Patrick Lei

All notes

D-modules and localization

Talks by Fan Zhou and Kevin Chang, Columbia, Spring 2022.

Geometric Representation Theory Learning Seminar
Spring 2022
Notes by Patrick Lei
Lectures by Various

Columbia University

Disclaimer

These notes were taken during the seminar using the vimtex package of the editor neovim. Any errors are mine and not the speakers’. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the lecturers. If you find any errors, please contact me at plei@math.columbia.edu.

1 Fan (Feb 09): D-modules, bare minimumz, part 1

For now, let X=Spec⁡A be a smooth affine variety over ℂ.

Definition 1.0.1.

A differential operator of order ≤i is a ℂ-linear D:A→A such that for any ϕ∈A, [D,ϕ] is a differential operator of order ≤i−1. In other words, for any ϕ0,…,ϕn∈A, we have

We will call the set of differential operators of order ≤i 𝒟i. We will see later that 𝒪X≅𝒟0, 𝒪X⊕𝒯X≅𝒟1, and Gri⁡𝒟≅Sym𝒪Xi⁡𝒯X. Some other facts are:

  • •

    𝒟 is generated by 𝒪X,𝒯X under the rules

  • •

    For any ϕ∈A, Aϕ⊗A𝒟i⁢(A)≅Di⁢(Aϕ).

We will now prove these facts. First, 𝒪X≅𝒟1⁢(X) by ϕ↦⋅ϕ and D↦D⁡(1). On the other hand, we note that 𝒪X⊕𝒯X→𝒟1 is given by ϕ⊕ξ↦(⋅ϕ)+ξ(−) and D↦D⁡(1)⊕[D,−].

Lemma 1.0.2.

We have 𝒟i⋅𝒟j⊆𝒟i+j and [𝒟i,𝒟j]⊆𝒟i+j−1.

Proposition 1.0.3.

There is a canonical map Sym𝒪X⁡𝒯X→𝒟X.

On one hand, note that 𝒯X≃Gr1⁡(D). Taking symmetric powers of this, we obtain a map Sym⁡𝒯X→Gr⁡𝒟.

To define the inverse, note that we have a morphism in degree 0 already. Inductively, we consider the map

This kills 𝒪X⊗𝒟Xn−1, so it descends to 𝒪X⊗Grn⁡𝒟X. Inductively, we use the adjunction to obtain

It is easy to check that this is an inverse.

We will now consider generation of 𝒟 by 𝒪 and 𝒯. We will endow 𝒪⊕𝒯 with the Lie bracket

We will show that

Clearly there is a map ϕ↦ϕ,ξ↦ξ, whose source we call 𝒰. The commutation relations imply that Gr⁡𝒰 is commutative, so we have a surjection

Clearly, we see that Gr⁡𝒰→DX is an isomorphism, so 𝒰→𝒟X is an isomorphism.

Finally, consider the map Aϕ⊗𝒟i⁢(A)→𝒟i⁢(Aϕ) given by

Clearly this is well-defined, so it suffices to check that Aϕ⊗Gri⁡𝒟⁡(A)→Gri⁡D⁡(Aϕ) is an isomorphism. But both sides are Aϕ⊗Symi⁡TA=Symi⁡TAϕ, and by exactness of localization we are done.

Now that we have a sheaf 𝒟X, we will define 𝒟X-modules.

Definition 1.0.4.

A left 𝒟X-module is a quasicoherent sheaf with a left 𝒟X-action.

Example 1.0.5.

Clearly 𝒟X and 𝒪X are 𝒟X-modules. On the other hand, if kx is the skyscraper sheaf at x∈X, then kx⊗𝒟X is a right 𝒟X-module.

Proposition 1.0.6.

Let ℱ,ℱ′∈𝒟X−𝖬𝗈𝖽. Then

  • •

    ℱ⊗ℱ′ is a left 𝒟X-module with the formula

  • •

    H⁢o⁢m¯𝒪X⁢(ℱ,ℱ′) is a left 𝒟X-module by

    The analogous statement is true for right modules.

  • •

    If 𝒢 is a right 𝒟X-module, then 𝒢⊗ℱ is a right module by

  • •

    Hom¯𝒪X⁢(ℱ,𝒢) is a right module by

Recall that 𝒯X acts on ΩXn by the Lie derivative. Then det⁡ΩX admits a right 𝒟X-module structure by w⋅ξ=−Lieξ⁡ω.

Definition 1.0.7.

Define the right 𝒟X-module Ω𝒟X by

Now if ℱ is a left 𝒟X-module, then we see that

is a right 𝒟X-module. Now if 𝒢 is a right 𝒟X-module, then

We will call this sheaf Ω𝒟X†.

Proposition 1.0.8.

The functors detΩX⊗𝒪X− and Hom¯𝒪X⁢(det⁡ΩX,−) give an equivalence of categories between 𝒟X−𝖬𝗈𝖽 and 𝖬𝗈𝖽−𝒟X.

Proof.

We will prove an adjunction

This will be defined by

Here, we note that f⁡(ξ⁢ω⊗v)−f⁡(ω⊗ξ⁢v)=f⁡((ω⊗v)⁢ξ), so we have the adjunction. We only need to check that the unit and counit are isomorphisms now. Here, we have

and the unit is similar. □

2 Fan (Feb 16): D-modules, bare minimumz, part 2

Our goal is the following underived theorem:

Theorem 2.0.1 (Kashiwara).

Let ι:X→Y be a closed immersion. Then there exist functors

giving an equivalence of categories.

Definition 2.0.2.

Let f:X→Y be a morphism. Then the 𝒟-module pullback f∗ is defined by

with the 𝒟X-action given by

2.1 Transfer modules

Definition 2.1.1.

Define the transfer module 𝒟X→Y≔f∗⁢𝒟Y.

Now that we have this definition, it is clear that f∗⁢ℱ=𝒟X→Y⊗f−1⁢𝒟Yf−1⁢ℱ.

Definition 2.1.2.

Define the functor f+:𝒟X−𝖬𝗈𝖽→𝒟Y−𝖬𝗈𝖽 by

After symbol pushing, we obtain

This motivates the following definition:

Definition 2.1.3.

Define the other transfer bimodule by

Now we have f+⁢ℱ=f∗⁢(𝒟Y←X⊗𝒟X⁢ℱ).

If ι:X→Y is a closed immersion, then ι+ is right exact. We hope that an adjoint exists, and in fact we will define a candidate.

Definition 2.1.4.

Define the functor ι+:𝒟Y−𝖬𝗈𝖽→𝒟X−𝖬𝗈𝖽 by

Example 2.1.5.

Let {xi,∂i} be a local étale coordinate system on X. Then write ξ⁡(x,∂)=∑αpα⁢(x)⁢∂α. Now we define

This is an anti-homomorphism of 𝒟X. Defining a right action on Ω𝒟X⊗ℱ by v⁢ξ=ξ+⁢v, this is the left-right flip.

Example 2.1.6.

Now let {yi,∂i} be a coordinate system on Y and suppose X is cut out by ym=⋯=yn=0. Then we have

On the other hand, we have 𝒟Y←X=ℂ⁡[∂m+1,…,∂m]⊗ℂ𝒟X.

2.2 Proof of Kashiwara

We will first prove an adjunction. In fact, we will prove the adjunction on the level of sheaf Homs. But this reduces to

There is a natural morphism coming from 𝒢→ι∗⁢ι−1⁢𝒢, so we only need to check locally that this is an isomorphism. But now the left hand side becomes

where ΓX⁢𝒢 is the subsheaf with supports in X (note notation is different from Hartshorne).

We will now prove that the units and counits are isomorphisms. We can do this affine locally, so we will choose an étale coordinate system {yi,∂i} on Y. We will assume that X=V⁡(yn+1,…,ym), and in particular we can reduce to the case where X=V⁡(ym). We will simply write y≔ym for simplicity. First, consider

This is determined by f⁡(1⊗1) such that y⁢f⁢(1⊗1)=0. Write f⁡(1⊗1)=∑k∂k⊗vk. Because [∂,y]=1, we have [y,∂k]=−k⁢∂k−1 and [∂,yk]=k⁢yk−1. Now we have

This is nonzero unless vk=0 for all k>0, so f⁡(1⊗1)=1⊗v0.

