Introduction to algebraic surfaces

Lecture Notes for the course at the

University of Mainz

Wintersemester 2009/2010

Arvid Perego
(preliminary draft)

October 30, 2009

2
Contents

Introduction 5

1 Background material 9
1.1 Complex and projective manifolds . . . . . . . . . . . . . . . . . 9
1.1.1 Complex manifolds . . . . . . . . . . . . . . . . . . . . . . 10
1.1.2 The algebraic world . . . . . . . . . . . . . . . . . . . . . 13
1.2 Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Sheaves and cohomology . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.1 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.2 Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3.3 Cohomology of coherent sheaves . . . . . . . . . . . . . . 29
1.3.4 The GAGA Theorem . . . . . . . . . . . . . . . . . . . . . 33
1.4 Hodge theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.4.1 Harmonic forms . . . . . . . . . . . . . . . . . . . . . . . 35
1.4.2 Kähler manifolds and Hodge decomposition . . . . . . . . 38

2 Line bundles 41
2.1 The Picard group . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2 Cartier and Weil divisors . . . . . . . . . . . . . . . . . . . . . . 45
2.2.1 Cartier divisors . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.2 Weil divisors . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 Ampleness and very ampleness . . . . . . . . . . . . . . . . . . . 52
2.4 Intersection theory on surfaces . . . . . . . . . . . . . . . . . . . 56
2.4.1 The Riemann-Roch Theorem for surfaces . . . . . . . . . 61
2.5 Nef line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5.1 Cup product and intersection product . . . . . . . . . . . 64
2.5.2 The Néron-Severi group . . . . . . . . . . . . . . . . . . . 65
2.5.3 The Nakai-Moishezon Criterion for ampleness . . . . . . . 69
2.5.4 Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3 Birational geometry 79
3.1 Contraction of curves . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.1.1 Rational and birational maps . . . . . . . . . . . . . . . . 79
3.1.2 The Zariski Main Theorem . . . . . . . . . . . . . . . . . 81
3.1.3 Blow-up of a point . . . . . . . . . . . . . . . . . . . . . . 84
3.1.4 Indeterminacies . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.5 Castelnuovo’s contraction theorem . . . . . . . . . . . . . 89

3
4 Contents

3.2 Kodaira dimension . . . . . . . . . . . . . . . . . . . . . . . . . . 93

3.2.1 Definition and geometrical interpretation . . . . . . . . . 94
3.2.2 Kodaira dimension and nefness . . . . . . . . . . . . . . . 100
3.3 Albanese tori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4 The Enriques-Castelnuovo Classification 107

4.1 The Kawamata-Mori Rationality Theorem . . . . . . . . . . . . . 109
4.2 Kodaira dimension −∞ . . . . . . . . . . . . . . . . . . . . . . . 111
4.2.1 Ruled surfaces . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2.2 Invariants of geometrically ruled surfaces . . . . . . . . . 114
4.2.3 Rational surfaces . . . . . . . . . . . . . . . . . . . . . . . 117
4.2.4 Minimal ruled surfaces . . . . . . . . . . . . . . . . . . . . 118
4.2.5 Classification for Kodaira dimension −∞ . . . . . . . . . 122
4.2.6 The Castelnuovo Rationality Criterion . . . . . . . . . . . 125
4.3 Non-negative Kodaira dimension . . . . . . . . . . . . . . . . . . 127
4.3.1 Definitions and examples . . . . . . . . . . . . . . . . . . 127
4.3.2 Fibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.3.3 Elliptic fibrations . . . . . . . . . . . . . . . . . . . . . . . 136
4.3.4 Classification for Kodaira dimension 0 and 1 . . . . . . . 139

5 Further properties and classifications 147

5.1 Ruled surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.1.1 Classification of non-rational ruled surfaces . . . . . . . . 148
5.1.2 Ruled and bielliptic surfaces . . . . . . . . . . . . . . . . . 151
5.2 Rational surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.1 Linear systems of conics . . . . . . . . . . . . . . . . . . . 154
5.2.2 Linear systems of cubics . . . . . . . . . . . . . . . . . . . 156
5.3 K3 surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.3.1 The Kummer surface . . . . . . . . . . . . . . . . . . . . . 161
5.3.2 Projectivity . . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.3.3 The Torelli Theorem for K3 surfaces . . . . . . . . . . . . 167
5.4 Enriques surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.4.1 Construction and projectivity . . . . . . . . . . . . . . . . 170
5.4.2 The Torelli Theorem for Enriques surfaces . . . . . . . . . 171
5.5 Bielliptic surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.6 Abelian surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.6.1 Properties of abelian surfaces . . . . . . . . . . . . . . . . 177
5.6.2 The Torelli Theorem for abelian surfaces . . . . . . . . . . 181
5.7 Elliptic surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.7.1 Classification of singular fibers . . . . . . . . . . . . . . . 183
5.7.2 Classification of elliptic surfaces . . . . . . . . . . . . . . . 186
5.8 Surfaces of general type . . . . . . . . . . . . . . . . . . . . . . . 194

6 Generalizations of the Enriques-Castelnuovo Classification 199

6.1 Singular projective surfaces . . . . . . . . . . . . . . . . . . . . . 199
6.2 The Enriques-Kodaira Classification . . . . . . . . . . . . . . . . 200
6.3 Characteristic p . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.4 Higher dimensional classification . . . . . . . . . . . . . . . . . . 207
Introduction

In this series of lectures I will present an introduction to the classification of

compact complex surfaces. This is a quite old subject in geometry, and involves
several elements coming from different areas of mathematics, like differential
geometry, topology, algebraic geometry (classical and modern). It is, to my
opinion, a very beautiful piece of mathematics, which is nowadays considered
classical, and which is very useful to modern research in geometry.
As usual when dealing with a geometric problem, one needs to define the
family of objects one wants to investigate, and to define a notion of equivalence
between them. Moreover, in order to understand if two objects of the chosen
family are equivalent, one studies their invariants. If one considers as objects
topological spaces, i. e. couples (X, τ ) where X is a set and τ is a topology on it,
then one can talk about continuous maps between two topological spaces, and
can introduce the notion of homeomorphism: namely, two topological spaces
(X1 , τ1 ) and (X2 , τ2 ) are homeomorphic if there are two continuous maps f1 :
X1 −→ X2 and f2 : X2 −→ X1 such that f2 ◦ f1 = idX1 and f1 ◦ f2 = idX2 . One
can then try to classify topological spaces up to homeomorphism, even if this
problem seems somehow impossible to solve, as every set can be considered as
a topological space (e.g. with the discrete topology). One can even change the
notion of equivalence, introducing, as instance, the homotopic equivalence: first
of all, one defines the notion of homotopy between two maps f, g : X −→ Y ;
then one says that two topological spaces (X1 , τ1 ) and (X2 , τ2 ) are homotopic
equivalent if there are two continuous maps f1 : X1 −→ X2 and f2 : X2 −→ X1
such that f2 ◦ f1 is homotopic to idX1 and f1 ◦ f2 is homotopic to idX2 . Still,
the problem seems very hard to solve. For simplicity’s sake, we will consider
only compact and connected topological spaces.
The next step in the classification is then to enrich the structure of a topolig-
ical space. This can be done in different ways: first of all, one only considers
topological spaces over which one can, in some way, perform analysis or alge-
bra. A first approach is to define the notion of a manifold : this can be done
for different purposes. A real manifold will be a topological space M which
is Hausdorff, admits a countable basis for the topology, and admits an open
cover {Ui }i∈I , where Ui is an open subset of M and for every i ∈ I there is a
homeomorphism ϕi : Ui −→ Rn for some fixed n (here Rn is a topological space
with the euclidean topology). This n is called the (topological) real dimension
of M . If one considers i, j ∈ I, then we have a map

ϕi ◦ ϕ−1 n n
j : ϕj (Ui ∩ Uj ) ⊆ R −→ R .

Asking for these maps to be differentiable, of class C k , of class C ∞ , ana-

5
6 Introduction

lytic, one gets the notion of differentiable manifold, C k −differentiable manifold,

C ∞ −differentiable manifold, analytic manifold. Moreover, one can introduce the
notion of continuous map, C k −, C ∞ −differentiable map or analytic map be-
tween two manifolds M1 and M2 , and hence of equivalence: one gets the notions
of homeomorphic manifolds and of diffeomorphic ( C k −, C ∞ −)differentiable or
analytic manifolds.
In a similar way one can introduce the notion of complex manifold, which
is a topological space X which is Hausdorff, admits a countable basis for the
topology, and admits an open cover {Ui }i∈I , where Ui is an open subset of X
and for every i ∈ I there is a homeomorphism ϕi : Ui −→ Cn for some fixed n
(here Cn is a topological space with the euclidean topology). This n is called
the (topological) complex dimension of X. Again, for every i, j ∈ I we have a
map
ϕi ◦ ϕ−1 n
j : ϕj (Ui ∩ Uj ) ⊆ C −→ C .
n

Asking for these maps to be holomorphic, one gets the notion of holomorphic
manifold, or simply of complex manifold. As in the previous case, one can
introduce the notion of holomorphic map between two complex manifolds, and
hence of equivalence: two complex manifolds X1 and X2 are equivalent if they
are biholomorphic.
Using the notion of sheaf, one can still enrich the structure of a manifold,
attaching to it a sheaf of rings: if M is a differentiable or analytic manifold, and
U ⊆ M is an open subset, then one can define the ring

OM (U ) := {f : U −→ R | f is differentiable/analytic}.

This assignement defines a sheaf of rings over M , and the couple (M, OM ) is
called a differentiable or analytic space. In a similar way one defines the notion
of a complex space (X, OX ).
There is a different approach, relying on algebra instead of analysis. To
any commutative ring A with unity 1A , we can associate a topological space
Spec(A), whose underlying set is the set of prime ideals of A, and whose topology
is defined using ideals of A, and is called Zariski topology. Moreover, using
localizations of A, one can define a sheaf of rings OSpec(A) , and the couple
(Spec(A), OSpec(A) ) is called an affine scheme. More generally one can define
the notion of a scheme: this is simply a couple (X, OX ), where X is a topological
space, OX is a sheaf of commutative rings with unity, and admitting an open
cover by affine schemes. Again, one can define the notion of morphism between
two schemes, and hence of equivalence: two schemes are equivalent if they are
isomorphic. As for manifolds, one can define the notion of (algebraic) dimension
of a scheme, relying on the notion of Krull dimension of a commutative ring.
In any case, the notion of dimension is crucial: if two spaces (or schemes) are
equivalent, they have the same dimension. This is a first important invariant,
and we can classify these objects at first with respect to their dimension. In
dimension 0, there is only the point, so there is no classification needed. In di-
mension 1, things are more complicated: in the complex universe, these objects
are called Riemann surfaces. These are, in particular, compact connected real
manifolds of dimension 2, whose classification started in the second half of the
XIX century by Moebius, Klein and others. Anyway, they contain more struc-
ture, namely the complex structure: using the theory of holomorphic functions,
Riemann was able to give a classification of these objects by means of the genus:
Introduction 7

roughly speaking, we can divide them in rational curves (i. e., genus 0 Riemann
surfaces), elliptic curves (i. e. genus 1 Riemann surfaces) and curves of general
type (i. e. Riemann surfaces with genus at least 2).
There are some crucial facts in the theory of Riemann surfaces: first of
all, they are all projective, i. e. any Riemann surface can be realized as a
curve in some projective space; second, classification of Riemann surfaces is
with respect to biholomorphism, i. e. two Riemann surfaces are equivalent if
they are homeomorphic as topological spaces and if they carry biholomorphic
complex structures.
Once the classification of curves was achieved, it was natural to try to classify
2−dimensional objects. Already at the end of the 1850’s there were important
studies on low degree surfaces in the 3−dimensional projective space by peo-
ple like Cayley, Kummer and Steiner. During the 1870’s the first generation
of italian algebraic geometers (Bertini, Cremona, del Pezzo, Segre, Veronese)
started to look at surfaces embedded in higher dimensional projective spaces;
independently, Noether was able to show some important properties, like what
is now known as the Noether formula, even if the proof was not complete. The
proof was corrected by Enriques, one of the major exponents of the second gen-
eration of algebraic geometers, which together with Castelnuovo was finally able
to produce at the end of the 1910’s a classification of smooth projective complex
surfaces.
Anyway, their classification was lacking of background material, which was
yet to be developed. Anyway, they pointed out two of the main problems of
the classification of compact complex surfaces. As told before, classification of
Riemann surfaces is a biholomorphic classification of smooth projective com-
plex curves. For surfaces, this is not the case: first of all, there are compact
complex surfaces which are not projective. Second, it is impossible to carry out
a biholomorphic classification of surfaces, even only for the projective ones: this
was pointed out by Enriques and Castelnuovo, who realized the existence of
an importan phenomenon, nowadays known as blow-up. They then introduced
the notion of rational morphism, and they were able to perform (even lacking
of the foundations of algebraic geometry) a birational classification of smooth
projective complex surfaces.
The first attempt to formulate in a clearer way the results obtained by
Enriques and Castelnuovo were made by Zariski, Weil and van der Waerden
in the classical projective geometry, and by Hodge, de Rham and Lefschetz
on the analytic side (during the 1940’s). The main techniques in algebraic
geometry were carried out by Serre, Grothendieck and Hirzebruch in the 1950’s
and 1960’s, who introduced the notion of sheaf, of cohomology, of scheme and
their properties.
Using these tools, Kodaira was able to complete the classification of compact
complex surfaces, including even singular and non-projective ones. The obtained
classification goes under the name of Enriques-Kodaira birational classification
of compact complex surfaces, and which is one of the most important results
in the whole complex geometry. During the 1970’s, the work of Kodaira was
generalized to compact surfaces defined over the an algebraically closed field of
characteristic p, by Bombieri and Mumford: these classification is the same as
the Kodaira-Enriques one if p > 3, but presents some important differencies if
p = 2, 3. The classification of smooth projective complex surfaces as presented
by Kodaira is obtained in the following way: one needs to attach to any surface
8 Introduction

some invariants, which Enriques and Castelnuovo called plurigenera, and which
nowadays goes under the name of Kodaira dimension. Roughly speaking, then
count the number of independent differential forms one has on a surface, and
we have 4 possible values of it: −∞, 0, 1 and 2. This invariant gives a first
classification of surfaces, which can be completed as follows:

1. Kodaira dimension −∞:

• rational surfaces (birational to P2 );
• ruled surfaces (projectivized of a rank 2 vector bundle on a smooth
projective curve of genus g ≥ 1);

2. Kodaira dimension 0:
• abelian surfaces (projective quotients of C2 by a maximal rank lat-
tice);
• K3 surfaces;
• Enriques surfaces;
• bielliptic surfaces;
3. Kodaira dimension 1: proper elliptic surfaces;
4. Kodaira dimension 2: surfaces of general type.

As one can see easily from this picture, the birational classification of surfaces
is quite coarse, in the same spirit of the classification of Riemann surfaces: using
Kodaira dimension, the classification for Riemann surfaces is:

1. Kodaira dimension −∞: rational curves (P1 );

2. Kodaira dimension 0: elliptic curves (quotient of C by a maximal rank

lattice);
3. Kodaira dimension 1: curves of general type.

In this sense, the classification of surfaces as given by Kodaira and Enriques

is not complete: as for curves, the family of curves of general type is very coarse,
the same can be said for surfaces of general type, and up to now there is no
good understanding of this huge family.
Let me conclude this introduction by some mentions about MMP. Starting
from the end of the 1970’s and the beginning of the 1980’s, there were some
attempts to classify higher dimensional smooth projective complex varieties,
imitating the process used to classify surfaces. The obtained results incouraged
the development of what is nowaday known as Minimal Model Program, or sim-
ply MMP. This generalization, which is still to be completed, introduced new
techniques to adapt the methods of the classification of surfaces to higher di-
mensional varieties. Using notions like nefness and cones, one is able to simplify
the proof of the classification of surfaces. In the following lectures I will try to
present this more modern approach to the theory of surfaces.
Chapter 1

Background material

This first chapter is dedicated to basic facts that will be needed all along these
lectures. First, I will recall the notions of complex manifold, of analytic vari-
ety and of projective variety, linking these two notions via the Chow Theorem.
Then, I will briefly recall the definition and important properties of vector bun-
dles on complex manifolds, giving useful examples.
I will continue with a brief recall about sheaves and their cohomology: this
will be done in axiomatic terms and by introducing the notion of Cech coho-
mology, but will be enough to define the notions of singular cohomology and
the de Rham and the Dolbeault Theorems. I will conclude with the statement
and corollaries of the GAGA Theorem, following Serre.
The next topic will be Hodge theory: I will recall the notion of hermitian
metric and of complex structure, in order to state the Hodge Theorem. I will
introduce the notion of Kähler manifold and of Hodge decomposition. Main
results will concern Hodge numbers, projectivity of Riemann surfaces and ex-
amples on non-Kähler surfaces, like the Hopf surface.

1.1 Complex and projective manifolds

In this first section I will recall two basic notions: complex manifolds and basic
properties, and projective manifolds. I will present important examples that
will be useful all along the lectures.
Before going on with definitions, let me recall some notations in complex
analysis that I will use in the following. If n ∈ N, I will use z = {z1 , ..., zn } for a
coordinate system on the complex affine space Cn . For every j = 1, ..., n, I will
write zj = xj + iyj , the usual decomposition of the complex coordinate zj into
real and imaginary part. For every j = 1, ..., n we can define the two differential
operator
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
:= −i , := +i .
∂zj 2 ∂xj ∂yj ∂z j 2 ∂xj ∂yj

In a similar way, one can define

dzj := dxj + idyj , dz j = dxj − idyj .

9
10 Background material

Using this definitions, if U ⊆ Cn is an open subset (in the euclidean topology)

and f : U −→ C is a differentiable function, one has
n n
X ∂f X ∂f
df = dzj + dz j =: ∂f + ∂f.
j=1
∂zj j=1
∂z j

Recall now the following definition: let U ⊆ Cn be an open subset and f :

U −→ C a C ∞ −function, that we can decompose in real and imaginary part as
f = u + iv, where u, v : U −→ R are two C ∞ −functions.

Definition 1.1.1. The function f is holomorphic on U if one of the following

(equivalent) conditions is verified:

1. Cauchy-Riemann condition: for every j = 1, ..., n the two following equa-

tions are satisfied on U :
∂u ∂v ∂u ∂v
= , =− .
∂xj ∂yj ∂xj ∂yj

2. Antiholomorphic condition: on U we have ∂f = 0.

3. Analytic condition: there is a power series which absolutely converges to

f on U .
If f = (f1 , ..., fm ) : U −→ Cm , then f is holomorphic if and only if fi is
holomorphic for every i = 1, ..., m. Notice that one provide the following matrix,
called the Jacobian of f :

J(f ) := [∂fi /∂zj ]i=1,...,m; j=1,...,n .

This is a n × m matrix whose enters are holomorphic functions over U . If n = m

and z0 ∈ U , we say that J(f ) is non-singular at z0 if det(J(f )(z0 )) 6= 0.

1.1.1 Complex manifolds

Let us start with the following basic definition:

Definition 1.1.2. An n−dimensional complex manifold is a topological space

M such that

1. M is Hausdorff;

2. M admits a countable basis for its topology;

3. there is a family U = {Ui , ϕi }i∈I , called atlas, where {Ui }i∈I is an open
cover of M , for every i ∈ I we have ϕi : Ui −→ Cn is a homeomorphism
and for every i, j ∈ I such that Uij := Ui ∩ Uj 6= ∅ the map

ϕji := ϕ−1
j ◦ ϕi : ϕj (Uij ) −→ ϕi (Uij )

is holomorphic.
Complex and projective manifolds 11

Definition 1.1.3. Let M be an n−dimensional complex manifold with atlas

U = {Ui , ϕi }i∈I , and let U ⊆ M be an open subset of M . A continuous map
f : U −→ C is called holomorphic on U if for every i ∈ I such that Ui ∩ U 6= ∅
the map
f ◦ ϕ−1
i : ϕi (Ui ∩ U ) −→ C

is holomorphic on ϕi (Ui ∩U ). A map f = (f1 , ..., fm ) : U −→ Cm is holomorphic

on U if and only if every fi : U −→ C is holomorphic on U . More generally, if
M is an n−dimensional complex manifold with atlas U = {Ui , ϕi }i∈I and N is
an m−dimensional complex manifold with atlas V = {Vj , ψj }j∈J , a continuous
map f : M −→ N is homolorphic if for every i ∈ I and for every j ∈ J such
that Vj ∩ f (Ui ) 6= ∅, the map

ψj ◦ f ◦ ϕi : Ui −→ ψ(Vj ∩ f (Ui )) ⊆ Cm

is holomorphic.

One can generalize the notion of Jacobian matrix for any holomorphic map
between two complex manifolds.
Example 1.1.1. A first trivial example of n−dimensional complex manifold is
simply any open subset U of Cn . This is an example of non-compact complex
manifold.
Example 1.1.2. Let n be any integer, and let us define PnC := Cn+1 / ', where
z = (z0 , ..., zn ) and w = (w0 , ..., wn ) are equivalent under ' if and only if there
is λ ∈ C∗ such that λz = w. Let the topology on PnC be the quotient topology.
This is the n−dimensional complex projective space, and its points are denoted
as z = (z0 : ... : zn ). It is easy to see that it is an n−dimensional complex
manifold: for every i = 0, ..., n, let

Ui := {z = (z0 : ... : zn ) | zi 6= 0}.

Sn
It is easy to see that Ui is an open subset of PnC , and that PnC = i=0 Ui .
Moreover, the map
 
z0 zi−1 zi+1 zn
ϕi : Ui −→ Cn , ϕi (z0 : ... : zn ) := , ..., , , ...,
zi zi zi zi

is clearly a homeomorphism and holomorphic. This is a very important example

of compact complex manifold.
Example 1.1.3. Let γi , ..., γ2n ∈ Cn be 2n R−linearily independent elements, and
let Γ := Zγ1 ⊕...⊕Zγ2n ⊆ Cn be the lattice generated by them. Let TΓ := Cn /Γ,
and let its topology be the quotient topology. This is the n−dimensional complex
torus, and it is not difficult to see that this is a compact n−dimensional complex
manifold.
12 Background material

Example 1.1.4. Let ∼ be the equivalence relation on Cn generated by the

homothety sending z to 2z. The quotient topological space Cn / ∼ is the
n−dimensional complex Hopf manifold. It is not hard to see that this is a
compact complex manifold.
Example 1.1.5. Any 1−dimensional compact connected complex manifold is
called Riemann surface. A 2−dimensional compact connected complex manifold
is called complex surface.

Definition 1.1.4. Let M be a complex manifold with atlas U = {Ui , ϕi }i∈I ,

and let m ∈ M . The holomorphic tangent space at M in m is the complex
h
vector space Tm (M ) spanned by ∂/∂z1 , ..., ∂/∂zn , where z = (z1 , ..., zn ) are
local coordinate at m. The antiholomorphic tangent space at M in m is the
a
complex vector Tm (M ) space spanned by ∂/∂z 1 , ..., ∂/∂z n . The complex tangent
space at M in m is
h a
TC,m (M ) := Tm (M ) ⊕ Tm (M ).

The (holomorphic, antiholomorphic, complex ) cotangent space at M in m is the

dual of the (holomorphic, antiholomorphic, complex) tangent space.

Proposition 1.1.1. Any complex manifold has a natural orientation.

Proof. Let U be an open subset of Cn , and let f = (f1 , ..., fn ) be a holomorphic

map. Let fj = uj + ivj for every j = 1, ..., n, and let fR = (u, v) : U −→
R2n : this is a differentiable map, and one can calculate its jacobian J(fR ),
which is a 2n × 2n matrix whose entries are ∂uj /∂xk and ∂vj /∂yk . As f is
holomorphic, it is easy to see that det(J(fR )) = |det(J(f ))|2 : if f is invertible,
then det(J(fR )) > 0. Now, if one considers a complex manifold M equipped
with an atlas U = {Ui , ϕi }i∈I , for every i, j ∈ i such that Uij 6= ∅ one has the
map ϕji which is holomorphic. Then J(ϕij ) has positive determinant, implying
that M has a natural orientation.

Let M be an n−dimensional complex manifold, m < n and let f : M −→ Cm

be a holomorphic function. We define

V (f ) := f −1 (0),

and we call it the zero locus of f . Using the Inverse Function and the Implicit
Function Theorems, one can easily show that if the rank of J(f ) is always equal
to m, then V (f ) is an (n − m)−dimensional complex manifold. This can be
generalized in the following:

Definition 1.1.5. Let M be an n−dimensional complex manifold with atlas

U = {Ui , ϕi }i∈I , and let N ⊆ M be a closed subset. Then N is called an
analytic subset of M if for every i ∈ I such that Ni := N ∩ Ui 6= ∅ there is a
holomorphic map f : Ui −→ Cm such that Ni = V (fi ).
Complex and projective manifolds 13

If N is irreducible (i. e. it is not union of two or more proper analytic

subsets of M ), then N is called analytic subvariety of M .
If m < n is constant and the rank of J(fi ) is m at every point, then N is
called a complex submanifold (or smooth or non-singular subvariety) of M , of
dimension n − m.
If M = PnC , all these are called projective.

Example 1.1.6. It is an important fact that every Riemann surface is projective.

This is not the case for complex surfaces: the Hopf surface is not projective.
We will see these in the following, as a consequence of Hodge theory.

1.1.2 The algebraic world

One can specialize the holomorphic functions allowed.

Definition 1.1.6. Let U = Cn+1 , and f = (f1 , ..., fm ) : U −→ Cm be such

that fi ∈ C[z0 , ..., zn ], i. e. fi is a polynomial for every i = 1, ..., m. Then its
zero locus V (f ) ⊆ Cn+1 is called affine algebraic subset of Cn+1 .

As polynomials are holomorphic functions, any affine algebraic subset of

n+1
C is an analytic subset. A particular property of affine algebraic subsets of
Cn+1 is the following: they do not depend on f1 , ..., fm , but only on the ideal
I = (f1 , ..., fm ) spanned by f1 , ..., fm in the commutative ring C[z0 , ..., zn ]. For
this reason, one writes V (I) := V (f ).

Definition 1.1.7. If V (I) is irreducible, i. e. it is not the union of two or more

proper affine subsets of Cn+1 , then it is called affine variety.

Example 1.1.7. It is not difficult to see that if I is a prime ideal, then V (I) is
an affine variety.

Definition 1.1.8. Let V ⊆ Cn+1 be an affine algebraic variety. A rational

map f on V is a map f = (f1 , ..., fm ) such that for every i = 1, ..., m there
are pi , qi ∈ C[z0 , ..., zn ] such that fi = pi /qi , with qi is not identically trivial on
V . If qi is everywhere non trivial on V for every i = 1, ..., m, then f is called
regular. A rational (resp. regular ) map between two affine algebraic varieties
V ⊆ Cn and W ⊆ Cm is simply a rational (resp. regular) map on V taking
values in W .

If f1 , ..., fm are homogenous polynomials in z0 , ..., zn , then V (f ) is closed

under multiplication by every element of C∗ : one can then consider its projection
to PnC , i. e. the subset of those z = (z0 : ... : zn ) ∈ PnC such that fi (z) = 0
for every i = 1, ..., m. This is clearly a closed subset V (f ) of PnC , which clearly
depends only on the homogenous ideal I spanned by f1 , ..., fm , and it is hence
denoted V (I). It is clear that V (I) is a projective analytic set.
14 Background material

Definition 1.1.9. Any such a projective analytic set is called projective alge-
braic set. If V (I) is irreducible, i. e. it is not union of two or more projective
algebraic sets, it is called projective algebraic variety.

Remark 1.1.1. Any projective algebraic variety is a projective analytic variety.

Definition 1.1.10. A projective algebraic variety is called projective algebraic

manifold if it is a complex manifold.

Example 1.1.8. The n−dimensional complex projective space is a projective

algebraic variety. If F is a homogenous polynomial of degree d in z0 , ..., zn , then
V (F ) := V ((F )) is a projective algebraic variety called hypersurface of degree d
in PnC .

Definition 1.1.11. Let V be a projective algebraic variety. A rational map f

on V is a map f = (f1 , ..., fm ) such that for every i = 1, ..., m there are pi , qi
homogenous polynomials of the same degree such that qi is not identically trivial
on V and fi = pi /qi . If for every i = 1, ..., m the homogenous polynomial is
everywhere non trivial, then f is called regular.

Notice that any rational map f = (p1 /q1 , ..., pm /qm ) on V defines in a clear
way a map
m
[
f :V \ V (qi ) −→ PnC ,
i=1

and a regular map is defined over the whole V . This map is called rational
(resp. regular ) morphism on V .

Definition 1.1.12. Let V ⊆ PnC and W ⊆ Pm C be two projective algebraic

varieties. A rational morphism (resp. regular morphism or simply morphism)
between V and W is simpla a rational (resp. regular) map f : V −→ Pm
C taking
values in W .

Remark 1.1.2. Notice that a regular morphism is a holomorphic map between

projective analytic varieties.
On PnC , and hence on every projective algebraic set, one can define a new
topology, called the Zariski topology: an open subset is the complement of a
projective algebraic set. Any regular morphism is continuous with respect to
this topology.
In conclusion, we have the following picture:

proj. analytic manifolds ⊂ proj. analytic varieties ⊂ complex manifolds

∪ ∪
proj. algebraic manifolds ⊂ proj. algebraic varieties
Vector bundles 15

A priori, there is no reason why a projective analytic variety, or manifold,

is projective. Anyway, we have there impressive result, that can be seen as a
corollary of the GAGA Theorem.

Theorem 1.1.2. (Chow Theorem). Every projective analytic variety is alge-

braic. Moreover, any holomorphic map between two projective algebraic varieties
is a regular morphism.

In general, moreover, there is no reason why a complex manifold should be

projective. In dimension 1, one has the following:

Theorem 1.1.3. Any Riemann surface is projective, hence algebraic.

This is not the case in higher dimension:

Theorem 1.1.4. The Hopf surface is not projective.

This is a very important point: to classify Riemann surfaces, we can consider

them to be projective. To classifiy compact complex surfaces, one needs to
choose between the whole family of compact complex surfaces, and the family
of those which are projective.

1.2 Vector bundles

Vector bundles are one of the main objects of study in complex geometry: in
particular, line bundles will be one of the main tools in this series of lectures,
and it seems useful to recall basic facts and definitions. Anyway, I prefer to
posticipate the discussion about properties of line bundles, which will be the
content of the second chapter of these notes. As in the following we will not
only use vector bundles over complex manifolds, but even on real manifolds, we
will start our discussion on a differentiable manifold M .

Definition 1.2.1. A differentiable vector bundle E on M is a differentiable

manifold E together with a differentiable morphism p : E −→ M such that

1. for every point m ∈ M , the fibre Em := p−1 (m) is a real vector space;

2. there is an open cover U = {Ui } of M , a real vector space T and a

diffeomorphism
ϕi : p−1 (Ui ) −→ Ui × T,

called trivialization, such that for every p ∈ Ui ϕUi |Ep : Ep −→ {p} × T is

an isomorphism of vector spaces.

For every i, j ∈ I such that Uij 6= ∅ there is a differentiable map

ϕij : Uij −→ GL(T ),

16 Background material

called transition function, such that for every p ∈ Uij and every t ∈ T

ϕ−1 −1
j (p, t) = ϕi (p, ϕij (p)(t))

and for every i, j, k ∈ I such that Uijk := Ui ∩ Uj ∩ Uk 6= ∅ we have

ϕij ◦ ϕjk ◦ ϕki = id.

Definition 1.2.2. Let p : E −→ M be a differentiable vector bundle on M . If

the rank of T is d, then E is called a differentiable vector bundle of rank d. A
differentiable vector bundle of rank 1 is called differentiable line bundle.

Definition 1.2.3. Let pE : E −→ M and pF : F −→ M be two differentiable

vector bundles A morphism of vector bundles is a differentiable map f : E −→ F
such that

1. pF ◦ f = pE

2. f|Ep : Ep −→ Fp is a linear map for every p ∈ M .

It is not difficult to show that the category VM of differentiable vector bundles

on M is an abelian category.

Definition 1.2.4. Let p : E −→ M be a differentiable vector bundle on M .

A differentiable section of E is a differentiable map s : M −→ E such that
pE ◦ s = idM . The set of differentiable sections of E is denoted Γ(M, E) or
H 0 (M, E), and has a natural structure of real vector space.

All the definitions we have given up to now go through if one replaces real by
complex and differentiable by holomorphic: in this way one gets the notion of
holomorphic vector bundle, morphism between two holomorphic vector bundles
and holomorphic section.
Example 1.2.1. Let T be a real (resp. complex) vector space, M a differen-
tiable (resp. complex) manifold. Then T × M is clearly a differentiable (resp.
holomorphic) vector bundle, called the trivial bundle.
Example 1.2.2. Let M be a differentiable manifold, m ∈ M . Let TM :=
S
m∈M Tm (M ) with the natural differentiable structure coming from M and
the natural projection p : TM −→ M . Then TM is a differentiable vector bundle
of rank equal to the dimension of M , and it is called differentiable tangent bun-
dle. A section of the differentiable tangent bundle is called differentiable vector
field.
If M is a complex manifold, one can define in a similar manner the holomor-
phic tangent bundle and holomorphic vector fields.
There are important constructions one can perform using vector bundles.
Sheaves and cohomology 17

Definition 1.2.5. If p : E −→ M is a differentiable (resp. holomorphic)

vector bundle, a differentiable (resp. holomorphic) sub-vector bundle of E is
a closed submanifold F ⊆ E such that p|F : F −→ M is a vector bundle. If
F is a differential sub-vector bundle of E, then the quotient vector bundle is
the differentiable (resp. holomorphic) vector bundle whose fiber over the point
m ∈ M is the quotien vector space Em /Fm .

Definition 1.2.6. If p : E −→ M is a differentiable (resp. holomorphic)

vector bundle, then the dual vector bundle is E ∗ := m∈M Em ∗
S
, with the clear
differentiable (resp. holomorphic) structure and the clear projection to M .

Example 1.2.3. If M is a differentiable manifold, the differentiable cotangent

∗
bundle if TM := (TM )∗ . Similarly, if M is a complex manifold, one has the
notions of holomorphic cotangent bundle ΩM .

Definition 1.2.7. If pi : Ei −→ M is a differentiable (resp. holomorphic) vector

bundle, i = 1, 2, then one can define easily the direct sum, tensor product, wedge
product vector bundle, denoted, E1 ⊕ E2 , E1 ⊗ E2 and E1 ∧ E2 . If E is a vector
bundle of rank k, we define the determinant line bundle of E as

k
^
det(E) := E.

Example 1.2.4. Let M be a differentiable manifold. The canonical differentiable

line bundle of M is
∗
KM := det(TM ).

If M is a complex manifold, the canonical holomorphic line bundle is

KM := det(ΩM ),

where ΩM is the holomorphic cotangent bundle of M .

Example 1.2.5. Let M be a differentiable (resp. complex manifold), and let

N ⊆ M be a differentiable (resp. complex) submanifold. The differentiable
(resp. holomorphic) normal bundle to N in M is the quotient bundle

NN/M := TM |N /TN .

In the algebraic setting, one can define algebraic vector bundles and their
sections in a completely analogue way, only replacing complex manifold by al-
gebraic manifold, and holomorphic by regular. Every algebraic vector bundle is
then clearly holomorphic and, as we will see, on a projective manifold even the
converse is true.
18 Background material

1.3 Sheaves and cohomology

In this section I will recall the basic notion of sheaf on a topological space,
and the fundamental theory of cohomology. In particular, I will define the
notions of presheaf, separated presheaf and sheaf, with the properties that we
will use in the following lectures, and I will present some important examples
and constructions.
Next, I will concentrate in cohomology: this will be introduce in a complete
axiomatic way, and the main aim will be to define some important cohomology
theories. Namely, I will present the singular (co)homology of a manifold, intro-
ducing Betti numbers, the Euler charcateristic and the Poincaré duality. Then I
will define the Čech chomology and its properties, aiming to a way to calculate
cohomology of sheaves on a manifold: as a consequence I will state the de Rham
and the Dolbeault Theorems, linking singular cohomology to differential forms.
To conclude, I will present a brief discussion on properties of cohomology
of sheaves on projective varieties, namely the Serre Vanishing and Finiteness
Theorem, the Serre Duality Theorem and other properties.
To conclude I will state the GAGA Theorem, linking the algebraic and the
complex worlds, and present come corollaries.

1.3.1 Sheaves
Sheaves theory is one of the main technical tools in modern mathematics. We
will use it almost everywhere in the next lectures, and I will present it in a
very elemenatry and general way in this section. In the following, let M be a
topological space, whose topology is denotet τ , and R a principal domain (i. e.
a commutative ring with unity, without zero divisors and such that every ideal
admits one generator).

Definition 1.3.1. A presheaf F of R−modules on M is given by the following

data:

1. an R−module F (U ) for every open subset U of M ;

2. a morphism of R−modules ρU V : F (U ) −→ F (V ), called restriction mor-

phism, for every V ⊆ U open subsets of M , such that

• ρU
U = idF (U ) for every open subset U of M ;

• ρVW ◦ ρU U
V = ρW for every W ⊆ V ⊆ U open subsets of M .

The elements in F (U ) are called sections of F on U .

Definition 1.3.2. A separated presheaf F of R−modules on M is a presheaf

F of R−modules on M such that for every open subset U ⊆ M and every open
Sheaves and cohomology 19

cover U = {Ui }i∈I of U , if f ∈ F is such that ρU

Ui (f ) = 0 for every i ∈ I, then
f = 0.

Definition 1.3.3. A sheaf F of R−modules on M is a separated presheaf of

R−modules on M such that for every open subset U of M , every open cover
U = {Ui }i∈I of U and every collection {fi }i∈I , where fi ∈ F (Ui ) are such
Uj
that for every i, j ∈ I such that Uij 6= ∅ we have ρU
Uij (fi ) = ρUij (fj ), there is
i

f ∈ F (U ) such that ρUUi (f ) = fi .

Definition 1.3.4. Let F and G be two presheaves of R−modules on M . A

morphism of presheaves from F to G is a collection f = {fU }U ∈τ , where fU :
F (U ) −→ G (U ) is a morphism of R−modules for every open subset U of M ,
and for every V ⊆ U open subsets of M we have ρU U
V ◦ fU = fV ◦ ρV .
If F and G are separated presheaves or sheaves, then a morphism between
them is simply a morphism of presheaves.

Definition 1.3.5. Let F be a (pre)sheaf of R−modules on M , and let U ⊆ M

be an open subset. The restriction of F to U is the (pre)sheaf F|U such that
for every V ⊆ U open we have F|U (V ) := F (V ), with the obvious restriction
morphisms.

Now, let F and G be two (pre)sheaves of R−modules on M . For every open

subset U ⊆ M , the set Hom(F|U , G|U ) has a natural structure of R−module,
and if V ⊆ U is open, there is an obvious restriction morphism. We have then
H om(F , G ), which is the (pre)sheaf defined as

H om(F , G )(U ) := Hom(F|U , G|U )

and obvious restriction morphisms.

Definition 1.3.6. The (pre)sheaf H om(F , G ) is called (pre)sheaf of mor-

phisms from F to G .

There are important definitions that are useful in sheaf theory. First of all,
notice that if m ∈ M and τm is the set of the open subsets of M containing m,
then the family {F (U )}U ∈τm is a direct system of R−modules: on τm one has
a partial order relation, given by V ≤ U if and only if V ⊆ U , and if V ≤ U one
has a map ρU V : F (U ) −→ F (V ).

Definition 1.3.7. Let m ∈ M and F be a presheaf of R−modules on M . The

stalk of F at m is
Fm := lim F (U ).
−→
U ∈τm

If F is a separated presheaf or a sheaf, then its stalk at m is simply its stalk at

m as presheaf.
20 Background material

Notice that for every U ∈ τm there is a canonical map of R−modules ρU

m :
F (U ) −→ Fm . Moreover, if f : F −→ G is a morphism of presheaves, then it
induces a map
fm : Fm −→ Gm .

There are two important construcions one can perform on presheaves or sheaves.
First of all, let F be a presheaf of R−modules on M , and let U be an open
subset of M . Define: Sh(F )(U ) as the set of maps s : U −→ x∈U Fx verifying
`

these two properties:

1. s(x) ∈ Fx for every x ∈ U ;

2. for every x ∈ U there is V ⊆ U open neighborhood of x and an element

sV ∈ F (V ) such that s(y) = ρVy (sV ) for every y ∈ V .

Clearly, Sh(F )(U ) is an R−module for every open subset U of M . Moreover,

if V ⊆ U are open subsets of M , define

ρU
V : Sh(F )(U ) −→ Sh(F )(V ), ρU
V (s) := s|V .

It is easy to show that such a map is a well defined morphism of R−modules,

and that Sh(F ) is a sheaf.

Definition 1.3.8. Let F be a presheaf of R−modules on M . The sheaf Sh(F )

is the sheaf associated to F .

Remark 1.3.1. If m ∈ M , it is easy to see that Fm = Sh(F )m .

Now, let M and N be two topological spaces, and let f : M −→ N be a
continuous map. Let F be a (pre)sheaf of R−modules on M , and let U be an
open subset of N . Define

f∗ F (U ) := F (f −1 (U )),
f −1 (U )
and if V ⊆ U are open subsets of N , define ρU
V := ρf −1 (V ) . It is easy to see that
if F is a sheaf, then f∗ F is a sheaf too.

Definition 1.3.9. The (pre)sheaf f∗ F is called push-forward of F by f .

Let now F be a (pre)sheaf of R−modules on N , and U ⊆ M be an open

subset. Define:
f −1 F (U ) := lim F (V ),
−→
f (U )⊆V

with obvious restriction morphisms. In general, this is not a sheaf, even if F is.

Definition 1.3.10. The inverse image of F by f is the sheaf associated to the

presheaf f −1 F . This will still be denoted f −1 F .
Sheaves and cohomology 21

Other important constructions are those of kernel, cokernel and image of

a morphism of sheaves of R−modules. Let F and G be two (pre)sheaves of
R−modules on M , and let f : F −→ G be a morphism. For every open subset
U of M let
ker f (U ) := ker(fU )
with the obvious restriction morphism. If F is a sheaf, then ker f is a sheaf,
called the kernel of f . Similarily one can define the cokernel and the image of
f , but in order to get a sheaf one has to consider the associated sheaf to the
naif presheaf. It is not difficult then to see that the category whose objects are
sheaves of R−modules and morphisms are morphisms of presheaves between
them is abelian.

Definition 1.3.11. Let F , G and H be three presheaves of R−modules on

M , and let
f g
F −→ G −→ H .
This sequence is exact if and only if for every open subset U of M the sequence
fU gU
F (U ) −→ G (U ) −→ H (U )

is an exact sequence of R−modules. If F , G and H are sheaves, then the

sequence above is exact as sequence of sheaves if and only if for every m ∈ M
the sequence
fm gm
Fm −→ Gm −→ Hm
is an exact sequence of R−modules.

Remark 1.3.2. As we will see, an exact sequence of sheaves is not, in general,

an exact sequence of presheaves.
Clearly one can perform constructions as direct sums, tensor products and
wedge products even with sheaves, getting new sheaves. To conclude this sec-
tion, we present some important examples.
Example 1.3.1. Let M be a topological space, R a principal domain and G any
R−module. Then one can define the presheaf G as follows: for every open subset
U of M let G(U ) := G and restriction morphisms are simply the identity of G.
This in general is not a sheaf (e. g. when M is not connected). The associated
presheaf is denoted GM and is called constant sheaf (of value G). Important
examples are when R = Z, Q, R or C, getting as instance the sheaves ZM , QM ,
RM and CM .
Example 1.3.2. Let M be a differentiable manifold. For every open subset U of
M , let
OM (U ) := {f : U −→ R | f is differentiable},
with clear restriction functions. It is easy to see thar OM is a sheaf, called the
structural sheaf of M . In a similar way, if M is a complex manifold, one can
22 Background material

define the structural sheaf of M as the sheaf of holomorphic functions. More

generally, one can define the sheaf MM of meromorphic functions on M .
If M is an affine algebraic variety in Cn , one can define the structural sheaf
OM simply considering polynomials: if U is a Zariski open subset of M , whose
complementary is V (f ) for some polynomial f , then we have

OM = {p/f k | k ∈ Z, p ∈ C[M ]},

where C[M ] is the ring of regular functions on M . Similar definitions hold if M

is a projective algebraic variety.
Example 1.3.3. Let M be a (differentiable, complex) manifold, and p : E −→ M
be a (differentiable, holomorphic) vector bundle. For every open subset U of M
define
OM (E)(U ) := Γ(U, E),

with obvious restriction morphisms. It is easy to see that OM (E) is a sheaf,

called the sheaf of sections associated to E.

Definition 1.3.12. Let M be a differentiable (resp. complex) manifold, and

let F be a sheaf of R−vector spaces (resp. C−vector spaces) on M . The
F is an OM −module if there is a morphism of sheaves c : OM −→ F . A
morphism of OM −modules is a morphism of sheaves respecting the struc-
ture of OM −modules. Using this morphisms, one can define the OM −module
H omOM (F , G ).

Remark 1.3.3. Notice that if F is an OM −module, there is a natural isomor-

phism of OM (M )−modules

h : F (M ) −→ HomOM (OM , F ), h(f )(s) := s · f.

Definition 1.3.13. Let M be a (differentiable, complex) manifold, and let F be

an OM −module. The dual of F is the OM −module F ∗ := H omOM (F , OM ).

Notice that if F and G are two OM −modules, then one can define F ⊗OM G
in the obvious way, getting a new OM −module.

Definition 1.3.14. Let M and N be two complex (resp. algebraic) manifolds.

A morphism between (M, OM ) and (N, ON ) is a couple (f, f ] ), where f : M −→
N is a holomorphic (res. regular) map, and f ] : f −1 ON −→ OM is a map of
sheaves.

In the following, by a morphism between two complex (resp. algebraic)

manifolds (or varieties) we will always mean a morphism in the sense of the
previous definition. As we have a map f ] : f −1 ON −→ OM , we can think of
OM as an f −1 ON −module. Then, the following definition makes sense:
Sheaves and cohomology 23

Definition 1.3.15. Let M and N be two complex (resp. algebraic) manifolds,

and let f : M −→ N be a holomorphic (res. regular) map between them. Let
F be an ON −module. The pull-back of F by f is the OM −module

f ∗ F := f −1 F ⊗f −1 ON OM .

Remark 1.3.4. Notice that the pull-back of any ON −module has a natural struc-
ture of OM −module. If F is an OM −module, then f∗ F has a natural structure
of f∗ OM −module. Now, f∗ OM has a natural structure of ON −module, as by
adjunction the morphism f ] gives a morphism

f [ : ON −→ f∗ OM .

In conclusion, f∗ F has a natural structure of ON −module.

Lr
Definition 1.3.16. An OM − module F is free of rank r if F ' i=1 OM .
It is locally free of rank r if there is an open cover {Ui }i∈I of M such that for
Lr
every i ∈ I there is an isomorphism F|Ui ' i=1 OM |Ui . It is called coherent
if there are n, m ∈ N and an open cover {Ui }i∈I of M such that for every i ∈ I
there is an exact sequence of sheaves
⊕m ⊕n
OM |Ui −→ OM |Ui −→ F|Ui −→ 0.

Remark 1.3.5. Notice that any locally free sheaf is coherent.

Example 1.3.4. Let p : E −→ M be a vector bundle of rank r. It is easy to verify
that the sheaf OM (E) of sections associated to E is a locally free OM −module of
rank r. Moreover, one can verify that to any locally free OM −module F of rank
r one can associate a vector bundle pF : EF −→ M such that F is canonically
isomorphic (as OM −module) to OM (EF ), and EOM (E) is canonically isomorphic
to E as vector bundles.
Similarily, to any morphism of vector bundles one can associate a morphism
of OM −modules between the corresponding sheaves of sections, and viceversa.
In this way one sees that the category of vector bundles on M and the category
of locally free OM −modules are equivalent.
In particular, any line bundle can be viewed as a locally free sheaf of rank
1: using the previous notations, one can write OM (KM ) for the canonical line
bundle.

Definition 1.3.17. Let F be an OM −module. A resolution of F is an exact

sequence of sheaves

0 −→ F −→ F 1 −→ F 2 −→ ...

Example 1.3.5. Let M be a differentiable manifold, p ∈ N, and let

p
^
p
EM := ∗
TM .
24 Background material

This is the sheaf of differentiable p−forms on M : it is easy to see that this

is a vector bundle whose rank is np , where n = dim(M ). Using the local

coordinates and the definition of differential of a p−form, one can produce a

morphism of sheaves
p p+1
dp : EM −→ EM .
It is well known that dp+1 ◦ dp = 0 for every p ∈ N. Moreover, there is a clear
injective map RM −→ EM 1
whose image is in the kernel of d1 , so that we get a
complex of sheaves
d0 d1 d2
0 −→ RM −→ OM −→ EM
1
−→ EM
2
−→ ...

By the Poincaré Lemma, this complex is a resolution of RM .

If M is a complex manifold, one can consider ΩpM , the vector bundle of
holomorphic p−forms on M . More generally, one can consider the sheaf Ωp,q
M
of (p, q)−forms on M . Using the local coordinate and the definition of the
∂−differential, one can produce a morphism of sheaves
q
∂ : Ωp,q p,q+1
M −→ ΩM .
q+1 q
Again, it is well known that ∂ ◦ ∂ = 0 for every q ∈ N. Moreover, there is
p p,0 1
a clear injective map ΩM −→ ΩM whose image lies in the kernel of ∂ , so that
we get a complex of sheaves
0 1 2
∂ ∂ ∂
0 −→ ΩpM −→ Ωp,0 p,1 p,2
M −→ ΩM −→ ΩM −→ ...

By the ∂−Poincaré Lemma, this complex is a resolution of ΩpM .

To conclude this section, let me define two important classes of sheaves.

Definition 1.3.18. A sheaf F is fine if for every locally finite open cover
{Ui }i∈I of M , for every i ∈ I there is an endomorphism

hi : F −→ F

whose support (i. e. the set of m ∈ M such that hi,m 6= 0) is contained in Ui .

A sheaf F is flasque if for every V ⊆ U open subsets of M , the restriction
morphism ρU
V is surjective.
p
Example 1.3.6. All the sheaves EM are fine for a differentiable manifold M .
p,q
Similarily, all the sheaves ΩM are fine for a complex manifold M . In particular,
the resolutions presented in the previous example are fine resolutions, i. e.
resolutions whose sheaves are fine.

Definition 1.3.19. Let F be a sheaf of R−modules on M . The sheaf of

discontinuous sections of F is the sheaf associating to any open subset U of M
the R−module
a
C 0 (F )(U ) := {s : U −→ Fx | s(x) ∈ Fx },
x∈U
Sheaves and cohomology 25

with obvious restriction maps.

Remark 1.3.6. The sheaf of discontinuous sections of a sheaf F is a flasque

sheaf. In particular, this implies that any sheaf admits a flasque resolution: we
have an injective map
s : F −→ C 0 (F ),

sending f ∈ F (U ) to
a
sU (f ) : U −→ Fx , sU (f )(y) := ρU
y (f ).
x∈U

The fact that the map is injective follows from the fact that F is a sheaf. Now,
let Z 1 (F ) := C 0 (F )/F , and let C 1 (F ) := C 0 (Z 1 (F )). Clearly, there is a
well-defined map d0 : C 0 (F ) −→ C 1 (F ). Going on by induction, one gets a
complex of sheaves

s d0 d1
0 −→ F −→ C 0 (F ) −→ C 1 (F ) −→ ...

and it is easy to show that this is a resolution of F .

1.3.2 Cohomology
In this section I will introduce one of the main tools we are going to use, namely
sheaf and singular cohomology. I will introduce it in a pure axiomatic way, and
then introduce some examples. Consider a topological space M .

Definition 1.3.20. A cohomology theory on M is given by the following data:

1. for every sheaf of R−modules on M define an R−module H q (M, F ) for

every q ∈ Z,

2. for every morphism of sheaves of R−modules f : F −→ G define a mor-

phism of R−modules

H q (f ) : H q (M, F ) −→ H q (M, G ),

for every q ∈ Z,

such that the following properties are satisfied:

• for every sheaf of R−modules on M and q < 0 we have H q (M, F ) = 0;

for q = 0 there is an isomorphism of R−modules

iF : H 0 (M, F ) −→ F (M )

such that for every f : F −→ G morphism of sheaves of R−modules on

M we have iG ◦ H 0 (f ) = H 0 (g) ◦ iF ;
26 Background material

• if F is flasque or fine, then H q (M, F ) = 0 for every q > 0;

• for every sheaf F of R−modules we have H q (idF ) = idH q (M,F ) for every
q ∈ Z, and for every morphism f : F −→ G and g : G −→ H we have
H q (g ◦ f ) = H q (g) ◦ H q (f ) for every q ∈ Z.

• for every exact sequence of sheaves of R−modules

f g
0 −→ F −→ G −→ H −→ 0

and every q ∈ Z there is δ q : H q (M, H ) −→ H q+1 (M, F ), morphism of

R−modules called connecting morphism, such that the sequence
δ q−1 H q (f ) H q (g) δq
... −→ H q (M, F ) −→ H q (M, G ) −→ H q (M, H ) −→ ...

is exact. This is called long exact sequence induced in cohomology.

