ∆, ⟨λ − ρ, α∨⟩ = 0}, Preg = P − Psing.
Define a shifted action of W on P by
w⋆λ = w(λ − ρ) + ρ.
Theorem 1 (Borel-Weil-Bott,[HTT] 9.11.2.). Assume λ∈ L ⊂ P .
(i) If ⟨λ, α∨⟩ ≤ 0 for all α ∈ ∆+, then L(λ) is generated by global sections. That is, the natural morphism
OX ⊗kΓ(X,L(λ)) → L(λ)
is surjective.
(ii) L(λ) is ample if and only if ⟨λ, α∨⟩ < 0 for all α ∈ ∆+. (iii) Assume char(k) = 0.
(a) If λ∈ Psing, then Hi(X,L(λ)) = 0 for all i ≥ 0.
(b) Let λ∈ Preg and take w∈ W such that w⋆λ ∈ −P+. Then
Hi(X,L(λ)) =
L−(w⋆λ) if i = l(w),
0 otherwise.
0.2 Berlinson-Bernstein Theorems
From now on we assume k = C. For every smooth variety Y and locally free OY-module of finite rank V, we consider the sheaf of diffenertial operators on V, DYV ⊂ E ndCY(V). DYV is isomorphic to V ⊗OY DY ⊗OY V∗. There’s a natural filtration
Fp(DVY) = 0 for all p < 0,
Fp(DVY) ={P | fP − P f ∈ Fp−1(DYV)∀f ∈ OY} for all p ≥ 0.
Assume K is alinear algebraic group acting on Y and V is a K-equivlent vector bundle. There is a natural morphism ∂ : U (k) to Γ(Y, DVY). Let a ∈ k, then ∂a is defined by
(∂as)(y) = d
dt(exp(ta)s(exp(−ta)y))|t=0 (s∈ V, y ∈ Y ).
Here exp is the exponential map w.r.t right invariant vector fields. Algebraically, let φ : p∗2V ∼= σ∗V, then ∂a is determined by
ϕ((a⊗ 1) · φ−1(σ∗s)) = σ∗(∂as)
Here a is regarded as right invariant vector fields on K acting on k[K]([HTT]Equation 11.1.7).
Consider X. Let Dλ := DL(λ+ρ)X . We have Φλ : U (g) → Γ(X, Dλ).
Definition 1. Let z be the center of U (g) = U (h) ⊕ (n−U (g) + U (g)n)([HTT]
Equation 9.4.7). Let p be the projection from U (g) to U (h). f be the automorphism of U (h) defined by f (h) = h− ρ(h)1 for h ∈ h. For each λ ∈ h∗, define the central character
χλ(z) = (f◦ p(z))(λ) for all z ∈ z.
Proposition 1 ([HTT]Theorem 11.2.2). Let λ∈ L. Then Φλ : U (g) → Γ(X, Dλ) is surjective. Let z be the center of U (g). Then Φλ(z) = χλ(z) for all z ∈ z. Moreover, ker(Φλ) = U (g)(ker(χλ)).
We assume the proposition.
Let Modqc(Dλ) be the abelian category of Dλ-modules which are quasi-coherent overOX and Mod(g) be the categoriy of U (g)-modules. We have additive functors
Γ(X,·) : Modqc(Dλ)→ Mod(g), Dλ ⊗U (g)(·) : Mod(g) → Modqc(Dλ).
We have adjointness
HomDλ(Dλ⊗U (g)M,N ) ∼= HomU (g)(M, Γ(X,N )).
Let Mod(g, χ) be the category of U (g)-modules with central character χ and Modf(g, χ) be the full subcategory of Mod(g, χ) of finitely generated U (g)-modules. The propo- sition shows that Mod(g, χλ) ∼= Mod(Γ(X, Dλ)).
Theorem 2 ([HTT] Theorem 11.2.3 & 11.2.4). Let λ∈ L.
1. Suppose
⟨λ, α∨⟩ ≤ 0 for all α ∈ ∆+. (1) That is, λ∈ −P+. Then for all M ∈ Modqc(Dλ) we have Hk(X,M) = 0 for all k > 0.
2. Suppose
⟨λ, α∨⟩ < 0 for all α ∈ ∆+. (2) Then for all M ∈ Modqc(Dλ), the natural morphism
Dλ⊗U (g)Γ(X,M) → M is surjective.
Proof. For v ∈ −P+, Borel-Weil-Bott theorem says Γ(X,L(v)) = H0(X,L(v)) = L−(v) and pv : OX ⊗CL−(v) → L(v) is surjective. Since H omOX(L(v), OX) = L(−v) and HomC(L−(v),C) = L+(−v), we have L(−v) ,→ OX ⊗CL+(−v). Apply L(v) ⊗OX (·), we have iv : OX ,→ L(v) ⊗CL+(−v). Since L(v) is a line bundle, ker(pv) is a direct summand of OX ⊗CL−(v) as an OX-module locally. Therefore, im(iv) is a direct summand of L(v) ⊗CL+(−v) as an OX-module locally.
