Holonomic D-modules and

minimal extensions

Andrés Sarrazola Alzate

Master thesis in mathematics, year 2015-2016

Université de Rennes 1
Under the direction of Tobias Schmidt
Institute de recherche mathématiques de Rennes.
Équipe de géométrie arithmétique.
A mi padres, Jesús Ubeimar y Ana Gisela, ellos confiaron plena-
mente en mi incluso en los momentos en que yo dejé de hacerlo.

1
Contents

Contents 2

Introduction 3

1 6
1.1 Basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Algebraic properties of D-modules . . . . . . . . . . . . . . . . 8

2 12
2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Homological properties of D-modules . . . . . . . . . . . . . . 16
2.2.1 The Spencer resolution . . . . . . . . . . . . . . . . . . 19
2.2.2 The de Rham complex . . . . . . . . . . . . . . . . . . 20
2.2.3 The Koszul complex for a closed embedding . . . . . . 21
2.2.4 Kashiwara’s equivalence . . . . . . . . . . . . . . . . . 22

3 24
3.1 Characteristic varieties and holonomic D-modules . . . . . . . 24
3.2 Duality functors . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1 Functorial relations under a proper morphism . . . . . 28

4 33
4.1 Preservation of holonomicity . . . . . . . . . . . . . . . . . . . 33
4.2 Minimal extensions . . . . . . . . . . . . . . . . . . . . . . . . 38

Appendix 42

5 44
5.1 Derived categories and derived functors . . . . . . . . . . . . . 44

2
Bibliography 51

3
Introduction

Let X be a smooth complex algebraic variety and let DX be its sheaf of

differential operators (section 1.1). The objective of this work is to give
a geometric classification of the simple holonomic DX -modules (3.1.5 and
section 4.2). More exactly, we will show that every simple holonomic DX -
module is isomorphic to the minimal extension L(Y, M) (4.2.1) for some pair
(Y, M), where Y is a locally closed subvariety of X such that the inclusion
map i : Y ֒→ X is affine and M is a simple integrable connection on Y
(2.1.1). To understand this classification we will develop the following con-
cepts. First of all, we will denote by D + (DX ) the bounded above derived
category and D − (DX ) denotes the bounded below derived category associ-
ated to the abelian category Mod(DX ). In the second chapter we will study
the duality functor DX : D − (DX ) → D + (DX )op defined by

DX M• := RHomDX (M• , DX ) ⊗OX Ω⊗−1

X [dim X], (M• ∈ D − (DX )),

and in the third chapter we will prove that if f : X → Y is a morphism of

smooth algebraic varieties, Dcb (DX ) denotes the bounded derived category
of bounded complexes of coherent DX -modules and Dhb (DX ) denotes the
full subcategory of Dcb (DX ) consisting of complexes M• ∈ Dcb (DX ) such that
H i (M• ) is a holonomic DX -module, then there exists a morphism of functors
((2.2) and (4.3))
Z Z
→ : Dhb (DX ) → Dhb (DY ).
f! f
R R R
Here, f denotes the D-module direct image functor and f ! := DY f DX .
Therefore, for X, Y and i as in the beginning,R and M R∈ Modh (DX ), the
last morphism of functors, gives us a morphism i! M → i M in Modh (DY )
whose image, denoted by L(Y, M), is called the minimal extension of M.
Finally, we explain the content of the main theorem in the case of the affine

4
line and rederive the classical description of simple modules over the first
Weyl algebra given by [6].

5
Chapter 1

1.1 Basic notions

Throughout this work X will denote a smooth algebraic variety of dimension
dX over the complex field C with OX the structural sheaf. TX will denote the
tangent sheaf (whose sections are called vector fields) and which is defined
over an arbitrary open subset U ⊂ X by

TX (U) := {θ ∈ EndC OX (U)|θ(f g) = θ(f )g + f θ(g) (f, g ∈ OX (U))}.

On the other hand, we know that the diagonal morphism ∆ : X → X ×C X

is a closed embedding and therefore we can define the cotangent sheaf of X
by Ω1X := ∆∗ (I/I 2 ), where I is the ideal sheaf of ∆(X). Moreover, in our
case, Ω1X is a locally free sheaf of rank dX .
Finally, we have a derivation d : OX → Ω1X which satisfies d(f g) = d(f )g +
f d(g) , for f, g ∈ OX and gives rise to an isomorphism HomOX (Ω1X , OX ) ≃
TX given by composition with d (the last reasoning is just the global version
of [2] section 25 and the local description of d given in [1] remark 8.9.2).

The following theorem introduces one of the most important tools in our
work.

1.1.1 Theorem. For each point p ∈ X, there exists an affine open neigh-
borhood V of p, regular functions xi ∈ OX (V ), and vector fields ∂i ∈ TX (V )
(1 ≤ i ≤ dX ) satisfying the conditions

 [∂i , ∂j ] = 0, ∂i (xj ) = δij (1 ≤ i, j ≤ dX ),
 LdX
TV = i=1 OV ∂i .

6
Moreover, we can choose the functions x1 , x2 , ..., xdX so that they generate
the maximal ideal mp of the local ring OX,p at p.

Proof. To soften the notation we will assume that n = dX . It’s a consequence

of [1] (chapter II proposition 8.7) that there exist functions x1 , x2 , ..., xn gen-
erating the maximal ideal mp of OX,p and that dx1 , dx2 , ..., dxn is a basis of
the free OX,p -module Ω1X,p . Therefore there exists an affine open neighbor-
hood V of p such that Ω1X (V ) is a free module with basis dx1 , dx2 , ..., dxn
over OX (V P)n andl taking the dual basis we get ∂i (xj ) = δij . Finally, writing
l
[∂i , ∂j ] = l=1 gij ∂l , we have gij = [∂i , ∂j ]xl = 0. Hence [∂i , ∂j ] = 0.

1.1.2 Definition. The set {xi , ∂i | 1 ≤ i ≤ n = dX } defined over an affine

open neigborhood of p satisfying the conditions of the last theorem is called a
local coordinate system at p.

Hereafter, if there is no risk of confusion, we use the notation f ∈ OX for a

local section f of OX . We will identify OX with a subsheaf of EndC (OX ) by
identifying f ∈ OX with the endomorphism of OX given by g 7→ f g.

1.1.3 Definition. The C-subalgebra of EndC (OX ) generated by OX and TX

is called the sheaf of differential operators and is denoted by DX .

The theorem 1.1.1 gives us a local description of the sheaf DX . If x ∈ X, we

can take a coordinate affine neighborhood (U, {xi , ∂i })1≤i≤dX at x. Hence
we have
M
DU = OU ∂xα ,
α∈Nn

αd
where α = (α1 , ..., αdX ) is a multi index and ∂xα = ∂1α1 ...∂dXX .

1.1 Remark. Let U be an affine open subset of X and R the C-algebra

generated by elements {f˜, θ̃| f ∈ OX (U), θ ∈ TX (U)} satisfying the following
fundamental relations:

(1) f˜1 + f˜2 = f^

1 + f2 , f˜1 f˜2 = fg
1 f2 ,

(2) θ˜1 + θ˜2 = θ^

1 + θ2 ,
^
[θ˜1 , θ˜2 ] = [θ1 , θ2 ],

(3) f˜θ̃ = ffθ, g).

[θ̃, f˜] = θ(f

7
The map defined by f˜ 7→ f and θ̃ 7→ θ defines an isomorphism between R
and DX (U).

We end this section with the following definition.

1.1.4 Definition. We say that a sheaf M is a left DX -module if M(U) is
endowed with a left DX (U)-module structure for each open subset U of X
and these actions are compatible with the restriction morphisms.

1.2 Algebraic properties of D-modules

We start this section giving to DX a structure of filtered sheaf of noncommu-
tative rings. First of all, let (U, {xi , ∂i }) be a coordinate affine open subset
of X. We define the order filtration F of DU by
X
Fl DU = OU ∂xα ,
|α|≤l

and more generally, for an arbitrary open subset V of X we can define the
order filtration F of DX over V by

(Fl DX )(V ) =
{P ∈ DX (V )| ρU,V (P ) ∈ Fl DX (U) for any affine open subset U ⊂ V },

where ρU,V : DX (V ) → DX (U) is the restriction map and for convenience we

set Fl DX = 0 if l < 0.
1.2.1 Proposition. (i) {Fl DX }l∈Z is an increasing filtration of DX
[
such that DX = Fl DX and each Fl DX is a locally free module over
l∈Z
OX .

(ii) F0 DX = OX , (Fl DX )(Fm DX ) = Fl+m DX .

(iii) If P ∈ Fl DX and Q ∈ Fm DX , then [P, Q] ∈ Fl+m−1 DX .
Proof. (i) and (ii) are obvious, so we will give the proof of (iii). As the Lie
product is a bilinear operation is enough to see this property on a expression
of the form ∂ α ∈ Fl DX and ∂ β ∈ Fm DX . In this case we can apply induction
on the equality

8
 [∂ α , ∂ β ] = ∂ α1 [∂ α−α1 e1 , ∂ β ] + [∂ α1 , ∂ β ]∂ α−α1 e1 .

1.2 Remark. We have the following formula

Fl DX = {P ∈ EndC (OX )| [P, f ] ∈ Fl−1 DX (∀f ∈ OX )},
and the terms of degree 0 and 1 are OX and TX respectively (cf. [3] chapter
3).
The above results tell us that if we take an affine chart (U, {xi , ∂i }) then
for the sheaf of commutative graded rings
∞
M
F
gr DX = grl DX ,
l=0

where grl DX = Fl DX /Fl−1 DX , we have

M
grl DU = Fl DU /Fl−1 DU = OU ξ α (ξi := ∂i mod(F0 DU )),
|α|=l

and therefore
grDU = OU [ξ1 , ..., ξdX ].
In the rest of the section we will show some algebraic and homological prop-
erties of certain special categories of DX -modules. We will be interested in
the category of quasi-coherent OX -modules (resp. of coherent OX -modules)
which will be denoted by Modqc (OX ) (resp. by Modc (OX )) and we regard
the notation Modqc (DX ) for the category of OX -quasi-coherent DX -modules
and Modc (DX ) the category of coherent DX -modules.
1.3 Remark. It can be shown that M is a coherent DX -module if and only
if is quasi-coherent over OX and locally finitely generated over DX .
1.2.2 Definition. A smooth algebraic variety X is called D-affine if the
following conditions are satisfied:
(a) The global section functor Γ(X, •) : Modqc (DX ) → Mod(Γ(X, DX )) is
exact,

(b) if Γ(X, M) = 0 for M ∈ Modqc (DX ), then M = 0.