This abstract definition does not say anything about the existence of coho-
mology theory. There are anyway several ways of producing them. A first way
is to consider a flasque resolution of a sheaf F , as
d0 d1
0 −→ F −→ F 0 −→ F 1 −→ ...

Then take the complex

d0 d1
0 −→ Γ(M, F 0 ) −→
M
Γ(M, F 1 ) −→
M
...

and let
H q (M, F ) := ker(dqM )/im(dq−1
M ).

One can replace flasque resolutions with fine resolutions, and in this way, up
to providing the existence of a flasque or fine resolution for every sheaf, one
produces a cohomology theory.
Example 1.3.7. Let M be a differentiable manifold: then there is a fine resolution
of the sheaf RM , which is
d0 d1 d2
0 −→ RM −→ OM −→ EM
1
−→ EM
2
−→ ...

Then one can calculate the cohomology of RM as just mentioned, getting the
p
de Rham cohomology HdR (M ).
If M is a complex manifold, for every p ∈ N0 there is a fine resolution of the
sheaf ΩpM , which is
0 1 2
∂ ∂ ∂
0 −→ ΩpM −→ Ωp,0 p,1 p,2
M −→ ΩM −→ ΩM −→ ...

One can then calculate the cohomology of ΩpM as before, getting the Dolbeault
of ∂−cohomology H∂p,q (M ) := H∂q (M, ΩpM ).
Sheaves and cohomology 27

Another way to produce a cohomology theory is Čech cohomology. Let

U = {Ui }i∈I be an open cover of M . For every i1 , ..., in ∈ I, where ij 6= ik for
i 6= k, let
n
\
Ui1 ,...,in := Uij .
j=1

Then define
Y
C p (U , F ) := F (Ui0 ,...,ip ),
i0 ,...,ip ∈I

and the map

dp : C p (U , F ) −→ C p+1 (U , F )

as
p+1
X 
(dp (α))i0 ,...,ip+1 := (−1)j (α)i0 ,...,ij−1 ,ij+1 ,...,ip+1 ,
j=0 |Ui0 ,...,ip+1

where if β ∈ C q (U , F ), we denote (β)i0 ,...,iq its component in F (Ui0 ,...,iq ). It

is not difficult to show that dp+1 ◦ dp = 0 for every p ∈ N0 , so that we get a
complex
d0 d1
0 −→ C 0 (U , F ) −→ C 1 (U , F ) −→ ...

Now define
Ȟ q (M, U , F ) := ker(dq )/im(dq−1 ).

Now, let us consider two open covers U = {Ui }i∈I and V = {Vj }j∈J of M . We
have a partial order relation on the family of open covers of M , putting V ≤ U
if and only if for every j ∈ J there is i ∈ I such that Vj ⊆ Ui . In particular,
then for every j1 , ..., jp ∈ J there are i1 , ..., ip ∈ I such that Vj1 ,...,jp ⊆ Ui1 ,...,ip .
Hence, we have a restriction map
Ui
ρU
V : C (U , F ) −→ C (V , F ),
p p
ρU
,...,i
V |F (Ui := ρVj 1,...,jpp ,
1 ,...,ip 1

which, as easily seen, pass through cohomology. From this, one sees that the
family of Ȟ p (M, U , F ) forms a direct system, and we can define

Ȟ p (M, F ) := lim Ȟ p (M, U , F ).

−→
U

An important result is that if M is Hausdorff and admits a countable basis for

its topology, then one can show that Čech cohomology is a cohomology theory.
Example 1.3.8. For any topological space M one can define the notion of singular
cohomology of M . To do that, one needs the notion of p−simplex, which is the
topological space

∆p := {(x1 , ..., xp ) ∈ Rp | xi ≥ 0, x1 + ..., xp = 1}.

28 Background material

A p−simplex in M is simply a continuous map fp : ∆p −→ M . One can then de-

fine the group Sp (M ), which is the free abelian group generated by p−simplexes
in M , and S p (M ) := HomZ (Sp (M ), Z). Moreover one has morphisms

dp : Sp (M ) −→ Sp−1 (M ), dp : S p (M ) −→ S p+1 (M ),

such that dp ◦ dp+1 = 0 and dp+1 ◦ dp = 0. Finally, one has

Hp (M, Z) := ker(dp )/im(dp+1 ), H p (M, Z) := ker(dp )/im(dp−1 ).

These are called the singular homology and cohomology of M . One can modify
the contruction in order to get H p (M, Q) (resp. H p (M, R), H p (M, C)): by the
Universal Coefficient Theorem, this is simply H p (M, Z) ⊗Z Q (resp. tensoring
by R, by C). If M is Hausdorff and admits a countable basis for its topology,
one can show that
H p (M, Z) ' Ȟ p (M, ZM ),

and similar for Q, R or C.

Definition 1.3.21. Let M be a topological space. The i−th Betti number of

M is
bi (M ) := rk(H i (M, Z)).

The topological Euler characteristic of M is

X
e(M ) := (−1)i bi (M ).
i

These two examples are indeed strictly linked. We have the following:

Theorem 1.3.1. (de Rham, Dolbeault). Let M be a differentiable manifold.

For every p ∈ Z we have
p
H p (M, R) ' HdR (M ).

If M is a complex manifold, then for every p, q we have

H p (M, ΩqM ) ' H∂p,q (M ).

To conclude this section, I just want to mention an important property of

singular cohomology of a complex manifold. First of all, recall that any complex
manifold admits an orientation. We have the following:

Proposition 1.3.2. Let M be a connected compact real manifold of dimension

n. Then M admits an orientation if and only if Hn (M, Z) ' Z.

From this, one sees that if M is a complex manifold of real dimension 2n,
then b2n (M ) = 1. A generator of H2n (M, Z) is called fundamental class of M ,
and is denoted ηM .
Sheaves and cohomology 29

Now, let M be a real manifold of dimension n. For every integer p by

definition we have
S p (M ) × Sp (M ) −→ Z
which is a well defined pairing, passing through cohomology, getting hence a
pairing, called Kronecker pairing

(., .) : H p (M, Z) × Hp (M, Z) −→ Z.

Passing to Q, one has a perfect pairing, so that H p (M, Q) ' Hp (M, Q)∗ .
One can introduce a Z−graded ring structure on H ∗ (M, Z) := p H p (M, Z):
L

the grading is clear, while the product is the following, called cup product. Let
f ∈ S p (X), g ∈ S q (X), and α ∈ Sp+q (X). Let αp be the p−simplex obtained
considering only the first p + 1 elements of α, and αq be the q−simplex obtained
by the remaining elements. Then let

f ∪ g(α) := f (αp )g(αq ).

This product passes through cohomology, and we get the desired cup product
on H ∗ (M, Z), still denoted ∪.
Now we can define the following

DM : H p (M, Z) −→ Hn−p (M, Z), (a, DM (b)) = (a ∪ b, ηM ).

Theorem 1.3.3. (Poincaré duality). For every p the morphism DM is an

isomorphism.

Corollary 1.3.4. Let M be a real manifold of dimension n. Then bi (M ) =

bn−i (M ). In particular bi (M ) = 0 if i < 0 or i > n.

Example 1.3.9. If M is a Riemann surface, then b0 (M ) = b2 (M ) = 1. For

b1 (M ), recall that H 1 (M, Z) ' Z2g(M ) , where g(M ) is the genus of M . In
conclusion, we have b1 (M ) = 2g(M ).

1.3.3 Cohomology of coherent sheaves

It seems now useful to recall some basic facts about the cohomology of coherent
sheaves. In particular, I will present some results on the cohomology of coherent
sheaves on affine varieties and on PnC . As a conclusion of this, I will introduce
an important theorem, due to Serre, about the vanishing and the dimension of
cohomology C−vector spaces of coherent sheaves. Let me start with coherent
sheaves on Cn . In the following, let

R := C[z1 , ..., zn ].

As we are concerned with Cn as an affine variety, we put on it the Zariski

topology: in this way, we see that the structural sheaf of regular functions
30 Background material

on Cn is completely determined by R: if f ∈ R is a regular map on Cn and

U (f ) = Cn \ V (f ) is the open subset of Cn associated to f , we have

OCn (U (f )) = Rf ,

where Rf is the localization of R at f (i. e. Rf = {p/f k | f ∈ R, k ∈ N0 }).

More generally, if M is any R−module, one can define the following sheaf: for
any open subset U (f ) as before, define

M
f(U (f )) := Mf ,

where again Mf is the localisation of M at f . Notice that if U (g) ⊆ U (f ), this

implies that V (f ) ⊆ V (g), meaning that f is contained in the ideal generated
by g. Then there is k0 ∈ N such that f = g k0 , so that there is a well defined
map
U (f ) f
ρU (g) : M (U (f )) −→ M
f(U (g)).

f is an OCn −module, called quasi-coherent sheaf asso-

It is easy to verify that M
ciated to M . If M and N are two R−modules and h : M −→ N is a morphism
of R−modules, then we can define a morphism of OCn −modules

e f −→ N
h:M e.

Remark 1.3.7. It is easy to verify that M f is coherent if and only if M is a finitely

generated R−module: indeed, M is a finitely generated R−module if and only
if its localization Mf is a finitely generated Rf −module, i. e. a quotient of some
Ln
i=1 Rf by the submodule of relations, for every f ∈ R. But this is equivalent
to ask that M f is a coherent OCn −module.
Conversely, to any OCn −module F one can associate Γ(Cn , F ), which has
a natural structure of R−module.

Proposition 1.3.5. The functor sending any R−module M to OCn −module M f

and any morphism h to h is an equivalence of between the category of R−modules
e
and the category of quasi-coherent OCn −modules. The inverse functor is the
one sending a quasi-coherent sheaf F to the R−module of its global sections.
Moreover, this functor is exact, i. e. if
f g
0 −→ F −→ G −→ H −→ 0

is an exact sequence of quasi-coherent sheaves on C n , the sequence

f n g n
0 −→ Γ(Cn , F ) −→
C
Γ(Cn , G ) −→
C
Γ(Cn , H ) −→ 0

is exact.

All these constructions and results can be generalized to any affine variety.
In this generality, one can prove the following:
Sheaves and cohomology 31

Proposition 1.3.6. Let X be an affine variety in Cn , and F a quasi-coherent

sheaf on X. Then Ȟ p (X, F ) = 0 for every p > 0.

Let us pass to coherent sheaves on PnC . Let

S := C[z0 , ..., zn ],

and consider it as a Z−graded ring, where every zi has degree 1 and α ∈ C has
degree 0. Let d ∈ Z and S(d) be the same as S, but the degree is shifted by d, i.
e. α ∈ C has degree −d and zi has degree 1 − d. As we are concerned with PnC as
a projective algebraic variety, we put on it the Zariski topology. In this way, we
see that the structural sheaf of regular functions on PnC is completely determined
by S. Indeed, if f is a homogenous polynomial in S and U (f ) = PnC \ V (f ) is
the open subset of PnC determined by f , then

OPnC (U (f )) = Sf ,

where
Sf := {p/f k | p ∈ S, k ∈ Z, deg(p) = deg(f k )}.

More generally, if M is a Z−graded S−module, one can define the following

OPnC −module: for every open subset U of PnC as before, let

M
f(U (f )) := Mf .

As in the previous case, if U (g) ⊆ U (f ), then one can define a restriction map
U (f ) f
ρU (g) : M (U (f )) −→ M
f(U (g)).

This is the quasi-coherent sheaf associated to M .

Remark 1.3.8. As in the previous case, Mf is coherent if and only if M is finitely
generated.
Example 1.3.10. Let d ∈ Z and consider S(d), which is a graded S−module.
The associated sheaf is
O(d) = OPnC (d) := S(d).
g

It is not difficult to see that for every d ∈ Z, the sheaf Od is a locally free sheaf
of rank 1 on PnC . Notice that if d1 , d2 ∈ Z, then

O(d1 ) ⊗OPn O(d2 ) ' O(d1 + d2 ).

C

These locally free sheaves are very important in the study of PnC and, in
general, of projective varieties. An important result is the following, describing
their cohomology:

Proposition 1.3.7. The following hold (here the cohomology is the Čech coho-
mology):
32 Background material

H 0 (PnC , O(d)) is a graded isomorphism.

L
1. The natural map S −→ d∈Z

2. For every d ∈ Z and every 0 < i < n we have H i (PnC , O(d)) = 0.

3. We have H n (PnC , O(−n − 1)) ' C, and H n (PnC , O(d)) = 0 for every d >
−n − 1.

4. For every d ≥ 0 the natural map

H 0 (PnC , O(d)) × H n (PnC , O(−k − n − 1)) −→ H n (PnC , O(−n − 1)) ' C

is a perfect pairing. This is called Serre duality for PnC .

5. Let KPnC be the canonical line bundle of PnC . Then

OPnC (KPnC ) ' O(−n − 1).

6. Let F be a coherent sheaf on PnC . Then there are k ∈ N and n1 , ..., nk ∈ Z

such that there exists a surjective morphism
k
M
O(ni ) −→ F −→ 0.
i=1

From this one can deduce the following important theorem:

Theorem 1.3.8. (Serre’s Vanishing and Finiteness). Let X be a projec-

tive algebraic variety in PnC , and let j : X −→ PnC be the inclusion. Let F be a
coherent OX −module, and for every d ∈ Z let F (d) := F ⊗OX j ∗ O(d). Then

1. for every i ∈ Z the C−vector space H i (X, F ) is finite dimensional;

2. there is n0 = n0 (F ) ∈ Z such that H i (X, F (m)) = 0 for every m > n0

and every i > 0;

3. if dim(X) = n, then H p (X, F ) = 0 for p > n.

To conclude, let me recall some basic properties of pull-back and push-

forward of coherent sheaves. First, let me introduce the following notion:

Definition 1.3.22. Let M and N be two topological spaces, f : M −→ N

a continuous map, and let R be a principal domain. Let F be a sheaf of
R−modules on M and q ∈ Z. The q−th direct image sheaf of F by f is the
sheaf Rq f∗ F associated to the presheaf H q (F ) defined as follows: for every
open subset U of N , let

H q (F )(U ) := H q (f −1 (U ), F|f −1 (U ) ),

with clear restriction maps.

Sheaves and cohomology 33

Remark 1.3.9. Notice that Rq f∗ F = 0 if q < 0, and R0 f∗ F = f∗ F .

Proposition 1.3.9. Let X and Y be two projective algebraic varieties, and let
f : X −→ Y be a morphism.

1. If F is a coherent OY −module, then f ∗ F is a coherent OX −module.

2. If F is a coherent OX −module, then Rq f∗ F is a coherent OY −module

for every q ∈ Z.

3. Let F be a coherent OX −module and E a locally free OY −module of finite

rank. Then there is a canonical isomorphism
∼
Rq f∗ (F ⊗OX f ∗ E) −→ Rp f∗ F ⊗OY E

for every p ∈ Z. This is called projection formula.

1.3.4 The GAGA Theorem

Up to now we have been dealing with two a priori different worlds: the com-
plex one, with complex manifolds, analytic varieties, holomorphic maps and
holomorphic vector bundles; and the algebraic one, with projective algebraic
varieties, regular maps and algebraic vector bundles. we have already seen that
every projective analytic variety (or manifold) is a projective algebraic variety
(or manifold), and that holomorphic maps between two projective algebraic va-
rieties are regular. In this section we want to mention that the same kind of
properties hold even for vector bundles, sheaves and cohomology.
If X is a projective variety, it can clearly be viewed as a complex subvariety
X of PnC : one of the main difference is that on X one has the Zariski topology,
h

which is not Hausdorff, while on X h we have the ordinary topology, which is

Hausdorff. Similarily, if X and Y are two projective varieties and f : X −→ Y
is a regular map, it can be viewed as an holomorphic map f h : X h −→ Y h .
Now, on X one has a natural structure sheaf of regular functions on X,
which is OX . Similarily, on X h we have the sheaf of holomorphic functions on
X, which is OX h . Now, consider the identity map

i : X h −→ X,

which is continuous, and induces a natural map

i] : i−1 OX −→ OX h ,

as regular maps are holomorphic.

More generally, if F is an OX −module, then one can consider i∗ F , which
has a natural structure of OX h −module. If F is coherent, we can define the
34 Background material

following: as F is coherent, there is a Zariski open cover {Ui }i∈I of X such that
for every i ∈ I there is an exact sequence
M
OU⊕m
i
i
−→ OU⊕n
i
−→ F|Ui −→ 0.

In other words, we have F|Ui = coker(Mi ). If one considers Uih , the analytic
variety associated to Ui , it still makes sense to consider OUih , and we clearly have
an analogue map Mih . We can then define the sheaf F h as the OX h −module
such that for every i ∈ I we have
Mh
OU⊕m ⊕n
h −→ OU h −→ F |U h −→ 0.
i
i
i i

Then F h is a coherent OX h −module. It is easy to see that F h ' i∗ F . In a

similar way, to any morphism f : F −→ G of OX −modules on can associate a
morphism f h : F h −→ G h of OX h −modules.
Next step is to compare cohomologies. Let F be a coherent OX −module on
the projective variety X, and let p ∈ Z. As F h ' i∗ F , one can easily produce
a natural map
H p (i) : H p (X, F ) −→ H p (X h , F h )
for every p (here we use Čech cohomology).
The most surprising facts about these constructions, is that one can go in
the opposite direction:

Theorem 1.3.10. (Serre’s GAGA). Let X be a projective variety.

1. For any coherent OX h −module F there is a coherent OX −module F a

such that (F a )h ' F . Moreover, if F is a coherent OX −module, then
(F h )a ' F .

2. For every morphism f : F −→ G of coherent OX h −modules there is a

morphism f a : F a −→ G a of coherent OX −modules such that

• if F = G and f = idF , then f a = idF a ;

• if f : F −→ G and g : G −→ H are two morphisms of coherent
OX h −modules, then (g ◦ f )a = g a ◦ f a ;
• we have (f a )h = f . If f : F −→ G is a morphism of OX −modules,
then (f h )a = f .

3. the natural maps H p (i) are isomorphisms for every p ∈ Z.

By this Theorem one then sees that if we are in the projective world, it is
the same thing to work in the algebraic or in the analytic setting: from now
on, as soon as we are in the projective world, we will only talk about projective
varieties, coherent sheaves and cohomology, without specifying if this is to be
considered in the analytic or in the algebraic sense.
Sheaves and cohomology 35

1.4 Hodge theory

An important piece of theory we are going to use at several points in what follows
is the nowadays called Hodge theory. The aim of this section is to present basic
material on this interesting subject. In particular, we will introduce the notion of
Hermitian metrics, Hodge operators and harmonic forms on a compact complex
manifold. Then we will pass to the case of Kähler manifolds, among which sit
projective manifolds, and to state the main result of the section, namely the
Hodge decomposition theorem. This will lead us to the notion of the Hodge
numbers and of the Hodge diamond, and we will state one of the main tools
of algebraic geometry, which is the Serre Duality Theorem. To conclude, I will
present some remarks about Riemann surfaces and complex surfaces.

1.4.1 Harmonic forms

Let M be a compact complex manifold of dimension n, and let p : E −→ M be
a holomorphic vector bundle.

Definition 1.4.1. A Hermitian metric on E is a C ∞ −function h : E ×E −→ C

such that for every m ∈ M the map hm := h|Em ×Em is a Hermitian inner
product. A holomorphic vector bundle equipped with a Hermitian metric is
called Hermitian vector bundle. The complex manifold M is called Hermitian
manifold if TM ⊗ TM is a Hermitian vector bundle.

Now, let M be a Hermitian manifold with Hermitian metric h on TM ⊗ TM .

Let m ∈ M and consider {z1 , ..., zn } to be a coordinate system at m. Then we
can define a basis for TM,m , which is given by {∂/∂z1 , ..., ∂/∂zn }, and a basis
for TM , which is {∂/∂z 1 , ..., ∂/∂z n }. It is then clear that for every j, k = 1, ..., n
there must be C ∞ −functions hjk (defined on a neighborhood of m) such that
locally at m we have
X n
h= hjk dzj ⊗ dz k .
j,k=1

We can associate to the Hermitian metric h a real (1, 1)−form ωh , which locally
at m is
n
i X
ωh = hjk dzj ∧ dz k .
2
j,k=1

Now, as h is a Hermitian metric on TM ⊗ TM , it induces a metric on every

associated tensor bundle: in particular, if p, q ∈ N0 , we have a Hermitian metric
on ^ p  ^ q 
p,q ∗
EM := ∗
TM ⊗ TM .

Let us now define M p,q

EM
l
:= EM .
p+q=l
36 Background material

We have then a Hermitian metric on EM

l
for every l ∈ N0 . Let
n
^
volh := ωh ,

which is a 2n−form on M , and let the Hodge inner product be

Z
(., .) := (., .)m volh ,
M

where (., .)m is the Hermitian inner product induced by h on the fiber over m.
With respect to the Hodge inner product, the decomposition of EM
l
is orthogonal
for every l.

Definition 1.4.2. Let l ∈ Z. The Hodge ∗−operator of index l is the linear

map
l
^ 2n−l
^
∗ ∗
∗: TM −→ TM
defined by the following formula: for every point m ∈ M , and every α, β we
have α ∧ ∗β = (α, β)volh (m).

The Hodge ∗−operators induce linear operators

p,q n−p,n−q
∗ : EM −→ EM , ∗ : EM
l
−→ EM
2n−l
.

The corresponding conjugate operators are

p,q n−q,n−p
∗ : EM −→ EM , ∗(α) := ∗(α).
p,q
As we have inner products on EM l
and EM , one can define a formal adjoint
∗
operator to the differentials d, ∂ and ∂, which will be denoted d∗ , ∂ ∗ and ∂ .

Definition 1.4.3. The d−Laplacian (with respect to h) is

∆d := dd∗ + d∗ d.

Analogously, one defines ∂−Laplacian ∆∂ and ∂−Laplacian ∆∂ . An m−form

(resp. (p, q)−form) α is d−harmonic (resp. ∂−harmonic, ∂−harmonic) if
∆d (α) = 0 (resp. if ∆∂ (α) = 0, ∆∂ (α) = 0). The complex vector space
of d−harmonic (resp. ∂−harmonic) m−forms (resp. (p, q)−forms) is denoted
Harmm (M ) (resp. Harmp,q (M )).
p,q
Now, assume that the complex vector spaces EM (M ) are finite dimensional
(the infinite dimensional case is more complicated, but the results still hold),
and consider the complex
p,q−1 p,q
p,q−1 ∂
p,q p,q+1 ∂
EM M
(M ) −→ EM M
(M ) −→ EM (M ).

By linear algebra one gets a direct sum decomposition

p,q p,q p,q∗ p,q−1 p,q−1∗
EM (M ) = (ker(∂ ) ∩ ker(∂ )) ⊕ im(∂ ) ⊕ im(∂ ).
Sheaves and cohomology 37

Notice that the first summand consists of ∂−harmonic forms, and ∆∂ is an

isomorphism on the other two direct summands. The main result of this section
are the following:

Theorem 1.4.1. Let M be a compact differentiable manifold equipped with an

Hermitian metric of dimension n, and let m ∈ N0 .

1. The complex vector space Harmm (M ) has finite dimension.

2. Let H : EMm
(M ) −→ Harmm (M ) be the orthogonal projection to the com-
plex vector space of ∂−harmonic m−forms. There is a unique linear en-
domorphism G of EM m
(M ) such that

• Harmm (M ) ⊆ ker(G);
• H + ∆∂ · G = idEM
m (M ) .

In particular there is a direct sum decomposition

∗ ∗
EM
m m
(M ) = Harmm (M ) ⊕ ∂ ∂ GEM m
(M ) ⊕ ∂ ∂GEM (M ).

3. The map H induces an isomorphism Harmm (M ) ' HdR

m
(M ).

Theorem 1.4.2. Let M be a Hermitian manifold of dimension n, and let p, q ∈

N0 .

1. The complex vector space Harmp,q (M ) has finite dimension.

p,q
2. Let H : EM (M ) −→ Harmp,q (M ) be the orthogonal projection to the
complex vector space of ∂−harmonic (p, q)−forms. There is a unique lin-
p,q
ear endomorphism G of EM (M ) such that

• Harmp,q (M ) ⊆ ker(G);
• H + ∆∂ · G = idEM
p,q
(M ) .

In particular there is a direct sum decomposition

p,q ∗
p,q ∗
p,q
EM (M ) = Harmp,q (M ) ⊕ ∂ ∂ GEM (M ) ⊕ ∂ ∂GEM (M ).

3. The map H induces an isomorphism Harmp,q (M ) ' H∂p,q (M ).

Corollary 1.4.3. Let M be a Hermitian manifold and p, q ∈ N0 . Then the

complex vector spaces H p (M, ΩqM ) are finite dimensional.

Definition 1.4.4. Let M be a Hermitian manifold and p, q ∈ N0 . The (p, q)−th

Hodge number of M is

hp,q (M ) := dimC (H p (M, ΩqM )).

38 Background material

Corollary 1.4.4. (Serre Duality). Let M be a Hermitian manifold of dimen-

sion n, p, q ∈ N0 . Then

1. There is an isomorphism H n (M, ΩnM ) ' C, so that hn,n (M ) = 1.

2. The Hodge ∗−operator defines an isomorphism of complex vector spaces

H p (M, ΩqM (E)) −→ H n−p (M, Ωn−q ∗ ∗

M (E )) .

In particular hp,q (M ) = hn−p,n−q (M ).

p,q
Notice that as p+q > 2n we have EM (M ) = 0, we can consider the following,
called Hodge diamond of M :
hn,n (M )
hn,n−1 (M ) hn−1,n (M )
...
n,0 n−1,1
h (M ) h (M ) ... h1,n−1 (M ) h0,n (M )
...
h1,0 (M ) h0,1 (M )
h0,0 (M )
Notice that by Serre Duality, one just needs to known the lower half of the
Hodge diamond in order to determine it completely.

1.4.2 Kähler manifolds and Hodge decomposition

If one restricts to particular families of Hermitian manifolds, then one can get
more informations about the Hodge numbers, which will play a fundamental
role in the classification of surfaces.

Definition 1.4.5. Let M be a compact complex manifold. A Hermitian metric

h on M is called Kähler metric if the associated real (1, 1)−form ωh is closed. A
compact complex manifold admitting a Kähler metric is called Kähler manifold.

First of all, let me give some examples of important Kähler manifolds.

Example 1.4.1. Let M be a Riemann surface. Then, any Hermitian metric h on
M is Kähler: this is clear, as ωh is a (1, 1)−form over a 1−dimensional manifold,
so it is closed.
Example 1.4.2. Let n ∈ N, and set for every z ∈ Cn+1 \ {0}:
1
h(z) := ∂∂ log k z k2 .
2π
It is easy to verify that h is an Hermitian metric which is C∗ −invariant. In
particular, this induces an Hermitian metric on PnC , which is still denoted h.
One can even show that h is a Kähler metric, called Fubini-Study metric.
Sheaves and cohomology 39

Example 1.4.3. Let M be a Kähler manifold, and N an analytic submanifold of

M . Then N is Kähler: indeed, any Hermitian metric h on M clearly defines an
Hermitian metric h|N on N . If the induced (1, 1)−form ωh is closed, then ωh|N
is closed, hence N is Kähler. In particular, every projective manifold is Kähler.
The reason why Kähler manifolds are so important is the following: let M
be a Kähler manifold with Kähler metric h, and let

p,q p+1,q+1
Lp,q : EM −→ EM , L(α) := ωh ∧ α.

p+1,q+1 p,q
Moreover, let Λp,q : EM −→ EM be its formal adjoint.

Theorem 1.4.5. (Kähler identities). Let M be a Kähler manifold, and let

p, q ∈ Z. The following identities are verified:
p,q
∂ p,q∗ = i[Λp,q , ∂ ],

p,q∗
∂ = −i[Λp,q , ∂ p,q ].

From these identities, one has the following:

Corollary 1.4.6. Let M be a Kähler manifold. Then ∆d = 2∆∂ . In particular,

the Laplacian is real, preserves type (i. e. the Laplacian of a (p, q)−form is still
a (p, q)−form) and preserves ∂−harmonic forms.

Proof. The proof is easy: by Theorem 1.4.5 we have

∗ ∗
i(∂ ∂ + ∂ ∂) = ∂[Λ, ∂] + [Λ, ∂]∂ = 0.
∗ ∗
Hence ∂ ∂ + ∂ ∂ = 0. Then
∗ ∗
∆d = (∂ + ∂)(∂ ∗ + ∂ ) + (∂ ∗ + ∂ )(∂ + ∂) = ∆∂ + ∆∂ .

But now
i∆∂ = −∂[Λ, ∂] − [Λ, ∂]∂ = ∂[Λ, ∂] + [Λ, ∂]∂ = i∆∂ ,

so that ∆d = 2∆∂ .

As a corollary one gets:

Theorem 1.4.7. (Hodge Decomposition). Let M be a Kähler manifold, and

let k ∈ Z. Then we have the following direct sum decomposition
M
k
HdR (M ) ⊗ C = H∂p,q (M ).
p+q=k

Moreover, for every p, q ∈ Z we have H∂p,q (M ) = H∂q,p (M ).

40 Background material

Proof. As the Laplacian preserves type, we have

M
Harmk (M ) = Harmp,q (M ).
p+q=k

From the Hodge Theorems 1.4.1 and 1.4.2 we get the first statement. As the
Laplacian is real, the last statement follows.

As a conclusion, for Kähler manifolds one has hp,q = hq,p , so that the Hodge
diamond is known if one knows hp,q for 0 ≤ p ≤ n and 0 ≤ q ≤ p. Moreover,
k
recall that by the de Rham Theorem we have HdR (M ) ' H k (M, R), so that
tensoring with C we finally have
M
H k (M, C) ' H∂p,q (M ).
p+q=k

It even follows that for every k ∈ Z we have

X
bk (M ) = hp,q (M ).
p+q=k

As a simple corollary we have then:

Corollary 1.4.8. Let M be a Kähler manifold. Then b1 (M ) is even.

Proof. As M is a Kähler manifold, we have h1,0 (M ) = h0,1 (M ) and b1 (M ) =

h1,0 (M ) + h0,1 (M ), and we are done.

In particular, using this we are finally able to show that there are compact
complex surfaces which are not Kähler, hence non projective: one example
is given by the Hopf surface X described in section 1. Topologically, X is
homeomorphic to S 1 × S 3 , so that

b1 (X) = b1 (S 1 ) + b1 (S 3 ).

by the Künneth Formula. But b1 (S 1 ) = 1 and b1 (S 3 ) = 0, so that b1 (X) = 1,

which is not even. By Corollary 1.4.8 we can then conclude that X is not Kähler.
Chapter 2

Line bundles

In this chapter we introduce the most important tool in the study of the ge-
ometry of surfaces: line bundles and their properties. First, I will recall the
notion of Picard group and of Néron-Severi group of a surface: the first is the
group of isomorphism classes of line bundles on a surface, with respect to the
composition law given by the tensor product; the second is strictly related to
the Picard group, and we are going to see various definitions of it.
Then I will introduce the notion of divisor on a surface. There two ways of
doing this: one is the notion of Weil divisor, which is simply a formal sum of
curves; the other one is the notion of Cartier divisor, namely the locus of zeroes
and poles of a meromorphic function. These two notions are strictly related: if
the surface is smooth, they are equivalent. Moreover, these are strictly related
to line bundles: to any line bundle one can associate a Cartier, or Weil divisor,
and conversely.
The next step will be to introduce basic facts about intersection theory on
surfaces, very ampleness, ampleness, nefness and bigness. Among these we are
going to define the nef cone, the ample cone and we are going to state the
Kleiman Criterion for ampleness and nefness, the Hodge Index Theorem, the
Kodaira Vanishing Theorem, the Riemann-Roch Theorem, the Noether Formula
and so on. These are very fundamental tools in the theory of surfaces.

2.1 The Picard group

As seen in the first chapter, line bundles can be viewed in two equivalent ways:
vector bundles of rank 1 or locally free sheaves of rank 1. We are going to
switch from one point of view to the other. Moreover, as our goal is to work
on projective surfaces, we are not going to specify if a line bundle is considered
to be holomorphic or algebraic, this would be clear from the setting. In this
section, X will be a (not necessarily projective) complex surface, and OX will

41
42 Line bundles

be the structural sheaf.

A line bundle L on X is then a locally free OX −module of rank 1, and
if L 0 is another line bundle, we can define their tensor product L ⊗OX L 0 .
This is again a locally free OX −module (as easily seen) whose rank is clearly 1.
Consider the following set

P ic(X) := {L , |, L is a line, bundle}/ ',

i. e. the set of isomorphism classes of line bundles on X: this is closed under

tensor product of OX −modules.

Proposition 2.1.1. For every X, P ic(X) is an abelian group with respect to

the tensor product.

Proof. As P ic(X) consists of isomorphism classes of line bundle, it is easy to see

that ⊗OX is associative and commutative. Moreover, OX is a line bundle, and
for every L ∈ P ic(X) we have L ⊗OX OX ' OX ⊗OX L ' L , i. e. P ic(X) is
an abelian modoid. In order to show that P ic(X) is an abelian group, we only
need to provide an inverse element to any line bundle L : indeed, consider

L −1 := H omOX (L , OX ) = L ∗ .

It is easy to show that L −1 is a locally free OX −module of rank 1, and we have

L −1 ⊗OX L ' H omOX (L , L ) = OX .

Definition 2.1.1. For every ringed space (X, OX ), the group P ic(X) is called
Picard group.

Before going on with the properties of the Picard group, let me introduce
some notation that will be useful in the following. Let X be a complex manifold,
and let L be a line bundle. Then we set

hi (X, L ) := dimC H i (X, L ).

Definition 2.1.2. The Euler characteristic of L is

dim(X)
X
χ(X, L ) := (−1)i hi (X, L ).
i=0

A very useful theorem that we are going to use several times in the following,
and that we present here without proof, is the following:
The Picard group 43

Theorem 2.1.2. (Serre duality). Let X be a smooth projective manifold,

and let L ∈ P ic(X). Then H n (X, OX (KX )) is a 1−dimensional vector space.
Moreover, for every 0 ≤ i ≤ n there is a perfect pairing

H i (X, L ) ⊗ H n−i (X, L −1 ⊗ OX (KX )) −→ H n (X, OX (KX )) ' C.

In particular hi (X, L ) = hn−i (X, L ⊗ OX (KX )) for every i.

One of the main properties of the Picard group of the ringed space (X, OX )
is that it has a purely cohomological description in terms of Čech cohomology.
We have indeed the following:

Proposition 2.1.3. Let (X, OX ) be a ringed space. Then there is an isomor-

phism of groups P ic(X) ' H 1 (X, OX
∗
).

Proof. To prove this, let L be a line bundle on X. As this is a locally free

OX −module of rank 1, there is an open cover U = {Ui }i∈I of X such that
for every i ∈ I there is an isomorphism ϕi : OUi −→ L|Ui . For every i, j ∈ I
let ϕij := ϕ−1i ◦ ϕj , which is an automorphism of OUij , i. e. an element
in OX (Uij ). Notice that for every i, j, k ∈ I we have ϕii = id, ϕij = ϕ−1
∗
ji
and ϕij ◦ ϕjk ◦ ϕki = id, i. e. the family {ϕij }i,j∈I defines a 1−cocycle: its
cohomology class defines then an element in H 1 (X, OX ∗
). It is moreover easy to
see that if L and M are isomorphic line bundles, their corresponding classes
in H 1 (X, OX
∗
) are equal. In conclusion, we defined a group morphism

P ic(X) −→ H 1 (X, OX
∗
).

We need to show that this is an isomorphism. For the injectivity, let us suppose
that the class corresponding to L is trivial: without loss of generality we can
suppose ϕij = id for every i, j ∈ I. But this implies that ϕi|Uij = ϕj|Uij : we can
then glue them together, getting an isomorphism ϕ : OX −→ L , and injectivity
is shown.
For surjectivity, let γ = {ϕij }i,j∈I be a 1−cocyle, where ϕij : OUij −→ OUij
is an isomorphism. As it is a 1−cocyle, for every i, j, k ∈ I we have ϕii = id,
ϕij = ϕ−1ji and ϕij ◦ ϕjk ◦ ϕki = id. We can then glue together the sheaves
OUi along this gluing datum, getting a sheaf L such that for every i ∈ I we
have L|Ui ' OUi via some isomorphism ϕi . Then L is a line bundle whose
corresponding class in H 1 (X, OX ∗
) is the class of γ.

Notice that if X, Y are two complex spaces and f : X −→ Y is a map

between them, if L is a line bundle on Y , then f ∗ L is a line bundle on X. We
have then
f ∗ : P ic(Y ) −→ P ic(X),

which is a group morphism.

44 Line bundles

An important application of this description is via the exponential sequence,

i. e. the short exact sequence of sheaves

0 −→ ZX −→ OX −→ OX
∗
−→ 0.

It induces a long exact sequence in cohomology

c
· · ·H 1 (X, OX ) −→ H 1 (X, OX
∗ 1
) −→ H 2 (X, Z) −→ H 2 (X, OX ) · ··

By Proposition 2.1.3, we have then the exact sequence

c
· · ·H 1 (X, OX ) −→ P ic(X) −→
1
H 2 (X, Z) −→ H 2 (X, OX ) · ·· (2.1)

Definition 2.1.3. The morphism c1 in the exact sequence (2.1) is called first
Chern class.

Let us see how one can apply this sequence:

Example 2.1.1. If X is a connected curve, then H 2 (X, OX ) = 0 and H 2 (X, Z) '
Z. Choosing a generator for H 2 (X, Z), we can consider then H 2 (X, Z) = Z, and
c1 : P ic(X) −→ Z is rather called the degree, and denoted deg.
Example 2.1.2. If X = Pn , then one has H 2 (Pn , OPn ) = H 1 (Pn , OPn ) = 0 by
Proposition 1.3.7. Then c1 is an isomorphism. In the next section we prove that
P ic(Pn ) ' Z: this will imply even b2 (Pn ) = 1.
Example 2.1.3. More generally, if X is simply connected then c1 is injective:
indeed, in this case one has H 1 (X, OX ) = 0. Similarily, if H 2 (X, OX ) = 0, then
c1 is surjective.
There is a more intrinsec description of the first Chern class of a line bundle
on a complex manifold X. Let L be a holomorphic line bundle, and let h be
an Hermitian metric on L . Then one can define the following (1, 1)−form

i
c(L , h) := − ∂∂ log h.
2π

This is a ∂−closed (1, 1)−form on X. One of the most important properties of

Hermitian metrics on a holomorphic line bundle asserts that if h and h0 are two
such metrics, then there is a holomorphic function f such that h0 = hef : then
it is easy to see that
c(L , h0 ) = c(L , h) + d(∂f ).

In conclusion, the cohomology class of c(L , h) does not depend on the chosen
Hermitian metric, so that one can associate an element c(L ) ∈ H∂2 (X) to any
line bundle L . Now, recall that there is an isomorphism between H∂2 (X) and
H 2 (X, C), so that one gets finally an element c(L ) ∈ H 2 (X, C). The main
property of this class is the following:
Cartier and Weil divisors 45

Theorem 2.1.4. Let X be a complex manifold, and let L be a holomorphic

line bundle on X. Then c(L ) ∈ H 2 (X, Z) and c(L ) = c1 (L ). In particular
c1 (L ) ∈ H 2 (X, Z) ∩ H 1,1 (X).

Now, let P ic0 (X) := ker(c1 ). Using the exponential sequence, it is easy to
see that
H 1 (X, OX )
P ic0 (X) ' .
im(H 1 (X, Z))
Notice that H 1 (X, OX ) is a complex vector space, and the image of H 2 (X, Z)
is a lattice in it. One can prove that it is always a maximal rank lattice:
indeed, by the Hodge theory recall that H 1 (X, C) = H 1,0 (X) ⊕ H 0,1 (X). So,
if [α] ∈ H 1 (X, OX ) is the class of a (1, 0)−real form, then [α] = [α]. Then, two
of these classes are linearily independent over C if and only if they are linearily
independent over R. This implies that im(H 1 (X, Z)) is a maximal rank lattice
in H 1 (X, OX ). In conclusion, P ic0 (X) has the structure of a complex torus.

Definition 2.1.4. The variety P ic0 (X) is called Picard variety of X. If X is a

curve, then P ic0 (X) is the Jacobian of X.

Definition 2.1.5. The group N S(X) := P ic(X)/P ic0 (X) is the Néron-Severi
group of X.

By the previous Theorem we have N S(X) ⊆ H 2 (X, Z) ∩ H 1,1 (X). We will

return on the Néron-Severi group of surfaces in the following sections.

2.2 Cartier and Weil divisors

In this section we introduce two different notions of divisor on a complex man-
ifold. The first one is due to Cartier, and is more recent than the second one,
due to Weil. On a singular variety, these two notions can be different: gener-
ally, equivalence classes of Cartier divisors are in bijective correspondence with
line bundles; this is not true for Weil divisors, since any line bundle defines a
Weil divisor but the inverse statement is generally not true. However, as we are
usually going to consider only smooth projective surfaces, these notions will be
equivalent, giving different interpretation of the notion of line bundle.

2.2.1 Cartier divisors

The notion of Cartier divisor was introduced in the 1950s by Pierre Cartier.
Let X be any complex variety (or manifold), and let OX be the sheaf of regular
functions on it.

Definition 2.2.1. A Cartier divisor on X is a global section of the sheaf

MX∗ /OX
∗
.
46 Line bundles

Using Čech cohomology one can represent a Cartier divisor in the following
way: let U = {Ui }i∈I be an open cover of X, and for every i ∈ I let fi ∈ MX∗ (Ui )
(i. e. fi is a non-trivial meromorphic function on Ui if X is a complex manifold,
and fi is a rational function on Ui if X is algebraic). In order to define a Cartier
divisor, we need the following property to be verified: for every i, j ∈ I, we have
fi /fj ∈ OX ∗
(Uij ).
In other words, a Cartier divisor is locally the locus of zeroes and poles
of some meromorphic functions. Notice that the set of Cartier divisors has a
natural structure of abelian group.
Notice that we have a short exact sequence of sheaves

0 −→ OX
∗
−→ MX∗ −→ MX∗ /OX
∗
−→ 0,

inducing the following exact sequence

α
H 0 (X, MX∗ ) −→ H 0 (X, MX∗ /OX
∗
) −→ H 1 (X, OX
∗
) = P ic(X). (2.2)

In particular, then, to every Cartier divisor one can associate an isomorphism

class of line bundles on X. Let us see the map α more explicitely: let D =
{Ui , fi }i∈I be a Cartier divisor. Then consider the family {ϕij }i,j∈I where
ϕij := fi /fj . As D is a Cartier divisor, we have ϕij ∈ OX∗
(Uij ). Moreover, it
−1
is easy to see that ϕii = 1, ϕij = ϕji and ϕij ϕjk ϕki = 1, so that {ϕij }i,j∈I is
a 1−cocycle whose class is an element of H 1 (X, OX ∗
). The corresponding line
bundle is denoted OX (D), and it is called line bundle associated to the Cartier
divisor D. We can be more explicit in the definition of OX (D): if U ⊆ X is
open, we have

OX (D)(U ) = {f ∈ MX (U ) | f · fi ∈ OX (U ∩ Ui ) ∀ i ∈ I s. t. U ∩ Ui 6= ∅}.

Indeed, for every i ∈ I the map

·1/fi
OUi −→ OX (D)|Ui

is well-defined and it is an isomorphism. Hence OX (D) is a line bundle and a

trivialization is given. The corresponding cycle is then clearly {fi /fj }i,j∈I .

Definition 2.2.2. A Cartier divisor is principal if it is in im(H 0 (X, MX∗ )). Two
Cartier divisors are linearily equivalent if their difference is a principal Cartier
divisor.

We can then define

H 0 (X, MX∗ /OX∗
)
Car(X) := .
im(H (X, MX ))
0 ∗

Notice that the exact sequence (2.2) gives then an injection

α : Car(X) −→ P ic(X).
Cartier and Weil divisors 47

In many situations, this map is an isomorphism; however, there is an example

due to Kleiman where this is not the case. Nakai shows that α is an isomorphism
whenever X is projective. We can show the following:

Proposition 2.2.1. If X is irreducible, then α : Car(X) −→ P ic(X) is an

isomorphism.

Proof. The main remark is that if X is irreducible, then MX∗ is simply the
constant sheaf k(X)∗X . As this sheaf is flasque, then H 1 (X, MX∗ ) = 0, and the
map α is surjective.

2.2.2 Weil divisors

The notion of Weil divisor is born first in algebra (this is why we still use the
name divisor), where it was introduced by Kummer, and only some decades
later was generalized by André Weil to algebraic varieties. Their nature is
hence strictly algebraic, and we define them only on algebraic varieties having
particular properties (via the GAGA Theorem one can rewrite all this section
in analytic terms). In this section, X will be an integral projective variety
which is regular in codimension 1, i. e. the local ring OX,x is regular for every
point x ∈ X of codimension 1 (which means that the Krull dimension of the
commutative ring OX,x is 1).

Definition 2.2.3. A prime divisor of X is a closed irreducible subvariety of

X of codimension 1. The set of prime divisors of X is denoted P (X). A Weil
divisor is an element in the free abelian group Div(X) = ZP (X) generated by
P (X).

In particular, a Weil divisor is written in the form

X
D= nY Y,
Y ∈P (X)

where nY ∈ Z for every Y ∈ P (X), and nY = 0 for all but a finite number of
prime divisors.
If Y ⊆ X is a prime divisor, then Y is irreducible, and it has exactly one
generic point ηY . The local ring OX,ηY is a 1−dimensional commutative local
ring (since the codimension of Y is 1), hence it is regular by our assumptions.
By basic commutative algebra, one can show that hence OX,ηY is a discrete
valuation ring whose quotient field is k(X). The discrete valuation is

vY : k(X)∗ −→ Z.

It is not difficult to show that if f ∈ k(X)∗ , then there are at most a finite
number of prime divisors Y ∈ P (X) such that vY (f ) 6= 0. Then, the following
48 Line bundles

definition makes sense:

X
(f ) := vY (f )Y.
Y ∈P (X)

Definition 2.2.4. A Weil divisor D is principal if there is f ∈ k(X)∗ such

that D = (f ). Two Weil divisors D1 , D2 are linearily equivalent, and we write
D1 ∼ D2 , if their difference is a principal Weil divisor.

Let us define
A1 (X) := Div(X)/ ∼ .

This is an abelian group which is strictly related to the Picard group and to
Car(X). Let us begin with this second relation: it is easy to associate to a
Cartier divisor {Ui , fi }i∈I a Weil divisor. Indeed, let
X
D := vY (fi(Y ) )Y,
Y ∈P (X)

where i(Y ) ∈ I is chosen among all those i ∈ I such that Ui ∩ Y 6= ∅. Clearly,

there can be many of these: however, if i, j ∈ I are such that Uij ∩ Y 6= ∅, then
fi /fj ∈ OX
∗
(Uij ), hence vY (fi ) = vY (fj ). In conclusion, D is a well-defined Weil
divisor.
Moreover, if the initial Cartier divisor is principal, then D is principal. In-
deed, there is f ∈ k(X)∗ such that fi = f|Ui , and vY (fi ) = vY (f ), so that
D = (f ). We have then a group morphism

β : Car(X) −→ A1 (X).

Composing with the previous α−1 , we get a map

β ◦ α−1 : P ic(X) −→ A1 (X).

Proposition 2.2.2. If X is regular, then β : Car(X) −→ A1 (X) is an isomor-

phism.

More generally, one has that if X is normal and integral, then β is injective.
In order to have β an isomorphism, we can only suppose X to be locally factorial,
i. e. the local ring OX,x is a unique factorization domain for every point x ∈ X.
In general, however, if X is not regular then β can be not surjective.
Example 2.2.1. If X is a Riemann surface, then a prime divisor on X is simply a
P
point. Hence a Weil divisor is written as D = P ∈X nP P , where nP ∈ Z is zero
for all but a finite number of points. As a Riemann surface is smooth, we have
that A1 (X) ' P ic(X), and in this form one can even give an interpretation
of the degree of a line bundle: if L ∈ P ic(X), then there is a Weil divisor
D ∈ A1 (X) such that L ' OX (D). Let D = P ∈X nP P : then deg(L ) =
P
P
P ∈X nP .
Cartier and Weil divisors 49

Example 2.2.2. Let us calculate A1 (Pn ). Let D = Y ∈P (Pn ) nY Y be a Weil

P

divisor in Pn . As Y ∈ P (Pn ), we have that Y is an irreducible hypersurface in

Pn , i. e. it is the zero locus of an irreducible homogenous polynomial gY (more
precisely, Y = V (I) where I is a prime ideal generated by an element gY ). We
can then define the degree of Y as deg(Y ) := deg(FY ). Using this, we can define
the following group morphism
 X  X
n
deg : Div(P ) −→ Z, deg nY Y := nY deg(Y ).
Y ∈P (Pn ) Y ∈P (Pn )

Then, it is easy to see that deg is well-defined on A1 (Pn ): let f be a rational

function on Pn , so that there are p, q ∈ C[z0 , ..., zn ] homogenous polynomials
of the same degree such that f = p/q. If Y ∈ P (Pn ), we have then vY (f ) =
vY (p) − vY (q) = 0 as p and q have the same degree, so that (f ) = 0. In
conclusion, we have a group morphism

deg : A1 (Pn ) −→ Z.

We show that this is an isomorphism: let g ∈ C[z0 , ..., zn ] be a homogenous

polynomial of degree d. We can write it (in a unique way up to a constant) as
n
g = g1n1 ...gp p , where gi is irreducible and ni ∈ N for every i = 1, ..., p. Then
V (gi ) = Yi is a prime divisor, i. e. an hypersurface of degree di , and we have
Pp
d = i=1 ni di . Let
Xp
(g) := ni Yi .
i=1
P
This is an effective Weil divisor of degree d. Now, let D = Y ∈P (X) nY Y be
any Weil divisor of degree d, and write it as D = D+ − D− , where D+ =
P P
Y ∈P (X), nY >0 nY Y and D− = − Y ∈P (X), nY <0 nY Y . If d+ := deg(D+ ) and
d− = deg(D− ), we have that D+ and D− are two effective Weil divisors of
respective degree d+ and d− and d = d+ − d− . Let now Y = V (gY ), where
gY is an homogenous polynomial, and let g+ = Y ∈P (X), nY >0 gYnY and g− =
Q
Q −nY
Y ∈P (X) gY . We have then D+ = (g+ ) and D− = (g− ). Finally, let H be
the hyperplane whose equation is z0 = 0, and let f := g+ /z0d g− . Then it is easy
to see that f is a rational function on Pn and that

D − dH = (f ).

In conclusion, if D is a Weil divisor of degree d, we have that D ∼ dH. Now,

simply notice that deg(H) = 1, so that deg is an isomorphism.

We can be then more precise about the relationship between Weil divisors
P
and line bundles. Let D be a Weil divisor, and write D = Y ∈P (X) nY Y . Let
us suppose that D is actually Cartier, so that it corresponds to a line bundle
50 Line bundles

OX (D). Let us represent D as a Cartier divisor by {Ui , fi }i∈I . Recall that if

U ⊆ X is open, we have

OX (D)(U ) = {f ∈ MX (U ) | f · fi ∈ OX (U ∩ Ui ) ∀ i ∈ I s. t. U ∩ Ui 6= ∅}.

Now, let Y ∈ P (X) be such that Y ∩ U 6= ∅, and let i ∈ I be such that

Ui ∩U ∩Y 6= ∅. As f ·fi ∈ OX (Ui ), we have vY (f ·fi ) > 0, so that vY (f ) > vY (fi ).
Now, recall that vY (fi ) = nY , so that we have finally

OX (D)(U ) = {f ∈ MX (U ) | vY (f ) > nY ∀ Y ∈ P (X) s. t. U ∩ Y 6= ∅}. (2.3)

Now, let us introduce the following definition:

P
Definition 2.2.5. A Weil divisor D = Y ∈P (X) nY Y is effective, and we write
D ≥ 0, if nY ≥ 0 for every Y ∈ P (X). If D1 , D2 are two Weil divisors on X,
we write D1 ≥ D2 if and only if D1 − D2 is effective.

Using this, we can describe the space of global sections of the line bundle
associated to D: we have

OX (D)(X) = {f ∈ k(X) | (f ) ≥ −D}. (2.4)

In other words, if L is a line bundle and D is an associated Weil divisor, then

the space H 0 (X, L ) is the space of all effective divisors D0 which are linearily
equivalent to D, i. e. all Weil divisors associated to L . This will be used in
the following.
Remark 2.2.1. Let X be a smooth projective variety, and let Y be a prime divisor
on X. As Y is an irreducible subvariety of X of codimension 1, it defines in a
natural way an element [c(Y )] ∈ H2n−2 (X, Z). By linearity, every Weil divisor
−1
D on X defines a class [c(D)] ∈ H2n−2 (X, Z), whose Poincaré dual DM ([c(D)])
2 −1
is in H (X, Z). Then it is possible to show that c1 (OX (D)) = DM ([c(D)]).
Let me now state some interesting properties of Weil divisors.

Proposition 2.2.3. Let X be a projective variety which is regular in codimen-

sion 1, and let Z ⊆ X be a closed subset whose complementary open subset is
U . Let i : U −→ X be the inclusion of U in X.

1. If codimX (Z) ≥ 2, then i induces an isomorphism between A1 (X) and

A1 (U ).

2. If codimX (Z) = 1, then we have an exact sequence

z i∗
Z −→ A1 (X) −→ A1 (U ) −→ 0,

where z(1) is the class of Z in A1 (X).

