You are on page 1of 63

Contents

1 Triangulated and derived categories 1


1.1 Additive categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Triangulated categories . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Derived categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Derived functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Derived categories of coherent sheaves 13


2.1 Coherent sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Derived categories of coherent sheaves . . . . . . . . . . . . . . . . . 14
2.3 Derived functors in algebraic geometry . . . . . . . . . . . . . . . . . 17
2.4 Fourier-Mukai transform and Orlov’s theorem . . . . . . . . . . . . . 20
2.5 Criterion for equivalence . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Abelian varieties 27
3.1 Definition and basic properties . . . . . . . . . . . . . . . . . . . . . 27
3.2 Line bundles on abelian varieties . . . . . . . . . . . . . . . . . . . . 31
3.3 Dual abelian variety . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Equivalences between derived categories 43


4.1 From equivalence to isomorphism . . . . . . . . . . . . . . . . . . . . 43
4.2 From isomorphism to equivalence . . . . . . . . . . . . . . . . . . . . 53

Bibliography 61

i
Chapter 1

Triangulated and derived


categories

1.1 Additive categories


We start with a brief description of a few principal notions, which will be used
throughout the paper. For the proofs see [Huy06, § 1] and and for a more in-depth
reading see, for example, [GM13].
Let A and B be two categories. A (covariant) functor F : A → B is called faithful
(resp. full ) if, for any two objects A, B in A, the induced map
F : Hom(A, B) −→ Hom(F (A), F (B))
is injective (resp. surjective). In particular, if the induced map is bijective, the
functor is called fully faithful.
Definition 1.1. Two functors F, F 0 : A → B are isomorphic if there exists a mor-
phism of functors ϕ : F → F 0 such that for any object A ∈ A the induced morphism
ϕA : F (A) → F 0 (A) is an isomorphism in B.
Definition 1.2. A functor F : A → B is called an equivalence if there exists a
functor F −1 : B → A (called an inverse or a quasi-inverse) such that F ◦ F −1 is
isomorphic to idB and F −1 ◦ F is isomorphic to idA .
Two categories are called equivalent if there is an equivalence between them.
As we will see later, the notion of equivalence between categories plays an im-
portant role, so we want to know whether a functor is an equivalence.
Clearly, an equivalence is fully faithful. For the converse, we have:
Proposition 1.3. Let F : A → B be a fully faithful functor. Then F is an equiva-
lence if and only if, for every object B ∈ B, there exists an object A ∈ A such that
B is isomorphic to F (A).
We are not, in general, interested in completely arbitrary categories. We will
work with categories that are at least additive.
Definition 1.4. A category A is called additive if for every two objects A, B ∈ A,
the set Hom(A, B) is equipped with a structure of abelian group such that the
following conditions are satisfied:

1
2 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

1. The compositions Hom(A, B) × Hom(B, C) → Hom(A, C) are bilinear.


2. There exists a zero object 0 ∈ A, which is an object such that Hom(0, 0) is the
trivial group with one element.
3. For any two objects A1 , A2 ∈ A there exists an object B ∈ A with morphisms
ji : Ai → B and pi : B → Ai such that
pi ◦ ji = id, p2 ◦ j1 = p1 ◦ j2 = 0, j1 ◦ p1 + j2 ◦ p2 = id .
These make B the direct sum and the direct product of A1 and A2 .
A functor F : A → B between additive categories is additive if the induced maps
Hom(A, B) → Hom(F (A), F (B)) are group homomorphisms.
Functors between additive categories will be always assumed to be additive.
We will eventually be interested in a special type of additive category, that is,
categories defined in terms of varieties over a base field.
Definition 1.5. A k-linear category is an additive category A such that the groups
Hom(A, B) are k-vector spaces an such that all compositions are k-bilinear.
Any additive functor F : A → B between two k-linear categories over a common
base field k is assumed to be k-linear, i.e. for any two objects A, B ∈ A the map
F : Hom(A, B) → Hom(F (A), F (B)) is k-linear.
All the properties of additive categories hold for k-linear ones. However, it could
(in principle) happen that two k-linear categories are equivalent only as additive
categories, without being equivalent as k-linear ones.
Furthermore, we can add a condition to Definition 1.4 in order to work with
complexes and exact sequences of objects in a category.
Definition 1.6. An additive category A is called abelian if a further condition holds:
4. Every morphism f ∈ Hom(A, B) admits a kernel and a cokernel, and the
natural map Coim(f ) → Im(f ) is an isomorphism.
Definition 1.7. Let F : A → B be an additive functor between abelian categories.
The functor F is left exact (resp. right exact) if any short exact sequence
f g
0 A1 A2 A3 0
is mapped to a sequence
F (f ) F (g)
0 F (A1 ) F (A2 ) F (A3 ) 0

which is exact except possibly in F (A3 ) (resp. in F (A1 )).


The functor F is called exact if it is both left and right exact.
Definition 1.8. Let F : A → B be a functor between arbitrary categories. A functor
G : B → A is left adjoint (G a F ) to F if there exist isomorphisms Hom(B, F (A)) '
Hom(G(B), A) for any two objects A ∈ A and B ∈ B which are functorial in A and
B.
A functor H : B → A is right adjoint (F a H) to F if there exist isomorphisms
Hom(F (A), B) ' Hom(A, H(B)) for any two objects A ∈ A and B ∈ B which are
functorial in A and B.
1.2. TRIANGULATED CATEGORIES 3

We have that H is the right adjoint of F if and only if F is the left adjoint of H.
The existence of an adjoint of a functor F can be used to check whether F is
fully faithful:

Proposition 1.9. Let F : A → B and G : B → A be two functors such that G a F .


Then F is fully faithful if and only if the induced functor morphism G ◦ F → idA is
an isomorphism.
Similarly, let F : A → B and H : B → A be two functors such that F a H. Then
F is fully faithful if and only if the induced functor morphism idA → H ◦ F is an
isomorphism.

The case that interest us most is the case of equivalences:

Proposition 1.10. Let F : A → B be an equivalence of categories. Then, if F 0 is


an inverse functor of F , F a F 0 a F .

1.2 Triangulated categories


In this section we introduce the notion of triangulated category. This type of
category was defined for the first time by Verdier in his Ph.D. thesis in the 60’s,
with the help of his advisor Grothendieck.
Why do we introduce this new structure? In general, in homological algebra and
algebraic geometry, we are interested in studying complexes and exact sequences of
objects. But making complexes is possible only in a category that admits kernels
and images of morphisms, that is, an abelian category. However, also in a non-
abelian category, it is sometimes possible to provide a structure, the structure of
triangulated category, that allow us to do homological algebra.

Definition 1.11. Let A be an additive category. Let

T : A −→ A

be an additive equivalence, called the shift functor. We will use the notation A[1] :=
T (A), for any object A ∈ A.
Now, we can define a triangle as a sequence

A B C A[1]

of objects and relative morphisms in A.


Given two triangles (A, B, C) and (A0 , B 0 , C 0 ), a morphism of triangles is a triple
of morphisms (f, g, h) such that the diagram

A B C A[1]
f g h f [1]

A0 B0 C0 A0 [1],

is commutative.
4 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

Definition 1.12. Let A be an additive category. The structure of triangulated


category on A is given by a shift functor and a set of triangles, called distinguished
triangles, which satisfy the following axioms:

T1. (a) Any triangle of the form

id
A A 0 A[1]

is distinguished.
(b) Any triangle isomorphic to a distinguished triangle is distinguished.
(c) Any morphism f : A → B can be completed to a distinguished triangle

f
A B C A[1].

T2. The triangle


f g h
A B C A[1]
is distinguished if and only if the triangle

g h −f [1]
B C A[1] B[1]

is distinguished.

T3. Given a commutative diagram of distinguished triangles

A B C A[1]
f g f [1]

A0 B0 C0 A0 [1],

there exists a (possibly not unique) morphism h : C → C 0 such that (f, g, h)


is a morphism of triangles.

T4. Given objects A, B, C ∈ A, morphisms f : A → B, g : B → C and distin-


guished triangles
f p1 q1
A B D1 A[1]

g◦f p2 q2
A C D2 A[1]
g p3 q3
B C D3 B[1],
there exist morphisms a : D1 → D2 , b : D2 → D3 such that

(a) The triangle


a b p1 [1]◦q3
D1 D2 D3 D1 [1]
is distinguished.
1.2. TRIANGULATED CATEGORIES 5

(b) The triple (idA , g, a) is a morphism of triangles.


(c) The triple (f, idC , b) is a morphism of triangles.

The last axiom can be seen in this way: suppose to have a nested inclusion
of abelian group A ⊂ B ⊂ C. There exists a canonical isomorphism C/B '
(C/A)(B/A). Then, if we replace the short exact sequences A → B → B/A,
A → C → C/A and B → C → C/B by distinguished triangles in a triangulated
category, the axiom, roughly speaking, requires B/A → C/A → C/B to be distin-
guished as well.

We can easily prove, using first and third axioms, that in a distinguished triangle
A → B → C → A[1] the composition A → C has to be zero.
So, roughly speaking, we should think of the distinguished triangles as a replace-
ment for short exact sequences, and this triangle turn out to be enough to develop
much of the theory linked with complexes and exact sequences. However, for an ad-
ditive category to be abelian is a property entirely depending on the category itself.
On the other hand, a triangulated structure is an extra piece of data, consisting of a
shift functor and some distinguished triangles suitably chosen to satisfy the previous
axioms.
Definition 1.13. An additive functor F : A → A0 between triangulated categories
is called exact if the following conditions hold:
1. There exists a functor isomorphism F ◦ TA ' TA0 ◦ F .

2. Any distinguished triangle

A B C A[1]

in A is mapped in a distinguished triangle

F (A) F (B) F (C) F (A)[1]

in A0 , where F (A[1]) is identified with F (A)[1] via the isomorphism of the first
point.
Proposition 1.14. Let F : A → A0 be an exact functor between triangulated cate-
gories. If G a F a H, then G and H are exact funtors.
Remark 1.15. The notion of triangulated category can be adapted for k-linear cat-
egories. In this case, the shift functor should be k-linear.
Definition 1.16. Two triangulated categories A, A0 are equivalent if there exists an
exact equivalence F : A → A0 .
If A is a triangulated category, the set Aut(A) of isomorphism classes of equiv-
alences F : A → A forms the group of autoequivalences of A.
We state also a definition that will be useful in the next section.
Definition 1.17. A triangulated category A is decomposed into triangulated sub-
categories A1 , A2 ⊂ A if the following conditions hold:
6 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

1. Both categories A1 , A2 contain objects non isomorphic to 0.

2. For all objects A ∈ A there exists a distinguished triangle

B1 A B2 B1 [1]

with Bi ∈ Ai , i = 1, 2.

3. Hom(B1 , B2 ) = Hom(B2 ,1 ) = 0 for all objects Bi ∈ Ai , i = 1, 2.

A triangulated category that cannot be decomposed is called indecomposable.

Equivalences of triangulated categories


As we said before, we want to find criteria that show whether an exact functor is
fully faithful or even an equivalence. In order to do this, we begin with the definition
of a spanning class.

Definition 1.18. A collection Ω of objects in a triangulated category A is a spanning


class of A if for all objects B ∈ A the following two conditions hold:

1. If Hom(A, B[i]) = 0 for all A ∈ Ω and all i ∈ Z, then B ' 0.

2. If Hom(B[i], A) = 0 for all A ∈ Ω and all i ∈ Z, then B ' 0.

Here we have a first way to se if a functor is fully faithful.

Proposition 1.19. Let F : A → A0 be an exact functor between triangulated cate-


gories, with left and right adjoints G and H.
Suppose Ω is a spanning class of A such that for all objects A, B ∈ Ω and all
i ∈ Z the natural isomorphisms

F : Hom(A, B[i]) −→ Hom(F (A), F (B[i]))

are bijective. Then F is fully faithful.

Now, suppose we arleady now that a functor is fully faithful. Then, to see if it
is an equivalence, we have this results.

Lemma 1.20. Let F : A → A0 be a fully faithful exact functor between triangulated


categories such that F has a right adjoint H.
Then F is an equivalence if and only if, for any object C ∈ A0 , the triviality of
H(C) implies C ' 0.

Proposition 1.21. Let F : A → A0 be a fully faithful exact functor between trian-


gulated categories. Suppose that A contains objects non isomorphic to 0 and A0 is
indecomposable.
Then F is an equivalence if and only if F has left and right adjoints G a F a H
such that, for any object B ∈ A0 , H(B) ' 0 implies G(B) ' 0.
1.3. DERIVED CATEGORIES 7

1.3 Derived categories


In this section we describe structure of the derived category, stated for the first
time by Verdier and Grothendieck.
The principal reason for introducing derived categories is that working with
complexes is better than working with their (co)homology. Indeed, the (co)homology
of a complex contains less information than the complex itself: for example there
exists spaces with same homology but different homotopy type, so the homology
invariant is in fact limited. On the other hand, the complex from that we obtain
the (co)homology contains lot more information.

When working with abelian categories, it is natural to study also complexes of


objects in those categories. The category of complexes Kom(A) of an abelian cate-
gory A is the category whose objects are complexes A• in A and whose morphisms
are morphisms of complexes.

Proposition 1.22. The category of complexes Kom(A) of an abelian category is


again abelian.

Definition 1.23. Let A• ∈ Kom(A) be a complex. Then A• [1] is the complex with
(A• [1])i := Ai+1 and differential diA[1] := −di+1
A .
The shift f [1] of a morphism of complexes f : A• → B • is the morphism of
complexes A• [1] → B • [1] given by f [1]i := f i+1 .

Corollary 1.24. The shift functor T : Kom(A) → Kom(A), A• 7→ A• [1] defines an


equivalence of abelian categories.

Note that Kom(A) equipped with the shift functor T does not define a triangu-
lated category.

Definition 1.25. The cohomology H i (A• ) of a complex A• is defined as:

Ker(di )
H i (A• ) := ∈ A,
Im(di−1 )

and any morphism of complexes f : A• → B • induces natural homomorphisms

H i (f ) : H i (A• ) −→ H i (B • ).

Moreover, a complex A• is called acyclic if H i (A• ) = 0 for all i ∈ Z.

These definition are used to describe quasi-isomorphisms.

Definition 1.26. A morphism of complexes f : A• → B • is a quasi-isomorphism if


for all i ∈ Z the induced map H i (f ) : H(A• ) → H(B • ) is an isomorphism.

Now we are ready to introduce the derived category. The idea is that quasi-
isomorphisms between complexes should become actual isomorphisms in the derived
category.
8 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

Theorem 1.27. Let A be an abelian category and let Kom(A) be its category of
complexes. Then there exists a category D(A), the derived category of A, and a
functor
Q : Kom(A) −→ D(A)

such that:

1. If f : A• → B • is a quasi-isomorphism, then Q(f ) is an isomorphism in D(A).

2. Any functor F : Kom(A) → D satisfying the property in the first point factor-
izes uniquely over Q, i.e. there exists a unique functor (up to isomorphism)
G : D(A) → D such that F ∼ = G ◦ Q.

Construction of the derived category


Theorem 1.27 is an existence result, but it does not provide a description of
objects and morphisms of D(A). So, in order to work with the derived category, we
sketch its construction. For more details, see for example [Huy06, § 2].
We start with a quasi-isomorphism C • → A• . As in the derived category a
quasi-isomorphism becomes an isomorphism, any morphism C • → B • must count
as a morphism A• → B • in the derived category.
This observation leads to the definition of morphisms in D(A) as diagrams (called
roofs)
C•
q.is

A• B•.

To make this a formal definition, we need to show when two roofs are equal and
how to compose them. So, we introduce an intermediate step between Kom(A) and
D(A), that is, the homotopy category of complexes K(A).
This category has the same objects of Kom(A), and its morphisms are classes
of equivalence morphisms of Kom(A), where the equivalence is provided by the
definition below.

Definition 1.28. Two morphism of complexes f, g : A• → B • are called homotopi-


cally equivalent if there exists a collection of homomorphisms hi : Ai → B i−1 , i ∈ Z,
such that
i−1
f i − g i = hi+1 ◦ diA + dB ◦ hi .

Now we can state the precise definition of the derived category. The objects of
D(A) are simply the objects of Kom(A), and the set of morphism HomD(A) (A• , B • )
is the set of all equivalence classes of roofs

C•
q.is

A• B•,
1.3. DERIVED CATEGORIES 9

where C • → A• is a quasi-isomorphism. Two roofs are in the same equivalence class


if there exists a commutative diagram in K(A) of the form

C•
q.is

C1• C2•

A• B•.

Once we have defined morphisms, we have to describe how they compose: let
f : A• → B • and g : B • → C • be two morphisms in the derived category, with C1• and
C2• top parts of their roofs. The composition of these morphisms is a commutative
(in K(A)) diagram of the form

C0•
q.is

C1• C2•
q.is q.is

A• B• C •.

It can be proved that this diagram exists and is unique up to equivalence. Hence,
we have defined the derived cateogory in a constructive way.
Unfortunately, contrary to Kom(A), the category D(A) is in general not abelian,
but it is always a triangulated category. To define the triangulated structure on K(A)
and D(A), we need the following definitions:
Definition 1.29. Let f : A• → B • be a morphism of complexes. Its mapping cone
is the complex C(f ) with
 i+1 
i i+1 i i −dA 0
C(f ) := A ⊕B and dC(f ) := .
f i+1 diB

Definition 1.30. A triangle

A•1 A•2 A•3 A•1 [1]

in K(A) (or D(A)) is called distinguished if is isomorphic in K(A) (or D(A)) to a


triangle
f g h
A• B• C(f ) A• [1],
with f a morphism of complexes, and g and h natural morphisms.
Proposition 1.31. The set of distinguished triangles defined in Definition 1.30 and
the natural shift functor for complexes give to the homotopy category K(A) and the
derived category D(A) of an abelian category the structure of a triangulated category.
Moreover, the natural functor QA : K(A) → D(A) is an exact functor of trian-
gulated categories.
10 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

The construction of the derived category is a specific application of a general


procedure, called localization. Roughly speaking, we can construct the localization
of a category with respect to a localizing class of morphisms. In what we did before,
these morphisms are the quasi-isomorphisms.