Let λ∈ L and M be a Dλ-module. Apply M ⊗OX (·), we get pv :M ⊗CL−(v)↠ M ⊗OX L(v),
iv :M ,→ M ⊗OX L(v) ⊗CL+(−v).
Proposition 2 ([HTT] Proposition 11.4.1). (i) If λ satisfies (2), then ker(pv) is a direct summand ofM ⊗CL−(v) as a sheaf of abelian groups.
(ii) If λ satisfies (1), then im(iv) is a direct summand of M ⊗OX L(v) ⊗CL+(−v) as a sheaf of abelian groups.
Suppose λ satisfies (1). For all M ∈ Modqc(Dλ), Hk(X,M) = lim−→Hk(X,N )
where N runs over all coherent OX-submodule of M. It suffices to prove that the natural map Hk(X,N ) → Hk(X,M) is the zero map. Fix N . Borel-Weil- Bott theorem says L(v) is ample if and only if v satisfies (2). Hence there is a v ∈ L ∩ −P+ such that Hk(X,N ⊗OX L(v)) = 0 for all k > 0. For this v, consider the commutative diagram
Hk(X,N ) Hk(X,M)
Hk(X,N ⊗OX L(v) ⊗CL+(−v)) Hk(X,M ⊗OX L(v) ⊗CL+(−v)).
iv∗
iv∗ is injective. On the other hand, Hk(X,N ⊗OXL(v)⊗CL+(−v)) = Hk(X,N ⊗OX
L(v)) ⊗CL+(−v) = 0 for all k > 0. So Hk(X,N ) → Hk(X,M) is the zero map.
Suppose λ satisfies (2). For given M ∈ Modqc(Dλ), set M′ be the image of Dλ⊗U (g)Γ(X,M) → M and M′′ be the cokernel of it. If M′′̸= 0, let N ⊂ M′′ be a nonzero coherent OX-submodule. There is a v∈ L ∩ −P+ such that N ⊗OX L(v) is generated by global sections. In this case, Γ(X,N ⊗OX L(v)) ̸= 0, neither is Γ(X,M′′⊗OX L(v)) On the other hand,
pv∗ : Γ(X,M′′)⊗CL−(v) = Γ(X,M′′⊗CL−(v))→ Γ(X, M′′⊗OX L(v)) is surjective. So Γ(X,M′′)̸= 0. Consider the exact sequence
0→ Γ(X, M′)→ Γ(X, M) → Γ(X, M′′)→ 0.
By definition, Γ(X,M) = Γ(X, Dλ ⊗U (g) Γ(X,M)) ↠ Γ(X, M′). So Γ(X,M′) = Γ(X,M) and hence Γ(X, M′′) = 0. SoM′′must be 0. The isomorphism Γ(X,M) = Γ(X, Dλ⊗U (g)Γ(X,M)) is proved in the proof of Corollary 1.
0.3 Equivalences of Categories
For λ ∈ L satisfying (1), we denote Modeqc(Dλ) the full subcategory of Modqc(Dλ) consisting of objectsM satisfying that
(a) Dλ ⊗U (g)Γ(X,M) → M is surjective.
(b) For all nonzero subobject N ⊂ M in Modqc(Dλ), we have Γ(X,N ) ̸= 0.
Set Modec(Dλ) = Modeqc(Dλ)∩ Modc(Dλ).
Corollary 1. Γ(X,·) induces equivlences of categories
Modeqc(Dλ) ∼= Mod(g, χλ), Modec(Dλ) ∼= Modf(g, χλ).
Proof. We first prove that M → Γ(X, Dλ⊗U (g)M ) is an isomorphism for all M ∈ Mod(g, χλ). For given M ∈ Mod(g, χλ), consider an exact sequence
Γ(X, Dλ)⊕I → Γ(X, Dλ)⊕J → M → 0.
From Theorem 2.1, Γ(X,·) : Modqc(Dλ) → Mod(g, χλ) is an exact functor. So Γ(X, Dλ⊗U (g)(·)) is right exact. We have an commutative diagram with exact rows
Γ(X, Dλ)⊕I Γ(X, Dλ)⊕J M 0
Γ(X, Dλ)⊕I Γ(X, Dλ)⊕J Γ(X, Dλ⊗U (g)M ) 0.
id id
So M ∼= Γ(X, Dλ⊗U (g)M ).
Now we show that Γ(X,·) : Modeqc(Dλ)→ Mod(g, χλ) is fully faithful. That is, for all M1,M2 ∈ Modeqc(Dλ),
Γ : HomDλ(M1,M2)→ HomU (g)(Γ(X,M1), Γ(X,M2))
is an isomorphism. Since Dλ ⊗U (g)Γ(X,M1)→ M1 is surjective, we have HomDλ(M1,M2) ,→ HomDλ(Dλ⊗U (g)Γ(X,M1),M2)
∼= HomU (g)(Γ(X,M1), Γ(X,M2)).