9
1.4 Remark. Any smooth affine algebraic variety is D-affine.
Let’s suppose that X is D-affine and let’s take M ∈ Modqc (DX ). From
the above definition, if we apply the global sections functor to the exact
sequence
0 → M0 → M → M/M0 → 0,
where M0 is the image of the natural morphism DX ⊗Γ(X,DX ) Γ(X, M) → M,
we obtain Γ(X, M/M0) = 0 and therefore M = M0 .
From [4] we have that for G ∈ Mod(Γ(X, DX )) the canonical morphisms
αM : DX ⊗Γ(X,DX ) Γ(X, M) → M, βG : G → (Γ(X, DX ⊗Γ(X,DX ) G))
are isomorphisms and is straightforward to verify that DX ⊗Γ(X,DX ) (•) is left
adjoint of Γ(X, •).
1.2.3 Proposition. Let’s suppose that X is D-affine.
(i) Every M ∈ Modqc (DX ) is generated over DX by its global sections.

(ii) The functor

Γ(X, •) : Modqc (DX ) → Mod(Γ(X, DX ))

gives an equivalence of categories.

Proof. We have already proved (i). To see (ii) we must check that Γ(X, •) is
surjective (which comes from αM ) and fully faithful. So, let ψ : Γ(X, M) →
Γ(X, N ) where N ∈ Modqc (DX ), be a morphism of Γ(X, DX )-modules and
−1
let’s define φ := αN ◦ (idDX ⊗ ψ) ◦ αM . We easily can verify that Γ(X, φ) =
ψ.

Now, let’s consider the case when G ∈ Modf (Γ(X, DX )) and M ∈ Modc (DX ).
It’s clear that DX ⊗Γ(X,DX ) G belongs to Modc (DX ). On the other hand, as
M is generated by its global sections and X is quasi-compact, we see that
M is globally generated by finitely many elements of Γ(X, M). This means
that we have a surjective morphism DX m
→ M → 0 for some m ∈ N and
therefore Γ(X, M) belongs to Modf (Γ(X, DX )). We have proved,
1.2.4 Proposition. Let’s suppose that X is D-affine. The equivalence given
in proposition 1.2.3 induces the equivalence

10
 Modc (DX ) ≃ Modf (Γ(X, DX )).
Finally, let’s suppose again M ∈ Modqc (DX ). Let’s take an affine open
covering {Ui }i and let jk : Uk → X be the open embedding. By proposition
1.2.3 we canLembed jk∗ M into an injective object Ik ∈ Modqc (DUk ) and
setting I = jk∗ Ik , we get a canonical embedding into an injective object
of Modqc (DX ). Moreover, if N ∈ Modqc (OUk ) the relation
HomOUk (N , Ik ) ≃ HomDUk (DUk ⊗OUk N , Ik ),
tells us that Ik is flabby and we get,
1.2.5 Proposition. Any M ∈ Modqc (DX ) can be embedded into an injective
object I of Modqc (DX ) which is flabby.
When X is quasi-projective we have a dual version of the last proposition
(cf. [4], 1.4.18). More exactly,
1.2.6 Proposition. Let’s suppose that X is a quasi-projective variety. Then,
any M ∈ Modqc (DX ) is a quotient of a locally free DX -module.
1.5 Remark. Hereafter, all algebraic varieties are assumed to be quasi-
projective.
In the last section we have shown that over an affine chart (U, {xi , ∂i }i )
we get
grF DX (U) = OX (U)[ξ1 , ..., , ξdX ], grF DX,x = OX,x [ξ1 , ..., ξdX ].
In particular, grF DX (U) and grF DX,x are noetherian rings with global di-
mension 2dX which implies that both DX (U) and DX,x are left noetherian
rings whose global dimensions are smaller than or equal to 2dX (cf. [4] D.2.6).
Applying this fact to the resolution
... → P1 → P0 → M → 0
of M by locally free DX -modules given by the proposition 1.2.6 we see that
if we set Q = Coker(P2dX +1 → P2dX ) then Q|U is a projective object of
Modqc (DU ) and we get
1.2.7 Corollary. Let M ∈ Modqc (DX ). There exists a finite resolution
0 → P2dX → P2dX −1 → ... → P0 → M → 0
of M by locally projective DX -modules. If M ∈ Modc (DX ), we can take all
Pi′ s to be of finite rank.
1.6 Remark. One can show that M actually admits a finite resolution of
length dX (cf. [4] 2.6.11).

11
Chapter 2

2.1 Preliminaries
In this section we will define the transfer bimodules for a morphism f :
X → Y of smooth algebraic varieties and the side-changing operations. We
will specify the objects and the reader can verify the left or right D-module
structure (cf. [4]).
We start with a useful lemma.
2.1 Lemma. Let M be an OX -module. To extend the OX -module structure
to DX is equivalent to give a C-linear morphism
∇ : TX → EndC (M),
satisfying the following conditions:
(1) ∇f θ (s) = f ∇θ (s),

(2) ∇θ (f s) = θ(f )s + f ∇θ (s),

(3) ∇[θ1 ,θ2 ] (s) = [∇θ1 , ∇θ2 ](s).

Proof. Let’s suppose that M is a DX -module. For θ ∈ TX and s ∈ M we
define ∇θ (s) := θ.s (where the multiplication is given by the action of DX ).
We must verify the conditions (1) − (3).
The first condition is just the compatibility between the OX -module structure
and the action of TX and (3) is the definition of the bracket. Finally by the
remark 1.1 we get the relation [θ, f ] = θ(f ) and therefore, for s ∈ M we see
that
θ(f )(s) = [θ, f ](s) = ∇θ (f s) − f ∇θ (s).

12
Reciprocally, is straightforward to check that the operations f˜.s = f s and
θ̃.s = ∇θ (s), respect the relations given in remark 1.1 and therefore define
on M a structure of DX -module compatible with OX .

2.1 Remark. We have an analogous version for right DX -modules. In this

case, it is necessary to define the right DX -module structure in terms of ∇
by

s.θ = −∇θ (s).

2.1.1 Definition. We say that a DX -module M is an integrable connection if

it is locally free of finite rank over OX . The category of integrable connections
on X will be denoted by Conn(X).

2.2 Remark. The category Conn(X) is an abelian category (cf [4] 1.4.11).

In contrast with OX -modules, is important to specify if we are dealing

with a left or a right DX -module. In what follows we will study an OX -
module that will help us to interchange those structures. To start with it, we
op
will denote by Mod(DX ) the category of left DX -modules and by Mod(DX )
the category of right DX -modules.
Let’s consider the canonical sheaf
dX
^
ΩX := Ω1X .

In the last section we have identified Ω1X ≃ HomOX (TX , OX ) and therefore,
we can consider ΩX as the sheaf AltdX (TX × ... × TX ; OX ). With this identifi-
cation we can define a natural action of TX on ΩX , called the Lie derivative.
If θ, θ1 , θ2 , ..., θdX ∈ TX and ω ∈ Ω it is defined by:
dX
X
Lieθ ω(θ1 , ..., θdX ) := θ(ω(θ1 , ..., θdX )) − ω(θ1 , ..., [θ, θi ], ..., θdX ).
i=1

It is straightforward to verify that Lie satisfies the implicit conditions given

in the remark 2.1. Hence we can define a structure of a right DX -module on
ΩX by

ωθ := −Lieθ ω.

13
 op
Given M, N ∈ Mod(DX ) and M′ , N ′ ∈ Mod(DX ) it’s not hard to see
op
′
that M ⊗OX N ∈ Mod(DX ), M ⊗OX N ∈ Mod(DX ), HomOX (M′ , N ′) ∈
op
Mod(DX ), HomOX (M′ , N ′) ∈ Mod(DX ), HomOX (M, N ′) ∈ Mod(DX ) (cf.
[4] 1.2.9) and the following lemma is a straightforward calculation using these
facts.
2.2 Lemma. We have the following isomorphisms

(M′ ⊗OX N ) ⊗DX M ≃ M′ ⊗DX (M ⊗OX N ) ≃ (M′ ⊗OX M) ⊗DX N

of CX -vector spaces (where CX is the constant sheaf ).

Let’s denote by Ω⊗−1

X the dual of the invertible OX -module ΩX . The above
results give us,
op
2.1.2 Proposition. The functor ΩX ⊗OX (•) : Mod(DX ) → Mod(DX ) is
⊗−1
an equivalence of categories. Its quasi-inverse is given by ΩX ⊗OX (•).

The operations described by the above proposition are called side-changing

operations.

Our next objective is to exhibit some functorial properties of D-modules.

More exactly, for a morphism f : X → Y of smooth algebraic varieties, we
will introduce two operations called the inverse image and the direct image.

Let’s start with the inverse image. Let M be a left DY -module. By [1]
we get a homomorphism of OX -modules, f ∗ Ω1Y → Ω1X and therefore taking
the OX -dual we get the OX -homomorphism

TX → f ∗ TY , (θ → θ̃). (2.1)

Based on the above homomorphism we can put a left DX -module structure

to f ∗ M by defining
θ(ψ ⊗ s) = θ(ψ) ⊗ s + ψ θ̃(s),
where ψ ∈ OX , θ ∈ TX and s ∈ M. Taking a local coordinate system {yi , ∂i }
on Y the action of TX can be written as
dY
X
θ(ψ ⊗ s) = θ(ψ) ⊗ s + ψ θ(yi ◦ f ) ⊗ ∂i s.
i=1

14
Applying the above construction to DY , f ∗ DY turns out to be a (DX , f −1 DY )-
bimodule which we denote by DX→Y . With this definition, it’s clear that
f ∗ M ≃ DX→Y ⊗f −1 DY f −1 M and we get a right exact functor

DX→Y ⊗f −1 DY f −1 (•) : Mod(DY ) → Mod(DX ).

2.1.1 Example. For a closed embedding i : X → Y , at any point we can

choose an affine chart (U, {yk , ∂yk }1≤k≤dY ) such that ydX +1 = ... = ydY = 0
is a well-defining equation of X, {xk = yk ◦ i, ∂xk } is a local coordinate of X
and the morphism 2.1 is given by ∂xk 7→ ∂yk . In this particular case, DX→Y
is locally described by

DX→Y ≃ DX ⊗C [∂dX +1 , ..., ∂dY ].