1.4 Derived functors


Complexes in K(A) and D(A) are in general unbounded, but we often prefer to
work with bounded ones.

Definition 1.32. Let Kom∗ (A), with ∗ ∈ {+, −, b}, be the category of complexes
A• with Ai = 0 for i  0, i  0, |i|  0 respectively.
From this we can define also K∗ (A) and D∗ (A).

We have the natural functor D∗ (A) → D(A) given by forgetting the boundedness
condition. This functor define equivalences of D∗ (A) with the full triangulated
subcategories of all complexes A• ∈ D(A) with H i (A• ) = 0 for i  0, i  0, |i|  0
respectively.

Let F : A → B be an additive functor between abelian categories. We will work


with exact functors, or at least left/right exact ones. Indeed, if F is not exact, the
image of an acyclic complex (i.e. a complex that becomes trivial in D(A)) is not in
general acyclic, so an extension of F to the derived categories does not make sense.

Lemma 1.33. Let F : K∗ (A) → K∗ (B) be an exact functor of triangulated cate-


gories, and suppose that one of the following (equivalent) conditions is true:

1. Under F a quasi-isomorphism is mapped to a quasi-isomorphism.

2. The image of an acyclic complex is again acyclic.

Then F naturally induces a commutative diagram

K∗ (A) K∗ (B)

D∗ (A) D∗ (B).

However, especially in the geometric context, we use functors that are not exact,
but only semi-exact. Thus, in order to introduce a natural functor between derived
categories, we need a more complicated construction.
So, we introduce a paticular kind of abelian category.

We recall that an object I ∈ A is called injective if Hom(_, I) is exact, and


similarly an object P ∈ A is called projective if Hom(P, _) is exact.

Definition 1.34. An abelian category A has enough injectives (resp. enough projec-
tives) if for any object A ∈ A there exist an injective object I ∈ A and an injective
morphism A → I (resp. a projective object P ∈ A and a projective morphism
P → A).
1.4. DERIVED FUNCTORS 11

From now on, we will focus our attention only on injective objects.
An injective resolution of an object A ∈ A is an exact sequence

0 A I0 I1 I2 ...

with all I i ∈ A injective. Obviously, if A has enough injectives, any object A ∈ A


admits an injective resolution.
It can be proven that, if A has enough injectives, any A• with H i (A• ) = 0 for
i  0 is isomorphic to a complex I • of injective objects with I i = 0 for i  0.

We consider now the full additive subcategory I ⊂ A of all injectives of an


abelian category A. The composition of the inclusion I ⊂ A with the natural
exact functor QA : K∗ (A) → D∗ (A) (from Proposition 1.31) yields the exact functor
ι : K∗ (I) → D∗ (A).

Proposition 1.35. Suppose that A contains enough injectives. Then the natural
functor
ι : K+ (I) −→ D+ (A)
is an equivalence.

Proof. See [Huy06, § 2].

Let F : A → B be a functor of abelian categories, and suppose that is left


exact. Furthermore, suppose that A has enough injectives. Then we have the
equivalence ι : K+ (IA ) → D+ (A) defined above, and we denote by ι−1 one of its
quasi-inverse, given by choosing a complex of injective objects quasi-isomorphic to
any given complex buonded below. Hence, we have the diagram

K(F )
K+ (IA ) K+ (A) K+ (B)
QA QB
ι−1
D+ (A) D+ (B),

where the map K(F ) is piecewise-defined in the homotopy categories.

Definition 1.36. The right derived functor of F is the functor

RF := QB ◦ K(F ) ◦ ι−1 : D∗ (A) −→ D+ (B).

Proposition 1.37. 1. There exists a natural morphism of functors

QB ◦ K(F ) −→ RF ◦ QA .

2. The right derived functor is an exact functor of triangulated categories.

3. Suppose G : D∗ (A) → D+ (B) is an exact functor. Then any morphism

QB ◦ K(F ) −→ RF ◦ QA

factorizes through a unique functor morphism RF → G.


12 CHAPTER 1. TRIANGULATED AND DERIVED CATEGORIES

Proof. See [Huy06, § 2].

These properties determine the right derived functor of a left exact functor up
to unique isomorphism.

Definition 1.38. Let F : A → B be a left exact functor and RF : D+ (A) → D+ (B)


be its right derived functor. Then, for any complex A• ∈ D+ (A), we can define:

Ri F (A• ) := H i (RF (A• )) ∈ B.

The induced additive functors

Ri F : A −→ B

are called higher derived functors of F .

We have that Ri F (A) = 0 for i < 0 and R0 F (A) ' F (A) for any A ∈ A.

We have chosen to describe RF , because in the following part we will deal with
categories with enough injectives and mostly with left exact functors. However, it
is possible define an analogous derived functor for a right exact functor G : A → B,
where A has enough projectives, called the left derived functor

LG : D− (A) −→ D− (B).

As constructed above, we also have the higher derived functors Li G.


Chapter 2

Derived categories of coherent


sheaves

2.1 Coherent sheaves


In this section we recall the definition of an important kind of sheaves, namely
coherent sheaves, which will be used extensively throughout the paper. For more
details about this topic, see, for example, [Har13].

Definition 2.1. Let (X, OX ) be a ringed space. A sheaf of OX -modules (or an


OX -module) is a sheaf F on X such that, for each open U ⊂ X, the group F(U )
is an OX (U )-module, and for each inclusion of open sets V ⊂ U , the restriction
homomorphism F(U ) → F(V ) is compatible with the module structure, via the
ring homomorphism OX (U ) → OX (V ).
A morphism F → G of sheaves of OX -modules is a morphism of sheaves such
that, for each open U ⊂ X, the map F(U ) → G(U ) is a homomorphism of OX (U )-
modules.

Definition 2.2. Let F be an OX -module. F is free if it is isomorphic to a direct


sum of copies of OX . It is locally free if X can be covered by open sets U for which
F|U is a free OX |U -module.

Let A be a ring and let M be an A-module. We want to define the sheaf


associated to M on Spec(A) (denoted by M f). For each prime ideal p ⊂ A, let MP
be the localization of M at p. For any open set U ⊂ Spec(A), we define the group
f(U ) as the set of functions s : U → `
M p∈U Mp such that, for each p ∈ U , s(p) ∈ Mp ,
and such that s is locally a fraction m/f , where m ∈ M and f ∈ A.
More precisely, we require that, for each p ∈ U , there is a neighbourhood V of
p, and elements m ∈ M and f ∈ A such that, for each q ∈ V , we have f ∈ q and
s(q) = m/f ∈ Mp .
We can make M f into a sheaf by using the obvious restriction maps.
Now we can state the definition of quasi-coherent and coherent sheaves.

Definition 2.3. Let (X, OX ) be a scheme. A sheaf of OX -modules F is quasi-


coherent if X can be covered by open affine subsets Ui = Spec(Ai ) such that, for
each i, there is an Ai -module Mi with F|Ui ' M
fi .

13
14 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

Moreover, F is coherent if is quasi-coherent and each Mi can be taken to be a


finitely generated Ai -module.

Quasi-coherent sheaves and coherent sheaves on X form two categories, denoted


respectively by Qcoh(X) and Coh(X).

There is another definition of quasi-coherent and coherent sheaves, that holds


more generally in ringed spaces.

Definition 2.4. Let (X, OX ) be a ringed space. A sheaf of OX -modules F is


quasi-coherent if it has a local presentation, that is, every point in X has an open
neighbourhood U such that the sequence

⊕I ⊕J
OX |U OX |U F|U 0

is exact, for some sets I and J, possibly infinite.


Moreover, F is coherent if it is quasi-coherent and if it satisfies two properties:

1. F is of finite type, that is, evrey point x ∈ X has an open neighbourhood U


such that there is a surjective morphism OX n | → F| for some n ∈ N.
U U

2. For any open U ∈ X, any m ∈ N and any morphism ϕ : OX m|


U → F|U of
OX |U -modules, the kernel of ϕ is finitely generated.

The notion of coherent sheaf behaves well on the category of locally noetherian
schemes. In particular, if X is a locally noetherian scheme, we have that F is
coherent if and only if it is a finitely presented OX -module, that is, every point in
X has an open neighbourhood U such that the sequence

n|
OX m|
OX F|U 0
U U

is exact, for some n, m ∈ N.

The categories Qcoh(X) of quasi-coherent sheaves and Coh(X) of coherent


sheaves on X are abelian categories.

2.2 Derived categories of coherent sheaves


We are mainly interested in the derived category of coherent sheaves on a pro-
jective variety (or, more in general, on a noetherian scheme).

Definition 2.5. Let X be a scheme. Its derived category Db (X) is, by definition,
the bounded derived category of the abelian category Coh(X), i.e.

Db (X) := Db (Coh(X)).

Definition 2.6. Two schemes X and Y defined over a field k are called derived
equivalent if there exists a k-linear exact equivalence Db (X) ' Db (Y ).

In the case of an affine scheme, we have this equivalence:


2.2. DERIVED CATEGORIES OF COHERENT SHEAVES 15

Proposition 2.7. Let X be an affine scheme, namely X = Spec(A) for some ring
A. Then the category of quasi-coherent sheaves on X is equivalent to the category
of A-modules.
Moreover, suppose that X is noetherian. Then the category of coherent sheaves
on X is equivalent to the category of finitely generated A-modules.

Unfortunately, the category Coh(X) usually does not have non-trivial injective
objects. Here is a simple example that shows the lack of injectives.

Example 2.8. Let X = Spec(Z) and assume that I is an injective object of


Coh(X). By Proposition 2.7, we have that Coh(X) is equivalent to the category of
finitely generated Z-modules, that is, finitely generated abelian groups. Let a ∈ I
and let ϕ : Z → I be a morphism of modules that sends 1 to a. By the fact that I
is injective, we can construct a commutative diagram

I
ϕ h

·n
0 Z Z,

where ·n is the multiplication for n ∈ N+ . So we have that a = ϕ(1) = h(n) = nh(1),


and hence I is a divisible abelian group. On the other hand, by the structure theorem
for finitely generated abelian groups, we have that non-trivial groups cannot be
divisible. Hence, I must be trivial.

Thus, in order to compute derived functors, we have to skip to bigger abelian


categories. In almost all the cases, this category will be the category of quasi-
coherent sheaves Qcoh(X), and sometimes the abelian category of OX -modules
ShOX (X).
Notation. To avoid any confusion between sheaf cohomology H i (X, F) and the co-
homology H i (F • ) of a complex of sheaves, we will write Hi (F • ) for the latter.
The following proposition shows that the category of quasi-coherent sheaves on
a noetherian scheme X has enough injectives.

Proposition 2.9. Let X be a noetherian scheme. Then any quasi-coherent sheaf F


admits a resolution

0 F I0 I1 ...

where the quasi-coherent sheaves I i are injective as OX -modules.

This result is useful in our case, i.e. when X is a smooth projective variety.

The problem arise in the passage from the quasi-coherent to the coherent sheaves.
In fact, we cannot hope to find an injective resolution of a coherent sheaf made by
coherent sheaves. However, we have a result that allow us to overcome the problem
of lack of injectives in Coh(X).
16 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

Proposition 2.10. Let X be a noetherian scheme. Then the natural functor

Db (X) −→ Db (Qcoh(X))

defines an equivalence between the derived category Db (X) and the full triangulated
b (Qcoh(X)) of bounded complexes of quasi-coherent sheaves with
subcategory Dcoh
coherent cohomology.

Proof. Let G • be a bounded complex of quasi-coherent sheaves

0 Gn ... Gm 0

with coherent cohomology Hi . Suppose G i is coherent for i > j. We need the


following lemma:
Lemma 2.11. Let X be a noetherian scheme. If G → F is an OX -module surjective
homomorphism from a quasi-coherent sheaf G onto a coherent sheaf F, then there
exists a coherent subsheaf G 0 ⊂ G such that the composition G 0 → G → F is still
surjective.
Applying this result to the surjections

dj : G j −→ Im(dj ) ⊂ G j+1 and Ker(dj ) −→ Hj

we obtain the subsheaves G1j ⊂ G j and G2j ⊂ Ker(dj ) ⊂ G j . Thus, we can replace }|
by the coherent sheaf generated by Gij , i = 1, 2, and G j−1 by the pre-image of the
new G j under G j−1 → G j . The inclusion defines a quasi-isomorphism of the new
complex to the old one and now G i is coherent for i ≥ j.

Before introducing the derived functors used in algebraic geometry, we state a


couple of useful results.

Definition 2.12. The support of a complex F • ∈ Db (X) is the union of the supports
of all its cohomology sheaves, that is, it is the closed subset
[
supp(F • ) := supp(Hi (F • )).
i∈Z

Lemma 2.13. Let F • ∈ Db (X) and suppose supp(F • ) = Z1 tZ2 , where Z1 , Z2 ⊂ X


are disjoint closed subsets. Then F • ' F1• ⊕ F2• , with supp(Fj• ) ∈ Zj for j = 1, 2.

Proof. See [Huy06, § 3].

Proposition 2.14. Let X be a noetherian scheme. Then Db (X) is an indecompos-


able triangulated category if and only if X is connected.

Proof. See [Huy06, § 3].

Proposition 2.15. Let X be a smooth projective variety. Then the objects of the
form k(x), with x ∈ X closed point, span the derived category Db (X).

Proof. See [Huy06, § 3].


2.3. DERIVED FUNCTORS IN ALGEBRAIC GEOMETRY 17

2.3 Derived functors in algebraic geometry


We now discuss the derived functors which we will need in the sequel.

Global section. Let X be a noetherian scheme over a field k. The global section
functor
Γ : Qcoh(X) −→ Vec(k)
F 7−→ Γ(X, F)
is a left exact functor. The category Qcoh(X) has enough injectives, so we can
define the right derived functor
RΓ : D+ (Qcoh(X)) −→ D+ (Vec(k)).
The higher derived functor is denoted by H i (X, F • ) := Ri Γ(F • ). If F is an actual
sheaf, these are the cohomology groups.
The following result is a special case of a more general one.
Theorem 2.16 (Grothendieck). Let X be a noetherian scheme and F be a quasi-
coherent sheaf on X. Then H i (X; F) = 0 for i > dim(X).
Using this theorem, we can prove that the functor RΓ induces an exact functor
RΓ : Db (Qcoh(X)) −→ Db (Vec(k)).
Now, we restrict the functor to the full subcategory of coherent sheaves.
Theorem 2.17 (Serre). Let X be a projective scheme on a field k and F be a
coherent sheaf on X. Then all cohomology groups H i (X, F) are of finite dimension.
Then, the global section functor yields the left exact functor Γ : Coh(X) →
Vecf in (k), where Vecf in (k) is the category of vector spaces of finite dimension over
k.
As we said before, Coh(X) does not have enough injectives, so in general we
cannot construct the derived functor. However, thanks to the theorem above, the
right derived functor
RΓ : Db (X) −→ Db (Vecf in (k))
can be obtained as the composition of exact functors

Db (X) Db (Qcoh(X)) Db (Vec(k)).

Direct image. Let f : X → Y be a morphism between two noetherian schemes.


The direct image is the left exact functor
f∗ : Qcoh(X) −→ Qcoh(Y ).
The category Qcoh(X) has enough injectives, so we can define the right derived
functor
Rf∗ : D+ (Qcoh(X)) −→ D+ (Qcoh(Y )).
We can notice that, if X is a noetherian scheme over a field k, the direct image of
a sheaf under the structure morphism X → Spec(k) is nothing but Γ. So, we can
consider the following result as a generalization of Theorem 2.16:
18 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

Theorem 2.18. Let f : X → Y be a morphism of noetherian schemes, and F be


a quasi-coherent sheaf on X. Then the higher direct images Ri f∗ (F) are trivial for
i > dim(X).

Thus, the higher direct image funcot induces an exact functor

Rf∗ : Db (Qcoh(X)) −→ Db (Qcoh(Y )).

If we restrict to coherent sheaves, we can use also the following theorem.

Theorem 2.19. Let f : X → Y be a projective (or proper) morphism of noetherian


schemes and let F be a coherent sheaf on X. Then the higher direct images Ri f∗ (F)
are again coherent.

Hence, for a proper morphism f : X → Y of noetherian schemes, we obtain the


right derived functor
Rf∗ : Db (X) −→ Db (Y )

as the composition

Db (X) Db (Qcoh(X)) Db (Qcoh(Y )).

Other functors. Global section and direct image functors are not the only derived
functors which we will use in the sequel. We now present some other ones, but we will
not show how to contruct the derived functors. Indeed, the construction is similar
to the previous ones (for a more explicit construction, see for example [Huy06, § 3]).

Local Hom Let F ∈ Qcoh(X). Then

Hom(F, _) : Qcoh(X) −→ Qcoh(X)

is a left exact functor, where Hom(F, E) is the sheaf of OX -modules U 7→


Hom(F|U , E|U ). Its derived functor is

RHom(F, _) : D+ (Qcoh(X)) −→ D+ (Qcoh(X)).

For any quasi-coherent sheaves F and E, we have Exti (F, E) := Ri Hom(F, E).
The restriction to coherent sheaves yields the functor

RHom(F, _) : Db ((X)) −→ Db ((X)).

Dual We define the dual F •∨ of a complex F • ∈ D− (Qcoh(X)) of quasi-coherent


sheaves as
F •∨ := RHom(F • , OX ) ∈ D+ (Qcoh(X)).

If X is regular, for any F • ∈ Db (X), we have F •∨ := RHom(F • , OX ) ∈


Db (X).
2.3. DERIVED FUNCTORS IN ALGEBRAIC GEOMETRY 19

Tensor product Let F • ∈ K− (Coh(X)) be a complex bounded above. We define


the exact functor

F • ⊗ _ : K− (Coh(X)) −→ K− (Coh(X))
M
(F • ⊗ E • )i := F p ⊗ E q with d = dF ⊗ 1 + (−1)i 1 ⊗ dE .
p+q=i

We have the left derived functor

F • ⊗L _ : D− (X) −→ D− (X).