Assume ϕ ∈ HomU (g)(Γ(X,M1), Γ(X,M2)). Let K1 be the kernel of Dλ ⊗U (g)
Γ(X,M1)→ M1. Apply the exact functor Γ(X,·) on the exact sequenct 0→ K1 → Dλ⊗U (g)Γ(X,M1)→ M1 → 0
and we get the exact sequence
0→ Γ(X, K1)→ Γ(X, M1)→ Γ(X, M1)→ 0.
So Γ(X,K1) = 0. Let K2 be the image of
K1 Dλ⊗U (g)Γ(X,M1) 1⊗ϕ Dλ⊗U (g)Γ(X,M2) M2.
Since Γ(X,·) is exact and Γ(X, K1) = 0, Γ(X,K2) = 0. So K2 = 0. hence we obtain ψ :M1 ∼= Dλ⊗U (g)Γ(X,M1)/K1 → M2 with Γ(ψ) = ϕ.
Next, we prove that Γ(X,·) : Modeqc(Dλ)→ Mod(g, χλ) is essentially surjective.
Given M ∈ Mod(g, χλ). Let L be a maximal element of the set of subobjects K of Dλ ⊗U (g)M in Modqc(Dλ) satisfying that Γ(X,K) = 0. Set M = Dλ⊗U (g)M /L.
Then Γ(X,M) = Γ(X, Dλ⊗U (g)M )/Γ(X,L) = M. Dλ⊗U (g)M → M is surjective.
For allN ⊂ M, the maximality of L shows that Γ(X, N ) ̸= 0. So M ∈ Modeqc(Dλ).
Finally, we have to show that Modec(Dλ) and Modf(g, χλ) correspond to each other. Let M ∈ Modf(g, χλ). Γ(X, Dλ) is left-noetherian. There is an exact se- quence
Γ(X, Dλ)⊕I → Γ(X, Dλ)⊕J → M → 0
with |I|, |J| < ∞. Apply the right exact functor Dλ⊗U (g)(·) on it and we get the exact sequence
Dλ⊕I → Dλ⊕J → Dλ⊗Ug M → 0.
We get Dλ⊗Ug M ∈ Modec(Dλ) and henceM = Dλ⊗UgM /L.
Conversely, let M ∈ Modec(Dλ). Since Dλ⊗UgΓ(X,M) → M is surjective, M is locally generated by finitely many global sections. Since X is quasi-compact, M is globally generated by finitely many global sections. We have an exact sequence
D⊕Iλ → M → 0
where|I| < ∞. Apply Γ(X, ·) on it and we get the exact sequence
Γ(X, Dλ)⊕I → Γ(X, M) → 0 Hence Γ(X,M) is an finitely generated U(g)-module.
Suppose λ satisfies (2). Then Modqc(Dλ) = Modeqc(Dλ). In this case, we have Corollary 2. Γ(X,·) induces equivlences of categories
Modqc(Dλ) ∼= Mod(g, χλ), Modc(Dλ) ∼= Modf(g, χλ).
Let K be a closed subgroup of G. We consider K-equivariant g-modules. That is, a g-module with a K-action satisfying that
k-actions obtained from the g-action and the K-action coincide. (3) k· (a · m) = Ad(k)(a) · (k · m) for all k ∈ K, a ∈ g, and m ∈ M. (4) We denote the full subcategory consisting of K-equivariant objects of Mod(g, χ) and Modf(g, χ) by Mod(g, χ, K) and Modf(g, χ, K), respectively.
We also introduce K-equivariant D-modules. Let K acts on Y . Consider mor- phisms p2 : K× Y → Y , σ : K × Y → Y , m : K × K → K defined by p2(k, y) = y, σ(k, y) = ky, m(k1, k2) = k1, k2. A K-equivariant DY-module is a DY-module M with a isomorphism of DK×Y-modules
φ : p∗2M ∼= σ∗M satisfying the cocycle condition.
We consider categories Modqc(DY, K) and Modc(DY, K). For λ =−ρ, we have Mod(g, χ−ρ) ∼= Modqc(DX) and Modf(g, χ−ρ) ∼= Modqc(DX).
Theorem 3. For any closed subgroup K ≤ G, we have Mod(g, χ−ρ, K) ∼= Modqc(DX, K) and Modf(g, χ−ρ, K) ∼= Modqc(DX, K).
Proof. What we have to prove is K-equivariances defined on Mod(g, χ−ρ) and Modqc(DX) coincide.