Before introducing the direct image, we need the following notion. Let’s
take an affine chart (U;P{xi , ∂i }1≤i≤dX ). We define a transposition τ of DU as
follows. For P (x, ∂) = α aα (x)∂ α ∈ DU we set
X
τ (P (x, ∂)) := (−1)α ∂ α aα (x),
α

τ satisfies τ (P.Q) = τ (Q).τ (P ) and therefore, for a left DU -module M we

can define on M a structure of right DU -module via τ which we will denote
by Mτ .

2.3 Remark. The definition of τ a priori depends on the coordinated system

taken. The global notion is just proposition 2.1.2.

2.1.3 Definition. We have a (f −1 DY , DX )-bimodule defined by

DY ←X := ΩX ⊗OX DX→Y ⊗f −1 OY f −1 Ω⊗−1

Y

2.1.2 Example. Using the same notation of example 2.1.1 for a closed em-
bedding i : X → Y we have that locally

DY ←X ≃ C[∂dX +1 , ..., ∂dY ] ⊗C DX .

τ
In fact, DY ←X = DX→Y (cf. [4] 1.3.5)

15
2.2 Homological properties of D-modules
In this section we will define several functors on derived categories of D-
modules and show some fundamental properties concerning them. At the
end of the section we will introduce the Spencer resolution, the Koszul
resolution of OX and the de Rham complex. The second one is an im-
portant tool in the proof of Kashiwara’s equivalence. For a general study on
Derived categories the reader is invited to look up the appendix (cf. [4]).
Let’s suppose that R is a sheaf of rings. Throughout this section we will
denote by D(R), D + (R), D − (R) and D b (R) the derived categories associ-
ated to the abelian category Mod(R). The key behind the following def-
initions comes from the fundamental fact that any object M• ∈ D + (R)
(resp. D − (R)) is quasi-isomorphic to a complex I • (resp. P • ) of injective
(resp. flat) R-modules belonging to D + (R) (resp. D − (R)). Moreover, if
b
Dqc (DX ) denotes the bounded derived category associated to the abelian
category Modqc (DX ) then by proposition 1.2.5 and corollary 1.2.7, any ob-
b
ject of D(DX ) (resp. Dqc (DX )) is represented by a bounded complex of flat
DX -modules (resp. locally projective DX -modules belonging to Modqc (DX )).
Let f : X → Y be a morphism of smooth algebraic varieties. As a classical re-
sult in algebraic geometry we know that the functor f ∗ is right exact and the
functor f∗ is left exact, consequently if M• ∈ D b (DY ) and N • ∈ D b (f −1 DY )
we can define the functors

Rf∗ : D b (f −1 (DY )) → D b (DY ) (N • 7→ f∗ (N • ))

by using an injective resolution on N • , and

Lf ∗ : D b (DY ) → D b (DX ) (M• → DX→Y ⊗Lf −1 (DY ) f −1 M• )

by using a flat resolution of M• .

We call Lf ∗ the inverse image. We also use the shifted inverse image

f † := Lf ∗ [dX − dY ] : D b (DY ) → D b (DX ),

defined by f † M• = Lf ∗ M• [dX − dY ]. The shifted inverse image will play an

important role in Kashiwara’s equivalence.
Let g : Y → Z be another morphism of smooth algebraic varieties. As a result
of the fact that D is a locally free O-module we obtain an isomorphism

DX→Z ≃ DX→Y ⊗Lf −1 DY f −1 DY →Z

16
of (DX , (g ◦f )−1DZ )-bimodules and therefore L(g ◦f )∗ M• = Lf ∗ (Lg ∗ (M• )).
In conclusion,

2.2.1 Proposition. Let f : X → Y and g : Y → Z be morphisms of smooth

algebraic varieties. Then we have

L(g ◦ f )∗ ≃ Lf ∗ ◦ Lg ∗ , (g ◦ f )† ≃ f † ◦ g † .

The following functors will play an important role in the next section and
they will be considered in depth at the last part of this work.

2.2.2 Definition. For a closed embedding i : X → Y of smooth algebraic

varieties we define a left exact functor

i♮ : Mod(DY ) → Mod(DX )

by i♮ M := Homi−1 DY (DY ←X , i−1 M).

We will need the following relation (cf. [4] 1.5.16).

2.2.3 Proposition. Let i : X → Y be a closed embedding. Then we have

i† M• ≃ Ri♮ M•

for any M• ∈ D + (DY ).

According to the above reasoning, we can define functors

D b (DX ) ∋ M• 7→ DY ←X ⊗LDX M• ∈ D b (f −1 DY ),

D b (f −1 DY ) ∋ N • 7→ Rf∗ (N • ) ∈ D b (DY )

by using a flat resolution of M• and an injective resolution of N • . Taking

the composition we obtain the functor
Z
: D b (DX ) → D b (DY ) (2.2)
f

given by
Z
M• = Rf∗ (DY ←X ⊗LDX M• ),
f

17
and for an integer k we set
Z k Z 
• k •
M := H M .
f f

Finally, we will need the following relation (cf. [4] 1.5.21).

2.2.4 Proposition. Let f : X → Y and g :→ Z be morphisms of smooth

algebraic varieties. Then we have
Z Z Z
= .
g◦f g f

Continuing with the ideas given in 2.1.1 and bearing in mind the exactness
of the functor i∗ , forR i : X → Y a closed embedding, we have the following
local description of i M for M ∈ Mod(DX ):
Z 0 Z k6=0
M = C[∂dX +1 , ..., ∂dY ] ⊗C i∗ M, M = 0. (2.3)
i i
R0
In particular i
is exact and sends Modqc (DX ) to Modqc (DY ).

2.2.5 Proposition. Let i : X → Y be a closed embedding of smooth algebraic

varieties.

(i) There exists a functorial isomorphism

Z 
• •
RHomDY M ,N ≃ i∗ RHomDX (M• , Ri♮ N • )
i

for M• ∈ D − (DX ) and N • ∈ D + (DY ).

Z
♮ b b
(ii) The functor Ri : D (DY ) → D (DX ) is right adjoint to : D b (DX ) →
i
D b (DY ).

Proof. We only have to show the statement (i) because taking H 0 (RΓ(Y, •))
we obtain (ii). From the canonical ismorphism
HomDX (M, Homi−1 DY (DY ←X , i−1 N )) ≃ Homi−1 DY (DY ←X ⊗DX M, i−1 N )

18
we obtain

RHomDX (M• , RHomi−1 DY (DY ←X , i−1 N • ))

≃ RHomi−1 DY (DY ←X ⊗LDX M• , i−1 N • ),

and from this we can see by definition of i♮ , that

Z 
• ♮ • • •
i∗ RHomDX (M , Ri N ) ≃ RHomDY M ,N
i

2.2.1 The Spencer resolution

k
^
Let’s denote ΩkX Ω1X for 0 ≤ k ≤ dX . It’s clear that for k = 0 we have
=
^0
a canonical map DX ⊗OX TX → OX and for 0 < k ≤ dX we can define a
morphism
k
^ k−1
^
dk : DX ⊗OX TX → DX ⊗OX TX

by

dk (P ⊗ θ1 ∧ ... ∧ θk )
X
= (−1)i+1 P θi ⊗ θ1 ∧ ... ∧ θbi ∧ ... ∧ θk
i
X
+ (−1)i+j P ⊗ [θi , θj ] ∧ θ1 ∧ ... ∧ θbi ∧ ... ∧ θbj ∧ ... ∧ θk .
i<j

It is easy to see that dk ◦ dk−1 = 0.

In order to show that the last complex, that we will denote by S • , is acyclic
we consider the filtration {Fp S • }p :
" dX 0
#
^ ^
Fp S • = Fp−dX DX ⊗OX TX → ... → Fp DX ⊗OX TX → Fp OX ,

(Fp OX = OX if p ≥ 0 and Fp OX = 0 if p < 0 ) and we will show that the

associated graded complex

19
 dX
^
• F
grS =gr DX ⊗OX TX → ... →grF DX ⊗OX TX →grF DX → OX

is acyclic. It is straightforward to verify that in the associated graduated

complex the morphisms come up from the relation
X
dk (P ⊗ θ1 ∧ ... ∧ θk ) = (−1)i+1 P θi ⊗ θ1 ∧ ... ∧ θbi ∧ ... ∧ θn
i

and therefore, if T ∗ X is the cotangent bundle, π : T ∗ X → X is the projection

and i : X → T ∗ X is the embedding by the zero-section then grF S • ≃ π∗ L• ,
where L• is the Koszul’s complex of the OT ∗ X - module i∗ OX . Hence L• is
acyclic and then, also π∗ L• .
The complex S • is called the Spencer resolution of OX . Using the side-
changing operation we obtain the complex

0 → Ω0X ⊗OX DX → ... → ΩdXX ⊗OX DX → ΩX → 0. (2.4)

2.2.2 The de Rham complex

Let Y and Z be smooth algebraic varieties and set X = Y ×Z. Let f : X → Y
and g : X → Z be the projections. For M ∈ Mod(OY ) and N ∈ Mod(OZ )
we define the bifunctor
M ⊠ N := OX ⊗f −1 OY ⊗C g−1 OZ (f −1 M ⊗C g −1 N ) ∈ Mod(OX )
which is exact with respect to both factors and extends to a functor on the
derived categories. It is easily seen that this extension, which we denote by
b b
(•) ⊠ (•), sends Dqc (DY ) × Dqc (DZ ) (resp. Dcb (DY ) × Dcb (DZ )) to Dqc
b
(DX )
b
(resp. Dc (DX )). We also note that
f ∗ M ≃ M ⊠ OZ , g ∗N ≃ OY ⊠ N .
Now, let’s suppose that M ∈ Modqc (DX ). To compute DY ←X ⊗LDX M we
use the resolution of the right DX -module DY ←X = DY ⊠ ΩZ induced by the
resolution of the right DZ -module ΩZ given in the previous section and we
define the de Rham complex DRX/Y (M) by
 dZ +k
 ΩX/Y ⊗OX M if −dZ ≤ k ≤ 0,
k
(DRX/Y (M)) =

0 otherwise.