Inverse image Let f : (X, OX ) → (Y, OY ) be a morphism of ringed spaces. Then

f ∗ : ShOY (Y ) −→ ShOX (X)

is the composition of the exact functor

f −1 : ShOY (Y ) −→ Shf −1 (OY ) (X)

and the right exact functor

OX ⊗f −1 (OY ) _ : Shf −1 (OY ) (X) −→ ShOX (X).

Thus, f ∗ is right exact, and its left derived functor is

Lf ∗ := OX ⊗L −1
: D− (Y ) −→ D− (X).

f −1 (OY ) _ ◦ f

However, in most of our applications, f is often flat, so we do not even have


to derive f ∗ because it is exact.
In many cases, we have to deal with a combination of these functors. So, we need
to check their compatibilities. Let us start with a morphism f : X → Y between
projective schemes over k. Then for any F • , E • ∈ Db (Y ), there exists a natural
isomorphism:
Lf ∗ (F • ) ⊗L Lf ∗ (E • ) ' Lf ∗ (F • ⊗L E • ).
If f is proper, for any F • ∈ Db (X) and any E • ∈ Db (Y ), there exists a natural
isomorphism, called projection formula:

Rf∗ (F • ) ⊗L E • ' Rf∗ (F • ⊗L Lf ∗ (E • )).

Moreover, consider a fibre product diagram


v
X ×Z Y Y
g f
u
x Z
with u : X → Z flat and f : Y → Z proper. Then the flat base change asserts the
existence of a functorial isomorphism:

u∗ Rf∗ F • ' Rg∗ v ∗ F •

for any F • ∈ D(Qcoh(Y )).


20 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

2.4 Fourier-Mukai transform and Orlov’s theorem


Let X and Y be two smooth projective varieties, and let

p: X × Y → X and q: X × Y → Y

be the two projections.

Definition 2.20. Let P ∈ Db (X ×Y ) (we will denote it without the bullet, although
it is in general a true complex). The induced Fourier-Mukai transform is the functor

ΦP : Db (X) −→ Db (Y )
E • 7−→ Rq∗ (p∗ E • ⊗L P).

The object P is called the Fourier-Mukai kernel of the Fourier-Mukai transform.

Remark 2.21. Any Fourier-Mukai transform ΦP is the composition of three exact


functors p∗ , _ ⊗L P and Rq∗ . Thus ΦP is exact.
The kernel P can also be used to define an exact functor in the opposite direction.
To avoid confusion, when necessary we will denote by ΨP the functor

ΨP : Db (Y ) −→ Db (X)
E • 7−→ Rp∗ (q ∗ E • ⊗L P).

Example 2.22. Here we have some examples of equivalences which are in fact
Fourier-Mukai transforms.

1. The identity
id : Db (X) −→ Db (X)
is isomorphic to the Fourier-Mukai transform ΦO∆ , where O∆ is the structure
sheaf of the diagonal ∆ ⊂ X × X. In fact, if we denote with i : X → X × X
the diagonal embedding, we have

ΦO∆ (E • ) = Rq∗ (p∗ E • ⊗L O∆ ) = Rq∗ (p∗ E • ⊗L i∗ OX ) '


' Rq∗ (Ri∗ (i∗ p∗ E • ⊗L OX )) ' R(q ◦ i)∗ (p ◦ i)∗ E • ' E • ,

where OX is the structure sheaf of X.

2. Let f : X → Y be a morphism. Then the direct image Rf∗ : Db (X) → Db (Y )


is isomorphic to the transform ΦOΓ , where Γ ⊂ X ×Y is the graph of f . Indeed,
denoting with j : X → Γ ⊂ X × Y the function that sends x to (x, f (x)), we
have:

ΦOΓ (E • ) = Rq∗ (p∗ E • ⊗L OΓ ) = Rq∗ (p∗ E • ⊗L j∗ OX ) '


' Rq∗ (Rj∗ (j ∗ p∗ E • ⊗L OX )) ' R(q ◦ j)∗ (p ◦ j)∗ E • ' Rf∗ (E • ).

We can also use OΓ as the kernel for a Fourier-Mukai transform in the opposite
direction to obtain the inverse image functor (that is, ΨOΓ is isomorphic to
Lf ∗ ).
2.4. FOURIER-MUKAI TRANSFORM AND ORLOV’S THEOREM 21

3. Let L be a line bundle on X. Then E • 7→ E • ⊗ L defines an autoequivalence


Db (X) → Db (X) which is isomorphic to the transform with kernel i∗ (L), where
i is the diagonal embedding.

4. Let P be a coherent sheaf on X ×Y flat over X, and consider the Fourier-Mukai


transform ΦP : Db (X) → Db (Y ). If x ∈ X is a closed point with k(x) ' k,
then
ΦP (k(x)) ' Px ,
where Px := P|{x}×Y is considered as a sheaf on Y via the projection {x}×Y →
Y.

5. Let P ∈ Db (X × Y ) be a coherent sheaf on X × Y flat over X. This can be


seen as a family of coherent sheaves Px on Y , or as a deformation of the sheaf
Px0 for a distinguished closed point x0 ∈ X (we assume k(x0 ) ' k). A tangent
vector v at x0 is determined by a subscheme Zv ⊂ X. Pulling back

0 k(x) OZv k(x) 0

and tensoring with P yields

0 Px0 P|Zv ×Y Px0 0.

If we see this as a sequence on Y , it gives rise to a class in Ext1Y (Px0 , Px0 ). So


we obtain a linear map, called Kodaira-Spencer map,

κ(x0 ) : Tx0 X −→ Ext1Y (Px0 , Px0 ).

By construction, κ(x0 ) iscompatible with ΦP , that is, we have the commutative


diagram
κ(x0 )
Tx0 X ' Ext1X (k(x), k(x)) Ext1Y (Px0 , Px0 )
' '
ΦP
HomDb (X) (k(x0 ), k(x0 )[1]) HomDb (Y ) (Px0 , Px0 [1]).

We know that any equivalence has a left and a right adjoint. In the case of
Fourier-Mukai transforms, any transform has left and right adjoints. Moreover,
these adjoints are again Fourier-Mukai tranforms, and their kernels can be described
explicitely.
Proposition 2.23 (Mukai). Let ΦP : Db (X) → Db (Y ) be a Fourier-Mukai trans-
form with kernel P, and let

PL := P ∨ ⊗ q ∗ ωY [dim(Y )] and PR := P ∨ ⊗ p∗ ωX [dim(X)],

objects in Db (X × Y ), where ωX and ωY are the canonical bundles of X and Y .


Then
ΨPL : Db (Y ) −→ Db (X) and ΨPR : Db (Y ) −→ Db (X)
are left and right adjoints of ΦP .
22 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

We are also interested in the composition of Fourier-Mukai transforms. It turns


out that the composition of two tranforms is again a transform, and we can obtain
an explicit formula using kernels.

Proposition 2.24 (Mukai). Let X,Y and Z be smooth projective varieties over a
field k. Suppose that the objects P ∈ Db (X × Y ) and Q ∈ Db (Y × Z) are kernels of
transforms. Then the composition

ΦP ΦQ
Db (X) Db (Y ) Db (Z)

is isomorphic to the Fourier-Mukai transform ΦR : Db (X) −→ Db (Z), where



R := πXZ∗ (πXY P × πY∗ Z Q)

and πXY , πY Z and πXZ are the projections from X × Y × Z to X × Y , Y × Z and


X × Z.

Notation. Let Xi and Yi be smooth projective varieties, i = 1, 2, and Pi ∈ Db (Xi ×


Yi ). We denote with P1 P2 ∈ Db ((X1 ×X2 )×(Y1 ×Y2 )) the exterior tensor product
p∗13 P1 ⊗ p∗24 P2 , where pij are the projection maps.

Lemma 2.25. Let ΦPi : Db (Xi ) → Db (Yi ), i = 1, 2, be two Fourier-Mukai trans-


forms. Then there exist isomorphisms

ΦP1 P2 (F1•  F2• ) ' ΦP1 (F1• )  ΦP2 (F2• ),

which are functorial in Fi• . Moreover, let R ∈ Db (X1 × X2 ) and let S := ΦP1 P2 (R)
be its image in Db (Y1 × Y2 ). Then the diagram

ΨP1
Db (X1 ) Db (Y1 )
ΦR ΦS

Db (X2 ) ΦP2
Db (Y2 )

is commutative.

Remark 2.26. Let ΦPi : Db (Xi ) → Db (Yi ) and ΦQi : Db (Yi ) → Db (Zi ), i = 1, 2, be
Fourier-Mukai transforms, and let Ri be the kernels of the compositions ΦQi ◦ ΦPi .
There is a natural relation

ΦQ1 Q2 ◦ ΦP1 P2 ' ΦR1 R2 .

The relation between arbitrary functors and Fourier-Mukai ones is well described
in the following theorem due to Orlov.

Theorem 2.27 (Orlov). Let X an Y be two smooth projective varieties and let
F : Db (X) → Db (Y ) be a fully faithful exact functor. If F admits left and right
adjoints, then there exists an object P ∈ Db (X × Y ), unique up to isomorphism,
such that F is isomorphic to the Fourier-Mukai transform ΦP .
2.4. FOURIER-MUKAI TRANSFORM AND ORLOV’S THEOREM 23

Proof. See [Orl03, § 3].

Hence, since we will work only in a geometric point of view, we can focus our
attention on Fourier-Mukai transforms. In particular, we can state Orlov’s theorem
for equivalences:
Corollary 2.28. Let F : Db (X) → Db (Y ) be an equivalence between the derived
categories of two smooth projective varieties. Then F is isomorphic to a Fourier-
Mukai transform ΦP whose kernel P ∈ Db (X × Y ) is unique up to isomorphism.
Proposition 2.29. Let ΦPi : Db (Xi ) → Db (Yi ), i = 1, 2, be two Fourier-Mukai
transforms that are fully faithful (resp. equivalences). Then the functor

ΦP1 P2 : Db (X1 × X2 ) −→ Db (Y1 × Y2 )

is fully faithful (resp. an equivalence).


Proof. The functors ΦPi have left adjoints ΨQi , and since they are fully faithful, the
compositions ΨQi ◦ ΦPi are isomorphic to the identity functors, represented by O∆i ,
where ∆i are the diagonals of Xi × Xi . It is easy to see that O∆1  O∆2 ' O∆ ,
where ∆ is the diagonal of (X1 × X2 ) × (X1 × X2 ). Then, from Remark 2.26, we
have that the composition ΨQ1 Q2 ◦ ΦP1 P2 is represented by O∆ , and hence it is
isomorphic to the identity. Thus ΦP1 P2 is fully faithful. Similarly we can prove the
equivalence part.

Remark 2.30. Suppose that ΦP : Db (X) → Db (Y ) is an equivalence and that an


object Q ∈ Db (X × Y ) satisfies ΨQ ' Φ−1
P . Then we denote the functor

ΦPQ : Db (X × X) −→ Db (Y × Y )

by AdP . By Proposition 2.29, AdP is an equivalence, and by Lemma 2.25, for each
Q ∈ Db (X × X), there is an isomorphism of functors

ΦAdP (Q) ' ΦP ◦ ΦQ ◦ Φ−1


P .

We have some interesting applications of Orlov’s existence theorem.


Corollary 2.31. Let X and Y be smooth projective varieties with equivalent derived
categories Db (X) ' Db (Y ). Then dim(X) = dim(Y ).
Proof. Suppose that F : Db (X) → Db (Y ) is the equivalence. By Theorem 2.27,
there exists an object P ∈ Db (X × Y ) such that F ' ΦP . So F has a left and right
adjoints given by the Fourier-Mukai transforms with kernels PL and PR respectively.
Since F is an equivalence, the two adjoints are both quasi-inverses of F . Then, by
the uniqueness of the Fourier-Mukai kernel, we have that PL and PR are isomorphic
objects in Db (X × Y ). Using Proposition 2.23, we have

P ∨ ⊗ q ∗ ωY [dim(Y )] ' P ∨ ⊗ p∗ ωX [dim(X)]

and then
P ∨ ' P ∨ ⊗ (p∗ ωX ⊗ q ∗ ωY [dim(X) − dim(Y )]).
P is an object of a bounded derived category, hence we have dim(X) = dim(Y ).
24 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

Corollary 2.32. Let k be an algebraically closed field, and let Φ : Db (X) → Db (Y )


be an equivalence such that, for any closed point x ∈ X, there exists a closed point
f (x) ∈ Y with
Φ(k(x)) ' k(f (x)).
Then f : X → Y defines an isomorphism and Φ is the composition of f∗ with the
twist of some line bundle M on Y , i.e.

Φ ' (M ⊗L _) ◦ Rf∗ .

Proof. First, we want to show that there exists a morphism X → Y which on the
set of closed points induces the given map f . Since Φ is a Fourier-Mukai transform,
it has a kernel P. We need the following lemma:
Lemma 2.33. Let S → X be a moprhism of schemes. Suppose Q ∈ Db (S) and
assume that, for all closed points x ∈ X, the derived pull-back Li∗x ∈ Db (Sx ) is a
complex concentrated in degree zero, i.e. a sheaf. Then Q is isomorphic to a sheaf
which is flat over X.
This lemma implies that P is a sheaf on X × Y flat over X. By assumption,
P|{x}×Y ' k(f (x)), so choosing local sections of P shows that it defines a morphism
X → Y inducing f on the closed points. We call this morphism again f .
Next, we want to prove that f is an isomorphism. Since the sheaves k(x) span
Db (X), their images span Db (Y ). Then, for any closed point y ∈ Y , there exists a
closed point x ∈ X and an integer m with Hom(Φ(k(x)), k(y)[m]) 6= 0. This implies
that any k(y) is of the form k(f (x)) for some closed point x ∈ X, that is, f is
surjective on the set of closed points.
In the same way, two different points x1 , x2 ∈ X have different images f (x1 ) 6=
f (x2 ). If char(k) = 0, this is sufficent to conclude that f is an isomorphism between
two smooth varieties. If char(k) 6= 0, we can use a quasi-inverse Φ−1 to obtain an
inverse f −1 .
Now, if we see P has a sheaf on its support, which is the graph of f , we have
that it is a sheaf of constant fibre of dimension one, that is, a line bundle. Using
supp(P) ' Y given by the second projection, we can consider this line bundle as a
line bundle M on Y .

2.5 Criterion for equivalence


In this section, we make a further (but crucial) hypotesis: we assume that k is
an algebraically closed field of characteristic zero.
Let X and Y be two smooth projective varieties over k, and let ΦP : Db (X) →
b
D (Y ) be the Fourier-Mukai transform between them given by the object P ∈
Db (X × Y ). We want to give a criterion for a Fourier-Mukai transform to be fully
faithful, and to see when such a transformation is an equivalence.
Theorem 2.34 (Bondal, Orlov). The functor ΦP is fully faithful if and only if, for
any two closed points x, y ∈ X, we have
(
k if x = y and i = 0
Hom(ΦP (k(x)), ΦP (k(y))[i]) =
0 if x 6= y or i < 0 or i > dim(X).
2.5. CRITERION FOR EQUIVALENCE 25

Proof. The goal of the proof is to show that, for arbitrary closed points x, y ∈ X
and any i ∈ Z, the homomorphisms

Hom(k(x), k(y)[i]) −→ Hom(ΦP (k(x)), ΦP (k(y))[i])

are bijective. Since the Fourier-Mukai transform F := ΦP has left and right adjoints
G := ΦP L and H := ΦP R , we can apply Proposition 1.19.
Step 1. If x 6= y, by assumption the claim is true. Thus, we have to discuss
only the case x = y. In this case, by the assumption we already have the bijectivity
for i < 0 and i > dim(X).
We know that the bijectivity of

Hom(k(x), k(x)[i]) −→ Hom(F (k(x)), F (k(x))[i])

is equivalent to the bijectivity of


_◦gk(x)
Hom(k(x), k(x)[i]) Hom(G(F (k(x))), k(x)[i]), (2.1)

induced by the adjunction morphism g : G ◦ F → idDb (X) .


If we claim that G(F (k(x))) ' k(x), then the adjunction morphism

gk(x) : G(F (k(x))) −→ k(x)

is either an isomorphism (which yields bijectivity in (2.1)) or the zero map, because
k is an algebraically closed field. But gk(x) , cannot be the zero map, since the
composition of F (gk(x) ) : F (G(F (k(x)))) → F (k(x)) with the adjunction morphism
hF (k(x)) : F (k(x)) → F (G(F (k(x)))) yields the identity, and F (k(x)) 6= 0, because
in the assumption we have Hom(F (k(x)), F (k(x))) = k.
Thus, we only need to prove the claim, that is, G(F (k(x))) ' k(x).

Step 2. Let us fix a closed point x ∈ X. We want to show that

1. G(F (k(x))) is a sheaf and

2. the homomorphism (2.1) is at least injective for i = 1.

For the first point, we shall use the following lemma:


Lemma 2.35. Let X be a smooth projective variety, x ∈ X a closed point, and F • ∈
Db (X). Suppose that, for any closed point y 6= x and any i ∈ Z, Hom(F • , k(y)[i]) =
0, and Hom(F • , k(x)[i]) = 0 for i < 0 or i > dim(X). Then F • is isomorphic to a
sheaf concentrated in x ∈ X.
We have that G(F (k(x))) satisfies the assumption of the lemma, because

Hom(G(F (k(x))), k(y)[i]) ' Hom(F (k(x)), F (k(y))[i]) = 0

for i < 0 or i > dim(X) or x 6= y. Thus, using the lemma, the first point is proven.
We want now to prove the second point. The composition G ◦ F is a Fourier-
Mukai transform, and we denote its kernel by Q. Let x ∈ X and ix : {x} × X → X ×
X the inclusion map. We have seen that i∗x Q = G(F (k(x))) is a sheaf concentrated
26 CHAPTER 2. DERIVED CATEGORIES OF COHERENT SHEAVES

in x. By Lemma 2.33, we have that Q is a sheaf on X × X flat over the first factor.
Let x be a point of X (note that is enough to take a generic point). We want to
show that Hom(k(x), k(x)[1]) → Hom(F (k(x)), F (k(x))[1]) is injective, and this will
prove the second point.The composition with G yields the map

κ(x) : Hom(k(x), k(x)[1]) −→ Hom(G(F (k(x))), G(F (k(x)))[1]).