Cartier and Weil divisors 51

This has a very important consequence: let us suppose that X is a normal

projective variety. The normality assumption implies that codimX (Xsing ) ≥ 2,
so that by the previous Proposition we have A1 (X) = A1 (X s ), where X s :=
X \ Xsing . As X s is smooth, by Proposition 2.2.2 we have A1 (X s ) ' P ic(X s ):
in conclusion, we have A1 (X) ' P ic(X s ), so that every Weil divisor on X is
determined by a line bundle on X s .
In particular this allows us to define a canonical Weil divisor on every nor-
mal projective variety. Recall that the canonical line bundle is OX (KX ) :=
Vdim(X)
ΩX . If X is smooth, then ΩX is a vector bundle of rank dim(X), so
that OX (KX ) is a well-defined line bundle. If X is singular, this is no longer
true, so we have no canonical line bundle. Anyway, on X s we have OX s (KX s ),
which is a line bundle: if X is normal, this defines a natural Weil divisor KX ,
which is called canonical Weil divisor of X.
To conclude this section, let me make the following observation: if D is
any Weil divisor (not necessarily Cartier), equation (2.3) still makes sense. In
particular, then, we can associate an OX −module OX (D) to every Weil divisor
D on X. If D is Cartier, then OX (D) is known to be a line bundle; otherwise,
it still has some interesting properties.

Definition 2.2.6. A divisorial sheaf on X is a coherent OX −module which

verifies the three following properties:

1. L has no torsion;

2. rk(L ) = 1;

3. L = L ∗∗ , where L ∗ := H om(L , OX ) is the dual of L .

Theorem 2.2.4. Let X be a normal projective variety.

• Let D be a Weil divisor on X. Then OX (D) is a divisorial sheaf. More-

over, if D1 ∼ D2 are linearily equivalent Weil divisors, then OX (D1 ) '
OX (D2 ).

• Let L be a divisorial sheaf on X. Then there is a Weil divisor D on X

such that L ' OX (D).

As a conclusion of this section, we add the following theorem:

Theorem 2.2.5. (Bertini). Let X be a smooth projective variety. A generic

hypersurface of X is smooth.

Proof. Let us consider X as a d−dimensional subvariety of Pn for some n ∈ Z.

Let us consider hypersurfaces of degree k, which are obtained as H ∩ X, where
H is a hypersurface of degree k in Pn . The family of these hypersurfaces is
|OPn (k)|, which is a projective space of dimension nk . Let

BX,k := {(x, H) ∈ X × |OPn (k)| | Tx (X) ⊆ Tx (H)},

52 Line bundles

which is the set of couples (x, H) such that H is not transversal to the pro-
jectivized tangent space of X at x. As dim(X) = d, we have that P(Tx (X))
is has dimension d − 1, and the possible hyperplanes containing it form a pro-
jective space of dimension nk − d − 1. The canonical projection B −→ X


n
presents B as a projective bundle, whose fiber is then P(k )−d−1 . In conclusion
dim(B) = nk − 1, so that p : B −→ |OPn (k)| cannot be surjective. Let then

H ∈ |OPn (k)| \ p(B): then H ∩ X is a hypersurface of X of degree k which is

smooth at every point x ∈ X.

2.3 Ampleness and very ampleness

Let us start with the following basic definition. Let X be a projective manifold
of dimension n, and let L be a line bundle on X. As X is compact, the complex
vector space H 0 (X, L ) is finite dimensional. Let dim(H 0 (X, L ) = n + 1, and
let us suppose that this is not trivial.

Definition 2.3.1. The complete linear system associated to L is the set

|L | := {D ∈ Div(X) | D ≥ 0, OX (D) ' L }.

By equation (2.4), we have that |L | = H 0 (X, L ), so that |L | is a complex

vector space. Any linear subspace of |L | is called linear system. This allows us
to define the following:

Definition 2.3.2. The base locus of L is the set

\
Bs(L ) := D.
D∈|L |

Every point in Bs(L ) is called base point of L .

T
Using sections, we have that Bs(L ) = s∈H 0 (X,L ) V (s), i. e. it is the locus
of common zeroes of global sections of L . Now, let us consider {s0 , ..., sn } a
basis of H 0 (X, L ), and define

ϕs0 ,...,sn : X \ Bs(L ) −→ Pn , ϕs0 ,...,sn (x) := (s0 (x) : ... : sn (x)).

This map is clearly not defined on Bs(L ) and depends on the chosen basis.
Anyway, we will always write ϕL := ϕs0 ,...,sn .
Example 2.3.1. Let X be a projective manifold, and let f : X −→ Pn . If
L = f ∗ OPn (1), then Bs(L ) = ∅ and ϕL = f (for the canonical basis of
H 0 (X, f ∗ OPn (1))). Indeed, we have that h0 (Pn , OPn (1)) = n + 1 by Proposition
1.3.7, and a canonical basis is given by the homogenous coordinates z0 , ..., zn .
Now, a canonical basis for H 0 (X, L ) is given by s0 , ..., sn , where si = f ∗ zi for
every i = 0, ..., n, so that ϕL = f . If x ∈ Bs(L ), then si (x) = 0 for every
Ampleness and very ampleness 53

i = 0, ..., n: By definition, this means that zi (f (x)) = 0 for every i = 0, ..., n, i.

e. the coordinates of the point f (x) are all 0, so that Bs(L ) = ∅.

Definition 2.3.3. A line bundle L on X is globally generated if Bs(L ) = ∅.

It is called very ample if it is globally generated and the map ϕL is a closed
embedding.

Remark 2.3.1. If X is a complex manifold admitting a very ample divisor, then

X is clearly projective. We then write L = OX (1).
Remark 2.3.2. A line bundle L is globally generated if and only if the evaluation
map
ev : H 0 (X, L ) ⊗ OX −→ L , ev(s, t) := s(t),

is surjective. More precisely, if s ∈ H 0 (X, L ), then s corresponds to a morphism

s : OX −→ L , which can be evaluated on every section t of OX . The surjectivity
of ev is verified if and only if the corresponding map on stalks is surjective. If
x ∈ X, we have Lx ' OX,x , so that evx is surjective if and only if it is not
trivial. As evx (s, 1) = s(x), we have that evx 6= 0 if and only if x ∈/ Bs(L ).

Proposition 2.3.1. A line bundle L on X is globally generated (resp. very

ample) if and only if there is a map (resp. a closed embedding) f : X −→ Pn ,
where n = h0 (X, L ) − 1, such that L ' f ∗ OPn (1).

There is a very geometric criterion to test if a line bundle is very ample.

First of all, let me remark the following: a line bundle L is globally generated
if and only if ϕL is everywhere defined. By Remark 2.3.2, this means that for
every point x ∈ X there must be a section s ∈ H 0 (X, L ) such that s(x) 6= 0.
Let mx be the ideal of OX,x of those germes of functions vanishing at x: this is
the maximal ideal of OX,x . we have then the following exact sequence

0 −→ mx L −→ L −→ Lx /mx Lx −→ 0.

Clearly, there is a section s of L such that s(x) 6= 0 if and only if sx ∈

/ mx Lx ,
i. e. if and only if H 1 (X, mx L ) = 0 for every x ∈ X.

Definition 2.3.4. A line bundle L separates points if and only if for every
x 6= y ∈ X we have H 1 (X, mx · my L ) = 0. A line bundle L separates tangent
directions if and only if for every x ∈ X we have H 1 (X, m2x L ) = 0.

Proposition 2.3.2. A line bundle L is very ample if and only if it is globally

generated, it separates points and tangent directions.

Proof. If a line bundle is very ample, then clearly it is globally generated. More-
over, the morphism ϕL is an closed embedding. In particular, it is injective, so
54 Line bundles

that for every x 6= y ∈ X there must be s, t ∈ H 0 (X, L ) such that s(x) = 0,

t(x) 6= 0 and s(y) 6= 0 ad t(y) = 0. Consider the exact sequence

0 −→ mx · my L −→ L −→ Lx /mx Lx ⊕ Ly /my Ly −→ 0.

Then the two sections above exists if and only if H 1 (X, mx · my L ) = 0, i. e. if

and only if L separates points.
Moreover, ϕL is an embedding if for every x ∈ X and every cotangent vector
v ∈ Tx (X)∗ there is a section s ∈ H 0 (X, L ) such that s(x) = 0 and ds(x) = v.
Now, there is a well-defined map

dx : mx L −→ Tx (X)∗ Lx /mx Lx , dx (s) := ds(x).

Moreover we have an exact sequence

0 −→ m2x L −→ mx L −→ Tx (X)∗ Lx /mx Lx −→ 0.

Then ϕL is an embedding if and only if H 1 (X, m2x L ) = 0, i. e. if and only if

L separates tangent directions.

Example 2.3.2. As an application of the previous criterion, we can show that

if X is a Riemann surface, then X is projective. We already know that every
Riemann surface is Kähler, but here we prove something more. Let L be a
line bundle on X: as X is compact, the complex vector spaces H 0 (X, L ) and
H 1 (X, L ) are finite dimensional. Moreover, let d = deg(L) ∈ Z: if d < 0,
then H 0 (X, L ) = 0. Indeed, if H 0 (X, L ) 6= 0, there is a section s of L
corresponding to a Weil divisor D which is trivial or effective, such that L '
OX (D). Then we have deg(D) ≥ 0, so that deg(L ) ≥ 0. By Serre’s duality we
have H 1 (X, L ) ' H 0 (X, L −1 ⊗ OX (KX ))∗ . It is an easy calculation to show
that deg(KX ) = 2g(X) − 2, so that H 1 (X, L ) = 0 if deg(L ) > 2g(X) − 2.
Now, let D = (2g(X) + 1)P for some P ∈ X, and let L := OX (D). If
Q ∈ X is any point, it is easy to see that mQ = OX (−Q): then mQ L =
OX ((2g(X)+1)P −Q), so that deg(mQ L ) = 2g(X); moreover, if R ∈ X we have
mR · mQ L = OX ((2g(X) + 1)P − Q − R), so that deg(mR · mQ L ) = 2g(X) − 1.
By the previous remark we have H 1 (X, mQ L ) = 0 for every point Q ∈ X, so
that the map
ϕL : X −→ Pn

is well-defined. Moreover H 1 (X, mR · mQ L ) = 0 for every choice of R, Q ∈ X,

so that L is very ample and ϕL is a closed embedding. In conclusion, every
Riemann surface is projective.
Now, we introduce one of the main definitions in the theory of line bundles.

Definition 2.3.5. A line bundle L on X is ample if there is n ∈ N such that

L ⊗n is very ample.
Ampleness and very ampleness 55

Remark 2.3.3. By Proposition 2.3.1, a line bundle is ample if and only if there
is a closed embedding f : X −→ PN , for some N ∈ N, and n ∈ N such
that L ⊗n ' f ∗ OPn (1). Clearly, if X admits an ample line bundle, then X is
projective, and the embedding is given by ϕL ⊗n .

Remark 2.3.4. Notice that if two line bundles L and M are ample, then L ⊗M
is ample.

To conclude this section, we show the following important criterion for am-
pleness:

Theorem 2.3.3. Let X be a projective manifold and L be a line bundle on X.

The following are equivalent:

1. L is ample;

2. for every coherent sheaf F there is n0 ≥ 0 such that for every n > n0 we
have H i (X, F ⊗ L ⊗n ) = 0 for every i > 0;

3. for every coherent sheaf F there is m0 ≥ 0 such that for every n > m0
we have F ⊗ L ⊗n is globally generated;

4. there is p0 ≥ 0 such that L ⊗m is ample for every m > p0 .

Proof. The fact that 1 implies 2 is simply a corollary of Serre’s Vanishing: if

L is ample, then L ⊗k is very ample for some k ∈ N. Then by Proposition
2.3.1 there is a closed embedding f : X −→ Pn , for some n ∈ N, such that
L ⊗k ' f ∗ OPn (1). By Serre’s Vanishing, for every coherent sheaf F there is
l  0 such that H i (X, F ⊗ L ⊗lk ) = 0 for every i > 0. Now, let n ∈ B, and
suppose n = lk + r, for 0 ≤ r < k: then we have

H i (X, F ⊗ L ⊗lk+r ) = 0

for every i > 0, and we are done.

The fact that 2 implies 3 is easy: to show that F ⊗L m is globally generated,
we just need to show that for every point x ∈ X we have H 1 (X, mx F ⊗L ⊗m ) =
0. As mx F is coherent, by assumption this is the case for m  0. By compact-
ness of X, we finish the proof.
The fact that 3 implies 1 is more complicated.

As a corollary to this, we have the following:

Corollary 2.3.4. Let X, Y be two projective manifolds, and let f : X −→ Y be

a finite morphism. If L is an ample line bundle on Y , then f ∗ L is an ample
line bundle on X.
56 Line bundles

Proof. Let F be a coherent sheaf on X. As f is finite, the sheaf f∗ F is coherent,

and we have H i (Y, f∗ F ⊗L ⊗n ) = 0 for i > 0 and n  0. By projection formula,
we have f∗ F ⊗ L n ' f∗ (F ⊗ f ∗ L n ), so that

H i (X, F ⊗ f ∗ L ⊗n ) ' H i (Y, f∗ (F ⊗ f ∗ L ⊗n )) = 0

for i > 0 and n  0, as f is a finite morphism.

There is another important theorem we need to state:

Theorem 2.3.5. (Kodaira’s Vanishing). Let X be a smooth projective va-

riety of dimension n, and let p, q ∈ Z. If L is ample line bundle on X, then

H q (X, L ⊗ ΩpX ) = 0

for every p + q > n. In particular H i (X, L ⊗ OX (KX )) = 0 for every i > 0.

The Kodaira Vanishing Theorem holds in greater generality: one can get
rid of the projectivity of M , and consider any complex manifold. To do so, one
would have to introduce the notion of positive line bundle, which is the complex
analogue of ample line bundle. There are other various generalizations of this
theorem, but we will not be concerned with them.

2.4 Intersection theory on surfaces

From now on, X will be a smooth projective surface. We introduce here one
of the basic element in the theory of classification of surfaces: the intersection
theory of divisors on a surface. Notice that if X is a smooth projective surface,
a prime divisor is simply an irreducible curve on X, so that a Weil divisor is
simply a formal finite linear combination of irreducible curves.
We introduce the following definition:

Definition 2.4.1. Let X be a smooth projective surface, Y any smooth projec-

tive variety and f : X −→ Y a finite morphism. If C ∈ P (X) is an irreducible
curve, we define f∗ C := 0 if f (C) is a point, otherwise we define f∗ C := df (C),
where d is the degree of the map C −→ f (C) induced by f . By linearity we can
define f∗ D for every Weil divisor D on X.

Remark 2.4.1. If D is a Weil divisor on X, then D defines a line bundle OX (D):

as f is finite, we have that f∗ OX (D) is a coherent sheaf on Y , which in general
is not a line bundle. The previous definition is really defined on Weil divisors.
It is easily verified that if D ∼ D0 , then f∗ D ∼ f∗ D0 , so that f induces a map
f∗ : A1 (X) −→ A1 (Y ).
Remark 2.4.2. If D ∈ A1 (Y ), we have that f∗ f ∗ D ∼ dD.
We introduce now the basic definition for the intersection of divisors.
Intersection theory on surfaces 57

Definition 2.4.2. Let C and C 0 be two distinct irreducible curves on X, and

let x ∈ C ∩ C 0 . Let f, g ∈ OX,x be two local equations respectively for C and
C 0 . Then the intersection multiplicity of C and C 0 at x is

multx (C, C 0 ) := dimC OX,x /(f, g),

where (f, g) ⊆ OX,x is the ideal generated by f and g. If multx (C, C 0 ) = 1, we

say that C and C 0 are transverse (or meet transversally) at x. The intersection
multiplicity of C and C 0 is
X
C · C 0 := multx (C, C 0 ).
x∈C∩C 0

Remark 2.4.3. The fact that OX,x /(f, g) is a finite dimensional complex vector
space comes from the Hilbert Nullstellensatz. The intersection multiplicity is a
well-defined integer as C ∩ C 0 is a finite set.
Notice that we cannot define by linearity the intersection multiplicity of two
Weil divisors, as we do not know how to define it for C with C itself. We
use instead a different approach, following Beauville. Recall that if C is an
irreducible curve, then we have IC = OX (−C). If C and C 0 are two irreducible
curves on X, let

OC∩C 0 := OX /(OX (−C) + OX (−C 0 )).

This is a skyscraper sheaf supported on C ∩ C 0 , and it is easy to verify that if

x ∈ C ∩ C 0 we have
(OC∩C 0 )x = OX,x /(f, g),

where f and g are local equations for C and C 0 respectively. By this, it follows
that
C · C 0 = h0 (X, OC∩C 0 ).

Definition 2.4.3. Let L and M be two line bundles on X. Let

L · M := χ(OX ) − χ(L −1 ) − χ(M −1 ) + χ(L −1 ⊗ M −1 ),

which is called intersection product of L and M .

We want to prove that this is a well-defined bilinear and non-degenerated

form, so it is an intersection pairing. Let us start with the following:

Lemma 2.4.1. Let C and C 0 be two distinct irreducible curves on X. Then

OX (C) · OX (C 0 ) = C · C 0 .

Proof. The proof is rather easy: consider the following sequence

(t,−s) (s,t)
0 −→ OX (−C − C 0 ) −→ OX (−C) ⊕ OX (−C 0 ) −→ OX −→ OC∩C 0 −→ 0,
58 Line bundles

where s ∈ H 0 (X, OX (C)) is a section defining C and t ∈ H 0 (X, OX (C 0 )) is a

section defining C 0 . It is easy to verify that this sequence is exact: namely, we
need to show that for every point x ∈ C ∩ C 0 the sequence
(g,−f ) (f,g)
0 −→ OX,x −→ OX,x
2
−→ OX,x −→ OX,x /(f, g) −→ 0

is exact, where f and g are local equations for C and C 0 respectively. To do so,
we just need to check that if a, b ∈ OX,x are such that af = bg, then there is
k ∈ OX,x such that a = kg and b = kf . But this is clear: we have that f and
g are coprime (since C and C 0 are two distinct irreducible curves), and OX,x
is a unique factorization domain (as X is smooth). Now, the exactness of the
sequence implies that

χ(OX (−C − C 0 )) + χ(OX ) = χ(OX (−C) ⊕ OX (−C 0 )) + χ(OC∩C 0 ).

It is easy to show that χ(OX (−C) ⊕ OX (−C 0 )) = χ(OX (−C)) + χ(OX (−C 0 )),
and that OX (−C) = OX (C)∗ and OX (−C − C 0 ) = OX (C) ⊗ OX (C 0 ). Now,
recall that OC∩C 0 is a skyscraper sheaf, so that hi (X, OC∩C 0 ) = 0 for every
i > 0. Then χ(OC∩C 0 ) = h0 (X, OC∩C 0 ) = C · C 0 , and the proof is finished.

The next property we need to show is the following

Lemma 2.4.2. Let C be a smooth curve on X and L ∈ P ic(X). Then

OX (C) · L = deg(L|C ).

Proof. Consider the exact sequence

0 −→ OX (−C) −→ OX −→ OC −→ 0.
−1
Then χ(OC ) = χ(OX ) − χ(OX (−C)). Tensoring with L −1 we get χ(L|C )=
χ(L ) − χ(L ⊗ OX (−C)). Now, by definition we have
−1 −1

OX (C) · L = χ(OX ) − χ(OX (−C)) − χ(L −1 ) + χ(OX (−C) ⊗ L −1 ) =

−1
= χ(OC ) − χ(L|C ).
Now, use the Riemann-Roch Theorem on curves, i. e. χ(OC ) = 1 − g(C) and
χ(L|C ) = deg(L|C ) + 1 − g(C), so that

OX (C) · L = deg(L|C ),

and we are done.

We are finally able to show the following:

Theorem 2.4.3. The intersection product

· : P ic(X) × P ic(X) −→ Z

is a symmetryc bilinear form such that OX (C) · OX (C 0 ) = C · C 0 for every two

distinct irreducible curves C, C 0 on X.
Intersection theory on surfaces 59

Proof. The symmetry is obvious. Now, consider L1 , L2 , L3 ∈ P ic(X), and let

s(L1 , L2 , L3 ) := L1 · L2 ⊗ L3 − L1 · L2 − L1 · L3 .

If Li = OX (C) for some i = 1, 2, 3 and for some irreducible curve C. Then

s(L1 , L2 , L3 ) = 0 by Lemma 2.4.2, so that in this case bilinearity is shown.
For the general case, we need the following fact: let L ∈ P ic(X). Then there
is n ≥ 0 such that L ⊗ OX (n) is very ample. In particular, if D is any Weil
divisor, then there are A, B smooth curves such that D ∼ A−B (simply consider
B a smooth curve in |OX (n)| and A a smooth curve in |OX (D) ⊗ OX (n)|).
Now, let L , M ∈ P ic(X). Write M = OX (A − B) for A, B smooth curves
on X, so that
L · M = L · OX (A) − L · OX (B).

From this and Lemma 2.4.2 we deduce the linearity in L , and bilinearity is
shown. The remaining part is simply Lemma 2.4.1.

If D and D0 are two Weil divisors, let D · D0 := OX (D) · OX (D0 ). In

particular, if D = D0 , we have D2 := D · D. The previous Theorem guarantees
that if D ∼ D,e then D · D0 = De · D0 for every Weil divisor D0 . In particular, if
D0 ∼ D, we have D · D0 = D2 . Let me now give two first very useful application
of this principle.

Proposition 2.4.4. Let X and Y be two smooth projective surfaces, and let C
be a smooth curve.

1. Let f : X −→ C be a surjective morphism, and let F be the class of a

fiber. Then F 2 = 0.

2. Let g : X −→ Y be a generically finite morphism of degree d, and let D, D0

be two Weil divisors on Y . Then g ∗ D · g ∗ D0 = dD · D0 .

Proof. Let us begin with the first point. As X is a smooth surface and C is a
smooth curve, the fiber f −1 (P ) =: XP of f over the point P ∈ C is a closed
subvariety of X of codimension 1 (it is not empty as f is surjective). It is then
a Weil divisor, and we write F for its class in A1 (X). Notice that P ∈ C is a
Weil divisor on X, and OX (XP ) ' f ∗ OC (P ). The Weil divisor P is ample on
C, so that there is n ∈ N such that nP is very ample. Notice that for n  0
we have h0 (nP ) > 1: indeed, we have h0 (KC − nP ) = 0 for n  0 (since for
n  0 we have deg(KC − nP ) = 2g(C) − 2 − n < 0). By Serre duality this
implies that for n  0 we have h0 (nP ) = χ(nP ). Using Riemann-Roch for
curves, we then have h0 (nP ) = n + 1 − g(C) > 1, as n  0. Hence there
is a section s ∈ H 0 (X, OC (nP )) such that D = div(s) is an effective divisor
which is linearily equivalent to nP and P does not appear in D. Notice that if
D = Q∈C nQ Q, we have that f ∗ OC (D) ' OX ( Q∈C nQ XQ ). In conclusion
P P
60 Line bundles

P
the Weil divisor nXP is linearily equivalent to Q∈C nQ XQ , with nP = 0.
Clearly, we have XQ ∩ XP = ∅ for every Q 6= P , so that XQ · XP = 0 in this
case. Hence, notice that
X X
n2 F 2 = n2 XP2 = nXP · nQ XQ = nnQ XQ · XP = 0,
Q∈C Q6=P

and the first item is shown.

For the second point, recall that there are A, B, A0 , B 0 smooth curves on Y
such that D ∼ A − B and D0 ∼ A0 − B 0 . If the statement is true for smooth
curves, we have

g ∗ D · g ∗ D0 = g ∗ A · g ∗ A0 + g ∗ B · g ∗ B 0 − g ∗ A · g ∗ B 0 − g ∗ A0 · g ∗ B = dD · D0 .

In conclusion, we can suppose D and D0 to be smooth curves. As g is gener-

ically finite of degree d, there is an open subset U of Y such that the map
g : g −1 (U ) −→ U is finite of degree d and not ramified. Up to changing D and
D0 by linear equivalent ones, we can suppose that D and D0 meet transversally
at every point, and that D ∩ D0 ⊆ U . Then it is easy to see that

g ∗ D · g ∗ D0 = g −1 (D ∩ D0 ) = dD · D0 ,

and we are done.

I will conclude this section with two important examples.

Example 2.4.1. We can easily describe the intersection form of P2 . Indeed, we
know that P ic(P2 ) ' Z, and that a generator for it is the line bundle OP2 (1).
A Weil divisor corresponding to this line bundle is obtained as the zero locus
of a section: As H 0 (P2 , OP2 (1)) is given by homogenous polynomials of degree
1, a Weil divisor is simply a curve of degree 1 in P2 , i. e. a line L. As
L2 = L1 · L2 , where L1 and L2 are two lines in P2 , we can choose L1 and L2
meeting transversally: take Li = (zi ), where zi are homogenous coordinates on
P2 , so that
L1 · L2 = ](L1 ∩ L2 ) = 1,

as L1 ∩L2 = {(1 : 0 : 0)}. Then OP2 (1)·OP2 (1) = 1, and the intersection product
is described by linearity. As a consequence, we simply get Bezout’s Theorem:
if Ci is a curve of degree di in P2 , for i = 1, 2, then C1 · C2 = d1 d2 .
Example 2.4.2. In this example we introduce a new surface X = P1 × P1 . Define
the following curves

h1 := {(0 : 1)} × P1 , h2 := P1 × {(0 : 1)}.

Moreover, let U := X \ (h1 ∪ h2 ): it is easy to see that U ' C2 . Indeed, if

P = ((x0 : x1 ), (y0 , y1 )) ∈ U , then x0 , y0 6= 0, so that one can associate to P
Intersection theory on surfaces 61

the point (x1 /x0 , y1 /y0 ) of C2 . Conversely, to every Q = (x, y) ∈ C2 we can

associate the point ((1 : x), (1 : y)) ∈ U , and we have the isomorphism.
Now, it is a basic fact that P ic(C2 ) = 0, so that P ic(U ) = 0. Hence, if D is
a Weil divisor on X, we have that D|U = (f ) for some function f . There must
then be n1 , n2 ∈ Z such that D = (f ) + n1 h1 + n2 h2 . In conclusion, we get that

P ic(X) ' A1 (X) ' Z ⊕ Z,

and generators are the classes of h1 and h2 . The line bundle OX (D) will be
denoted OX (n1 , n2 ), and (n1 , n2 ) is called the bidegree of D.
In order to describe the intersection pairing, we then just need to know
hi · hj for every i, j. By the very definition of h1 and h2 , we have that they meet
transversally, so that

h1 · h2 = ](h1 ∩ h2 ) = ]{((0 : 1), (0 : 1))} = 1.

We then need only to determine h2i . Now, notice that the projection

p1 : X −→ P1

to the first factor is surjective and has h1 as fiber over (0 : 1). Then h21 = 0
by Proposition 2.4.4. Similarily for h2 : in conclusion, the intersection pairing is
given by h1 · h2 = h2 · h1 = 1 and h21 = h22 = 0.

2.4.1 The Riemann-Roch Theorem for surfaces

In this subsection we introduce one of the most important formulas in the whole
theory of surfaces. Recall a basic result in the theory of curves: if D is a Weil
divisor on a smooth curve C, then we have

χ(OC (D)) = deg(D) − g(C) + 1.

This is called Riemann-Roch Theorem for curves. We now want to introduce

an analogue of this for smooth projective surfaces. Here is the statement:

Theorem 2.4.5. (Riemann-Roch). Let X be a smooth projective surface,

and let L ∈ P ic(X). Then
1
χ(L ) = χ(OS ) + (L 2 − L · OX (KX )).
2
Proof. The proof uses the Serre Duality. By the very definition of the intersec-
tion pairing, we have:

L −1 · (L ⊗ OX (−KX )) = χ(OX ) − χ(L ) − χ(L −1 ⊗ OX (KX )) + χ(OX (KX )).

By Serre duality χ(OX ) = χ(OX (KX )) and χ(L ) = χ(L −1 ⊗ OX (KX )), so
that we finally get
1
χ(L ) = χ(OX ) − L −1 · (L ⊗ OX (KX )).
2
62 Line bundles

But now notice that by bilinearity of the intersection pairing we have L −1 ·

(L ⊗ OX (KX )) = −L 2 + L · OX (KX ), and we are done.

Using the additive notations of Weil divisors and the definition of χ, the
Riemann-Roch Formula becomes
1
h0 (D) − h1 (D) + h2 (D) = χ(OX ) + (D2 − D · KX ).
2
Moreover, we can use Serre’s Duality to replace h1 (D) by h1 (KX − D) and
h2 (D) by h0 (KX − D). As h1 (D) ≥ 0 for every Weil divisor D, we have the
following inequality

1
h0 (D) + h0 (KX − D) ≥ χ(OX ) + (D2 − D · KX ).
2
It is moreover interesting to note that as χ(OX (D)) is an integer, for every Weil
divisor we have D2 − D · KX ∈ 2Z.

Definition 2.4.4. The irregularity of a smooth projective surface X is q(X) :=

h1 (OX ) = h0,1 (X). The geometric genus of X is pg (X) := h2 (OX ) = h0,2 (X).

Using this classical definitions, we have

χ(OX ) = 1 − q(X) + pg (X),

and one could even state the Riemann-Roch Formula using this.
Remark 2.4.4. Notice that as X is a smooth projective surface, it has a natural
Kähler structure, so that h0,1 (X) = h1,0 (X) and h2,0 (X) = h0,2 (X). In partic-
ular, then, we see that b1 (X) = 2q(X), and that b2 (X) = 2pg (X) + h1,1 (X).
From this, it follows that q(X) is a topological invariant of any smooth projective
surface.
Example 2.4.3. If X = P2 , then we have q(X) = pg (X) = 0, so that χ(OP2 ) = 1.
The Riemann-Roch formula is then very easy:

d2 + 3d
χ(OP2 (d)) = 1 + .
2

Example 2.4.4. If X = P1 × P1 , then we have the following properties: let

pi : X −→ P1 be the projection to the i−th factor. Then it is easy to see
that pi∗ OX ' OP1 , so that by projection formula we get q(X) = pg (X) = 0.
Moreover, we have that TX ' p∗1 TP1 ⊕ p∗2 TP1 . Hence KX = OX (−2, −2). In
conclusion, we have that χ(X) = 1 and that

χ(OX (n, m)) = 1 + n + m + nm.

We conclude this section by stating two important properties more.

Intersection theory on surfaces 63

Theorem 2.4.6. (Noether’s Formula). Let X be a smooth projective for-

mula, and let e(X) be the topological Euler characteristic of X. Then
1
χ(OX ) = (K 2 + e(X)).
12 X
This formula represents a connection between topological and complex char-
acteristics of a smooth projective surface. An important consequence of the
Riemann-Roch formula for surfaces is the following:

Theorem 2.4.7. (Genus formula). Let X be a smooth projective surface,

and let C be an irreducible curve on X. The genus of C is
1
g(C) = 1 + (C 2 + C · KX ).
2
Proof. Recall that the genus of a curve is g(C) = h1 (C, OC ). Using the following
exact sequence
0 −→ OX (−C) −→ OX −→ C −→ 0

we get χ(OX ) = χ(OX (−C)) + χ(OC ). By the Riemann-Roch Formula for X

we then get
1
χ(OC ) = − (C 2 + C · KX ).
2
Now, using Riemann-Roch for C we have that χ(OC ) = 1 − g(C), and we are
done.

Remark 2.4.5. Recall that if C is a curve in X, then we have an exact sequence

0 −→ OX (−C)|C −→ ΩX|C −→ OC (KC ) −→ 0.

From this it follows easily that OC (KC ) ' OX (KX + C)|C or, written in Weil
divisor form, we have KC ∼ (KX +C)|C . This is called the Adjunction Formula.
Now, consider the degrees

2g(C) − 2 = deg(KC ) = deg((KX + C)|C ) = C 2 + KX · C,

where the last equality follows from 2.4.2. Notice that this again implies the
genus formula.
Example 2.4.5. Notice that the genus formula for X = P2 gives us the following:
if C is a smooth curve of degree d in P2 , then we have

d2 − 3d (d − 1)(d − 2)
g(C) = 1 + = ,
2 2
which is the well-known formula relating genus and degree for a smooth pro-
jective curve. If X = P1 × P1 and C is a curve of bidegree (n, m), then we
have
g(C) = 1 + nm − n − m.
64 Line bundles

2.5 Nef line bundles

In this section we introduce the notion of nefness of a line bundle, which can be
seen as a limit of ampleness. This will be made precise in the following, and in
order to do this we need to introduce some important material. First of all we
discuss the intersection pairing on the Néron-Severi group of a smooth projective
surface, and we prove that it is a finite rank lattice. We then introduce the notion
of ample cone, effective cone and nef cone, and we study their relations. We
will then prove some important criterion for ampleness and nefness (Kleiman,
Nakai-Moishezon) which will be used frequently in the following.

2.5.1 Cup product and intersection product

Let us first make some remarks on the intersection pairing. If C and C 0 are two
irreducible curves on a smooth projective surface, and we suppose them to meet
transversally, we defined
C · C 0 = ](C ∩ C 0 ).

This definition has to be modified in order to intersect curves which don’t meet
transversally or which are not irreducible One way of doing this is simply to
consider the intersection pairing OX (C) · OX (C 0 ) we defined in the previous
section. There is another way of doing this, namely using algebraic cycles: C
and C 0 are two algebraic cycles in X, which is a complex manifold of dimension
2, and their dimension is 1. Let p ∈ C ∩ C 0 , and let v ∈ Tp (C), w ∈ Tp (C 0 )
be two oriented basis for these two 1−dimensional complex vector spaces. As
C and C 0 meet transversally at p, we have that v, w is a basis for Tp (M ). We
define ip (C, C 0 ) = 1 if this is an oriented basis (with the natural orientation of
M ), otherwise ip (C, C 0 ) = −1. Then we define
X
]
(C · C 0 ) := ip (C, C 0 ).
p∈C∩C 0

The main point is that as we are on complex manifolds, then for every point
p ∈ C ∩ C 0 we have ip (C ∩ C 0 ) = +1, so that ] (C · C 0 ) = C · C 0 . Another
important point is to show is the following:

Lemma 2.5.1. Let C ∼ C e and C 0 ∼ C e 0 be homologically equivalent cycles, i.

e. [C] = [C] ∈ H2 (X, Z) and [C ] = [C ] ∈ H2 (X, Z). Then ] (C · C 0 ) =] (C
e 0 e 0 e 0 ).
e·C

As a consequence we have an intersection pairing

∩ : H2 (X, Z) × H2 (X, Z) −→ Z,

which is unimodular. This is the real content of the Poincaré duality: this
implies that for every element α ∈ H 2 (X, Z), which corresponds to a linear map
Nef line bundles 65

H2 (X, Z) −→ Z by definition, there is γ ∈ H2 (X, Z) such that α = γ∩. In this

way we can interpret the cup product

∪ : H 2 (X, Z) × H 2 (X, Z) −→ Z.

Indeed, we can show that if α, β ∈ H 2 (X, Z), then α ∪ β = DM (α) ∩ DM (β), i.

e. the intersection pairing is Poincaré-dual to the cup product.
Now, let us consider OX (D) and OX (D0 ), for D, D0 two Weil divisors on X.
Then by the properties of the intersection pairing on line bundles and the one
on cycles, we have that

OX (D) · OX (D0 ) = [D] ∩ [D0 ].

By the previous duality we have then

−1 −1
OX (D) · OX (D0 ) = DM ([D]) ∪ DM ([D0 ]).
−1
Now, by Remark 2.2.1 we have that DM ([D]) = c1 (OX (D)), so that

OX (D) · OX (D0 ) = c1 (OX (D)) ∪ c1 (OX (D0 )).

There is another interpretation of the intersection product which can be ob-

tained from the de Rham Theorem: this result states, in particular, that if X is
a smooth projective surface, we have an isomorphism i : H 2 (X, R) −→ HdR
2
(X).
2
Recall that an element in HdR (X) is a 2−form. we have then the following pair-
ing Z
2
HdR 2
(X) × HdR (X) −→ R, (ω, ω 0 ) 7→ ω ∧ ω0 .
X
The explicit formulation of the de Rham Theorem states that for every α, β ∈
H 2 (X, R) we have Z
α∪β = i(α) ∧ i(β).
X

2.5.2 The Néron-Severi group

Let X be a smooth projective surface. We defined the Néron-Severi group of X
as
N S(X) := P ic(X)/P ic0 (X).
In this way, the Néron-Severi group is a group of equivalence classes of line
bundles on X. Moreover, we have in injective map N S(X) −→ H 2 (X, Z), so
that we can view the Néron-Severi group as a submodule of H 2 (X, Z). Moreover,
we have that N S(X) ⊆ H 2 (X, Z) ∩ H 1,1 (X). A first important result is the
following:

Theorem 2.5.2. (Lefschetz). If X is a smooth projective surface, then

N S(X) = H 2 (X, Z) ∩ H 1,1 (X)

66 Line bundles

i j
Proof. Recall that there is a sequence of injections ZX −→ CX −→ OX . Now,
the injection j induces the map H k (X, C) −→ H k (X, OX ) = H 0,k (X) for every
k, which is the projection onto the factor in the Hodge decomposition. In
particular we have

H 2 (i) H 2 (j)
H 2 (X, Z) −→ H 2 (X, C) −→ H 2 (X, OX ).
h
This combination is exactly the map H 2 (X, Z) −→ H 2 (X, OX ) coming from
the exponential sequence. Hence, the kernel of h is exactly N S(X). Now, let
α ∈ H 2 (X, Z) ∩ H 1,1 (X). Then H 2 (j) ◦ H 2 (i)(α) = 0 (as α is of type (1, 1)), so
that α ∈ ker(h) = N S(X).

There is an alternative definition of the Néron-Severi group. First of all, let

me define the following:

Definition 2.5.1. Let D be a Weil divisor on X. Then D is numerically

equivalent to 0 if and only if for every irreducible curve C of X we have D·C = 0.
Two Weil divisors D, D0 are numerically equivalent, and we write D ≡ D0 , if
D − D0 is numerically equivalent to 0.

Let us define N 1 (X) := Div(X)/ ≡. Notice that since the intersection

pairing depends only on the linear class of a Weil divisor, we have that N 1 (X) =
P ic(X)/ ≡. First of all, notice that if L ∈ P ic0 (X), then L ≡ 0: indeed, let
C be an irreducible curve on X. Then

L · OX (C) = c1 (L ) · c1 (OX (C)) = 0,

as c1 (L ) = 0. Then we have a map N S(X) −→ N 1 (X), which is clearly

surjective. Now, consider the following theorem

Theorem 2.5.3. (Kleiman). A line bundle L is numerically equivalent to 0

if and only if there is m ∈ Z, m 6= 0 such that L ⊗m ∈ P ic0 (X).

Let L ∈ P ic(X) be a numerically trivial line bundle. Then there is m ∈ Z,

m 6= 0, such that L ⊗m ∈ P ic0 (X). This implies that c1 (L ⊗m ) = 0, so that
c1 (L ) ∈ H 2 (X, Z)tors . By the Lefschetz Theorem on (1, 1)−classes, we then
get
N 1 (X) = H 2 (X, Z)f ree ∩ H 1,1 (X),

so that N 1 (X) = N S(X)/tors. As a corollary of this we have:

Theorem 2.5.4. Let X be a smooth projective surface. Then N 1 (X) is a finitely

generated free abelian group. Hence N S(X) is a finitely generated abelian group.

Definition 2.5.2. The rank ρ(X) of N S(X) is called Picard number of X.

Nef line bundles 67

As N S(X) (or N 1 (X)) is a submodule of H 2 (X, Z), and on this one we have
the intersection pairing given by the cup product, we can consider the inter-
section pairing induced on N S(X) (or N 1 (X)). In this way, N 1 (X) becomes a
lattice of rank ρ(X). Our aim is to study this intersection pairing. Let me first
introduce the following definitions:

Definition 2.5.3. A Weil Q−divisor on X is an element of DivQ (X) :=

Div(X) ⊗ Q.

In particular, a Weil Q−divisor D on X can be written as

X
D= nY Y,
Y ∈P (X)

where nY ∈ Q is 0 for all but a finite number of Y ∈ P (X). A Weil Q−divisor D

on X has the following property: there is n ∈ Z such that nD is a Weil divisor.
On Weil Q−divisors one can define a linear equivalence relation by linearity:
namely, two Weil Q−divisors D, D0 are linearily equivalent if and only if there
are n, n0 ∈ Z such that nD and n0 D0 are linearily equivalent Weil divisors.
One has then A1Q (X), which is naturally isomorphic to A1 (X) ⊗ Q. As X is
smooth, we have A1 (X) ' P ic(X), so that A1Q (X) is canonically isomorphic to
P ic(X) ⊗ Q. Notice that while the objects in the first vector space have a clear
geometric interpretation, the second vector space has no real meaning.
Recall, that on Div(X) there is an intersection pairing. By linearity one can
then define an intersection pairing

· : DivQ (X) × DivQ (X) −→ Q.

Definition 2.5.4. Two Weil Q−divisors D, D0 are numerically equivalent, and

we write D ≡ D0 , if and only if for every irreducible curve C on X we have
D · C = D0 · C. We denote by NQ1 (X) := DivQ (X)/ ≡.

Clearly, we have NQ1 (X) = N 1 (X)⊗Q. Moreover, the intersection pairing on

DivQ (X) defines an intersection pairing on NQ1 (X), which coincides with the lin-
ear extension of the one we have on N 1 (X). Recall that N 1 (X) ' N S(X)/tors,
so that N 1 (X) ⊗ Q ' N S(X) ⊗ Q, so that NQ1 (X) ' N S(X) ⊗ Q. This is a
Q−vector space of dimension ρ(X).
Using the same formal extension of coefficients, we can define R−divisors and
1
NR (X), which is isomorphic to N S(X) ⊗ R, and which is a real vector space
of dimension ρ(X). On this vector space we have a non-degenerate symmetryc
bilinear form, given by the intersection pairing

· : NR1 (X) × NR1 (X) −→ R.

In a completely similar way, one can extend the cup product on H 2 (X, Z) to

∪ : H 2 (X, R) × H 2 (X, R) −→ R,
68 Line bundles

and get an injection c1 : NR1 (X) −→ H 2 (X, R), so that by the Lefschetz Theorem
on (1, 1)−classes we even get NR1 (X) = H 2 (X, R) ∩ H 1,1 (X). Here is the main
result of this section:

Theorem 2.5.5. (Hodge Index Theorem). The cup product restricts non-
degenerately to N 1 (X) with signature (1, ρ(X) − 1).

Proof. The proof goes as follows: let ω be the real (1, 1)−form associated to the
Fubini-Study metric (the metric induced on X by some embedding in PN ). This
is a Kähler form, hence it is closed, and defines a class [ω] ∈ H 1,1 (X)∩H 2 (X, R),
i. e. an element of NR1 (X). Moreover, the form ω is pointwise positive definite.
Now, let
2
Hprim (X, R) := {[α] | [α ∧ ω] = 0}.
2
(X, R) = [ω]⊥ (where the orthogo-
R
As [α] ∪ [ω] = X α ∧ ω, we have that Hprim
nality is with respect to the cup product). In conclusion we have a direct sum
decomposition H 2 (X, R) = Hprim 2
(X, R) ⊕ R[ω], orthogonal with respect to the
cup product, so that

NR1 (X) = (Hprim

2
(X, R) ∩ H 1,1 (X)) ⊕ R[ω].
R
As ω is pointwise positive definite, we have X ω∧ω > 0, so that [ω]∪[ω] > 0 and
the intersection pairing is positive on R[ω]. Let us study the intersection pairing
2
on Hprim (X, R) ∩ H 1,1 (X): let α be a real (1, 1)−form with α ∧ ω = 0. Choose
a C ∞ −trivialization of TX ∗
by two 1−forms β1 and β2 , which is everywhere
orthonormal with respect to the Kähler metric corresponding to ω. We then
have (locally)
i X
ω = (β1 ∧ β 1 + β2 ∧ β 2 ), α= αjk βj ∧ β k .
2
j,k

Since α is real, we must have α11 and α22 purely imaginary and α12 = α21 . The
fact that α ∧ ω = 0 implies α11 = −α22 . In conclusion, we have
1
α ∧ α = (|α11 |2 + |α12 |2 )β1 ∧ β 1 ∧ β2 ∧ β 2 = −(|α11 |2 + |α12 |2 )ω ∧ ω.
2
R
Hence X α ∧ α ≤ 0, and α ∧ α = 0 if and only if α = 0. But this clearly means
2
that the intersection form is negative definite on Hprim (X, R) ∩ H 1,1 (X). To
conclude, simply remark that R[ω] ∩ N 1 (X) 6= ∅, as there is surely the class of
an ample divisor.

Remark 2.5.1. An equivalent reformulation of the Hodge Index Theorem is the

following: let D be a Weil divisor such that D2 > 0. If there is a Weil divisor
C such that D · C = 0, then C 2 ≤ 0, and we have C 2 = 0 if and only if C ≡ 0.
Indeed, we can produce a basis of NR1 (X) such that D is an element of it. Then,
as D2 > 0, every element C in NR1 (X) orthogonal to D is such that C 2 ≤ 0 and
it is 0 if and only if C ≡ 0. The converse direction is similar.
Nef line bundles 69

2.5.3 The Nakai-Moishezon Criterion for ampleness

First of all, let me introduce the following definition:

Definition 2.5.5. A Weil divisor D on X is ample if the corresponding line

bundle OX (D) is ample.

Here is a corollary of the Hodge Index Theorem:

Lemma 2.5.6. Let D be a Weil divisor on a smooth projective surface X such

that D2 > 0. If H is an ample Weil divisor, then D · H 6= 0. Moreover, if
D · H > 0 (resp. D · H < 0) then there is m ∈ N such that mD (resp. −mD)
is effective.

Proof. Let D2 > 0. If D · H = 0, then D · (mH) = 0 for every m ∈ Z. We can

then suppose H to be very ample. By the Hodge Index Theorem (see Remark
2.5.1) we would have H 2 ≤ 0. As H is very ample, it is easy to see that H 2 > 0,
hence D · H 6= 0.
Now, let us suppose D · H > 0. Consider m ∈ N: then the linear system
|KX − mD| is empty for m  0. Indeed, if there is an effective divisor D0 such
that D0 ∼ KX − mD, we have D0 · H > 0 as D0 is effective and H is ample (see
Lemma 2.5.7), but
D0 · H = KX · H − mD · H < 0

for m  0 as D · H > 0. Hence h0 (KX − mD) = 0 for m  0. Using the

inequality coming from the Riemann-Roch Theorem we get
1 2 2 1
h0 (mD) + h0 (KX − mD) ≥ D m − (D · KX )m + χ(OX ).
2 2
If m ≥ 0, as D2 > 0 we have h0 (mD) > 0, so that mD is effective, hence D is
effective. The other case follows.

We have the following lemma:

Lemma 2.5.7. Let X be a smooth projective surface, and let L be an ample

line bundle on X. If C is an irreducible curve on X, then L · OX (C) > 0.
Equivalently, if a Weil divisor D on X is ample, then D · C > 0 for every
irreducible curve C on X.

Proof. The proof is easy: let i : C −→ X be the inclusion. As the generic

curve in the linear system |OX (C)| is smooth by Bertini’s Theorem, we can
suppose C smooth. Since i is a finite morphism, the line bundle L|C := i∗ L is
⊗m
ample on C. There is then m ∈ N such that L|C is very ample, hence there
⊗m
is an inclusion f : C −→ P such that f OPn (1) = L|C
n ∗
. Now, notice that
deg(f OPn (1)) = H · f (C), where H is an hyperplane in Pn . Since f (C) is a
∗

curve in Pn we easily see that H · f (C) > 0. In conclusion deg(L|C

m
) > 0. Now,
70 Line bundles

m > 0, so that L · OX (C) = deg(L|C ) > 0, where the first equality is Lemma
2.4.2.

In general, the converse statement is non true, and we are going to see an
example in the following section. Anyway, we have the following important
criterion:

Theorem 2.5.8. (Nakai-Moishezon’s Criterion for Ampleness). Let X

be a smooth projective surface. A line bundle L on X is ample if and only if
L 2 > 0 and L · OX (C) > 0 for every irreducible curve C on X.

Proof. Let D be a Weil divisor corresponding to L . If L is ample, then D·C > 0

for every irreducible curve on X. Moreover, we have D2 > 0: indeed, as D is
ample there is m ∈ N such that mD is very ample, so that (mD)2 > 0. Hence
m2 D2 > 0, so that D2 > 0. For the converse, let us begin with the following
lemma:

Lemma 2.5.9. Let D = D1 + D2 be the sum of two effective divisors on X.

Then there is an exact sequence

0 −→ OD1 (−D2 ) −→ OD −→ OD2 −→ 0.

Proof. It is sufficient to prove the lemma in the affine case, so let D1 = (f1 ) and
D2 = (f2 ) for some f1 , f2 ∈ C[x, y]. Then D = (f1 f2 ), so that the restriction
map r2 is the natural surjection

r2 : C[x, y]/(f1 f2 ) −→ C[x, y]/(f2 ).

Its kernel is (f2 ) · C[x, y]/(f1 f2 ), which is readily seen to be isomorphic to (f2 ) ·
C[x, y]/(f1 ).

Let H be a very ample divisor. It is then effective, so that D · H > 0.

Moreover D2 > 0, and we can apply Lemma 2.5.6 to guarantee that there is
m ∈ N such that mD is effective. Notice that D is ample if and only if mD is
ample, so we can suppose D to be effective.
We are noy goint to show that if D0 is a divisor supported on D and n ∈ N,
then H 1 (D0 , OD0 (nD)) = 0 for n  0. Let us suppose D0 to be irreducible.
Then consider its normalisation ν : D00 −→ D0 . As ν is a finite map, we
have H 1 (D00 , ν ∗ OD0 (nD)) ' H 1 (D0 , OD0 (nD)). Moreover deg(ν ∗ OD0 (nD)) =
deg(OD0 (nD)), and this last is nD · D0 . As D · D0 > 0 (since D0 is supported
on D), for n  0 we then have H 1 (D00 , ν ∗ OD0 (nD)) = 0, and we are done.
If D0 is reducible, we can write it as D0 = D00 + R, where D00 is irreducible
and R is effective. The exact sequence in Lemma 2.5.9 twisted by OD0 (nD)
then gives an exact sequence

H 1 (D00 , OD00 (nD − R)) −→ H 1 (D0 , OD0 (nD)) −→ H 1 (R, OR (nD)).

Nef line bundles 71

Now, the first group is trivial for n  0 (follow the same argument as before,
as D00 is irreducible). Now we can proceed by induction on the number of
components of R to get that the last group is trivial for n  0, and we are
done.
Let us now consider the following exact sequence

0 −→ OX ((n − 1)D) −→ OX (nD) −→ OD (nD) −→ 0. (2.5)

By the previous part, we have that H 1 (D, OD (nD)) = 0 for n  0. The long
exact sequence induced by (2.5) in cohomology then gives a surjective map

H 1 (X, OX ((n − 1)D)) −→ H 1 (X, OX (nD))

for n  0. As these two spaces are finite dimensional, this surjective map
is an isomorphism for n  0, so that the exact sequence induced by (2.5) in
cohomology implies that the map

H 0 (X, OX (nD)) −→ H 0 (D, OD (nD))

is surjective for n  0. Using a similar argument as before, one can show that
OD (nD) is globally generated, so that OX (nD) is generated by global sections
on points of D. But recall that D is effective, hence there must be a section of
OX (D) whose zero locus is exactly D: hence OX (nD) is globally generated. By
properties of globally generated line bundles, we have then a map

ϕnD : X −→ PN .

Let us suppose there is a curve C of X such that ϕnD (C) = {P }, where P ∈ PN .

Let L be a hyperplane of PN such that P ∈ / L. Hence

0 = OX (C) · ϕ∗nD OPN (L) = OX (C) · OX (nD).

But this implies that there is a curve C of X such that D · C = 0, contradicting

the hypothesis. Hence ϕnD is a finite map. From Corollary 2.3.4 we then get
that nD is ample, so that D is ample.

Another characterization of ampleness of a line bundle can be given in terms

of its first Chern class: if X is a complex manifold, we can define a line bundle
L to be positive if and only if c1 (L ) is a positive (1, 1)−form. Hence if X is
projective, then it has to carry a positive line bundle by the Nakai-Moishezon
Criterion. What is more surprising, is that even the converse is true, and we
have the following result, whose proof is omitted:

Theorem 2.5.10. (Kodaira’s Projectivity Criterion). Let X be a com-

pact complex manifold. Then X is projective if and only if there is a positive
(1, 1)−form ω ∈ H 1,1 (X) ∩ H 2 (X, Q).
72 Line bundles

2.5.4 Cones
In the previous section we have shown that if D is an ample divisor on X,
then D · C > 0 for every irreducible curve on X. We now, define the following
definition:

Definition 2.5.6. A line bundle L on X is nef if L · OX (C) ≥ 0 for every

irreducible curve C on X. A Weil divisor D on X is nef if the corresponding
line bundle OX (D) is nef.