Now consider

Locally, the left hand side is simply ι∗⁢(ℂ⁡[∂]⊗ℂHom¯ι−1⁢𝒟Y⁢(ℂ⁡[∂]⊗ℂ𝒟X,ι−1⁢ℱ)). Consider the operator y⁢∂:ℱ→ℱ. Then define ℱk=ker(y∂−⋅k). We now have

and

so we see that y,∂ act like shift operators. Now define ℱk=ker⁡(yk)⊂ℱ. Then we will show that ℱk⊆ℱ−1⊕⋯⊕ℱ−k. To show this, we will use induction. If v∈ker⁡y, then y⁢∂v=−v, so v∈ℱ−1. For the inductive step, we note that if v∈ℱk, then

Thus ∂y⁢v∈ℱ−2⊕⋯⊕ℱ−k. Then

so y⁢∂v+k⁢v∈ℱ−1⊕⋯⊕ℱ−(k−1), and thus

Because ℱ is supported on X, we have ℱ=⋃k=1∞ℱk⊆ℱ−1⊕⋯, and thus

But now f∈Hom¯ι−1⁢𝒟Y⁢(ℂ⁡[∂]⊗ℂ𝒟X,ι−1⁢(ℂ⁡[∂]⊗ℂℱ−1)) is determined by f⁡(1⊗1) and must be killed by y, so Hom¯=ℱ−1.

3 Fan (Mar 02): D-modules, bare minimumz, part 3

Recall that we have an underived version of Kashiwara. We will prove that

Proposition 3.0.1.

If a 𝒟X-module is coherent over 𝒪X, then it is locally free over 𝒪X.

Also, we will prove

Proposition 3.0.2.

Let X⊂Y be a smooth closed subvariety. Then there exists a functor ι† such that ι†=R⁢ι+.

Finally, we will prove

Theorem 3.0.3.

There exists a functor i⋆:Db⁢(𝒟X−𝖬𝗈𝖽)→Db⁢(𝒟Y−𝖬𝗈𝖽X) such that ι†,ι⋆ give an equivalence of categories.

3.1 Proof of first proposition

Note that local freeness can be checked at stalks. Also, flatness can be checked at stalks. Over local rings, flat is equivalent to free. Finally, for finite modules over a reduced Noetherian ring, flatness is the same as the rank of M⊗k⁡(𝔭) being locally constant. Then, any two closed points on a smooth variety can be connected by (a chain of) smooth curves. Therefore, we can reduce to the case when X is a smooth curve.

First, we will show that ℱ is torsion-free. If ℱ has torsion at x, then there exists 𝒢⊂ℱ supported at x, so by Kashiwara, there exists ℋ∈𝒟x−𝖬𝗈𝖽 such that

which is not coherent. But now ℱ is not coherent. Now because X is smooth, 𝒪X,x is a DVR, and in this case, being torsion free is the same as being free.

3.2 D-modules on singular varieties

Now suppose X is singular and affine. Choose X↪Y=𝔸n, and then define 𝒟X−𝖬𝗈𝖽 to be 𝒟Y−𝖬𝗈𝖽X, and this automatically satisfies Kashiwara. This is compatible with diagrams of the form

We will not consider singular things, so we will not need this in the future.

3.3 Construction of functors

We will define the functors we want to study. First define f†:Db⁢(𝒟Y−𝖬𝗈𝖽)→Db⁢(𝒟X−𝖬𝗈𝖽) by

where the shift apparently makes the Riemann-Hilbert correspondence more convenient.

On the other side, define f⋆:Db⁢(𝒟X−𝖬𝗈𝖽)→Db⁢(𝒟Y−𝖬𝗈𝖽) by

Theorem 3.3.1 (Bernstein).

Let X be separated and Noetherian. If ℛ is a sheaf of 𝒪X-algebras, quasicoherent as an 𝒪X-module. Let Dq⁢cb⁢(ℛ−𝖬𝗈𝖽) be those with quasicoherent cohomology. Then

is induced by the inclusion of quasicoherent modules into all modules.

Example 3.3.2.

If j:X→Y is an open embedding, then j∗ is exact and 𝒟X→Y is precisely 𝒟X, so in fact j†=j−1. Also,

and therefore j+=j∗(𝒟Y←X⊗−)=j∗, and thus j⋆=R⁢j∗.

Example 3.3.3.

Let ι:X→Y be a closed embedding. Choose coordinates yi,∂i of Y such that X is cut out by yn+1=⋯=ym=0. Then we know

wnere 0→Km−n→⋯→K0→𝒪X→0 is a resolution of ι−1⁢𝒪Y-modules. Note that

and that

Applying −⊗ι−1𝒟Y, we obtain a resolution

of locally free ι−1⁢𝒟Y-modules. Therefore we have

On the other hand, 𝒟Y←X is locally free as a right 𝒟X-modules, so ι⋆=ι+.

3.4 Proof of second proposition

First, we will show that L⁢ι∗⁢𝒢⁢[n−m]=R⁢Hom¯ι−1⁢𝒟Y⁢(𝒟Y←X,ι−1⁢𝒢). But it suffices to see that 𝒟X→Y⁢[n−m]=R⁢Hom¯ι−1⁢𝒟Y⁢(𝒟Y←X,ι−1⁢𝒟Y). Now it suffices to show that

Note that Km−n=ι−1⁢det⁡ΩY has rank 1, so there exists a canonical pairing Kj⊗Km−n−j→Km−n such that

Therefore

Comparing this to

we have the desired result.

Next, we will show that R⁢Hom¯ι−1⁢𝒟Y⁢(𝒟Y←X,ι−1−)≅R⁢ι+ in Db⁢(𝖬𝗈𝖽⁡(Y)). Recall from the proof of Kashiwara that

We will prove the derived version of this. Note that ι−1⁢ΓX is exact, so it sends injectives to injectives. We now claim that the map

induced from R⁢ΓX⁢𝒢→𝒢 is an isomorphism. This is by an exact triangle

It suffices to show that R⁢Hom¯ι−1⁢𝒟Y⁢(𝒟Y←X,ι−1⁢R⁢j∗⁢j−1⁢𝒢)=0. But this is because

By vanishing of j−1⁢ι∗⁢𝒪X=0, we are done.

3.5 Proof of Theorem

Finally, the proof of the derived version of Kashiwara follows from the underived version by an inductive argument using truncation functors, which is not reproduced here.1

4 Fan (Mar 09): brief interlude: refresher on algebraic groups

We will call G⁢Ln the general linear group, Bn the upper triangular matrices, Un the nilpotent part, and Tn the maximal torus.

Example 4.0.1.

Here are some examples of group schemes. First, we have 𝔾m=Spec⁡ℂ⁡[t,t−1], whose functor of points is A↦A×. Analogously, we have G⁢Ln=Spec⁡ℂ⁡[ti⁢j,det−1] which has functor of points A↦G⁢Ln⁢(A).

Proposition 4.0.2 (Cartier).

In characteristic 0, all algebraic groups are smooth.

Proposition 4.0.3.

For algebraic groups, smooth is the same as geometrically reduced.

It should be clear what it means to be a sub-group scheme and a normal sub-group scheme. In fact, if H⊆G is an algebraic subgroup, the inclusion H→G is a closed immersion.

Proposition 4.0.4.

For algebraic groups, connected is the same as being irreducible.

For a homomorphism of algebraic groups G→H, we have a kernel. Then

is exact if A maps isomorphically to the kernel of B→C and B→C is faithfully flat. In fact, any G→H factors as G→image↪H into a faithfully flat map followed by a closed immersion. As one would expect, we can construct stabilizers of Y⊆X when G acts on X. As a special case of this, we can construct normalizers and centralizers of subgroups.

Example 4.0.5.

Some other classical examples of algebraic groups are given by

  • •

    S⁢Ln=Spec⁡ℂ⁡[ti⁢j]/(det−1);

  • •

    S⁢O2⁢n+1=A↦{g∈S⁢L2⁢n+1⁢(A)∣gT⁢(1InIn)⁢g=(1InIn)}.

  • •

    S⁢O2⁢n=A↦{g∈S⁢L2⁢n⁢(A)∣gT⁢(InIn)⁢g=(InIn)}.

  • •

    S⁢p2⁢n=A↦{g∈S⁢L2⁢n⁢(A)∣gT⁢(−InIn)⁢g=(In−In)}.

Of course tori are defined to be 𝔾mn.

Definition 4.0.6.