By the flatness of Q, we know that the map κ(x) is the Kodaira-Spencer map of the
flat family Q over X ×X defining G◦F . On the other hand, since Qx is concentrated
in x, the map f : x 7→ Qx is injective. Thus, the tangent map κ(x) := df (x) is
injective for X ∈ X generic.

Step 3. Now we can prove the claim at the end of Step 1, namely that
G(F (k(x))) ' k(x). W denote G(F (k(x))) by F, which is a sheaf from Step 2.
Since Hom(F, k(y)) = 0 for any y 6= x, F is concentrated in x. We already know
that the adjunction morphism gk(x) : F → k(x) is not trivial and hencde surjective.
So consider the short exact sequence
gk(x)
0 Ker(gk(x) ) F k(x) 0.

Obviously, Ker(gk(x) ) is also concentrated in k(x), so in order to show the bijectivity


of gk(x) we only need to prove that Hom(Ker(gk(x) ), k(x)) = 0. Thus we apply
Hom(_, k(x)) to the exact sequence, and using Hom(F, k(x)) = k we obtain the
exact sequence
_◦gk(x)
0 Hom(Ker(gk(x) ), k(x)) Hom(k(x), k(x)[1]) Hom(F, k(x)[1]).

The map _◦gk(x) is injective due to the second point of Step 2, and hence Ker(gk(x) )
is zero, and the claim is proved.
Chapter 3

Abelian varieties

3.1 Definition and basic properties


Definition 3.1. A group variety over a field k is an algebraic variety V over k
together with k-morphisms

m : V ×k V −→ V (group law)
inv : V −→ V (inverse)

and a distinguished element e ∈ V (k), such that:

1. the maps
idV ×e m e×idV m
V V ×k V V, V V ×k V V

are both the identity map, and so e is the identity element.

2. The maps
∆ idV × inv m
V V ×k V V ×k V V

∆ inv × idV m
V V ×k V V ×k V V
are both equal to
e
V Spec(k) V,
so inv is the map taking an element to its inverse.

3. the diagram
idV ×m
V ×k V ×k V V ×k V
m×idV m
m
V ×k V V
commutes, that is, associativity holds.

The quadruple (V, m, inv, e) defined above is a group in the category of varieties
over k.

27
28 CHAPTER 3. ABELIAN VARIETIES

Definition 3.2. Let V be a group variety over k, and let x ∈ V be a k-rational


point. We define the (right) translation ta : V → V to be the composition
idV ×a m
V V ×V V,

so on points this map is given by ta (x) = m(x, a). Moreover, ta is a bijective map
with inverse tinv(a) .
A group variety has some properties descending directly from its definition and
the definition of translation.
Proposition 3.3. Let V be a group variety over k. Then V is smooth over k.
Moreover, if V is also connected, then it is irreducible.
We now state the definition of a complete variety, that is an analogue of topo-
logical compactness in algebraic geometry.
Definition 3.4. A complete variety is an algebraic variety X such that, for any
variety Y , the projection morphism π : X × Y → Y is a closed map.
Finally we can define the object of study of this section:
Definition 3.5. An abelian variety is a complete and connected group variety.
Theorem 3.6 (Rigidity lemma). Let X, Y and Z be three algebraic varieties such
that X is complete, and let f : X × Y → Z be a morphism such that, for some
Y0 ∈ Y , the fibre X × {y0 } is mapped to a single point z0 ∈ Z. Then f factors
through the projection p2 : X × Y → Y , i.e. there exists a morphism g : Y → Z such
that f = g ◦ p2 .
Proof. We may assume that k = k̄. We choose a point x0 ∈ X and we define
g(y) = f (x0 , y). Since X × Y is a variety, to prove that f = g ◦ p2 it is sufficient
to show that these morphisms coincide on some open subset of X × Y . So, let
U ⊂ Z be an affine open neighbourhood of z0 . Since X is complete, the projection
p2 is a closed map, then V := p2 (f −1 (Z \ U )) is closed in Y . By construction,
y0 ∈/ V , so Y \ V is a non-empty open subset of Y . For each y ∈ Y \ V , X × {y} is
mapped by f into U , and since X × {y} is complete and U is affine, we have that
f maps X × {y} into a single point in U . Hence, for any x ∈ X and y ∈ Y \ V ,
f (x, y) = f (x0 , y) = (g ◦ p2 )(x, y), and this proves the theorem.

Definition 3.7. Let V and W be two group varieties. A morphism f : V → W is


called a homomorphism if the diagram
f ×f
V ×V W ×W
mV mW
f
V W

is commutative. If this holds, we have also f (eV ) = eW and f ◦ invV = invW ◦f .


The rigidity of abelian varieties is evidenced by the fact that every morphism is
an homomorphism, up to a translation.
3.1. DEFINITION AND BASIC PROPERTIES 29

Corollary 3.8. Let A and B be two abelian varieties, and let f : A → B be a


morphism. Then f is the composition of a homomorphism and a translation.

Proof. In this proof we will use the additive notation (although we have not yet
proven that an abelian variety is commutative), in order to have a better compre-
hension of the morphism we will construct.
Replacing f by f − f (0) (i.e. composing with the translation t−f (0) ), we may
assume f (0) = 0. Consider the map

ϕ : A × A −→ B

definded by ϕ(x, y) = f (x + y) − f (x) − f (y). Then ϕ is the difference of the two


morphisms f ◦ mA and mB ◦ (f × f ), and then it is a morphism too. We have that
ϕ(A × {0}) = ϕ({0} × A) = 0, so by Rigidity lemma ϕ ≡ 0 on A × A. This means
that f ◦ mA = mB ◦ (f × f ), that is, f is an homomorphism.

Remark 3.9. The corollary shows that the group structure on an abelian variety is
uniquely determined by the choice of the identity element.

Corollary 3.10. The group law on an abelian variety is commutative.

Proof. In general, a group is commutative if the inverse map is an homomorphism.


In this case, the map inv takes the identity element to itself, so by previous corollary,
it is an homomorphism.

Notation. From now on, since abelian varieties are commutative groups, we will use
the additive notation, that is, x + y for m(x, y), −x for inv(x) and 0 for e.
The rigidity of abelian varieties can be seen also in the rational maps between
them. In general, a rational map from a variety X to a variety Y is defined only on
a dense open set of X, and we may fail to extend it to a morphism on the whole X.
In this sense, the following is the best result possible.

Theorem 3.11. Let X be a normal variety and Y be a complete variety. Then a


rational map f : X 99K Y is defined on an open subset U ⊂ X with codim(X\U ) ≥ 2.

Proof. We first assume that X is a (non-singular) curve. Thus we have a regular


map f : U → Y from a subset U ⊂ X that we want to extend to X. We have the
maps
X
p

idU ×f
U X ×Y

q
Y,
where p and q are the projections. Let U 0 be the image of U in X × Y , and let Z be
its closure. Since Y is complete, p(Z) is closed, and it contains U , which is an open
dense in X, so p(Z) = X. Then the maps U → U 0 → U (restrictions of the maps
on the diagram) are isomorphisms. Hence, Z → X is a surjective map between
complete curves that is an isomorphism on open subsets. Such a map must be an
30 CHAPTER 3. ABELIAN VARIETIES

isomorphism, and then Z ' X. Thus, the restriction to Z of the map q : X ×Y → Y


is the extension of f to X that we are looking for.
We can reduce the general case to the one-dimensional case. Let U be the
maximal open subset on which f is defined, and suppose that W := X \ U has
codimension one. Then there is a prime divisor D in W . By normality of X, its
associated local ring OD is a descrete valuation ring, with field of fraction k(X).
By the valuative criterion of properness, the map f : Spec(k(X)) → Y extends to a
morphism Spec(OD ) → Y . This implies that f has a representative defined on an
open subset that meets D in a non-empty set, and this is a contradiction. Hence
codim(W ) ≥ 2.

Lemma 3.12. Let X be a smooth variety, V a group variety, and let f : X 99K V
be a rational map. Then either f is defined on the whole X or the points where is
not defined form a closed subset of pure codimension 1 in X.

Proof. We define a rational map

F : X × X 99K V
(x, y) 7−→ f (x) · f (y)−1 .

Clearly, if f is defined at a point x ∈ X, F is defined at (x, x), and then F (x, x) = e.


Conversely, if F is defined at (x, x), then it is defined on an open neighbourhood of
(x, x), and in particular there exists an open subset U ⊂ X such that F is defined
on {x} × U . Up to shrinking U , f , will be defined on U , but it can happen that x
is not contained in U . For u ∈ U , the formula F (x, u)f (u) = f (x) defines f at x.
Thus, f is defined at x if and only if F is defined at (x, x).
The map F defines a map

F ∗ : OV,e −→ k(X × X).

Since F sends (x, x) to e (if it is defined there), we have that F is defined in (x, x) if
and only if F ∗ (OV,e ) ⊂ OX×X,(x,x) . For a nonzero rational function g ∈ k(X × X),
we write div(g) = div(g)0 − div(g)∞ , with div(g)0 and div(g)∞ effective divisors.
Then

OX×X,(x,x) = {g ∈ k(X × X) | div(g)∞ does not contain (x, x)} ∪ {0}.

Suppose f is not defined in x. Then for some g ∈ Im(f ∗ ), (x, x) ∈ div(g)∞ and F
is not defined at the points of ∆ ∩ div(g)∞ . This is a subset of pure codimension
one, and identifying it with a subset of X, we obtain a subset of codimension one
passing through x on which f is not defined.

Combining this two results, we have:

Proposition 3.13. Let A be an abelian variety over k. If X is a smooth k-variety,


then any rational map f : X 99K A extends to a morphism X → A.
3.2. LINE BUNDLES ON ABELIAN VARIETIES 31

3.2 Line bundles on abelian varieties


In this section we study bundles and divisors on abelian varieties, in order to
prove that these varieties are projective. Since abelian varieties are nonsingular, a
Weil divisor defines a Cartier divisor and a line bundle, and we have the natural

isomorphism Cl(X) − → Pic(X).
Notation. If L is a line bundle on a product variety X × Y , x ∈ X and y ∈ Y , then
we will write Lx for the restriction L|{x}×Y and Ly for L|X×{y} .
Here we have a general theorem from algebraic geometry that can be applied to
abelian varieties.

Theorem 3.14. Let X and Y be two varieties, with X complete, and let L and M
be two line bundles on X × Y . If, for all closed points y ∈ Y , we have Ly ' My ,
there exists a line bundle N on Y such that L ' M ⊗ p∗ N , where p is the projection
onto Y .

In order to prove the theorem, we need the following lemma (see [Mum70, § II.5]
for the proof):

Lemma 3.15. Let X be a noetherian scheme, Y be a reduced and connected noethe-


rian scheme, f : X → Y be a proper morphism and F be a coherent sheaf on X flat
over Y . Then for all p ∈ Z the following are equivalent:

1. y 7→ dimk(y) H p (Xy , Fy ) is a constant function.

2. Rp f∗ (F) is a locally free sheaf E on Y , and for all y ∈ Y , the natural map

E ⊗ k(y) −→ H p (Xy , Fy )

is an isomorphism.

Moreover, if these conditions hold, we have that

Rp−1 f∗ (F) ⊗ k(y) −→ H p−1 (Xy , Fy )

is an isomorphism for all y ∈ Y .

Proof of Theorem 3.14. Since Ly ⊗ My−1 is trivial and Xy is complete, the space of
sections H 0 (Xy , Ly ⊗ My−1 ) is isomorphic to k(y). So, using Lemma 3.15, we have
that p∗ (L ⊗ M −1 ) is locally free of rank one, and hence a line bundle. Thus, it
remains to prove that the natural map

α : p∗ p∗ (L ⊗ M −1 ) −→ L ⊗ M −1

is an isomorphism. The restriction to a fibre yields to the map

OXy ⊗ Γ(Xy , OXy ) −→ OXy ,

which is an isomorphism. Using Nakayama’s Lemma, this implies that α is surjective


and then, comparing the ranks, we have that it is an isomorphism.
32 CHAPTER 3. ABELIAN VARIETIES

Corollary 3.16 (See-saw principle). In addition to the assumptions of the Theo-


rem 3.14, suppose that Lx ' Mx for at least one point x ∈ X. Then L ' M .
Proof. The previous theorem shows that L ' M ⊗ p∗ N , for some line bundle N on
Y . If we restrict to {x} × Y , we have Lx ' Mx ⊗ (p∗ N )x . As Lx ' Mx , (p∗ N )x is
trivial. But (p∗ N )x = N , so N is trivial.

Proposition 3.17. Let X and Y be two varieties, with X complete. For a line
bundle L on X × Y , the set {y ∈ Y | Ly is trivial} is closed in Y .
The following theorem is again a general result from algebraic geometry, and it
is of crucial importance for the theory of abelian varieties.
Theorem 3.18 (Theorem of the cube). Let X and Y be two complete varieties and
let Z be a connected, locally noetherian scheme. Let x0 ∈ X, y0 ∈ Y and z0 ∈ Z.
If L is a line bundle on X × Y × Z whose restrictions to each of {x0 } × Y × Z,
X × {y0 } × Z and ×X × Y × {z0 } are trivial, L is trivial.
Proof. Since the projection X × Y × Z → Z is flat, we can see L as a family of
line bundles parametrized by Z. Let Z 0 be the maximal closed subscheme of Z over
which L is trivial (the existence of such a closed subscheme is due to a refinement
of Proposition 3.17). We want to show that Z 0 = Z. Since Z 0 is non empty, because
z0 ∈ Z 0 , and Z is connected, it is enough to prove that Z 0 is also an open subscheme.
Let ζ ∈ Z 0 , m be the maximal ideal of OZ,ζ and I ⊂ OZ,ζ be the ideal defining the
germ of Z 0 . We want to show that OZ,ζ and OZ 0T ,ζ are isomorphic, that is, I = (0).
Assume that I 6= (0). OZ,ζ is noetherian, so n mn = (0). Let n ∈ N such that
I ⊂ mn and I 6⊆ mn+1 , and let a1 = (I, mn+1 ). Then there exist an ideal a2 such
that
mn+1 ⊂ a2 ⊂ a1 and dimk(ζ) (a1 /a2 ) = 1.
Indeed, we have a chain 0 ⊂ a1 /mn+1 ⊂ mn /mn+1 of k(ζ)-vector spaces, and
a1 /mn+1 has dimension greater than zero. Hence we can find a vector subspace of
a2 /mn+1 ⊂ a1 /mn+1 of codimension one. Then a1 = a2 + k(ζ)a for some a ∈ a1 ,
and I ⊂ a1 but I 6⊂ a2 . Let a0 = m.
Let Zi ⊂ Spec(OZ,ζ ) be the closed subscheme defined by the ideals ai , i = 0, 1, 2.
We will show that the restriction of L to X × Y × Z2 is trivial. This implies that
Z2 is contained in Z 0 , which is a contradiction, since I 6⊂ a2 .
Let Li be the restrictions of L to X × Y × Zi , i = 0, 1, 2. We have that L0 and
L1 are trivial. So we can take a trivializing global section s ∈ Γ(L1 ). The inclusion
Z1 ⊂ Z2 induces a restriction map Γ(L2 ) → Γ(L1 ). We claim that L2 is trivial if
and only if s can be lifted to a global section s0 ∈ Γ(L2 ). To see this, suppose that
we have such a lift s0 . Since the underlying point set of both Z1 and Z2 is the same,
also the schemes X × Y × Z1 and X × Y × Z2 have the same point sets. If s(P ) = 0
for some point P , then also s(P ) = 0, but this contradicts the assumption that s
trivializes L1 . So s0 is nowhere zero, and we know that L2 is locally free of rank
1, and hence s0 trivializes L2 . Conversely, if L2 is trivial, then the restriction map
Γ(L2 ) → Γ(L1 ) is nothing but Γ(OZ2 ) → Γ(OZ1 ), and this map is surjective.
We know that the structure sheaves of Zi are related by the exact sequence

0 OZ0 OZ2 OZ1 0,


3.2. LINE BUNDLES ON ABELIAN VARIETIES 33

and from this we have the exact sequence

0 L0 L2 L1 0.

We fix an isomorphism L0 ' OX×Y . A part of the long exact cohomology sequence
is
... Γ(L2 ) Γ(L1 ) δ H 1 (X × Y, OX×Y ) ...

Suppose that δ(s) = ξ ∈ H 1 (X × Y, OX×Y ). Then s can be lifted if ξ = 0.


By hypotesis, the restrictions of L2 to {x0 }×Y ×Z2 and X ×{y0 }×Z2 are trivial.
We can write the maps i1 = (idX , y0 ) : X → X × Y and i2 = (x0 , idY ) : Y → X × Y ,
and being trivial means that ξ has trivial image under i∗1 : H 1 (X × Y, OX×Y ) →
H 1 (X, OX ) and under i∗1 : H 1 (X × Y, OX×Y ) → H 1 (Y, OY ). Finally, since X and Y
are complete, we have that H 1 (X, OX )⊗H 0 (X, OX ) ' H 1 (X, OX ) and H 1 (Y, OY )⊗
H 0 (Y, OY ) ' H 1 (Y, OY ). Then, using Künneth formula, we have the isomorphism

H 1 (X × Y, OX×Y ) ' H 1 (X, OX ) ⊕ H 1 (Y, OY ),

so ξ = 0 and s can be lifted.