It follows from Lemma 2.5.7 that every ample line bundle is nef. Let us
moreover introduce the following

Definition 2.5.7. A Weil Q−divisor (resp. R−divisor) D on X is effective if

Pk
D ∼ i=1 di Ai for some integer k, where Ai is a Weil divisor and di ∈ Q>0
(resp. di ∈ R>0 ) for every i = 1, ..., k.
A Weil Q−divisor (resp. R−divisor) D on X is ample if there are D ∼
Pk
i=1 di Ai for some integer k, where Ai is an ample Weil divisor and di ∈ Q>0
(resp. di ∈ R>0 ) for every i = 1, ..., k.
A Weil Q−divisor (resp. R−divisor) D on X is nef if there are D ∼
Pk
i=1 di Ai for some integer k, where Ai is a nef Weil divisor and di ∈ Q>0
(resp. di ∈ R>0 ) for every i = 1, ..., k.

Remark 2.5.2. Let D be a Weil Q−divisor, and let us suppose n ∈ N such that
nD ∈ Div(X). It is easy to see that D is effective if and only if nD is effective.
Moreover D is ample if and only if nD is ample. Indeed, if nD is ample,
Pk
simply take k = 1, d1 = 1/n and A1 = nD. If D is ample, then D = i=1 di Ai
Pk
for di ∈ Q>0 and Ai is ample. Then nD = i=1 ndi Ai is a Weil divisor. As Ai
is ample and di > 0 for every i = 1, ..., k, we then have ndi Ai ample for every i,
so that D is ample.
Similiarily, D is nef if and only if nD is nef. Indeed, if nD is nef, then do as
Pk Pk
before. If D = i=1 di Ai for di ∈ Q>0 and Ai is nef. Then nD = i=1 ndi Ai
is a Weil divisor. Let C be an irreducible curve: then Ai · C ≥ 0 for every i. As
di > 0, we have nD · C > 0, so that nD is nef.
Remark 2.5.3. Let D be a Weil R−divisor. If D is ample, then D · C > 0
for every irreducible curve C. Similarily, if D is nef, then D · C ≥ 0 for every
irreducible curve C on X.
Let us now define the following set:
X
QEf f (X) := {α = ai [Ci ] ∈ NR1 (X) | Ci is an irr. curve, ai ≥ 0}.
i

Clearly QEf f (X) is a cone in NR1 (X). Now, let Ef f (X) := QEf f (X), the
closure of QEf f (X) in the euclidean topology of NR1 (X). Then Ef f (X) is a
Nef line bundles 73

cone, called the effective cone of X. In general we have QEf f )(X) ⊆ Ef f (X),
but not the equality.
Notice that for classes in NR1 (X) we can define ampleness and nefness: a
class α ∈ NR1 (X) is ample (resp. nef ) if it is the numerical class of an ample
(resp. of a nef) Weil R−divisor. Before going on, we cite the following result:

Theorem 2.5.11. (Kleiman’s Criterion for nefness). Let X be a smooth

projective surface, and let D be a Weil R−divisor. Then D is nef if and only if
D2 ≥ 0 and D · C ≥ 0 for every irreducible curve on X.

We have then

Amp(X) := {α ∈ NR1 (X) | α is ample},

which is called the ample cone of X, and

N ef (X) := {α ∈ NR1 (X) | α is nef},

called the nef cone of X. Notice that Amp(X) ⊆ N ef (X), and that these are
cones.

Proposition 2.5.12. Let X be a smooth projective surface. Then N ef (X)

is the dual cone of Ef f (X) (and conversely) with respect to the intersection
pairing on NR1 (X).

Proof. If V is a real vector space with an intersection form (., .) and C is a cone
in V , then the dual cone of C with respect to (., .) is

C ∨ := {α ∈ V | (α, β) ≥ 0 ∀ β ∈ C}.

Let now α ∈ N ef (X). Then α · [C] ≥ 0 for every irreducible curve C of X.

This then implies that α · γ ≥ 0 for every γ ∈ QEf f (X), hence of Ef f (X). In
conclusion, α ∈ Ef f (X)∨ . Conversely, if α ∈ Ef f (X)∨ , then α · γ ≥ 0 for every
γ ∈ Ef f (X). In particular this is true for every irreducible curve of X, so that
by Theorem 2.5.11 we get that α is the class of a nef R−divisor. In conclusion
N ef (X) = Ef f (X)∨ .

Next, we study the relation between the nef and the ample cone:

Proposition 2.5.13. Let X be a smooth projective surface. Then Amp(X) =

N ef (X)◦ and N ef (X) = Amp(X). In particular, the nef cone is closed and
the ample cone is open.

Proof. As ample line bundles are nef, we easily see that Amp(X) ⊆ N ef (X).
The fact that N ef (X) is closed is evident, and we need to show that Amp(X)
is open. In order to do that, let us prove the following:
74 Line bundles

Lemma 2.5.14. Let H be an ample Q−divisor on X, and let E be a Q−divisor.

Then H + E is ample for every  ∈ Q such that 0 < ||  1.

Proof. We need to show that there is n  0 such that n(H + E) is an ample
divisor. First of all we can clear the denominators, and we can suppose that H
and E are Weil divisors, with H ample. Then there is m  0 such that mH + E
is ample. Indeed, as H is ample, we can choose m1 , m2 ∈ N such that nH is
very ample for n ≥ m1 and nH + E is globally generated for n ≥ m2 . Hence, if
n ≥ m1 + m2 , then nH + E is very ample. Now, let us take H + n1 E: this is an
ample Q−divisor for n  0, and we are done.

Now, let us consider H an ample Q−divisor, and let E1 , ..., Er be arbitrary

Q−divisor. It follows from the previous Lemma that H + 1 E1 + ... + r Er
is ample for i ∈ Q and 0 < |i |  1 for every i = 1, ..., r. Notice that we
can even suppose i ∈ R. We can now consider H an ample R−divisor and
E1 , ..., Er arbitrary R−divisors. As every Ei is a finite linear combination (with
real coefficients) of Weil divisors, we can suppose Ei to be a Weil divisor for
P
every i = 1, ..., r. Let us write H = i di Ai , where di ∈ R>0 and Ai is an ample
Weil divisor. Let q ∈ Q be such that 0 < q < d1 , so that we can write
r
X r
X X
H+ j Aj = (qA1 + j Ej ) + (d1 − q)A1 + di Ai .
j=1 j=1 i>1

The second and the third summands are ample R−divisors, and the first one
is ample provided that 0 < |j |  1 for every i by the previous part, so that
H + 1 E1 + ... + r Er is an ample R−divisor.
From this it follows easily that Amp(X) is an ample cone. Hence Amp(X) ⊆
N ef (X)◦ and N ef (X) ⊆ Amp(X). We need to show the opposite inclusions.

Lemma 2.5.15. Let X be a smooth projective surface, and let D be a nef

R−divisor on X. If H is an ample R−divisor on X, then D + H is ample for
every  > 0. Conversely, if D and H are two R−divisors such that D + H is
ample for every  > 0, then D is nef

Proof. Let D and H be R−divisors such that D + H is ample for every  > 0.
Then, if C is any effective curve on X, we have

0 < (D + H) · C = D · C + H · C.

Then, the limit for  going to 0 is non-negative, and D is nef.

Conversely, let D be nef and H ample. We can replace H by H, so that we
just need to prove that D + H is ample. If D and H are Q−divisors, then we
can clear denominators and suppose that D is a nef divisor, and H is ample. If
C is an irreducible curve, we then get

(D + H) · C = D · C + H · C > 0,
Nef line bundles 75

as D · C ≥ 0 (since D is nef) and H · C > (since H is ample). Moreover, we

have
(D + H)2 = D2 + H 2 + 2D · H > 0,

since D2 ≥ 0 by the Kleiman Criterion for nefness, H 2 > 0 by the Nakai-

Moishezon Criterion for ampleness and D · H ≥ 0 by the Kleiman Criterion for
nefness (indeed, we have D · H ≥ 0 if and only if D · (nH) ≥ 0 for some n ∈ N;
as H is ample, then there is n ∈ N such that nH is very ample, hence nH is
effective, and D · (nH) ≥ 0 as D is nef).
Now, let us suppose D and H to be R−divisors. Then, let H1 , ..., Hρ(X) be
ample divisors giving a basis for NR1 (X) (this is possible as Amp(X) is open).
Pρ(X)
By Lemma 2.5.14 the R−divisor H(1 , ..., ρ(X) ) := H − i=1 i Hi is ample
for 0 < |i |  1. We can find 1 , ..., ρ(X) ∈ R such that the divisor D0 :=
D + H(1 , ..., ρ(X) ) is a Q−divisor, and by the previous part then D0 is ample.
Finally, simply note that

D + H = D0 + 1 H1 + ... + ρ(X) Hρ(X) ,

which is ample as D0 is ample and every Hi is ample.

From this lemma it follows immediately the Proposition: any nef divisor is
the limit of ample divisors, and conversely.

Now, notice that if D is a nef divisor, then N ef (X) ⊆ Ef f (X): indeed,

if D is a nef class in NR1 (X), and H is an ample R−divisor, then D + H is
ample for every  > 0. Let us suppose D, H be Q−divisors, so that clearing
the denominators we can suppose D and H to be Weil divisors. As D + H is
ample, then there is n  0 such that n(D + H) is very ample, hence its class in
NR1 (X) is effective. Hence D + H is effective, and taking the limit for  going
ot 0, we then get that D is a class in Ef f (X).
Moreover, notice that Ef f (X) ⊆ N ef (X) if and only if for every irreducible
curve C we have C 2 ≥ 0: indeed, every irreducible curve intersects a distinct
curve non-negatively, hence C is nef if and only if C 2 ≥ 0. By Kleiman’s
Criterion for nefness we then conclude. If there is an irreducible curve C such
that C 2 ≤ 0, then Ef f (X) is spanned by R≥0 · C and by the following cone

Ef f (X)C≥0 := {γ ∈ Ef f (X) | γ · C ≥ 0}.

Indeed, if C 0 is an effective curve on X such that C is not contained in C 0

as a component, the C · C 0 ≥ 0. Now, C ∈ Ef f (X), and C 2 < 0, so that
C∈ / Ef f (X)C≥0 , and the assertion is proved.
Moreover, if C 2 < 0, then the ray spanned by C is not contained in the nef
cone. It then spans an extremal ray in Ef f (X).
76 Line bundles

Example 2.5.1. Let us consider X = P2 . Then the line bundle O(d) is ample
if and only if d > 0, and it is nef if and only if d ≥ 0. In this case we have
NR1 (X) = R. Then Ef f (X) = N ef (X) = R≥0 , and Amp(X) = R>0 . Notice
that OX (KX ) = O(−3) is not nef.
Example 2.5.2. Let us consider X = P1 × P1 . Then NR1 (X) ' R2 , and we
consider as a basis h1 , h2 . A class α = ah1 + bh2 is effective if and only if a ≥ 0
and b ≥, so that Ef f (X) = R≥0 × R≥0 . Let now OX (n, m) be the line bundle
on X of bidegree (n, m). Then OX (n, m) · α = am + bn, and if OX (n, m) then
am + bn ≥ 0 for every a, b ≥ 0. If b = 0, then we get m ≥ 0, and if a = 0 then
we get n ≥ 0. Moreover, OX (n, m)2 = 2nm, which is non-negative if n, m ≥ 0,
so that N ef (X) = Ef f (X). The ample cone is clearly the interior of N ef (X).
Recall that OX (KX ) = OX (−2, −2), hence KX is not nef.
Example 2.5.3. We describe now a new example of smooth projective surface.
Let C be a smooth curve of genus g, and let E be a rank 2 vector bundle on C.
Let X := P(E) be the projective bundle associated to E, and let π : X −→ C be
the projection morphism. Without loss of generality, we can suppose deg(E) =
0. Let ξ := c1 (OX (1)) be the first Chern class of the tautological line bundle
on X (i. e. OX (1) is the line bundle such that π∗ OX (1) ' E), and let f be the
class of a fiber. Then NR1 (X) ' R2 , where a basis is given by ξ and f (notice
that if P, Q ∈ C are different points, then it is not true in general that P ∼ Q,
so that f ∗ P ∼ f ∗ Q is not necessarily true. Anyway, P − Q ∈ P ic0 , so that by
passing to N 1 (X) thew give the same class). We have the following intersection
properties: f 2 = 0 by Proposition 2.4.4; ξ · f = 1, as ξ represents (locally) a
section of π; ξ 2 = deg(E) = 0. Hence, the intersection form is the same as in
Example 2.5.2.
Let us consider af + bξ ∈ NR1 (X). We have (af + bξ)2 = 2ab, which is
non-negative if and only if a, b ≥ 0 or a, b ≤ 0. As f is evidently nef, we then
get that N ef (X) ⊆ R≥0 × R≥0 , and ξ = 0 gives a boundary of it. Similarily, if
a curve af + bξ is effective, then a ≥ 0; moreover, f is effective, so that Ef f (X)
has ξ = 0 as a boundary. The remaining part of these cones depends on the
geometry of E.
Case 1. Let us suppose that there is a line bundle A ⊆ E such that deg(A) =
d < 0. Hence we have C := P(A) ⊆ X is an effective curve on X, and its class
in N 1 (X) is df + ξ. Indeed, if C = af + bξ, we have C · f = 1, so that b = 1,
and C · ξ = deg(A), so that a = d. In particular, notice that C 2 = 2d < 0,
hence we have an effective divisor which is not nef. Moreover, Ef f (X) is the
cone spanned by f and af + ξ, and N ef (X) is strictly contained in it. In order
to determine N ef (X), let nf + mξ be a nef line bundle. Hence n, m ≥ 0, and

n + md = (nf + mξ) · (df + ξ) ≥ 0.

Then N ef (X) is the cone spanned by f and ξ − df . Notice in this case that ξ
Nef line bundles 77

is an effective divisor such that ξ · (af + bξ) > 0 for every a, b > 0, but ξ is not
ample: indeed ξ 2 = 0.
Case 2. Let us suppose that if A ⊆ E is a line bundle of degree d, then d ≥ 0.
Any effective curve on X arises then by a line bundle of the form OX (m) ⊗ π ∗ A
for some m ∈ N0 and some line bundle A ⊆ E. The class of this curve is then
df + mξ, and here d ≥ 0, so that Ef f (X) ⊆ R≥0 × R≥0 . As ξ and f are effective
curves, then it is an equality. Now, it is easy to show that ξ is nef, so that
N ef (X) = Ef f (X).
Example 2.5.4. Let C be a smooth curve of genus g, and let X := C × C. Then
X is a smooth surface having two natural projections p1 , p2 : X −→ C. Let us
consider the following curves on X: f1 is a fiber of p1 , f2 is a fiber of p2 and δ
is the class of the diagonal. If g ≥ 1, these classes are independent. Indeed, it
is clear that f1 and f2 are independent, and that fi2 = 0 and f1 · f2 = 1. Now,
we have δ 2 = 2 − 2g, and clearly δ · fi = 1. If there are n1 , n2 ∈ Z such that
δ = n1 f1 + n2 f2 , we would then get n1 = n2 = 1, and δ 2 = 2. Hence g = 0,
but we supposed g > 0, so that δ is independent from f1 and f2 . We have then
NR1 (X) ' R3 and a basis is given by f1 , f2 , δ. In this case it is more complicated
to describe the nef and the effective cone.
We conclude this section with two basic properties of nef line bundles:

Proposition 2.5.16. Let X be a smooth projective surface, and let L ∈

P ic(X). Moreover, let Y be a smooth projective variety and f : X −→ Y
be a morphism.

1. If L is nef, then f ∗ L is nef.

2. If f is surjective and f ∗ L is nef, then L is nef.

Proof. For the first item, let C be an irreducible curve on X. If f (C) is a point,
then f ∗ L · OX (C) = 0. If f (C) is a curve and f : C −→ f (C) is a map of
degree d, then f ∗ L · OX (C) = dL · OY (f (C)) ≥ 0 as L is nef and f (C) is an
irreducible curve. The second item is evident.
78 Line bundles
Chapter 3

Birational geometry

In this chapter we start with the investigation on the birational geometry of

surfaces. We introduce some important notions we will need in the following,
namely the blow-up of a surface in a point, and the birational map between two
surfaces. The main role will be played by (−1)−curves, i. e. curves E such that
E 2 = −1. We will show that a surface X admits a (−1)−curve if and only if it
is a blow-up of another surface at a smooth point, and E is exactly the curve
coming from the blow-up. This will allow us to introduce the notion of minimal
model of a smooth projective surface, so that the birational classification of
smooth projective surfaces is simply the classification of thei minimal models.
In particular we recall some properties of normalizations, and we prove the
Zariski Main Theorem about fibers of a birational morphism.
Moreover, we introduce the notion of Kodaira dimension of a line bundle
and of a surface, and we prove that this is a birational invariant of a smooth
projective surface: the birational classification then will proceed first by dividing
surfaces by means of thei Kodaira dimension.

3.1 Contraction of curves

In this section we recall some basic facts about rational and birational maps,
blow-ups and related subjects. This will be one of the basic steps in order to
understand how the Enriques-Castelnuovo classification works.

3.1.1 Rational and birational maps

Let us start from the following easy lemma:

Lemma 3.1.1. Let X and Y be two projective varieties, and let f, g : X −→ Y

be two morphisms. If there is a non-empty open subset U ⊆ X such that f|U =
g|U , then f = g.

79
80 Birational geometry

Proof. Let i : Y −→ Pn be a closed immersion. Clearly, if the Lemma is true

for i ◦ f and i ◦ g, then it is true for f and g, so we can suppose Y = Pn . The
morphisms f and g give then a map (f, g) : X −→ Pn × Pn , and let ∆ be the
diagonal of Pn × Pn . By hypothesis we have (f, g)(U ) ⊆ ∆. As U is dense in X
and ∆ is closed, we get (f, g)(X) ⊆ ∆, so that f = g.

The idea of a rational map is that we want to allow morphisms which are
not defined on the whole X, but only on some dense open subset of X. So, the
naif definition should be that a rational map f : X 99K Y is a couple (U, f ),
where U is a dense open subset of X, and f : U −→ Y is a morphism. Anyway,
the correct definition is more complicated.

Definition 3.1.1. Let X and Y be two projective varieties. A rational map

f : X 99K Y is an equivalence class of pairs (U, f ), where U is a dense open subset
of X and f : U −→ Y is a morphism, where (U, f ) and (U, g) are equivalent if
and only if U ∩ V 6= ∅ and f|U ∩V = g|U ∩V . A rational map is dominant if f (U )
is dense in Y for every pair (U, f ) representing f .

Notice that the Lemma implies that the relation described is an equivalence
relation. We can compose rational (resp. dominant) maps getting another
rational (resp. dominant) map.

Definition 3.1.2. A birational map f : X 99K Y is a rational map adimitting

an inverse f −1 : Y 99K X, which is a rational map such that f ◦ f −1 and f −1 ◦ f
are the identity (as rational maps). If there is a birational map from X to Y ,
we say that X is birational to Y .

Example 3.1.1. Let us describe an important example of birational map. First

of all, let us consider P1 × P1 , and then the Segre embedding

s : P1 × P1 −→ P3 , s((x0 : x1 ), (y0 : y1 )) := (x0 y0 : x0 y1 : x1 y0 : x1 y1 ),

which is a well-defined injective map. The image X := im(i) is then isomorphic

to P1 × P1 . Notice that if on P3 we use coordinates (u : v : w : z), then
X = (uz − vw), which is a smooth quadric in P3 . Now, let us consider the
following map:

f : X \ {(0 : 0 : 0 : 1)} −→ P2 , f (u : v : w : z) := (u : v : w),

which is the projection of X from the point (0 : 0 : 0 : 1). This gives a rational
map f : X 99K P2 , and we can easily write its inverse map

g : P2 −→ P3 , g((x0 : x1 : x2 )) := (x20 : x0 x1 : x0 x2 : x1 x2 ).

It is clear that im(g) ⊆ X, and these two maps are inverse to each other. Then
P2 is birational to P1 × P1 . Notice that these two varieties are anyway not
isomorphic (their Picard group are different).
Contraction of curves 81

Now, let ϕ : X 99K Y be a dominant rational map, and let f MY (Y ) be a

rational function on Y . Then ϕ can be represented by a pair (U, ϕU ), and f
by a pair (V, fV ), where V is an open subset of Y where f is regular. As ϕ is
dominant, the open subset ϕU (U ) is dense in Y , and V ∩ ϕU (U ) 6= ∅. Then
ϕ−1
U (V ∩ ϕU (U )) is a dense open subset of X, so that f ◦ ϕU is a regular function
on it. Hence, ϕ∗ f := f ◦ ϕ is a rational function on X. In conclusion, we get

ϕ : MY (Y ) −→ MX (X).

One can define an inverse correspondence.

Proposition 3.1.2. Let X and Y be two projective varieties. The following are
equivalent:

1. X and Y are birational;

2. there is a dense open subset U of X and a dense open subset V of Y which

are isomorphic;

3. MX (X) ' MY (Y ) as C−algebras.

3.1.2 The Zariski Main Theorem

In this subsection I birefly recall properties of birational maps and of fibers of
morphism. Let me recall the following definition:

Definition 3.1.3. A projective variety X is normal if for every x ∈ X the local

ring OX,x is a normal ring, i. e. it is integrally closed in its quotient field.

If x ∈ X is a smooth point, then OX,x is normal, so that any smooth

projective variety is normal. If X is normal and if x ∈ X is such that the
local ring OX,x has dimension 1, then OX,x is regular. Hence if X is normal,
the singular locus of X has codimension at least 2. In particular, every normal
curve is smooth.

Definition 3.1.4. Let X be a projective variety. A normalisation of X is a

pair (X 0 , f ) such that X 0 is a normal projective variety, and f : X 0 −→ X is a
morphism which is finite and birational.

This definition coincides with the usual definition of normalisation. Indeed,

we have the following:

Proposition 3.1.3. Let X be a projective variety, and let (X 0 , f ) be a nor-

malisation of X. If (X 00 , g) is another normalization of X, then there is an
isomorphism h : X 0 −→ X 00 such that g ◦ h = f .
82 Birational geometry

Proof. We start with X affine, and let R(X) := OX . It is a Noetherian

ring whose integral closure in k(X) is denoted R0 . Then R0 is a finitely gen-
erated R(X)−module, so that there is an ideal I of C[x1 , ...xn ] such that
R0 ' C[x1 , ..., xn ]/I. The closed subvariety of Cn defined by I is denoted X 0 :
it is a normal variety admitting a finite morphism f : X 0 −→ X defined by the
inclusion R(X) ⊆ R0 . Moreover, the quotient field of R(X) and of R0 are equal,
hence by Proposition 3.1.2 f is birational. Now, X 00 is normal if and only if
R(X 00 ) is normal, and as X 00 is a normalisation of X, then R(X) ⊆ R(X 00 ).
Hence X 00 and X 0 correspond to the integral closure of R(X) in its quotient
field, and we are done.
If X is projective, one produces a normalisation as in the previous case, and
applies the statement to local pieces.

We can now proceed with the proof of the Zariski Main Theorem:

Theorem 3.1.4. (Zariski’s Main Theorem). Let Y be a normal projective

variety, and let f : X −→ Y be a birational morphism. Then the fibers of f are
connected.

Proof. The first step is to prove the following lemma:

Lemma 3.1.5. Let X, Y be two projective varieties, and let f : X −→ Y be a

morphism.

1. If the natural map OY −→ f∗ OX is an isomorphism, then the fibers of f

are connected and non empty.

2. If Y is normal, f is surjective and its fibers are connected, then the natural
map OY −→ f∗ OX is an isomorphism.

Proof. We can work in the euclidean topology (i. e. we consider X and Y to

be complex varieties). Let y ∈ Y be such that f −1 (y) =: Xy is not connected.
As f is a morphism between two projective varieties, then f is proper. Hence,
there is an open subset V of Y (in the euclidean topology) such that y ∈ V
and f −1 (V ) is not connected. Indeed, if F = A ∪ B is a disconnected fiber
whose connected components are A and B, we have that A and B are compact.
Then, let U be a neighborhood of A and W a neighborhood of B in X which
are disjoint. Again by compactness (following from the fact that f is proper)
all points of X which are sufficiently close to F are in U ∪ W . Now, simply take
V := f (W ).
Now, let y := f (F ). The inverse image of a sufficiently small ball centered
at y cannot be connected, and the canonical map OY,y −→ (f∗ OX )y cannot be
surjective. Hence if the hypethesis of item 1 is verified, we need that every fiber
has to be connected. For non-emptyness, just remark that if Xy = ∅, then the
map OY,y −→ (f∗ OX )y is not injective, and we are done.
Contraction of curves 83

For the converse, let y ∈ Y and let V be an open neighborhood of y in the

complex topology. Let g ∈ OX (f −1 (V )) be a bounded holomorphic function.
On every smooth fiber F of f the function g has the same value by definition, as
F is connected. Hence, let V 0 be the open subset of V given by smooth points
of V where f has maximal rank. Then there is a bounded continuous function
h on V 0 such that g|f −1 (V 0 ) = h0 ◦ f . As f has maximal rank at every point of
f −1 (V 0 ), for every such a point there is a neighborhood of the form U × V 00 for
some V 00 open subset of V 0 , and f is simply the projection to V 00 . Hence on V 00
the map h has to be holomorphic (as g is), hence h ∈ OY (V 0 ). As V is normal,
we can extended h to a holomorphic function on V , and in this way we see that
OY,y −→ (f∗ OX )y is an isomorphism.

Now, let us suppose that f : X −→ Y is a birational morphism, and that Y

is normal. By Lemma 3.1.5, in order to prove that the fibers of f are connected,
we just need to prove that OY −→ f∗ OX is an isomophism. As the statement
is local, we can suppose X and Y to be affine. As f is birational, f∗ OX is
coherent, hence H 0 (Y, f∗ OX ) is a finitely generated H 0 (Y, OY )−module. But
H 0 (Y, OY ) is normal as Y is normal, and H 0 (Y, f∗ OX ) is normal too. As f is
birational, their fields of fractions are isomorphic, hence we need H 0 (Y, f∗ OX ) '
H 0 (Y, f∗ OX ). But this implies that the morphism OY −→ f∗ OX is an isomor-
phism, as X and Y are affine, and we are done.

Let me first prove the following corollary:

Corollary 3.1.6. Let X and Y be smooth projective surface. If f : X −→ Y ,

is a surjective and birational morphism, then q(X) = q(Y ) and pg (X) = pg (Y ).
In particular, χ(OX ) = χ(OY ).

Proof. Let f : X 99K Y be a birational morphism. By the Zariski Main Theorem

the fibers of f are connected, and we can apply Lemma 3.1.5 to get f∗ OX ' OY .
Now, let us use the projection formula:

hi (Y, OY ) = hi (Y, f∗ f ∗ OY ) = hi (X, f ∗ OY ) = hi (X, OX ),

as OX ' f ∗ OY . Hence h0,1 (X) = h0,1 (Y ) and h0,2 (X) = h0,2 (Y ), and we are
done.

Another important corollary is the following: if f : X 99K Y is a rational

map, we can consider Γf ⊆ X × Y to be the closure of the graph of f . We have
then the two natural projections p : Γf −→ X and q : Γf −→ Y .

Corollary 3.1.7. Let X be a normal projective variety, and let Y be any pro-
jective variety. Let f : X 99K Y be a rational map. Then for every point x ∈ X
the set f (x) := q(p−1 (x)) is connected.
84 Birational geometry

Proof. The map p is a birational morphism, and we can apply to it the Zariski
Main Theorem as X is normal. Hence p−1 (x) is connected for every x ∈ X, so
that f (x) is connected.

Corollary 3.1.8. Let X and Y be two projective varieties, and let (X 0 , f )

and (Y 0 , g) be their normalisations. Every morphism h : X −→ Y induces a
morphism kh : X 0 −→ Y 0 such that g ◦ kh = h ◦ f .

Proof. The morphisms f and g are birational, so we have a rational inverse f −1

and g −1 . Let kh := g −1 ◦ h ◦ f : X 0 −→ Y 0 . We need to show that kh is a
morphism: if x ∈ X 0 , we have g −1 (h(f (x))) is a finite set. This is connected by
the previous corollary as X 0 is normal, so it consists of only one point, and kh
is a morphism.

Another important application is the following:

Theorem 3.1.9. (Stein’s factorization). Let X, Y be projective varieties,

and f : X −→ Y a surjective morphism. Then there is projective variety Y 0 , a
map f 0 : X −→ Y 0 with connected fibers and a finite morphism g 0 : Y 0 −→ Y ,
such that f = g 0 ◦ f 0 . Moreover, if X is normal, then Y 0 is normal.

Corollary 3.1.10. Let f : X −→ Y be a morphism of projective varieties. If Y

is normal and the generic fiber of f is connected, then every fiber is connected.

Proof. Let us take the Stein factorisation g 0 : Y 0 −→ Y . as the generic fiber of

f is connected, the degree of g 0 is generically 1, so that g 0 is a birational map.
By the Zariski Main Theorem the fibers of g 0 are connected, hence the consist of
only 1 point, and g 0 is an isomorphism. Then f = f 0 has connected fibers.

Corollary 3.1.11. Let f : X 99K Y be a proper map, and suppose X to be

normal. Moreover let U ⊆ X be the maximal open subset over which f is
defined. Then codimX (X \ U ) ≥ 2.

Proof. In the first case, the fibers of f are connected by the Zariski Main The-
orem. If y ∈ Y is a point such that p−1 (y) consists of a single point, then we
can find an open subset V ⊆ Y , y ∈ V , such that f : f −1 (V ) −→ V is finite.
Hence X \ U is exactly the locus where the fibers have positive dimension, so
that codimX (X \ U ) ≥ 2.

3.1.3 Blow-up of a point

In this subsection, let X be a complex surface, and let q ∈ X. Moreover, let U
be an open neighborhood of x in X, and let (x, y) be local coordinates, so that
the point q corresponds to (0, 0). Define the following:

e := {((x, y), (z : w)) ∈ U × P1 | xw = yz}.

U
Contraction of curves 85

Clearly, we have a map pU : U e −→ U , which is the projection onto the first fac-
tor. Notice that if (x, y) 6= (0, 0), then p−1
U ((x, y)) = {((x, y), (x : y)}. Moreover,
−1 1
we have that pU (q) = {q} × P . Hence,

pU : p−1
U (U \ {q}) −→ U \ {q}

is an isomorphism, and p−1 1

U (q) ' P is a curve contracted by pU . Now, let us
take the gluing of X and U along X \ {q} and U
e e \ {q} ' U \ {q}. In this way
e −→ X. Notice that p
we obtain a surface Se together with a morphism p : X
−1
gives an isomorphism between X \ {q} and Xe \ p (q), and contracts the curve
1 −1
P ' p (q) to the point q.

Definition 3.1.5. The morphism p : Xe −→ X is called blow-up of X along q.

e The curve E := p−1 (q) ' P1 is called exceptional
We often write Blq (S) := S.
curve of the blow-up.

The first remark is that the blow-up of a surface X along a point q is a

birational map. Hence X e and X are birational equivalent surfaces. Moreover,
the morphism p is surjective, hence by Corollary 3.1.6 we have q(X)e = q(X),
pg (X)
e = pg (X) and χ(O e ) = χ(OX ).
X

Remark 3.1.1. If X is projective, even X e is projective. Indeed, we can define

n fn ⊆ Pn × Pn−1
the blow-up of P at the point (0 : .. : 0 : 1) as a subvariety P
in a way completely analogue to the previous one (here instead of P1 we need
Pn−1 ): simply take the subvariety defined by equations xi yj = xj yi for every
i, j = 0, ..., n − 1, where xi are coordinates for Pn and yj are coordinates for
2
Pn−1 . Next take the Segre embedding of Pn × Pn−1 in Pn +n−1 . Now, every
subvariety of Pn × Pn−1 of bihomogenous equation correspond to a projective
2
subvariety of Pn +n−1 via the Segre embedding. It is now easy to see that the
blow up of a projective variety at one point is still projective.

Definition 3.1.6. Let C be a curve in X. The strict transform of C under he

blow-up p of X along q is the closure C e of p−1 (C \ {q}).
e in X

Notice that the strict transform of a curve is always a curve. If q ∈

/ C, then
−1 ∗
C = p (C), so that C = p C. More generally, one has the following:
e e

Lemma 3.1.12. Let C be an irreducible curve passing through q with multi-

plicity m. Then
p∗ C = Ce + mE.

Proof. Clearly we have p∗ C = C e + kE for some k ∈ Z. Use local coordinates

e ∩ (U × {z 6= 0})
(x, y) around q (in the open subset U ). On the open subset U
we can then consider coordinates u = x and v = w/z. In these coordinates,
86 Birational geometry

we see that p(u, v) = (u, uv). Let f be a local equation of C at q. As C has

multiplicity m at q, we have

f = fm (x, y) + O(m),

where fm (x, y) is homogenous of degree m. Hence

f ◦ p(u, v) = fm (u, uv) + O(m) = um (fm (1, v) + O(m)).

Then the component E has multiplicity m.

We now prove the following, which we will use frequently in the following:

Proposition 3.1.13. Let X be a smooth projective surface, and p : X e −→ X

be the blow-up of X along q ∈ X, whose exceptional curve is denoted E.

1. The morphism of groups

P ic(X) ⊕ Z −→ P ic(X),
e (OX (D), n) 7→ OXe (p∗ D + nE)

is an isomorphism, and similarly for the Néron-Severi groups. In partic-

ular h1,1 (X)
e = h1,1 (X) + 1 and ρ(X)
e = ρ(X) + 1.

2. For every two divisor D, D0 on X we have

p∗ D · p∗ D0 = D · D0 , p∗ D · E = 0, E · E = −1.

3. We have KXe = p∗ KX + E.

Proof. Let us start from the second item: let C and C 0 be two curves on X such
/ C ∩ C 0 . Then
that q ∈

C · C 0 = p∗ C · p∗ C 0 = C e0 ,
e·C

where the first equality comes by Proposition 2.4.4 as p is finite of degree 1

outside q, and the second comes from Lemma 3.1.12. Moreover, it is clear that
Ce · E = 0, by projection formula, as E is contracted to q. For E · E consider the
following: let C be a curve of X passing through q with multiplicity 1. Hence
Ce meets E transversally, so that C e · E = 1. Now, by Lemma 3.1.12 we have
∗ ∗
e = p C − E, so that (p C − E) · E = 1. But now, by the projection formula
C
one has p∗ C · E = 0 as E is contracted to q, so that E 2 = −1. The remaining
cases follow from this.
Let us now show item 1: every curve on X e is either a multiple of E, or it is
not. If it is not, then it is not contracted to a point, and it is then the proper
transform of a curve on X. The map P ic(X) ⊕ Z −→ P ic(X) e is surjective.
∗
To show injectivity, simply suppose that p D + nE is linearily equivalent to 0.
Then
0 = (p∗ D + nE) · E = −n,
Contraction of curves 87

by the previous part of the proof, so that n = 0. Hence p∗ D ∼ 0, so that D ∼ 0

on X, and injectivity is shown.
To conclude, we have that KXe = p∗ KX + kE for some k ∈ Z. Now, use the
adjunction formula:

−2 = deg(KE ) = deg((KXe + E)|E ) = KXe · E + E 2 = p∗ KX · E + (k + 1)E 2 .

Now, p∗ KX · E = 0 by the projection formula, and E 2 = −1 by item 2, so that

2 = k + 1. In conclusion, k = 1 and we are done.

3.1.4 Indeterminacies
In this section we prove a fundamental result of surface theory: every rational
map is composition of blow-ups. Let me first show the relationship between
resolution of indeterminacies of a rational map, and blow-up. Let us consider
f : S 99K Pn , and let Indf be the indeterminacy locus of f . Then let f (S) :=
S \ Indf be the Zariski closure of the image of f .

Proposition 3.1.14. There is finite sequence

pn pn−1 p1
Sn −→ Sn−1 −→ ... S 0 := S,

where for each n the map pn is the blow-up of Sn−1 along a point, such that the
map
fn := f ◦ p1 ◦ ... ◦ pn : Sn −→ Pn

is a morphism.

Proof. We can assume that f (S) is not contained in a hyperplane of Pn . If

one considers a system of hyperplanes on Pn , we then get a linear system |D|
on X associated to a divisor D, whose base locus is contained in Indf . If
Indf = ∅, then we are done. So let us suppose Indf 6= ∅, and let q ∈ Indf .
Let S1 := Blq (S), and let p1 be the associated blow-up. Then p∗1 = D1 + mE1 ,
where D1 is the proper transform of D and E1 = exc(p1 ), for some integer
m1 ∈ Z. We can choose m1 so that E1 ∈ / Bs(D1 ). If Bs(D1 ) = ∅, then we are
done. If Bs(D1 ) 6= ∅, let q1 ∈ Bs(D1 ) and proceed as before. In this way we
produce a surface Sk = Blqk−1 (Sk−1 ) and pk : Sk −→ Sk−1 , and on Sk we have
Dk such that p∗ Dk−1 = Dk + mk Ek , where mk ∈ Z is such that Ek := exc(pk )
is not in the base locus of Dk . Now, by Proposition 3.1.13 we have

0 ≤ Dk2 = Dk−1
2
− m2k < Dk−1
2
,

where the first inequality comes from the fact that Dk has no curve in its base
locus (since the indeterminacy locus has codimension at least 2 by Corollary
3.1.11). Hence the sequence {Dk2 } must stabilise for some k  0. Hence for
k  0 we have that |Dk | has no base locus, and we are done.
88 Birational geometry

Example 3.1.2. 2.1.2Let X = P2 , and let q ∈ X. The projection from q gives a

map
πq : P2 99K P1 ,

which is not defined only at q. Blowing-up q in X, we easily solve the indeter-

minacy.
Example 3.1.3. Using the same notations as in Example 3.1.1, we see that the
map Q 99K P2 is not defined at 0, and that the two lines passing through 0
contained in Q are projected to two distint points A and B. Then, blowing up
Q at 0 we solve the indeterminacy, and we can show that Bl0 Q is the blow-up
of P2 at A and B.
Now, the locus where a rational map is not defined has codimension at least
2. For a smooth (or normal) surface, this implies that the indeterminacy locus
is given by a finite number of points. To solve the indeterminacies we have then
to blow them up, getting a finite number of curves which are contracted.

Lemma 3.1.15. Let X and Y be two projective surfaces, and suppose Y to be

smooth. Let f : X −→ Y be a birational map. If p ∈ Y is such that the inverse
of f is not defined at p, then f −1 (p) is a curve.

Proof. If the inverse map g : Y 99K X of f is not defined at p, this means that
f −1 (p) consists at least of 2 distinct points. By the Zariski Main Theorem the
fibers of f are connected, hence f −1 (p) is a curve.

Corollary 3.1.16. Let X, Y be two smooth projective surfaces, and let f :

X 99K Y be a birational map. If f is not defined at x ∈ X, then the inverse
map g : Y 99K X contracts a curve to x.

Proof. Let us suppose that the maximal open subset of X over which f is defined
is U . Let Γf ⊆ X × Y be the closure of the graph of f , and let p : Γf −→ X and
q : Γf −→ Y be the two projections. If f is not defined at x, then the inverse of
q (which is birational) is not defined at x. Hence q −1 (x) is a curve C by Lemma
3.1.15, which is contracted to x by q. Now, notice that g = q ◦ (q 0 )−1 , hence C
is contracted to x by g.

Here we can finally prove the main theorem on blow-ups of surfaces.

Theorem 3.1.17. Let X, Y be two smooth projective surfaces, and f : X −→ Y

a morphism which is birational. If y ∈ Y is a point where the inverse g of f is
not defined, then there is a morphism h : X −→ Bly Y which is birational and
such that f = p ◦ h, where p : Bly Y −→ Y is the blow-up morphism.

Proof. Let p−1 be the inverse of p, and let h := p−1 ◦ f . Moreover, let h−1
be the inverse of h. we want to show that h is a morphism, and to do so we
assume the contrary. Hence, by Corollary 3.1.16 there is a curve C on Bly Y
Contraction of curves 89

contracted by h−1 to a point x ∈ X. As h−1 = f −1 ◦ p, and h−1 (C) = x, we

have f −1 (p(C)) = x, so that p(C) = f (x). Hence C is a curve contracted by p.
As p contracts only its exceptional curve E, we have E = C and f (x) = y. Let
u be a local coordinate at y on Y . If f ∗ u is not a local coordinate at x on X,
we would have f ∗ u ∈ m2x . Hence p∗ u = (h−1 )∗ f ∗ u ∈ m2e for every e ∈ E where
it is defined, i. e. at every e ∈ E but a finite number. The blow-up has the
property that any local coordinate at y on Y lifts to a local coordinate at every
point of E, but one. Hence f ∗ u has to be a local coordinate at x on X. But this
implies that h is defined at x, so that h is a morphism, and we are done.

From this, one can prove the following:

Theorem 3.1.18. Let f : X −→ Y be a birational morphism between two

smooth projective surfaces. Hence f is the composition of a finite number of
blow-ups and isomorphisms.

Proof. Let us suppose f is not an isomorphism. Then there is a point y1 ∈ Y

over which the inverse map f −1 is not defined. By Theorem 3.1.17, there is
a blow-up p1 : Y1 := Bly1 Y −→ Y such that f = p1 ◦ f1 , for some birational
morphism f1 : X −→ Y1 . Notice that f contracts all those curves which f1
contracts, and contracts a curve more, namely the curve which f1 maps to
exc(p1 ). Hence the number of curves that f1 contracts is strictly smaller than
the one of the curves contracted by f . If no curve is contracted by f1 , then f1−1
is everywhere defined and f1 is an isomorphism, and we are done. Otherwise
there is a point y2 ∈ Y1 on which f1−1 is not defined, and we can proceed as
before. Notice that this procedure has to finish since the number of contracted
curves strictly decreases.

As a corollary we have the following:

Corollary 3.1.19. Let X, Y be smooth projective surfaces, and f : X 99K

Y a birational map. Hence there is a smooth projective surface Z and two
morphisms g : Z −→ X and h : Z −→ Y which are composition of blow-ups and
isomorphisms and such that h = f ◦ g.

Notice that from this it follows that if X and Y are two birational surfaces,
then q(X) = q(Y ), pg (X) = pg (Y ) and χ(OX ) = χ(OY ).

3.1.5 Castelnuovo’s contraction theorem

In this section we prove one of the main results in the classification theory
of surfaces. As we have seen in the previous section, the only birational maps
between smooth projective surfaces are isomorphisms and blow-ups along points.
If we blow up a smooth projective surface X at a point x ∈ X, we obtain a
smooth projective surface X
e on which there is a curve E, the exceptional curve
90 Birational geometry

of the blow-up, such that E 2 = −1. One of the most surprising properties of
surfaces is that every curve C such that C 2 = −1 is the exceptional curve of
some blow-up. Before proving this, let me introduce a definition:

Definition 3.1.7. A (−1)−curve on a surface X is a smooth rational curve C

such that C 2 = −1.

Theorem 3.1.20. (Castelnuovo’s contraction). Let X be a smooth pro-

jective surface, and suppose there is (−1)−curve E. Then there is a smooth
projective surface Y and a point y ∈ Y such that X ' Bly Y and E = p−1 (y),
where p : X −→ Y is the blow-up morphism.

Proof. Let H be a very ample divisor on X, and suppose that H 1 (X, OX (H)) =
(this is always possible up to changing H with mH for m  0). Let d := H · E,
and H 0 := H + dE. Then H 0 · E = H · E + dE 2 = 0 as E 2 = −1. Let us
consider the map ϕH 0 , which is defined on X \ Bs(H 0 ). We want to show that
ϕH 0 is defined on the whole X. In order to do so, we define a good basis for
H 0 (X, OX (H 0 )): let i = 1, ..., d, and consider the exact sequence of sheaves

0 −→ OX (H + (i − 1)E) −→ OX (H + iE) −→ OE (d − i) −→ 0.

This is exact as E 2 = 1 and

deg((H + iE)|E ) = (H + iE) · E = H · E + iE 2 = d − i.

Now, this exact sequence induces a long exact sequence in cohomology, and we
have

H 1 (X, OX (H + (i − 1)E) −→ H 1 (X, OX (H + iE) −→ H 1 (E, OE (d − i)).

The last term of this exact sequence is trivial for every i = 1, ..., d as E ' P1
and d − i ≥ 0. Moreover, if i = 1, then the first element is H 1 (X, OX (H)) = 0
by hypothesis, so that H 1 (X, OX (H + E)) = 0. Using this process in successive
steps, we finally get that H 1 (X, OX (H + iE)) = 0 for every i = 1, ..., d. Hence
the restriction map

H 0 (X, OX (H + iE)) −→ H 0 (E, OE (d − i))

is surjective for every i = 1, ..., d. Then

H 0 (X, OX (H + dE)) ' H 0 (X, OX (H)) ⊕ H 0 (E, OE (1)) ⊕ ... ⊕ H 0 (E, OX (d)).

Hence to determine a basis of H 0 (X, OX (H 0 )) we need to fix a basis {s0 , ..., sn }

of H 0 (X, OX (H)), and for every i = 1, ..., d a set {si,0 , ..., si,d−i } of sections
of H 0 (X, OX (H + iE)) which restricts to a basis of H 0 (E, OE (d − i)) (here
Contraction of curves 91

h0 (P1 , O(d − i)) = d − i + 1). Hence, if s ∈ H 0 (X, OX (E)) (it exists as E is

effective), we get the basis

{sd s0 , ..., sd sn , sd−1 s1,0 , ..., sd−1 s1,d−1 , ..., ssd−1,0 , ssd−1,1 , sd,0 }

for H 0 (X, OX (H 0 )). Now, let’s look at the map ϕH 0 associated to this basis: if
x ∈ E, then s(x) = 0 and sd,0 (x) = 1, so that ϕH 0 (E) = (0 : 0 : ... : 0 : 1) := P ,
and E is contracted to a point. If x ∈ / E, then s(x) 6= 0, and ϕH 0 is defined at
x: indeed H has no base locus, so that there is j = 1, ..., n such that sj (x) 6= 0.
Hence the map ϕH 0 is everywhere well defined, and notice that if x ∈ / E, then
ϕH 0 (x) 6= P . . Moreover, it is an isomorphism between X \ E and ϕH 0 (X \ E) =
ϕH 0 (X) \ {P }. Indeed, outside of E we have that H = H 0 , so that H 0 separates
points outside E as H does it (more precisely, if x, y ∈ X \ E are such that
ϕH 0 (x) = ϕH 0 (y), then sj (x) = sj (y) for every j = 0, ..., n, as sd (x), sd (y) 6= 0;
but as H is very ample, it separates points, so that this implies x = y).
Now, let us show that P is smooth in ϕH 0 (X). As sd,0 gives a basis for
H (E, OE ), there is an open neighborhood U of E in X over which sd,0 6= 0.
0

The sheaf OU (−E) has then two sections t0 := sd−1,0 /sd,0 and t1 := sd−1,1 /sd,0 .
Clearly they restrict to a basis of H 0 (E, OE (1)). Up to shrinking U , we can
suppose that t0 and t1 are not simultaneously trivial on U . Hence we get a map

z2 : U −→ P1 , z2 (x) := (t0 (x) : t1 (x)).

Moreover, let

z1 : U −→ C2 , z1 (x) := (s(x)t0 (x), s(x)t1 (x)),

which is trivial exactly on E. If one looks at the map

(z1 , z2 ) : U −→ C2 × P1 , (z1 , z2 )(x) = (z1 (x), z2 (x)),

we easily see that it maps to the blow-up C f2 of C2 at the point (0, 0) inside
2 1 f2 −→ C2 be the
f2 . Let now p : C
C × P . Hence we get a map z : U −→ C
blow-up morphism, and let ze : ϕH 0 (U ) −→ C2 be the map induced by z. Hence
we have p ◦ z = ze ◦ ϕH 0 .
Let now Y := ϕH 0 (X), and notice that ϕH 0 (U ) ∩ Y ⊆ Y is an open subset.
Observe the following: the map z maps E to the exceptional divisor F of p,
and it is an isomorphism. Using the complex topology, we can see that z is
an isomorphism locally around every point of E. Indeed, let (u, v) coordinates
on C2 , and let (U : V ) be coordinates on P1 . The local equation of C f2 is
uV = U v. Let q ∈ E, and let suppose z(q) = ((0, 1), (0 : 1)). Hence v and
U/V are local coordinates on Cf2 around z(q), and we easily see that z ∗ v = st1
and z U/V = t0 /t1 . Then z v vanishes with multiplicity 1 on E, while z ∗ U/V
∗ ∗

is a local coordinate around q. Hence, in a neighborhood of q where these two

92 Birational geometry

functions are defined, they give a coordinate system, and z is an isomorphism.

Hence, as E is compact in X, there is an open neighborhood U of E which maps
isomorphically to an open neighborhood W of F . Let now

−1
ϕH 0 ◦ z|W : W −→ ϕH 0 (U ).

As this contracts the exceptional curve F in W , by Theorem 3.1.17 there must

−1
be a map g : p(W ) −→ ϕH 0 (U ) such that g ◦ p = ϕH 0 ◦ z|W . Hence, by
−1
commutativity properties, one easily gets g = ze . Hence, we see that ϕH 0 (U )
is isomorphic to p(W ), which is smooth, so that P is a smooth point in Y . It
is now clear that ϕH 0 : X −→ Y is exactly the blow-up of Y along P , and the
exceptional locus is E.

The main conclusion of this theorem is the following: whenever on a smooth

surface X there is a (−1)−curve, we can contract it as it is the product of a blow-
up. The obtained surface Y is again smooth and projective, but ρ(Y ) = ρ(X)−1
be Proposition 3.1.13. Let me define the following:

Definition 3.1.8. A surface X is called minimal if every birational morphism

f : X −→ Y is an isomorphism. A minimal model for a surface X is a surface
X 0 which is minimal and birational to X.

Here is an important corollary of the Castelnuovo Contraction Theorem:

Corollary 3.1.21. A surface is minimal if and only if it has no (−1)−curve.

In particular, every surface has a minimal model.

Proof. Let X be a smooth projective surface, and let us suppose it is minimal.

If there is a (−1)−curve E on E, by the Castelnuovo Contraction Theorem there
is a smooth projective surface Y and y ∈ Y such that X = Bly Y . In particular,
there is a birational morphism p : X −→ Y which is not an isomorphism, hence
X cannot be minimal, and we get a contradiction. Hence every minimal smooth
projective surface has no (−1)−curve.
For the converse, let us suppose that X has no (−1)−curve, but X is not
minimal. Then there is a birational morphism f : X −→ Y to a smooth
projective surface Y which is not an isomorphism. Hence, by Theorem 3.1.18
f is the composition of isomorphisms and blow-ups, and there is at least one
blow-up. As f is defined on X, then this introduces an exceptional curve on X,
which is a (−1)−curve. This is a contradiction, and we are done.
Let us prove the remaining part. Let X be a smooth projective surface. If
it is minimal, then it is a minimal model for itself. If it is not minimal, then
there is a (−1)−curve E on X. By the Castelnuovo Contraction Theorem, E is
the exceptional curve of a blow-up, so that contracting it we get a new smooth
projective surface X1 , such that ρ(X) = ρ(X1 ) + 1 by Proposition 3.1.13. If X1
Kodaira dimension 93

is minimal, we are done. Otherwise one can still contract exceptional curves,
producing a sequence of blow-ups
p0 p1 pn−1 pn
X = X0 −→ X1 −→ ... −→ Xn −→ ...

Notice that as ρ(Xn ) = ρ(X) − n, and ρ(Xn ) > 0, this sequence has to termi-
nate for some n0 < ρ(X). Hence Xn0 cannot contain (−1)−curves, being then
minimal. Hence it is a minimal model for X.

This is a very important point: as every surface admits a minimal model,

once a birational classification of minimal surfaces is completed, even the bira-
tional classification of surfaces is completed. The classification we are going to
present is hence only a classification of minimal surfaces. There is anyway an
important point to remark: every surfaces admits a minimal model. How many
are them? Let us present a simple example:
Example 3.1.4. Let X := P2 and Y := P1 ×P1 . In Example 3.1.1 we have shown
that these two smooth projective surfaces are birational and not isomorphic.
Anyway, they are minimal. Indeed, on X every irreducible curve C is such that
C 2 > 0, hence there is no (−1)−curve. Similarily, every curve C on Y is such
that C 2 is even, so there is no (−1)−curve on Y , and these two surfaces are
minimal. Hence, if S is a smooth projective surface birational to P2 it can have
at least two non-isomorphic birational models.

3.2 Kodaira dimension

In this section we introduce the notion of Kodaira dimension of a normal pro-
jective variety. In the case of surfaces, it respresents an important birational
invariant, which is used to give a first rough classification. In the following, let
X be a normal projective variety, and let D be an effective Cartier divisor on
X. Consider the following field

Q(X, D) := {s/t | s, t ∈ H 0 (X, OX (kD)), t 6= 0, k ≥ 0},

which is the homogenous field of fractions of the ring

M
R(X, D) := H 0 (X, OX (kD)).
k≥0

Notice that if D is a hyperplane section, then H 0 (X, OX (kD)) is given by ho-

mogenous regular functions on X of degree k, so that Q(X, D) = MX (X), the
field of functions of X. More generally, we have the following:

Lemma 3.2.1. Let X be a normal projective variety, and let D be an effective

divisor. The field Q(X, D) is algebraically closed in MX (X). In particular, its
transcendence degree d is finite and 0 ≤ d ≤ dim(X).
94 Birational geometry

Proof. Let f be a rational function on X, and suppose that there are a1 , ...an ∈
Q(X, D) such that f n +a1 f n−1 +...+an = 0. We can suppose ai = si /t for some
si , t ∈ H 0 (X, OX (kD) for every i = 1, ..., n. Hence, h := f t is a meromorphic
section of OX (kD) verifying

hn + s1 hn−1 + ... + sn = 0,

where si are all regular. Let now x ∈ X: as X is normal, OX,x is a normal

ring, and si,x ∈ OX,x . Hence hx ∈ OX,x for every x ∈ X, so that h is a regular
section of OX (kD). Hence f = h/t ∈ Q(X, D), and we are done.