Suppose G,H are smooth and connected with φ:G→H. Then φ is an isogeny if ker⁡φ is a finite algebraic group. In general G and H are isogenous if they are connected by a zigzag of isogenies.

Theorem 4.0.7.

An algebraic group G is affine if and only if G is a closed subscheme of G⁢Ln.

We will define G⁢LV to be the functor A↦AutA⁡(V⊗kA), and a representation is a morphism ρ:G→G⁢LV.

Definition 4.0.8.

Let H⊆G be a subgroup. Then f:G→X is H-invariant if G×H→π1G→𝑓X is the same as G×H→𝜇G→𝑓X. Then f:G→X is a quotient of G by H if f is faithfully flat, f is H-invariant, and if

is an isomorphism.

All quotients G/H exist and are quasiprojective varieties. If H is a normal subgroup of G, then G/H is also an algebraic groups. If G,H are affine and H is a normal subgroup, then G/H is also affine. Note tha this fails for H not normal, because G/B is always projective.

Definition 4.0.9.

A filtration or a subnormal series is a sequence

A composition series is a filtration where dimG0>dimG1>⋯ For composition series, the quotients are unique up to reordering and isogeny.

Definition 4.0.10.

Define D⁢G=[G,G]. Then the derived series of G is G⊇D⁢G⊇D2⁢G⊇⋯.

Definition 4.0.11.

A subgroup G is solvable if Dn⁢G=1. Equivalently, there exists a filtration with abelian quotients.

Definition 4.0.12.

A group G is unipotent if every representaiton has a fixed vector. Equivalently, G is isomorphic to a subgroup of Un.

Definition 4.0.13.

Let G be a smooth connected linear algebraic group. Then the radical Rad⁡G is the maximal smooth connected solvable normal subgroup. In addition, the unipotent radical is the smallest unipotent normal subgroup. G is semisimple if Rad⁡G=1 and G is reductive if Radu⁡G=1.

Theorem 4.0.14 (Jordan decomposition).

For g∈G⁡(k), there exists a unique g=gs⁢s⁢gu=gu⁢gs⁢s such that

  1. 1.

    If G=G⁢Ln, then gs⁢s is semisimple (diagonalizable after field extension) and (gu−1)n=0 for some n.

  2. 2.

    This decomposition is functorial under G→H.

Definition 4.0.15.

A group G is diagonalizable if G is isomorphic to a subgroup of Tn. Equivalently, 𝒪G⁢(G) is spanned by a such that Δ⁡(a)=a⊗a (grouplike elements).

Proposition 4.0.16.

There exists an equivalence of categories between the opposite category of finitely-generated ℤ-modules and diagonalizable algebraic groups given by M↦D⁡(M)=(A↦Hom⁡(M,A×)) and G↦χ∗⁢(G)=Hom⁡(G,𝔾m).

Definition 4.0.17.

A group G is of multiplicative type if Gk¯=D⁢(M)k¯ for some finitely generated abelian group M.

Having defined the characters χ∗⁢(G), we can define the cocharacters

Definition 4.0.18.

Let f:𝔾m→X. If there is an extension f̂:𝔸1→X, then we can define limt→0f⁡(t)=f~⁢(0). For λ:𝔾m→G, we have an action of 𝔾m on G by conjugation. Then we can define

These are all subgroups of G.

Note that an action of 𝒢m on G induces a decomposition

Then U⁡(λ) corresponds to positive weights, P⁡(λ) corresponds to nonnegative weights, and Z⁡(λ) corresponds to the weight 0 subspace.

Proposition 4.0.19.

In characteristic 0, unipotent algebraic groups are the same as finite-dimensional nilpotent Lie algebras.

Definition 4.0.20.

A group G is trigonalizable if G is isomorphic to a subgroup of Bn. Equivalently, every irreducible representation is finite-dimensional. Also, this is the same as G being an extension of a diagonalizable group by a unipotent group.

Theorem 4.0.21 (Lie-Kolchin).

(Split) solvable implies trigonalizable. In fact, for smooth connected groups over an algebraically closed fields, the two conditions are equivalent.

Definition 4.0.22.

A Borel subgroup of a group G is a maximal smooth connected solvable subgroup.

Borels are minimal parabolic subgroups. We then have G⊃B=NG⁢B⊃CG⁢T⊃T. We also have N=NG⁢T and B∩N=T. We define NG⁢T/T to be the Weyl group. Also, all pairs (B,T) of a Borel and a maximal torus are conjugate to each other. For G reductive, we have

The quotient G/Z⁢G is semisimple. For example, G⁢Ln/𝔾m=P⁢G⁢Ln, which is isogenous to S⁢Ln.

5 Fan (Mar 23): localization: d-modules and representationZ

We will warn that the things that we say may be slightly wrong because every source uses a different convention. Here, localization will take the form

where the 𝒟-modules live on G/B and everything is twisted in some way. The three ways to get the 𝒟 are by:

  1. 1.

    Using line bundles to twist 𝒟X;

  2. 2.

    Doing something like (𝒟G/⟨λ−ξ⁡(λ)⟩)B;

  3. 3.

    Taking some quotient of 𝒰⁢𝔤~ as a Lie algebroid.

All of these strategies are the same.

Definition 5.0.1.

A sheaf ℱ on X is G-equivariant if, considering act,π2:G×X→X, we equip ℱ with an isomorphism

such that

  1. 1.

    φℱ|1×X=id;

  2. 2.

    We have the commutative diagram

    on G×G×X.

5.1 Universal enveloping algebras

Recall that a central character ϑ:Z⁢𝔤→ℂ is the same as a W-orbit (by the dot action) on 𝔥∗ by Harish-Chandra. Then we define

Now let G(n)=Spec⁡𝒪⁡(G)/𝔪1n+1 be the n-th infinitesimal neighborhood of the identity. Then for an open set U⊂X, we obtain a sequence of morphisms

This is the same as the data of a map

Therefore, we have a morphism aℱ:𝒰⁢𝔤→Endℂ⁡ℱ⁡(U). This satisfies for all ξ∈𝔤,ϕ∈𝒪X,v∈ℱ the identity

and in particular a𝒪X:𝔤→𝒯X⁢(U).

Definition 5.1.1.

A Lie algebroid 𝔤~ is a sheaf isomorphic to 𝒪X⊗ℂ𝔤 extending 𝔤 such that

The corresponding construction for the universal enveloping algebra is 𝒰⁢𝔤~=𝒪X⊗ℂ𝒰⁢𝔤 with a product extending that of 𝒰⁢𝔤 such that

Now aℱ extends to a~ℱ:𝒰⁢𝔤~→End𝒪X⁡ℱ. We can also write Z⁢𝔤~=𝒪X⊗ℂZ⁢𝔤. Here, we have

Then for X=G/B, points of G/B are the same as choices of a Borel subgroup B⊂G. We can similarly define 𝔫~=[𝔟~,𝔟~]. For all λ∈𝔥∗, we have

In particular, we have a ϱ~.

Now we may define

This is in fact a two-sided ideal, and so we define

In this perspective, we have a natural projection 𝒰⁢𝔤~↠𝒟Xλ and a natural sequence 𝒰⁢𝔤↪Γ⁡(X,𝒪X)⊗𝒰⁢𝔤→Γ⁡(X,𝒟Xλ).

5.2 Line bundles

Let V be a module over B. Then define ℒ⁡(V) on X=G/B by

Then we have a projection π:G×BV↠G/B. When V is actually a representation of G, then ℒ⁡(V)≅𝒪X⊗ℂV. Then for a character λ, we can define

Definition 5.2.1.

An ℒ-twisted differential operator of order ≤n is D:ℒ→ℒ such that for any ϕi∈𝒪X, we have [ϕn,…,[ϕ0,𝒟]]=0. The sheaf of such twisted operators is called 𝒟X,ℒ.

Proposition 5.2.2.

For any λ, we have 𝒟Xλ≅𝒟X,ℒλ−ϱ.

Theorem 5.2.3 (Beilinson-Bernstein localization).

There are equivalences of categories

whenever λ is regular and dominant.

5.3 Taking B-invariants

Definition 5.3.1.

A weakly B-equivariant 𝒟-module is a 𝒟-module that is B-equivariant as a quasicoherent sheaf such that the morphism

is B-equivariant.