This theorem reveals the quadratic character of line bundles, which becomes
clearer in the sequel. Indeed, if f (x) = ax2 + bx + c is a quadratic function on the
real line, then

f (x + y + z) − f (x + y) − f (x + z) − f (y + z) + f (x) + f (y) + f (z)

is constant. In the corollary below we state the analogue of this for line bundles on
abelian varieties.
We will use the following notation. Let A be an abelian variety, and let pi : A ×
A×A → A be the projection onto the ith factor (e.g. p1 (x, y, z) = x), let pij = pi +pj
(e.g. p23 (x, y, z) = y +z) and let p123 = p1 +p2 +p3 (so that p123 (x, y, z) = x+y +z).
Corollary 3.19. Let A be an abelian variety and let L be a line bundle on A. Then
the line bundle

Θ(L) := p∗123 L ⊗ p∗12 L−1 ⊗ p∗13 L−1 ⊗ p∗23 L−1 ⊗ p∗1 L ⊗ p∗2 L ⊗ p∗3 L

on A × A × A is trivial.
Proof. Let m, p, q be the maps from A × A to A sending (x, y) to x + y, x and y
respectively. The restriction of Θ(L) to A × A × {0} ' A × A is

m∗ L ⊗ m∗ L−1 ⊗ p∗ L−1 ⊗ q ∗ L−1 ⊗ p∗ L ⊗ q ∗ L ⊗ OA×A

which is obviously trivial. In the same way, restricitons on A×{0}×A and {0}×A×A
are trivial, and so Θ(L) is trivial.

Corollary 3.20. Let X be a variety and A be an abelian variety, and let f , g and
h be morphism form X to A. Then for every line bundle L on A, the bundle

(f + g + h)∗ L ⊗ (f + g)∗ L−1 ⊗ (f + h)∗ L−1 ⊗ (g + h)∗ L−1 ⊗ f ∗ L ⊗ g ∗ L ⊗ h∗ L

on X is trivial.
34 CHAPTER 3. ABELIAN VARIETIES

Proof. The bundle in question is the inverse image of the bundle Θ(L) of Corol-
lary 3.19 by the map (f, g, h) : X → A × A × A, and hence it is trivial.

An important corollary is the following:

Corollary 3.21 (Theorem of the square). Let A be an abelian variety and let L be
a line bundle on A. Then, for all a, b ∈ A,

t∗a+b L ⊗ L ' t∗a L ⊗ t∗b L.

Proof. Using Corollary 3.20 with f the identity on A and g and h the constant maps
in A with images a and b, we have that

t∗a+b L ⊗ t∗a L−1 ⊗ t∗b L−1 ⊗ L

is trivial.

Remark 3.22. The results showed in this section can be restated in terms of divisors.
In particular, for any divisor D on A, and for all a, b ∈ A, the above corollary asserts
that
Da+b + D ≡ Da + Db ,
where Dx is the translate D + x (for x ∈ A) and ≡ is the linear equivalence between
divisors.

Corollary 3.23. Let A be an abelian variety and let L be a line bundle on A. Let
Pic(A) be the group of isomorphism classes of line bundles on A. Then the map

φL : A −→ Pic(A) (3.1)
a 7−→ t∗a L ⊗ L−1

is a homomorphism.

The homomorphism φL will play a very important role in the sequel.

Let A be an abelian variety. For every n ∈ Z we can define a homomorphism


[n] := [n]A : A → A called multiplication by n. Indeed, for n ≥ 1, [n] sends a ∈ A to
its nth multiple, i.e. [n](a) = a + a + · · · + a (n summands); for n = −m ≤ −1, we
have [n] := invA ◦ [m]. In general, we simply denote [n] by n.

Corollary 3.24. For every line bundle L on an abelian variety A, we have


2 +n)/2 2 −n)/2
n∗ L ' L(n ⊗ (−1)∗ L(n .

In paticular,
2
• n∗ L ' Ln if L is symmetric, that is, (−1)∗ L ' L.

• n∗ L ' Ln if L is antisymmetric, that is, (−1)∗ L ' L−1 .


3.2. LINE BUNDLES ON ABELIAN VARIETIES 35

Proof. Using Corollary 3.20 where f, g, h : A → A are the maps [n], [1] and [−1]
respectively, we obtain that

n∗ L ⊗ (n + 1)∗ L−1 ⊗ (n − 1)∗ L−1 ⊗ n∗ L ⊗ L ⊗ (−1)∗ L

is trivial, or in other words

(n + 1)∗ L ' n∗ L2 ⊗ (n − 1)∗ L−1 ⊗ L ⊗ (−1)∗ L.

Then the assertion follows by induction, starting from n = 1.

We want now to prove that an abelian variety is projective. To do this, we first


recall some definitions and results that involve divisors.
Remark 3.25. In what follows, we assume for simplicity that k is an algebraically
closed field. Later we will explain how to remove this condition.

Definition 3.26. Let X be a complete nonsingular variety over k. If D is a divisor


on X, we define the complete linear system of D as the set of all effective divisors
which are linearly equivalent to D:

D := {D + div(f ) | f ∈ L(D)}.

For any subspace W ⊂ L(D), the set {D + div(f ) | f ∈ W } is called a linear system.

Let D be a divisor in D, and let f0 , f1 , . . . , fn be a basis for L(D). There is a


rational map
P 7→ [f0 (P ) : f1 (P ) : · · · : fn (P )] : X 99K Pn .
Up to a projective linear transformation, this rational map depends only on the
linear system D.

Definition 3.27. Let D be a complete linear system.

• An effective divisor E is called a fixed divisor if, for every D ∈ D, D ≥ E.

• A point P ∈ X is called a base point of D if, for all D ∈ D, P ∈ supp(D).

Without loss of generality, we can assume that D has no fixed divisor (but there
may be some base points).

Proposition 3.28. Ther rational map X 99K Pn defined by D is defined at a point


P ∈ X if and only if P is not a base point of D.

Definition 3.29. Let D be a linear system.

• D is said to separate points if for any pair of points P, Q ∈ X, there exists


D ∈ D such that P ∈ supp(D) and Q ∈ / supp(D).

• D is said to separate tangent directions if for every P ∈ X and every non-zero


tangent t to X at P , there exists a divisor D ∈ D such that P ∈ D but
t∈/ TP D.
36 CHAPTER 3. ABELIAN VARIETIES

Definition 3.30. Let X be a non-singular complete variety. A divisor D on X is


called very ample if the complete linear system it defines gives a closed immersion
of X into Pn . D is called ample if nD is very ample for some n ∈ N+ .
Proposition 3.31. Let D be a complete linear system with no base points. Then
the map X 99K Pn defined by D is a closed immersion (that is, the element defining
D is a very ample divisors) if and only if D separates points and separates tangent
directions.
We are ready to state and prove the following theorem.
Theorem 3.32. Every abelian variety A is projective.
Proof. The first step is to prove that there exists a finite set of prime divisors Zi
such that
T
1. Zi = {0},
T
2. T0 (Zi ) = {0},
P
that is, Zi separates 0 from the other points of A an separates the tangent direc-
tions at 0. So we show that for any point a ∈ A, 0 and a are contained in an open
affine subvariety of A. Let U be an open affine neighbourhood of 0, and let U + a
be its translation by a. Choose a point u ∈ U ∩ (U + a). Then

u ∈ U + a =⇒ 0 ∈ U + a − u,
u ∈ U =⇒ u + a ∈ U + a =⇒ a ∈ U + a − u,

so U 0 = U + a − u is an open affine neighbourhood of both 0 and a. We can identify


U 0 with a subset of An for some n ∈ N. There is an hyperplane H in An passing
through 0 but not a, and we take Z1 to be the closure of H ∩ U 0 in A. If there is
a point b on Z1 other than 0, we can choose Z2 to pass through 0 but not b, and
we can iterate this process. Since A has the descending chain condition for closed
subsets, the process will end T in a finite number of iterations, that is, we obtain a
finite set of Zi , such that Zi = {0}. Now, we take an open affine neighbourhood
V of a, and let t ∈ T0 (a). Suppose t ∈ T0 (Zi ) for all i. If we embed V in An , we
can choose an hyperplane H through 0 such that t ∈ / H, and we add the closure Z
of H ∩ V in A to the set of divisors Zi . We can continue in this way until the new
set of Zi has Pthe property (2).
Let D = Zi , where the set of Zi satisfies conditions (1) and (2). We wanto to
show that 3D defines an embedding of A into Pn , for some nßN, i.e. that D is an
ample divisor. For any family of points {a1 , . . . , an ; b1 , . . . , bn } of A, Corollary 3.21
shows that X X
(Zi,ai + Zi,bi + Zi,−ai −bi ) ≡ 3Zi = 3D,
i i
where Zi,x is the translate Zi + x. Let a, b ∈ A be two distinct points. By condition
(1), there exists a Zi (say Z1 ) which does not contain b − a. Choose a1 = a. Then
Z1,a1 passes through a but not b. The sets

{b1 | Z1,b1 passes through b}


{b1 | Z1,−a1 −b1 passes through b}
3.3. DUAL ABELIAN VARIETY 37

are proper closed subsets of A. Thus, it is possible to choose a b1 that lies on


neither of these. In the same way, ai and bi for i ≥ 2 can be chosen so that
P of Zi,ai , Zi,bi and Zi,−ai −bi passes through b. Then a is in the support of
none
i (Zi,ai + Zi,bi + Zi,−ai −bi ), while b is not, and this prove that the linear system
defined by 3D separates points. The proof that 3D separates tangent directions is
similar.

Finally, the next proposition allow us to remove the condition that k is alge-
braically closed from Theorem 3.32.

Proposition 3.33. Let X be a complete non-singular variety over a field k.

1. If D and E are ample divisors on X, so is D + E.

2. If D is an ample divisor on X, then D|Y is ample for any closed subvariety Y


of X, assuming that D|Y is defined.

3. A divisor D on X is ample if and only if its extension of scalars to k is ample


on Xk .

4. X has an ample divisor if Xk has an ample divisor.

3.3 Dual abelian variety


Definition 3.34. Let A and B be abelian varieties. The homomorphism f : A → B
is said to be an isogeny if it is surjective and has finite kernel.

Let L be a line bundle on A, and consider the bundle m∗ L ⊗ p∗1 L−1 , where the
maps m, p1 : A × A → A send (a, b) respectively to a + b and a. We can see it as
a family of line bundles on the first factor A parametrized by the second factor A.
We define the set

K(L) := {a ∈ A | (m∗ L ⊗ p∗1 L−1 )|A×{a} is trivial}.

By Proposition 3.17, this set is a closed subset of A.


We note that m ◦ j = ta and p1 ◦ j = idA , where j : A × {a} ,→ A × A is the
inclusion map, and so

(m∗ L ⊗ p∗1 L−1 )|A×{a} = j ∗ (m∗ L ⊗ p∗1 L−1 ) = t∗a L ⊗ L−1 .

Hence, we have

K(L) = {a ∈ A | t∗a L ⊗ L−1 is trivial} = {a ∈ A | φL (a) = 0}, (3.2)

where φL is the homomorphism (3.1) defined before. Thus, K(L) is the kernel of
φL .

Proposition 3.35. Let L be a line bundle such that H 0 (A, L) 6= 0. Then L is ample
if and only if K(L) has dimension zero, i.e. is finite.

Proof. See [Mil08, § 8].


38 CHAPTER 3. ABELIAN VARIETIES

Proposition 3.36. Let A be an abelian variety and let L be a line bundle on A.


The following conditions are equivalent:
1. K(L) = A.

2. t∗a L ' L on Ak , for all a ∈ A(k).

3. m∗ L ' p∗1 L ⊗ p∗2 L.


Proof. The first two conditions are equivalent by definiton of K(L). From condition
(3) we obtain that
(m∗ L ⊗ p∗1 L−1 )|A×{a} ' p∗2 L|A×{a} ,
which is trivial, so (3) implies (1). For the converse, we note that, fore every a ∈
A p∗2 L|A×{a} is trivial and, by condition (1), (m∗ L ⊗ p∗1 L−1 )|A×{a} is trivial too.
Moreover, (m∗ L ⊗ p∗1 L−1 )|{0}×A = p∗2 L|{0}×A , so we can apply the See-saw principle
and obtain (3).

Definition 3.37. Let A be an abelian variety. We define Pic0 (A) as the subgroup
of Pic(A) such that its elements are isomorphism classes of line bundles satisfying
the conditions in Proposition 3.36.
Notation. As in Pic(A), we usually write L ∈ Pic0 (A) to mean the isomorphism
class represented by L.
So, Pic0 (A) can be seen as the set of line bundles L such that the homomorphism
φL is identically zero. We can make some observations about Pic0 (A):
Proposition 3.38. Let A be an abelian variety, X be a variety and f, g : X → A
be two morphisms. Let L ∈ Pic0 (A). Then

(f + g)∗ L ' f ∗ L ⊗ g ∗ L.

In particular, we have that n∗ L ' Ln .


Proof. Since f + g = m ◦ (f × g), the result descends from applying (f × g)∗ to the
isomorphism (3) in Proposition 3.36. The second part is an easy induction on the
first one.

Proposition 3.39. Let A be an abelian variety, and let L ∈ Pic0 (A) with L non
trivial. Then H i (A, L) = 0 for every i ∈ Z.
Proof. The first step is to show that H 0 (A, L) = 0. Suppose that H 0 (A, L) 6= 0. So
there exists a non-negative divisor D such that L ' OA (D). Then L−1 ' (−1)∗ L '
OA ((−1)∗ D), and hence OA ' L ⊗ L−1 ' OA (D + (−1)∗ D). Thus, D + (−1)∗ D = 0
and then D = 0, that is, L = OA , which is a contradiction.
Now, let k ∈ N be the minimal integer such that H k (A, L) 6= 0, so k ≥ 1. Let
s1 : A → A × A be the map that sends a to (a, 0). Then the composition
s1 m
A A×A A

is the identity, and from this we have the composition

m∗ s∗1
H k (A, L) H k (A × A, m∗ L) H k (A, L),
3.3. DUAL ABELIAN VARIETY 39

which is again the identity. Using (3) from Proposition 3.36 and the Künneth
formula, we obtain that
X
H k (A × A, m∗ L) ' H k (A × A, p∗1 L ⊗ p∗2 L) ' H i (A, L) ⊗ H j (A, L).
i+j=k

Since k ≥ 1, either i < k or j < k, and then H i (A, L) ⊗ H j (A, L) = 0 for every i, j
such that i + j = k, that is, H k (A × A, m∗ L) = 0. Thus the identity of H k (A, L)
factors through zero, so H k (A, L) = 0.

We now want to endow Pic0 (A) with a geometric structure, showing that it is
naturally isomorphic to a particular abelian variety. To do this, we provide the
following definitions:

Definition 3.40. Let A be an abelian variety. Let Ab be an abelian variety and P


be a line bundle on A × A
b which have the following properties:

1. P|A×{α} ∈ Pic0 (A) for all α ∈ A.


b

2. P|{0}×Ab is trivial.

The variety A b is called the dual abelian variety and the bundle P is called the
b P) has the following universal property: for any pair
Poincaré bundle if the pair (A,
(T, L) consisting of a variety T and a line bundle L such that

i. L|A×{t} ∈ Pic0 (A) for all t ∈ T , and

ii. L|{0}×T is trivial,

b such that (1 × f )∗ P ' L.


there exists a unique morphism f : T → A
b P) (if
The universal property in the definition determines uniquely the pair (A,
it exists) up to a unique isomorphism. Moreover, it says that
b ' {line bundles on A × T satisfying (i) and (ii)}/ ' .
Hom(T, A)

So, if we consider T = Spec(k), using the fact that A(k)


b = Hom(Spec(k), A),b we have
0
A(k)
b ' Pic (A), that is, the rational points of the dual abelian variety correspond
to the elements of Pic0 (A).

The next step, in order to ensure the existence of the dual abelian variety, is the
construction of the pair (A,b P). We will do this when the field k is of characteristic
zero, since it is an easy case. Furthermore, we assume that k is algebraically closed.
The construction uses quotients of varieties by the action of a finite group, so we
give some results about them.

Proposition 3.41. Let X be an algebraic variety over an algebraically closed field


k, and let G be a finite group of automorphisms of X. Assume that for any x ∈ X,
the orbit Gx of x is contained in an affine open subset of X. Then there exists a pair
(Y, π), where Y is a variety and π : X → Y is a morphism, satisfying the following
conditions:
40 CHAPTER 3. ABELIAN VARIETIES

1. as a topological space, (Y, π) is the quotient of X by the action of G,

2. if π∗ (OX )G denotes the subsheaf pf G-invariants of π∗ (OX ) for the action of


G on π∗ (OX ), the natural homomorphism OY → π∗ (OX )G is an isomorphism.

The pair (Y, π) is determined up to an isomorphism by these conditions. The map


π is finite, surjective and separable, and Y is affine if X is affine. If further G acts
freely on X, π is an étale morphism.

Proof. See [Mum70, § II.7].

The variety Y in the theorem is denoted by X/G and it is called the quotient of
X by G.
Remark 3.42. Let A be an abelian variety over k (with k algebraically closed), and
let G be a finite subgroup of A. Then the quotient B = A/G is an abelian variety
and the map π : A → B is an isogeny with kernel G.
We define a coherent G-sheaf on a variety X as a coherent sheaf F of OX -
modules together with an action of the finite group G on F compatible with its
action on X.

Proposition 3.43. Assume that the finite group G acts freely on X, and let Y =
X/G, with the quotient map π : X → Y . Then the functor F 7→ π ∗ (F) is an
equivalence between the category of coherent OY -modules and that of coherent G-
sheaves on X, under which locally free sheaves correspond to locally free sheaves of
the same rank.

Proof. See [Mum70, § II.7].

The next result is the key point in the theory of Pic0 (A).