Notice that id D is not effective, we cannot define Q(X, D) in the same way.
If for every k ∈ N we have h0 (X, OX (kD)) = 0, then R(X, D) = Q(X, D) = C.
Otherwise, consider the set

N(D) := {k ∈ N | h0 (X, OX (kD)) > 0}.

Then we can define Q(X, D) to be the homogenous quotient field of the ring
k∈N(D) H (X, OX (D)). Obviously, Lemma 3.2.1 holds in this
0
L
R(X, D) :=
case.

3.2.1 Definition and geometrical interpretation

We can then give the following definition:

Definition 3.2.1. Let X be a normal projective variety, and let D be a divisor.

If N(D) 6= ∅, we call Kodaira dimension of D the degree of transcendence κ(D)
of Q(X, D). Otherwise we define κ(D) := −∞. The Kodaira dimension of X is

κ(X) := κ(KX ) ∈ {−∞, 0, ..., dim(X)}.

Remark 3.2.1. Notice that if X is a normal projective surface, and KX is nef,

then κ(X) ≥ 0. Indeed, as KX is nef, there is m  0 such that mKX is effective.
Hence h0 (X, OX (mKX )) > 0, and κ(X) ≥ 0.
This abstract definition is strictly related to the geometry of X. Let us take
k ∈ N(D), so that h0 (X, OX (kD)) > 0. Hence we can define the morphism

ϕkD : X 99K PNkD ,

where NkD := h0 (X, OX (kD)) − 1. Let UkD := X \ Bs(kD), and let

WkD := ϕkD (UkD ).

This is a projective variety whose fields of functions is easily described in terms

of sections of kD: indeed, if {s0 , ..., sNkD } is a basis for H 0 (X, OX (kD)), then
it is easy to see that

M (WkD ) = C(s1 /s0 , ..., sNkD /s0 ).

As a corollary of this, we see that Q(X, D) = k∈N(D) M (WkD ).

S
Kodaira dimension 95

Proposition 3.2.2. Let X be a normal projective variety, and let D be a divisor

such that N(D) 6= ∅. Then

κ(D) = max{dim(WkD ) | k ∈ N(D)}.

Proof. Let us suppose that N(D) = N, and let k ∈ N. The natural injec-
tion OX (kD) −→ OX ((k + 1)D) induces a natural inclusion H 0 (X, OX (kD)) ⊆
H 0 (X, OX ((k + 1)D)). By the previous description of M (WkD ) we then get an
inclusion ik : M (WkD ) −→ M (W(k+1)D ). Now, the quotient field M (X) of X
is finitely generated over C (indeed, the transcendence degree of M (X) is the di-
mension of X), hence Q(X, D) is finitely generated over C. So the sequence of in-
clusions ik must stabilize, hence there is k0 ∈ N such that Q(X, D) = M (Wk0 D ).
Their transcendence degree is then equal, so that κ(D) = dim(Wk0 D ). Now,
simply notice that as ik is an inclusion, we have dim(WkD ) ≤ dim(W(k+1)D )
for every k ∈ N, so that we finally prove the proposition if N(D) = N. The
remaining case is similar.

Corollary 3.2.3. If X is a normal projective variety such that KX is ample,

then κ(X) = dim(X).

Proof. If KX is ample, there is m  0 such that mKX is very ample, and

the map ϕmKX is a closed embedding. Hence dim(X) = dim(WmKX ), so that
κ(X) = dim(X).

Another important result is the following:

Proposition 3.2.4. Let X be a normal projective surface, and let D be a divisor

such that N(D) 6= ∅. Then the generic fiber of ϕkD is connected.

Proof. Let me start by supposing that there is k ∈ N(D) such that ϕkD is
everywhere defined on X, i. e. Bs(kD) = ∅. Let f := ϕkD : X −→ WkD . We
can consider the Stein factorization of f , so that there is a normal projective
variety Y and two morphisms
f0 g
X −→ Y −→ WkD ,

such that g ◦ f 0 = f , g is finite and f 0 has connected fibers. We have then

M (WkD ) ⊆ M (Y ) ⊆ M (X). Moreover, notice that if kD has no base locus,
the same is true for every k 0 D, with k 0 ≥ k: hence we can suppose k  0 so that
M (WkD ) = Q(X, D). As Q(X, D) is algebraically closed in M (X), we then
have M (Y ) = M (WkD ). The morphism g is then birational by Proposition
3.1.2. Generically, then, the fibers of f and f 0 are equal, so that the generic
fiber of f is connected.
In general, the situation is more complicated. Let k ∈ N(D), and let f :=
ϕkD : by Corollary 3.1.11, as X is normal we have that codimX (Bs(kD)) ≥ 2,
96 Birational geometry

since Bs(kD) = Indf , so that Bs(kD) is given by a finite number of points. By

Proposition 3.1.14 we know that to solve the indeterminacies of f we just need
to blow up X in Bs(kD), and we get a projective surface X e together with a
morphism p : Xe −→ X such that the map

fe := f ◦ p : X
e −→ WkD

is a morphism. As X is normal, the fibers of p are connected by the Zariski

Main Theorem. Moreover, we have

p∗ : H 0 (X, OX (kD)) −→ H 0 (X,

e p∗ OX (kD))

is an isomorphism. Up to replacing X e with its normalisation, we can suppose X e

to be normal. Hence we can replace X with another smooth projective surface
X 0 , and the morphism f is replaced by f 0 , which is defined by means of the same
basis (up to the identification of global sections). We can apply the previous
part and guarantee that the generic fiber of ϕkD is always connected, for every
k ∈ N(D).

We conclude with the following, which is a beautiful equivalent definition of

the Kodaira dimension. Let me introduce the following:

Definition 3.2.2. Let X be a normal projective variety. The m−th plurigenus

of X is
Pm (X) := h0 (X, OX (mKX )).

Proposition 3.2.5. Let X be a normal projective variety, and let D be a divisor

such that N(D) 6= ∅. There are α, β ∈ R>0 such that if k  0 we have

αk κ(D) ≤ h0 (X, OX (kD)) ≤ βk κ(D) .

In particular, we have Pm (X) ∼ mκ(X) .

Proof. Up to replace D with some multiple and X with some blow-up, we can
suppose that f := ϕD is everywhere defined, and that dim(W ) = κ(D), where
W = WD . Let F be the fixed part of |D|, so that D = F + f ∗ H, where H is a
hyperplane section of W . We have the following sequence of inequalities:

h0 (X, OX (kD)) ≥ h0 (X, f ∗ OW (kH)) ≥ h0 (W, OW (kH)) ≥ αk κ(D) ,

by the Hilbert polynomial.

For the remaining inequality, let us first suppose that the fixed part F does
not contain any component mapping surjectively on W . Then F is contained
in the pull-back of some divisor G on W , so that

h0 (X, OX (kD) ≤ h0 (X, f ∗ OW (kH + kG).

Kodaira dimension 97

Up to adding a very ample divisor to G, we can suppose that H + G is very

ample. By the Hilbert polynomial we then get

h0 (X, OW (kH + kG)) ≤ βk κ(D) ,

and we are done. So now suppose that F contains a component which surjects
onto W . Let us write F = F 0 + F 00 , where F 0 is the maximal divisor such that
f (F 0 ) = W . Clearly we have |D| = |D − F 0 | + F 0 . If |kD| = |kD − kF 0 | + kF 0 ,
we can just replace D by D − F 0 in the preceding part, and we are done.
We then need to show that |kD| = |kD − kF 0 | + kF 0 for every k. Clearly
we have |kD − kF 0 | + kF 0 ⊆ |kD|, and suppose there is E ∈ |kD|, but E ∈ /
|kD − kF 0 | + kF 0 . Let G be the maximal divisor such that E ∈ |kD − G| + G
and which does not contain kF 0 . Then we can write E = E 0 + G, where
E 0 ∈ |kD − G|, and E 0 passes through some points of F 0 , but not all. Now,
f (F 0 ) = W , so the fibers of f meet F 0 . If x ∈ E 0 , but x ∈/ F 0 , we have that
f −1 (f (x)) intersects F 0 in some point. If y ∈ f −1 f (x) ∩ F 0 , we have then
ϕkD (x) 6= ϕkD (y).
The generic fiber of f and of ϕkD are connected by Proposition 3.2.4, and
the seconds are contained in the firsts. as ϕkD is not constant on the generic
fiber of f , then the generic fiber of ϕkD is a lower dimensional subvariety of the
generic fiber of f . Now, since

dim(WkD ) + dim(ϕ−1
kD (pt)) = dim(X) = dim(W ) + dim(f
−1
(pt)),

we get that dim(WkD ) > dim(W ), which is impossible as dim(W ) = κ(D) by

hypothesis. Hence |kD| = |kD − kF 0 | + kF 0 for every k, and we are done.

Here is a basic property of the Kodaira dimension:

Proposition 3.2.6. If X and Y are two birational normal projective surfaces,

then Pm (X) = Pm (Y ) for every m > 0, and κ(X) = κ(Y ).

Proof. Let f : X 99K Y be a birational map. By Corollary 3.1.19 there is

projective surface Z and two maps g : Z −→ X and h : Z −→ Y which are
compositions of blow-ups and isomorphisms, and such that f ◦ g = h. If we are
able to show that the Kodaira dimension does not change under blow-up, we
e −→ X
are then done. Now, let X be a normal projective surface, and let p : X
be the blow-up of X along x ∈ X, and let E be the exceptional divisor. By
Proposition 3.1.13 we have that KXe = p∗ KX + E.
Let D be an effective divisor on X
e which is linear equivalent to mK e for
X
some m ∈ N. Then we have

D · E = mKXe · E = mp∗ KX · E + mE 2 = −m.

98 Birational geometry

e we can write it as D = D0 + kE, where k ≥ 0 and D0 is

As D is a divisor on X,
e which does not contain E. Hence D0 = D − kE, and
an effective divisor on X
as D0 and E are effective and E is not contained in D0 we have

0 ≤ D0 · E = D · E − kE 2 = k − m.

Hence k ≥ m, so that the divisor D − mE is still effective: indeed D − mE =

D0 + (k − m)E, where D0 is effective and k ≥ m. As D ∼ mKXe , we get
D − mE ∼ mKXe − mE. The map

|mKEe | −→ |m(KXe − E)|, D 7→ D − mE

is then well-defined and it is clearly an isomorphism. Hence

e O e (mK e )) = h0 (X,
h0 (X, e O e (m(K e − E))) =
X X X X

e p∗ OX (mKX )) = h0 (X, OX (mKX )).

= h0 (X,

Hence κ(X)
e = κ(X), and we are done.

The final result we need is the following:

Proposition 3.2.7. Let X and Y be to smooth projective varieties, ad suppose

that f : X −→ Y is an unramified covering. If D is a divisor on Y , then
κ(D) = κ(f ∗ D). In particular, κ(X) = κ(Y ).

Proof. Has f is an unramified covering, one has f ∗ KY = KX . Hence the second

part is a consequence of the first one. Now, f is surjective, so that the induced
map
f ∗ : H 0 (Y, OY (D)) −→ H 0 (X, f ∗ OY (D))

is injective. Hence κ(D) ≤ κ(f ∗ (D)). We then just need to show the converse.
As f is a covering, the map f∗ : π1 (Y ) −→ π1 (X) is injective, and there is
a normal subgroup N
π1 (X) such that N ⊆ f∗ (π1 (Y )), and such that the
quotient π1 (X)/N is a finite group occurring as a group of desk transformations
of a Galois covering f 0 : X 0 −→ Y factoring f . One the reduces to the case
where f is a Galois covering.
Clearly, if κ(f ∗ (D)) = −∞, then κ(D) = −∞, so we assume κ(f ∗ (D)) ≥ 0.
Let k  0 such that dim(Wf ∗ (kD) ) = κ(f ∗ (D)). If one fixes a basis {s0 , ..., sn }
of H 0 (X, f ∗ OY (kD)), we have that κ(f ∗ (D)) is the transcendence degree of
L := C(s1 /s0 , ..., sn /s0 ). As f : X −→ Y is a Galois cover, let G be its Galois
group, which acts on X, and hence on L, and consider the invariant subfield
K ⊆ L. Consider the following:
Y
(x − g ∗ (si /s0 )) = xm + a1 (si /s0 )xm−1 + ... + am (si /s0 ),
g∈G
Kodaira dimension 99

for every i = 1, ..., n. Hence the field K, which is G−invariant, is generated by

{a1 (s1 /s0 ), ..., a1 (sn /s0 ), ..., am (s1 /s0 ), ..., am (sn /s0 )}.

Notice that t0 := g∈G g ∗ (s0 ) is a G−invariant section of f ∗ OY (kmD), and

Q

similarily for t0 aj (si /s0 ). Hence, they define sections of OW (km), where W =
im(ϕkD ). As aj (si /s0 ) is a quotient of two G−invariant global sections of
f ∗ OY (kmD), they define sections of OWkmD , so that there is finally a canonical
inclusion of K inside the field Q of functions of WkmD . Hence the transcendence
degree of K is at most the transcendence degree of Q, which is κ(D). As the
transcendence degree of L is κ(f ∗ (D)), and it is at most the transcendence
degree of K, we are done.

To conclude the section, we give some examples:

Example 3.2.1. The first example I describe is the case of a Riemann surface
X. The degree of the canonical divisor is very easy to compute, as we have

deg(KX ) = 2g(X) − 2,

where g(X) is the genus of X. Using Riemann-Roch for curves and Serre’s
duality, this is easy to prove: indeed, hi (X, OX ) = h1−i (X, OX (KX )), so that
χ(OX ) = −χ(KK ). Hence

deg(KX ) − g(X) + 1 = χ(KX ) = −χ(OX ) = −g(X) + 1,

so that deg(KX ) = 2g(X) − 2.

If g(X) = 0, i. e. X ' P1 , then deg(mKX ) = −2m < 0, so that Pm (X) = 0
for every m > 0. This implies that κ(P1 ) = −∞.
If g(X) = 1, i. e. X is an elliptic curve, then deg(mKX ) = 0 for every
m > 0. One can actually show more, namely that OX (KX ) ' OX , so that
Pm (X) = 1 for every m > 0. Hence κ(X) = 0.
If g(X) ≥ 2, then deg(KX ) > 0. Hence KX is ample, so that κ(X) = 1.

Example 3.2.2. The Kodaira dimension of P2 is κ(P2 ) = −∞. Indeed, we have

OP2 (mKP2 ) = O(−3m), so that Pm (P2 ) = 0 for every m > 0. As the Kodaira
dimension is a birational invariant, we see that κ(P1 × P1 ) = −∞.

Example 3.2.3. Let X be a generic hypersurface of degree 4 in P3 , i. e. X ∈

|O(4)|. As it is generic, by Bertini’s Theorem X is smooth, hence normal.
Moreover, we can use the adjunction formula to calculate the canonical bundle
of X: we have OX (KX ) ' OP3 (KP3 + X)|X . But now notice that OP3 (KP3 ) =
O(−4), so that KX = 0. Hence Pm (X) = 1 for every m > 0. By Proposition
3.2.5, we then get κ(X) = 0.
100 Birational geometry

3.2.2 Kodaira dimension and nefness

We want to give a precise connection between the Kodaira dimension and the
nefness of the canonical bundle. we have already shown that if X is a smooth
projective surface and KX is nef, then κ(X) ≥ 0, and that if KX is ample, then
κ(X) = 2. It then remains to look at the cases where KX is not nef. The main
result is the following:

Proposition 3.2.8. Let X be a smooth projective surface.

1. If there is an irreducible curve C on X such that KX · C < 0 and C 2 ≥ 0,

then Pm (X) = 0 for every m > 0.

2. If there is m > 0 such that Pm (X) 6= 0, and there is an irreducible curve

C on X such that KX · C < 0, then C is a (−1)−curve.

Proof. For the first item, let me suppose there is m > 0 such that Pm (X) 6= 0.
Then |mKX | 6= ∅, and let D ∈ |mKX |. Clearly, one can write D = nC + R,
where k ∈ Z and R is an effective divisor which does not contain C. Since
m > 0 and KX · C < 0, we get D · C < 0, so that k 6= 0 (otherwise D · C = 0).
Moreover, as D is effective, we need k ≥ 0, so that finally k ≥ 1. But now

0 > mKX · C = D · C = kC 2 + R · C ≥ kC 2 ≥ 0,

and we get a contradiction. Hence Pm (X) = 0 for every m > 0.

For the second item, we can have only two possibilities, namely KX · C ≤ −2
or KX · C = −1. In the first case, use the genus formula:
1
0 ≤ g(C) = 1 + (KX · C + C 2 ) ≤ C 2 ,
2
so that one can apply item 1 to get that Pm (X) = 0 for every m > 0. This is in
contradiction with the hypothesis, so that we need KX · C = −1. By the first
item, one needs C 2 < 0, and by the genus formula we have
1
0 ≤ g(C) = 1 + (C 2 − 1),
2
so that C 2 ≥ −1. Hence C 2 = −1, and the genus formula gives g(C) = 0, so
that C is a (−1)−curve.

We can use this result to prove the following:

Proposition 3.2.9. Let X be a smooth projective surface, and suppose that

KX is not nef. Hence either X is not minimal, or κ(X) = −∞.

Proof. As KX is not nef, there must be an irreducible curve C on X such that

KX ·C < 0. Let us suppose that κ(X) ≥ 0. Then there must be m > 0 such that
Pm (X) 6= 0, and we can apply Proposition 3.2.8 to get that C is a (−1)−curve.
Hence X is not minimal.
Kodaira dimension 101

Corollary 3.2.10. Let X be a minimal surface. Then κ(X) = −∞ if and only

if KX is not nef.

Proof. If KX is not nef and X is minimal, then κ(X) = −∞ by Proposition

3.2.9. If κ(X) = −∞, then KX cannot be nef: by Remark 3.2.1, if KX is nef
we have κ(X) ≥ 0.

There is another important result we need to show, and as a corollary of

it we prove that every smooth projective surface whose Kodaira dimension is
non-negative has a unique minimal model.

Proposition 3.2.11. Let X and Y be two smooth projective surfaces, and let
f : X 99K Y be a birational map.

1. If KX is nef, then f : X −→ Y is a morphism.

2. If KX and KY are nef, then f : X −→ Y is an isomorphism.

Proof. Before, we need to prove the following:

Lemma 3.2.12. Let S be a surface, and let f : S 0 −→ S be a birational

morphism. If KS is nef and C 0 is a curve on S 0 which is not contracted by f ,
then KS 0 · C 0 ≥ 0.

Proof. Since every birational morphism f is a composition of blow-ups and

isomorphisms, we just need to look at the case where f s the blow-up of S at
some point q. Let E be the exceptional divisor on S 0 , and let us suppose that
C 0 is a curve such that C := f (C 0 ) is a curve on X. Hence we have

KS 0 · C 0 = (f ∗ KS + E) · (f ∗ C − mE) = KS · C + m ≥ 0,

as m is the multiplicity of C at q (hence m ≥ 0).

Now, let f : X 99K Y be a birational map. Let us suppose that KX is nef,

but f is not a morphism: hence we blow up X along at least one point in order
to solve the indeterminacies of f . Let E be an exceptional curve on the blow-up
Xe of X: then K e · E = −1. Indeed, by the genus formula and the fact that E
X
is a (−1)−curve we have
1 1
0 = g(E) = 1 + (KXe · E + E 2 = (KXe · E + 1).
2 2
e −→ X is obtained by the minimal number of blow-ups
Now, notice that if p : X
so that f ◦p : X −→ Y is a morphism, the exceptional divisor of the last blow-up
e
is not contracted by f , so that by Lemma 3.2.12 we have a contradiction. In
conclusion, f is a morphism
For the remaining item, we just apply item 1 to f −1 : Y 99K X: as KY is
nef, it is a morphism, which is inverse to f , so that f is an isomorphism.
102 Birational geometry

Corollary 3.2.13. Let X be a smooth projective surface such that κ(X) ≥ 0.

Then X has a unique (up to isomorphism) minimal model.

Proof. Let X 0 and X 00 be two minimal models for X. In particular, X 0 and

X 00 are birational to X, and hence there is a birational map f : X 0 99K X 00 .
As the Kodaira dimension is a birational invariant, we have that κ(X 0 ) =
κ(X 00 ) = κ(X) ≥ 0. Moreover, as X 0 and X 00 are minimal, the fact that
κ(X 0 ) = κ(X 00 ) ≥ 0 implies that KX 0 and KX 00 are nef: indeed, if KX 0 is not
nef, as X 0 is minimal this implies that κ(X 0 ) = −∞ by Proposition 3.2.9. Hence
we can apply Proposition 3.2.11, and f is an isomorphism.

3.3 Albanese tori

This section is devoted to introduce a very general tool in algebraic geometry,
which is the Albanese torus of every complex manifold. This will be used later in
the proof of the Enriques-Castelnuovo birational classification of smooth projec-
tive surfaces. Let me start with the following: let V be a complex vector space
of dimension d, and let Γ be a maximal rank lattice in V . Let T := V /Γ, which
has the structure of compact complex manifold, and which is called complex
torus of dimension d.

Lemma 3.3.1. Let γ ∈ Γ, and let tT (γ) be the algebraic 1−cycle on T defined
by the line [0, γ] ⊆ V . Then the map

tT : Γ −→ H1 (T, Z), γ 7→ tT (γ)

is an isomorphism. The map

τT : V ∗ −→ H 0 (T, ΩT ), τT (f ) := df

is an isomorphism. Moreover, for every γ ∈ Γ and every f ∈ V ∗ we have

R
τ (f ) = f (γ).
tT (γ) T

Using this lemma, one can easily produce a morphism

j : H1 (X, Z) −→ H 0 (X, ΩX )∗ .

Indeed, if α ∈ H1 (X, Z), by Lemma 3.3.1 there is γ ∈ Γ such that α = tT (γ).

Now simply define
Z
0
j(α) : H (X, ΩX ) −→ C, j(α)(ω) := ω.
γ

The following definition hence makes sense:

Definition 3.3.1. The Albanese torus of a compact Kähler manifold X is

Alb(X) := H 0 (X, ΩX )∗ /im(H1 (X, Z)).

Albanese tori 103

Proposition 3.3.2. If X is a compact Kähler manifold, then Alb(X) is a com-

plex torus of dimension q(X).

Proof. One just needs to verify that im(H1 (X, Z)) is a maximal rank lattice in
H 0 (X, ΩX )∗ . To do this, first of all notice that as X is compact and Kähler, then
the Hodge decomposition shows that b1 (X) = 2q(X). Let {ω1 , ..., ωq(X) } be a
basis for H 0 (X, ΩX ), and let {α1 , ..., α2q(X) } be a basis for H1 (X, Z). For every
i = 1, ..., 2q(X), let vi ∈ Cq(X) be the complex vector whose j−th component
R
is t−1 (αi ) ωj . If these vectors are linearily dependent over R, then there are
T
non-trivial λ1 , ..., λ2q(X) ∈ R such that for every j = 1, ..., q(X) we have
2q(X) Z
X
λi ωj = 0.
i=1 t−1
T (αi )

Hence for every ω ∈ H 0 (X, ΩX ) we have

2q(X) Z
X
λi ω = 0.
i=1 t−1
T (αi )

As this is over R, this equality holds for ω too. By the Hodge decomposition,
P2q(X) R
the linear functional i=1 λi t−1 (αi ) is then trivial on H 1 (X, C). As this is
T
P2q(X)
the dual of i=1 λi αi , we finally get λi = 0 for every i = 1, ..., 2q(X), so that
all the vectors vi are linearily independent over R. But clearly this means that
im(H1 (X, Z)) is a maximal rank lattice of H 0 (X, ΩX )∗ , and we are done.

The main property of the Albanese torus of X is that one can define the
following map: fix a point x0 ∈ X. For every point x ∈ X we can consider any
path γ(x0 , x) from x0 to x, and we can define
Z
aX : X −→ Alb(X), aX (x) := .
γ(x0 ,x)

This is a well defined map (it does not depend on γ(x0 , x)), and moreover it is
holomorphic.

Definition 3.3.2. The map aX is the Albanese map of X.

Proposition 3.3.3. The map a∗X : H 0 (Alb(X), ΩAlb(X) ) −→ H 0 (X, ΩX ) in-

−1
duced by aX is an isomorphism. Moreover, we have that a∗X = τAlb(X) .
−1
Proof. As τAlb(X) is an isomorphism, we just need to show that a∗X = τAlb(X) .
0 ∗ ∗
Let ω ∈ H (X, ΩX ): we show that aX (τAlb(X) (ω)) = ω. As aX (τAlb(X) (ω)) is
a holomorphic 1−form on X, we can evaluate it at every point x ∈ X. Let U
be an open neighborhood of x in X, and let q : H 0 (X, ΩX )∗ −→ Alb(X) be the
projection. Hence

a∗X (τAlb(X) (ω))(x) = a∗X (q ∗ (τAlb(X) (ω)))(x) = a∗X (d((ω, .)))(x) =

104 Birational geometry

Z 
= d((ω, aX (x))) = d ω = ω(x).
γ(x0 ,x)

Example 3.3.1. Let X be a Riemann surface: then Alb(X) = Jac(X). Indeed

Alb(X) = H 0 (X, OX (KX ))∗ /im(H1 (X, Z)),

and
Jac(X) = H 1 (X, OX )/im(H 1 (X, Z)).

But now using Serre’s duality one sees that H 0 (X, OX (KX ))∗ ' H 1 (X, OX ).
Moreover, this respects the duality between H1 (X, Z) and H 1 (X, Z), and we are
done. Moreover, one has that the Albanese map is the Abel-Jacobi map for X.
The Albanese variety and the Albanese map satisfy a universal property as
follows:

Theorem 3.3.4. (Universal property of the Albanese). Let X be a com-

pact Kähler manifold, T a complex torus and f : X −→ T a morphism. Then
there is a (unique) map g : Alb(X) −→ T such that f = g ◦ aX .

Proof. The uniqueness of g is given by the commutativity. Let us prove the

following:

Lemma 3.3.5. Let V and V 0 be two complex vector spaces, and let Γ ⊆ V and
Γ0 ⊆ V 0 be two maximal rank lattices. Moreover, let T := V /Γ and T 0 := V 0 /Γ0 .

1. Every linear map H : V −→ V 0 such that H(Γ) ⊆ Γ0 induces a morphism

h : T −→ T 0 .

2. Every morphism h : T −→ T 0 is induced by the composition of a transla-

tion and of a linear map H : V −→ V 0 such that H(Γ) ⊆ Γ0 . If H t is the
transpose of H, then we have τT 0 ◦ H t = h∗ ◦ τT .

Proof. Item 1 is clear. For item 2: by the universal property of liftings, the map
h lifts to a holomorphic map e h : V −→ V 0 such that for every v ∈ V and every
γ ∈ Γ we have e h(v + γ) − eh(v) ∈ Γ0 . As e
h is holomorphic and Γ0 is discrete,
the value of eh(v + γ) − e
h(v) does not depend on v, hence all partial derivatives
of h are invariant under translation via an element in Γ. They define hence a
e
holomorphic function on T . As T is compact, they have to be constant, so that
h(v) = H(v) + k, where k is a constant vector and H : V −→ V 0 is a linear map
e
such that H(Γ) ⊆ Γ0 . The rest is clear.

Let us now consider f : X −→ T , where T = V /Γ for some complex vector

space V and some maximal rank lattice Γ in V . Consider the linear map

G := τAlb(X) ◦ f ∗ : H 0 (T, ΩT ) ' V −→ H 0 (Alb(X), ΩAlb(X) ).

Albanese tori 105

−1
By Proposition 3.3.3 we have τAlb(X) = a∗X , so that we finally get f ∗ = a∗X ◦ G.
0 ∗ −1
Moreover, let Fe : H (X, ΩX ) −→ V be a map such that τAlb(X) ◦ Fet = G ◦ τT .
By construction hence Fet = f ∗ ◦ τT .
Let me prove that Fe preserves the lattices: let γ ∈ H1 (X, Z), and let h ∈ V ∗ .
Hence  Z  Z Z Z
h Fe = Fet (h) = (f ∗ ◦ τT )(h) = τT (h).
γ γ γ f∗ γ

But now by Lemma 3.3.1 this last is h(τT−1 f∗ γ). As this is true for every
γ ∈ H1 (X, Z) and every h ∈ V ∗ , this simply means that τT−1 ◦ f∗ = Fe. Hence
f∗ = τT ◦ Fe, and the lattice are hence preserved.
Now, by Lemma 3.3.5 the map Fe induces a map g : Alb(X) −→ T , which is
unique up to translation. But g(0) = f (x0 ), so that g is unique.

Remark 3.3.1. If q(X) = 0, then Alb(X) is simply a point. The universal

property of the Albanese variety then implies that every map f : X −→ T ,
where T is a complex torus, is constant.
Remark 3.3.2. The Albanese variety Alb(X) is the smallest subtorus of Alb(X)
generated by the image of aX . This follows easily by the universal property of
the Albanese.
Remark 3.3.3. The Albanese variety is functorial: this follows immediately from
the universal property. In particular, if f : X −→ Y is a morphism, it induces
a morphism Alb(f ) : Alb(X) −→ Alb(Y ). If f is surjective, by the previous
remark one gets that Alb(f ) is surjective.
Let me now prove the last result we need on Albanese varieties: in general
the Albanese map is not surjective, but one can say something in the case where
the image is a curve.

Proposition 3.3.6. If the image of aX is a curve C in Alb(X), then aX has

connected fibers. Moreover, C is smooth and g(C) = q(X).

Proof. We have aX : X −→ C, and consider its Stein factorisation, i. e. a

manifold Y together with two morphisms f : X −→ Y with connected fibers,
and g : Y −→ C which is finite. The manifold X is normal, hence Y is normal.
Moreover, it has to be a curve, hence Y is a smooth curve. The morphism f
induces a morphism Alb(f ) : Alb(X) −→ Alb(Y ), and we know that Alb(Y ) =
Jac(Y ). Moreover, the map g induces a morphism Alb(g) : Jac(Y ) −→ Alb(X).
Moreover, as f is surjective, we have Alb(f ) surjective. By functoriality we have

Alb(g) ◦ Alb(f ) ◦ aX = g ◦ f = aX .

By the universal property of the Albanese, we then get Alb(g) ◦ Alb(f ) =

idAlb(X) . Now, Alb(f ) is surjective, hence it has to be an isomorphism with
inverse Alb(g). But now the map aY is injective (as Y is a smooth curve), so
106 Birational geometry

that g = Alb(g) ◦ aY is injective, hence g is an isomorphism. In conclusion

the fibers of aX are the fibers of f , hence they are connected. Moreover C is
smooth. Finally, simply notice that g(C) = g(Y ), and g(Y ) = dim(Jac(Y )).
As Alb(f ) is an isomorphism, then

g(C) = dim(Alb(X)) = q(X),

and we are done.

Chapter 4

The Enriques-Castelnuovo
Classification

This chapter is devoted to the proof of the Enriques-Castelnuovo birational

classification of smooth projective surfaces. As we have seen in the previous
chapter, if X is a smooth projective surface, one can perform a very natural
transformation on X, which is the blow-up of X along a point x. The product of
such a transformation is a smooth projective surface X e together with a proper
morphism p : X −→ X. This map is not an isomorphism: its restriction to
e
Xe \ p−1 (x) gives an isomorphism with X \ {x}, but is contracts a the smooth
rational curve p−1 (x) to the point x. Hence p is not an isomorphism. Moreover,
on Xe there is a curve more than on X, so that ρ(X) e = ρ(X) + 1, so that X e and
X cannot be isomorphic.
It is then unreasonable to expect a classification of smooth projective sur-
faces up to isomorphism: one can consider p1 , ..., pn different points on X, and
consider X ei := Blp ,...,p (X): we obtain n+1 surfaces which are not isomorphic,
1 i

but which are strictly related. They are, indeed, birational. Hence, it seems
reasonable to expect that it is possible to perform a birational classification of
smooth projective surfaces: this was the main idea Enriques and Castelnuovo
had, and it is the starting point of the classification. Indeed, from Theorem
3.1.18, we know that every birational map between two smooth projective sur-
faces is composition of isomorphisms and blow-ups. Moreover, from Theorem
3.1.20 we know that a surface is the blow-up of another one if and only if it
carries a (−1)−curve. A surface which does not contain (−1)−curves is mini-
mal, and we have seen that every smooth projective surface admits a minimal
model, which is again a smooth projective surface. The birational classification
of smooth projective surfaces is then the classification, up to isomorphism, of
minimal smooth projective surfaces.
A first attempt to classify minimal smooth projective surfaces is to use their

107
108 The Enriques-Castelnuovo Classification

Kodaira dimension: it is a birational invariant, hence two minimal surfaces with

different Kodaira dimension cannot be isomorphic. As seen in the previous
chapter, there is a first distinction: surfaces with Kodaira dimension −∞, and
surfaces of non-negative Kodaira dimension. For minimal surfaces, this can be
read in terms of the nefness of their canonical bundle: those surfaces whose
canonical bundle is not nef are in the first family, otherwise they are in the
second one.
Anyway, a classification by means of Kodaira dimension is too rough: there
are several examples of surfaces having completely different characteristics, but
having the same Kodaira dimension. One could then try to give a finer clas-
sification, for example using topological and analytic invariants like the Betti
numbers, the irregularity and the plurigenera. The Betti numbers are not bira-
tional invariants, hence they cannot e used for our purposes; anyway, irregularity
and plurigenera are important birational invariants, as we are going to show.
Our goal il to show the following:

Theorem 4.0.7. (Enriques-Castelnuovo’s birational classification). Let

X be a smooth projective surface. A minimal model of X is then one of the
following:

1. Kodaira dimension −∞:

• rational surface;

• ruled surface;

2. Kodaira dimension 0:

• abelian surface;

• K3 surface;

• Enriques surface;

• bielliptic surface;

3. Kodaira dimension 1: proper elliptic surface;

4. Kodaira dimension 2: surface of general type.

Clearly, there are still several things to define in the statement of the previous
Theorem, and this will be one of the major goals of the following sections. We
start with the classification of minimal surfaces of Kodaira dimension −∞, and
we will define rational and ruled surfaces. We will next prove the remaining
part of the classification, together with definitions.
The Kawamata Rationality Theorem 109

4.1 The Kawamata-Mori Rationality Theorem

The Kawamata-Mori Rationality Theorem is one of the most important results
in Mori Theory, and was proven at the beginning of the 80s. It came hence
much later then the Enriques-Castelnuovo Classification: it was however soon
realized that this Theorem, which holds in greater generality than the one we
are going to state, gives a serious shortcut in the proof of the classification.
As this result is basic in the context of the MMP, I decided to prove and use
it (following the modern fashion of view the original birational classification of
surfaces as the toy case for the MMP).
Before stating and proving the theorem, I need to recall some basic facts we
have seen in Chapter 2. If X is a smooth projective surface, on the real vector
space NR1 (X) we have a non-degenerate intersection form, and one may view a
divisor D either as an element of NR1 (X), or as the hyperplane

D⊥ := {α ∈ NR1 (X) | α · D = 0}.

Moreover, in NR1 (X) we have the important cone Ef f (X), whose dual cone is
N ef (X): a divisor D is nef if and only if D · α ≥ 0 for every α ∈ Ef f (X). If
one uses the notation

NR1 (X)D≥0 := {α ∈ NR1 (X) | α · D ≥ 0},

and similar for NR1 (X)D≥0 , NR1 (X)D>0 and NR1 (X)D<0 . Using this notation, we
easily see that D is nef if and only if Ef f (X) is contained in NR1 (X)D≥0 , and
that D is ample if and only if Ef f (X) \ {0} is contained in NR1 (X)D>0 .
If D is not nef, then there is a part of Ef f (X) lying in NR1 (X)D<0 . Moreover,
if s ∈ R and H is ample, we know that H + sD is ample for 0 < |s|  0, but
for s  0 we have that H + sD is not nef, as D is not nef. The hyperplane
(H +sD)⊥ moves in NR1 (X) as s changes: for small values of s the cone Ef f (X)
is contained in the positive hyperspace it defines, but for high values of s there
will be a part of Ef f (X) contained in the negative hyperspace. Hence, there is
s0 (H, D) ∈ R such that H + sD is nef for every s ≤ s0 (H, D), and H + sD is
not nef for every s > s0 (H, D), i. e.

s0 (H, D) := sup{s ∈ R | H + sD ∈ N ef (X)}.

Example 4.1.1. Let X = P1 ×P1 , and consider D = KX . Using the isomorphism

NR1 (X) ' R2 which sends h1 , h2 to the canonical basis of R2 , we have that
D = (−2, −2). As we have seen in Example 2.5.2, D is not nef, and

Ef f (X) = N ef (X) = {(x, y) ∈ R2 | x, y ≥ 0}.

Let us consider H = (1, 2), which is an ample divisor, so that H + sD =

(1 − 2s, 2 − 2s). The line (H + sD)⊥ is then

(H + sD)⊥ = {(x, y) ∈ R2 | (2 − 2s)y + (1 − 2s)x = 0}.

110 The Enriques-Castelnuovo Classification

Now, notice that if s = 0, then the line H ⊥ has equation y = −x/2, and for
0 ≤ s ≤ 1/2 the effective cone is always in the positive half-plane defined by
H + sD. For s > 1/2, this is no longer true, so that s0 (H, KX ) = 1/2.
In this example we have seen that s0 (H, KX ) ∈ Q: this is the general phe-
nomenon described by the Kawamata-Mori Rationality Theorem:

Theorem 4.1.1. (Kawamata-Mori’s Rationality Theorem). Let X be a

smooth projective surface, H an ample divisor on X. Suppose that KX is not
nef. Then s0 (H, KX ) ∈ Q.

Proof. For the proof we need the following result in number theory, which I
won’t prove:

Lemma 4.1.2. (Hardy-Wright). Let α ∈ R, α 6= 0, and let  ∈ Q>0 . More-

over, let
Wα, := {(x, y) ∈ R2 | αx < y < αx + }.

If α ∈
/ Q, then Wα, ∩ (N × N) is infinite.

Now, let u, v ∈ N, and let Du,v := uH + vKX , and let W := Ws0 (H,KX ),1
as in the previous lemma. If (u, v) ∈ W , then Du,v − KX is ample: we have
Du,v − KX = uH + (v − 1)KX , and s0 (H, KX ) < v/u < s0 (H, KX ) + 1/u; hence
(v − 1)/u = v/u − 1/u < s0 (H, KX ), and by definition of s0 (H, KX we have that
Du,v − KX is ample. By the Kodaira Vanishing Theorem we can then conclude
that hi (X, OX (Du,v )) = 0 for i = 1, 2 and for every (u, v) ∈ W . Let now

P (u, v) := χ(uH + vKX ) =

1 2 2 2 2 2
= (H u + KX v + 2H · KX uv + H · KX u + KX v) + χ(OX ),
2
which is a quadratic polynomial in u, v. Notice that the coefficient of u2 is
H 2 > 0, as H is ample, so that P (u, v) is not identically 0.
Let us suppose that s0 (H, KX ) ∈ / Q. By Lemma 4.1.2 there is in infinite
number of u, v ∈ N such that (u, v) ∈ W . There are two possibilities: either
P (u, v) = 0 for every (u, v) ∈ W , or there is (u, v) ∈ W such taht P (u, v) 6= 0.
Let us show that the first possibility can be excluded.
Indeed, since s0 (H, KX ) ∈ / Q, then there are arbitrary large p, q ∈ N such
that s0 (H, KX ) < p/q < s0 (H, KX ) + 1/3q. Hence, if k = 1, 2, 3 then the
couple (kq, kp) ∈ W for arbitrary large integers p, q. If P (u, v) = 0 for every
(u, v) ∈ W , then P (kq, kp) = 0 for k = 1, 2, 3, so that P (u, v) restricted to the
line R>0 (p, q) vanishes at three different points. As it is a quadratic polynomial,
it has to vanish on the whole line. As one can consider infinite many lines, then
P (u, v) would be identically zero, which is not possible.
In conclusion, there is (u, v) ∈ W such that P (u, v) 6= 0. Now, recall that for
this choice of (u, v) we have h0 (XOX (Du,v )) = P (u, v), so that there must be
Kodaira dimension −∞ 111

an effective divisor L ∼ uH + vKX : we can then suppose Du,v to be effective.

But as (u, v) ∈ W , we have v/u > s0 (H, KX ), so that Du,v is not nef. There
is then an irreducible curve C such that Du,v · C < 0. As Du,v is effective, this
forces C to be one of the components of Du,v . Let these be D1 , ..., Dr : without
loss of generality, we can suppose C = D1 . Resuming, we have
 
v
H + KX · D1 < 0, (H + s0 (H, KX )KX ) · D1 ≥ 0.
u
2
We can even say more: as H · KX , KX ∈ Z and s0 (H, KX ) ∈/ Q, then (H +
s0 (H, KX )KX ) · D1 > 0. Then, there must be some ρ1 ∈ Q>0 such that (H +
ρ1 KX ) · D1 ≥ 0 and s0 (H, KX ) < ρ1 < v/u. We can repeat the same argument
for every irreducible component of Du,v , getting for each i = 1, ..., r a positive
rational number ρi such that s0 (H, KX ) < ρi < v/u, and (H + ρi KX ) · Di > 0.
Let now
ρ := min{ρ1 , ..., ρr }.

Then s0 (H, KX ) < ρ < v/u and (H + ρKX ) · Di ≥ 0 for every i = 1, ..., r.
Now, the divisor H + ρKX is nef. Let C be an irreducible curve different
from D1 , ..., Dr ; we need to check that (H + ρKX ) · C ≥ 0. We have
      
1 u 1 u
(H + ρKX ) · C = ρ H + KX · C = ρ H + KX · C + − H ·C =
ρ v ρ v
   
1 1 u
= ρ Du,v · C + − H · C > 0,
v ρ v
as ρ > 0, Du,v is effective and C is not one of its components, ρ < v/u and H
is ample.
Hence H + ρKX is nef, so that ρ ≤ s0 (H, KX ), which is a contradiction. In
conclusion, we need s0 (H, KX ) ∈ Q.

This result is appearently innocent, but it is really far from that, as we are
going to see in the next section.

4.2 Kodaira dimension −∞

Using the Kawamata-Mori Rationality Theorem we can proceed to the classifi-
cation of smooth projective surfaces having Kodaira dimension −∞. We first
introduce some examples, in particular rational and geometrically ruled surfaces.
We will then proceed to a classification.

4.2.1 Ruled surfaces

Here we introduce three main definitions: rational surfaces, ruled surfaces and
geometrically ruled surfaces. In this section we show some of their important
112 The Enriques-Castelnuovo Classification

features, in particular which of them are minimal, and their numerical birational
invariants.

Definition 4.2.1. A smooth projective surface X is ruled if it is birational to a

product C ×P1 , where C is a smooth curve. If C is P1 , then X is called rational.

Notice that every rational surface is then birational to P2 , as P1 × P1 is

birational to P2 . Conversely, every smooth projective surface which is birational
to P2 is rational.

Definition 4.2.2. A smooth projective surface X is geometrically ruled if there

is a surjective morphism p : X −→ C of maximal rank, where C is a smooth
curve, and the fibers of p are all isomorphic to P1 .

A first important remark is that it is by no means clear which is the realtion

between ruled and geometrically ruled surfaces. Here is the first property:

Proposition 4.2.1. Let X be a surface, C a curve and p : X −→ C a surjective

morphism. If c ∈ C is a regular value for p such that p−1 (c) ' P1 , then there is
a Zariski-open neighborhood U of c in C such that p−1 (U ) ' U × P1 in a fiber
preserving manner. In particular, every geometrically ruled surface is ruled.

Proof. The proof is rather long, so it will be divided in several steps.

Step 1. We show that H 2 (X, OX ) = 0. This is easy: indeed, by Serre’s
Duality we have H 2 (X, OX ) = 0 if and only if H 0 (X, OX (KX )) = 0. Let us
suppose that this last is not verified, so that there is an effective divisor KX .
Let F be the fiber over c, which is a smooth rational curve. Since F 2 = 0, by
the genus formula we get
1
0 = 1 + KX · F,
2
so that KX · F = −2. However, KX is effective, so that KX = nF + G, where
n ∈ N and G is effective and not containing F . Hence

−2 = KX · F = (nF + G) · F ) = G · F ≥ 0,

which is not possible. In conclusion pg (X) = 0.

Step 2. As pg (X) = 0, the first Chern class morphism is surjective. We want
to produce a divisor H such that H · F = 1. To do that, it is then sufficient
to find a class h ∈ H 2 (X, Z) such that h · f = 1, where f = c1 (OX (F )). Let
α ∈ H 2 (X, Z)/tors, and let us consider α · f ∈ Z: the set

If := {α · f ∈ Z | α ∈ H 2 (X, Z)/tors}

is an ideal of Z, hence there must be d ∈ Z such that If = (d). Let us consider

the linear functional
α·f
δ : H 2 (X, Z)/tors −→ Z, δ(α) := .
d
Kodaira dimension −∞ 113

This is then an element in (H 2 (X, Z)/tors)∗ , so that by Poincaré duality there

must be f 0 ∈ H 2 (X, Z)/tors such that δ(α) = α·f 0 for every α ∈ H 2 (X, Z)/tors.
Hence we have f · α = df 0 · α for every α ∈ H 2 (X, Z)/tors, so that f = df 0 .
If we show that f = f 0 , i. e. d = 1, we are done: we would have If = Z, so
there is a divisor H such that H · F = 1. To do so, let k = c1 (OX (KX )). As
f · k = −2 by Step 1, we need f 0 · k = −2/d, so that d = 1, 2. Now, let α be the
class of an irreducible smooth curve: then we can use the genus formula to get
that α2 +α·k is even. By linearity, this is then true on the whole H 2 (X, Z)/tors.
Moreover, notice that f 0 · k + (f 0 )2 = f 0 · k, which has then to be even, so that
−2/d is even. In conclusion d = 1.
Step 3. Consider now the divisor H such that H · F = 1, and consider the
exact sequence

0 −→ OX (H + (r − 1)F ) −→ OX (H + rF ) −→ OF (H + rF ) −→ 0.

Notice that

deg(OF (H + rF )) = (H + rF ) · F = H · F = 1,

so that OF (H + rF ) = O(1). As H 2 (F, O(1)) = 0, the map

br : H 1 (X, OX (H + (r − 1)F )) −→ H 1 (X, OX (H + rF ))

is surjective for every r ∈ N. The sequence decreases for r increasing, so that

there is r0 ∈ N such that for every r > r0 the map br is an isomorphism. Hence,
the map
ar : H 0 (X, OX (H + rF )) −→ H 0 (F, O(1))
is surjective for r  0.
Consider a plane V of H 0 (X, OX (H + rF )) which is mapped isomorphically
by ar onto H 0 (F, O(1)), and consider the corresponding pencil P = P(V ). As
O(1) is very ample and V ' H 0 (F, O(1)) via ar , we have that P separates
points of F . The possible fixed locus Z of P is then given by points lying in
fibers distinct from F , or of curves in fibers disjoint from F . Let us consider
U := C \ p(Z), which is a Zariski-open subset of C containing c. Moreover, let
P 0 := P|p−1 (U ) . The generic member Ct of the moving part of P meets F in
only one point (as Ct ∼ H + rF , so that Ct · F = H · F = 1), so that if it
is reducible is must contain some fibers. If it was the case, another Cs in the
moving part of P , with s 6= t, would meet Ct in the intersection points of these
fibers, and hence they would be base points. Hence P 0 , which is the moving
part of P , consists exactly of sections of p|p−1 (U ) . It then defines a map

g : p−1 (U ) −→ P1 ,

sending a point x ∈ p−1 (U ) to the intersection point of Ct with F , where Ct is

the only element of P 0 passing through x.
114 The Enriques-Castelnuovo Classification

Notice that the fibers of g meet the fibers of p at exactly one point: the
morphism
(p, g) : p−1 (U ) −→ U × P1

is then an isomorphism preserving the fibers.

Remark 4.2.1. In the proof we have shown that if X is a geometrically ruled

surface, then pg (X) = 0. Moreover, we have shown that there is always a divisor
H such that H · F = 1, where F is the class of a fiber.
We have a lot more informations about the structure of geometrically ruled
surfaces: we know indeed how to produce them all.

Theorem 4.2.2. Let p : X −→ C be a geometrically ruled surface. Then there

is a rank 2 vector bundle E on C and an isomorphism f : X −→ P(E) such
that π ◦ f = p, where π : P(E) −→ C is the canonical projection. Moreover, two
geometrically ruled surfaces P(E) and P(E 0 ) over C are isomorphic if and only
if there is a line bundle L on C such that E ' E 0 ⊗ L .

Proof. Consider the sheaf GL(2, OC ) defined in the following way: for every
open subset U ⊆ C, let

GL(2, OC )(U ) := GL(2, OC (U )),

with the obvious restriction morphisms. The quotient of GL(2, OC ) under the
natural action of C∗ defines another sheaf, denoted P GL(2, OC ). It is easy to
show that the isomorphism classes of rank 2 vector bundles on C are classified
by H 1 (C, GL(2, OC )), and that the isomorphism classes of P1 −bundles on C are
classified by H 1 (C, P GL(2, OC )). We have an evident exact sequence of sheaves

1 −→ OC∗ −→ GL(2, OC ) −→ P GL(2, OC ) −→ 1,

which in cohomology gives the exact sequence

a p
P ic(C) −→ H 1 (C, GL(2, OC )) −→ H 1 (C, P GL(2, OC )) −→ H 2 (C, OC∗ ).

First of all, notice that p(e) = p(e0 ) if and only if there is L ∈ P ic(X) such that
e = e0 · a(L ), and the second part is shown. For the first part, one just needs
to prove that H 2 (C, OC ) = 0, which follows immediately from the exponential
sequence of C, as C is a curve.

4.2.2 Invariants of geometrically ruled surfaces

Let me calculate some important invariants of geometrically ruled surfaces. By
Theorem 4.2.2, every geometrically ruled surface X on a smooth curve C is the
projective bundle associated to a rank 2 vector bundle E on C. By Remark
4.2.1 there is a divisor H such that H · F = 1, and we denote the corresponding
Kodaira dimension −∞ 115

line bundle as OX (1). Let h := c1 (OX (1)). I recall how one can define this line
bundle: we have p : X = P(E) −→ X, and every point x ∈ X corresponds hence
to a line Dp(x) in the 2−dimensional vector space Ep(x) . Consider the sub-line
bundle N ⊆ p∗ E such that for every x ∈ X we have Nx = Dp(x) . Then OX (1)
is the quotient of p∗ E by N . First of all, let me describe the Picard group and
the intersection form on a geometrically ruled surface.

Proposition 4.2.3. Let p : X = P(E) −→ C be a geometrically ruled surface.

1. The Picard group of X is P ic(X) ' p∗ P ic(C) ⊕ Z · OX (1). In particular,

H 2 (X, Z) is generated by the class f of a fiber and by h.

2. Let d := deg(E). Then h2 = d.

2
3. We have c1 (KX ) = −2h+(2g(C)−2+d)f . In particular KX = 8(1−g(C)).

Proof. We clearly have a map

p∗ P ic(C) ⊕ Z · OX (1) −→ P ic(X), (p∗ L , n) 7→ p∗ L ⊗ OX (n),

where OX (n) := OX (1)⊗n . This map is injective: if p∗ L ⊗ OX (n) ' p∗ M ⊗

OX (m) we have p∗ L ' p∗ M ⊗ OX (m − n). As OX (1) is not contracted to a
point by p, we then need n = m. Hence p∗ L ' p∗ M .
We then need to show the surjectivity: let D be an effective divisor on X,
and write D = nH + G, where n ∈ N and G is effective and does not contain
H. Then we need to show that G is the pull-back of a line bundle on C. If
G · F 6= 0, then by construction we need G = kH for some integer k. Hence
G · F = 0, so that G has components lying on fibers, and G is the pull-back
of some divisor on C. For non effective divisors D, it is more complicated: the
idea is to show that for n  0 we have Dn := D + nF effective. Then item 1
follows from what we have just shown. Now, consider Dn . As D · F = 0, we
have D2 = Dn2 , and

Dn · KX = D · KX + nF · KX = D · KX + ndeg(KF ) = D · KX − 2n.

By Riemann-Roch we then get

1
h0 (X, OX (Dn )) + h0 (X, OX (KX − Dn )) ≥ χ(OX ) + n + (D2 − D · KX ).
2
But if n  0 we have h0 (X, OX (KX − Dn )) = 0: if A is a very ample divisor on
X, we have that (KX − Dn ) · A = KX · A − D · A − nF · A < 0 for n  0, so that
KX − Dn is not effective for n  0. In conclusion, we get h0 (X, OX (Dn )) ∼ n
for n  0, and we are done.
For the H 2 (X, Z), simply use Remark 4.2.1: we have H 2 (X, OX ) = 0, so
that c1 is surjective. Hence H 2 (X, Z) is generated by f and h.
116 The Enriques-Castelnuovo Classification

For the second item, consider the exact sequence defining OX (1):

0 −→ N −→ p∗ E −→ OX (1) −→ 0.