Consider the map a𝒪:𝔟→End⁡(𝒪⁡(−))⊂𝒟X, which acts on ℱ. Then we define aℱ♮≔aℱ−a𝒪. This is in fact a Lie algebra homomorphism to End⁡(ℱ).

Definition 5.3.2.

A 𝒟-module is strongly B-equivariant if aℱ♮=0.

We now define the sheaf 𝒟Gλ,𝔟≔𝒟G/𝒟G⁢⟨ξ−λ⁡(ξ)⟩, which is weakly B-equivariant and has a♮=−λ. We will denote the category of weakly B-equivariant 𝒟-modules with a♮=−λ by 𝒟G−𝖬𝗈𝖽B,λ. To relate all of the notions we have defined so far, we have

In this situation, localization becomes

Theorem 5.3.3.

We have an equivalence of categories

6 Fan (Mar 30): proof of localization: d-modules and representationz, part 1

6.1 An example with S⁢L2

Let V be a module over B and consider π:G→G/B=X. We will see that

Here, ℒ⁡(λ) is the map B→ℂ× with weight λ. We will consider the case when G=S⁢L2 and thus G/B=ℙ1.

If we set λ=n⁢ρ, then we see that

But this implies that ϕ∈x−n⁢ℂ⁢[y,x], and a similar result is true for D⁡(y). Therefore,

Doing more computations, we note that the action of ∂z acts on D⁡(x⁢y), and in fact we obtain ∂z=−w2⁢∂w+n⁢w. We also have an action of G on 𝒟X,ℒ⁡(λ). On the chart D⁡(x), we have

and we have ϕ=1xn⁢P⁢(z) on D⁡(x) and 1yn⁢Q⁢(w) on D⁡(y). For the element e∈𝔰⁢𝔩2, we note that

on D⁡(x) and −∂wϕ⁡(w) on D⁡(y). Therefore we have

Now we know that ℒ⁡(−n⁢ρ)=𝒪⁡(−n)∈𝒟ℙ1−n⁢ρ−𝖬𝗈𝖽. Taking global sections, consider the basis vi=(−1)i⁢wi. Then we see that e⁢vi=i⁢vn−i,f⁢vi=(n−i)⁢vi+1,h⁢vi=(n−2⁢i)⁢vi, and this is the representation Ln⁢ρ.

6.2 Main results

There are a few key results:

Theorem 6.2.1.

We have the identity

Theorem 6.2.2.
  1. 1.

    If λ is ρ-antidominant, then

    is exact.

  2. 2.

    If λ is ρ-regular, then Γ is faithful.

Theorem 6.2.3.

For the action of G on X, then

The logical equivalences are that the first two theorems imply localization and that the third implies the first. Also, Kostant implies the third theorem.

6.3 Proof of localization

Let 𝒜 be a quasicoherent sheaf of algebras on X and A=Γ⁡(X,𝒜). Then 𝒜⊗A− and Γ⁡(X,−) give an equivalence between 𝒜-modules and A-modules. It is clear that ηM is an isomorphism to all free M because Γ⁡(X,𝒜⊗AA)=A. But then for any M, if we consider a presentation

and note that ηPi are isomorphisms and the sequence stays exact after applying Γ(𝒜⊗A−), we see that ηM is an isomorphism.

To prove that the counit is an isomorphism, note that Γ is conservative, so it suffices to show that Γ⁡(εℱ) is an isomorphism. But here we have

so εℱ is an isomorphism.

6.4 The technical results

Next, note that if M is a 𝒰⁢𝔤/ϑ-module for ϑ≠ϑλ, then 𝖫𝗈𝖼λ+2⁢ρ⁢M=𝒟Xλ+2⁢ρ⊗M=0 because

Proposition 6.4.1.

The map pλ+2⁢ρ:𝒰⁢𝔤→Γ⁡(X,𝒟Xλ+2⁢ρ) factors through 𝒰⁢𝔤/ϑλ.

Proof.

Let 𝒥λ≔⟨z−ϑλ⁢(z)∣z∈Z⁢𝔤~⟩⊂𝒰⁢𝔤. It suffices to show that

is the zero map. For x∈X, note that

The action of Z⁢𝔤 on this is actually ϑλ.

By the discussion from last semester, in the diagram

χλ=λ⁡(ψH⁢C⁢(z)) is independent of the choice of Borel. Also, χλ=ϑλ−ρ because Z⁢𝔤 acts on 𝒰⁢𝔤⊗ℂλ by ϑλ, on 𝒰⁢𝔤⊗𝔟xℂ⁢λ by χλ+ρx, and therefore, taking the transpose, 𝒰⁢𝔤 acts on ℂλ⊗𝔟x𝒰⁢𝔤, and Z⁢𝔤 acts by χλ−ρx=ϑλ−2⁢ρx. □

Lemma 6.4.2.

The inclusion Z⁢𝔤↪𝒰⁢𝔤 induces an isomorphism Gr⁡Z⁢𝔤≅(S∙⁢𝔤)G.

Third theorem implies first theorem.

Note that 𝒰⁢𝔤 inherits the filtration from 𝒰⁢𝔤 and consider the map pλ:𝒰⁢𝔤/ϑλ→Γ⁡(X,𝒟Xλ+2⁢ρ). It suffices to show that the graded version of this is an isomorphism. Here we have a sequence of maps

whose total composition is an isomorphism. Here, the second map is Gr⁡pλ, the third map comes from something like

and using left exactness of Γ. Finally, S∙⁢𝔤→Gr⁡𝒰⁢𝔤→Gr⁡𝒰⁢𝔤/ϑλ factors through (Gr⁡Z⁢𝔤)+. □

7 Fan (Apr 06): proof of localization: d-modules and representationz, part 2

7.1 Comments on equivariant things

Let B act on Y. Recall the definitions of weakly B-equivariant and strongly B-equivariant 𝒟-modules.

Proposition 7.1.1.

Suppose ϖ:Y→X is a principal B-bundle. Then there is an equivalence between 𝒟X-modules and strongly B-equivariant 𝒟)⁢Y-modules given by ℱ↦ϖ∗⁢ℱ and 𝒢↦𝒢B.

Proposition 7.1.2.

We have an equivalence 𝒟Xλ−𝖬𝗈𝖽≅𝒟Y−𝖬𝗈𝖽B,a♮=−λ.

Proof.

In general, if ℱ is a quasicoherent sheaf on X, ϖ∗⁢ℱ is B-equivariant because the diagram

commutes. We also need to check compatibility with the 𝒟Y-action, which means that

commutes. Here, we are assuming that Y is locally X×B in the Zariski topology (this is true in the cases we are interested in), so ϖ∗⁢ℱ=ℱ⊠𝒪B. Now ℱ locally has trivial B-equivariant structure and 𝒪B is strongly B-equivariant by definition, so ϖ∗⁢ℱ is also strongly equivariant.

Next, it is clear that

This implies that our assignment is fully faithful. Now for a strongly equivariant 𝒟Y-module 𝒢, we can write 𝒢=ℱ⊠𝒪B for some ℱ on X. Finally, we check that

7.2 Proof of Theorem 6.2.2

First we will prove exactness. It suffices to show that

is exact. However, we identify the first term with 𝒟G−𝖬𝗈𝖽B,a♮=−(λ+2⁢ρ), and then global sections take us to 𝔤−𝖬𝗈𝖽. This lands inside a non-finitely generated version of category 𝒪, and then we take M↦(M𝔫)−(λ+2⁢ρ) to get a vector space. But now

This is exact exactly when M−(λ+2⁢ρ) is a projective Verma module, which follows from our assumptions on λ.

Definition 7.2.1.

A sheaf of twisted differential operators is a sheaf of 𝒪X-algebras with a filtration such that

  1. 1.

    𝒪X↪𝒟 induces 𝒪X≅𝒟≤0;

  2. 2.

    The natural map S𝒪X⁢𝒟≤1/𝒟≤0→Gr⁡𝒟 is an isomorphism;

  3. 3.

    The map 𝒟≤1/𝒟≤0→𝒯X given by ξ↦(ϕ↦ξ⁢ϕ−ϕ⁢ξ) is an isomorphism.