20
 dZ
X
d(ω ⊗ s) = dω ⊗ s + (dzi ∧ ω) ⊗ ∂i s.
i=0

Here {zi , ∂i }1≤i≤dZ is a local coordinate of Z and the complex 2.4 gives us
DY ←X ⊗LDX M ≃ DRX/Y (M).
Finally, given that f∗ has cohomological dimension dZ we get
2.2.6 Proposition. Let Y and Z be smooth algebraic varieties, and let f :
X = Y × Z → Y be the projection.
R
(i) For M ∈ Mod(DX ) we have f M ≃ Rf∗ (DRX/Y (M).
Rj
(ii) For M ∈ Mod(DX ) we have f M = 0 unless −dZ ≤ j ≤ dZ .

2.2.3 The Koszul complex for a closed embedding

In this section we will give a description of the cohomology sheaves of Li∗ M
for M ∈ Modqc (DX ) and i : X → Y a closed embedding (cf. [4] for a
local description of the DX -module structure). Taking the notation given in
example 2.1.1 we have a locally free resolution
0 → KdY −dX → ... → K0 := i−1 OY → OX → 0
where
j dY
!
^ M
Kj = i−1 OY dyk
k=dX +1

and the morphism Kj → Kj−1 is given by

j
X
f dyk1 ∧ ... ∧ dykj 7→ d
(−1)p+1 ykp f dyk1 ∧ ... ∧ dy kp ∧ ... ∧ dykj .
p=1

Taking the tensor product with i−1 DY over i−1 OY we get a locally free reso-
lution of the right i−1 DY -modules DX→Y and hence Li∗ M is represented by
the complex
... → 0 → KdY −dX ⊗i−1 OY i−1 M → ... → K0 ⊗i−1 OY i−1 M → 0 → ...
From this complex we see that
2.2.7 Proposition. Let i : X → Y be a closed embedding of smooth algebraic
varieties. Set d = codimY (X). For M ∈ Mod(DY ) we have H j (Li∗ M) = 0
unless −d ≤ j ≤ 0.

21
2.2.4 Kashiwara’s equivalence
Throughout this subsection i : X → Y will denote a closed embedding R
between smooth algebraic varieties. According to (2.3) the functor i :
Modqc (DX ) → Modqc (DDY ) is an exact functor and the image of a DX -
module is an object of the full subcategory ModXqc (DY ) of quasi-coherent
DY -modules (we keep the notation for coherent DY -modules) supported by
X.
2.2.8 Theorem. (Kashiwara’s
R0 equivalence). Let i : X → Y be a closed
0 †
embedding. The functors i : Modqc (DX ) → ModX ♮
qc (DY ) and i = H i :
X
Modqc (DY ) → Modqc (DX ) are inverse and define an equivalence of cate-
gories.
Proof. Let’s take M ∈ Modqc (DX ) and N ∈ ModX qc (DY ). We need to show
that the canonical homomorphisms
Z 0 Z 0
M →i ♮
M, i♮ N → N
i i
are isomorphisms. It’s sufficient to verify locally, so we can suppose that Y is
affine and by induction on the number of generators of the ideal defining X,
we can assume that X is a hypersurface. Over a locally coordinate system
{yk , ∂yk }1≤k≤dY as in example 2.1.1, in particular ydY = 0 on X, we set
y = ydY , ∂ = ∂ydY , θ = y∂. From the above section, we can calculate the
homology using the Koszul complex
y
...0 → i−1 N −
→ i−1 N → 0 → ...
and consequently, considering the shifting of degree −1 we get H 0 i† N =
ker(y) and H 1 i† N = coker(y).
Let’s consider the eigenspaces N j := {s ∈ N | θs = js} (j ∈ Z). It’s a
y ∂
straightforward calculation to show that both morphisms N j − → Nj
→ N j+1 −
are isomorphisms for j < −1. Let’s see by induction that
k
M
k
ker(y : N → N ) ⊂ N −j . (2.5)
j=1

For k = 1, the relation ys = 0 gives us θs = (∂y − 1)s = −s. Let’s suppose

k−1
M
k > 1. Then for s ∈ ker(y k ) we have y k s = y k−1(ys) = 0 and ys ∈ N −j
j=1

22
 k
M
by hypothesis of induction. Hence ∂ys ∈ N −j and
j=2

k
M
θs + s = y∂s + s = ∂ys ∈ N −j (2.6)
j=2

On the other hand, we have y k−1(θs + ks) = ∂y k s = 0. Therefore by the

hypothesis of induction
k−1
M
θs + ks ∈ N −j . (2.7)
j=1

k
M
From (2.6)-(2.7) we get (k − 1)s ∈ N −j . Moreover, the fact that N is a
j=1
quasi-coherent OY -module supported on X tell us that (2.5) is an equality.
In particular H 1 i† N = 0 ( for all k 6= 0 by 2.2.7), N ≃ C[∂] ⊗C N −1 and
i♮ N ≃ i−1 N −1. From this we get the theorem.

23
Chapter 3

3.1 Characteristic varieties and holonomic D-

modules
In this section we will develop the central definition of our work, namely, the
notion of a holonomic D-module. For this, we will globalize to the theory of
D-modules some facts of the theory of graded rings (cf. [4] and [3]).
We start with the following property (cf [4] 2.1.1)
3.1.1 Proposition. [Definition]. Let (M, F ) be a filtered DX -module.
Then the following condition are equivalent to each other:
(i) grF M is coherent over grF DX .

(ii) Fi M is coherent over OX for each i, and there exists i0 >> 0 satisfying
Fj DX Fi M = Fi+j M, for j ≥ 0 and i ≥ i0 .
⊕m
(iii) There exist locally a surjective DX -homomorphism Φ : DX → M and
integers n1 , ..., nm such that Φ(Fi−n1 DX ⊕ ... ⊕ Fi−nm DX ) = Fi M.
If F satisfies one of the above equivalent conditions we say that F is a good
filtration.
3.1 Remark. A DX -module admits a good filtration if and only if it is a
coherent DX -module.
Let M be a coherent DX -module and choose a good filtration on it.
Let π : T ∗ X → X be the cotangent bundle of X. Regarding ξ1 , ..., ξdX
as the coordinate system of the cotangent space ⊕ni=1 Cdxi , OU [ξ1 , ..., ξdX ]
is identified with the sheaf π∗ OT ∗ X |U of algebras and therefore we obtain

24
a canonical identification grF DX ≃ π∗ OT ∗ X . We define the characteristic
variety of M, denoted by Ch(M), as the support of the coherent OT ∗ X -
module
^
gr F M := O ∗ ⊗ −1
T X
−1 F
π π∗ OT ∗ X π (gr M).

3.2 Remark. The characteristic variety does not depend on the choice of a
good filtration F .
By a classic result in algebraic geometry (cf. [4] D.3.1) we have the
following
3.1.2 Theorem. For a short exact sequence 0 → M → L → N → 0 of
coherent DX -modules, we have
Ch(L) = Ch(M) ∪ Ch(N ).
Let’s introduce the notion of multiplicity for M ∈ Modc (DX ). Let C be
an irreducible component of the support of M, this set will be denoted by
I(supp(M)), and U an affine open subset of X such that C ∩ U = C and
denote the defining ideal of C ∩ U by pC ⊂ OU (U). Taking the localization
at pC we get an artinian OU (U)pc -module and its length mC (M) is defined.
We call it the multiplicity of M along C. The next fact is just a global
version of the additive property of the length.
3.1.3 Proposition. Let 0 → M → N → L → 0 be an exact sequence of
DX -modules. Then for an irreducible subvariety C of T ∗ X such that C is an
irreducible component of Ch(N ) we have
^
mC (gr C
^
F (N )) = m (gr ^
F (M)) + m (gr
C
F (L))

Moreover, if we define the characteristic cycle of M by

X
CC(N ) = mC (gr^F (N ))C,

C∈I(Ch(N ))

and CCd (N ) the sum taking only the irreducible components of degree d,
then in particular for d = dim Ch(N ) we have
CCd (N ) = CCd (M) + CCd (L).
The heart of our work is the following result.

25
3.1.4 Proposition. Let M be a nonzero coherent DX -module. Then
dim Ch(M) ≥ dX .
To prove the above proposition we will need the following relation (cf [4]
2.3.5).

R 0 closed subvariety of X and let M be a co-

3.1 Lemma. Let Z be a smooth
herent DZ -module. Set N = i M, where i : Z ֒→ X denotes the embedding.
Let ρi : Z ×X T ∗ X → T ∗ Z and ωi : Z ×X T ∗ X ֒→ T ∗ X be the natural
morphisms induced by i. Then we have Ch(N ) = ωi ρ−1
i (Ch(M)).

Proof of proposition 3.1.4. We will proceed by induction on dX . The case

dX = 0 is trivial and we can assume that dX > 0. Moreover, if TX∗ denotes
the zero section of T ∗ X and supp(M) = X then Ch(M) ⊃ TX∗ X, and
hence dim Ch(X) ≥dim TX∗ X = dX . Therefore we can assume from the
beginning that supp(M) is a proper closed subset of X and replace X by
a suitable smooth hypersurface Z, with i : Z ֒→ X the closed embedding,
that contains the support of M. By Kashiwara’sR equivalence there exists a
0
non-zero coherent DZ -module L such that M = i L. By the above lemma
we have Ch(M) = ωi ρ−1 i (Ch(L)) and hence dim Ch(M) =dim Ch(L) + 1.
On the other hand, according to the hypothesis of induction dim Ch(L) ≥
dZ = dX − 1 and therefore dim Ch(M) ≥ dZ + 1 = dX .
3.1.5 Definition. A coherent DX -module M is called a holonomic DX -
module if it satisfies dim Ch(M) ≤ dX .
3.1.1 Example. Let’s suppose that M is an integrable connection. If Fi M =
0 for i < 0 and Fi M = M for i ≥ 0, then F defines good filtration on M
and grF M ≃ M. Moreover, since TX =Annπ∗ OT ∗ X (gr F M), we get
Ch(M) = TX∗ X
(where TX∗ X is the image of the zero section) and by the proof of proposition
3.1.4, M is a holonomic DX -module.

3.2 Duality functors

3.2.1 Definition. We define the duality functor DX : D − (DX ) → D + (DX )op
by
DX (M• ) := RHomDX (M• , DX ) ⊗OX Ω⊗−1
X [dX ]
• ⊗−1
= RHomDX (M , DX ⊗OX ΩX [dX ]).

26
3.3 Remark. (cf. [4] 2.6.8) If M is a holonomic DX -module, then
DX (M) = ExtdDXX (M, DX ) ⊗OX Ω⊗−1
X .