Hence N ⊗ OX (1) ' det(E), so that h = p∗ e − c1 (N ), where e = c1 (det(E)).

Hence h2 = h · p∗ e − h · c1 (N ). Now, for every exact sequence

0 −→ L −→ F −→ M −→ 0,

where L, M are line bundles on a surface, and F is a rank 2 vector bundle, we

have
L · M = L−1 · M −1 = χ(OX ) − χ(L) − χ(M ) + χ(L ⊗ M ) =
= χ(OX ) − χ(F ) + χ(det(F )),
which depends only on F . In our case we have then

c1 (N ) · h = χ(OX ) − χ(p∗ (E)) − χ(det(E)).

As E is a rank 2 vector bundle on a curve C, there are two line bundles L, M

on C and an exact sequence

0 −→ L −→ E −→ M −→ 0,

(see Lemma 5.1.1 below), so that

c1 (N ) · h = p∗ L · p∗ M = 0,

hence h2 = h · p∗ e = deg(det(E)) = deg(E), and we are done with item 2.

For the third item, we have c1 (KX ) = ah + bf . Moreover

−2 = c1 (KX ) · f = ah + bf = a,

so that c1 (KX ) = −2h + bf . Then KX · H = −2d + b. The genus formula gives

2g(X) − 2 = KX · H + H 2 = −2d + b + d = b − d,

so that b = 2g(X) + d − 2. Finally

2
KX = (−2h + (2g(X) + d − 2)f )2 = 4d − 8g(X) − 4d + 8 = 8(1 − g(X)).

Remark 4.2.2. Let X be a geometrically ruled surface. Then KX is not nef:

indeed we have that f is an effective class, and that KX · f = −2.
To conclude, we show the following:

Proposition 4.2.4. Let X be a ruled surface birational to C × P1 . Then

pg (X) = Pm (X) = 0 for every m > 0, and q(X) = g(C). In particular
κ(X) = −∞.
Kodaira dimension −∞ 117

Proof. As pg (X), Pm (X) and q(X) are birational invariants, we can use C × P1
to calculate them. Here notice that

ΩC×P1 ' p∗ ΩC ⊕ q ∗ ΩP1 ,

where q : C ×P1 −→ P1 is the projection. Hence KC×P1 ' p∗ KC +q ∗ KP1 . Then

q(X) = h0 (C × P1 , ΩC×P1 ) = h0 (C, OC (KC )) + h0 (P1 , O(−2)) = g(C),

pg (X) = h0 (C × P1 , OC×P1 (KC×P1 )) = h0 (C, OC (KC )) · h0 (P1 , O(−2)) = 0.

Similar calculations show that Pm (X) = 0 for every m > 0.

4.2.3 Rational surfaces

A rational surface X is a ruled surface which is birational to P1 × P1 . Hence, a
rational surface is birational to P2 and one can calculate its numerical invariants
from this (without using Proposition 4.2.4). Namely, if X is rational, then
q(X) = pg (X) = 0 and Pm (X) = 0 for every m > 0.
If a rational surface is geometrically ruled, then by Theorem 4.2.2 it is iso-
morphic to a projective bundle associated to a rank 2 vector bundle on P1 . One
can be more explicit in the classification of geometrically ruled rational surfaces.
Here is the main result of the section:

Theorem 4.2.5. Let E be a vector bundle of rank n on P1 . Then there are

a1 , ..., an ∈ Z such that E ' O(a1 ) ⊕ ... ⊕ O(an ).

Proof. Use the following lemma:

Lemma 4.2.6. Let U, V, W be three vector bundles on a complex manifold X

and consider the exact sequence
a b
0 −→ U −→ V −→ W −→ 0.

It splits if and only if H 1 (X, H om(W, U )) = 0.

Proof. Apply the functor H om(W, .) to the exact sequence: As V, W, U are

locally free, one gets an exact sequence
a◦ b◦
0 −→ H om(W, U ) −→ H om(W, V ) −→ H om(W, W ) −→ 0. (4.1)

The exact sequence of the lemma splits if and only if there is a morphism
h : W −→ V such that b ◦ h = idW . Looking at the long exact sequence induced
by the exact sequence (4.1), we have
b◦
Hom(W, V ) −→ Hom(W, W ) −→ H 1 (X, H om(W, U )).

The splitting of the exact sequence of the lemma is then equivalent to the
vanishing of H 1 (X, H om(W, U )).
118 The Enriques-Castelnuovo Classification

Now, let E be a vector bundle on P1 , and for every p ∈ Z let E(p) :=

E ⊗ O(p). Now, fix k ∈ Z such that H 0 (P1 , E(k)) 6= 0, but for every h < k we
have H 0 (P1 , E(h)) = 0. Let s ∈ H 0 (P1 , E(k)): we have Z(s) = ∅. Indeed, if
Z(s) 6= ∅, it would define an effective line bundle O(p), p > 0, so that E(k − p)
has a section. As we supposed that for every h < 0 the vector bundle E(h) has
no section, this is not possible, so that Z(s) = ∅. Then s defines the trivial line
bundle OP1 , and an injection

0 −→ OP1 −→ E(k) −→ F −→ 0.

Twisting by O(−1) we then get

0 −→ O(−1) −→ E(k − 1) −→ F (−1) −→ 0.

Recall that H 0 (P1 , E(k − 1)) = 0. Moreover, we have H 1 (P1 , O(−1)) = 0, so

that H 0 (P1 , F (−1)) = 0, and H 0 (P1 , F (−2)) = 0. Using Serre’s Duality we
then get H 1 (P1 , F ∗ ) = 0. But this means H 1 (P1 , H om(F, OP1 )) = 0. Apply
Lemma 4.2.6 to get hence that E ' O(−k) ⊕ F .
But now F has rank n − 1, so that one can apply induction to see that there
are a2 , ..., an ∈ Z such that F ' O(a2 ) ⊕ ... ⊕ O(an ). Now simply let a1 := −k,
and we are done.

Theorem 4.2.2 together with Theorem 4.2.5 implies that every geometrically
ruled rational surface X is of the form X ' P(O(a1 ) ⊕ O(a2 )). Notice that
tensoring with O(−a1 ) does not change X, so that one can always suppose that
X ' P(OP1 ⊕ OP1 (n)) for some n ≥ 0. Now, let me introduce the following
notation: for every n ≥ 0, let

Fn := P(OP1 ⊕ OP1 (n)).

Definition 4.2.3. The surface Fn is called Hirzebruch surface.

Resuming, we have shown that every geometrically ruled rational surface is

a Hirzebruch surface.

4.2.4 Minimal ruled surfaces

In this section we show some important results about minimal ruled surfaces.
Up to now we have seen that every geometrically ruled surface is ruled, and
that every ruled surface is birational to a geometrically ruled one. It is by no
means clear that a ruled surface is geometrically ruled: this is, indeed, not true
in general.

Proposition 4.2.7. (Noether-Enriques). Let X be a minimal surface, and

let us suppose that there is a surjective morphism p : X −→ C to a smooth
curve such that the generic fiber is a smooth rational curve. Then p gives X the
structure of a geometrically ruled surface.
Kodaira dimension −∞ 119

Proof. Let c ∈ C be a generic point and consider F := p−1 (c). Then F ' P1 ,
and F 2 = 0. Then
F · KX = 2g(F ) − 2 = −2.
Following the same argument as in Step 2 of Proposition 4.2.1, we have that
F is primitive, i. e. if divisor F 0 is such that F = nF 0 for some n ∈ Z, then
n = ±1. Now, let us consider G to be any fiber (no more only generic). We
have G2 = 0 and G · KX = −2, so that g(G) = 0 for every fiber.
We now show that there are no reducible fibers. Let us suppose that F =
P
i ni Ci is a reducible fiber. We have
X X
ni Ci2 = Ci · (F − nj Cj ) = − nj Ci · Cj < 0
j6=i j6=i

as Ci · F = 0 and there is at least one j 6= i such that Ci · Cj 6= 0 (as the fibers

are connected: this is true for the generic fiber, hence is everywhere true by
easy topological argument). Hence nj > 0 and Ci · Cj ≥ 0, for every j 6= i, and
there is j 6= i such that Ci · Cj > 0. In conclusion Ci2 < 0. By the genus formula
we have
−2 = 2g(Ci ) − 2 = Ci−2 + Ci · KX ,
so that Ci · KX ≥ −1, and Ci · KX = −1 if and only if Ci2 = −1. In this last
case Ci would be a (−1)−curve, but as X is minimal, by Corollary 3.1.21 the
surface X does not contain any (−1)−curve. Hence Ci · KX ≥ 0. As this is true
for every i, we then get
X X
−2 = KX · F = KX · ni Ci = ni Ci · KX ≥ 0,
i i

which is impossible. In conclusion, there are no reducible fibers.

Now, every point c ∈ C is a regular value for p. By Proposition 4.2.1 there
is a Zariski-open neighborhood U of c in C such that p−1 (U ) ' U × P1 in a
fiber preserving manner. But this implies that p : X −→ C is a geometrically
ruled surface.

The main importance of this is the following:

Theorem 4.2.8. Let X be a ruled surface which is not rational. Then X is

geometrically ruled if and only if it is minimal.

Proof. As X is ruled, there is a smooth curve C such that X is birational to

C × P1 . We have then a birational map X 99K C × P1 , and composing with
the projection to C we get a rational map p : X 99K C. If this map is not
a morphism, we need to blow-up X along the indeterminacies of p in order to
e −→ C. On X
get a morphism pe : X e there will then be (−1)−curves: all these
are then mapped to a point of C (otherwise the image would be a rational
curve, but C is not rational as X is not rational). But this is not possible if
120 The Enriques-Castelnuovo Classification

we are blowing-up indeterminacies, and p is then a morphism. Now, we have a

surjective map p : X −→ C whose generic fiber is a smooth rational curve. As
X is minimal, we can apply Proposition 4.2.7 to get that X has the structure
of a geometrically ruled surface.
To conclude we just need to show that every geometrically ruled non-rational
surface is minimal. We have p : X −→ C, where C is not rational. If X is not
minimal, then there is a (−1)−curve E on X. As E 2 = −1, E is not contained
in a fiber, hence p(E) = C. But this implies C is rational, which is not the case,
and we are done.

In conclusion, the minimal models of ruled surfaces which are not rational
are geometrically ruled surfaces. Moreover, one sees from this that minimal
models of ruled non-rational surfaces are of the form P(E), where E is a rank 2
vector bundle on a smooth curve C. This implies two things: first of all, there
is no uniqueness of the minimal model; second, one can in principle classify all
minimal models of non-rational ruled surfaces simply knowing a classification
of rank 2 vector bundles on curves.
Let us then look at rational surfaces. A first remark is the following:
Remark 4.2.3. There is an example of minimal rational surface which is not
geometrically ruled: it is simply P2 . Indeed, if P2 was geometrically ruled, then
there would be some n ≥ 0 such that P2 ' Fn . But recall that P ic(P2 ) ' Z,
while P ic(Fn ) ' P ic(P1 ) ⊕ Z ' Z2 by Proposition 4.2.3. Hence P2 cannot be
isomorphic to any Fn , so it is not geometrically ruled.
Anyway, one can give a complete list of minimal rational surfaces. First of
all one has the following:

Proposition 4.2.9. The Hirzebruch surface F1 is isomorphic to the blow-up of

P2 along a point. In particular, F1 is not minimal.

Proof. Let x0 , x1 , x2 be homogenous coordinates on P2 , and let y0 , y1 be homoge-

nous coordinates on P1 . Consider the point P = (1 : 0 : 0) ∈ P2 . The blow-up
of P2 along P is then a smooth projective surface V ⊆ P2 × P1 of equation
x0 y1 = x1 y0 . Now, F1 = P(OP1 ⊕ OP1 (1)). The vector bundle E := OP1 ⊕ OP1 (1)
is generated by its global sections 1 of OP1 and y0 , y1 of O(1).
The algebraic variety F1 is then P roj(S(E)), where S(E) is the symmetric
algebra of E. we have a map of graded rings

C[x0 , x1 , x2 ] ⊗ C[y0 , y1 ] −→ S(E),

which sends x0 to y0 , x1 to y1 , x2 to 1 and yo and y1 to themself. This map

is clearly surjective, and it kernel is the ideal generated by x0 y1 − x1 y0 . Hence
F1 is the subvariety of P2 × P1 of equation x0 y1 = y1 x0 . In conclusion we have
F1 ' BlP P2 , and we are done.
Kodaira dimension −∞ 121

All the other Hirzebruch surfaces are minimal:

Theorem 4.2.10. Let X be a minimal rational surface. If it is geometrically

ruled, then X ' Fn for some n 6= 1.

Proof. First of all, P2 is ruled. If n = 0, then F0 ' P1 × P1 , as easily seen, so

this one is minimal too. Let now consider n > 1, and let pn : Fn −→ P1 be the
canonical projection. On Fn there are two natural type of sections of pn one
can describe: one is

Sb : P1 −→ Fn , S(x)
b := (0, 1) ∈ OP21 ,x .

The other one is obtained considering any homogenous polynomial of degree n

in two variables, and considering

Sp : P1 −→ Fn , Sp (x) := (1, p(x)) ∈ OP21 ,x .

Notice that Sb and Sp define curves in Fn , and that Sp ∼ Sq for every two
homogenous polynomials p, q of degree n, so that one can define S := Sp for
some homogenous polynomial p of degree n. It is easy to see that S · Sb = 0, and
one has S · S = n (as Sp ∩ Sq coincides with the set of common zeroes of p and
q: as p and q can be chosen to have no common factors, and they have degree
n, we have that this set is given by n points).
By Proposition 4.2.3, there must be some a ∈ N0 such that Sb = S + aF .
Hence
0 = Sb · S = S 2 + aS · F = a + n,
hence Sb = S − nF . Moreover, we have

Sb · Sb = (S − nF ) · (S − N F ) = S 2 − 2nS · F + n2 F 2 = −n.

As n > 1, we see that Sb is not a (−1)−curve. Now, let us consider an irreducible

effective curve C = aS +bF , and let us suppose that C is not linearily equivalent
to S.
b Hence

C · Sb = (aS + bF ) · (S − nF ) = aS 2 + (b − na)S · F − nbF 2 = b ≥ 0.

Moreover, we have C · F = (aS + bF ) · F = a ≥ 0. Finally, let us consider

C 2 = (aS + bF )2 = a2 n + 2ab ≥ 0,

so that the only irreducible curve of negative self-intersection is S.

b In conclusion,
Fn is minimal for every n 6= 0.

To conclude, let us first notice that P2 is not isomorphic to any Fn by Remark

4.2.3, and that if n 6= m, then Fn is not isomorphic to Fm : on Fn there is a curve
of self-intersection −n, and on Fm there is no such a curve. This implies that
even for rational surfaces there is no unique minimal model. Anyway, notice
that we have not proved yet that P2 and the Hirzebruch surfaces are the only
minimal rational surfaces. This will be proven in the next section.
122 The Enriques-Castelnuovo Classification

4.2.5 Classification for Kodaira dimension −∞

As far as now, we introduced several examples of surfaces having Kodaira dimen-
sion −∞. We have the following, which completes the classification of surfaces
of Kodaira dimension −∞.

Theorem 4.2.11. Let X be a minimal surface such that κ(X) = −∞. Then
X is either geometrically ruled or X ' P2 .

Proof. The proof uses the Kawamata-Mori Rationality Theorem. First of all,
by Corollary 3.2.10 as X is minimal and κ(X) = −∞ we have that KX is not
nef. Consider the positive half ray R := Q>0 · (−KX ) in NQ1 (X). We have only
two possibilites: either P ic(X) ⊆ R, i. e. −KX is ample and ρ(X) = 1; or there
is H ∈ P ic(X) which is not a positive multiple of −KX .
Let us begin with the first case, so we suppose that −KX is ample and that
ρ(X) = 1. As −KX is ample, by the Kodaira Vanishing Theorem we have that
hi (X, OX ) = 0 for every i > 0, i. e. pg (X) = q(X) = 0. Hence, by Example
2.1.3 the first Chern class is an isomorphism
c
1
P ic(X) −→ H 2 (X, Z).

Hence b2 (X) = 1. As q(X) = 0 and X is Kähler we get b1 (X) = 0. In conclusion

we get e(X) = 3. Apply Noether’s Formula:
1
1 = 1 − q(X) + pg (X) = χ(OX ) = (K 2 + e(X)),
12 X
2
so that KX = 9.
Now, as P ic(X) as rank 1, one can consider an ample generator H. As −KX
is ample, H − KX is ample too, so that by the Kodaira Vanishing Theorem one
gets hi (X, OX (H)) = 0 for i = 1, 2. In conclusion we have
1
h0 (X, OX (H)) = χ(H) = 1 − q(X) + pg (X) + (H 2 − H · KX ).
2
Moreover, as H − KX is ample, we need H − KX = nH for some n ∈ N, hence
2
KX = (n + 1)H, where n + 1 > 1. As KX = 9 we get
2
9 = KX = (n + 1)2 H 2 ,

so that H 2 properly divises 9. In conclusion H 2 = 1 and KX = −3H. Using

this, we finally get h0 (X, OX (H)) = 3. As H is ample, it defines a dominant
rational map
f : X 99K P2 ,
mapping H to a line. Now, we have H 2 = 1, hence |H| has no base points, so
that f : X −→ P2 is everywhere defined. Now, f cannot contract curves, as
ρ(X) = ρ(P2 ) = 1, hence f is finite of degree d. Recall now that

1 = H 2 = (f ∗ O(1))2 = dO12 = d,
Kodaira dimension −∞ 123

so that d = 1, and the map f is birational. If f −1 is not defined at some points,

then f contracts a curve to that point, but as f does not contract curves, f −1
is everywhere defined. In conclusion X ' P2 .
Let us consider the remaining case, namely there is an ample divisor H such
that H ∈/ R. Let s := s0 (H, KX ), which is a rational number by the Kawamata-
Mori Rationality Theorem, and let L := uH + vKX for u, v ∈ N such that
v/u = s. Then L is a nef divisor, so that L2 ≥ 0. Moreover, for every m ≥ 1
1
the Q−divisor L − m KX is ample: indeed
 
1 1
L − KX = uH + v − KX ,
m m
and (v − (1/m))/u < v/u = s. Multiplying by m we get the ample divisor
mL − KX , for every m ≥ 1. Using the Kodaira Vanishing Theorem we then
have hi (X, OX (mL)) = 0 for i = 1, 2. By Riemann-Roch we finally get
1
h0 (X, OX (mL)) = χ(OX ) + (m2 L2 − mL · KX ).
2
Recall that L2 ≥ 0, so that we have only two possibilities: either L2 > 0, or
L2 = 0.
Let us begin with the case L2 > 0. As L is nef, for every effective divisor D
we have L · D ≥ 0. If D is such that L · D = 0, then

0 = L · D = (uH + vKX ) · D = uH · D + vKX · D.

As u, v ∈ N and H · D > 0 as H is ample, we need KX · D < 0. Moreover, as

L2 > 0 and L · D = 0, by the Hodge Index Theorem we get D2 < 0. Now, use
the genus formula:
1
g(D) = 1 + (KX · D + D2 ) ≤ 0,
2
so that g(D) = 0. Moreover, by the same formula we get

KX · D + D2 = −2.

If D2 < −1, then we would have KX · D ≥ 0, which is not possible, so that

D2 = −1. In conclusion D is an exceptional curve. As X is minimal, such
a D cannot exist on X, so that for every irreducible effective divisor we have
L · D > 0. By the Nakai-Moishezon Criterion, we can finally conclude that L
is ample: but L is on the border of N ef (X), hence it cannot be ample, so we
must have L2 = 0.
So, we consider L2 = 0. As L is nef we need L · H ≥ 0. If L · H = 0, the
Hodge Index Theorem implies L2 ≡ 0. In conclusion, we have uH ≡ −vKX , i.
e. H ∈ R, which is not possible. In conclusion, we have L · H > 0. Now
1 2
0= L = L · (H + sKX ) = L · H + sL · KX .
u
124 The Enriques-Castelnuovo Classification

As s > 0 and L · H > 0, we get L · KX < 0. As we have seen before we have

1
h0 (X, OX (mL)) ≥ − L · KX m + χ(OX ),
2
so that h0 (X, OX (mL)) ∼ m. Up to replacing L with mL we can then suppose
dim|L| > 1.
Let us write L = L0 + M , where M is the fixed part, and L0 the moving
part. As L0 moves in a linear system, it is clear that L0 is nef. Hence (L0 )2 ≥ 0.
Moreover M is effective, so L0 · M ≥ 0 as L0 is nef. Then

0 = L2 = L · L0 + L · M ≥ 0,

so that we need L · L0 = L · M = 0. Hence

0 = L · L0 = (L0 )2 + L0 · M ≥ 0,

so that (L0 )2 = L0 ·M = 0. Now, let D be an irreducible component of a member

of |L0 |. As L · L0 = 0, we have L · D = 0. Similarily, as M · L0 = 0 we have
M · D = 0. Hence

L0 · D = (L − M ) · D = L · D − M · D = 0,

which implies D2 ≤ 0. But now, remark that

0 = L · D = uH · D + vKX · D,

so that D · KX < 0, as H is ample and u, v > 0. As shown previously, this

implies that g(D) = 0 and that D2 = 0.
What we have just done applies to every linear system of L containing no
fixed part. Consider then a 1−dimensional V ⊆ |L|, and take out its fixed part,
so that we get a pencil P of curves F such that F 2 = 0. Every irreducible
component of a member of |F | is a smooth rational curve. As F 2 = 0, there
can be no fixed points, hence |F | gives a morphism

f : X −→ P1 ,

as dim|F | = 1. Consider now the Stein factorization of f : there is a smooth

curve C and a map f 0 : S −→ C whose fibers are connected, and are the
components of the members of F . Hence this is a P1 −fibration, and X has the
structure of a geometrically ruled surface.

Notice that this theorem implies that the only smooth projective surfaces of
Kodaira dimension −∞ are the ruled surfaces. If X it is not rational, then we
had already seen that its minimal model is geometrically ruled. If X is rational
we can now conclude:
Kodaira dimension −∞ 125

Corollary 4.2.12. Let X be a minimal rational surface. Then X is either P2

or Fn for some n 6= 1.

Proof. As X is a minimal rational surface, it is a minimal surface such that

κ(X) = −∞. Hence by Theorem 4.2.11 X is either geometrically ruled or
P2 . But if X is geometrically ruled, by Theorem 4.2.10 we have that X is a
Hirzebruch surface Fn for n 6= 1 and we are done.

4.2.6 The Castelnuovo Rationality Criterion

In this subsection I want to prove an important corollary of the Kawamata-
Mori Rationality Theorem that will be useful in the following, in the proof of
the Castelnuovo-Enriques Classification.

Proposition 4.2.13. Let X be a smooth projective surface, and let us suppose

that KX is nef. Then one of the following is possible:
2
1. KX > 0. In this case we have even P2 (X) > 0 and

1 2
Pm (X) ≥ m(m − 1)KX + 1 − q(X) + pq (X)
2
for every m ≥ 2.
2
2. KX = 0 and q(X) = 0. In this case P2 (X) > 0.
2
3. KX = 0, pg (X) > 0 and q(X) > 0.
2
4. KX = 0, pq (X) = 0, q(X) = 1 and b2 (X) = 2.
2 2 2
Proof. As KX is nef, then KX ≥ 0: then we can have only KX > 0 or KX = 0.
First of all, notice that if pg (X) > 0, then Pm (X) > 0 for every m > 0. In order
to show that P2 > 0 we can then only look at the case where pg (X) = 0.
So let us suppose that pg (X) = 0. Use the Noether Formula to get

2
12(1 − q(X)) = KX + 2 − 4q(X) + b2 (X).

2
If KX = 0 and q(X) = 1, then one gets b2 (X) = 2, and the last case is shown.
If q(X) > 1, one gets
b2 (X) = 10 − 8q(X) < 0,
2
which is not possibile. Hence, if KX = 0 and pg (X) = 0, we need q(X) = 1 and
b2 (X) = 2, or q(X) = 0 (so that b2 (X) = 10).
Now, use the following easy lemma:

Lemma 4.2.14. Let X be a smooth projective surface, and let L be a nef line
bundle on X. If h0 (X, L −1 ) 6= 0, then L ' OX .Recall that a minimal surface
X is such that κ(X) = −∞ if and only if KX is not nef.
126 The Enriques-Castelnuovo Classification

Proof. Suppose L is not trivial, and let C be an effective curve transversal to

a divisor defined by L −1 : then

0 ≤ L · OX (C) = −L −1 · OX (C) < 0,

and we are done.

As KX is nef, we need h0 (X, OX (rKX )) = 0 for every r < 0 by Lemma

4.2.14. Hence if m > 1 we have h0 (X, OX (−(m − 1)KX )), and

Pm (X) = h0 (X, OX (mKX )) = h0 (X, OX (mKX )) + h0 (X, OX (−(m − 1)KX )) ≥

1 2
≥ m(m − 1)KX + 1 − q(X) + pg (X).
2
2
If KX > 0 and m = 2 we have then
2
P2 (X) ≥ KX + 1 − q(X) + pg (X) > 0.

2
If KX = 0 and q(X) = 0, then P2 (X) ≥ 1 + pg (X) > 0.

The reason why I decided to put this Proposition here is that it as an easy
corollary on rational surfaces. This Theorem, which here appears as an easy
corollary of previously shown results was one of the main points in the birational
classification of smooth projective surfaces given by Enriques and Castelnuovo.
Without the Kawamata-Mori Rationality Theorem, its proof is hard and tricky.

Theorem 4.2.15. (Castelnuovo’s Rationality Theorem). Let X be a min-

imal smooth projective surface. Then X is rational if and only if q(X) = 0 and
P2 (X) = 0.

Proof. Let us suppose X rational. Then by Proposition 4.2.4 we have q(X) =

P2 (X) = 0.
Let us show the converse. If X is smooth projective surface such that q(X) =
P2 (X) = 0, then by Proposition 4.2.13 KX is not nef (if X is nef, we have
P2 (X) > 0 or q(X) > 0). Then, as X is minimal, we have κ(X) = −∞, so that
X can only be either geometrically ruled or P2 . In any case, as q(X) = 0, we
need X to be rational.

Remark 4.2.4. Notice that if P2 (X) = 0, then even pg (X) = 0 by Serre’s Dual-
ity. Hence it seems quite reasonble to expect that the Castelnuovo Rationality
Theorem works even with the hypothesis pg (X) = q(X) = 0. This is not the
case, as we are going to see with Enriques surfaces.
An interesting corollary of the Castelnuovo Rationality Theorem concerns
unirational surfaces.

Definition 4.2.4. A smooth projective variety X is unirational if there is a

dominant rational map Pn 99K X, where n = dim(X).
Non-negative Kodaira dimension 127

It is a well known result that every unirational curve is rational: indeed if

C is a smooth curve which is unirational then there is a surjective morphism
f : P1 −→ C. If ω is a holomorphic form on C, then f ∗ ω is a holomorphic form
on P1 . But as h1 (P1 , OP1 (KP1 )) = 0, we have f ∗ ω = 0, so that ω = 0. Hence
g(C) = 0, and C is rational. This result is known as the Lüroth Theorem, and
one would like to generalize it to any dimension.

Corollary 4.2.16. Let X be a smooth projective surface which is unirational.

Then X is rational.

Proof. As X is unirational, there is a dominant map P2 99K X. Resolving the

indeterminacies we get a dominant morphism R −→ X, where R is a rational
surface. Hence q(R) = P2 (R) = 0 by the Castelnuovo Rationality Theorem, so
that q(X) = P2 (X) = 0. Again by the Castelnuovo Rationality Theorem, one
then concludes that X is rational.

This result is not true in higher dimensions.

4.3 Non-negative Kodaira dimension

In this section we continue with the classification of surfaces. As for Kodaira
dimension −∞ we are done, we need to look at those smooth projective surfaces
of non-negative Kodaira dimension.
Thus, we can have Kodaira dimension 0, 1 or 2. We start by giving some
definitions and important examples, introducing the notion of elliptic surface,
K3 surface, abelian surface, Enriques surface, bielliptic surface, surface of gen-
eral type. The classification proceeds using important properties of fibrations
and of Abelian tori.

4.3.1 Definitions and examples

The first definition we need id the following, which generalizes the notion of
geometrically ruled surface.

Definition 4.3.1. Let X be a surface and C be a smooth curve. A morphism

f : X −→ C is called fibration if it is surjective and has connected fibers. If the
generic fiber of f is a smooth curve of genus g, then f is called g−fibration.

Notice that if all the fibers of f have genus 0, then f gives X the structure
of a geometrically ruled surface. If X is minimal and f is a 0−fibration, then
X is a geometrically ruled surface. A particular and very important case is the
following:

Definition 4.3.2. A 1−fibration is called elliptic fibration. If a surface X

admits an elliptic fibration, then X is called elliptic surface.
128 The Enriques-Castelnuovo classification

Let me give some easy example:

Example 4.3.1. Let C, F be two smooth curves, and consider X := C × F . We
have then two fibrations on X: one is pC : X −→ C, the projection onto C,
which is a g(F )−fibration; the other is pF : X −→ F , the projection onto F ,
which is a g(C)−fibration. For this kind of surfaces is very easy to calculate
invariants. Indeed, we have ΩX ' p∗C ΩC ⊕ p∗F ΩF , so that

q(X) = h0 (X, ΩX ) = h0 (C, ΩC ) + h0 (F, ΩF ) = g(C) + g(F ).

Moreover, we have OX (KX ) ' p∗C OC (KC ) ⊗ p∗F OF (KF ), so that

pg (X) = h0 (X, OX (KX )) = h0 (C, OC (KC )) · h0 (F, OF (KF )) = g(C) · g(F ),

and similarily we get

Pm (X) = Pm (C) · Pm (F ).
Hence, we have the following cases:

1. one between C and F is a smooth rational curve, and without loss of

generality we can suppose that it is F . Then Pm (X) = 0, so that κ(X) =
−∞. Moreover q(X) = g(C) and pg (X) = 0. In fact, X is a geometrically
ruled surface.

2. the two curves C and F are elliptic. Hence q(X) = 1, pg (X) = 1 and
κ(X) = 0.

3. one of the two curves is elliptic and the other has higher genus. Then
q(X) ≥ 3, pg (X) ≥ 2 and κ(X) = 1.

4. the two curves are not elliptic nor rational. Hence q(X) ≥ 4, pg (X) ≥ 4
and κ(X) = 2.

This is then an important family of examples, as in this way we are able to pro-
duce examples of smooth projective surfaces of any possible Kodaira dimension.
Example 4.3.2. Let C and F be two smooth curves, and consider the group
Aut(F ) of automorphisms of F . Consider an open cover {Ui }i∈I of C, and for
every i ∈ I let Xi := Ui × F . For every i, j ∈ I such that Uij 6= ∅, consider a
transition function ϕij : Uij −→ Aut(F ). Then we can glue together Xi and Xj
using ϕij . This gives a surface X with the structure of a g(F )−fibration.
For the next example we need the following lemma:

Lemma 4.3.1. Let D and D0 be two smooth curves, and suppose that D0 is the
quotient of D under the action of a finite group G. Then we have
 m   
X 1
OD (mKD ) ' f ∗ OD0 mKD0 + 1− Qj ,
j=1
rj

for every m > 0.

Non-negative Kodaira dimension 129

Proof. For this, let us consider the following: let D and D0 be two smooth
curves, and let f : D −→ D0 be a morphism. Consider the ramification points
P1 , ..., Pn ∈ D, and let ri ∈ N be the ramification index at Pi . The ramification
divisor is
Xn
R := (ri − 1)Pi ,
i=1

and the Hurwitz formula gives OD (KD ) = f ∗ OD0 (KD0 ) ⊗ OD (R). A special
case is when D −→ D0 is the quotient under the action of a group G: then
the ramification index is the same for every point of a fiber. If Q1 , ..., Qm ∈ D0
are points over which we have ramification, we have a ramification index rj for
every j = 1, ..., m for every point Rk lying over Qj , and
m
X
R= (rj − 1)f −1 (Qj ).
j=1

Now, recall that f ∗ Qj = ej f −1 (Qj ), so that

m  
X 1
R= 1− f ∗ Qj .
j=1
rj

As a consequence, we get a formula for the pluricanonical bundle of D:

 m   
X 1
OD (mKD ) ' f OD0 mKD0 +
∗
1− Qj .
j=1
rj

Example 4.3.3. Let a, b ∈ Z, and consider G := (Z/Za) ⊕ (Z/Zb). Let F be an

elliptic curve such that G is a subgroup of the group ot translations of F , i. e.
G acts on F as translations. Moreover, let C −→ P1 be a Galois cover of P1
ramified along three different points, and whose covering group is G. Hence G
acts on C × F , and we can consider X := (C × F )/G. It has two fibrations: one
is p : X −→ P1 whose fiber is F , and one is q : X −→ F/G whose fiber is C.
We have the following equalities:

H 0 (X, ΩX ) = H 0 (C × F, ΩC×F )G = H 0 (C, ΩC )G ⊕ H 0 (F, ΩF )G ,

and similar for

H 0 (X, OX (mKX )) = H 0 (C, OC (mKC ))G ⊗ H 0 (F, OF (mKF ))G .

By Lemma 4.3.1 we have

   X 
1
h (C, OC (mKC )) = h P , O m − 2 +
0 G 1 0
1− ,
j
rj
130 The Enriques-Castelnuovo classification

and
  X  
1
h (F, OF (mKF )) = h
0 G 0
F/G, OF/G mKF/G + 1− Qj ,
j
sj

where sj is the ramification index for F and G. As a conclusion, we get

q(X) = g(P1 ) + g(F/G) = g(F/G), pg (X) = g(P1 )g(F/G) = 0,

Pm (X) = h0 (P1 , O(−2m + Rm (C, G))) · h0 (F/G, OF/G (mKF/G + Rm (F, G))),

where Rm (D, G) := j (1 − r1j )Qj is the ramification divisor on D0 = D/G.

P

More generally, on can consider C to be the Galois cover of C 0 with Galois

group G, and one gets similar formulas for the invariants:

q(X) = g(C/G) + g(F/G), pg (X) = g(C/G) · g(F/G),

Pm (X) = h0 (C/G, O(mKC/G + Rm (C, G))) · h0 (F/G, O(mKF/G + Rm (F, G))).

As a particular case of this example, we get an important definition:

Definition 4.3.3. Let E and F be two elliptic curves, and let G be a group
of translations of F acting on E. The surface X := (E × F )/G is a bielliptic
surface if pg (X) = 0.

By Example 4.3.3 the surface X is bielliptic if and only if one of the two
quotients between E/G and F/G is a smooth rational curve. We now introduce
the remaining definitions and examples we will need in the following.

Definition 4.3.4. A surface X is a K3 surface if q(X) = 0 and KX = 0.

A surface X is an Enriques surface if q(X) = 0, pg (X) = 0 and 2KX = 0.
A surface X is an abelian surface if it is a 2−dimensional complex torus
admitting an embedding in a projective space.
A surface X is a surface of general type if κ(X) = 2.

Let me just finish this section with some examples:

Example 4.3.4. An example of abelian surface is the following: let E and F be
two elliptic curves. Then X := E × F is an abelian surface. Indeed, we have
E = C/Γ and F = C/Γ0 for some lattices Γ, Γ0 , and as they are curves they admit
an embedding in some projective spaces jE : E −→ Pn and jF : F −→ Pm .
Hence X admits an embedding in a projective space
(jE ,jF ) s
X −→ Pn × Pm −→ PN ,

where s is the Segre embedding. But now notice that X ' C2 /L, where L =
Γ × Γ0 . Hence X is a complex torus. In conclusion, X is an abelian surface.
Non-negative Kodaira dimension 131

Example 4.3.5. We already met an example of K3 surface: in P3 consider a

generic quadric surface X. As we have seen in Example 3.2.3, we have that X
is smooth and KX = 0. To show that it is indeed a K3 surface, we need to
prove that q(X) = 0. This follows from a Theorem of Lefschetz: as P3 is simply
connected, a generic hypersurface is simply connected too, so that b1 (X) = 0.
Then q(X) = 0, and X is a K3 surface. More generally, we have the following
example: let n ∈ N and consider d1 , ..., dn ∈ N. In Pn+2 let X be the complete
intersection of multidegree (d1 , ..., dn ), i. e. X = H1 ∩ ... ∩ Hn , where Hj is a
hypersurface of Pn+2 of degree dj . Then we have

OX (KX ) = OPn+2 (−n − 3 + d1 + ... + dn )|X .

If d1 + ... + dn = n + 3, then we have that KX = 0. Again by the Lefschetz

Theorem we have that q(X) = 0, so that X is a K3 surface. Notice that if n = 1,
then we need d1 = 4, so that X is the generic quartic in P3 we saw before. If
n = 2, then we need d1 + d2 = 5, so that d1 = 2 and d2 = 3; if n = 3 we have
d1 + d2 + d3 = 6, so that d1 = d2 = d3 = 2, and so on.
Example 4.3.6. We now describe an example of Enriques surface. If X is a K3
surface with an involution i : X −→ X, i. e. i ◦ i = idX , which has no fixed
points, we can consider the quotient X e := X/i. We have p : X −→ X e which is
∗
a degree 2 cover. If C is a curve on X, it is easy to see that p∗ p C = 2C, hence
e
by linearity we get p∗ p∗ D = 2D for every divisor D. In particular we have
p∗ p∗ KXe = 2KXe . But now notice that p∗ KXe = KX , hence it is trivial since X
is a K3 surface. Then 2KXe = 0. Moreover, we have χ(OX ) = 2χ(OXe ). Notice
that as X is a K3 surface, we have q(X) = 0 and pg (X) = 1, so that χ(OX ) = 2.
Hence χ(OXe ) = 1, which implies q(X) e = pg (X). e Now, as p is a double cover,
we easily get that q(X)e = q(X) = 0, so that pg (X) e = 0. Notice that since
h (X, OXe (KXe )) = 0, we need to have KXe 6= 0. Hence X is an Enriques surface.
0 e

The only problem now is to find an example of K3 surface having an invo-

lution without fixed points. In P5 consider the surface X obtained as complete
intersection of three quadrics. If x0 , ..., x5 are homogenous coordinates on P5 ,
we can consider the quadrics of the form

Pi (x0 , x1 , x2 ) + Qi (x3 , x4 , x5 ) = 0,

where i = 1, 2, 3 and Pi , Qi are homogenous polynomials of degree 2 for every

i = 1, 2, 3. For generic choices of Pi and Qj , the surface X is smooth (by
Bertini), and it has a natural involution

i : X −→ X, i(x0 : x1 : x2 : x3 : x4 : x5 ) := (x0 : x1 : x2 : −x3 : −x4 : −x5 ).

This involution has no fixed points if and only if the three quadrics P1 , P2 , P3
(resp. Q1 , Q2 , Q3 ) have no common points. Hence, choosing them generic we
have an involution without fixed points. In conclusion X/i is an Enriques sur-
face.
132 The Enriques-Castelnuovo classification

4.3.2 Fibrations
Before starting the proof of the Enriques-Castelnuovo Classification for non-
negative Kodaira dimension, we need to collect some basic facts about fibrations.
Let me start with the following.

Lemma 4.3.2. (Zariski’s Lemma). Let X be a surface and X a smooth

P
curve. Let f : X −→ C be a fibration, and let F = i mi Ci be a fiber, where
P
mi ∈ N and Ci are the irreducible components of F . Moreover, let D := i ri Ci ,
with ri ∈ Q. Then D2 ≤ 0, and D2 = 0 if and only if D = rF for some r ∈ Q.
P
Proof. Let Fi := mi Ci , so that we can write F := i Fi . Let si := ri /mi , so
P
that D = i si Fi . We have
X X
D2 = s2i Fi2 + 2 si sj Fi · Fj =
i i<j

X X X
= s2i Fi · F − (s2i + s2j − 2si sj )Fi · Fj = − (si − sj )2 Fi Fj ≤ 0,
i i<j i<j

as Fi · F = 0 and Fi · Fj ≥ 0 as F is connected (f is a fibration). Notice that in

order to have D2 = 0 we need si = sj for every i, j (as there is at least one pair
P
i, j such that Fi · Fj > 0. Hence let r := si , so that D = i rFi = rF , and we
are done.

Let me introduce the following:

Definition 4.3.5. Let X be a surface, C a smooth curve and f : X −→ C a

P
fibration. Let F = i mi Ci be a fiber, where mi ∈ N and Ci are the irreducible
components of F . The multiplicity of F is m(F ) = gcd(mi ). The fiber F is a
multiple fiber if and only if m(F ) > 1.

Let now f : X −→ C be a fibration on the surface X, and let c ∈ C. Consider

F = f −1 (c) the fiber over c, and suppose it is multiple, i. e. F = mF 0 where
F 0 is a fiber and m > 1. Consider an open neighborhood U of c ∈ C, and let z
be a local coordinate around c on U . Then, the function z ◦ f : f −1 (U ) −→ C
vanishes with order m along F 0 . As the fibers are connected, every holomorphic
function g over f −1 (U ) is of the form h ◦ f , where h is a holomorphic function
on U . Hence every g ∈ OX (f −1 (U )) vanishes with order at least m along F 0 .
Now, consider the line bundle Of −1 (U ) (mF 0 ). By the previous discussion,
this sheaf is trivial, hence every function in Of −1 (F 0 ) has order at most m. If
a function g ∈ Of −1 (U ) (F 0 ) vanishes with order smaller than m, then we would
get a function on f −1 (U ) vanishing with order smaller than m along F 0 , which
is not possible. Hence, the line bundle Of −1 (U ) (F 0 ) is torsion of order m in
P ic(f −1 (U )). we have the following:
Non-negative Kodaira dimension 133

Proposition 4.3.3. Let f : X −→ C be a fibration on the surface X, and let

F = mF 0 be a fiber, m ∈ N. The line bundle OF 0 (F 0 ) is torsion of order m in
P ic(F 0 ).

Proof. For the proof of this proposition we need the following lemma:

Lemma 4.3.4. Let X be a surface, C an effective divisor on X. Consider

rC : OX −→ OC the restcriction morphism, and its analogue rC ∗
: OX
∗
−→ OC∗ .
∗
Then we have exp ◦ rC = rC ◦ exp. In particular, the exponential sequence of X
remains exact when restricted to C.

Proof. We can assume that C is connected. Let g ∈ IC , so that rC (g) = 0 and

∗
exp(rC (g)) = 1. We need to show that rC (exp(g)) = 1. Notice that exp(g) =
1 + m≥1 (2πig )/(m!), so that exp(g) − 1 ∈ IC . Hence rC
m ∗
P
(exp(g) − 1) = 0,
∗
so that rC (exp(g)) = 1, and we are done.
For the remaining part, it is clear that the exponential is the surjective, and
we only need to prove the following: let f ∈ OX such that rC ∗
(exp(f )) = 1. Then
f|C = n for some n ∈ Z. Indeed, we have exp(f ) − 1 ∈ IC . Let C = i ni Ci ,
P

where ni ∈ N and Ci is an irreducible curve. We have that f|Ci = mi by the

classical exponential sequence, and by connectedness of C we need mi to be
constant, i. e. f|Ci = n for some integer n. Hence f − n vanishes with some
order mi along every component mi . Then even exp(f ) − 1 vanishes with order
mi along every Ci , but as exp(g) − 1 ∈ IC we need mi = ni . In conclusion we
have f − n ∈ IC , so that f|C = n.

It is clear that F 0 is torsion in P ic(F 0 ) and that its period divises m. By

topological facts, the fiber F 0 is deformation retract of some open neighborhood,
and we can suppose it to be of the form V := f −1 (U ) for some open subset U
of C. The restriction map

H p (V, Z) −→ H p (U, Z)

is then an isomorphism for every p. By the previous lemma we then get a

commutative diagram
a b c
H 0 (V, Z) −−−−→ H 0 (OV ) −−−−→ P ic(V ) −−−−→ H 2 (V, Z)
 
 
ey

dy
a0 b0 c0
H 0 (F 0 , Z) −−−−→ H 0 (OF 0 ) −−−−→ P ic(F 0 ) −−−−→ H 2 (F 0 , Z).
Now, let us suppose that OF 0 (F 0 ) is torsion of period p, and consider pF 0 ∈
P ic(V ). As c(pF 0 ) = c0 (pF 0 ) = 0, there is α ∈ H 1 (V, OV ) such that b(α)pF 0 . By
commutativity we have e(b(α)) = b0 (d(α)) = 0, so that there is β ∈ H 1 (F 0 , Z)
such that d(α) = a0 (β). Hence α = d(β), so that pF 0 = 0 in P ic(V ). But F 0
as period m in P ic(V ), so that p is a multiple of m. Hence p = m, and we are
done.
134 The Enriques-Castelnuovo classification

There are still two results we will need on fibrations, namely a topological
one and something about isotrivial fibrations. The first one is the following, for
which we need the following lemma:
P
Lemma 4.3.5. Let C = i Ci be a curve on a surface X, where Ci are the
irreducible components of C. Then e(C) ≥ χ(OC ), and e(C) = 2χ(OC ) if and
only if C is smooth.

Proof. Consider the normalization ν : C e −→ C of the curve C. We have

CC ⊆ ν∗ CCe , and let δ be the quotient. Similarily, we have OC ⊆ ν∗ OCe , and
let ∆ be the quotient. Hence we have e(C) e = e(C) + h0 (C, δ) and χ(O e ) =
C
0 e = 2χ(O e ), since χ(O e ) = 1 − g(C)
χ(OC ) + h (C, ∆). Moreover, we have e(C) C C
e
and e(C) = 2 − b1 (X), where b1 (C) = 2g(C). Hence we have
e e e e

e(C) = 2χ(OC ) + 2h0 (C, ∆) − 2h0 (C, δ).

The natural map δ −→ ∆ is an injection, so that h0 (C, δ) ≤ h0 (C, ∆), so that

e(C) ≥ 2χ(OC ), and equality holds if and only if C is smooth.

Using this one proves the following:

Proposition 4.3.6. Let f : X −→ C be a fibration, and suppose that X is a

surface with KX nef. Moreover, let δ(f ) be the set of critical values of f , i. e.
δ(f ) := {c ∈ C | df (x) = 0 f or every x ∈ f −1 (c)}. Let F be a smooth fiber of
f , and for every c ∈ δ(f ) let Fc be the fiber over c. Then
X
e(X) = e(C)e(F ) + (e(Fc ) − e(F )).
c∈δ(f )

Moreover, we have e(Fc ) ≥ e(F ) for every c ∈ δ(f ), and e(Fc ) = e(F ) if and
only if Fc supports a smooth elliptic curve.

Proof. As f is locally a trivial fibration over the set C \ δ(f ), we can show that
X
e(X) = e(X \ f −1 (δ(f ))) + e(f −1 (δ(f ))) = e(C \ δ(f ))e(F ) + e(Fc ).
c∈δ(f )
P
Notice that e(C \ δ(f )) = e(C) − c∈δ(f ) 1, so that we finally get the first part
P
of the statement. Now, consider a singular fiber Fc = i mi Ci , where mi ∈ N
and Ci are the irreducible components of Fc . Let Fc0 := i Ci . By Lemma
P

4.3.5 we have e(Fc0 ) ≥ χ(OFc0 with equality if and only if Fc0 supports a smooth
elliptic curve. By the adjunction formula we have

e(Fc0 ) ≥ 2χ(OFc0 ) = −(Fc0 )2 − Fc0 · KX .

By Lemma 4.3.2 we have (Fc0 )2 ≤ 0, and

X X
− Ci · KX ≥ − mi Ci · KX = −Fc0 · KX = −F · KX = e(F ).
i i
Non-negative Kodaira dimension 135

The first inequality follows indeed from the fact that KX is nef, and the last
equality follows from andjunction formula, the fact that F 2 = 0 and Lemma
4.3.5. In conclusion we get e(Fc ) ≥ e(F ). Equality happens if and only if
(Fc0 )2 = 0, i. e. by the Zariski Lemma if and only if Fc = nFc0 for some positive
integer n. We then get e(Fc0 ) = e(F )/n. Notice that as KX is nef the generic
fiber is not rational, so that e(F ) ≤ 0. Hence e(Fc0 ) = e(F ) = 0, and we are
done.

The conclusion of this section deals with isotrivial fibrations.

Definition 4.3.6. Let X and Y be two manifolds, and f : X −→ Y be a

fibration. Then f is isotrivial if there is a finite unramified cover g : Y 0 −→ Y
such that the pull-back f 0 : X ×Y Y 0 −→ Y 0 is isomorphic to a trivial fibration
Y 0 × F −→ Y 0 for some projective manifold F .

Let me give a first example of an isotrivial fibration.

Example 4.3.7. Let Y be a manifold, and let G be a finite group which is a
quotient of π1 (Y ) and which acts on a projective manifold F . Consider g :
Y 0 −→ Y be the cover defined by G, and let X := (Y 0 × F )/G. Then X admits
a natural isotrivial fibration on Y . Even a converse is true: if X −→ Y is a fiber
bundle whose fiber F has a finite automorphism group is an isotrivial fibration
arising in this way.
The following proposition will be used in the proof of the classification.

Proposition 4.3.7. Let f : X −→ C be a fibration of a surface onto a curve

of genus 0 or 1. Then f is isotrivial.

Proof. The proof uses the theory of periods of curves, so we are just going to
sketch it. As f : X −→ C is a fibration, it is proper and surjective, hence it
is differentiably locally trivial (this is Ehresman’s Theorem). This implies that
the family ∪c∈C H1 (Fc , Z) is locally trivial in the euclidean topology of C. As
Fc is a curve, H1 (Fc , Z) is a rank 2g(Fc ) lattice. If one considers the sheaf
f∗ OX (KX ), it is coherent as f is proper, and as H 0 (Fc , OFc (KFc )) is locally of
constant dimension, hence f∗ OX (KX ) is a locally free sheaf whose rank is g(F ).
For every point y ∈ C there is an open neighborhood U over which H1 (Fc , Z) is
constant on U and f∗ OX (KX ) is free on U . This means that we can find a basis
γ1 , ..., γ2g(F ) of H1 (Fy , Z) over U , and a basis ω1 (y), ..., ωg(F ) (y) which depends
R
holomorphically on y, so that the matrix Ω := [ γi ωj (y)] is normalized, i. e. it
is symmetric and has positive definite imaginary part. Let now

Hg := {Ω ∈ Cg,g | Ω = ΩT , Im(Ω) > 0},

over which the symplectic group Γg acts. We then get a holomorphic map

U −→ Hg ,
136 The Enriques-Castelnuovo classification

which does not extend to C, but to C, e the universal cover of C, getting the
period map p : C −→ Hg .
e
Now, as g(C) = 0, 1, the universal cover of C is either P1 or C. As g(F ) =
0, 1, we have H1 bounded, so that p is constant. Then all the fibers have the same
period, so that by Torelli’s Theorem on curves we need them to be isomorphic.
The isotriviality follows from Example 4.3.7 if g(F ) ≥ 2, as in this case the
fiber has a finite group of automorphism. The case of genus 1 fiber is more
involved: the action of π1 (C) on the group of n−torsion points of the fiber can
be trivialized up to passing to a finite unramified cover. Hence, globally one
gets a zero section: but then the group of automorphism of an elliptic curve
which preserves the origin is finite, and one can use again Example 4.3.7.

4.3.3 Elliptic fibrations

Let me now come to elliptic fibrations. The first result we need is the following:

Lemma 4.3.8. Let f : X −→ C be an elliptic fibration, and let F = mF 0 be

a multiple fiber. Then F 0 is either an elliptic curve, a rational curve with one
ordinary double point or a cycle of smooth rational curves.

Proof. If F is a smooth fiber, we have KX · F = 0: indeed a smooth fiber is an

elliptic curve, so that g(F ) = 1. Moreover F 2 , hence the genus formula gives
KX · F = 0. Hence KX · F 0 = 0, so that χ(OF 0 ) = 0. If F is irreducible, then F 0
is either an elliptic curve or a rational curve with an ordinary double point. If
F 0 is reducible and Ci is a component of F 0 , we have Ci2 < 0 by Lemma 4.3.2.
Now, Ci2 6= −1 as X is minimal, and Ci · KX = 0. Hence by the genus formula
we get 2g(Ci ) − 2 = Ci2 < 0. Hence g(Ci ) = 0 and Ci2 = −2. Now, let us
consider two different components Ci and Cj : then Ci · Cj = 0, 1, 2. Indeed as
F is connected we have Ci · Cj ≥ 0. Moreover by Lemma 4.3.2 we have

0 ≥ (C1 + C2 )2 = 2C1 · C2 − 4,

so that C1 · C2 ≤ 2.
Now, F 0 is not simply connected. Indeed, in P ic(F 0 ) there is a non-trivial
torsion line bundle by Proposition 4.3.3, hence H 1 (F 0 , OF 0 ) 6= 0, so that b1 (F 0 )
is not 0. Now, let F 0 = k nk Ck , where ni ∈ N. If there are i 6= j such that
P

Ci · Cj = 2, then we have
X X
0 = Ci · F 0 = nk Ci · Ck = −2ni + nk Ci · Ck .
k k6=i

If ni ≤ nj we get
X
2ni = 2nj + nk Ci · Ck ,
k6=i,j
Non-negative Kodaira dimension 137

then we need nk = 1 for every k 6= i, j, and we have only two components,

which hence form a cycle. If Ci · Cj = 1, and three pairwise distinct components
meet in a point, one uses a similar argument to show that there are no more
components. But this implies that F 0 is simply connected, which is not the
case, so we need that the components of F 0 form a cycle of smooth rational
curves.