Note here that 𝒟Xλ=𝒰⁢𝔤~/⟨𝔟~−λ⁡(𝔟~)⟩ is a sheaf of twisted differential operators. This is because the image of ℐλ in 𝒰⁢𝔤~≤1/𝒰⁢𝔤~≤0 is 𝔟~, so we get an exact sequence

Next, the image of ℐλ in 𝒰⁢𝔤≤i/𝒰⁢𝔤≤i−1 is 𝔟~⋅S𝒪Xi⁢𝔤~, so we have

The next fact is that for any λ∈Λ, 𝒟Xλ=𝒟X,ℒ⁡(λ). To see this note that ℒ⁡(λ) is G-equivariant, so the map

factors through ℐλ because 𝔟 acts on ℒ⁡(λ) by eλ for any x∈X.

Finally, we will accept as fact that G=⨆w∈WB⁢w⁢B, that X=⨆w∈WB⁢w⁢B/B, and if B⁢w⁢B/B≕Xw, then

Let ιw:Xw↪X. Here, we have

Now we are ready to prove faithfulness. Let ℱ∈𝒟Xλ+2⁢ρ−𝖬𝗈𝖽. We want to show that Γ⁡(X,ℱ)≠0. Let Zk=X¯k=⨆ℓ⁡(w)≤kXw, where Xk=⨆ℓ⁡(w)=kXw. Now if ℓ is the maximum length in the Weyl group, consider

consider the filtration

Now if ji:X∖Zk↪X is the inclusion, we have an exact triangle

which gives us an exact sequence

and in particular, we obtain

The upshot is that we have a filtration of Γ⁡(X,ℱ) such that

Now we have a spectral sequence

But now the E1 page is actually

8 Fan (Apr 13): some more on D-modules and identifying some highest-weight modules in the setup

We will discuss important notions like coherence, holonomicity, duality, and then identify the 𝒟X-modules corresponding to important modules in representation theory under localization.

8.1 Coherence and holonomicity

Definition 8.1.1.

A filtration for ℱ∈𝒟X−𝖬𝗈𝖽 is compatible with 𝒟X if

Definition 8.1.2.

A good filtration of ℱ satisfies one of the following conditions:

  • •

    Gr⁡ℱ is coherent as a 𝒪ΩX-module;

  • •

    ℱ≤i is a coherent 𝒪X-module for all i and 𝒟X≤1⋅ℱ≤i=ℱ≤i+1 for all sufficiently large i.

Definition 8.1.3.

Define the singular support Supps⁡ℱ≔Supp⁡Gr⁡ℱ⊂ΩX, where we take any good filtration on ℱ. This is in fact independent of the good filtration.

Proposition 8.1.4.

A 𝒟X-module is 𝒟X-coherent if and only if it has a good filtration.

Theorem 8.1.5 (Bernstein’s inequality).

Let ℱ be a nonzero coherent 𝒟X-module. Then

Proposition 8.1.6.

Let i:X↪Y be a closed embedding. For every ℱ∈𝖢𝗈𝗁⁡(𝒟X), i+⁢ℱ∈𝖢𝗈𝗁⁡(𝒟Y). In addition,

If j:X→Y is an open embedding and 𝒢∈𝖢𝗈𝗁⁡(𝒟Y), j∗⁢𝒢∈𝖢𝗈𝗁⁡(𝒟X). In addition,

Another fact is that if 𝒢⊂ℱ, in fact Supps⁡ℱ=Supps⁡𝒢∪Supps⁡ℱ/𝒢.

Definition 8.1.7.

ℱ is holonomic if the inequality in Bernstein’s inequality is in fact an equality, that is if

Define Dhol⁢(𝒟X−𝖬𝗈𝖽) and Dcoh⁢(𝒟X−𝖬𝗈𝖽) to be the categories with holonomic and coherent cohomology, respectively.

Theorem 8.1.8.

Let f:X→Y. Then f!,f⋆ preserve Dhol.

Now consider the duality functor 𝔻X:Dcoh⁢(𝖣X−𝖬𝗈𝖽)op→Dcoh⁢(𝒟X−𝖬𝗈𝖽) given by

Fortunately, we can compute this by taking a projective resolution, so

Theorem 8.1.9.

If ℱ is holonomic, then 𝔻X is concentrated in degree 0, and so 𝔻X is a duality functor on holonomic 𝒟X-modules.

This allows us to construct a six functors formalism for 𝒟-modules. Define the pushforward with compact support

and the pullback

Theorem 8.1.10.

This, combined with the existing f⋆,f!, form a six functors formalism. In addition, the natural map f!→f⋆ is an isomorphism if f is proper and f!=f∗[2(dimX−dimY)] if f is smooth.

8.2 Representation theory

For any subgroup K⊆G, define a (𝔤,K)-module to be a module where the action of 𝔨 comes from differentiating an action of K. Then we obtain

Sticking to ϑ0, define Mw≔Mw∘(−2⁢ρ) and Lw≔Lw∘(−2⁢ρ). Then denote their images under localization by ℳw,ℒw.

Now consider the inclusion ιw:Xw↪X. Define

Lemma 8.2.1.
  1. 1.

    χΓ⁡(X,𝒩w)=χMw;

  2. 2.

    There is no nontrivial submodule of 𝒩w supported on X¯w∖Xw.

Proof.

Define

These give algebraic groups N1,N2. For w∈W=NG⁢(H)/H, choose a lift w~∈NG⁢(H) of w and define

Note that φ⁡(id,N2)=X2=B⁢w⁢B/B. We now have a diagram

Then we compute

We then have

because

so we obtain

If we let H act on N1×N2 by h⁡(n1,n2)=(h⁢n1⁢h−1,h⁢n2⁢h−1), this becomes an isomorphism of H-representations, and so we can compute the characters. First, we have

Next, we obtain

The action of 𝔥 on det⁡𝔫1 has weight

so χdet⁡𝔫1=e−w⁢ρ−ρ=ew∘(−2⁢ρ). In total, the character is

We now prove the second part of the lemma. Let j:Imφ↪X and ι⁢Z=:X∖Imφ↪X. We have an exact triangle

and because j∗⁢j∗⁢𝒩w=𝒩w, we see that ΓZ⁢𝒩w=0. □

Proposition 8.2.2.

We have

Proof.

Induct on ℓ⁡(w). First, for [M]=[Mw†], if Hom⁡(Lw′,M)=0 for all ℓ⁡(w′)<ℓ⁡(w), then M=Mw†. To see this, consider the map Mw→M†. If this is not surjective, then there exists a morphism M†→Lw′ and thus Lw′ is a Jordan-Holder factor of M. But then Lw′ is a Jordan-Holder factor of Mw, so ℓ⁡(w′)<ℓ⁡(w). this gives a nonzero map Lw′→M.

By the inductive hypothesis, Lw′⊂Mw′⟂, so ℒw′⊂𝒩w′ for all ℓ⁡(w′)<ℓ⁡(w). Because ℒw′ is supported on X¯w′, which does not intersect Xw, by adjunction we see that Hom⁡(ℱ,ιw⁣∗⁢𝒪Xw)=0 for all ℱ supported inside X¯w′. Applying global sections, we see that Hom⁡(Lw′,M)=0. □

One can also show that ℒw is the image under ι!→ι⋆ of 𝒪Xw. This is very difficult and uses the Riemann-Hilbert correspondence. Also, 𝔻X⁢𝒩w=ℳw.

9 Kevin (Apr 20): the Riemann-Hilbert correspondence

Note: these are the speaker’s notes. Minor modifications were made to adapt to my typographical style choices. Some references include Hotta-Takeuchi-Tanisaki (D-Modules, Perverse Sheaves, and Representation Theory, Part I), Dimca (Sheaves in Topology, Chapters 4-5), and Borel (Algebraic D-modules).

9.1 Review of constructible sheaves

Since Caleb’s talk last Friday covered this material and (presumably) most of you were there, I’ll briefly review constructible sheaves and the six functor formalism.

Let X be a complex variety. A stratification of X is a locally finite partition X=⨆jXj, where each Xj is a smooth connected locally closed subvariety such that each X¯j is a union of strata. A ℂ-sheaf ℱ on X is constructible (with respect to this stratification) if ℱ|Xj is a local system with finite-dimensional fibers for each j. A complex ℱ∙∈Db⁢(X)≔Db⁢(X,ℂ) is constructible (with respect to this stratification) if each cohomology sheaf ℋi⁢(ℱ∙) is. We use Dcb⁢(X) to denote the full subcategory of Db⁢(X) consisting of constructible complexes.