Moreover, in this case, DX (M) is also holonomic by [4] 2.6.7.

3.4 Remark. (cf. [4] 2.6.5) D2 ≃id on Dcb (DX ).
3.2 Lemma. For M• ∈ Dcb (DX ) and N • ∈ D b (DX ), we have
RHomDX (M• , N •) ≃ RHomDX (M• , DX ) ⊗LDX N • .
Proof. First of all let’s note that we have a canonical morphism
RHomDX (M• , DX ) ⊗LDX N • → RHomDX (M• , N • ). (3.1)
By 1.2.7 and the additive property of Hom, for a finite direct sum, we can
assume that M• ≃ DX and the assertion follows since both sides of 3.1 are
isomorphic to N • .
3.2.2 Proposition. For M• ∈ Dcb (DX ), and N • ∈ D b (DX ) we have an
isomorphism
RHomDX (M• , N • ) ≃ RHomDX (OX , DX M• ⊗LOX N • ). (3.2)
in D b (CX ) (with CX the constant sheaf ). In particular we have
RHomDX (OX , N • ) ≃ ΩX ⊗LDX N • [−dX ] (3.3)
Proof. Let’s see 3.3. By the above lemma we can assume that N • = DX . In
this case by 2.4 we have
RHomDX (OX , DX )
" dX
!#
^
= HomDX (DX , DX ) → ... → HomDX DX ⊗OX TX , DX
" dX
!#
^
≃ DX → ... → HomOX TX , DX
" dX
#
^
≃ DX → ... → Ω1X ⊗OX DX

≃ ΩX [−dX ]
Now, by definition and the last lemma we have

27
 RHomDX (M• , N • ) ≃ (ΩX ⊗LOX DX M• ) ⊗LDX N • [−dX ],
and therefore by taking M• = OX we get 3.3. The lemma 3.2 give us
(3.2).

3.2.1 Functorial relations under a proper morphism

The goal of this subsection is to show the following result.

3.2.3 Theorem. Let f : X → Y be a proper morphism. Then we have a

canonical isomorphism
Z Z
DX → DY : Dcb (DX ) → Dcb (DY )
f f

of functors.

To prove the above theorem we will need the following notions.

Let Z be a closed subset of X and U := X \ Z the complementary open
subset. If i : Z → X and j : U → X denote the embeddings, then for any
injective sheaf I on X we get an exact sequence

0 → ΓZ (I) → I → j∗ j −1 I → 0

and by proposition 5.1.10, for any M• ∈ Dqc

b
(DX ), we get a distinguished
triangle
+1
RΓZ (M• ) → M• → Rj∗ j −1 M• −→ . (3.4)

If Z is a smooth closed subvariety, then

Z Z
†
i M |U = 0 and RΓZ (M ) ≃ i† M•
• •
(3.5)
j i

On the other hand, let f : X → Y be a proper morphism of smooth algebraic

varieties. We want to construct a morphism
Z
T rf : : OX [dX ] → OY [dY ] (3.6)
f

called the trace map of f .

First of all, as X is quasi projective, we have an embedding j : X → Pn

28
and therefore we can consider f as the composition of the closed embedding
i : X → Y × Pn (x 7→ (f (x), j(x)) and the projectionR p† : Y × P → Y . The
n

case f = i is just to apply the canonical morphism i i → id to OY ×Pn and

take a shift of degree dY .
For the case of the projection X = Pn × Y → Y , by OX = OPn ⊠ OY , we can
suppose that Y consists of a single point p : Pn → pt. By proposition 2.2.6
Z
OPn = Rp∗ (DRPn /pt (OPn )) = RΓ(Pn , [OPn → ... → ΩnPn ])
p

and we have isomorphisms

Z  Z 
0 ≥0
H OPn [n] ≃ τ OPn [n] ≃ H n (Pn , ΩPn ).
p p

By [1] and the last isomorphism we get the desired morphism

Z Z 
≥0
OPn → τ OPn [n] ≃ C ≃ Opt .
p p

Finally, from the preceding reasoning, we define the trace of f as the com-
position of
Z Z Z Z
OX [dX ] = OX [dX ] → OPn ×Y [dY + dX ] → OY [dY ].
f p i p

Before proving our theorem we recall the projection formula (cf. [4] 1.7.5).

3.2.4 Proposition. Let f : X → Y be a morphism of smooth algebraic

varieties. Then for M• ∈ Dqc
b
(DX ) and N • ∈ Dqcb
(DY ) we have
Z Z 
• ∗ •
L
(M ⊗OX Lf N ) ≃ M ⊗LOY N • .
•
f f

Proof of theorem 3.2.3. By definition, we know that

Z
DX M• = Rf∗ (RHomDX (M• , DX ) ⊗LDX DX→Y ) ⊗LOY Ω⊗−1
Y [dX ]
f
= Rf∗ (RHomDX (M• , DX→Y )) ⊗LOY Ω⊗−1
Y [dX ],

and

29
 Z Z 
• •
DY M = RHomDY M , DY ⊗LOY Ω⊗−1
Y [dY ].
f f

Let’s construct a canonical morphism

Z 
• • •
Φ(M ) : Rf∗ (RHomDX (M , DX→Y [dX ])) → RHomDY M , DY [dY ]
f

in Dcb (DYop ). By proposition 3.2.4 we have

Z Z Z
∗
DX→Y [dX ] = OX ⊗OX Lf DY [dX ] ≃ OX [dX ] ⊗LOY DY .
L
f f f

With the
R projection formula, the trace map of f induces a canonical mor-
•
phism f DX→Y [dX ] → DY [dY ] and Φ(M ) is defined as the composite of

Rf∗ (RHomDX (M• , DX→Y [dX ]))

→ Rf∗ RHomf −1 DY (DY ←X ⊗LDX M• , DY ←X ⊗LDX DX→Y [dX ])
→ RHomDY (Rf∗ (DY ←X ⊗LDX M• ), Rf∗ (DY ←X ⊗LDX DX→Y )[dX ])
Z Z 
•
= RHomDY M , DX→Y [dX ]
f f
Z 
•
→ RHomDY M , DY [dY ] .
f

The first morphism is given by taking the tensor product with DY ←X , the
second one is just the definition of f∗ and the last one is induced by the trace
morphism. Let’s prove that Φ(M• ) is an isomorphism.
By decomposing f as a closed embedding and a projection we can reduce the
proof to these cases and we may assume from the beginning that M• = DX
(If f is a closed embedding then this assertion is corollary 1.2.7 and if f is
the projection the assertion comes from the fact that a product between a
projetive space and a smooth affine variety is D-affine [4] 1.6.5).
If f = i : X → Y is a closed embedding then Φ(M• ) is given by the

30
composition
Ri∗ (HomDX (DX , i∗ DY ))[dX ]
Z Z 
∗
≃ RHomDY DX , i DY [dX ]
i
Z Zi 
†
≃ RHomDY DX , i DY [dY ]
i i
Z 
→ RHomDY DX , DY [dY ],
i

where the first isomorphism is given by Kashiwara’s equivalence, the second

id
one is just definition of i† applied to the resolution 0R→ DY −→ DY → 0 and
the last one is induced by the canonical morphism i i† → id. We want to
show that this map is an ismorphism. Setting U := Y \ X and j : U → Y the
open embedding, the propositions 2.2.4, 2.2.5 and the relations (3.5) give us
Z Z   Z 
∗ † ∗
RHomDY DX , j DY ≃ i∗ RHomDX DX , i j DY
i j j
Z
†
≃ i∗ i j ∗ DY = 0
j

and (3.4) gives us the result. Finally if f = p : Pn × Y → Y is the projection,

then by DX = DPn ⊠ DY w can suppose that Y = {pt} is a single point. In
this case, DPn →pt = OPn , Dpt←Pn = ΩPn and hence it is easy to see that
Rp∗ (RHomDX (DX , DX→Y [dX ])) ≃ C[n]
and
Z 
RHomDY DX , DY [dY ] ≃ C[n].
p

Hence we only have to show that Φ(DPn ) is non trivial. But, Φ(Pn )[−n] is
given by
RHomDPn (DP n , OPn )
→ RHomC (ΩPn , ΩPn ⊗LDPn OPn )
→ RHomC (RΓ(Pn , ΩPn ), RΓ(Pn , ΩPn ⊗LDPn OPn ))
→ RHomC (RΓ(Pn , ΩPn ), τ ≥n RΓ(Pn , ΩPn ⊗LDPn OPn ))
≃ RHomC (RΓ(Pn , ΩPn ), C[−n])
≃ RHomC (RΓ(Pn , ΩPn ), RΓ(Pn , ΩPn ))

31
which is induced by the canonical morphism OPn → HomOPn (ΩPn , ΩPn ) which
is non-trivial.
We end this subsection with the following results (cf. [4] 3.1.6 and 3.1.7).

3.2.5 Proposition. Let M be a holonomic DX -module. Then there exist

an open dense subset U ⊂ X such that M|U is coherent over OU . In other
words, M|U is an integrable connection on U.

3.2.6 Proposition. Let M ∈ Modqc (DX ). For an open dense subset U ⊂ X

suppose that we are giving a holonomic submodule N of M|U . Then there
exists a holonomic submodule Ñ of M such that Ñ |U = N .

Proof. (Proposition 3.2.6). By a standard reasoning in algebraic geometry

we may assume that M is coherent and M|U = N . Set L = H 0 (DX (M)). By
[4] 2.6.7 we have codim Ch(L) ≥ dX and hence L is holonomic. Then N e =
DX (L) is holonomic by 3.3. By definition, L = H 0 (DX (M)) ≃ τ ≥0 DX (M)
and by (5.1) we have a distinguished triangle
+1
τ ≤−1 DX (M) → DX (M) → L −→.
By applying DX we obtain
+1
e → M → DX (τ ≤−1 DX (M)) −
N →,
and since the duality functors commute with restrictions to open sets and by
holonomicity of M|U we have
e |U = DU (L|U ) = D2 (M|U ) = M|U = N .
N U
e → M is injective and
With this, one verifies that the canonical morphism N
we get the result.