As a corollary we have the following:

Corollary 4.3.9. Let f : X −→ C be an elliptic fibration, and let F be a

primitive fiber. Then the dualising sheaf ωF := OF (KX + F ) is trivial.

Proof. If F is elliptic, it is smooth and ωF = OF (KF ). Hence ωF = OF and

we are done in this case. If F is not smooth, then by Lemma 4.3.8 F can only
be either a rational curve with an ordinary double point, or a cycle of smooth
rational curves. Hence we have either F is P1 or its components are P1 . There
is a global meromorphic 2−form that has two poles with opposite residues, and
we get a trivializing section of ωF .

The final result of this section is the following:

Theorem 4.3.10. Let f : X −→ C be an elliptic fibration, and suppose that

KX is nef. Let F1 , ..., Fm be the multiple fibers of f , and let ni := m(Fi ), so that
Fi = ni Fi0 . Then there is a divisor D on C such that deg(D) = χ(OX ) − χ(OC )
and
m
X
KX = f ∗ D + (ni − 1)Fi0 .
i=1

Proof. The proof is divided in two steps: first of all we show the formula for the
canonical bundle, and the we calculate the degree of D.
Consider N smooth fibers G1 , ..., GN of f , and consider the exact sequence
 N
X  N
M
0 −→ OX (KX ) −→ OX KX + Gi −→ OGi (KX + Gi ) −→ 0.
i=1 i=1

By adjunction formula we have OGi (KX + Gi ) ' OGi , as Gi is an elliptic curve.

That the long exact sequence induced by this gives
  N
X 
h X, OX KX +
0
Gi ≥ pg (X) − q(X) + (N − 1).
i=1
P
If N  0 then there is an effective divisor L ∈ |KX + i Gi |. Let now F be any
fiber, so that F · Gi = 0 and F · KX = 0 by the genus formula. Hence F · L = 0,
so that L consists of components Li of fibers, for i = 1, ..., M . As L is effective,
we have X
L · Li = KX · Li + Gj · Li = KX · Li ≥ 0,
j
138 The Enriques-Castelnuovo classification

as KX is nef. Now, let me prove the following:

Lemma 4.3.11. Let f : X −→ C be an elliptic fibration with KX nef. Then

2
KX = 0.
2
Proof. As KX is nef, we have KX ≥ 0. Moreover, we have κ(X) ≥ 0, so there
is n ∈ n such that |nKX | is not empty, and take D ∈ |nKX |. As D · F = 0, the
components of D are components of fibers. Then n2 KX 2
= D2 ≤ 0 by Lemma
2
4.3.2, so that KX = 0.

Using this, we get

X
2
0 = KX = KX · (L − Gi ) = KX · L,
i

so that KX · Li = L · Li = 0. In particular, we get L2 = 0. Now, let L = i L0i ,

P

where L0i is the part supported on a fiber. Hence (L0i )2 = 0, so that L0i is a
rational multiple of a fiber by Lemma 4.3.2. Any part supported on a multiple
fiber Fi can then be written in the form ki Fi0 + ri Fi , where 0 < ki < ni and
ri ∈ Z. The part supported on a non-multiple fiber is of the form rj Gj for some
rj ∈ Z. In conclusion we get
X
KX = f ∗ D + ki Fi0 .
i

Now, notice that

OFi0 ((ki + 1)Fi0 ) ' OFi0 (KX + Fi0 ) ' OFi0 ,

where the last isomorphism comes from Coroloary 4.3.9. By Proposition 4.3.3
we then have that ki + 1 is a multiple of ni . As ki + 1 ≤ ni we finally get
ni − 1 = ki , and the first step is proved.
Let us now calculate the degree of D. We compute h0 (X, OX (KX + j Gj ))
P

in two ways. The first one is the following: the divisor i (ni − 1)Fi0 is a fixed
P
P
component of the linear system |KX + j Gj |. Writing cj := f (Gj ) we then
get X X
h0 (X, OX (KX + Gj )) = h0 (X, OX (f ∗ D + Gj )) =
j j
X
= h0 (C, OC (D + cj )) = deg(D) + N + 1 − g(C),
j

for N  0, by Riemann-Roch. However, using the previous part we have

X
h0 (X, OX (KX + Gj )) = χ(OX ) + N − 1 + d,
j

where d := dim(im(H 1 (X, OX (KX )) −→ H 1 (X, OX (KX +

P
j Gj )))). Hence

deg(D) = χ(OX ) + g(C) − 2 + d.

Non-negative Kodaira dimension 139

We then only need to calculate d. By Serre’s Duality this is simply

X
d = dim(im(H 1 (X, OX (− GJ )) −→ H 1 (X, OX (KX )))) =
j

M
= dim(ker(H 1 (X, OX ) −→ H 1 (Gj , OGj ))).
j

Use now the Hodge decomposition: we have H 1 (X, OX ) ' H 0 (X, ΩX ), and
that H 1 (Gj , OGj ) ' H 0 (Gj , ΩGj ), hence

d = dim(ker(H 0 (X, ΩX ) −→ ⊕j H 0 (GJ , ΩGj ))).

This kernel is given by those holomorphic 1−forms which are pull-back of holo-
morphic 1−forms on C, so that d = g(C). In conclusion

deg(D) = χ(OX ) − 2 + 2g(C) = χ(OX ) − χ(OC ),

and we are done.

As a corollary of this, we have the following:

Corollary 4.3.12. Let f : X −→ C be an elliptic fibration. Then κ(X) ≤ 1.

P
In particular, κ(X) = 1 if and only if 2g(C) − 2 + χ(OX ) + i (ni − 1)/ni 6= 0.

Proof. The formula for the canonical bundle of X can be written an KX = f ∗ L,

where L is a divisor of degree d = 2g(C) − 2 + χ(OX ) + i nin−1
P
i
. Hence
mKX = f ∗ Lm , where Lm is a line bundle of degree md on C. Hence
 X ni − 1 
Pm (X) = h0 (C, OC (Lm )) ∼ m 2g(C) − 2 + χ(OX ) + .
i
ni

In conclusion κ(X) ≤ 1, and the remaining part is clear.

4.3.4 Classification for Kodaira dimension 0 and 1

We are finally able to conclude the classification for Kodaira dimension 0 or 1.
The first result we need is the following:

Lemma 4.3.13. Let X be a smooth minimal projective surface with κ(X) = 0.

2
Then KX = 0 and Pm (X) = 0, 1 for every m > 0, and there is m > 0 such that
Pm (X) = 1.

Proof. Let us first show the first item. As κ(X) = 0, there is m > 0 such that
2
|mKX | =6 ∅. Let D ∈ |mKX |, so that D · mKX ≥ 0 implies KX ≥ 0. Now,
suppose KX > 0. Then by Riemann-Roch we get
 
n
h (X, OX (nKX )) + h (X, OX ((1 − n)KX )) ≥ χ(OX ) +
0 0
KX2
.
2
140 The Enriques-Castelnuovo classification

For n > 1 we have |(1 − n)KX | = ∅, otherwise there would be E ∼ (1 − n)KX

2
such that E · KX ≥ 0, getting KX ≤ 0. In conclusion Pn (X) ∼ n, so that
2
κ(X) > 0, hence KX = 0.
The remaining part is easier: suppose that m > 0 is such that Pm (X) > 1, i.
e. there are at least two linearily independent σ, τ ∈ H 0 (X, OX (mKX )). Hence
in for every n > 0 the elements σ n , σ n−1 τ, ..., τ n are linearily independent in
H 0 (X, OX (nmKX )), i. e. Pnm (X) ≥ n + 1, so that κ(X) ≥ 1. But since
κ(X) = 0, this is not possibile, so that Pm (X) = 0, 1. Moreover, if for every
m > 0 we have Pm (X) = 0, then κ(X) = −∞, so that in order to have κ(X) = 0
there must be m > 0 such that Pm (X) = 1.

Here is the first point of the classification.

Proposition 4.3.14. Let X be a smooth minimal projective surface, and sup-

2
pose that KX is nef. If KX = 0, q(X) = 1 and pg (X) = 0, then κ(X) = 0, 1.
Moreover, in this case κ(X) = 0 if and only if X is a bielliptic surface.

Proof. As q(X) = 1, the Albanese torus Alb(X) associated to X is a curve. As

it has to be abelian, Alb(X) is an elliptic curve. Recall that

aX : X −→ Alb(X)

2
has connected fibers by Proposition 3.3.6. Now, as KX is nef, KX = 0, q(X) = 1
and pg (X) = 0, by Proposition 4.2.13 we need that b2 (X) = 2. Moreover, as
q(X) = 1 we have b1 (X) = 2, so that e(X) = 0. By Proposition 4.3.6 we have
then X
e(Alb(X))e(F ) = − (e(Fc ) − e(F )) ≤ 0.
c∈δ(aX )

By Lemma 4.3.5 we have for every smooth curve C that e(C) = 2 − 2g(C) ≤ 2.
As KX is nef, the generic fiber F cannot be rational, so that e(F ) ≤ 0. We
have then only two possibilities: either e(F ) < 0, so that g(F ) ≥ 2; or e(F ) = 0,
so that the generic fiber is elliptic. But then for every c ∈ δ(aX ) we have
e(Fc ) = e(F ), which by Proposition 4.3.6 happens if and only if Fc supports on
an elliptic curve.
Let us start with the first case, i. e. g(F ) ≥ 2 for the generic fiber, so that
e(F ) < 0. As e(Alb(X)) · e(F ) ≤ 0, we can only have e(Alb(X)) = 0, 1. Hence
we can apply Proposition 4.3.7 to get that aX is the isotrivial. Hence, there is
a finite unramified cover Xe −→ X such that X e = Alb(X) × F . Recall that the
κ(X) = κ(X) by Proposition 3.2.7, hence κ(X) = κ(Alb(X) × F ). Now, notice
e
that g(F ) > 1, and g(Alb(X)) = 0, 1. If g(Alb(X)) = 0, then e(Fc ) = e(F ) for
every c ∈ δ(aX ), so that by Proposition 4.3.6 the fiber F is elliptic. This is not
possibile, so that Alb(X) is an elliptic curve. By Example 4.3.1 we then get
that κ(X) = 1. Notice that in this case X is not bielliptic, as every bielliptic
surface has Kodaira dimension 0.
Non-negative Kodaira dimension 141

Let us then suppose that g(F ) = 1, so that every fiber is elliptic. If there are
multiple fibers, then one can use Corollary 4.3.12 to get that κ(X) = 1: indeed
P
χ(OX ) = 0, so that κ(X) = 1 if and only if 2g(Alb(X)) − 2 + i (ni − 1)/ni 6= 0.
If g(Alb(X)) > 0 and there are multiple fibers, this is automathic. Let us
then suppose that g(Alb(X)) = 0, so that aX is an elliptic fibration over P1 .
Pk
If κ(X) = 0 then χ(OX ) − 2 + k − i=1 1/ni = 0 by Corollary 4.3.12. As
Pk
χ(OX ) ≥ 0, we then get 2 − k + i=1 1/ni ≥ 0. Hence there are only the
following possibilites:

χ(OX ) = 1, k = 2, n1 = n2 = 2,

χ(OX ) = 2, k = 3, n1 = n2 = n3 = 3,

χ(OX ) = 2, k = 3, n1 = 2, n2 = n3 = 4,

χ(OX ) = 2, k = 3, n1 = 2, n2 = 3, n3 = 6,

χ(OX ) = 2, k = 4, n1 = n2 = n3 = n4 = 4.

In conclusion, we get that κ(X) = 0 if χ(OX ) = 1, 2, but we have χ(OX ) = 0,

so that κ(X) = 1. We can then suppose that there are no multiple fibers.
We have then a surface with an elliptic fibration aX which is everywhere
of maximal rank. Then by Proposition 4.3.6 we get that e(Alb(X)) = 0, so
that g(Alb(X)) = 1, and one can then apply Proposition 4.3.7 to get that aX
is an isotrivial fibration. In conclusion there are two elliptic curves E and F
and a group G of automorphisms of F acting as translations on E such that
X = (E × F )/G. As by hypothesis we have pg (X) = 0, then X is a bielliptic
surface.

Let me first conclude the classification for Kodaira dimension 1.

Theorem 4.3.15. Let X be a smooth minimal projective surface such that KX

2
is nef and KX = 0. Then κ(X) = 0, 1. Moreover, every smooth minimal
projective surface of Kodaira dimension 1 admits an elliptic fibration.

Proof. As KX is nef and X is minimal, we have κ(X) ≥ 0. Let us suppose tha

κ(X) ≥ 1. Hence, for n  0 the linear system |nKX | has at least dimension 1.
Let Df be the fixed part of this linear system, and let D be a moving divisor in
it. Then

0 = n2 KX = KX · (D + Df ) = KX · D + KX · Df ≥ 0,

as KX is nef. Hence KX · D = KX · Df = 0. Now

0 = nD · KX = D2 + D · Df ≥ 0,

as D is moving, so that D2 = D · Df = 0.
142 The Enriques-Castelnuovo classification

In particular, this implies that the map Sf := ϕnKX : X −→ |nKX | maps

every divisor in |D| to a point, hence the image of this map is a curve. As this
is true for every n  0, we then get that |nKX | has dimension at most 1, so
that κ(X) = 1.
Let us then suppose that κ(X) = 1. By the genus formula, as F · KX = 0
for every smooth fiber F , we get that g(F ) = 1, hence the generic fiber is an
elliptic curve, and X admits an elliptic fibration.

Definition 4.3.7. An elliptic surface of Kodaira dimension 1 is called proper

elliptic surface.

Using this terminology, the second part of the previous statement can be
restated by saying that every surface of Kodaira dimension 1 is proper elliptic.
The final step in the classification is the following:

Theorem 4.3.16. Let X be a smooth minimal projective surface with κ(X) = 0.

Then X can only be one of the following:

1. a K3 surface;

2. an abelian surface;

3. a bielliptic surface;

4. an Enriques surface.
2
Proof. As κ(X) = 0, we have that KX = 0 and KX is nef. Hence, by Proposition
4.2.13 we have only three possibilites. The first one is q(X) = 1 and pg (X) = 0,
and in this case X is bielliptic by Proposition 4.3.14.
For the remaining cases, notice that we have only two possibilities: either
pg (X) = 0, or pg (X) > 0. In the first case, we need q(X) = 0 (otherwise
q(X) = 1 by Proposition 4.2.13, and hence X is bielliptic), so that P2 (X) = 1.
Let us consider pg (X) = 0, and let us suppose that P3 (X) 6= 0. Then
P3 (X) = 1 as κ(X) = 0. As P2 (X) = P3 (X) = 1, then let D2 ∈ |2KX | and
D3 ∈ |3KX |, so that 3D2 , 2D3 ∈ |6KX |. Since P6 (X) ≤ 1 as κ(X) = 0, then we
need 3D2 = 2D3 , and there must be an effective divisor D such that 2D = D2
and 3D = D3 . But then D ∈ |KX |, but as pg (X) = 0 then this is not possible,
hence P3 (X) = 0. Hence h0 (X, OX (3KX )) = 0, so that Riemann-Roch we get
h0 (X, OX (−2KX )) ≥ 1. But as P2 (X) = h0 (X, OX (2KX )) = 1, this is possible
if and only if 2KX = 0. Hence in this case X is an Enriques surface.
We are then left with the case pg (X) > 0. This implies that pg (X) = 1, and
consider the Noether Formula

12(2 − q(X)) = e(X) = 2 − 4q(X) + b2 (X).

Then b2 (X) = 22 − 8q(X), which clearly implies that q(X) = 0, 1, 2.

Non-negative Kodaira dimension 143

If q(X) = 0, then by the Riemann-Roch inequality we get

h0 (X, OX (2KX )) + h0 (X, OX (−KX )) ≥ 2.

As h0 (X, OX (2KX )) = 0, 1 we then get h0 (X, OX (−KS )) ≥ 1. As pg (X) = 1,

this implies that KX has to be trivial. In conclusion, in this case X is a K3
surface.
We can now suppose pg (X) = 1 and q(X) > 0, so that q(X) = 1, 2. As
q(X) > 0, we have P ic0 (X) 6= 0, so that there is a non-tivial line bundle OX (τ )
such that OX (2τ ) ' OX . If q(X) = 1, then use the Riemann-Roch inequality
to get that
h0 (X, OX (τ )) + h0 (X, OX (KX − τ )) ≥ 1.

Then h0 (X, OX (KX − τ )) ≥ 1, so that there is D ∈ |KX − τ |. Then 2D = 2KX ,

which implies D = KX , so that τ is trivial. But this is not possible, hence
q(X) 6= 1.
In conclusion we need q(X) = 2, consider the canonical bundle KX =
P
i mi Ci , where Ci are the components of KX . As KX is nef we get
X
2
0 = KX = mi Ci · KX ,
i

so that Ci · KX = for every i. Similarily we have

X
0 = Ci · KX = mi Ci2 + mj Ci · Cj ≥ mi Ci2 ≥ 0.
j6=i

In conclusion Ci2 ≤ 0 and we have only two possibilities: as

2g(Ci ) − 2 = Ci2 ≤ 0,

we need either that Ci2 = −2 and g(Ci ) = 0, or Ci2 = 0 and g(Ci ) = 1. In this
last case we get Ci · Cj = 0 fpor every i, j, hence we can divide the components
Ci of KX in two families: one is given by smooth rational curves, the other by
smooth elliptic curves or rational curves with an ordinary double point.
Consider the Albanese map aX : X −→ Alb(X). As q(X) = 2, then Alb(X)
is an abelian surface, and the image of aX is either a curve C or the whole
Alb(X). In the first case, as q(X) = 2, the curve C has genus 2 by Proposition
3.3.6, and we have a fibration aX : X −→ C. As the components of the canonical
bundle are either rational or elliptic, they cannot be mapped via aX on C, so
they must be contracted to points. By the Zariski Lemma, this implies that KX
is a multiple of a fiber. Let F be such a fiber, and consider D = (a/b)F , where
a, b ∈ N. Then bD = a∗X (aP ) for some point P ∈ C: hence h0 (X, OX (nbKX ))
and h0 (X, OX (nbD)) grow indefinitely with n. As κ(X) = 0 this is not possible,
so KX has to be a trivial multiple of a fiber, i. e. KX = 0. Now, consider a cover
C 0 −→ C which is unramified and of degree d ≥ 2. Pulling-back the fibration
144 The Enriques-Castelnuovo classification

we get a surface X 0 together with a fibration f 0 : X 0 −→ C 0 and an unramified

cover X 0 −→ X of degree d. Hence KX 0 = 0, χ(OX 0 ) = dχ(OX ) = 0, so that
q(X 0 ) = 2. But q(X 0 ) ≥ g(C 0 ) ≥ 3, so that the possibility of having im(aX ) of
dimension 1 is not possible.
Hence aX has 2-dimensional image, which is the torus Alb(X). As every
2−cycle on Alb(X) is homologous to a 2−cycle which lifts to X, the map

a∗X : H 2 (Alb(X), C) −→ H 2 (X, C)

is injective. Now, notice that b2 (X) = 22 − 8q(X) = 6, and b2 (Alb(X)) = 6

as it is a torus, so that a∗X is an isomorphism: in particular, no fundamental
cohomology class of a curve is mapped to zero. Hence aX does not contract any
curve, and it has then to be a finite morphism.
Consider D to be a component of KX . As aX (D) is not a point, it has to
be an elliptic curve mapping to an elliptic curve E of Alb(X). Consider E 0 :=
Alb(X)/E, which is an elliptic curve, and consider the surjective morphism
X −→ E 0 . The Stein factorization of this gives an elliptic fibration on X, and
D is one of the fibers. As D2 = 0, the Zariski Lemma implies that D is a multiple
of a fiber. Following arguments as before, one gets κ(X) = 1, so that this is not
possible. In conclusion KX = 0. By the formula of the canonical divisor under
coverings we conclude that aX is an unramified covering. In conclusion, X is a
torus, and we are done.

The previous Theorem then concludes the proof of the Enriques-Castelnuovo

birational classification of smooth projective surfaces. We can now to present
all the invariants of surfaces of Kodaira dimension 0:

1. Let X be a K3 surface: then KX = 0, Pm (X) = 1 for every m > 0,

q(X) = 0, pg = 1, χ(OX ) = 2,

e(X) = 24, b1 (X) = 0 b2 (X) = 22.

2. Let X be an abelian surface: then KX = 0, Pm (X) = 1 for every m > 0,

q(X) = 2, pg (X) = 1, χ(OX ) = 0,

e(X) = 0, b1 (X) = 4, b2 (X) = 6.

3. Let X be an Enriques surface: then KX 6= 0 but 2KX = 0, Pm (X) = 0 if

m is odd and Pm (X) = 1 if m is even,

q(X) = 0, pg (X) = 0, χ(OX ) = 1,

e(X) = 12, b1 (X) = 0, b2 (X) = 10.

Non-negative Kodaira dimension 145

4. Let X be a bielliptic surface: then KX 6= 0 but torsion of period 4 or 6

(we will see this),

q(X) = 1, pg (X) = 0, χ(OX ) = 0,

e(X) = 0, b1 (X) = 2, b2 (X) = 2.

146 The Enriques-Castelnuovo classification
Chapter 5

Further properties and

classifications

In the previous chapter we completed the proof of the Enriques-Castelnuovo

birational classification of smooth projective surfaces. If X is such a surface,
then it can be ot two types only: either X is minimal, or it is not. The difference
between these two cases is on the existence of (−1)−curves by the Castelnuovo
Contraction Theorem. As every smooth projective surface is birational to a
minimal one, the classification proceeds by classifying only minimal surfaces.
Then, the classification divides in 4 families, by means of the Kodaira dimension,
which can be −∞, 0, 1 or 2.
For surfaces of Kodaira dimension 2, the Enriques-Castelnuovo Theorem
doesn’t say anything: we only know that they exist (see Example 4.3.1), and
that they do not admit elliptic or rational fibrations. For surfaces of Kodaira
dimension 1, the Enriques-Castelnuovo Classification says only that they are all
proper elliptic, and that they cannot have pg = 0 and q(X) = 1.
A more complete classification is given for Kodaira dimension 0: these sur-
faces can only be K3, Enriques, abelian or bielliptic, and we have seen how
to chaacterize them in terms of self-intersection of the canonical bundle and of
birational invariants like q and pg . Finally, for Kodaira dimension −∞ we have
shown that minimal surfaces of this type can only be either P2 or geometrically
ruled. Moreover, we have a complete list of rational geometrically ruled surfaces,
namely the Hirzebruch surfaces Fn .
It is then natural to ask if it is possibile to give a complete birational classi-
fication of smooth projective surfaces, going beyond the Enriques-Castelnuovo
Theorem. The desired result would be a complete list of minimal surfaces. This
is the aim of this chapter, i. e. to study further properties and characteristics of
surfaces aiming to a finer classification. In the following we will then introduce
some important features, examples and properties of ruled surface, rational sur-

147
148 Further properties and classifications

faces, K3 surfacess, Enriques surfaces, abelian surface, bielliptic surfaces, elliptic

surfaces ans surfaces of general type.

5.1 Ruled surfaces

In this section we start by the first family we encounter in the birational classi-
fication by Enriques and Castelnuovo. A ruled surface X over a smooth curve
C has κ(X) = −∞, so that Pm (X) = 0 for every m > 0, q(X) ≥ g(C) and
pg (X) = 0. Moreover, if X is not ruled, i. e. q(X) > 0, if is minimal if and only
if is geometrically ruled, i. e. there is rank 2 vector bundle E on C such that
X ' P(E). If X is ruled, i. e. q(X) = 0, then using properties of rank 2 vector
bundles on P1 we have been able to present a complete list of the possibilities of
X: if X is not geometrically ruled, then it can be only P2 ; if it is geometrically
ruled it can only be one of the Hirzebruch surfaces Fn , for n 6= 1.
In this section we analyze the situation for non-rational minimal ruled sur-
faces, i. e. geometrically ruled surfaces with q > 0. Moreover, we will present
some criterion to distinguish between non-rational ruled surface and bielliptic
surface (which have the same irregularity and geometrical genus).

5.1.1 Classification of non-rational ruled surfaces

If X is a non-rational ruled surface, by Theorem 4.2.8 X is minimal if and only if
it is geometrically ruled, and by Theorem 4.2.2 this is the case if and only if it is
a projective bundle over a genus g > 0 curve. Hence, a birational classification
of non-rational ruled surfaces is equivalent to a classification of rank 2 vector
bundles on curves up to a tensor with a line bundle.
So, let X be a ruled surface birational tu C × P1 , where C is a smooth curve
of genus g > 0. Let E be a rank 2 vector bundle on C, and L ∈ P ic(C). Notice
that
deg(E × L) = deg(E) + 2deg(L),

so that we can always suppose E to be of degree 0 (or af any desired degree).

Lemma 5.1.1. Let C be a smooth curve of genus g, and suppose E to be a rank

2 vector bundle on C.

1. There are L, M ∈ P ic(X) such that there is an exact sequence

0 −→ L −→ E −→ M −→ 0,

i. e. E defines an element in Ext1 (M, L).

2. If h0 (C, E) > 0, then we can choose L ' OC (D), where D is a divisor on

C which can be either trivial or effective.
Ruled surfaces 149

3. If h0 (C, E) > 1, then we can choose D to be effective.

Proof. Up to changing E with E × H for some ample line bundle, we can always
suppose h0 (C, E) > 0 and deg(E) > 0.
There is then s ∈ H 0 (C, E) which is not zero, so it defines a morphism
s∗ : E ∗ −→ OC . The image is clearly an ideal of OC , i. e. it is of the form
OC (−D) where D is a the divisor D = Z(s). We have then a surjective map
s∗ : E ∗ −→ OC (−D), and the kernel of s∗ is clearly a line bundle. Taking duals,
this implies item 1 and 2.
For item 3, we just need to show that we can choose s to to vanish at some
points. As h0 (C, E) > 1 there are at least two linearily independent sections
s, t ∈ H 0 (C, E). As we supposed deg(E) > 0, we have deg(det(E)) > 0, so that
the section s ∧ t ∈ H 0 (C, det(E)) has to vanish at some points. Let p ∈ Z(s ∧ t).
Hence there are µ, λ ∈ C \ {0} such that µs(p) + λt(p) = 0. In conclusion µs + λt
vanishes at p, so that there is global section of E which vanishes at some points,
and we are done.

Corollary 5.1.2. (Riemann-Roch’s Theorem for rank 2 vector bun-

dles). Let C be a smooth curve of genus g and E a vector bundle of rank 2 on
C. Then
χ(E) = deg(E) − 2g + 2.

Proof. By Lemma 5.1.1 there are two line bundles L, M ∈ P ic(C) such that
χ(E) = χ(L) + χ(M ). By the Riemann-Roch Theorem for curves we then get

χ(E) = deg(L) + deg(M ) − 2g + 2.

But notice that

deg(E) = deg(det(E)) = deg(L × M ) = deg(L) + deg(M ),

and we are done.

As every rank two vector bundle E on C is the extension of a line bundle

M by another line bundle L on C, it seems natural to ask when the sequence
splits, i. e. E ' L ⊕ M . The sequence can be rewritten as

0 −→ L ⊗ M −1 −→ E ⊗ M −1 −→ OC −→ 0,

so that E ' L ⊕ M if and only if H 1 (C, L ⊗ M −1 ) = 0. Here is one of the two

main results of this section:

Theorem 5.1.3. Let C be an elliptic curve. A rank 2 vector bundle E on C

can be one of the following:

1. there are two line bundles L, M ∈ P ic(C) such that E ' L ⊕ M ;

150 Further properties and classifications

2. there is a line bundle L ∈ P ic(C) such that E is isomorphic to the non-

trivial extension of L by L.

3. there is a line bundle L ∈ P ic(C) and a point p ∈ C such that E is

isomorphic to the non-trivial extension of L ⊗ OC (p) by L.

Proof. We can assume E to have degree d = 1, 2, so that h0 (C, E) > 0 by

Corollary 5.1.2. By Lemma 5.1.1 there are then a line bundle Lk and a line
bundle Md−k of degrees k ≥ 0 and d − k respectively, such that there is an exact
sequence
0 −→ Lk −→ E −→ Md−k −→ 0.
If d = 2, we can even suppose k > 0 by Lemma 5.1.1. If k = 0, we can
−1
suppose L0 ' OC . Then E defines an element in H 1 (C, Lk ⊗ Md−k ), where
−1
deg(Lk ⊗ Md−k ) = 2k − d. If 2k > d, then, this is 0, so that E is split. We then
need to analyze the case 2k ≤ d. As d = 1, 2, this is 2k ≤ 1 or 2k ≤ 2, and as
k ≥ 0 the first is the case k = 0, the second is k = 1 (as if d = 2 we assume
k > 0).
In the first case we have then deg(E) = 1, and E is an extension of a line bun-
dle of degree 1 by OC . Notice that in this case h1 (C, OC (p) = h0 (C, OC (−p)) =
0 by Serre’s duality, as C is elliptic, so that

h1 (C, OC (−p)) = h0 (C, OC (p)) = deg(OC (p)) = 1,

by Riemann-Roch. Hence E is a non-trivial extension, and we get item 3.

In the second case we have deg(E) = 2, and E is an extension of a line bundle
of degree 1 by a line bundle of degree 1. Up to torsion, we have L1 ⊗M1−1 ' OC ,
so that L1 ' M1 . Again, notice that

h1 (C, L1 ⊗ M1−1 ) = h1 (C, OC ) = g(C) = 1,

as C is elliptic, so that E is a non-trivial extension. We finally get item 2, and

we are done.

This gives a complete classification of geometrically ruled surfaces of with

q = 1. They are all of the form P(OC ⊕ L) for every L ∈ P ic(X), or of the
form P(E) where E is the non-trivial extension of OC via OC , or the non-trivial
extension of OC (p) via OC (for any point p ∈ C). The first case is the same
one we have for rational geometrically ruled surfaces (as P ic(P1 ) ' Z, it is even
more precise).
It then remains to study the last case, i. e. whene g > 1. Here the result is
much less precise, and we have the following:

Theorem 5.1.4. Let C be a smooth curve of genus g ≥ 2. For every line bundle
L ∈ P ic(C) there is a quasi-projective variety SL of dimension dim(SL ) ≥ 2g−3
parameterizing indecomposable rank 2 vector bundles E such that det(E) ' L.
Ruled surfaces 151

Proof. Let L ∈ P ic(C) be a line bundle of degree g − 1 > 0, and suppose that
h0 (C, L) = 0. Then h0 (C, L−1 ) = 0, so that

h1 (C, L−1 ) = 2g − 2 ≥ 2.

Consider an affine hyperplane SL in H 1 (C, L−1 ) which does not pass through
the origin. Notice that SL is a quasi-projective variety of dimension 2g − 3.
Moreover, every point s ∈ SL corresponds to an exact sequence

0 −→ OC −→ Es −→ L −→ 0.

Notice that h0 (C, Es ) = 0, so that s determines a unique extension which is

indecomposable, and deg(Es ) = deg(L).

5.1.2 Ruled and bielliptic surfaces

According to the Enriques-Castelnuovo Classification, there are two types of
surfaces having pg = 0 and q > 0: namely, non-rational ruled surfaces and
bielliptic surfaces. Here we want to give criteria to characterize those surfaces
in the first or in the second class. The first result is the following:

Proposition 5.1.5. Let X be a smooth projective surface.

2 2
1. If pg (X) = 0 and q(X) ≥ 1, then KX ≤ 0, and KX = 0 if and only if
q(X) = 1 and b2 (X) = 2.
2
2. If KX < 0 and X is minimal, then pg (X) = 0 and q(X) ≥ 1.

Proof. For the first item, as that pg (X) = 0 Noether’s Formula gives

2
12 − 12q(X) = KX + 2 − 4q(X) + b2 (X),

2 2
so that KX = 10 − 8q(X) − b2 (X). As q(X) ≥ 1, we have that KX ≤ 0 if
and only if b2 (X) ≥ 2 when q(X) = 1. To prove this consider the Albanese
map aX : X −→ Alb(X). As q(X) = 1, the Albanese torus is an elliptic curve,
and the generic fiber F is connected. Since F 2 = 0 and H · F > 0 for every
hyperplane section H, we see that F and H are linearily independent, so that
b2 (X) ≥ 2.
2
For the second item, as X is minimal and KX < 0, then KX is not nef.
Hence X is either geometrically ruled or P , so that pg (X) = 0. If X is P2 ,
2

2
then KX = 9, so that X has to be geometrically ruled. If q(X) = 0, then X is
2
rational. But in this case KX = 8 by Proposition 4.2.3, so that q(X) > 0,

Notice then that one can distinguish between non-rational ruled surfaces and
bielliptic surfaces by means of the self-intersection of the canonical bundle: X is
2
bielliptic if and only if KX = 0. Moreover, we have shown that the only minimal
152 Further properties and classifications

surfaces whose canonical bundle has negative self-intersection are non-rational

geometrically ruled surfaces. To conclude this section, let me resume all the
invariants and numbers of a non-rational minimal ruled surface X over a curve
C of genus g > 0: we have
2
KX = 8 − 8g, Pm (X) = 0, κ(X) = −∞,

q(X) = g, pg (X) = 0, χ(OX ) = 1 − g,

e(X) = 4 − 4g, b1 (X) = 2g, b2 (X) = 2.
Moreover, as b2 (X) = 2pg (X)+h1,1 (X), we have h1,1 (X) = 2, so that ρ(X) = 2.

5.2 Rational surfaces

In this section we want to present several examples of ruled surfaces. Recall
that a ruled surface is minimal if and only if it is P2 or Fn where n = 0 or
n > 1. Hence we hava complete classification of ruled surfaces. Let me resume
the invariants of a rational surface X: if X ' P2 then
2
KX = 9, Pm (X) = 0, κ(X) = −∞,

q(X) = 0, pg (X) = 0, χ(OX ) = 1,

e(X) = 3, b1 (X) = 0, b2 (X) = 1.
Moreover, as b2 (X) = 2pg (X)+h1,1 (X), we have h1,1 (X) = 1, so that ρ(X) = 1.
If X is Fn for n 6= 1, we have
2
KX = 8, Pm (X) = 0, κ(X) = −∞,

q(X) = 0, pg (X) = 0, χ(OX ) = 1,

e(X) = 4, b1 (X) = 0, b2 (X) = 2.
Moreover, h1,1 (X) = 2, so that ρ(X) = 2. Notice that every rational surface X
is birational to P2 , so that every rational map P2 99K PN gives a rational map
X 99K PN . Then we just need to study rational maps from P2 to some projective
space. The advantage is that such rational maps correspond to linear systems
associated to some line bundle O(n) on P2 , which have no fixed components.
We need the following steps:
First of all, we need to determine N , which is N = n+2


n . Then we need
to determine if the map associated to O(n) is everywhere defined, i. e. if there
are base points. In general the linear system |O(n)| has base points, so that to
have a well-defined map we need to blow up P2 in the base locus of |O(n)|. This
base locus is given by a finite set of points {p1 , ..., pr }, so we have to consider
the blow-up Pr := Blp1 ,...,pr P2 , and we get a map
n+2
fr,n : Pr −→ P( n ) .
Rational surfaces 153

It corresponds to a linear system V on Pr : if H = f ∗ O(1), mi is the minimum

of the multiplicities of the members of |O(n)| at the point pi , and we let Ei be
the exceptional divisor obtained blowing-up pi , then we have
r
X
V ⊆ |nH − mi Ei |.
i=1

we are particularily interested in the case where fr,n is an embedding. If this is

n+2
the case, fr,n (Pr ) ⊆ P( n ) is a rational surface in a projective space. The aim
of this section is to study the geometry of fr,n (Pr ).
Notice that the Picard group of fr,n (Pr ) has an orthogonal basis given by
H, E1 , ..., Er , where H 2 = 1 and Ei2 = −1. Moreover, every hyperplane section
Pr
L is linearily equivalent to nH + i=1 mi Ei , so that
r
X
L2 = n2 − mi .
i=1

This is the degree of fr,n (Pr ) in PN . Other important points are the number of
lines on fr,n (Pr ), i. e. of curves on it having intersection 1 with a hyperplane
section, an possible equations of fr,n (Pr ).
There is again another point which is interesting for us: if H ∈ |O(n)| is a
hyperplane section, it corresponds to a point h ∈ |O(n)|∗ . The linear system
|H| then defines a rational map

P2 99K |H|∗ ,

which is easily seen to be the composition of ϕn with the projection of |O(n)|

onto |H| away from h. If x ∈ Pr , then the rational surface obtained from the
system of curves of |O(n)| passing through x is exactly the projection of fr,n (Pr )
away from fr,n (x). Then it is natural to consider projections of fr,n (Pr ) into
projective spaces of dimension smaller than n+2


n . We will then need two results
on this kind of projections:

Lemma 5.2.1. Let X be a surface in PN , and p ∈ PN \ X (resp. p ∈ X). Let

fp : X −→ PN −1 be the projection from p (resp. fp : Blp (X) −→ PN −1 is the
projection from p). Then fp is an embedding if and only if there is no line L
of PN passing through p such that L ∩ X consists of at least 2 (resp. 3) points
counted with multiplicities.

Proof. The proof is immediate.

As the set of bisecants (resp. tangents) to X is parameterized by the com-

plement of the diagonal in X × X (resp. by P(TX )), it follows that the union of
bisecants (resp. tangents) to X lies in a subvariety of PN of dimension d, where
d ≤ 5 (resp. d ≤ 4).
154 Further properties and classifications

Lemma 5.2.2. Every surface is isomorphic, via generic projection, to a smooth

surface in P5 .

The cases we want to study are more precisely when n = 2 and n = 3.

5.2.1 Linear systems of conics

Consider the linear system |O(2)|, i. e. the complete linear system of conics.
The map defined by it is

j : P2 −→ P5 , j(x0 : x1 : x2 ) := (x20 : x0 x1 : x0 x2 : x21 : x1 x2 : x22 ).

It is the easy to see that j is an embedding, which is called the Veronese embed-
ding. The surface V := f0,2 (P2 ) is called the Veronese surface, and its degree is
(2H)2 = 4, where H is an hyperplane section of P5 .

Proposition 5.2.3. The Veronese surface does not contain any line, but con-
tains a 2−dimensional linear system of conics.

Proof. If L is a line on V , then we would have

1 = L · H = O(1) · O(2) = 2,

which is clearly not possible. If L is a line in P2 , then j(L) is a curve in P5 , and

j(L) · H = 2, so that j(L) is a conic. But notice that h0 (P2 , O(1)) = 2, so V
contains a 2−dimensional linear system of conics.

Corollary 5.2.4. Let p be a generic point of P5 , and consider fp : V −→ P4

to be the projection from V . Then fp gives an isomorphism between V and
fp (V ) ⊆ P4 .

Proof. Let L be a line in P2 , and consider the plane P (L) defined by the conic
j(L). Moreover, let X := ∪L∈|O(1)| P (L). Then we have dim(X) ≤ 4. Consider
now
Z := {(L, x) ∈ |O(1)| × P5 | x ∈ P (L)},

which is a P2 −bundle over |O(1)|. Notice that Xs the projection to P5 of Z. Let

now x, y ∈ V , where x 6= y. The line L(x, y) defined by x and y is contained in
the plane P (L), where L is the line of P2 defined by j −1 (x) and j −1 (y). Hence
every bisecant of V lieas in X, and the result follows.

The generic projection of fp (V ) to P3 is called Steiner surface, which is a

singular surface: it has 3 double lines meeting in a point, which is triple.
Let us now look at the projection of V from a point p ∈ V . As V contains
no lines, there is a well defined morphism

fp : V \ {p} −→ S,
Rational surfaces 155

where S is a cubic surface in P4 . The map fp is 1 to 1 where it is defined, so

that to define it every where we need to blow-up the point p. As a consequence
we get an embedding
j : F1 −→ P4
corresponding to the linear system |H + F |, where H and F are the two gener-
ators of the Picard group of F1 . Using this, we can determine lines on S.

Proposition 5.2.5. The cubic surface S containes two types of lines: a linear
system {Lt }t∈P1 such that Lt ∩ Lt0 = ∅ for every t 6= t0 , and a line L such that
L · Lt = 1 for every t ∈ P1 .

Proof. As F · (H + F ) = 1, we have that j(F ) is a line on S, so that on S we

have a family of lines {Lt }t∈P1 , with the property that Lt ∩ Lt0 = ∅ for t 6= t0 .
Now, let E be the exceptional curve on F1 . Then E = H − F , so that

E · (H + F ) = (H − F ) · (H + F ) = H 2 = 1

by Proposition 4.2.3, so that j(E) is a line on S which meets every line Lt :

indeed
j(E) · Lt = E · F = H · F = 1.
If C is a curve on F1 , then C = aH + bF with a, b ≥ 0, so that

C · (H + F ) = (aH + bF ) · (H + F ) = 2a + b,

which is 1 if and only if a = 0 and b = 1, i. e. C = F . In conclusion, these are

the only lines on S.

To conclude, we study projections of S from points of P4 . We begin with

the following

Lemma 5.2.6. The cubic S is contained in a 2−dimensional linear system of

quadrics in P4 , of which it is the intersection. For every pencil {λ1 Q1 + λ2 Q2 }
of quadrics in P4 we have Q1 ∩ Q2 = S ∪ P , where P is a plane and S ∩ P is a
conic. Conversely, for every conic on S lying in a plane P , we have that S ∪ P
is the intersection of two quadrics.

Proof. A quadric on P4 cuts on S the strict transform of a quartic on P2 passing

through 0 with multiplicity 2. Now, the linear systems |OP4 (2)| and |OP2 (4)| have
the same dimension, and as passing through a point with multiplicity 2 gives
three conditions on a system of plane curves, there must be at least 3 linearily
independent quadrics of P4 containing S. Any two of them is irreducible, and
their intersection is a degree 4 surface containing S. As S has degree 3, we
need that the intersection of the two quadrics is S ∪ P , where P is a plane. If
P has equations L = M = 0, then two quadrics containing S have equations
LAi +M Bi = 0, where i = 1, 2. The determinant A1 B2 −A2 B1 vanishes at every
156 Further properties and classifications

point of S \ (S ∩ P ), hence S lies in the quadric of equation A1 B2 − A2 B1 = 0.

Then S is the intersection of these three quadrics, so that the linear system of
quadrics containing S has dimension 2. The intersectionnS ∩ P is easily seen
to be a conic. If conversely C is a conic in P ∩ S, then any quadric contains P
if and only if if contains a point of P \ C, so that there is a pencil of quadrics
containing S ∩P , and the intersection of the members of this pencil is S ∩P .

Corollary 5.2.7. The projection of S from a point p ∈ P4 \ S is a cubic surface

in P3 whose singularities are a double line. The projection of S from a point
p ∈ S is a quadric in P3 , which is smooth if and only if p ∈
/ L.

Proof. By Lemma 5.2.6 there are two quadrics Q1 and Q2 containing S and p.
Then Q1 ∩ Q2 = S ∪ P , where P is a plane passing through p. Every bisecant of
S through p cuts Qi in three points, thus it lies in Q1 ∩ Q2 , hence in P . Hence
fp is an isomorphism outside C := S ∩ P , which is a conic. The restriction to
C is 2 to 1, so that fp (S) has a double line of singularities.
If p ∈ S, then every line through p cuts S in a single point if p ∈
/ L. In this
case, the line L is projected to a singular point.

5.2.2 Linear systems of cubics

Consider p1 , ..., pr ∈ P2 be r distinct points in general position (i. e. none of
then are collinear, nor 6 of them lie on a quadric), where r ≤ 6, and take V
to be the linear system of cubics passing through p1 , ..., pr . Then V defines a
rational map
P2 99K Pd ,

and it is easy to calculate d: indeed h0 (P2 , O(3)) = 10, so that |O(3)| has dimen-
sion 9. As we are taking cubics passing through p1 , ..., pr in general position, we
have r independent conditions, so that d = 9 − r. Consider now Pr , the blow
up of P2 along p1 , ..., pr , and consider the morphism

j : Pr −→ Pd .
Pr
Notice that j is the morphism associated to |3H − i=1 Ei |, where H is a line
on P2 and Ei is the exceptional divisor over pi .

Definition 5.2.1. The image Sd := j(Pr ) is called Del Pezzo surface of degree
d.

Proposition 5.2.8. The map j is an embedding, and the Del Pezzo surface Sd
is a surface of degree d in Pd .

Proof. In order to show that j is an embedding, we need to show that the linear
system V separates points and tangent directions. In order to do this, we just
need to look at the case r = 6, the others follow. Consider i < j ∈ {1, ..., 6},
Rational surfaces 157

and let x ∈ P6 . Moreover, let π : P6 −→ P2 be the blow-up. Suppose that

x∈ / L(pi , pj ), where L(pi , pj ) is the line in P2 passing through pi and pj .
As the points are in general position, there is only one conic Qxij passing
through π(x) and pk for every k 6= i, j: indeed if x ∈ / Ei ∪ Ej (where Ei =
−1 x
π (pi )), then π(x) 6= pi , pj , hence the conic Qij is defined by 5 independent
conditions, so it is unique. If x ∈ Ei , then there is a 1−dimensional pencil of
conics passing through pk with k 6= i, j, and to pass through x means that this
conic has to pass through pi with tangent direction given by x. Hence we have
again 5 independent conditions, and there is only one conic Qxij .
Similarily, there is only one conic Qi through the points pj with j 6= i. If Q ei
is its proper transform, we have that Q ei ∩ Qe j = ∅ for i 6= j.
Now, consider x, y ∈ P6 , with x 6= y, and let i be such that pi 6= π(x), π(y)
and x ∈ /Q e i . Then Q ex ∩ Q e x = {x} for every j 6= k, with j, k 6= i and pk 6= π(x).
ij ik
x
Hence y ∈ Q e for at most one value of j 6= i. Hence there is at most one j 6= i
ij
such that the cubic Q e x ∪ L(p
e i , pj ) passes through x but not through j. Hence
ij
the morphism j separates points.
Let x ∈ P2 \ {p1 , ..., p6 }, and consider the cubics Qi ∪ L(pi , x) for i = 1, ..., 6.
They have different tangents at x, so that j is an immersion on P6 \ ∪6i=1 Ei .
Let now x ∈ E1 : the conics Qx23 and Qx24 intersect at x with multiplicity 2. The
cubics Q e x ∪ L(p
e 2 , p3 ) and Q e x ∪ L(p2 , p4 ) then have different tangent directions
23 34
at x, so that j is an embedding on E1 . Similar calculations show that the same
is true for i = 2, ..., 6, so that j is an embedding.
To calculate the degree of Sd , simply notice that
 r
X 2
deg(Sd ) = 3H − Ei = 9 − r = d,
i=1

and we are done.

The next topic is to calculate how many lines we have on Sd .

Theorem 5.2.9. On the Del Pezzo surface of degree d there are finitely many
lines, which are the images under j : Pr −→ Sd of one of the following curves
on Pr :

1. the exceptional curves Ei , for i = 1, ..., r;

2. the curves L(p

e i , pj ) for i, j = 1, ..., r, i < j;

3. the conics Q
e i , for i = 1, ..., r.

Proof. By the formula of the canonical bundle of the blow up, we get that
r
X
KPr = π ∗ (−3H) + Ei .
i=1
158 Further properties and classifications

Pr
As j ∗ O(1) = OPr (3H − i=1 Ei ), we get that the very ample divisor defining j
is −KPr . Now, if Ei is an exceptional divisor, we have
r
X
−Ei · KX = 3Ei · π ∗ H − Ei · Ej = 1,
j=1

so that j(Ei ) is a line on Sd .

Now, let L be a line on Sd which is not of the form j(Ei ) for any i = 1, ..., r.
Then
Xr
1 = −L · KX = 3L · p∗ H − L · Ei .
j=1
∗
We have L · Ei ≥ 0 and L · p H ≥ 0 as L is effective. Notice that L =
Pr
mπ ∗ H + i=1 mi Ei for some m, m1 , ..., mr ∈ N0 , so that
r
X
1 = 3m − mi .
i=1

As L is not one of the Ei , we need mi = 0, 1, and we need m > 0. Hence we

have the following possible solutions: if m = 1, then we need
r
X
mi = 2,
i=1

so that the only possible case is r = 2 and m1 = m2 = 1. Hence

L = π ∗ H + Ei + Ej

for some i 6= j, so that L is the proper transform of a line passing through pi

and pj . But we have other possible cases: if m = 2, then we have
r
X
mi = 5,
i=1

so that there are only the following possibility, namely r = 5 and mi = 1 for
i = 1, ..., 5, i. e. L is the proper transform of a cubic passing through 5 of the
pi ’s. If m ≥ 3, we have
r
X
6≥r mi = 3m − 1 ≥ 8,
i=1

which is not possible, hence we are done.

We can then give a complete description of lines on Sd for every d = 3, ..., 8.

1. If d = 3, i. e. r = 6, then we have 6 exceptional divisors, 15 lines and 6

conics, so that on S3 we have 27 lines.

2. If d = 4, i. e. r = 5, then we have 5 exceptional divisors, 10 lines and 1

conic, so that on S4 we have 16 lines.
Rational surfaces 159

3. If d = 5, i. e. r = 4, then we have 4 exceptional divisors, 6 lines and no

conics, so that on S5 we have 10 lines.

4. If d = 6, i. e. r = 3, then we have 3 exceptional divisors, 3 lines and no

conics, so that on S6 we have 6 lines.

5. If d = 7, i. e. r = 2, then we have 2 exceptional divisors, 1 line and no

conics, so that on S7 we have 3 lines.

6. If d = 8, i. e. r = 1, then we have 1 exceptional divisor, no lines and no

conics, so that on S8 we have 1 line.

To conclude this section, we want to prove some results about cubics an

other surfaces. Let me start with two lemmas:

Lemma 5.2.10. Let S ⊆ P3 be a smooth cubic surface. Then S contains a line.

Proof. Let V := |O(3)| be the linear system of cubics in P3 , and let G be the
Grassmannian of lines in P3 . Consider

Z := {(L, C) ∈ G × V | L ⊆ G},

which has two projections p : Z −→ G and q : Z −→ V . A cubic in P3

contains the line x2 = x3 = 0 if and only if the coefficients of x30 , x20 x1 , x0 x21
and x31 vanish, so that the fiber of p over x2 = x3 = 0 has dimension equal to
dim(V ) − 4. As dim(G) = 4, then dim(Z) = dim(V ). If there is a cubic S such
that no line is contained in S, then q(Z) ⊆ V has codimension at least 1, and the
fiber q −1 (S) is either empty or positive dimensional. As S3 is a cubic containing
finitely many lines, we see that this is not possible so that q is surjective, and
we are done.

Lemma 5.2.11. Let S ⊆ P3 be a smooth cubic surface, and let L be a line on S.

Then there are exactly 10 distinct lines which are distinct from L and meeting
L. Moreover, these 10 lines fall into 5 disjoint pairs of concurrent lines. In
particular, S contains two disjoint lines.

Proof. Consider the pencil {Pt }t∈P1 of planes through L. Then S ∩ Pt = L ∪ Ct ,

where Ct is a conic. Notice that Ct does not contain L, and cannot be a double
line: indeed, if the plane x = 0 cuts S in the line y = 0 counted twice, then
S has equation xQ + yZ 2 = 0, where Q is a conic and Z is a line. But then
S would have singularities, which is not possibile. Hence Ct has to be either
smooth or singular, and in this second case it is the union of two cuncurrent
lines distinct from L but meeting it.
Conversely, if L0 is a line meeting L and distinct from L, they define a plane
P which meets S along L. Hence P ∩ S = L ∪ L0 ∪ M , and L0 and M are two
cuncurrent lines meeting L. In conclusion, all the lines meeting L are of this
160 Further properties and classifications

form. Let us now suppose that L is given by equation x2 = x3 = 0 in P3 . The

equation of S is then

ax20 + 2bx0 x1 + cx21 + 2dx0 + 2ex1 + f = 0,

where a, b, c, d, e, f are homogenous polynomials in x2 and x3 . The equation

of Ct is obtained setting x2 = tx3 , so that Ct is singular if and only if its
determinant ∆(x2 , x3 ) vanishes. This is a homogenous quintic in x2 , x3 which
has no multiple root: if x2 = 0 is a root, let x ∈ C0 be the singular point. If
x∈ / L we can suppose C0 to be described by equation x0 x1 = 0. Then every
coefficient of ∆(x2 , x3 ) is divisible by x2 , except b, so that x is smooth in S and
f is not divisible by x22 . In conclusion ∆(x2 , x3 ) is not divisible by x22 . If x ∈ L,
we can write C0 as x20 − x21 , and repeating the argument as before we get that
∆(x2 , x3 ) is not divisible by multiple roots.
In conclusion, {Ct }t∈P1 contains 5 singular conics, givins the 10 lines of the
statement. If three of these lines meet at a point x ∈ S, they are coplanar (they
lie in the tangent plane of S at p). If C0 and C1 are singular conics in the
pencil, suppose Ci = Di ∪ Di0 . Since D0 , L and D1 are not coplanar, we have
D0 ∩ D1 = ∅, and we are done.

These two lemmas are used to show the following important result:

Theorem 5.2.12. We have the two following properties:

1. A smooth surface S ⊆ P3 is a cubic if and only if it is S3 .

2. A smooth surface S ⊆ P4 is a complete intersection of two quadrics if and

only if it is S4 .