Consider M ∈ Modqc(DX). K, X and K × X are all D-affine. So DK×X- modules are Γ(K×X, DK×X) = Γ(K, DK)⊗CΓ(X, DX)-modules. Since Γ(K, DK) ∼= Γ(K,OK)⊗CU (k) and Γ(X, DX) ∼= U (g)/U (g) ker(χ−ρ), DK×X-module structures are determined by actions of Γ(K,OK)⊗ 1, k ⊗ 1 and 1 ⊗ g.
Γ(K × X, p∗2M) ∼= Γ(K,OK)⊗CΓ(X,M). For σ∗M, consider isomorphisms ϵ1 : K × X → K × X, ϵ2 : K × X → K × X defined by ϵ1(k, x) = (k, kx), ϵ2(k, x) = (k, k−1x). ϵ1 = ϵ−12 and σ = p2◦ ϵ1. So
Γ(K× X, σ∗M) ∼= Γ(K× X, ϵ∗1p2∗M) ∼= Γ(K× X, (ϵ2)∗p∗2M)
∼= Γ(K× X, p∗2M) ∼= Γ(K,OK)⊗CΓ(X,M).
For given h ∈ Γ(X, OX) and m ∈ Γ(X, M), the element h ⊗ m ∈ Γ(K, OK)⊗C
Γ(X,M) corresponds to the global section h ◦ p1 ⊗ p−12 m of p2∗M = OK×X ⊗p−12 OX p−12 M and the global section h ◦ p1 ⊗ σ−1m : (k, x) 7→ (k, h(k)k−1 · m(kx)) of σ∗M = OK×X ⊗σ−1OX σ−1M. The Γ(K, DK)⊗CΓ(X, DX)-action on p∗2M. is
(f ⊗ 1) · (h ⊗ m) = fh ⊗ m for all f ∈ Γ(K, OK), (a⊗ 1) · (h ⊗ m) = a · h ⊗ m for all a ∈ k,
(1⊗ p) · (h ⊗ m) = h ⊗ p · m for all p ∈ g.
Consider the Γ(K, DK)⊗CΓ(X, DX)-action on σ∗M. (f ⊗ 1) · (h ⊗ m) = fh ⊗ m for all f ∈ Γ(K, OK). (a⊗ 1) · (h ⊗ m) = a · h ⊗ m − h ⊗ a · m for all a ∈ k. Finally,
d
dt exp(tp)k−1m(k exp(−tp)x)|t=0 = d
dtk−1exp(t Ad(k)(p))m(exp(t Ad(k)(p))kx)|t=0. Let Ad(k)(p) =∑
ihi(k)pi. We have (1⊗ p) · (h ⊗ m) = ∑
ihhi⊗ pi· m for all p ∈ g.
The K-equivariance of M is equivlent to an Γ(K, DK)⊗CΓ(X, DX)-module isomorphism from Γ(K,OK)⊗CΓ(X,M) ∼= Γ(K× X, p∗2M) to Γ(K × X, σ∗M) ∼= Γ(K,OK)⊗CΓ(X,M) satisfying the cocycle condition. Since Γ(K, OK)-actions on both sides are the same, the condition is aC-module homomorphism
e
φ : Γ(X,M) = 1 ⊗CΓ(X,M) → Γ(K, OK)⊗CΓ(X,M)
satisfying the cocycle condition and e
φ((a⊗ 1) · (1 ⊗ m)) = (a ⊗ 1) · eφ(1⊗ m) for all a ∈ k, m ∈ Γ(X, M). (5) e
φ((1⊗ p) · (1 ⊗ m)) = (1 ⊗ p) · eφ(1⊗ m) for all p ∈ g, m ∈ Γ(X, M). (6) Let φ(m) =e ∑
jgj ⊗ mj. The cocycle condition is equivlent to a K-representation structure of Γ(X,M). (5) is
0 = ∑
j
a· gj ⊗ mj− a · m.
a : m7→∑
ja·gj⊗mj is the k-action obtained from the K-action while a : m7→ a·m is the k-action obtained from the g-action. So (5) is (3).
(6) is
∑
j
gj⊗ p · mj =∑
i,j
higj ⊗ pi · mj, which is (4).
Theorem 4. Let Y be a smooth variety and K be a linear algebraic group action on Y . Suppose there are only finitely many K-orbits in Y . Then Modc(DY, K) ∼= Modrh(DY, K). Moreover, the simple objects in Modc(DY, K) is parametrized by Υ(Y, K), the set of pairs (O, L), where O⊂ Y is an irreducible K-orbit and L is a K-equivariant local system on Oan.
Proof. We use induction on the number of K-orbits of Y . Suppose Y is a ho- mogeneous K-space. Then Y ∼= K/K′ for some K′ ≤ K. Consider morphisms σ : K × Y → Y the natural action, p2 : K × Y → Y the second projection, l : K → Spec C, π : K → Y the quotient map, j : Spec(C) → K, j(x) = K′ and i : K → K × Y , i(k) := (k−1, kK′). Then for any M ∈ Modc(DY, K), we have
π∗M = (p2◦ i)∗M = i∗p∗2M ∼= i∗σ∗M = (σ ◦ i)∗M
= (j◦ l)∗M = l∗j∗M = OX ⊗C(j∗M).
j∗M is a finite dimensional C-vector space, so π∗M ∈ Modrh(DK). Since π is smooth,M ∈ Modrh(DY).