32
Chapter 4

4.1 Preservation of holonomicity

R
In this section we will show that the functors f and f † , for f : X → Y a
morphism of smooth algebraic varieties, preserve the holonomicity. More ex-
actly, Dh (DX ) denotes the full subcategory of D(DX ) consisting of complexes
of objects M• ∈ D(DX ) such that H i (M• ) is a holonomic DX -module. We
have,
4.1.1 Theorem. Let f : X → Y be a morphism of smooth algebraic vari-
eties.
R
(i) f sends Dhb (DX ) to Dhb (DY ).

(ii) f † sends Dhb (DY ) to Dhb (DX ).

To prove the above theorem we will need the following notions.
n
M
Let’s suppose X = Cn . In this case, TCn = C[x1 , ..., xn ]∂i and there-
i=1
fore
M
Γ(Cn , DCn ) = Cxα ∂ β =: Dn ,
α,β

where xα = xα1 1 ...xαnn and ∂ β = ∂1β1 ...∂nβn . The algebra Dn is called the Weyl
algebra, and as Cn is affine we have the following equivalences of categories,

Modqc (DCn ) → Mod(Dn ), Modc (DCn ) → Modf (Dn ). (4.1)

We have have the following property (cf. [4] 3.2.8)

33
4.1.2 Proposition. Let j : (C \ {0}) × Cn−1 → RCn be the open embedding.
If M is a holonomic DCn -module, then so is H 0 ( j j † M).

On the other hand, if N is a Dn -module, we can define its Fourier trans-

formation as follows. We set N = N b as an additive group but we redefine
the left Dn -module structure by

xi .s = −∂i s, ∂i .s = xi s.

and by 4.1 we get the following equivalences

c : Modqc (DCn ) → Modqc (DCn ),
(•) c : Modc (DCn ) → Modc (DCn ).
(•)

Using the above equivalences, the proposition 2.2.6 and the tools developed
in subsection 2.2.3 we can show

4.1.3 Proposition. Let p : Cn → Cn−1 be the projection and i : Cn−1 → Cn

be the embedding. For M ∈ Modqc (DCn ) we have
Z
\ 
Hk c
M ≃ H k (Li∗ M)
p

for any k.

4.1.4 Proposition (cf. [4] 3.2.7). A coherent DCn -module is holonomic if

and only if its Fourier transformation is holonomic.

Finally, we have the following technical lemmas.

4.1 Lemma. Let M and N be coherent DX -modules. Then

Ch(M ⊠ N ) =Ch(M)×Ch(N ).

Proof. If (M, F1 ) and (N , F2) are good filtrations of M and N . Then

X
Fk M ⊠ N := F1 i (M) ⊠ F2 j (N ) ⊂ M ⊠ N
i+j=k

is good filtration for M ⊠ N . With this is clear that grF (M ⊠ N )

grF (M ⊠ N )=grF1 (M)⊠grF2 (N ),
which implies the result.

34
4.2 Lemma. Let i : X → Y be a closed embedding. Then for M• ∈ Dcb (DX )
we have
Z
M ∈ Dh (DX ) ⇔ M• ∈ Dhb (DY ).
• b
i

Proof. The proof is essentially to reduce to lemma 3.1 and note that the
morphism ρ is a smooth morphism with fibers of dimension dY − dX (see the
proof of 4.1.3).
Proof of theorem 4.1.1. First of all we reduce the proof of (i) to the case
where f = p : Cn → Cn−1 is the projection. We can assume in (i) that M• ∈
Modh (DX ) (the general case is completely analogous), by the above lemma
we can suppose that f is the projection f : X ×Y → Y and since the problem
S
is local on Y , we may assume that Y is affine. Let’s take X = ri=0 Xi
T
an affine open covering of X and let’s denote by ji0 ...ik : kp=0 Xip → X
(0 ≤ i0 < ... < ik ≤ r) the open embedding. We know by [1] that M is
quasi-isomorphic to the Čech complex C • (M) with
M M Z
k
C (M) = T
ji0 ...ik ∗ M| k Xip ≃ ji∗0 ...ik M
p=0
i0 <...<ik i0 <...<ik ji0 ...ik
R
and therefore it will be sufficient to show f ◦ji ...i ji∗0 ...ik M ∈ Dhb (DY ) which
0 k
also implies that we can assume from the beginning that X is affine. Let’s fix
closed embeddings α : X → Cn , β : Y → Cm , and consider the commutative
diagram
f
X Y

X ×Y β

α×β

Cn+m p Cm ,

where g is the graph embedding. By proposition 2.2.4 we have the relation

Z Z Z Z
M= M,
β f p (α×β)◦g

35
 R
and since (α×β)◦g M ∈ Modh (DCm+n ) by the above lemma, and the fact
that Cm+n → Cn is a composite of morphisms Ck → Ck−1 , we only need to
consider the case when f is the projection Cn → Cn−1 .
Refreshing the notation we have f = p R: Cn → Cn−1 is the projection,
M ∈ Modh (DCn ) and we want to show p M ∈ Dhb (Cn ). By propositions
4.1.3 and 4.1.4 is sufficient to show i† M ∈ Dhb (Cn−1 ).
According to 3.4 we have a distinguished triangle
Z Z
+1
i M → M → j † M −→
†
(4.2)
i j
R
which, together with proposition 4.1.2, give us i i† M ∈ Dhb (DCn ) and hence
i† M ∈ Dhb (DCn−1 ) by the above lemma.
Finally let’s prove that (i) implies (ii). Decomposing f as a closed embedding
and a projection we can assume first that f is the projection X × Y → Y .
In this case f ∗ M ≃ OX ⊠ M and it is holonomic by lemma 4.1. For a closed
embedding the result comes from the distinguished triangle 4.2.
We end this section with the following adjunction formula. Letting f :
X → Y be a morphism of smooth algebraic varieties, we define new functors
Z Z
:= DY DX : Dhb (DX ) → Dhb (DY ). (4.3)
f! f
†
f⋆
:= DX f DY : Dhb (DY ) → Dhb (DX ). (4.4)
4.1.5 Theorem. For M• ∈ Dhb (DX ) and N • ∈ Dhb (DY ) we have a natural
isomorphism
Z 
• •
RHomDY M ,N ≃ Rf∗ RHomDX (M• , f †N • ),
f!  Z 
⋆ • • • •
Rf∗ RHomDX (f N , M ) ≃ RHomDY N , M .
f

Proof. We have
Rf∗ RHomDX (M• , f † N • )



≃ Rf∗ ΩX ⊗LOX DX M• ⊗LDX f † N • [−dX ]



≃ Rf∗ ΩX ⊗LOX DX M• ⊗LDX DX→Y ⊗Lf −1 DY f −1 N • [−dY ]



≃ Rf∗ ΩX ⊗LOX DX M• ⊗LDX DX→Y ⊗LDY N • [−dY ]
 Z 
≃ ΩY ⊗OYL
DY M ⊗DY N • [−dY ]
•
f

36
  Z 
≃ ΩY ⊗OY DY
L
M ⊗LDY N • [−dY ]
•
f!
Z 
• •
≃ RHomDY M ,N .
f!

The first isomorphism comes from the definition of DX , the third one is a
general property (called the derived projection formula for D-modules) of
the inverse and direct image (cf. [4] proof of 1.5.21) and the fifth one comes
from remark 3.4. This establish the first isomorphism, the second follows
from duality.
By applying H 0 (RΓ(Y, •)) in the above isomorphism we obtain the fol-
lowing
4.1.6 Corollary. With the same notation of theorem 5.1.5 we have a natural
isomorphism
Z 
• •
HomDhb (DY ) M ,N ≃ HomDhb (DX ) (M• , f † N •),
f!  Z 
⋆ • • • •
HomDhb (DX ) (f N , M ) ≃ HomDhb (DY ) N , M .
f

Finally,
4.1.7 Theorem. There exists a morphism of functors
Z Z
→ : Dhb (DX ) → Dhb (DY ).
f! f

Moreover, if f is proper, then this morphism is an isomorphism.

Proof. By Hironaka’s desingularization theorem, there exists a smooth com-
pletion Xe of X. Since X is quasi projective, a desingularization X
e of the
Zariski closure X of X in the projective space is such a completion and
therefore f can be factored as
i j p
X−
→X ×Y −
→Xe ×Y →
− Y,
where i is the graph embedding and p is the projection. In this case i and
p are proper (by separatedness) and j is a open embedding and we reduce
the problem to consider proper morphisms and open embeddings. For the
case of a proper morphism f by remark 3.4 and theorem 3.2.3 we have an
isomorphism

37
 Z Z Z
= DY DX → .
f! f f

On the other hand, if f : X → Y is an open embedding and M• ∈ Dhb (DX )

then by the above corollary we have
Z Z   Z 
• • • † •
HomDhb (DY ) M, M ≃ HomDhb (DX ) M , j M
j! j j
• •
≃ HomDhb (DX ) (M , M )

and we obtain the desired morphism as the image of id ∈ HomDhb (DX ) (M• , M• ).

4.2 Minimal extensions

We say that a DX -module M is simple if it contains no coherent DX -
submodules other that M and 0.
Let’s take M ∈ Modh (DX ). According to the proposition 2.2.6 the total
multiplicity of M defined by
X
m(M) = mC (gr^F (M))

C∈I(Ch(M))

is additive in the sense that we have m(M) = m(L) + m(N ) for any short
exact sequence

0→L→M→N →0

in Modh (DX ) (here we are using the fact that dim C = dX for any C ∈
I(Ch(M))). This implies that if we have a descending chain of holonomic
submodules of M

M = M0 ⊇ M1 ⊇ ... ⊇ Mk .

then m(Mi ) = m(Mi+1 ) + m(Mi /Mi+1) and therefore m(M) ≥ k which

implies that M cannot have an infinite descending chain. This means that,
Modh (DX ) is an artinian category. With this property we can construct a
Jordan-Hölder series of M as follows. Let’s suppose that we have constructed
a chain 0 = M0 ⊆ M1 ⊆ ... ⊆ Mk of submodules of M such that Mi /Mi−1
is simple. By the last reasoning we can take a submodule Mk+1 of M