Theorem 3.44. Let A be an abelian variety, let L be an ample bundle on A and let
M ∈ Pic0 (A). Then, for some a ∈ A,

M ' t∗a L ⊗ L−1 ,

that is, the map φL : A → Pic0 (A) is surjective.

Proof. See [Mum70, § II.8].

Now, we can sketch the construction. Let L be a line bundle on A, and consider
the line bundle
Λ(L) := m∗ L ⊗ p∗1 L−1 ⊗ p∗2 L−1

on A × A, which is called the Mumford bundle. Then we have that

• Λ(L)|{0}×A = L ⊗ L−1 = OA , and

• for every a ∈ A, Λ(L)|A×{a} = t∗a L ⊗ L−1 = φL (a) ∈ Pic0 (A).


3.3. DUAL ABELIAN VARIETY 41

Therefore, Λ(L) defines a family of bundles on A parametrized by A, such that


(Λ(L))a = φL (a). If L is ample, we know from Theorem 3.44 that each element
of Pic0 (A) is represented by (Λ(L))a for a finite number (different from zero) of
a ∈ A. Then, if (A, b P) exists, there is a unique isogeny φ : A → A b such that

(1 × φ) P = Λ(L); moreover, we have that φ = φL .
In characteristic zero, we know that the kernel of φ is determined by its un-
derlying set, which is K(L), as we see in (3.2). Since we have chosen an ample
L, by Proposition 3.35, K(L) is a finite group. Then we can define A b to be the
quotient A/K(L), whose existence is provided by Proposition 3.41. The action of
K(L) on the second factor of A × A lifts to an action on Λ(L) over A × A, which,
by Proposition 3.43, corresponds to a bundle P on A× A b such that (1×φ)∗ P = Λ(L).
We need to check that the pair A, b P) just constructed is really the one defined
in Definition 3.40, that is, it has the universal property, required in the definition,
for families of bundles parametrized by normal varieties over k (this will also imply
that it is independent of the choice of L). Let M on A × T be such a family, and
let N := p∗12 M ⊗ p∗13 P −1 on A × T × A.b Then we have

N |A×{(t,α)} ' Mt ⊗ Pα−1 .

If Γ denote the closed subset of T × A


b of points (t, α) such that N |A×{(t,α)} is trivial,
then Γ is the graph of a map T → A b that sends a point t ∈ T to the unique α ∈ A b
such that Pα ' Mt . Thinking of Γ as a closed subvariety of T × A, we see that
b
the projection Γ → T is a bijection. Since k has characteristic zero, this shows that
Γ and T are birationally equivalent varieties, and since T is normal, we obtain, by
Zariski’s Main Theorem, that Γ → T is an isomorphism.
Thus, we define the morphism f : T ' Γ → A b to be the restriction of the
projection on the second factor. We have that:
• the restrictions M |{0}×T and (1 × f )∗ P|{0}×T are both trivial, and

• for each t ∈ T , (1 × f )∗ P|A×{t} = Pf (t) ' Mt .


Then, by See-saw principle, we have (1 × f )∗ P = M , as required.
A similar construction can be made also when k has positive characteristic,
but the proofs become much more complicated. Indeed, in order to define the
dual variety Ab as the quotient A/K(L), we need some results of the theory of
group schemes and quotients of finite ones, since K(L) need not to be reduced.
See [Mum70, § III.11-13], [Pol03, § II.9] and [GM07, § 3-6] for the construction in
every characteristic.
Remark 3.45. If f : A → B is an homomorphism of abelian varieties, the induced
map Pic(B) → Pic(A) maps Pic0 (B) into Pic0 (A), and thus we ge the natural
map fb: Bb → A, b and this is a morphism. Indeed, if Q is the Poincaré bundle on
B × B, then (f × 1)∗ Q is a line bundle on A × B
b b such that, for every β ∈ B,
b
∗ ∗ 0
the restriction (f × 1) Q|A×{β} represents f (β) ∈ Pic (A), and by the universal
b→A
property, fb: B b is a morphism. Moreover, if f is an isogeny, so is fb.
In the end of this section we want to show that the relationship between A and
A is in reality symmetric, like the one between two vector spaces set up by a bilinear
b
pairing. To do this, we need the following definition:
42 CHAPTER 3. ABELIAN VARIETIES

Definition 3.46. Let A and B be abelian varieties. A divisorial correspondence


between A and B is a line bundle Q on A × B whose restrictions to {0} × B and
A × {0} are trivial.

Next proposition clarifies the idea of symmetry.

Proposition 3.47. Let A and B be abelian varieties of the same dimension over a
field of characteristic zero, and let Q be a divisorial correspondence between A and
B. Then the following are equivalent:

1. If Q|{a}×B is trivial, then a = 0.

2. If Q|A×{b} is trivial, then b = 0.

If these hold, then A ' B,


b with Q isomorphic to the Poincaré bundle PB of B, and
B ' A, with Q isomorphic to the Poincaré bundle PA of A.
b

Proof. By symmetry, we only need to prove (2) from (1). So we suppose that (1)
holds. Then, there exists an injective morphism f : A → B b such that (f ×1)∗ PB ' Q.
Since dim(A) = dim(B), f is also surjective, and this implies, in characteristic zero,
that f is an isomorphism, that is, A ' B.b Now, let g : B → A b be the morphism such

that (1 × g) PA ' Q. To prove (2), we have to show that g is injective. Suppose
that g is not injective. So we can find a non-empty finite subgroup K ⊂ Ker(g), and
g factorizes as
η g̃
B B/K A,
b

where η is the natural homomorphism. Then, if N is the line bundle (1 × g̃)∗ PA on


A × B/K, we have that (1 × η)∗ N ' Q. N induces a homomorphism ψ : A → B/K, [

and, by the isomorphism (1 × η) N ' Q, we have that the composition

ψ [ ηb
A B/K B
b

is the homomorphism f defined by Q, and thus the composition is an isomorphism.


Hence, ψ is injective, and, since dim(A) = dim(B), both ψ and ηb are isomorphisms.
But here we have a contradiction, because ηb has a non-zero kernel, namely the dual
abelian group of K. Hence, g has to be injective.

This proposition, which can be extended in the case of varieties over a field of
positive characteristic, leads to the following corollary, that explains why we call A
b
the dual variety.

Corollary 3.48 (Duality hypotesis). Let A be an abelian variety. Then the canon-
ical morphism i : A → A, defined by the Poincaré bundle P on A × A
b (regarded as a
b
b
family of line bundles on A
b parametrized by A), is an isomorphism.
Chapter 4

Equivalences between derived


categories

In the previous chapters we have introduced the derived categories of coherent


sheaves of a smooth variety (or, more in general, of a noetherian scheme). A funda-
mental question that arises is the following: when are derived categories of coherent
sheaves of two different smooth (projective) varieties equivalent, as triangulated cat-
egories? As we can immediately see, this is a quite wide topic, and we do not know
the complete answer to the question above.
However, we already have some interesting results in this direction. A first
example, that is, a theorem due to Bondal and Orlov, gives us the complete answer
to the question in the case when either the canonical or anticanonical bundle of the
variety is ample.

Theorem 4.1 (Bondal, Orlov). Let X be a smooth projective variety whose canoni-
cal (or anticanonical) bundle is ample, and let Y be a smooth algebraic variety, such
that the category Db (X) is equivalent to the derived category Db (Y ). Then X is
isomorphic to Y .

4.1 From equivalence to isomorphism


Here we are interested in the case of derived categories coherent sheaves of abelian
varieties, whose properties are described in the previous chapter. So, from now on,
we denote by A an abelian variety of dimension n, and by P its Poincaré bundle.
Notation. Suppose that A1 , . . . , Am are abelian varieties, and let (α1 , . . . , αm ) be
b1 × · · · × A
a rational point of A bm . Then we denote by Pα the line bundle on Ai
i
corresponding to the point αi , and by P(α1 ,...,αm ) the line bundle Pα1  · · ·  Pαm
on the product A1 × · · · × Am .
The Poincaré bundle provides an example of an exact equivalence between the
derived categories of coherent sheaves of the two varieties A and A,
b which are in
general non-isomorphic (this example underlines that two varieties with equivalent
derived categories do not have to be necessarily isomorphic).

43
44 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

Theorem 4.2. Let A an abelian variety of dimension n, and let P be its Poincaré
bundle. Then there are isomorphism of functors
ΨP ◦ ΦP ' (−1A )∗ [−n]
and
ΦP ◦ ΨP ' (−1Ab)∗ [−n].
In other words, ΦP : Db (A) → Db (A)b gives an equivalence of derived categories, and

its quasi-inverse is (−1Ab) ◦ ΨP [n].
Proof. See [Muk81, § 2] for the original proof for varieties over a field of characteristic
zero (the theorem is also true over an arbitrary field).

Remark 4.3. The definition of Fourier-Mukai transform has been introduced by


Mukai in [Muk81] in order to construct this equivalence between Db (A) and Db (A), b
in the case of abelian varieties, using the Poincaré bundle as the kernel of the trans-
form. After, the idea of Fourier-Mukai has been extended to the cateogires of every
projective variety.
Let (a, α) be a k-rational point in A × A.b This point determines a functor from
b
D (A) to itself by
Φ(a,α) (_) := ta∗ (_) ⊗ Pα = t∗−a (_) ⊗ Pα . (4.1)
This functor is a Fourier-Mukai transform, and its kernel is
S(a,α) = OΓa ⊗ p∗2 Pα (4.2)

on A × A, b where Γa is the graph of the translation map ta . Clearly, the functor


Φ(a,α) is an autoequivalence.
We now want to collect all the functors Φ(a,α) into a single functor from Db (A×A)
b
to Db (A × A) which maps k(a, α) to S(a,α) . To do this, we consider the object
PA := p∗14 O∆ ⊗ p∗23 P
in Db ((A × A)
b × (A × A)), and the morphism

µA : A × A −→ A × A,
(a1 , a2 ) 7−→ (a1 , m(a1 , a2 ))
where m is the group operation of A.
Definition 4.4. Using the object PA and the morphism µA , we define the functors
ΦPA : Db (A × A)
b −→ Db (A × A) and RµA∗ : Db (A × A) −→ Db (A × A),

and we denote their composition with ΦSA : Db (A × A)


b → Db (A × A).

The kernel SA can be described explicitely. Indeed, assuming that ∆ is the


diagonal of A × A and (m ◦ p13 , p4 ) : A × A b × A × A → A × A is the morphism that
sends (a1 , α, a3 , a4 ) to (m(a1 , a3 ), a4 ), we have
SA = (m ◦ p13 , p4 )∗ O∆ ⊗ p∗23 P.
However, we do not use this explicit formula in the sequel, so we do not prove it.
4.1. FROM EQUIVALENCE TO ISOMORPHISM 45

Lemma 4.5. The functor ΦSA defined before is an equivalence and, for every k-point
(a, α) ∈ A × A,
b

1. ΦSA maps k(a, α) to S(a,α) , defined in (4.2), and

b to the object O{−a}×A ⊗ p∗ Pα [−n].


2. ΦSA maps the line bundle P(α,a) on A × A 4

Proof. By Definition 4.4, ΦSA is the composition of two equivalences, hence it is an


equivalence too. For the first point, consider the commutative diagram

{(a, α)} × A × A A×A


b×A×A
p1 p23

{α} × A b × A,
A

where {(a, α)} × A × A ' A × A and {α} × A ' A (so p1 : A × A → A is the


projection on the first factor). Then, remembering that, for a sheaf E on X × Y and
x ∈ X, ΦE (k(x)) ' E|{x}×Y , we have

ΦPA (k(a, α)) ' (p∗14 O∆ ⊗ p∗23 P)|{(a,α)}×A×A ,

and, using the diagram above,

(p∗23 P)|{(a,α)}×A×A = p∗1 (P|{α}×A ) = p∗1 Pα .

In a similar way, we have

(p∗14 O∆ )|{(a,α)}×A×A = p∗2 (O∆ |{α}×A ) = p∗2 k(a) = OA×{a} ,

where p2 : A × A → A is the projection on the second factor, and hence we obtain


that ΦPA maps (k(a, α)) to OA×{a} ⊗p∗1 Pα . Now we apply the functor RµA∗ , keeping
in mind the commutative diagram

µA
A×A A×A
p1 p2

A A,

and we obtain that RµA∗ (OA×{a} ⊗ p∗1 Pα ) = S(a,α) . For the second point, we have
that

ΦPA (P(α,a) ) = Rp34∗ (p∗12 P(α,a) ⊗ p∗14 O∆ ⊗ p∗23 P)


= Rp34∗ (p∗14 (O∆ ⊗ q1∗ Pα ) ⊗ p∗23 (P ⊗ q2∗ Pa ))
= Rq3∗ (P ⊗ q2∗ Pa )  Rq4∗ (O∆ ⊗ q1∗ Pα )
= ΨP (Pa )  ΦO∆ (Pα ) = ΨP (Pa )  Pα ,
46 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

where the maps qi are the projections that commute with the usual ones (i.e., q1 ◦
p14 = p1 ◦ p12 , q2 ◦ p23 = p2 ◦ p12 , etc.), following the diagrams

A×A
b×A×A A×A
b×A×A
p14 p23

A×A b×A
A
q1 q4 q2 q3

A A A
b A

By Theorem 4.2, ΨP ◦ ΦP ' (−1A )∗ [−n], so we have

ΨP (Pa ) = ΨP (ΦP (k(a))) = (−1A )∗ (k(a))[−n] = k(−a)[−n],

and hence

ΦPA (P(α,a) ) = · · · = ΨP (Pa )  Pα = k(−a)[−n]  Pα


= p∗3 (k(−a)) ⊗ p∗4 (Pα )[−n] = O{−a}×A ⊗ p∗4 Pα [−n].

Finally we can see that RµA∗ maps O{−a}×A ⊗ p∗4 Pα [−n] to itself.

Let A and B be two abelian varieties whose derived categories of coherent sheaves
are equivalent, and let ΦE be the equivalence, with E ∈ Db (A × B).

Definition 4.6. We denote by J (E) the object that represents the functor

Φ−1
SB ◦ AdE ◦ΦSA ,

where AdE is the functor introduced in Remark 2.30. Thus, there is a commutative
diagram
ΦSA
Db (A × A)
b Db (A × A)
ΦJ (E) AdE

Db (B × B)
b
ΦSB
Db (B × B).

The next theorem is fundamental for the description of abelian varieties whose
derived categories of coherent sheaves are equivalent.

Theorem 4.7 (Orlov). Let ΦE : Db (A) → Db (B) be an equivalence, with kernel


E ∈ Db (A × B). Then there exist a homomorphism

fE : A × A
b −→ B × B
b

of abelian varieties and a line bundle LE on A × A


b such that

1. fE is an isomorphism, and

2. the object J (E), defined above, is isomorphic to i∗ (LE ),


4.1. FROM EQUIVALENCE TO ISOMORPHISM 47

where i is the embedding of A×A


b in (A×A)×(B×
b B)
b as the graph of the isomorphism
fE .
Before giving the proof of the theorem, we state a couple of lemmas in order to
assume that the field k is algebraically closed.
Notation. The algebraic closure of k is k, so, for a variety X, we put X := X ×Spec(k)
Spec(k) and we denote by F the inverse image of the sheaf F under the morphism
X → X.
Lemma 4.8. Let X be a smooth variety, and let F ∈ Coh(X). Suppose that we have
a closed subvariety j : Z ,→ X and an invertible sheaf L on Z such that F ' j∗ L.
Then there exists a closed subvariety i : Y ,→ X and an invertible sheaf M on Y
such that F ' i∗ M and j = i.
Proof. The argument of the lemma is local, so we can assume that X is an affine
variety, namely Spec(A), with A a commutative ring, and F is associated to an A-
module M . We have A := A ⊗k k and M := M ⊗k k. Let J ⊂ A be the annihilator
of M , and let J 0 ⊂ A be theL annihilator of M . If we consider a basis {ei }i∈I of
the field k over k, then M = i∈I M ei as a module over P A. Clearly, we have that
J ⊗k k ⊆ J 0 . On the other hand, if we take an element i∈I ai ⊗ ei ∈ A that belongs
to J 0 , we have that
P
P i∈I ma i ⊗ e i = 0 for every m ∈ M . Thus, for every i ∈ I,
ai ∈ J, and hence i∈I ai ⊗ ei ∈ J ⊗ k, that is, J ⊗ k = J . 0

Let B := A/J and B := A/J 0 . Then, since J ⊗ k = J 0 , we have B = B ⊗k k.


From the hypotheses, we have that M is a projective module of rank 1 over the
algebra B, and M = M ⊗B B. Since B is a strictly flat B-algebra, we obtain that
M is a projective B-module of rank 1.

The following lemma says that the property of a functor to be fully faithful (or
an equivalence) is stable with respect to field extension.
Lemma 4.9. Let X and Y be two smooth varieties over k, and let E ∈ Db (X × Y ).
Suppose that F/k is a field extension, so we have the varieties
X 0 = X ×Spec(k) Spec(F ) and Y 0 = Y ×Spec(k) Spec(F ).

Let E 0 be the lifting of E to the category Db (X 0 ×Y 0 ). Then, the functor ΦE : Db (X) →


Db (Y ) is fully faithful (resp. an equivalence) if and only if ΦE 0 : Db (X 0 ) → Db (Y 0 )
is fully faithful (resp. an equivalence).
Proof. First, suppose that ΦE is a fully faithful functor. We denote with ΦLE its left
adjoint. Then, the composition ΦLE ◦ ΦE is the identity functor of Db (X), which
is a Fourier-Mukai transform with kernel O∆ , where ∆ is the diagonal in X × X.
Using Proposition 2.24 and the flat base change, we have that the composition
ΦLE 0 ◦ ΦE 0 is represented by the structure sheaf O∆0 , where ∆0 is the diagonal in
X 0 × X 0 . Hence, the functor ΦE 0 is fully faithful.
Conversely, suppose that ΦE 0 is fully faithful. We consider the composition
ΦE ◦ ΦE , which is represented by an object Q ∈ Db (X × X). So there is a canonical
L

morphism φ : Q → OX . Since ΦE 0 is fully faithful, the lifted morphism φ0 : Q0 → OX 0


is an isomorphism. Thus, φ is an isomorphism too, and hence ΦE is a fully faithful
functor.
In the same way we can prove the statement for equivalences.
48 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

Now we can prove the principal theorem.