Now consider the general case. Let O be a closed K-orbit of Y and Y′ = Y −O.
Suppose i : O ,→ Y and j :,→ Y . Then we have the distinguish triagle
∫
ii†M M ∫
jj†M +1 .
We have i†M ∈ Dbc(DO) and j†M ∈ Dcb(DY′). By induction hypothesis, i†M ∈ Drhb (DO) and j†M ∈ Dbrh(DY′). We conclude that ∫
ii†M,∫
jj†M ∈ Drhb (DY) and henceM ∈ Modrh(DY).
By Riemann-Hilbert correspondence, Modrh(DY, K) ∼= Perv(CY, K), which is parametrized by Υ(Y, K).
In particular, B has only finite orbits in X. We conclude that simple objects in Modf(g, χ−ρ, B) are parametrized by Υ(X, B).
Bibliography
[HTT] R. Hotta, K. Takeuchi, T. Tanisaki, D-modules, Perverse Sheaves, and Rep- resentation Theory.
[S] T.A. Springer, Linear Algebraic Groups
D-module Final Report II
Hung Chiang
June 26, 2017
Proposition 1 ([HTT] Proposition 9.4.5, proved in[HC]). χλ = χµ if and only if λ and µ are in the same W -orbit.
⟨λ, α∨⟩ ≤ 0 for all α ∈ ∆+. (1)
⟨λ, α∨⟩ < 0 for all α ∈ ∆+. (2) For v ∈ −P+, Borel-Weil-Bott theorem says Γ(X,L(v)) = H0(X,L(v)) = L−(v) and pv :OX⊗CL−(v)→ L(v) is surjective. Since H omOX(L(v), OX) = L(−v) and HomC(L−(v),C) = L+(−v), we have L(−v) ,→ OX⊗CL+(−v). Apply L(v)⊗OX(·), we have iv :OX ,→ L(v) ⊗CL+(−v). Since L(v) is a line bundle, ker(pv) is a direct summand of OX ⊗CL−(v) as an OX-module locally. Therefore, im(iv) is a direct summand of L(v) ⊗CL+(−v) as an OX-module locally.
Let λ∈ L and M be a Dλ-module. Apply M ⊗OX (·), we get pv :M ⊗CL−(v)↠ M ⊗OX L(v),
iv :M ,→ M ⊗OX L(v) ⊗CL+(−v).
Proposition 2 ([HTT] Proposition 11.4.1). (i) If λ satisfies (2), then ker(pv) is a direct summand ofM ⊗CL−(v) as a sheaf of abelian groups.
(ii) If λ satisfies (1), then im(iv) is a direct summand of M ⊗OX L(v) ⊗CL+(−v) as a sheaf of abelian groups.
Proof. Let
L−(v) = L1 ⊃ L2 ⊃ · · · ⊃ Lr= 0
be a filtration of B-modules of L−(v) satisfying that Li/Li+1 is the character µi of B, µ1 = v, and µi < µj only if i < j. Then we obtain corresponding filtrations
OX ⊗CL−(v) = V1 ⊃ V2 ⊃ · · · ⊃ Vr = 0,
M ⊗CL−(v) =V1 ⊃ V2 ⊃ · · · ⊃ Vr = 0, and
M ⊗O L(v) ⊗CL+(−v) = Wr ⊃ Wr−1 ⊃ · · · ⊃ W1 = 0.
The corresponding composition factors are
Vi/Vi+1 ∼=M ⊗OX L(µi), Wi+1/Wi ∼=M ⊗OX L(v − µi).
SinceM ⊗OX L(µ) is a Dλ+µ-module, the action of z on it is χλ+µ. So we have
r−1
∏
i=1
(z− χλ+µi)(M ⊗CL−(v)) = 0
and r∏−1
i=1
(z− χλ+v−µi)(M ⊗OX L(v) ⊗CL+(−v)) = 0.
Seen as sheaves of abelian groups, M ⊗CL−(v) and M ⊗OX L(v) ⊗CL+(−v) are equipped with locally finite z-actions and thus have decompositions into χ-primary parts:
M ⊗CL−(v) =⊕
χ
(M ⊗CL−(v))χ, M ⊗OX L(v) ⊗CL+(−v) =⊕
χ
(M ⊗OX L(v) ⊗CL+(−v))χ.
The morphisms pv and iv are V1 → V1/V2 and W2 → Wr, respectively. It suffices to prove that
(i) If λ satisfies (2), then ker(pv) = (M ⊗CL−(v))χλ+v. That is, χλ+µi = χλ+v ⇔ i = 1.