38
such that Mk+1 /Mk is a minimal submodule of M/Mk and therefore is
simple. This process cannot continue indefinitely because, by remark 1.4
and proposition 1.2.3 M is also locally a noetherian module. So, we have a
finite sequence
M = M0 ⊃ M1 ⊃ ... ⊃ Mk = 0
of holonomic DX -submodules such that Mi /Mi+1 is simple for each i.
On the other hand, let Y be a locally closed smooth subvariety of X. Let’s
assume that the inclusion map i : Y ֒→ X is affine. Then, by example 2.1.2,
DX←Y is locally free over DY and the higher cohomology R groups ofj Ri∗ vanish.
j
Therefore, for a holonomic DY -module M we have H i M = H i! M = 0
for all j 6= 0. Finally, by the theorems 4.1.1 and 4.1.7 we have a morphism
Z Z
M→ M (4.5)
i! i

in Modh (DX ).
4.2.1 Definition. We call the image L(Y, M) of the morphism 4.5 the min-
imal extension of M.
4.2.2 Theorem. (i) Let Y be a locally closed smooth connected subvariety
of X such that i : X → Y is affine, and let M be a simple holonomic
DY -module. Then the minimal extension L(Y, M) is also simple, and
it is characterized asRthe unique simple
R submodule (resp. unique simple
quotient module) of i M (resp. of i! M).
(ii) Any simple holonomic DX -module is isomorphic to the minimal exten-
sion L(Y, M) for some pair (Y, M), where Y is as in (i) and M is a
simple integrable connection on Y .
(iii) Let L(Y, M) and L(Y ′ , M′ ) be as in (iii). Then we have L(Y, M) ≃
L(Y ′ , M′) if and only if Y = Y ′ and M|U ≃ M′ |U for an open dense
subset U of Y ∩ Y ′ .
Proof. Let’s choose an open subset U ⊂ X containing Y such that k : Y ֒→ U
is a closed embedding and let j : U → X be the embedding. We first show
the following results.
(a) For any E ∈ ModYqc (DX ) we have H l i† (E) = for l 6= 0. Hence H 0 i† :
ModYqc (DX ) → Modqc (DY ) is an exact functor.

39
 R
(b) For any non-zero holonomic submodule N of i M, we have i† N ≃ M.
R R
(c) i M (resp. i! M) has a unique simple holonomic submodule (resp.
simple holonomic quotient module).
R R
(d) For a sequence 0 6= N1 ⊂ N2 ⊂ i M of holonomic submodules of i M,
we have i† (N2 /N1 ) = 0.

As k is a closed embedding and for E ∈ ModYqc (DX ) we have i† = k † j † E =

k † j −1 E, supp(j −1E) ⊂ Y ∩ U = Y and j −1 is an exact functor then by
Kashiwara’s equivalence H l i† (E) = H l k † (j −1 E) = 0 for all l 6= 0 and (a) is
proved.
Now, by corollary 4.1.6
 Z   Z Z 
HomDX N , M = HomDX N , M
i j k
 Z 
⋆
≃ HomDU j N , M .
k

Since jRis an open embedding, we have j ⋆ = j † = j −1 . Therefore,

R the inclusion
R
†
N ֒→ i M induces a non-zero morphism φ : j N → k M. Since k M is
a simple holonomic DU -module by Kashiwara’s equivalence,Rφ is surjective.
Applying k † to φ we obtain a surjective morphism i† N →Rk † k M = M. On
the other hand, we have an injective morphism i† N → i† i M ≃ M because
i† is exact by (a). Hence we must have i† N ≃ M and (b) is proved.
To see (c) Rlet’s suppose that there exists two simple holonomic submodules
L= 6 L′ of i M. Set N = L ⊕ L′ . By (b) we have
M ≃ i† N ≃ i† L ⊕ i† L′ ≃ M ⊕ M′
which is a contradiction.
Finally, by exactness of i† we have i† NR2 /i† N1 ≃ i† (N2 /N1 ) then (b) implies
i† (N2 /N1 ) = 0, since i† N1 ⊂ i† N2 ⊂ i† i M = M.
Let’s see the Rtheorem. By (c) there exists a unique simple holonomic sub-
module L of i M and by corollary 4.1.6 there exist two isomorphisms
Z 
HomDX M, L ≃ HomDY (M, i† L) ≃ HomDY (M, M),
Z i!Z   Z 
HomDX M, M ≃ HomDY M, i† M ≃ HomDY (M, M),
i! i i

40
 R R
from
R which
R we have a decomposition i!
M → L ֒→ i
M of the morphism
i!
M → i M (which is non-zero since it corresponds to the identity). Since
L is simple, the image of the latter morphism is L and the proof of (i) is
complete.
To see (ii), let’s take L a simple holonomic DX -module. By the propositions
3.2.5 and 3.2.6 we can take an affine open dense subset Y of an irreducible
component of supp(L) such that if i : Y → X is the embedding then M = i† L
is an integrable connection on Y and it is simple. By corollary 4.1.6 we have
an isomoprhism
Z 
HomDX M, L = HomDY (M, i† L) ≃ HomDY (M, M) 6= 0
i!
R
which implies, by simplicity, that there exists a surjective morphism R i!
M→
L. Therefore, L is a simple holonomic quotient module of i! M and we ob-
tain L = L(Y, M) by (i).
Finally, let’s see (iii). First of all, let’s note that under the hypotheses
of (ii)R (in particular M is an integrable connection), supp(M) = Y , then
supp( i M) = Y and therefore Y ⊂ supp(L(Y, M)) ⊂ Y . But supp(L(Y M))
is a closed set and we obtain supp(L(Y, M)) = Y .
Now, if L(Y, M) ≃ L(Y ′ , M′ ) then Y = supp(L(Y, M)) = supp(L(Y ′ , M′)) =
Y ′ . Moreover, as Y is locally closed in X, we have that Y and Y ′ are open
in Y and Y ′ respectively. Let’s take U := Y \ (Y0 ∪ Y0′ ) where Y0 := Y \ Y
is a closed set in Y (resp. Y0′ is a closed set of Y0 ). It is clear that U is
an open subset contained in Y and in fact is dense in Y ∩ Y ′ . By (b), we
know that i† L(Y, M) ≃ M and L(Y, M)|U ≃ L(Y ′ , M′)|U which gives us
M|U ≃ M′ |U .
On the other hand, let’s denote by L := L(Y, M) and L′ := L(Y ′ , M′ ). Let
C be a element of I(supp(L)) and let’s take Y ′′ = U ∩ C. If i1 : Y ′′ → Y
and i2 : Y ′′ → Y ′ are the closed embeddings (of course, i1 = i2 ), then as in
(ii) we have L = L(Y ′′ , i†1 L). Similarly, L′ = L(Y ′′ , i†2 L′ ). It is enough to see
that i†1 L ≃ i†2 L′ . But i†1 L is the inverse image of M|U on Y ′′ and i†2 L′ is the
inverse image of M′ |U on Y ′′ , they are isomorphic by hypothesis and we get
(iii).

4.2.1 Example. Let X = A1 = Spec C[x] be the affine line, D1 := DA1 (A1 )
the first Weyl algebra and Y ⊂ X a subspace locally closed and connected.
Let’s take U ⊂ X an open subset such that Y ⊆ U is a closed subset in U,
and let’s consider the following cases. If Y ( U then dim Y < dim X = 1

41
and Y = {p} consist of a single point. In this case, by D-affinity, we have
L({p}, O{p} = C) = DX ⊗OX ,p C which is simple. On the other hand, if
Y = U then Y contains the generic point η ∈ X and Y = X. By (iii) in our
theorem, if M is a simple holonomic module on U and U ′ is another open
subset such that M′ is simple holonomic on U ′ , then L(U, M) ≃ L(U ′ , M′ )
if and only if M|η ≃ M′ |η . By (ii) we get that the simple holonomic DX -
modules are, up to isomorphisms, L({p}, O{p} ) and L(η, M) where M is a
simple holonomic Dη -module.
Finally, Γ(η, Dη ) = (D1 )(0) = C(x)[∂x ] and the global sections of a simple Dη -
module M form a simple Γ(η, Dη )-module. Every simple Γ(η, Dη )-module has
as D1 -socle (sum of all the simple D1 -submodules) a simple D1 -submodule
and therefore, the global sections of L(η, M) are equal to the D1 -socle of
Γ(η, M). P
Every element of Γ(η, Dη ) = C(x)[∂x ] can be written as d = α
α∈N qα ∂x ,
where qα ∈ C(x). Then, if we define the order of d as the highest power α
such that qα 6= 0 we can easily verify than Γ(η, Dη ) admits a left division
algorithm and therefore is a PID (noncommutative). Its simple submodules
are in a 1-1 correspondence with the irreducible elements. Our theorem gives
us that the simple D1 -modules are parameterised, up to isomoprhisms, by the
closed points of X and the irreducible elements of Γ(η, Dη ). This gives us
back the classical classification given in [6].

42
APPENDIX

43
Chapter 5

5.1 Derived categories and derived functors

Our intention in the next two sections is to give a brief reference to the results
about the derived categories used along this work. The reader can look up
the proof of the assertions in [4] or [5].
5.1.1 Definition. Let A be a category and S be a family morphism of A.
We say that S is a multiplicative system if it satisfies the following properties.
(i) For every object X ∈ ob(A) the identity morphism idX ∈ HomA (X, X)
is in S.
(ii) For every couple (f, g) of morphisms of S, if the composition f ◦ g is
well-defined in A, then f ◦ g ∈ S.
(iii) For every X, Y, Z ∈ ob(A), u ∈ HomA (X, Y ) (resp. u ∈ HomA (Y, X))
and s ∈ HomA (Z, Y ) ∩ S (resp. s ∈ HomA (Y, Z) ∩ S). There exist
W ∈ ob(A), v ∈ HomA (Z, W ) and t ∈ HomA (W, X) ∩ S such that
v
W → Z
↓t ↓s
u
X → Y
is a commutative diagram (resp. a diagram with the arrows reversed).
(iv) For every X, Y ∈ ob(A) and u, v ∈ HomA (X, Y ) the following condi-
tions are equivalent.
(a) There exist Y ′ ∈ ob(A) and s ∈ HomA (Y, Y ′ ) ∩S such that s ◦ u =
s ◦ v.