Proof of Theorem 4.7. By Lemma 4.8 and Lemma 4.9, we can assume that the field
k is algebraically closed.

Step 1. We denote with eA and eB the closed points of A × A b and B × B b


respectively that are the identity elements of the group structure, that is, eA =
(0A , OA ) and eB = (0B , OB ). We want to find the image of k(eA ) under the functor
ΦJ (E) . By Definition 4.6, we have that

ΦJ (E) = Φ−1
SB ◦ AdE ◦ΦSA ,

so we can the computation into three parts. Using Lemma 4.5, we have

ΦSA (k(eA )) = S(eA ) = S(0,OA ) = O∆A ⊗ p∗2 POA = O∆A .

Now we want to calculate AdE (O∆A ), then, by Remark 2.30,

ΦAdE (O∆ ) = ΦE ◦ ΦO∆A ◦ Φ−1 −1


E = ΦE ◦ ΦE = idDb (B) = ΦO∆B ,
A

and hence AdE (O∆A ) = O∆B . Finally, using again Lemma 4.5, we have that
Φ−1
SB (O∆B ) = k(eB ). Thus, we obtain that

ΦJ (E) (k(eA )) ' k(eB ). (4.3)

Step 2. By (4.3), we have the isomorphism

J (E)|{eA }×B×Bb ' k(eB ) ' k(eA , eB ).

In general, J (E) is not a sheaf. But, if we restrict to J 0 := J (E)|U ×B×Bb for some
affine open neighbourhood U := Spec(R) of eA ∈ A × A, b we can obtain a coherent
sheaf, whose support intersect the fibre {eA } × (B × B) at the point {eA } × {eB }.
b
Indeed, if we consider the inclusion i : {eA } × (B × B)
b ,→ (A × A) b × (B × B),b we
m ∗
know that H (i J (E)) = 0 for every m 6= 0. Then
b ∩ supp Hm (J (E)) = ∅
{eA } × (B × B)

for every m 6= 0, and since the support is a closed subset, there exists an (affine) open
neighbourhood of eA (that we have denoted by U ) where the cohomology groups
Hm (J (E)) (for m 6= 0) are zero. Thus, J 0 is quasi-isomorphic (and then isomorphic
in the derived category) to a complex concentrated in zero, that is, a coherent sheaf.
Let now V := Spec(S) be an affine open neighbourhood of eB in B × B. b We
want to show that
p12∗ (J 0 |U ×((B×B)\V
b )
)(eA ) = 0.
To do this, we need Lemma 3.15, which is a result about base change. Indeed, in
this case we have
b J 0|
H 0 (B × B, ) = 0.
{eA }×((B×B)\V
b )
Then we can apply the lemma, and we obtain

R0 p12∗ (J 0 |U ×((B×B)\V
b )
) ⊗ k(eA ) = 0,
4.1. FROM EQUIVALENCE TO ISOMORPHISM 49

so the point eB is not contained in the projection onto A × A b of supp J 0 ∩ (U ×


((B × B)
b \ V )).
Then, up to shrinking U , we can assume that the support of J 0 is contained in
U × V , which is affine. Hence, there exists a coherent sheaf F in U × V such that
j∗ (F) = J 0 , where j : U × V ,→ U × (B × B).b Since F is a coherent sheaf, it is
associated to a finitely generated R ⊗ S-module, that we denote by M . Moreover,
M is a finitely generated R-module since the direct image of the sheaf J 0 = j∗ (F)
under the projection is again a coherent sheaf.

Step 3. Let m be the maximal ideal of R corresponding to the point eA . We know


that F ⊗ k(eA , eB ) ' k(eA , eB ), and we also know that, by definition, k(eA , eB ) =
OU ×V,(eA ,eB ) /m(eA ,eB ) ). So, at the level of modules, we have M ⊗ Rm /m ' Rm /m,
and hence
M ⊗R R/m ' R/m.
Thus, there is a homomorphism ϕ : R → M which becomes an isomorphism if we
tensor it with R/m. Then, if we tensor the exact sequence
ϕ
0 Ker(ϕ) R M Coker(ϕ) 0

we obtain the sequence


'
Ker(ϕ) ⊗ R/m R/m M ⊗ R/m Coker(ϕ) ⊗ R/m,

which implies that Ker(ϕ) ⊗ R/m and Coker(ϕ) ⊗ R/m are both zero. Hence, the
supports of the coherent sheaves Ker(ϕ) and Coker(ϕ) do not contain the point
eA . So, up to shrinking U , we can suppose that U is disjoint from the supports of
Ker(ϕ) and Coker(ϕ). If we go back to associated sheaves, we have the isomorphism
'
e : OU −
ϕ → F, and hence J 0 = j∗ (OU ), that is, J 0 is the structure sheaf of a sub-
scheme X(U ) of U × (B × B). b Then, if π : U × (B × B) b → U is the projection on
the first factor, we obtain that the restriction X(U ) → U is an isomorphism and

J 0 = J (E)|U ×(B×B)
b ' OX(U ) .

Step 4. We now want to prove, for any closed point (a, α) ∈ A × A, b the same
result that occurred in Step 1 for the point eA , that is, the existence of a closed
point (b, β) ∈ B × B
b such that

ΦJ (E) (k(a, α)) ' k(b, β).

By the previous steps, this is true for any closed point in U .


As a general fact, every closed point in an abelian variety X can be written as
a sum of closed points in a (non-empty) open subset V ⊂ X that contains the zero
element of X. Indeed, we can denote by Y the abelian subvariety generated by V ,
and Y is closed. So, the closure of V is contained in Y , and, since V is dense in X,
its closure is exactly X. Thus, Y = X, so V generates the whole variety X.
Then we can write any closed point (a, α) ∈ A × A b as a sum (a1 , α1 ) + (a2 , α2 )
of points belonging to U . We denote by (b1 , β1 ) and (b2 , β2 ) the images of these
50 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

points under the functor ΦJ (E) . In order to calculate ΦJ (E) (k(a, α)), we decompose
the functor using Definition 4.6. By Lemma 4.5, the functor ΦSA sends k(a, α) to
S(a,α) . Denoting by G the object AdE (S(a,α) ) and using Remark 2.30, we have

ΦG ' ΦE ◦ Φ(a,α) ◦ Φ−1


E .

Then functor Φ(a,α) , since it is equal to t∗−a (_) ⊗ Pα , can be written as the compo-
sition Φ(a1 ,α1 ) ◦ Φ(a2 ,α2 ) . Thus we get

ΦG ' ΦE ◦ Φ(a,α) ◦ Φ−1 −1


E ' ΦE ◦ Φ(a1 ,α1 ) ◦ Φ(a2 ,α2 ) ◦ ΦE
' ΦE ◦ Φ(a1 ,α1 ) ◦ Φ−1 −1
E ◦ ΦE ◦ Φ(a2 ,α2 ) ◦ ΦE ' Φ(b1 ,β1 ) ◦ Φ(b2 ,β2 ) ' Φ(b,β) ,

where (b, β) = (b1 , β1 ) + (b2 , β2 ). Hence the object G is isomorphic to S(b,β) , and
finally, applying Φ−1
SB , we obtain that

ΦJ (E) (k(a, α)) ' k(b, β)

as required.

Step 5. From Step 4, for every closed point (a, α) we can repeat the procedure of
Steps 2 and 3 to find a neighbourhood W and a subscheme X(W ) ⊂ W ×(B×B) b such
that the projection X(W ) → W is an isomorphism and J (E)|W ×(B×B) b ' OX(W ) .
Then, we can glue all these neighbourhoods and find a subvariety i : X ,→ (A × A) b ×
(B × B)b such that the projection X → A × A b is an isomorphism. Moreover, J (E) is
locally isomorphic to a structure sheaf, and then it is an invertible sheaf isomorphic
to i∗ L, where L is a line bundle on X. The subvariety X can be seen as the graph of
a function f : A × A b → B × B,b which induces an equivalence of derived categories.
Hence, f has to be an isomorphism.

This theorem implies in particular that if two abelian varieties A and B have
b and B × B
equivalent derived categories of coherent sheaves, then the varieties A× A b
are isomorphic.

Corollary 4.10. The isomorphism fE in Theorem 4.7 maps a k-point (a, α) ∈ A× A


b
to a point (b, β) ∈ B × B
b if and only if the equivalences

Φ(a,α) : Db (A) −→ Db (A) and Φ(b,β) : Db (B) −→ Db (B)

defined in (4.1) satisfy the relation

Φ(b,β) ◦ ΦE ' ΦE ◦ Φ(a,α) ,

or, equivalently, the two points satisfy the following condition on kernels of the trans-
forms:
tb∗ E ⊗ Pβ ' t∗a E ⊗ Pα .

Proof. By Theorem 4.7, we have that ΦJ (E) (k(a, α)) ' k(b, β), where (b, β) =
fE (a, α). Lemma 4.5 implies that ΦSA maps k(a, α) to S(a,α) . Thus, using the
4.1. FROM EQUIVALENCE TO ISOMORPHISM 51

commutative diagram

ΦSA
Db (A × A)
b Db (A × A)
ΦJ (E) AdE

Db (B × B)
b
ΦSB
Db (B × B),

we see that fE maps the point (a, α) to (b, β) if and only if S(b,β) ' AdE (S(a,α) ).
Since S(a,α) and S(b,β) represent the functors Φ(a,α) and Φ(b,β) , using Remark 2.30
we find that Φ(b,β) ' ΦE ◦ Φ(a,α) ◦ Φ−1
E .

In the proposition below we give an explicit formula for the object J (E) in the
case when A = B and the equivalence is Φ(a,α) .

Proposition 4.11. Suppose that A = B, and consider the equivalence Φ(a,α) repre-
sented by the object S(a,α) . Then J (S(a,α) ) is equal to i∗ P(α,−a) , where i : A × A
b→
(A × A)
b × (A × A)b is the diagonal embedding.

Proof. By Lemma 4.5, ΦSA sends k(a0 , α0 ) to the sheaf S(a0 ,α0 ) on A × A. Then, we
apply the functor AdS(a,α) to S(a0 ,α0 ) : this object represents the functor

Φ(a,α) ◦ Φ(a0 ,α0 ) ◦ Φ−1


(a,α) ,

which turns out to be isomorphic to Φ(a0 ,α0 ) , since all such funcotrs commute. Thus,
we have that the functor with Fourier-Mukai kernel J (S(a,α) ) maps k(a0 , α0 ) to itself,
for any point (a0 , α0 ), that is, J (S(a,α) )|{(a0 ,α0 )}×A×A ' k(a0 , α0 ). Hence J (S(a,α) ) is
a line bundle L on A × A supported on the diagonal.
In order to write explicitely the bundle L, we find the image of P(α0 ,a0 ) under
this functor. Using again Lemma 4.5, we have that ΦSA maps P(α0 ,a0 ) to the object
O{−a0 }×A ⊗ p∗2 Pα0 [−n]. Applying AdS(a,α) to this object, we obtain O{−a0 +a}×A ⊗
p∗2 (Pα0 +α )[−n]. Hence, ΦJ (S(a,α) ) maps the bundle P(α0 ,a0 ) to the bundle P(α0 +α,a0 −a) .
Thus, L is isomorphic to P(α,−a) .

Definition 4.12. Let A and B be two abelian varieties. We denote by Eq(A, B)


the set of all exact equivalences between the derived categories of coherent sheaves
Db (A) and Db (B) up to isomorphism.

Let us now consider two groupoids A and D, that is, categories in which all the
morphisms are invertible. We want to characterize these categories. The objects
of both categories are abelian varieties. The morphisms of A are the isomorphism
between abelian varieties as algebraic groups, that is

HomA (A, B) := Iso(A, B),

while the morphisms of D are the exact equivalences between derived categories of
coherent sheaves on abelian vrieties, that is

HomD (A, B) := Eq(A, B).


52 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

By Theorem 4.7, there exists a map from the set Eq(A, B) to the set Iso(A × A,
b B×
B), which sends the equivalence ΦE to the isomorphism fE . So we consider a map
b
F : D → A that sends an abelian variety A to A × A b and works on morphisms as
explained above.

Proposition 4.13. The map F : D → A defined before is a functor.

Proof. We have to check that F preserves compositions of morphisms. So, let A,


B and C be abelian varieties, and let E ∈ Db (A × B) and F ∈ Db (B × C) be the
Fourier-Mukai kernels such that the functors

ΦE : Db (A) → Db (B) and ΦF : Db (B) → Db (C)

are equivalences. Moreover, we denote by G ∈ Db (A × C) the object that represents


their composition. From the relation in Remark 2.26, we obtain that AdG ' Ad F ◦
AdE , whence we have

ΦJ (F ) ◦ ΦJ (E) = (Φ−1 −1
SC ◦ AdF ◦ΦSB ) ◦ (ΦSB ◦ AdE ◦ΦSA )
= Φ−1
SC ◦ AdG ◦ΦSA = ΦJ (G) .

By Theorem 4.7, the objects J (E), J (F) and J (G) are line bundles supported on
the graphs of the isomorphisms fE , fF and fG . Therefore, we get fG = fF ◦ fE .

We now have the functor F : D → A, and we want to describe it. To do this, we


will introduce a particular kind of isomorphisms.
First, let f : A × A
b → B×B b be an arbitrary morphism. We write it as the
matrix  
x y
,
z w
where the entries are the morphisms x : A → B, y : A
b → B, z : A → B b→
b and w : A
b Each morphism f defines two other morphisms from B × B
B. b to A × A:
b
   
w b yb b −b
w y
f :=
b and f :=
e .
zb x
b −bz xb

Definition 4.14. Let f : A × A b → B×B b be an isomorphism between abelian


varieties. The isomorphism f is said to be isometric if its inverse f −1 coincides with
fe. We denote the set of isometric isomorphism with U (A × A, b B × B),b that is,

U (A × A,
b B × B)
b := {f ∈ Iso(A × A, b | f −1 = fe}.
b B × B)

Proposition 4.15. For every equivalence ΦE : Db (A) → Db (B) between abelian


varieties, the isomorphism fE is isometric.

Proof. Passing to the algebraic closure if necessary, we can suppose that the field
k is algebraically closed. In order to verify that fE−1 = ff
E , we only need to check
the equality on closed points. So, let (a, α) be a closed point on A × A, b and let
(b, β) ∈ B × B be the image of (a, α) under fE . We want to prove that
b

ff
E (b, β) = (a, α) or, equivalently, fc
E (−b, β) = (−a, α). (♠)
4.2. FROM ISOMORPHISM TO EQUIVALENCE 53

From the proof of Theorem 4.7, we know that fE is determined by an abelian sub-
variety X ∈ A × Ab × B × B.
b Then (♠) is true if and only if

P(0,0,β,−b) ⊗ OX ' P(α,−a,0,0) ⊗ OX

or, equivalently, the sheaf J 0 := P(−α,a,β,b) is isomorphic to the sheaf J (E). By


Proposition 4.11, the functor represented by J 0 is the composition of the functors
represented by the objects J (S(−a,−α) ), J (E) and J (S(b,β) ). Hence, J 0 coincides
with J (E 0 ), where E 0 is the Fourier-Mukai kernel of the functor

Φ(b,β) ◦ ΦE ◦ Φ(a,α) .

Using Corollary 4.10, we see that this composition is isomorphic to ΦE , and hence
E 0 ' E. Therefore, we have J 0 ' J (E).

At the end, we can gather Theorem 4.7 and Proposition 4.15, and we get

Theorem 4.16. Let A and B be two abelian varieties over a field k. If the de-
rived categories of coherent sheaves Db (A) and Db (B) are equivalent as triangulated
categories, then there exists an isometric isomorphism between A × A b and B × B.b

4.2 From isomorphism to equivalence


In this section we want to prove a converse of Theorem 4.16, that is, the existence
of an isometric isomorphism between two abelian varieties A × A b and B × B b implies
b b
the equivalence of the derived categories D (A) and D (B). Unfortunately, we can
not prove it for every field k. So, in the whole section, we assume that the field k is
algebraically closed and it has characteristic zero.
A different proof of this result can be found in [Pol95].

Let L be a line bundle on an abelian variety A. We recall that L determines a


map φL : A → A, b defined in (3.1), that sends a point a to the point corresponding
to the bundle ta L ⊗ L−1 ∈ Pic0 (A).

Definition 4.17. Let X be a smooth complete variety. We call Néron-Severi group


of X (denoted by NS(X)) the quotient of Pic(X) by its subgroup Pic0 (X). In other
words, we have the exact sequence

1 Pic0 (X) Pic(X) NS(X) 0.

The correspondence between line bundles L and maps φL yields an embedding


of NS(A) in Hom(A, A). b Moreover, we have, as a general fact, that a map ϕ ∈
Hom(A, A) belongs to the image of NS(A) if and only if ϕ
b b = ϕ.
To give a generalization of this phenomenon, we need to introduce shortly semi-
homogeneous vector bundles, defined by Mukai, and some results about them. For
more details and all the proofs, see [Muk78].

Definition 4.18. Let E be a vector bundle on an abelian variety A. E is said to


be homogeneous if t∗a E ' E for every point a ∈ A.
54 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

Definition 4.19. Let F be a vector bundle on an abelian variety A. F is said to


be unipotent if there exists a filtration

0 = F0 ⊂ F1 ⊂ · · · ⊂ Fn−1 ⊂ Fn = F

such that, for every i = 1, . . . , n, the quotient Fi /Fi−1 is isomorphic to OA .

With the following proposition, we characterize homogeneous vector bundles.

Proposition 4.20. Let E be a vector bundle on an abelian variety A. Then the


following conditions are equivalent:

1. E is homogeneous.
0
L line bundles Pi ∈ Pic (A) and some unipotent bundles Fi
2. There exist some
such that E ' i (Pi ⊗ Fi ).