(ii) If λ satisfies (1), then im(iv) = (M⊗OXL(v)⊗CL+(−v))χλ. That is, χλ+v−µi = χλ ⇔ i = 1.
Suppose λ satisfies (2). If χλ+µi = χλ+v, then there is a w∈ W such that w(λ+µi) = λ + v. That is, (w(λ)− λ) + (w(µi)− v) = 0. Since ⟨λ, α⟩ < 0 for all α ∈ ∆+, w(λ)− λ ≥ 0 and the equality holds if and only if w = id. Since w(µi) is a weight of L−(v), w(µi)≥ v. So w = id and thus µi = v. That is, i = 1.
Suppose λ satisfies (1). If χλ+v−µi = χλ, then there is a w ∈ W such that w(λ) = λ + v− µi. That is, (w(λ)− λ) + (µi − v) = 0. Since ⟨λ, α⟩ ≤ 0 for all α∈ ∆+, w(λ)≥ λ. Also, µi ≥ v. So µi = v and thus i = 1.
Theorem 1 ([HTT] Theorem 11.6.1). Let Y be a smooth variety and K be a linear algebraic group action on Y . Suppose there are only finitely many K-orbits in Y . Then Modc(DY, K) ∼= Modrh(DY, K). Moreover, the simple objects in Modc(DY, K) is parametrized by Υ(Y, K), the set of pairs (O, L), where O ⊂ Y is an irreducible K-orbit and L is a K-equivariant local system on Oan.
Proof. We use induction on the number of K-orbits of Y . Suppose Y is a ho- mogeneous K-space. Then Y ∼= K/K′ for some K′ ≤ K. Consider morphisms σ : K × Y → Y the natural action, p2 : K × Y → Y the second projection, l : K → Spec C, π : K → Y the quotient map, j : Spec(C) → K, j(x) = K′ and i : K → K × Y , i(k) := (k−1, kK′). Then for any M ∈ Modc(DY, K), we have
π∗M = (p2◦ i)∗M = i∗p∗2M ∼= i∗σ∗M = (σ ◦ i)∗M
= (j◦ l)∗M = l∗j∗M = OX ⊗C(j∗M).
j∗M is a finite dimensional C-vector space, so π∗M ∈ Modrh(DK). Since π is smooth,M ∈ Modrh(DY).
Now consider the general case. Let O be a closed K-orbit of Y and Y′ = Y −O.
Suppose i : O ,→ Y and j :,→ Y . Then we have the distinguish triagle
∫
ii†M M ∫
jj†M +1 .
We have i†M ∈ Dbc(DO) and j†M ∈ Dcb(DY′). By induction hypothesis, i†M ∈ Drhb (DO) and j†M ∈ Dbrh(DY′). We conclude that ∫
ii†M,∫
jj†M ∈ Drhb (DY) and henceM ∈ Modrh(DY).
By Riemann-Hilbert correspondence, Modrh(DY, K) ∼= Perv(CY, K), which is parametrized by Υ(Y, K).
In particular, B has only finite orbits in X. We conclude that simple objects in Modf(g, χ−ρ, B) are parametrized by Υ(X, B).
0.1 Highest Weight Module
Definition 1. Let λ∈ h∗ and M be a g-module. If there exists 0 ̸= m ∈ M such that m∈ M , nm = 0, and M = U (g)m, then M is called a highest weight module
with highest weight λ. m is called a highest weight vector.
In this case, M = U (n−)m and M = ⊕
µ≤λMµ. Mλ = C m. Since M is generated by m, M is the quotiend U (g)/N as a U (g)-module. The relation contains at least n and h− λ(h) for all h ∈ h.
Definition 2. The Verma module is defined as M (λ) := U (g)/(U (g)n +∑
h∈h
U (g)(h− λ(h)1)).
M (λ) is the unique maximal highest weight module. If M be a highest weight module, there is a unique surjective homomorphism f : M (λ) → M such that f (¯1) = m.
Lemma 1 ([HTT] Lemma 12.1.3). M (λ) is a free U (n−)-module. In particular, we compute that
ch(M (λ)) =∑
µ
dim(M (λ)µ)eµ=∑
β≤0
dim(U (n−)β)eλ+β
=eλ ∏
β∈∆+
(1 + e−β + e−2β+· · · )
= eλ
∏
β∈∆+(1− e−β). Proof. Let I = (U (g)n +∑
h∈hU (g)(h− λ(h)1)). We want to prove that U(g) = U (n−)⊕ I. By P BW theorem, we have a canonical isomorphism
U (n−)⊗ U(h) ⊗ U(n) ∼= U (g).