Why consider constructible sheaves? Local systems are all over the place, but they aren’t well-behaved under the usual sheaf operations. Constructible sheaves are. Here are the six operations we can perform (let’s say we have a map f:X→Y). All the operations are derived, so we will omit the R and L for the most part.

  • •

    f∗:Dcb⁢(X)→Dcb⁢(Y)

  • •

    f∗:Dcb⁢(Y)→Dcb⁢(X)

  • •

    f!:Dcb(X)→Dcb(Y)

  • •

    f!:Dcb(Y)→Dcb(X)

  • •

    ⊗L:Dcb⁢(X)×Dcb⁢(X)→Dcb⁢(X)

  • •

    R⁢Hom¯:Dcb⁢(X)op×Dcb⁢(X)→Dcb⁢(X)

We have adjunctions (f∗,f∗),(f!,f!),(−⊗Lℱ∙,RHom¯(ℱ∙,−)). Moreover, there is a Verdier duality functor 𝔻=R⁢Hom¯⁢(−,ωX):Dcb⁢(X)op→≅Dcb⁢(X) such that 𝔻2≅id. Here, ωX is the dualizing sheaf, which is ℂX⁢[2⁢d] (d is the dimension of X) when X is smooth. 𝔻 intertwines the functors: f!=𝔻f∗𝔻, f!=𝔻f∗𝔻.

9.2 The Riemann-Hilbert correspondence for integrable connections

For the remainder of this talk, all varieties will be smooth, since we want to talk about D-modules.

Let X be a smooth variety. The Riemann-Hilbert correspondence (for varieties) is an equivalence Drhb⁢(𝒟X)→≅Dcb⁢(X) (here, r⁢h stands for regular holonomic; I will define this shortly). To motivate the general correspondence, we will first discuss the simplest case: integrable connections. Let (ℰ,∇) be an integrable connection, i.e. a vector bundle with a flat connection. We can get a local system whose sections are horizontal sections of E, i.e. sections s for which ∇s=0. This produces an exact functor from the category of integrable connections on X to the category of local systems on X: 𝖢𝗈𝗇𝗇⁡(X)→𝖫𝗈𝖼⁡(X).

Is this functor an equivalence? If we consider the analytic analogue (we consider flat holomorphic connections on the complex manifold Xan), the answer is yes: 𝖢𝗈𝗇𝗇⁡(Xan)→𝖫𝗈𝖼⁡(X) is an equivalence. When X is proper, the answer is yes by GAGA. However, in general, in the algebraic case, this functor is not necessarily an equivalence, as the map 𝖢𝗈𝗇𝗇⁡(X)→𝖢𝗈𝗇𝗇⁡(Xan) is not necessarily an equivalence. We would like to apply GAGA; to do this, we must restrict to the subcategory of regular integrable connections on X, which are connections that behave well near the boundary of a compactification X¯.

We first define regularity for connections on a curve. A smooth curve C can be compactified to a smooth proper curve C¯ by adding a finite number of points. Let p∈C¯−C, and let (E,∇) be an integrable connection on C. Letting j:C↪C¯ be the inclusion, we get a meromorphic connection j∗⁢E on C¯ with possible poles at C¯−C. Let p∈C¯−C, let KC¯,p denote the fraction field of 𝒪C¯,p, and let x∈𝒪C¯,p be a local parameter at p. Taking the stalk of j∗⁢E at p, we get a KC¯,p-vector space M with a connection ∇:M→ΩC¯,p1⊗𝒪C¯,pM. We say that (E,∇) is regular at p if either of the following equivalent conditions holds:

  • •

    (Algebraic) There exists a finitely generated 𝒪C¯,p-submodule L⊂M such that M=KC¯,p⁢L and x⁢∇L⊂ΩC¯,p1⊗𝒪C¯,pL.

  • •

    (Analytic) All locally defined horizontal sections of E (local in the analytic topology) grow at most as fast as |x|−N for some N as x→0.

We say that (E,∇) is regular if the meromorphic connection j∗⁢E is regular at all p∈C¯−C.

One way to make sense of the algebraic condition is to pick an 𝒪𝒪,p basis y1,…,yr of L and work analytically. In terms of this basis, the set of conditions for horizontal sections of E near p is a set of r ODEs

where the Ai⁢j are some meromorphic functions defined near r (these form the connection matrix). The condition x⁢∇L⊂ΩC¯,p1 means that x⁢Ai⁢j is holomorphic for all i,j. This translates to the condition about moderate growth. For instance, the ODE 𝑑y𝑑x=A⁢xN⁢y has solution c⁢eA⁢xN+1N+1 for N≠−1 and c⁢xA for N=−1. When N≥−1, the solution is a meromorphic function with moderate growth. When N<−1, exN+1N+1 has an essential singularity at p and, in particular, doesn’t have moderate growth.

Now consider a smooth variety X of any dimension. By resolution of singularities, we can find a smooth compactification X¯ such that Y≔X¯−X is a divisor. We say that an integrable connection (E,∇) on X is regular if for every map from a smooth curve iC:C→X, the induced integrable connection iC∗⁢E is regular. It can be shown that this notion is independent of choice of X¯. For the complex manifold Xan, we can similarly define regular integrable connections as certain meromorphic connections on X¯an with respect to Yan, except that we consider maps from the unit disc i:B→X¯an such that i−1⁢(Yan)={0}. Unlike the algebraic case, analytic connections on Xan don’t have unique meromorphic extensions to X¯an, so it doesn’t make sense to say that a particular integrable connection on Xan is regular.

Here’s a big theorem of Deligne, known as Deligne’s Riemann-Hilbert correspondence.

Theorem 9.2.1 (Deligne).

The restriction defines an equivalence of categories

Corollary 9.2.2.

The functor

is an equivalence.

Sketch of proof.

After picking a compactification X¯ with divisor boundary Y, we can factor this functor as

The functor 𝖢𝗈𝗇𝗇reg⁡(X)→𝖢𝗈𝗇𝗇reg⁡(X¯,Y) is an equivalence because algebraic connections on X extend to unique meromorphic connections on X¯ with respect to Y. The functor 𝖢𝗈𝗇𝗇reg⁡(X¯,Y)→𝖢𝗈𝗇𝗇reg⁡(X¯an,Yan) is an equivalence by a GAGA argument; here, we need regularity because GAGA only applies to coherent sheaves on X¯. The functor 𝖢𝗈𝗇𝗇reg⁡(X¯an,Yan)→𝖢𝗈𝗇𝗇⁡(Xan) is an equivalence by the preceding theorem. We’ve already noted that the functor 𝖢𝗈𝗇𝗇⁡(Xan)→𝖫𝗈𝖼⁡(X) is an equivalence.

□

Deligne’s Riemann-Hilbert correspondence provides an algebraic description of the category 𝖫𝗈𝖼⁡(X). The Riemann-Hilbert correspondence of Mebkhout and Kashiwara does the same thing for Dcb⁢(X).

9.3 The Riemann-Hilbert correspondence for regular holonomic D-modules

Recall that a coherent D-module M is holonomic if its characteristic variety Ch⁡(M)=supp⁡grF⁡M⊂T∗⁢X has dimension d, where F is a good filtration. We can define the characteristic cycle CC⁡(M) of a holonomic D-module as the formal sum ∑ZmultZ⁡(M)⁢Z over the irreducible components Z of Ch⁡(M). Since Ch⁡(M) is conic Lagrangian, any irreducible component Z⊂Ch⁡(M) must be of the form TV∗⁢X¯, where V is a smooth locally closed subvariety of X.

For a short exact sequence of holonomic D-modules

CC⁡(N)=CC⁡(M)+CC⁡(L). Moreover, CC⁡(M)=0 iff M≅0. Thus, holonomic D-modules have finite length. The simple holonomic D-modules are the minimal extensions L⁡(V,E)=Im(∫jE→∫j!E), where V is a smooth locally closed subvariety such that j:V↪X is affine and E is a simple integrable connection on V. This means that we can think of holonomic D-modules as D-modules built from integrable connections, in the same way that constructible sheaves are built from local systems. CC⁡(L⁡(V,E))=rk⁡(E)⁢[TV∗⁢X¯], so CC⁡(M) tells us where the composition factors of M are supported.