Proof. See [Muk78, § 4].

We now define the objects we are interested in.

Definition 4.21. A vector bundle E on an abelian variety A is said to be semi-


homogeneous if, for every a ∈ A, there exists a line bundle L on A such that
t∗a E ' E ⊗ L.

The following is a useful result to check whether a simple vector bundle is semi
-homogeneous or not. We recall that a vector bundle is simple if its automorphism
algebra coincides with the field k.

Theorem 4.22. Let E be a simple vector bundle on an abelian variety A. Then the
following conditions are equivalent:

1. dimk H 1 (A, EndOA (E)) = n (n is the dimension of A),


 
j n
2. dimk H (A, EndOA (E)) = , for all j = 1, 2, . . . , n,
j
3. E is semi-homogeneous,

4. EndOA (E) is a homogeneous vector bundle,

5. there exists an isogeny π : B → A and a line bundle L on B such that E '


π∗ (L).

Proof. See [Muk78, § 5].

Let E be a vector bundle on an abelian variety A. We denote by µ(E) (called


slope of E) the equivalence class of det(E)
r(E) in NS(A) ⊗Z Q, where det(E) is the
[L]
determinant line bundle of E and r(E) is the rank of E. Every element µ = l ∈
NS(A) ⊗Z Q determines a correspondence Φµ ⊂ A × A b by

(l,φL )
Φµ = Im(A −−−→ A × A),
b (4.4)
4.2. FROM ISOMORPHISM TO EQUIVALENCE 55

where φL is the map (3.1). We have furthermore the two projections q1 : Φµ → A


and q2 : Φµ → A.b In particular, if the bundle is a line bundle L, we get the graph of
φL .
The following result gives a characterization of all semi-homogeneous vector bun-
dles in terms of simple bundles.

Proposition 4.23. Every semi-homogeneous vector bundle F of slope µ admits a


filtration
0 = F0 ⊂ F1 ⊂ · · · ⊂ Fn−1 ⊂ Fn = F

such that Ei = Fi /Fi−1 are simple semi-homogeneous vector bundles of the same
slope µ. Every simple semi-homogeneous bundle is stable.

Proof. See [Muk78, § 6].

Proposition 4.24. Let µ = [L] l , where [L] is the equivalence class of the bundle L
in NS(A) and l is a positive integer. Then:

1. there exists a simplesemi-homogeneous vector bundle E such that µ(E) = µ,

2. every simple semi-homogeneous vector bundle of slope µ is isomorphic to E⊗M


for some line bundle M ∈ Pic0 (A),

3. we have the equations r(E)2 = deg(q1 ) and χ(E)2 = deg(q2 ).

Proof. See [Muk78, § 7].

Moreover, we need a couple of lemmas on semi-homogeneous bundles, which are


consequences of the above results.

Lemma 4.25. Let E1 and E2 be two simple semi-homogeneous bundles on an abelian


variety A of the same slope µ. Then, E1 and E2 are either isomorphic or orthogonal
to each other, that is, either E1 ' E2 or, for every i ∈ Z,

Exti (E1 , E2 ) = 0 and Exti (E2 , E1 ) = 0.

Proof. Since E1 and E2 have the same slope, Proposition 4.24 implies that E2 '
E1 ⊗ M for some M ∈ Pic0 (A). Then, Hom(E1 , E2 ) ' Hom(E1 , E1 ⊗ M ) is a
homogeneous bundle. By Proposition 4.20, a homogeneous bundle is a direct sum of
unipotent bundles twisted by line bundles belonging to Pic0 (A). Thus, either all the
cohomology sheaves of Hom(E1 , E2 ) are zero, and so E1 and E2 are orthogonal, ot
the bundle Hom(E1 , E2 ) has a non-zero section. In the latter case, we get a non-zero
homomorphism from E1 to E2 ' E1 ⊗ M . Since this two bundles are simple and
have the same slope, this homomorphism is actually an isomorphism.

Lemma 4.26. Let E be a simple semi-homogeneous vector bundle on an abelian


variety A with slope µ. Then t∗a E ' E ⊗ Pα if and only if (a, α) ∈ Φµ , where Φµ is
the subset of A × A
b defined in (4.4).
56 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES

Proof. First we prove that t∗a E ' E ⊗ Pα for every point (a, α) ∈ Φµ . Let l =
r(E) and L = det(E). By definition of Φµ , there exists a point x ∈ A such that
(a, α) = (lx, φL (x)). Since E is semi-homogeneous, there exists a line bundle M ∈
Pic0 (A)such that
t∗x (E) ' E ⊗ M.
Comparing the determinants, we obtain that t∗x (L) ' L ⊗ M ⊗l . This means, by
definition of φL , that M ⊗l = PφL (x) . On the other hand, if we iterate the equation
t∗x (E) ' E ⊗ M , we get

t∗lx (E) ' E ⊗ M ⊗l = E ⊗ PφL (x) ,

and then t∗a (E) ' E ⊗ Pα .


Conversely, we define the set

Σ0 (E) := {α ∈ A
b | E ⊗ Pα ' E},

which is a subgroup of A. b Since E is semi-homogeneous, by Theorem 4.22 we have


that End(E) is an homogeneous L bundle, and then, using Proposition 4.20, we can
write End(E) as a direct sum i (Fi ⊗Pi ), where Pi ∈ Pic0 (A) and Fi are unipotent.
We know that, for a unipotent bundle F and a point α ∈ Pic0 (A) different from
j (F ⊗ P ) = 0 for each j ∈ Z. Hence, we have that H 0 (End(E) ⊗ Q) =
L H
zero, α
i H 0 (F i ⊗ P i ⊗ Q) 6= 0 for a line bundle Q ∈ Pic0 (A) only if Q is the dual of
one of the Pi . Denoting by r the rank of E, we have that r(End(E)) = r2 , and
since r(Fi ⊗ Pi ) ≥ 1, we can have at most r2 different line bundles Pi , that is,
H 0 (End(E) ⊗ Q) 6= 0 for at most r2 line bundles Q ∈ Pic0 (A). This means that
the order of the group Σ0 (E) does not exceed r2 . On the other hand, we know that
q2 (Ker(q1 )) ⊂ Σ0 (E), and by Proposition 4.24, deg(q1 ) = r2 . Hence, the order of
Σ0 (E) is exactly r2 .
Suppose that we have a point (a, α) ∈ A × A b such that t∗a (E) ' E ⊗ Pα . We
consider then the element α0 ∈ A b such that (a, α0 ) ∈ Φµ . Thus, we have that

ta E ' E ⊗ Pα0 , that is,
E ⊗ P(α−α0 ) ' E.
This means that (α − α0 ) ∈ Σ0 (E) = q2 (Ker(q1 )). Hence (0, α − α0 ) ∈ Φµ , and so
the point (0, α − α0 ) + (a, α0 ) = (a, α) belongs to Φµ .

Now we are ready to construct a special vector bundle starting from an isometric
isomorphism.
First, we fix an isometric isomorphism f : A × Ab → B × B, b and we denote its
graph with Γf . We write f in matrix form:
 
x y
f= .
z w

b → B is an isogeny. Then, from the map f we can determine


Suppose now that y : A
an element g ∈ Hom(A × B, Ab × B)
b ⊗Z Q by the matrix
 −1
y x −y −1

.
−by −1 wy − 1
4.2. FROM ISOMORPHISM TO EQUIVALENCE 57

The element g determines a correspondence on (A × B) × (A b × B),


b and, since f
is isometric, we have that g = gb. This fact implies that g belongs to the image of
the canonical embedding j : NS(A × B) ⊗Z Q ,→ Hom(A × B, A b × B)
b ⊗Z Q.
Thus, there exists an element µ ∈ NS(A × B) ⊗Z Q such that j(µ) = g, an we can
write µ as [L]
l , where L is a line bundle and l is an integer. The set Φµ defined in (4.4)
coincides with the graph of the correspondence g. Hence, by Proposition 4.24, there
exists a simple semi-homogeneous vector bundle E on A × B such that µ(E) = µ.
This bundle E can be used as the kernel of a Fourier-Mukai transform between Db (A)
and Db (B), and we will prove that this transform is in fact an equivalence.
Before proving that, we want to compare Γf and Φµ . Let (a, α, b, β) be a point
in Γf , then
b = x(a) + y(α) and β = z(a) + w(α).
Using the fact that y is an isogeny, we have that
(
α = y −1 (b) − y −1 x(a)
(♣)
β = z(a) + wy −1 (b) − wy −1 x(a) = (z − wy −1 x)(a) + wy −1 (b).

We know that f is isometric, that is,


    
x y wb −by 1 0
= ,
z w −b z x
b 0 1

and in particular
(
wbx − zb
y=1
and then z − wy −1 x = −b
y −1 .
xby = yb
x,

Thus, the equations in (♣) become


(
α = −y −1 x(a) + y −1 (b)
β = −by −1 (a) + wy −1 (b),

which means that the point (a, α, b, β) belongs to Γf if and only if (a, −α, b, β)
belongs to Φµ .
Hence, we have found that Φµ = (1A , −1Ab, 1B , 1Bb )Γf . In particular, since f is
an isomorphism, the projections of Φµ onto A × Ab and b × B b are also isomorphisms.

Proposition 4.27. Let E be the semi-homogeneous vector bundle on A × B con-


structed above. Then the Fourier-Mukai transform ΦE : Db (A) → Db (B) is an equiv-
alence.

Proof. Let Ea be the restriction of E to {a} × B. To prove that ΦE is fully faithful


(using Theorem 2.34), we must verify that, for any two different points a, a0 ∈ A,

Hom(ΦE (k(a)), ΦE (k(a0 ))[i]) = 0,

that is, Hom(Ea , Ea0 ) = 0 for a 6= a0 , and we have to show that Ea is simple for every
a ∈ A.
58 CHAPTER 4. EQUIVALENCES BETWEEN DERIVED CATEGORIES
p
Step 1. By Proposition 4.24, we have that r(Ea ) = deg(q), where q : Φµ →
A × B is the projection on the first factor. From the above part, we have that
Φµ |A×B = Γf |A×B , so q sends (a, b, α, β) to (a, x(a) + y(α)). Then, two elements
(a, x(a) + y(α)) and (a0 , x(a0 ) + y(α0 )) of the image of q are equal if and only if a = a0
and y(α) = y(α0 ). Hence, deg(q) = deg(y) (note that here y : A b → B is an isogeny
p
but not invertible, because we do not tensor with Q), and then r(Ea ) = deg(y).
Since E is semi-homogeneous, Ea is also semi-homogeneous. Moreover, we know
that the slope µ = µ(E) comes from g, that is, g is the image of µ under the
embedding NS(A × B) ⊗ Q ,→ Hom(A × B, A b × B)
b ⊗ Q. Then, restricting everything
to {a} × B, we have that the slope of Ea is the element ν in NS(B) ⊗ Q whose
image is wy −1 ∈ Hom(B, B) b ⊗ Q. By Proposition 4.24, there exists a simple semi-
homogeneous vector bundle F such that µ(F ) = ν. Then Φν is equal to

(y,w)
b−
Im(A −−→ B × B).
b

(y,w)
Since f is an isomorphism, the map A b −−−→ B × B b is an embedding, and then
the projection on B of Φν correspondsp to the map y : A b → B. Hence, using
again Proposition 4.24, we get r(F ) = deg(y) = r(Ea ), that is, F and Ea are
two semi-homogeneous bundles with the same rank and slope, and moreover F is
simple.
By Proposition 4.23, the bundle Ea has a filtration F0 = 0 ⊂ F1 ⊂ · · · ⊂ Fn = Ea
of vector bundles such that Fi /Fi−1 are simple semi-homogeneous bundles of slope
ν. Suppose that Ea is not simple. Then Ea strictly contains F1 , because F1 /F0 = F1
is simple. By Proposition 4.24, there exists a bundle M ∈ Pic0 (A) such that F1 =
F ⊗ M , and so r(F ) ≤ r(F1 ) ≤ r(Ea ), and then r(F1 ) = r(Ea ). But this means that
F1 = Ea , which is a contradiction since Ea is not simple. Thus, we have that Ea is a
simple bundle.

Step 2. We can now apply Lemma 4.25, which says that, for any two points
a1 6= a2 in A, the bundles Ea1 and Ea2 are either isomorphic or orthogonal. Suppose
that they are isomorphic. Since E is semi-homogeneous, we have

t∗(a1 −a2 ,0) E ' E ⊗ P(α,β) (4.5)

for some (α, β) ∈ A


b × B.
b If we restrict to {a2 } × B, we obtain

Ea1 ' Ea2 ⊗ Pβ

and then
Ea2 ⊗ Pβ ' Ea2 .
This means that β ∈ Σ0 (Ea2 ). From Lemma 4.26, we know that the orders of Σ0 (Ea2 )
and Σ0 (E) are equal, and we have the group homomorphism σ : Σ0 (E) → Σ0 (Ea2 ),
which is the restriction. So, the point (0, 0) is mapped into 0. Suppose that there
exists another point (α0 , 0) ∈ Σ0 (E) such that α0 6= 0 and σ(α0 , 0) = 0. Then
E ⊗ P(α0 ,0) ' E. Hence, by Lemma 4.26, we have that (0, α0 , 0, 0) ∈ Φµ , but this
is impossible, because f (0, −α0 ) 6= (0, 0), that is, (0, −α0 , 0, 0) ∈
/ Γf . Thus, σ is an
isomorphism.
4.2. FROM ISOMORPHISM TO EQUIVALENCE 59

This means that, since β ∈ Σ0 (Ea2 ), there exists a point α̃ ∈ A


b such that
0
(α̃, β) ∈ Σ (E), and then
E ' E ⊗ P(α̃,β) .
We can now rewrite (4.5) in the form

t∗(a1 −a2 ,0) E ' E ⊗ P(α̃,β) ⊗ P(α−α̃,0) ' E ⊗ P(α−α̃,0) ,

and then (a1 − a2 , α − α̃, 0, 0) ∈ Φµ . So, by the fact that Φµ = (1A , −1Ab, 1B , 1Bb )Γf ,
we have that a1 = a2 , that is, Ea1 = Ea2 .
Therefore, Ea1 and Ea2 are orthogonal for a1 6= a2 , and hence ΦE is a fully faithful
functor. Applying the same arguments to ΦE ∨ , we see that it is also fully faithful.
Hence, ΦE is an equivalence.

Proposition 4.28. Let E be the semi-homogeneous vector bundle on A × B con-


structed above, starting from an isometric isomorphism f . Then fE = f .

Proof. We denote by X the graph of the morphism fE . From Corollary 4.10, it


follows that a point (a, α, b, β) belongs to X if and only if

tb∗ E ⊗ Pβ ' t∗a E ⊗ Pα ,

which is equivalent to the equation

t∗(a,b)E ' E ⊗ P(−α,β) .

Hence, by Lemma 4.26, we have that X = (1A , −1Ab, 1B , 1Bb )Φµ , where µ is the slope
of E. On the other hand, the graph Γf is also equal to (1A , −1Ab, 1B , 1Bb )Φµ , and
hencethe isomorphisms fE and f coincide.

Remark 4.29. In the construction of E from f , we have assumed that the map
y: Ab → B is an isogeny. If this is not the case, we can write f as the composition
of two maps f1 ∈ U (A × A, b B × B)b and f2 ∈ U (A × A) b such that y1 and y2 are
isogenies. Then, for every i = 1, 2, we can construct the bundle Ei , and we take the
object that represent the composition of the functors ΦEi .
The results proved in this chapter can finally be combined into the following
theorem.

Theorem 4.30. Let A and B be two abelian varieties over an algebraically closed
field of characteristic zero. Then the bounded derived categories of coherent sheaves
Db (A) and Db (B) are equivalent as triangulated categories if and only if there exists
an isometric isomorphism f : A × A b → B × B.
b
Bibliography

[GM07] Gerard van der Geer and Ben Moonen. Abelian varieties. 2007. url:
https://www.math.ru.nl/~bmoonen/research.html.
[GM13] Sergei I. Gelfand and Yuri I. Manin. Methods of homological algebra.
Springer Science & Business Media, 2013.
[Har13] Robin Hartshorne. Algebraic geometry. Vol. 52. Springer Science & Busi-
ness Media, 2013.
[Huy06] Daniel Huybrechts. Fourier-Mukai transforms in algebraic geometry. Ox-
ford University Press on Demand, 2006.
[Mil08] James S. Milne. Abelian varieties (v2. 00). 2008. url: http : / / www .
jmilne.org/math.
[Muk78] Shigeru Mukai. “Semi-homogeneous vector bundles on an abelian vari-
ety”. In: Journal of Mathematics of Kyoto University 18.2 (1978), pp. 239–
272.
[Muk81] Shigeru Mukai. “Duality between D(X) and D(X̂) with its application
to Picard sheaves”. In: Nagoya Mathematical Journal 81 (1981), pp. 153–
175.
[Mum70] David Mumford. “Abelian varieties”. In: Studies in mathematics, Tata
Institute of Fundamental Research 5 (1970).
[Orl02] Dmitri O. Orlov. “Derived categories of coherent sheaves on abelian va-
rieties and equivalences between them”. In: Izvestiya: Mathematics 66.3
(2002), p. 569.
[Orl03] Dmitri O. Orlov. “Derived categories of coherent sheaves and equivalences
between them”. In: Russian Mathematical Surveys 58.3 (2003), p. 511.
[Pol03] Alexander Polishchuk. Abelian varieties, theta functions and the Fourier
transform. Vol. 153. Cambridge University Press, 2003.
[Pol95] Alexander Polishchuk. “Symplectic biextensions and a generalization of
the Fourier-Mukai transform”. In: arXiv preprint alg-geom/9511018 (1995).
[Tho00] Richard P. Thomas. “Derived categories for the working mathematician”.
In: arXiv preprint math/0001045 (2000).

61

You might also like