So we have
∑
h∈h
U (g)(h− λ(h)1)) =∑
h∈h
U (n−)U (h)U (n)(h− λ(h)1))
=∑
h∈h
U (n−)U (h)(C +U(n)n)(h − λ(h)1))
⊂U(n−)
(∑
h∈h
U (h)(h− λ(h)1)) )
+ U (g)n.
So I = U (n−)(∑
h∈hU (h)(h− λ(h)1)))
+ U (g)n. Finally we have the isomorphism
U (g) =U (n−)U (h)U (n)
=U (n−)U (h)(C ⊕U(n)n)
=U (n−)U (h)⊕ U(g)n
=U (n−) (
C ⊕∑
h∈h
U (h)(h− λ(h)1)) )
⊕ U(g)n
=U (n−)⊕ I.
Lemma 2 ([HTT]Lemma 12.1.4). There is a unique maximal proper U (g)-submodule N ⊂ M(λ).
Proof. Any proper U (g)-submodule of M (λ) is a weight module whose weights < λ.
So the sum of them is also a proper U (g)-submodule.
Define L(λ) = M (λ)/N . L(λ) is the minimal highest weight module.
Problem 1. Compute ch(L(λ)).
Example 1. If λ∈ ∆+, then L(λ) = L+(λ). Weyl’s character formula says ch(L(λ)) =
∑
w∈W(−1)l(w)ww(λ+ρ)−ρ
∏
β∈∆+(1− e−β)
= ∑
w∈W
(−1)l(w)ch(M (w(λ + ρ)− ρ)).
Lemma 3 ([HTT] Lemma 12.1.6). For z∈ z, zm = χλ+ρ(z)m.
Proof. We decompose z into u + v where z ∈ U(h) and v ∈ n−U (g) + U (g)n. Then zm = um = λ(u)m = χλ+ρ(z)m.
Proposition 3 ([HTT] Proposition 12.1.7). Let M be a highest weight module with highest weight λ. The M has a decomposition series with finite length and each composition factor of it has the form L(µ) where µ≤ λ and µ + ρ ∈ W (λ + ρ).
Proof. If M is simple then we are done. If M is not simple, then we take a nonzero proper submodule N ⊂ M. Let µ be a maximal weight of N and 0 ̸= n ∈ Nµ, then U (g)m ⊂ N is a highest weight module with highest weight µ. χµ+ρ(z)n = zn = χλ+ρ(z)n for all z ∈ z. So χµ+ρ = χλ+ρ. We have µ < λ and µ + ρ ∈ W (λ + ρ).
Replace M by N and repeat the process. We can repeat only finitely many times and obtain a simple U (g)-module N1, which is a highest weight module with highest weight µ1. N1 ∼= L(µ1). Replace M by M /N1 and repeat the process. We obtain a sequence
0 = N0 ⊂ N1 ⊂ N2 ⊂ · · · ,
the composition factors of which have the form L(µ) for some µ ≤ λ and µ + ρ ∈ W (λ + ρ). Since|W (λ + ρ)| < ∞ and L(µ) can occur no more than dim(Mµ) times, the sequence is finite.
Fix a equivlence class Λ = W (λ + ρ)− ρ. Let aµλ the the multiplicity of L(µ) appearing in the deconposition series of M (λ). We have aµλ ̸= 0 only if µ ∼ λ and µ≤ λ. aλλ = 1. Let (bµλ) be the inverse matrix of (aµλ). Then bµλ ∈ Z and
ch(M (λ)) =∑
µ∈Λ
aµλch(L(µ)).
ch(L(λ)) =∑
µ∈Λ
bµλch(M (µ)).
It suffices to compute bµλ.
0.2 Kazhdan-Lusztig Conjecture
The problem is answered when Λ = W (−ρ)−ρ. In this case, Λ ⊂ P . We are consid- ering objects M (−w(ρ) − ρ), L(−wρ − ρ) ∈ Modf(g, χρ, B) = Modrh(DX, B). Every object in M odf(g, χρ, B) has a composition series of finite length, which is proved similarly as in the proof above. We consider the Grothdieck group K(Modf(g, χρ, B)).
We have
[L(−wρ − ρ)] = ∑
y∈W
byw[M (−yρ − ρ)].
[M (−wρ − ρ)] = ∑
y∈W
ayw[L(−yρ − ρ)].
We want to compute byw.
Definition 3. The Hecke algebra H(W ) is the Z[q1, q−1] algebra which is freely generated by{Tw | w ∈ W } as a Z[q1, q−1]-module with multiplicative relations
TyTw = Tyw, if l(yw) = l(y) + l(w).
(Ts+ 1)(Ts− q) = 0, if s∈ W.
Proposition 4 ([HTT] Proposition 12.2.3). There exists a unique family {Py,w(q)} of polynomials in Z[q] satisfying the following conditions:
Py,w(q) = 0 if y̸≤ w, Pw,w(q) = 1, deg(Py,w(q))≤ l(w)− l(y) − 1
2 if y < w,
∑
y≤w
Py,w(q)Ty = ql(w)∑
y≤w
Py,w(q−1)Ty−1−1.