44
 (b) There exist X ′ ∈ ob(A) and s ∈ HomA (X ′ , X) ∩ S such that
u ◦ t = v ◦ t.
5.1.2 Definition. Let A be a category. We say that a functor of categories
T : A → A is a translation of A if T is an equivalence of categories.
5.1.3 Definition. Let A be a category with a translation T . A sequence
f g h
X−
→Y −
→Z−
→ T (X)
where X, Y, Z ∈ ob(A), f ∈ HomA (X, Y ), g ∈ HomA (Y, Z) and h ∈
HomA (Z, T (X)), is called a triangle in A. We usually denote a triangle
by
+1
X → Y → Z −→.
5.1.4 Definition. Let A be an additive category with a translation T . We
say that A is a triangulated category if there exists a family T of triangles of
A, called distinguished triangles, that satisfies the following properties:
(i) A triangle isomorphic to a distinguished triangle is a distinguished tri-
angle.
(ii) For all f ∈ HomA (X, Y ) there exist g ∈ HomA (Y, Z) and
f g h
h ∈ HomA (Z, T (X)) such that X −
→Y −
→Z−
→ T (X) is a distinguished
triangle.
(iii) For all X ∈ ob(A), the triangle X → X → 0 → T (X) is distinguished
triangle.
f g h
(iv) A triangle X −
→Y −
→Z−
→ T (X) is a distinguished triangle if and only
g h −T (f )
if Y −
→Z−
→ T (X) −−−→ T (Y ) is a distinguished triangle.
f g h
(v) for every couple of distinguished triangles X −
→ Y −
→ Z −
→ T (X)
f′ g′ h′
and X ′ −→ Y′ −
→ Z′ − → T (X ′ ), and for every couple of morphisms
α ∈ HomA (X, X ′ ) and β ∈ HomA (Y, Y ′ ) such that f ′ ◦ α = β ◦ f , there
exists γ ∈ HomA (Z, Z ′ ) such that the following diagram
f g h
X → Y → Z → T (X)
↓α ↓β ↓γ ↓T (α)
f′ g′ h′
X ′ → Y ′ → Z ′ → T (X ′ )
is commutative.

45
(iii) The family T must satisfy a complicated property called the octahedral
axiom. The reader is invited to take a look at [4] B.3.3 or [5] 10.1.6.

Furthermore, if T is a family of distinguished triangles and S is a multiplica-

tive system, we say that S is compatible with the triangulated structure of A
if the following conditions are satisfied.

(i) If s ∈ S then T (s) ∈ S.

(ii) For every morphism of distinguished triangles, this is, for a commuta-
tive diagram
f g h
X → Y → Z → T (X)
↓α ↓β ↓γ ↓T (α)
f′ g′ h′
X ′ → Y ′ → Z ′ → T (X ′ )

if α, β ∈ S then γ ∈ S.

5.1.5 Theorem. Let A be a category and S be a multiplicative system in A.

There exist a category AS and a functor QA : A → AS , called the localization
functor, such that:

(i) For every s ∈ S, QA (s) is an isomorphism of AS .

(ii) If F : A → A′ is a functor that transforms the elements of S in iso-

morphisms of A′ , then there exists a unique functor FS : AS → A′ such
that F ≃ FS ◦ QA .

5.1 Remark. If A is an additive category and S is a multiplicative system of

A. Then AS is an additive category and Q is an additive functor. Moreover,
if A is a triangulated category with translation T , and S is multiplicative sys-
tem compatible with the triangulated structure of A, then AS is a triangulated
category.

Let’s suppose that A is an abelian category and denote by C(A) the

category of complexes of objects of A and by K(A) the homotopy category
defined by

(i) ob(K(A)) = ob(C(A)).

P
(ii) HomK(A) (X • , Y • ) := HomC(A) (X • , Y • )/ C(A) (X
•
, Y • ). Where,

46
 P •
C(A) (X , Y • ) = {f • ∈ HomC(A) (X • , Y • )|f ∼ 0}.

5.1.6 Definition. A quasi-isomorphism in K(A) is a morphism [f • ] in K(A)

such that H i ([f • ]) is an isomorphism for every i ∈ Z.

5.2 Remark. QisA , the collection of quasi-isomorphisms of K(A), is a mul-

tiplicative system of K(A).

5.1.7 Definition. Let A be an abelian category. Under the notations of

theorem 5.1.5, the category K(A)QisA , denoted by D(A), is called the derived
category of A.

Now, for X • , Y • ∈ ob(C(A)) and f • ∈ HomC(A) (X • , Y • ). We can define

the shifted complex X • [k], k ∈ Z, by

(i) X n [k] := X n+k ,

n+k
(ii) dnX • [k] := (−1)k dX • ,

and for f • we can define the mapping cone Mf • ∈ ob(C(A)) by

(i) Mfn• = X n+1 ⊕ Y n ,

(ii) dMf • (xn+1 , y n) = (−dn+1

X • (x
n+1
), f n+1 (xn+1 ) + dnY • (y n )).

From this considerations, there exists a short exact sequence

α(f • ) β(f • )
0 → Y • −−−→ Mf • −−−→ X • [1] → 0,

where α(f •)n (y n ) = (0, y n ) and β(f • )n (xn+1 , y n ) = xn+1 .

5.3 Remark. The shift functor (•)[1] : K(A) → K(A) defines a translation
on K(A). The family of triangles X • → Y • → Z • → X • [1] for which there
exists a morphism f • ∈ HomK(A) (X0• , Y0• ) such that

f• α(f • ) β(f • )
X0• → Y0• → Mf • → X0• [1]
↓≀ ↓≀ ↓≀ ↓≀
X• → Y • → Z• → X • [1]

is a commutative diagram in K(A), defines a family of distinguished triangles

on K(A) in the sense of definition 5.1.4.

47
5.1.8 Theorem. Let A be an abelian category. The category D(A) is an
additive triangulated category. The distinguished triangles of D(A) are pre-
cisely the image under QA of distinguished triangles in K(A). The functor
QA : K(A) → D(A) is additive and triangulated, this means that, QA sends
distinguished triangles to distinguished triangles.

5.1.9 Definition. For a complex X • ∈ ob(C(A)), with A an abelian category,

we define the truncated complex by

τ ≤k X • := [... → X k−1 → Ker dk → 0 → 0 → ...],

τ ≥k X • := [... → 0 → 0 → Im dk+1 → X k+1 → ...].

The reader can easily verify that we have a short exact sequence

0 → τ ≤k X • → X • → τ ≥ X • → 0. (5.1)

5.1.10 Proposition. Let A be an abelian category. Any short exact sequence

f• g•
0 → X • −→ Y • −→ Z • → 0 in C(A) can be embedded into a distinguished
triangle
f• g•
X • −→ Y • −
→ Z • → X • [1]

in D(A).

Proof. Let’s denote by d the differential morphisms of the mapping cone

Mid•X . The morphism h• defined by hn+1 (xn+1 , xn ) = (xn , 0), satisfies

hn+2 d(xn+1 , xn ) = (xn+1 + dn (xn ), 0) (5.2)

and

dhn+1 (xn+1 , xn ) = d(xn , 0) = (−dn (xn ), xn ). (5.3)

•
Taking (5.2)+(5.3) it is clear that the complex Mid is quasi-isomorphic to 0.
With this and the exact sequence
A B
0 → Mid• − → Z• → 0
→ Mf • −

48
  
idX • 0
where A = and B = (0, g •), there exists an isomorphism φ :
0 f•
Mf • ≃ Z • in D(A). Hence there exists a commutative diagram
f• α(f • ) β(f • )
X• → Y • → Mf • → X • [1]
↓idX • ↓idY • ↓≀φ ↓idX • [1]
f• g• β(f • )◦φ−1
X• → Y• → Z• → X • [1]
f• g•
in D(A) which shows that X • −→ Y • −→ Z • → X • [1] is a distinguished
triangle.
5.1.11 Definition. Let A and B be abelian categories and ∗ ∈ {+, −, b}
(see the preliminaries to the section 2.2) and let F : K∗ (A) → K(B) a
triangulated functor. The right derived functor of F is a couple (R∗ F, ξF )
where R∗ F : D ∗ (A) → D ∗ (B) is a ∂-functor (i.e. compatible with translations
and preserving distinguished triangles) and ξF : QB ◦ F → R∗ F ◦ QA is a
morphism of functors (QA , QB the localization functors) that satisfies the
following universal property: for every triangulated functor G : D ∗ (A) →
D(B) and every morphism of functors Φ : QB ◦ F → G ◦ QA there exists a
unique morphism of functors η : R∗ F → G such that Φ = (η ◦ QA ) ◦ ξF .
5.1.12 Definition. Let F : A → B be an additive functor between abelian
categories. We say that a full additive subcategory J of A is F -injective if
the following condition are satisfied:
(i) For any X ∈ ob(A), there exists an object I ∈ ob(J ) and an exact
sequence 0 → X → I.

(ii) If 0 → X ′ → X → X ′′ → 0 is an exact sequence in A and X ′ , X ∈

ob(J ), then X ′′ ∈ ob(J ).

(iii) For any exact sequence 0 → X ′ → X → X ′′ → 0 such that X ′ , X, X ′′ ∈

ob(J ), the sequence 0 → F (X ′) → F (X) → F (X ′′ ) → 0 in B is also
exact.
Similarly we define F -projective categories reversing all arrows in the condi-
tions above.
5.4 Remark. If F : A → B is a functor as in the preceding definition, then
X • → F (X • ) defines a triangulated functor F : K∗ (A) → K(B).

49
5.1.13 Theorem. Let F : A → B be an additive functor between abelian
categories which is left (resp. right) exact functor and assume that there
exists an F -injective (resp. F -projective) subcategory of A. Then the right
(resp. left) derived functor R+ F : D + (C) → D + (B) of F exists (resp. the
left derived functor L− F : D − (A) → D − (B) of F exists).

5.5 Remark. Under the hypothesis of the last theorem, the right derived
functor is given by the following. By (cf. [5] 13.2.1) there exists a quasi-
isomorphism I • ≃ X • with I • ∈ K+ (J ), called an injective resolution of X • ,
and the right derived functor is given by RF (X •) = F (I • ).

50
Bibliography

[1] Robin Hartshorne, Algebraic Geometry, (Vol. 52). Springer Science

and Business Media.

[2] Matsumura, H., & Reid, M., Commutative ring theory, (Vol. 8).
Cambridge university press.

[3] Coutinho, S. C., A primer of algebraic D-modules, (Vol. 33). Cam-

bridge University Press.

[4] Hotta R., Takeuchi K., & Tanisaki T., D-modules, perverse
sheaves and representation theory, (Vol. 236). Springer Science & Busi-
ness Media.

[5] Kashiwara M., & Shapira P., Categories and sheaves, (Vol. 332).
Springer Science & Business Media.

[6] Block R. E. (1981), The irreducible representations of the Lie algebra

sl2 (C) and of the Weyl algebra, Advances in mathematics, 39(1), 69-110.

51