We say that a holonomic D-module M is regular holonomic if for all its composition factors are of the form L⁡(V,E) for E a regular integrable connection on V. We let Dhb⁢(𝒟X) (resp. Drhb⁢(𝒟X)) denote the full subcategory of Db⁢(𝒟X) consisting of complexes with holonomic (resp. regular holonomic) cohomology sheaves. To get a constructible complex from a holonomic complex M∙, we apply the de Rham functor DR⁡(M∙)≔ΩXand⊗𝒟XLMan⁣∙. Concretely,

where ΩXannd⊗𝒪XanMan is in degree 0. Here, if we pick local coordinates x1,…,xd, the map ΩXani⊗𝒪XanMan⁣∙→ΩXani+1⊗𝒪XanMan⁣∙ sends α⊗m↦𝑑α⊗m+(−1)|α|⁢∑j=n𝑑zj∧α⊗∂jm.

Theorem 9.3.1 (Kashiwara’s constructibility theorem).

For M∙∈Dhb⁢(𝒟X), DR⁡(M∙)∈Dcb⁢(X).

Theorem 9.3.2 (Riemann-Hilbert correspondence).

The functor DR:Drhb⁢(𝒟X)→Dcb⁢(X) is an equivalence respecting the six functor formalism on each category. The natural t-structure on Drhb⁢(𝒟X) corresponds to the perverse t-structure on Dcb⁢(X).

We can also do this with the solution functor Sol:Drhb⁢(X)→≅Dcb⁢(X)op defined by Sol⁡(M∙)≔R⁢Hom¯𝒟Xan⁢(M∙an,𝒪Xan)⁢[d]. The de Rham and solution functors are related by DR⁡(𝔻⁢M∙)≃Sol⁡(M∙).

When (M,∇) is an integrable connection on X, DR⁡(M∙) is the de Rham complex of M with differential ∇. By the holomorphic Poincaré lemma, DR⁡(M) is quasi-isomorphic to the local system of horizontal sections of M in degree −d (and Sol⁡(M) is the dual local system in degree −d). This matches our expectation from the Riemann-Hilbert correspondence for integrable connections.

10 Kevin (Apr 27): the Kazhdan-Lusztig conjectures

Note: These are the speaker’s notes. Minor modifications were made to adapt to my typographical conventions. Some references include Yi Sun’s notes “Perverse sheaves and the Kazhdan-Lusztig conjectures” and de Cataldo-Migliorini’s notes “The Decomposition Theorem, perverse sheaves and the topology of algebraic maps.”

10.1 The Kazhdan-Lusztig conjecture

I gave this talk last semester, but I forgot everything, so here it is again, possibly with different conventions. The Hecke algebra of a Weyl group W is the ℤ⁡[q1/2,q−1/2]-algebra ℋ⁡(W) with ℤ⁡[q1/2,q−1/2]-basis Tw for w∈W and multiplication given by T1=1 and relations

ℋ⁡(W) carries an involution (−)¯ given by q1/2¯=q−1/2 and Tw¯=Tw−1−1. There is a special basis Cw′ (w∈W) of ℋ⁡(W) satisfying the given properties:

  • •

    Cw′¯=Cw′.

  • •

    Cw′=q−ℓ⁡(w)/2⁢∑v≤wPv,w⁢(q)⁢Tv with Pv,w∈ℤ⁡[q] of degree ≤12⁢(ℓ⁡(w)−ℓ⁡(v)−1) for v<w and Pw,w=1.

The Kazhdan-Lusztig polynomials Pv,w describe the change of basis matrix from Tw to Cw′. The Kazhdan-Lusztig conjecture states that these polynomials describe the change of basis matrix from the Verma modules Mw≔M⁡(w⋅(−2⁢ρ)) to the simple modules Lw≔L⁡(w⋅(−2⁢ρ)) in K⁡(𝒪0):

10.2 Sheaves on the flag variety

To prove the Kazhdan-Lusztig conjecture, we translate everything to sheaves. Let Xw (resp. X¯w=⨆v≤wXv) be the Schubert cell (resp. Schubert variety) corresponding to w, and let jw:Xw↪X denote the inclusion.

We have the following functors from the last couple talks:

  • •

    Beilinson-Bernstein localization 𝖫𝗈𝖼:𝖬𝗈𝖽⁡(𝔤,χ0)→∼𝖬𝗈𝖽⁡(𝒟X).

  • •

    The Riemann-Hilbert correspondence DR:𝖬𝗈𝖽rh⁢(𝒟X)→∼𝖯𝖾𝗋𝗏⁡(X). I haven’t defined the category of perverse sheaves 𝖯𝖾𝗋𝗏⁡(X)⊂Dcb⁢(X), so here it is. 𝖯𝖾𝗋𝗏⁡(X) is the full subcategory of Dcb⁢(X) consisting of complexes K such that

    𝖯𝖾𝗋𝗏⁡(X) is the heart of the perverse t-structure, so it is naturally an abelian category. Examples of perverse sheaves include intersection cohomology complexes, which we will define shortly.

It can be shown that 𝖫𝗈𝖼 maps 𝒪0 to 𝖬𝗈𝖽rh⁡(𝒟X), so that we can consider the image DR⁡(𝖫𝗈𝖼⁡(𝒪0))⊂𝖯𝖾𝗋𝗏⁡(X). It is possible to upgrade 𝖫𝗈𝖼 and DR to functors on categories of B-equivariant sheaves, so that the image DR⁡(𝖫𝗈𝖼⁡(𝒪0)) is contained in the subcategory of 𝖯𝖾𝗋𝗏⁡(X) constructible with respect to the Schubert stratification of X. We are concerned with the images of the Vermas and simples in 𝖯𝖾𝗋𝗏⁡(X). Fan showed two weeks ago that

Since DR commutes with the six functors, we have

Implicitly, here we are using the fact that jw is an affine locally closed embedding, so that jw! and jw⁣∗ are exact functors (both for the ordinary t-structure on regular holonomic D-modules and for the perverse t-structure on Dcb⁢(X)). We’ll abuse notation by writing ℂXw for the sheaf jw!⁢ℂXw on X.

10.3 Categorification of the Hecke algebra

The Kazhdan-Lusztig conjecture predicts the change of basis matrix for two bases Lw,Mw of K⁡(𝒪0). Having done all this geometric work, we have shown that this is the change of basis matrix for two bases ICX¯w,ℂXw⁢[ℓ⁡(w)] of K⁡(𝖯𝖾𝗋𝗏⁡(X)W) (here, 𝖯𝖾𝗋𝗏⁡(X)W⊂𝖯𝖾𝗋𝗏⁡(X) is the full subcategory of complexes constructible with respect to the Schubert stratification). We can rephrase the Kazhdan-Lusztig conjecture purely in terms of sheaves:

Theorem 10.3.1 (Kazhdan-Lusztig conjecture).

In K⁡(𝖯𝖾𝗋𝗏⁡(X)W),

Figuring out the change of basis matrix amounts to figuring out the stalks of ICX¯w along the Schubert cells. For a complex of sheaves K on X constructible with respect to the Schubert stratification, let

Here, hi⁢(K)w is the dimension of any stalk of ℋi⁢(K) along Xw.

Theorem 10.3.2.

h⁡(ICX¯w)=Cw′

To prove the theorem, we’ll show that perverse sheaves categorify the Hecke algebra: ICX¯w will correspond to Cw′, whereas ℂXw will correspond to Tw. It’ll be convenient for us to work with X×X rather than X. Just as the Bruhat decomposition gives us a stratification of X by B-orbits, we get a stratification of X×X by G-orbits, where G acts diagonally. We have X×X=⨆wOw, where Ow is the G-orbit of (B,w⁢B). The projection p1:Ow→X (resp. p1:O¯w→X) is a Zariski fibration with fiber Xw (resp. X¯w). Thus, ICO¯w=p1∗⁢ICX¯w⁢[δ], where δ=dimX=number of positive roots. Thus, hi⁢(ICO¯w)v=hi+δ⁢(ICX¯w)v, and for a complex of sheaves K on X×X constructible with respect to the decomposition into G-orbits (we’ll call this the Schubert stratification as well), define

By the above observation, ĥ⁢(ICO¯w)=h⁡(ICX¯w), so we’ll want to show that ĥ⁢(ICO¯w)=Cw′.