Conjecture 1 (Kazhdan-Lusztig).
by,w = (−1)l(w)−l(y)Py,w(1).
Definition 4. For each w∈ W we define
Xw = BwB/B.
Here w is seen as an element in W = NG(H)/H. The Schubert variety is defined as Xw.
Proposition 5 ([HTT] Theorem 9.9.4, 9.9.5). X is the disjoint union of{Xw | w ∈ W}. Each Xw is isomorphic to Cl(w). Xw =⨿
y≤wXw.
We denote by IC(CXw) the intersection complex on Xw and set CπXw = IC(CXw)[− dim(Xw)].
We’ll show that the Kazhdan-Lusztig conjecture is reduced the theorem below.
Theorem 2 (Kazhdan-Lusztig,[HTT]Theorem 12.2.5). For any y, w∈ W , we have
∑
i
dim(Hi(CπXw)yB)qi/2 = Py,w(q).
In particular, We have Hi(CπXw)yB = 0 for all odd i and
∑
j
(−1)jdim(Hj(CπXw)yB) = Py,w(1).
Let Mw = DX ⊗U (g)M (−w(ρ) − ρ), Lw = DX ⊗U (g)L(−w(ρ) − ρ), and Nw =
∫
iw
OXw = (iw)∗(DX←Xw ⊗DXw OXw).
Nw ∈ Modc(DX, B) = Modrh(DX, B).
Lemma 4 ([HTT] Lemma 12.3.1). Let w∈ W . Then
(i) ch(Γ(X,Nw)) = ch(M (−w(ρ)−ρ)). In particular, [Mw] = [Nw] in K(Modrh(DX, B)).
(ii) The only DX submodule of Nw whose support is contained in Xw− Xw is 0.
Proof. Define two subalgebras of g:
n1 = ⊕
α∈∆+∩w(∆+)
g−α, n2 = ⊕
α∈∆+∩−w(∆+)
gα.
Let the corresponding unipotent subgroup of G be N1 and N2 respectively. Define a morphism φ : N1× N2 → X by
φ(n1, n2) = n1n2wB/B.
Then φ is an open embedding. φ({e} × N2) = Xw. Let V = im(φ), we have the commutative diagram
N1× N2 V X
N2 ={e} × N2 Xw φ
iw
So
Γ(X,Nw) =Γ(
X, (iw)∗(
DX←Xw ⊗DXw OXw
))
=Γ(Xw, DX←Xw ⊗DXw OXw)
=Γ(Xw, DV←Xw ⊗DXw OXw)
∼=Γ(N2, DN1×N2←N2 ⊗DN2 ON2).
DN1×N2←N2 ∼= (
DN1×{e}⊗ON1×{e} C)
⊗C
( Ω⊗−1N
1×{e}⊗ON1×{e} C)
⊗C⊗CΓ(N2,N2).
First, DN1 = U (n1)⊗COX, so DN1×e⊗ON1×{e} C ∼= U (n1). Second,
Ω⊗−1N
1×{e}⊗ON1×{e} C ∼=∧dim(n1)(n1⊗COX)⊗OX C = ∧dim(n1)n1.
Finally, the exponential map gives the isomorphism n2 ∼= N2. Therefore, Γ(N2,N2) ∼= Γ(n2,On2) = S(n∗2).
ch(Γ(X,Nw)) = ch(U (n1)) ch(∧dim(n1)n1) ch(S(n∗2)).
We compute that
ch(U (n1)) = ∏
α∈∆+∩w(∆+)
(1 + e−α+ e−2α+· · · ) = ∏
α∈∆+∩w(∆+)
1 1− e−α, ch(∧dim(n1)) = e
∑
α∈∆+∩w(∆+)−α = e−w(ρ)−ρ, and
ch(S(n∗2)) = ∏
α∈∆+∩−w(∆+)
(1 + e−α+ e−2α+· · · ) = ∏
α∈∆+∩−w(∆+)
1 1− e−α. So
ch(Γ(X,Nw)) = e−w(ρ)−ρ
∏
α∈∆+(1− e−α) = ch(M (−w(ρ) − ρ)).
Set Z = X− V and j : V → X be the open embedding, we have a distinguished traingle
RΓZ(Nw) Nw j∗(Nw) +1 .
By definition, Nw → j∗(Nw) is an isomorphism, so RΓZ(Nw) = 0. So ΓZ(Nw).
Hence the only DX submodule of Nw whose support is contained in Z is 0. Since Xw− Xw ⊂ Z, the assertion follows.
LetL(Xw,OXw) be the minimal extension of the DXw-module Xw. L(Xw,OXw)∈ Modrh(DX, B).
Proposition 6 ([HTT] Lemma 12.3.2). Let w ∈ W . Then we have (i)
L =L(X ,O ).