We’ll need an operation on sheaves to correspond to the multiplication on the Hecke algebra. Let 𝒞⊂Dcb⁢(X×X) be the full subcategory of complexes constructible with respect to the Schubert stratification such that ℋi⁢(K)≅0 either for all odd i or for all even i (this is called parity vanishing). Given K,K′∈𝒞, we can define the convolution K⋆K′ to be q∗⁢i∗⁢(p12∗⁢K⊗p34∗⁢K′), where the maps fit in the following diagram that I’ve stolen from de Cataldo-Migliorini:

[Uncaptioned image]
Proof of Theorem 10.3.2.

We’ll prove Theorem 10.3.2 by induction. The base cases consist of w=1 and w=s for a simple reflection s. We have O¯1=X, so ICO¯1≅ℂO¯1⁢[δ], and ICO¯1̂=T1=C1′. p1:O¯s→X is a Zariski fibration with fiber X¯s≅ℙ1, so O¯s is smooth of dimension 1+δ, and ICO¯s≅ℂO¯s⁢[1+δ]. The Schubert stratification gives O¯s=O1⊔Os, so ĥ⁢(ICO¯s)=q−1/2⁢(Ts+T1)=Cs′. In particular, ICO¯1,ICO¯s∈𝒞.

To induct, we’ll show that convolution with ICO¯s corresponds to multiplication by Cs′ in the Hecke algebra.

Lemma 10.3.3.

For K∈𝒞, ĥ⁢(ICO¯s⋆K)=Cs′⁢ĥ⁢(K). In particular, ICO¯s⋆K∈𝒞.

Proof.

This is proven by a direct computation. We have

so

Let p=(B,w⁢B) be a point of Ow. Then q−1⁢(p)∩i−1⁢(O¯s×X×X)={(B,x,w⁢B)|x∈X¯s}≅ℙ1. Let Y denote this ℙ1. Then ℋi⁢(ICO¯s⋆K)p≅Hi+δ+1⁢(Y,K|Y). By constructibility on X×X, K|Y is constructible with respect to Y=U⊔u0, where U≅ℂ and u0 is a point. Because U is contractible and K|U is a complex with locally constant cohomology, each ℋi⁢(K) is a constant sheaf, and K|U≃⨁iℋi⁢(K)⁢[−i]. The decomposition Y=U⊔u0 gives us a long exact sequence in compactly supported cohomology ⋯→Hci⁢(U,K|U)→Hi⁢(Y,K|Y)→ℋi⁢(K)|u0→⋯. By parity vanishing, this long exact sequence splices into short exact sequences 0→Hci⁢(U,K|U)→Hi⁢(Y,K|Y)→ℋi⁢(K)|u0→0. Poincaré duality tells us that dimHci⁢(U,K|U)=dimHi−2⁢(U,K|U)=dimℋi−2⁢(U)|u for any u∈U. Thus, hi⁢(ICO¯s⋆K)w=dimHi⁢(Y,K|Y)=dimℋi−2⁢(U)|u+dimℋi⁢(U)|u0.

To finish, we must identify which strata of X×X contain u and u0. If s⁢w>w, then u∈Os⁢w, and u0∈Ow, so that hi⁢(ICO¯s⋆K)w=hi+δ+1⁢(K)|w+hi+δ−1⁢(K)|s⁢w. If s⁢w<w, then u∈Ow, and u0∈Os⁢w, so that hi⁢(ICO¯s⋆K)w=hi+δ+1⁢(K)|s⁢w+hi+δ−1⁢(K)|w. This turns out to be the formula for multiplication by Cs′.

□

To finish the proof, we must compute the remaining ICO¯w. Unlike the O¯s, O¯w might not be smooth, so ICO¯w might not just be a shifted constant sheaf. To compute ICO¯w, we apply the decomposition theorem to the Bott-Samelson resolution π:O~w→O¯w. For a reduced expression w=s1⁢…⁢sℓ, we define O~w≔{(B1,…,Bℓ+1)∈Xℓ+1|(Bi,Bi+1)∈O¯si}. O~w is smooth because O~w→ℙ1 is a (ℙ1)ℓ-fibration. π:O~w→O¯w is defined by π⁡(B1,…,Bℓ+1)=(B1,Bℓ+1); this is an isomorphism over Ow, so it is a resolution of singularities.

By definition, π∗⁢ℂO~w⁢[δ+ℓ]=ICO¯s1⋆⋯⋆ICO¯sℓ. Thus, by Chapter 10.3.3, ĥ⁢(π∗⁢ℂO~w⁢[δ+ℓ])=Cs1′⁢⋯⁢Csℓ′=q−ℓ/2⁢(Ts1+T1)⁢⋯⁢(Tsℓ+T1). Since O~w is smooth, ℂO~w⁢[δ+ℓ]=ICO~w, and the decomposition theorem (which I will not state or explain) states that π∗⁢ℂO~w⁢[δ+ℓ] is a direct sum of shifted ICO¯v’s: π∗⁢ℂO~w⁢[δ+ℓ]≅ICO¯w⊕⨁v<w⨁iICO¯v⊗Vvi⁢[−i] for some finite-dimensional ℂ-vector spaces Vvi. Since ICO¯w is a direct summand of π∗⁢ℂO~w⁢[δ+ℓ]∈𝒞, ICO¯w∈𝒞. Applying ĥ, we have Cs1′⁢⋯⁢Csℓ′=ĥ⁢(ICO¯w)+∑v<wFv⁢(q)⁢ĥ⁢(ICO¯v), where Fv⁢(q)=∑j(dimVvj)⁢qj/2. Since ℂO~w⁢[δ+ℓ] is self-dual and π is proper, 𝔻⁢ℋip⁢(π∗⁢ℂO~w⁢[δ+ℓ])≅ℋ−i⁢(π∗⁢ℂO~w⁢[δ+ℓ]), and dimVvi=dimVv−i. This implies that Fv⁢(q)=Fv⁢(q−1).

By induction on length, we have ĥ⁢(ICO¯w)=Cs1′⁢⋯⁢Csℓ′−∑v<wPv⁢(q)⁢Cv′. Since everything on the RHS is self-dual in ℋ⁡(W), ĥ⁢(ICO¯w) must be self-dual. It remains to see that if we write ĥ⁢(ICO¯w)=q−ℓ/2⁢∑v≤wP~v,w⁢(q)⁢Tv, then (1) P~v,w⁢(q)∈ℤ⁡[q], (2) P~w,w=1, and (3) deg⁡P~v,w≤12⁢(ℓ⁡(w)−ℓ⁡(v)−1):

  1. (1)

    This follows from the fact that ICO¯w∈𝒞.

  2. (2)

    This follows from the equation ĥ⁢(ICO¯w)=Cs1′⁢⋯⁢Csℓ′−∑v<wPv⁢(q)⁢Cv′, as the first term contributes q−ℓ/2⁢Tw and no other terms contribute any.

  3. (3)

    This follows from the support condition on ICO¯w: ℋi−ℓ⁡(w)⁢(ICO¯w)v≅0 if dimOv≥−(i−ℓ⁡(w)). Since dimOv=δ+ℓ⁡(v), this says that i+δ≥ℓ⁡(w)−ℓ⁡(v). Thus, deg⁡P~v,w⁢(q)<12⁢(ℓ⁡(w)−ℓ⁡(v)), as desired.

By the uniqueness of the Kazhdan-Lusztig polynomials, P~v,w=Pv,w, and h⁡(ICX¯w)=ĥ⁢(ICO¯w)=Cw′, as desired.

□

To finish the proof of the Kazhdan-Lusztig conjecture, for K∈𝖯𝖾𝗋𝗏⁡(X)W, we define χW⁢(K)≔∑v∑i(−1)i⁢ℋi⁢(K)v⁢v∈ℤ⁡[W]. By parity vanishing (in particular, ℋi⁢(ICX¯w)≠0 only when i≡ℓ⁡(w)(mod2)),

We have χW⁢(ℂXw)=(−1)ℓ⁡(w)⁢w. From these computations, we see that χW:K⁡(𝖯𝖾𝗋𝗏⁡(X)W)→ℤ⁡[W] is an isomorphism of ℤ-modules. Since χW⁢(ICX¯w)=∑v(−1)ℓ⁡(w)−ℓ⁡(v)⁢Pv,w⁢(1)⁢χW⁢(ℂXv), we see that [ICX¯w]=∑v(−1)ℓ⁡(w)−ℓ⁡(v)⁢Pv,w⁢(1)⁢[ℂXv] in K⁡(𝖯𝖾𝗋𝗏⁡(X)W), which is exactly what we want.