In particular, every smooth cubic surface in P3 contains 27 lines.

Proof. We only prove the first item. The Del Pezzo surface S3 is a cubic by
Propositon 5.2.8, so that we need to show the converse. Let S be a smooth
cubic surface in P3 . By Lemmas 5.2.10 and 5.2.11 there are two disjoint lines
L, L0 on S, and consider the rational map

φ : L × L0 99K S

defined as follows: if (p, p0 ) is generic in L × L0 , the line through p and p0 meets

S at a third point p00 , so that φ(p, p0 ) := p00 . Consider the rational map

ψ : S 99K L × L0

defined as follows: if x ∈ S \ (L ∪ L0 ), let Px,L be the plane passing through L

and x, and let p := L ∩ Px,L0 and p0 := L0 ∩ Px,L . Then put ψ(x) := (p, p0 ).
It is easy to see that φ and ψ are inverse to each other, and we can define ψ
K3 surfaces 161

on the whole S: if x ∈ L, simply define p := L ∩ Tx (S) and p0 = L0 ∩ Px,L ,

and similarily on for L0 . Then we have a birational map which contracts curves
meeting L and L0 .
These lines come in pairs Di , Di0 for i = 1, ..., 5 defining a plane Pi by Lemma
5.2.11. The plane Pi meets L0 in a point, lying either on Di or on Di0 , so that
one line in each pair meets L0 . Hence the morphism ψ contracts 5 disjoint
lines. Hence S is isomorphic to P1 × P1 with 5 points blown-up. As P1 × P1 is
isomorphic to P2 blown-up in a point, so that S is isomorphic to P2 blow-up in
6 points. But S is embeddend in P3 via the anticanonical embedding, so that
S ' S3 .

5.3 K3 surfaces
Starting from this section, we collect properties and informations about the
geometry of smooth surfaces of non-negative Kodaira dimension. A first class
one meets in the Enriques-Castelnuovo Classification is the one of K3 surfaces.
This is a very interesting class of surfaces, whose geometry is very rich. It has
been investigated since 1950’s, and we are going to collect here some of the main
results. First of all, let me recall all the invariants. If X is a K3 surface, then
KX = 0, so that

2
KX = 0, Pm (X) = 1, κ(X) = 0,

q(X) = 0, pg (X) = 1, χ(OX ) = 2,

e(X) = 24, b1 (X) = 0, b2 (X) = 22.

Notice that the first Chern class c1 : P ic(X) −→ H 2 (X, Z) is injective, so that
0 < ρ(X) ≤ h1,1 (X), and as pg (X) = 1 we have h1,1 (X) = 20. In conclusion we
have different types of K3 surfaces according to ρ(X), which can be any integer
between 1 and 20.

5.3.1 The Kummer surface

We have already seen several examples of K3 surfaces: in Example 4.3.5 we have
shown that the generic complete intersection X of hypersurfaces H1 , ..., Hr of
Pr
degrees d1 , ..., dr in Pr+2 is a K3 surface if and only if i=1 di = r + 3. In this
first section I introduce a new example. Let A be an abelian surface, on which
we have a natural involution

i : A −→ A, i(x) = −x,

(recall that A = C2 /Γ for some maximal rank lattice Γ, so that A inherits a

group structure). The group A is isomorphic to (R/Z)4 , so the involution i has
162 Further Properties and classifications

16 fixed points, which are p1 , ..., p16 . Consider

e −→ A
π:X

to be the blow-up of A along p1 , ..., p16 , on which we have 16 exceptional

curves E1 , ..., E16 . The involution i on A extends to an involution ei on A.
e
Let Kum(A) := A/i and p : A −→ Kum(A) be the quotient map.
e e e

Definition 5.3.1. The surface Kum(A) is called Kummer surface associated

to A.

Proposition 5.3.1. Let A be an abelian surface. Then Kum(A) is a K3 sur-

face.

Proof. First of all, we need to show that Kum(A) is smooth. The involution i
e 16 Ei .
has no fixed points on A\{p1 , ..., p16 }, so that ei has no fixed points on A\∪ i=1
Hence p is étale outside ∪16 16
i=1 Ei , and Kum(A) \ p(∪i=1 Ei ) is smooth. Let then
q ∈ Ei for some i. Writing A = C2 /Γ we have local coordinates x, y around pi ,
with the property that i∗ x = −x and i∗ y = −y. Let x0 := π ∗ x and y 0 := π ∗ y. On
Ae we can then consider local coordinates around q which are x0 and t = y 0 /x0 .
The involution ei then acts as follows: ei∗ x0 = −x0 and ei∗ t = t. Hence t and
u := (x0 )2 form local coordinates on Kum(A) around p(q), so that Kum(A) is
smooth.
In order to show that X is a K3 surface, we need KKum(A) = 0 and
q(Kum(A)) = 0. Let us start with the canonical divisor. The abelian sur-
face A has a holomorphic 2−form which is nowhere zero. Writing A = C2 /Γ,
this is symple the form ω on A induced by dx ∧ dy. Notice that i∗ ω = ω, and
the two form π ∗ ω is invariant under the action of ei: hence π ∗ ω descends to a
holomorphic 2−form ω e on Kum(A). To show that KKum(A) is trivial, we sim-
ply need to show that ω e is nowhere zero on Kum(A). Let D be the zero divisor
e . As Z(ω) = 0, we have that Z(π ∗ ω) is concentrated on the exceptional
of ω
curves. Let Ci := p(Ei ), giving 16 curves on Kum(X). As p∗ ω e = π ∗ ω, we have
that D is concentrated on the Ci ’s. Let now q ∈ Ei . Then, in local coordinates
as before, we have
1
π ∗ ω = dx0 ∧ dy 0 = dx0 ∧ d(tx0 ) = x0 dx0 ∧ dt = du ∧ dt.
2
Then the form ω e is not zero at q, so that D = 0. In conclusion KKum(A) = 0.
Now, let us suppose we have a holomorphic 1−form on Kum(A), so that
q(Kum(A)) > 0. This defines a holomorphic 1−form on A e which is invariant
under ei. As the map

π ∗ : H 0 (A, ΩA ) −→ H 0 (A,
e Ω e)
A

is an isomorphism, we get that there exist a holomorphic 1−form on A which is

invariant under i. But H 0 (A, ΩA ) = C2 spanned by dx and dy, and these forms
K3 surfaces 163

are such that i∗ dx = −dx and i∗ dy = −dy. Hence for every holomorphic 1−form
α ∈ H 0 (A, ΩA ) we have i∗ α = −α, so that the only possible holomorphic
1−form invariant under i is α = 0. But this implies that q(Kum(A)) = 0, so
that Kum(A) is a K3 surface.

The Kummer surfaces are special among the class of K3 surfaces: the curves
Ci in the proof are independent, and Ci2 = −2. Hence ρ(Kum(A)) ≥ 16.
Moreover, the hyperplane section H of A induces an hyperplane section on
Kum(A), so that ρ(X) ≥ 17. We have even the following easy property: first
of all, we define a (−2)−curve on a surface X to be a curve C on X which is
smooth and rational, and C 2 = −2. Now, if X is a K3 surface and C is a curve
on X, then by the genus formula we have that C 2 is even and that C 2 ≥ −2, and
it is −2 if and only if g(C) = 0. Then every smooth curve C on a K3 surface
of negative self-intersection is a (−2)−curve. Using this, there is a beautiful
characterization of Kummer surfaces among K3 surfaces:

Proposition 5.3.2. Let X be a K3 surface containing 16 disjoint irreducible

P16
curves C1 , ..., C16 such that Ci is a (−2)−curve and the divisor i=1 Ci is
divisible by 2 in P ic(X). Then X is a Kummer surface.
P16
Proof. As i=1 Ci is divisible by 2 in P ic(X), there is a double cover p : Z −→
P16
X whose branching locus is exactly i=1 Ci . The curve Di := p−1 (Ci ) are 16
disjoint curves such that Di2 = Ci2 /2 = −1, and g(Di ) = g(Ci )/2 = 0, so they
are all (−1)−curves. Contracting them we get a map π : Z −→ Y , where Y is
a smooth surface.
P16 P16
We have that KZ = π ∗ KY + i=1 Di , and p∗ KX = KZ − i=1 Di , so
that π ∗ KY = p∗ KX = 0. In conclusion we need KY = 0. Moreover, we
have e(Y ) = e(Z) − 16, and e(Z) = 2(e(X) − 16). As X is a K3 surface, we
have e(X) = 24, so that e(Z) = 16 and e(Y ) = 0. Then q(Y ) = pg (Y ) + 1,
and pg (Y ) = 1 as KY = 0, so that q(Y ) = 2. By the Enriques-Castelnuovo
Classification, we have then that Y is an abelian surface.
To show that X = Kum(Y ) we still need to look at the involution on Y .
The involution on Z given by exchanging sheets descends to an involution i on
Y which has excatly 16 fixed points, namely the images of the Di ’s. Now, the
i-invariant part of H 1 (Y, Q) is isomorphic to H 1 (X, Q) = 0, so that i acts as
−id on H 1 (Y, Q), and X = Kum(Y ).

5.3.2 Projectivity
Let us now study maps from a K3 surfaces to some projective spaces. The main
result we need is the following:

Theorem 5.3.3. Let X be a K3 surface, and let C ⊆ X be a smooth curve of

genus g.
164 Further Properties and classifications

1. We have C 2 = 2g − 2 and h0 (X, OX (C)) = g + 1.

2. If g ≥ 1, then |C| has no base points, and it defines a morphism

φg : X −→ Pg .

The map φg|C : C −→ Pg−1 is the map associated to the canonical linear
system of C.

3. If g = 1, then φg is an elliptic fibration over P1 whose generic fiber is C.

4. If g = 2, then φg : X −→ P2 is a double cover of P2 ramified along a

sextic.

5. If g ≥ 3, then φg is either a birational morphism to its image and the

generic curve in |C| is non-hyperelliptic; or it is double cover of a rational
surface (possibly singular) of degree g − 1 in Pg , and the generic curve in
|C| is hyperelliptic.

6. If g = 2, then the map ϕ3C is birational. If g ≥ 3, then the map ϕ2C is

birational

Proof. Let us start with the first item. As g(C) = g and KX = as X is a K3

surface, then we get by the genus formula that C 2 = 2g − 2. Now, consider the
exact sequence
0 −→ OX (−C) −→ OX −→ OC −→ 0.

As H 0 (X, OX ) ' H 0 (C, OC ' C, and as H 1 (X, OX ) = 0 as X is a K3 surface,

we then get from this that H 1 (X, OX (−C)) = 0 and H 0 (X, OX (−C)) = 0.
As KX = 0, by Serre’s Duality this gives h2 (X, OX (C)) = h1 (X, OX (C)) = 0.
Then
h0 (X, OX (C)) = χ(OX (C)) = χ(OX ) + g − 1 = g + 1,

and we are done.

For the second item, as KX = 0 we have OC (KC ) = OC (C), by the adjunc-
tion formula, so that we have an exact sequence

0 −→ OX −→ OX (C) −→ OC (C) −→ 0.

As H 1 (X, OX ) = 0, the map H 0 (X, OX (C)) −→ H 0 (C, OC (C)) is surjective,

hence the complete linear system |C| cuts on C a complete linear system, which
is the canonical linear system on C. As g ≥ 1, deg(KC ) = 2g − 2 ≥ 0, so that
it has no base points on C. In conclusion, the linear system |C| has no base
points on X. Then φ is a morphism, and the rest is clear.
For the third item, as g = 1 we have C 2 = 0, and h0 (X, OX (C)) = 2 by item
1. Then we have a 1−dimensional pencil defined by |C|, every curve of which is
K3 surfaces 165

contracted by φ: hence we have a fibration whose generic memeber is a smooth

curve of genus 1, hence we have an elliptic fibration.
For the third item, as g = 2, we have C 2 = 2 and h0 (X, OX (C)) = 3. As
OX (C) = φ∗ O(1), we have that φ has degree 2. Let ∆ ⊆ P2 be the branche
curve, and let L be a line in P2 . Then L · ∆ = deg(∆). As φ−1 (L) is a double
cover of L ramified along deg(∆) points of L ∩ ∆. But notice that C = φ−1 and
g(C) = 2, so that the number of branching points is 6, and we are done.
For the fourth item, if C is non-hyperelliptic, then φ|C is an embedding,
hence C = φ−1 φ(C), so that φ is birational. If φ is not birational, every smooth
curve in |C| has to be hyperelliptic. Then for a geniric point x ∈ X the set
φ−1 φ(x) consists of 2 points, hence φ has degree 2. The image is the a surface
S of degree g − 1 in Pg , possibly singular, whose hyperplane sections are the
rational curves φ(C). Then S is rational.
The fifth item is easy: the restriction of φ3C (resp. φ2C ) is the 3−canonical
(resp. 2−canonical) map, hence it is an embedding, so that φ3C (resp. φ2C ) is
birational.

Let us now have a look to some particular cases, and let us suppose that X
is a K3 surface and C is a smooth curve of genus g ≥ 3 such that φ is birational.
If g = 3, then the map φ : X −→ P3 gives a surface φ(X) ⊆ P3 . The it has to
be a quartic surface, i. e. K3 surfaces of degree 4 in P3 are parameterized by
an open subset of the projective space |OP3 (4)|.
If g = 4, the situation is more complicated: we have that φ(X) is a surface of
degree 6 in P4 . As h0 (X, OX (2C)) = 14 by Riemann-Roch, and h0 (P4 , O(2)) =
15. Hence there is a quadric containing φ(X). In a similar way, we have
h0 (X, OX (3C)) = 29, and h0 (P4 , O(3)) = 35, so there is a cubic containing
φ(C). Hence φ(X) is complete intersection of a quadric and a cubic.
The case g = 5 is similar to the preceding one: it is a surface in P5 of
degree 8, and the generic one is complete intersection of 3 linearily independent
quadrics. Moreover, we have the following: in any of these cases, there is a
quasi-projective variety Tg parameterizing K3 surfaces of degree 2g − 2 in Pg .
As instance, if g = 3, this is an open subvariety in |OP3 (4)| ' P34 . It is then
quaite natural to ask about the dimension of Tg , i. e. to determine how many
K3 surfaces of degree 2g − 2 are in Pg . More precisely, on Tg acts the group
P GL(g + 1), and two K3 surfaces are isomorphic if they are in the orbit of this
group. One can prove that the stabilizer is finite, For g = 3, this group has
dimension 15, and hence the dimension of the moduli space of K3 surfaces of
degree 4 in P3 is 34 − 15 = 19.
For g = 2, we have that every K3 surface of degree 2 is double cover of
P ramified along a sextic, so that Tg is contained in |OP2 (6)| ' P27 , and the
2

dimension of P GL(3) is 8. Hence the dimension of the moduli space of K3

surfaces of degree 2 is 27 − 8 = 19. In a similar way one can see that the moduli
166 Further Properties and classifications

spaces of K3 surfaces of degree 6 in P4 and of degree 8 in P5 have dimension 19.

For higher g the situation is more complicated.

Proposition 5.3.4. For every g ≥ 3 there are K3 surfaces S2g−2 ⊆ Pg of degree

2g − 2.

Proof. By Theorem 5.3.3 we just need to produce a K3 surface admitting a very

ample divisor of self-intersection 2g −2. Actually, one can use the following fact:
if H is a hyperplane section on X and |E| is a linear system without base points,
then H + E is very ample. We have three cases:
Case 1 : g = 3k. Consider a quartic S ⊆ P3 , and suppose it contains a line.
Consider the pencil |H − L|, which is a pencil of elliptic curves. This pencil has
no base points, and if E ∈ |H − L| we have then that Dk := H + (k − 1)E is
very ample. Moreover

Dk2 = H 2 + (k − 1)2 E 2 + 2(k − 1)H · E = 4 + 6(k − 1) = 2g − 2.

Case 2 : g = 3k + 1. Let Q ⊆ P4 be a qaudric with a double point, and let

V ⊆ P4 be a cubic such that S = Q ∩ V is a smooth surface. We have two pencil
of planes on Q: one of them cuts on V a pencil |E| of elliptic curves. Again it
has no base points and Dk := H + (k − 1)E is a very ample divisor such that
Dk2 = 2g − 2.
Case 3 : g = 3k + 2. Let S ⊆ P3 be a smooth quartic containing a line L and
a twisted cubic T disjoint from L. Set E := H − L and H 0 := 2H − T . Again
one can show that Dk := H 0 + kE is very ample and that Dk2 = 2g − 2.

Even in this case, one can prove that there is a manifold Tg parameterizing
K3 surfaces of degree 2g − 2 in Pg , and that P GL(g + 1) acts on it. The moduli
space is again of dimension 19. Now, let me introduce the following definition:

Definition 5.3.2. A complex surface is called K3 surface if KX = 0 and

q(X) = 0.

If X is smooth and projective, then X is the familiar K3 surface on which we

have been talking up to now. As this definition makes sense for general complex
surfaces, one can ask if there are K3 surfaces which are not projective. To talk
of this, we need to introduce an importna notion, which is very general

Definition 5.3.3. Let X be any complex manifold. A deformation of X is the

datum of a complex manifold X together with a smooth morphism f : X −→ S
for some complex manifold S, such that there is s0 ∈ S such that f −1 (s0 ) = X.
The Kuranishi family for X is the universal deformation of X.

One of the most important results in deformation theory of surfaces is the

following:
K3 surfaces 167

Theorem 5.3.5. Let X be a complex surface. The minimal model of any de-
formation of X lies in the same class in the Enriques-Castelnuovo classification
as the minimal model of X.

In particular, any deformation of a K3 surface is again a K3 surface. One

can prove that the Kuranishi family exists for every K3 surface X, and let it be
f : X −→ S, where S is called the deformation space of X. By general results
on deformations of complex manifolds, one has that dim(S) = h1 (X, TX ), and
that S is smooth if and only if h2 (X, TX ) = 0. If X is a K3 surface, then we have
TX ' Ω∗X , so that Serre’s Duality gives h1 (X, TX ) = h1 (X, ΩX ) = h1,1 (X) = 20,
and h2 (X, TX ) = h0 (X, ΩX ) = q(X) = 0. Hence S is smooth and has dimension
20. Now if one deforms a projective K3 surface, and the deformation is still
projective, the deformation space would be of dimension 19 by the previous
part, hence this means that there are K3 surfaces which are not projective!
Here we have a first example of non-projective complex surfaces entering in the
Enriques-Castelnuovo Classification. There is another important result that I
want to state:

Theorem 5.3.6. (Siu). Every K3 surface is Kähler.

In conclusion, K3 surfaces are all Kähler surfaces which are not-necessarily

projective.

5.3.3 The Torelli Theorem for K3 surfaces

The last section on K3 surfaces is a digression on the Torelli Theorem, given an
important creterion to establish if two K3 surfaces are isomorphic, and hence
virtually completing the classification of K3 surface. First, let me introduce a
definition:

Definition 5.3.4. The K3 lattice is the lattice

ΛK3 := E8 (−1) ⊕ E8 (−1) ⊕ U ⊕ U ⊕ U,

where E8 (−1) is the rank 8 lattice whose intersection form is

 

 −2 0 1 0 0 0 0 0 

0 −2 0 1 0 0 0 0
 
 
 

 1 0 −2 1 0 0 0 0 

 0 1 1 −2 1 0 0 0 
,
 
−2


 0 0 0 1 1 0 0 


 0 0 0 0 1 −2 1 0 

0 0 0 0 0 1 −2 1
 
 
 
 0 0 0 0 0 0 1 −2 
168 Further Properties and classifications

and U is the rank 2 lattice whose interesection form is

 
 0 1 
.
 

 1 0 

The first property we need is the following:

Proposition 5.3.7. Let X be a K3 surface.

1. We have H 1 (X, Z) = H 3 (X, Z) = 0.

2. The Z−module H 2 (X, Z) is free of rank 22, and as a lattice with respect
to the cup produc it is isometric to ΛK3 .

Proof. To prove the first item and the fact that H 2 (X, Z) is free of rank 22 it
is sufficient to prove that H1 (X, Z) has no torsion. If there is an element of
n−torsion in H1 (X, Z), this would determine an unramified cover Y −→ X of
degree n. As KX = 0, this implies that KY = 0, so that pg (Y ) = 1. By the
Noether Formula we then get

2 − 2q(Y ) = χ(Y ) = nχ(X) = 2n,

so that n = 1 and q(Y ) = 0.

Now, the cup product on H 2 (X, Z) is even, and by Poincaré duality it is
unimodular. Moreover, the index is easily calculated to be −16. Then we have
two unimodular even lattices of the same index, ΛK3 and H 2 (X, Z), hence they
are isometric.

Definition 5.3.5. A marking on a K3 surface X is the choice of an isometry

between H 2 (X, Z) and ΛK3 . A marked K3 surface is the pair given by a K3 sur-
face and a marking. Two marked K3 surfaces (X, φ) and (Y, ψ) are isomorphic
if there is an isomorphism f : X −→ Y such that φ ◦ f ∗ = ψ.

Let now ΛC := ΛK3 ⊗Z C, which is a complex vector space of dimension 22,

and let P := P(ΛC ), which is a projective space of dimension 21. Inside of this,
we can consider

Ω := {[ω] ∈ P | (ω, ω) = 0, (ω, ω) > 0}.

Now, let X be a marked K3 surface, and let φ be the marking. As X is Kähler

by Siu’s Theorem, we have a Hodge decomposition

H 2 (X, C) ' H 2 (X, OX ) ⊕ H 1,1 (X) ⊕ H 2 (X, OX ).

As X is a K3 surface, we have that pg (X) = 1, so that H 2 (X, OX ) = C · σ. The

marking φ gives an isometry

φ : H 2 (X, C) −→ ΛC .
Enriques surfaces 169

Then the Hodge decomposition gives us a line C·φ(σ) ⊆ ΛC , i. e. a point [σ] ∈ P.

Now, As σ is a nowhere vanishing form we have (σ, σ) = 0 and (σ, σ) > 0, so
that [σ] ∈ Ω. In conclusion, we get a map

π : S −→ Ω, π(X) := [σ],

where S is the deformation space, which is called period map. It is a rather

important question the to understant which periods K3 surfaces can have, and
which informations on a K3 surface the period gathers.

Theorem 5.3.8. (Local Torelli Theorem). Two marked K3 surfaces X and

Y are isomorphic if and only if their periods are equal. Moreover, every ω ∈ Ω
is the period of a marked K3 surface.

Hence we can classify marked K3 surfaces by means of the period, and Ω is

the deformation space of K3 surfaces. But we have more:

Theorem 5.3.9. (Global Torelli Theorem). Two K3 surfaces X and Y are

isomorphic if and only if there is a Hodge isometry f : H 2 (X, Z) −→ H 2 (Y, Z).

In conclusion, K3 surfaces can be classified by means of the lattice structure

of H 2 (X, Z). Other important results on K3 surfaces are the following:

Theorem 5.3.10. Every two K3 surfaces are diffeomorphic. In particular every

K3 surface is simply connected.

5.4 Enriques surfaces

A second class of surfaces of Kodaira dimension 0 is the one of Enriques surfaces,
which are strictly related to K3 surfaces. Before starting with properties of
Enriques surfaces, let me recall their invariants. If X is an Enriques surfaces,
then KX 6= 0 but 2KX = 0, so that

2
KX = 0, Pm (X) ≡2 m − 1, κ(X) = 0,

q(X) = 0, pg (X) = 0, χ(OX ) = 1,

e(X) = 12, b1 (X) = 0, b2 (X) = 10.

As pg (X) = 0, we then conclude that h1,1 (X) = 10. Notice that as q(X) = 0 and
pg (X) = 0 the first Chern class is an isomorphism, so that H 2 (X, Z) ' P ic(X)
and ρ(X) = 10.
170 Further properties and classifications

5.4.1 Construction and projectivity

In Chapter 4 we have seen how one can produce examples of Enriques surfaces:
if X
e is a K3 surface admitting an involution i without fixed points, then X/i
e
is an Enriques surface. One of the main results of this section is that every
Enriques surface arises in this way.

Theorem 5.4.1. Let X be an Enriques surface. Then there is a K3 surface X

e
admitting an involution i without fixed points, such that X/i ' X.
e

Proof. As X is an Enriques surface, then KX is not trivial, but 2KX = 0, i. e.

KX has torsion 2 in P ic(X). Now, recall that the line bundle OX (KX ) can be
viewed as a rank 1 vector bundle KX −→ X, where KX is a complex manifold
of dimension 3. There is an isomorphism of line bundles α : OX (KX ) −→ OX ,
and a point s ∈ KX corresponds to a global section s of OX (KX ): let then

e := {s ∈ KX | α(s⊗2 ) = 1}.
X

Then X e comes with a natural map π : Xe −→ X, which is an unramified Galois

cover of degree 2, i. e. on X there is an involution i without fixed points such
e
that X/i
e = X.
It remains only to show that X is a K3 surface. First of all, as π is an
unramified cover of degree 2, we have χ(OXe ) = 2χ(OX ) = 2, as X is an Enriques
surface, so that q(X) e + 1. Moreover, we have that π ∗ KX = K e , but
e = pg (X)
X
e X KX ,
π ∗ KX has a nowhere vanishing section: indeed, we have that π ∗ KX = X×
and a section of π ∗ KX is obtained simply defining a map X e −→ X e ×X KX .
Consider then the map

σ:X e ×X K X ,
e −→ X σ(x) := (x, x),

which is everywhere non-vanishing. Hence π ∗ KX = 0, so that KXe = 0. This

then implies that pg (X)
e = 1, so that q(X)
e = 0, and X e is in conclusion a K3
surface.

Corollary 5.4.2. Let X be an Enriques surface. Then H1 (X, Z) ' Z/2Z.

Proof. By Theorem 5.4.1, there is a K3 surface X e admitting an involution i

such that X = X/i.
e By Proposition 5.3.10, a K3 surface is simply connected,
so that X is the universal cover of X, and π1 (X) is the Galois group of the
e
cover. As this is an unramified double cover, we get H1 (X, Z) ' Z/2Z.

In conclusion, Enriques surfaces are strictly related to K3 surfaces, and we

know hoy to produce them all. As for K3 surfaces, the definition of Enriques
surface can be extended to complex surfaces as follows:

Definition 5.4.1. A complex surface X is called Enriques surface if q(X) =

pg (X) = 0 and 2KX = 0.
Enriques surfaces 171

As there are K3 surfaces which are not projective, it is then quite natural
to ask if the same is true for Enriques surfaces, as these are quotients of K3
surfaces. The result is the following:

Proposition 5.4.3. Every Enriques surface is projective.

Proof. As X in an Enriques surface, then q(X) = pg (X) = 0, so that c1 :

P ic(X) −→ H 2 (X, Z) is an isomorphism. Hence for every class c ∈ H 2 (X, Z)
there is a unique (up to isomorphism) line bundle L such that c1 (L ) = c. Now,
recall the Hodge Index Theorem: the cup product on H 2 (X, Z) has signature
(1, 9), so there is a class c ∈ H 2 (X, Z) such that c2 > 0. Notice that c ∈ H 1,1 (X),
so it is a positive integral (1, 1)−form on X: by Kodaira Projectivity Criterion,
we finally get that X is projective.

As every Enriques surface is projective, and it is the quotient of a K3 surface

admitting an involution without fixed points, by the Kodaira Projectivity Cri-
terion it follows that if a K3 surface admits an involution without fixed points,
then it has to be projective. There are still interesting results on Enriques
surfaces that we won’t have the time to prove, like the following:

Theorem 5.4.4. Let X be an Enriques surface. Then it admits an elliptic

fibration.

5.4.2 The Torelli Theorem for Enriques surfaces

As for K3 surfaces, one can try to study the structure of the lattice H 2 (X, Z).
For K3 surfaces, this Z−module is free of rank 22, but for Enriques surfaces the
situation is different: its rank is 10, and it has torsion (as 2KX = 0). The first
result we need is the following:

Proposition 5.4.5. Let X be an Enriques surface.

1. There is an isomorphism P ic(X) ' Z10 ⊕ Z/2Z ' H 2 (X, Z)

2. The free Z−module H 2 (X, Z)f ree has rank 10 and it is isometric, as a
lattice, to ΛEn := E8 (−1) ⊕ U .

Proof. The first item follows immediately from ρ(X) = 10, the Universal Co-
efficients Theorem, and the fact that H 1 (X, Z) ' Z/2Z, so that H 2 (X, Z) '
H 2 (X, Z)f ree ⊕Z/2Z. As the first Chern class is an isomorphism, this completes
the proof of item 1.
For the second item, we have that H 2 (X, Z)f ree and ΛEn are two rank 10
unimodular even lattices with signature (1, 9) and the same index −8. Hence
the have to be isometric.
172 Further properties and classifications

For K3 surfaces, we have seen that the lattice H 2 (X, Z) is isometric to

ΛK3 = E8 (−1)⊕2 ⊕ U ⊕3 . This lattice has a natural involution, as follows:

σ : ΛK3 −→ ΛK3 , σ(x, y, a, b, c) := (y, x, −a, c, b).

Let us then define

Λ+
K3 := {l ∈ ΛK3 |, σ(l) = l}, Λ−
K3 := {l ∈ ΛK3 | σ(l) = −l}.

The lattice Λ+ +
K3 is the isometric to E8 (−2) ⊕ U (2), and l ∈ ΛK3 if and only if
(l, l) ≡4 0. We have then an isomoetry

1 +
Λ ' ΛEn .
2 K3
The first result we need is:

Proposition 5.4.6. Let X be an Enriques surface, and let p : X e −→ X be its

universal cover, where X
e is a K3 surface admitting an involution i. Then there
is an isometry
φ : H 2 (X,
e Z) −→ ΛK3

such that φ ◦ i∗ = σ ◦ φ.

Using this, the following definition makes sense:

Definition 5.4.2. A marking on an Enriques surface X is a marking φ on the

e such that φ ◦ i∗ = σ ◦ φ. A marked Enriques surface
associated K3 surface X,
is a pair given by an Enriques surface and a marking.

Let us now define the following: we defined the moduli space Ω, which is
locally isomorphic to the deformation space S of marked K3 surfaces via the
period map π. If (X, φ) is a marked Enriques surface, we can define its period
e φ). As φ◦i∗ = σ◦φ, the period π(X,
simply as the period of (X, e φ) satisfies some
more conditions. Indeed, if ω is a holomorphic 2−form on X e which is nowhere
∗
vanishing, then i ω = −ω (otherwise it would descend to a holomorphic 2−form
on X, but pg (X) = 0). Hence

σC (φ(ω)) = φC (i∗ (ω)) = −φC (ω),

so that φC (ω) ∈ Λ− −
C := ΛK3 ⊗Z C. Let us now define

Ω− := Ω ∩ P(Λ−
C ),

so that the period π(Y, φ) ∈ Ω−1 .

Theorem 5.4.7. (Torelli’s Theorem for Enriques surfaces). Two En-

riques surfaces are isomorphic if and only if their periods are equal.
Bielliptic surfaces 173

Proof. We give here an idea of the proof: let X, Y be two Enriques surfaces
and let pX : X e −→ X and pY : Ye −→ Y be the associated double covering
K3. We can choose markings φX and φY such that φX ◦ i∗X = σ ◦ φX and
similar for φY . The periods of (X, φX ) and (Y, φY ) are equal if and only if
the periods of (X, e φX ) and (Ye , φY ) are equal, hence by the Local and Global
Torelli Theorems for K3 surfaces, if and only if ψ := φY ◦ φ−1 2 e
X : H (Y , Z) −→
H 2 (X,
e Z) is a Hodge isometry, i. e. if and only if there is an isomorphism
g:X e −→ X e such that g ∗ = φY ◦ φ−1 . Notice that hence g ∗ ◦ i∗ = i∗ ◦ g ∗ , so
X Y X
that g ∗ ◦ i∗Y ◦ (g −1 )∗ ◦ i∗X = idH 2 (X,Z)
e . By the Global Torelli Theorem this is
−1
true if and only if g ◦ iY ◦ g ◦ iX = idXe , so that g ◦ iY = iX ◦ g, i. e. g descends
to an isomorphism between X and Y .

Another important result about Enriques surfaces is the following:

Theorem 5.4.8. Any two Enriques surfaces are equivalent under deformation.

5.5 Bielliptic surfaces

The next family of surfaces of Kodaira dimension 0 is given by bielliptic surfaces.
The aim of this section is the give a complete list of the possible bielliptic
surfaces. Let me recall the invariants: if X is a bielliptic surface, then KX 6= 0
(and as we will see we have 4KX = 0 or 6KX = 0), and
2
KX = 0, Pm (X) = 0, 1, κ(X) = 0,

q(X) = 1, pg (X) = 0, χ(OX ) = 0,

e(X) = 0, b1 (X) = 2, b2 (X) = 2.
Notice that as pg (X) = 0, we get that h1,1 (X) = 2. The first Chern class
is surjective, so that ρ(X) = 2. Moreover, notice that every bielliptic surface
carries a natural elliptic fibration, and they are all projective.
The first result we need to show is the following:

Lemma 5.5.1. Let E be an elliptic curve. Then every automorphism of E is

the composition of a translation and a group automorphism. The non-trivial
group automorphisms are the symmetry x 7→ −x and:

1. if E = Ei := C/(Z ⊕ Zi), the automorphism x 7→ ±ix;

2. if E = Eρ := C/(Z ⊕ Zρ, the automorphisms x 7→ ±ρx and x 7→ ±ρ2 x,

where ρ3 = 1, ρ 6= 1.

Proof. By Lemma 3.3.5, every automorphism of E is induced by an automor-

phism of C sending the lattice Γ (here E = C/Γ)itself. Hence every automor-
phism is the composition of a group translation with a group automorphism,
which are only those written above.
174 Further properties and classifications

Using this lemma, we can prove the following:

Proposition 5.5.2. Let X = (B × C)/G be a minimal surface which is not

ruled, where B and C are smooth curves and G is a finite group acting faithfully
on B and C. Suppose that pg (X) = 0 and q(X) ≥ 1. Then the curve B and
C are not rational, B/G is elliptic, C/G is rational and one of the following
possibilities is verified:

1. the curve B is elliptic and the group G is a group of translations on B;

2. the curve C is elliptic and the group G acts freely on C × B.

Moreover, the following properties are verified:

• we have P4 (X) 6= 0 or P6 (X) 6= 0. In particular P12 (X) = 1.

• If one between C and B is not elliptic, then κ(X) = 1; if B and C are

elliptic, then 4KX = 0 or 6KX = 0, so that 12KX = 0.

Proof. By Proposition 5.1.5, we know that if a surface X is minimal and has

2
pg = 0 and q(X) ≥ 1, then KX ≤ 0, and it is 0 if and only if q(X) = 1 and
2 2
b2 (X) = 2. As X is not ruled, then we need KX ≥ 0, so that finally KX = 0 and
q(X) = 1. Recall that as X = (B×C)/G, we have q(X) = g(B/G)+g(C/G). As
q(X) = 1, we then need that one between the two quotients is rational, and the
other is elliptic. Let us suppose that B/G is elliptic, so that C/G is rational.
If C = P1 , then C/G is no longer rational, so that C has to be irrational.
Similarily, B cannot be rational. Hence we have only two possibilites: either B
is elliptic, or C is elliptic. In any case, the action of G is free on B × C. In the
case where B is elliptic, the finite group G has to be a group of translantions
on B, so the first part is shown.
For the remaining part, notice that the second part of the second item follows
from the first item: as B and C are elliptic, then KB×C = 0. If D ∈ |4KX |,
we have π ∗ D = 0 if P4 (X) = 0, where π : B × C −→ X is the étale cover.
Similarily, if P6 (X) = 0, then 6KX = 0.
Let us first suppose that B is elliptic. We have

H 0 (Ω2X )⊗k ' (H 0 (B × C, Ω2B×C )⊗k )G ' (H 0 (B, ΩB )⊗k )G ⊗ (H 0 (C, ΩC )⊗k )G .

As G acts as translations on B by the first part, we have that (H 0 (B, ΩB ))G =

H 0 (B, ΩB ), so that
Pm (X) = h0 (C, OC (mKC ))G .

Now, let Lm := OP1 (−2m + p∈P1 m(1 − (1/ep ))), where ep is the ramification
P

index at p of the action og G on C. Hence

Pm (X) = h0 (P1 , Lm ),
Bielliptic surfaces 175

But now by Riemann-Roch

X  1
 
2g(C) − 2
Pm (X) ≥ −2m + m 1− −1 =m − r,
1
e p n
p∈P

where n is the order of G, r is the number of ramification points, by the Riemann-

Hurwitz Formula. If g(C) ≥ 2, then κ(X) = 1. Let us now suppose that C is
P
elliptic. As p (1 − (1/ep )) ≥ 2 by the Riemann-Hurwitz Formula, we have to
show that deg(Lm ) ≥ 0 for some k. We have the following possibilites:

1. r ≥ 4. Hence 2(1 − (1/ei )) ≥ 1, we have that deg(L2 ) ≥ 0.

2. r = 3. Then (1/e1 )+(1/e2 )+(1/e3 ) ≤ 1. If e1 ≥ 3, then 3(1−(1/e1 )) ≥ 2,

so that deg(L3 ) ≥ 0, and we can suppose e1 = 2, i. e. (1/e2 ) + (1/e3 ) ≤
1/2. If e2 ≥ 4, then again deg(L4 ) ≥ 0. If e2 = 3, then e3 ≥ 6, so that
deg(L6 ) ≥ 0.

In any case, hence P4 (X) 6= 0 or P6 (X) 6= 0, so that P12 (X) 6= 0.

Now, let us suppose that B is not elliptic, hence C has to be elliptic. Assume
that C 6= Ei , Eρ . Hence the only automorphisms on C can be translations on
C or −id, so that if ω ∈ H 0 (C, ΩC ), then ω 2 ∈ H 0 (C, ΩC )G . Hence we have
X
P2m (X) = h0 (B/G, OB/G ( (m(1 − (1/ep )))p)),
p

as B/G is an elliptic curve. If g(B) ≥ 2, then κ(X) = 1. If g(B) = 1, then

this expression is not zero, so that P2m (X) 6= 0. Similarily, if C = Ei , then we
need to consider ω ⊗4 , so that P4m (X) 6= 0 and the expression is the same. If
C = Eρ , we need ω ⊗6 , so that P6 (X) 6= 0.

As a corollary we have the following:

Corollary 5.5.3. A minimal surface X is ruled if and only if P12 (X) = 0.

Proof. If X is ruled, then clearly we have P12 (X) = 0. If P12 (X) = 0, then
pg (X) = P2 (X) = P4 (X) = P6 (X) = P3 (X) = 0. Hence if q(X) = 0, then
X is rational by the Castelnuovo Rationality Criterion. If q(X) ≥ 1, we need
2
KX < 0, so that KX is not nef, and κ(X) = −∞. In conclusion, X is ruled.

Another corollary is the following

Corollary 5.5.4. Every minimal surface X such that κ(X) = 0 has P12 (X) = 1
and 12KX = 0.

Proof. If X is a K3 or abelian surface, then P12 (X) = 1 and KX (X) = 0, so that

12KX = 0. If X is an Enriques surface, then 2KX = 0, so that 12KX = 0 and
P12 (X) = 0. If X is bielliptic, then 4KX = 0 or 6KX = 0 by Proposition 5.5.2.
Hence 12KX = 0, so that P12 (X) = 1. By The Enriques-Kodaira Classification,
this is all.
176 Further properties and classifications

We can now give the complete list of bielliptic surfaces:

Theorem 5.5.5. (Bagnera-De Franchis). Let X = (E ×F )/G be a bielliptic

surface. Suppose that G is a group of translations on E and acting of F . Then
E, F and G can be one and only one of the following types:

1. E and F are any two elliptic curves, G = Z/2Z acting on F by symmetry.

In this case we have 2KX = 0;

2. E and F are any two elliptic curves, G = Z/2Z ⊕ Z/2Z acting on F by

x 7→ −x, x 7→ x + , where  is a 2−torsion point of F . In this case we
have 2KX = 0;

3. E is any elliptic curve, F = Ei , G = Z/4Z acting on F by x 7→ ix. In

this case we have 4KX = 0;

4. E is any elliptic curve, F = Ei , G = Z/4Z⊕Z/2Z acting on F by x 7→ ix,

x 7→ x + ( 1+i
2 ). In this case we have 4KX = 0;

5. E is any elliptic curve, F = Eρ , G = Z/3Z acting on F by x 7→ ρx. In

this case we have 6KX = 0;

6. E is any elliptic curve, F = Eρ , G = Z/3Z⊕Z/3Z acting on F as x 7→ ρx,

x 7→ x + ( 1−ρ
3 ). In this case we have 6KX = 0;

7. E is any elliptic curve, F = Eρ , G = Z/6Z acting on F by x 7→ ρ2 x. In

this case we have 6KX = 0.

Proof. The group G is a subgroup of Aut(F ), so it has to be of the form G =

T o A, where T is a group of translations on F , and A is a group of group
automorphisms of F . Since F/G ' P1 , the group A is not trivial. By Lemma
5.5.1, A can be only of the form Z/2Z, Z/3Z, Z/4Z or Z/6Z. Moreover, G is
s group of translations on E, so that G = T ⊕ A, i. e. every element of T is
A−invariant. Now, the fixed points of A can be the following:

1. If A acts as x 7→ −x, the points of order 2.

1+i
2. If A acts as x 7→ ix, i. e. F = Ei , the points 0 and 2 .

1−ρ
3. If A acts as x 7→ ρx, i. e. F = Eρ the points 0, 3 and − 1+ρ
3 .

4. If A acts as x 7→ ρ2 x, i. e. F = Eρ , the point 0.

Now, simply notice that as G is a group of translations on E, it can be generated

by two elements, so that G 6= F [2] × A where F [2] is the subgroup of 2−torsion
points. The last remaining possibilities are those listed in the statement.
Abelian surfaces 177

5.6 Abelian surfaces

The last remaining class of surfaces of Kodaira dimension 0 is the one of abelian
surfaces. As the theory of abelian surfaces is almost the theory of abelian va-
rieties of any dimension, it is out of our goals to list their properties. Anyway,
I would like to collect some basic facts about the Torelli Theorem for complex
tori and projectivity criteria. I will also collect some basic facts about the coho-
mology of line bundles on abelian surfaces. Before, let me recall the invariants:
if X is an abelian surface, then KX = 0 and

2
KX = 0, Pm (X) = 1, κ(X) = 0,

q(X) = 2, pg (X) = 1, χ(OX ) = 0,

e(X) = 0, b1 (X) = 4, b2 (X) = 6.

We have then h1,1 (X) = 4, so that 0 ≤ ρ(X) ≤ 4. By definition, notice that

every complex torus is Kähler. One of the main question is then when a complex
torus is an abelian surface.

5.6.1 Properties of abelian surfaces

Recall that an abelian surface is defined as a complex torus of dimension 2
admitting an embedding into some projective space. Hence, any abelian surface
X is of the form X = C2 /Γ, where Γ is a maximal rank lattice (i. e. of
rank 4) in C2 . There are two stupid remarks to do immediately: as C2 is
simply connected, it is then the universal cover of X, so that it is obvious that
π1 (X) ' Γ. Moreover, as Γ ' Z4 is an abelian group, we have H1 (X, Z) ' Γ.
This gives an intrinsec description of the first singular cohomology group of an
abelian surface (actually, of every complex torus), which is then a free Z−module
of rank 4. Let me first introduce the following definition: if one fixes a basis
e1 , e2 of C2 and a basis λ1 , λ2 , λ3 , λ4 of Γ, we can write every λi in terms of e1 , e2 ,
getting λi = λi1 e1 + λi2 e2 , where λij ∈ C for every i = 1, ..., 4 and j = 1, 2. We
then get a matrix ΠX := [λij ]i,j , depending on the chosen basis.

Definition 5.6.1. The matrix ΠX ∈ M (2 × 4, C) is called period matrix of X

(with respect to the chosen basis).

Notice that ΠX determines the torus completely, but different matrices do

not give necessarily different abelian surfaces. Moreover, not every matrices
defines an abelian surface. We have the following:

Proposition 5.6.1. Let M ∈ M (2 × 4, C), and let M be its complex conjugate

matrix. Then there is a complex torus X such that ΠX = M if and only if the
matrix P := M


M ∈ M (4 × 4, C) is invertible.
178 Further properties and classifications

Proof. The the matrix M defines a lattice Γ ⊆ C2 , which defines a complex

torus if and only if it is of maximal rank, i. e. if and only if the columns are
linearily independent on R. Now, if M does not define a complex torus, two
columns have to be dependent, and let us suppose that thesare the first two
columns. Then there is x ∈ R4 , x 6= 0 such that M x = 0, so that P x = 0.
Hence det(P ) = 0, and P is not invertible. Conversely, if P is not invertible,
then there are x, y ∈ R4 , x, y 6= 0, such that P (x + iy) = 0. Then M (x + iy) = 0
and M (x − iy) = M (x + iy) = 0, so that M x = M y = 0, hence two columns of
M are linearily dependent on R, i. e. the lattice Γ has not maximal rank.

The period matrix is important even because it gives us informations on the

projectivity of the associated complex torus. Before looking at this problem, let
me discuss a little bit of singular cohomology of abelian surfaces. As H1 (X, Z)
is a free Z−module, the universal coefficient theorem tells us that H 1 (X, Z) '
Hom(H1 (X, Z), Z) is free of rank 4, and H 2 (X, Z) is free of rank 6. But we
have a more intrinsec description of H 2 (X, Z):

Proposition 5.6.2. The canonical map

2
^
H 1 (X, Z) −→ H 2 (X, Z)

induced by the cup product is an isomorphism.

Proof. This is just the Künneth formula.

As H 1 (X, Z) = Hom(Γ, Z), we have hence

2
^
H 2 (X, Z) ' Alt2 (Γ, Z) := Hom(Γ, Z),

which are alternating integral 2−forms on Γ. In particular, we have

H 2 (X, C) ' Alt2R (V, C),

where V := Γ ⊗Z C is a complex vector space of dimension 4, and Alt2R (V, C) is

the group of R−linear alternating 2−forms on V .
Notice that on H 1 (X, C) we have a Hodge structure of weight 1, i. e. a de-
composition H 1 (X, C) = H 1,0 (X) ⊕ H 0,1 (X), where H i,j (X) is the vector space
of global (i, j)−forms on X. We have that H 1 (X, C) is canonically isomorphic
to the complex vector space of 1−forms on C2 invariant under Γ, and similarily
we have that H i,j (X) is the vector space of Γ−invariant (i, j)−forms on C2 . By
Proposition 5.6.2 we have then that H 2 (X, C) is the vector space of Γ−invariant
2−forms on C2 , and the Hodge decomposition has the same property.
Now, let L ∈ P ic(X), so that c1 (L ) ∈ H 2 (X, Z) ∩ H 1,1 (X). Hence c1 (L )
can be seen either as a Γ−invariant (1, 1)−form on C2 , or as an integral alter-
nating 2−form on Γ. It is interesting to understant which alternating 2−forms
on Γ are corresponding to the first Chern class of a line bundle:
Abelian surfaces 179

Proposition 5.6.3. Let E ∈ Alt2 (V, R). The following are equivalent:

1. there is a line bundle L ∈ P ic(X) such that E represents c1 (L );

2. E(Γ, Γ) ⊆ Z and E(v, w) = E(iv, iw) for every v, w ∈ V .

Proof. Let Ω := HomC (V, C) and Ω = HomC (V, C). Then it is easy to show
that
2
^ ^2
Alt2R (V, C) ' Ω ⊕ (Ω × Ω) ⊕ Ω.
V2 V2
Hence we have natural identifications H 2 (X, C) ' Ω ⊕ (Ω × Ω) ⊕ Ω and
V2
H 0,2 (X) ' Ω. If L ∈ P ic(X), then c1 (L ) can be represented by an element
E ∈ Alt2R (V, C) that we can write as E = E1 +E2 +E3 according to the previous
decomposition. As E has values in R we need E1 = E3 , and E3 = 0 as c1 (L )
is a (1, 1)−form. Hence E = E2 and it is easy to see that it verifies the two
conditions in the statement. The converse is similar.

There is another way to represent the first Chern class of a line bundle:

Proposition 5.6.4. There is a 1 to 1 correspondence between alternating real

forms E on V such that E(Γ, Γ) ⊆ Z and E(v, w) = E(iv, iw) for every
v, w ∈ V , and Hermitian forms H on V , i. e. C−linear forms on V such
that H(v, w) = H(w, v) for every v, w ∈ V .

Proof. Let E : V × V −→ R be alternating and such that E(Γ, Γ) ⊆ Z and

E(v, w) = E(iv, iw) for every v, w ∈ V . Define

HE : V × V −→ C, HE (v, w) := E(iv, w) + iE(v, w).

Then H is C−linear and it is easy to verify that it is Hermitian. For the converse,
let H be Hermitian on V , and define

EH : V × V −→ R, EH (v, w) := Im(H(v, w)),

and it is easy to verify that EH is R−linear and satisfies the properties of the
statement.

In conclusion, the first Chern class of any line bundle corresponds to an

Hermitian form on V . Using this, we can finally prove the following:

Theorem 5.6.5. (Riemann’s Projectivity Criterion). Let X = V /Γ be

a 2−dimensional complex torus. Then X is an abelian surface if and only if
there is a non-degenerate alternating matrix A ∈ M (4 × 4, Z) such that the two
following relations are verified:

1. ΠX · A−1 · ΠTX = 0,
T
2. iΠX · A−1 · ΠX > 0.
180 Further properties and classifications

Proof. Consider an arbitrary non-degenerate alternating 2−form E on Γ (denote

by E even its real extension to V ), and consider

H : V × V −→ C, H(v, w) := E(iv, w) + iE(v, w).

Let A be the matrix representing E with respect to the basis λ1 , λ2 , λ3 , λ4 of

Γ, which is the basis with respect to which the period matrix of X is ΠX . By
Propositions 5.6.4 and 5.6.3, H is Hermitian if and only if there is a line bundle
L ∈ P ic(X) such that c1 (L ) is represented by E. Moreover, X is projective if
and only if c1 (L ) is a positive definite (1, 1)−form by the Kodaira Projectivity
Criterion. Notice that as c1 (L ) is represented by E, the positivity of c1 (L ) is
equivalent to the fact that H is positive definite. Hence the projectivity of X is
equivalent to the existence of a positive definite Hermitian form H on V . Le us
now read these two properties in terms of the matrix A.
Step 1. The form H is Hermitian if and only if ΠX · A−1 · ΠTX = 0. Indeed,
the form H is Hermitian if and only if E(v, w) = E(iv, iw) for every v, w ∈ V .
Consider the matrix
" #−1 " # " #
ΠX iI2 0 ΠX
M := · · .
ΠX 0 iI2 ΠX

It is easy to see that iΠX = ΠX · M . Moreover, we have

E(ΠX x, ΠX y) = xT Ay,

for every x, y ∈ R4 , so that H is Hermitian if and only if A = M T · A · M .

Equivalently, if and only if the following relation is verified:
" # " # " #
iI2 0 ΠX h T
i −1 iI2 0
−1 T
· · A · ΠX ΠX · =
0 −iI2 ΠX 0 −iI2
" #
i −1
ΠX −1
h T
= ·A · ΠTX ΠX .
ΠX
Comparing the 2 × 2 blocks on each side, we finish step 1.
Step 2. The Hermitian form H is positive definite if and only if iΠX · A−1 ·
T
ΠX > 0. Indeed, let x, y ∈ R4 , and let U := ΠX x and v := ΠX y. As H is
T
Hermitian, by Step 1 we know that ΠX · A−1 ΠX = 0, so that we get

E(iu, v) = xT M T Ay =
" # " # " #
iI2 0 ΠX h T
i −1 v
= [u u] · · · A−1 · ΠTX ΠX · =
0 −iI2 ΠX v
" # " #
0 i(ΠX · A−1 · ΠTX )−1 v
= [u u] · · =
−i(ΠX · A−1 · ΠTX )−1 0 v
Abelian surfaces 181

= iuT (ΠX · A−1 · ΠTX )−1 v − iuT (ΠX · A−1 · ΠTX )−1 v.

Similarily one computes

E(u, v) = uT (ΠX · A−1 · ΠTX )−1 v + uT (ΠX · A−1 · ΠTX )−1 v,

so that
H(u, v) = 2iuT (ΠX · A−1 · ΠTX )−1 v.

Hence H = 2i(ΠX · A−1 · ΠTX )−1 , so that it is positive definite if and only if
T
iΠX · A−1 · ΠX > 0.

Now, let L ∈ P ic(X), where X = V /Γ is any complex torus. We know

that c1 (L ) is represented by an Hermitian form H on V , whose alternating real
form E = Im(H) is integer valued on Γ. It is a basic result in matrix theory
that hence there is a basis λ1 , λ2 , µ1 , µ2 of Γ with respect to which the form E
is represented by a matrix
" #
0 D
ME := ,
−D 0

where D = diag(d1 , d2 ) is a diagonal matrix of eigenvalues d1 , d2 ∈ N0 , with

d1 |d2 . These elements are called elementary divisor, and are completely deter-
mined by E, hence by L .

Definition 5.6.2. The type of L is the pair (d1 , d2 ).

Notice that E is non-degenerate if and only if d1 , d2 > 0. If L is ample, i.

e. c1 (L ) is positive define, then L is called a polarization. If the type of L
is (1, 1), then L is called a principal polarization. An abelian surface is then
a 2−dimensional complex torus admitting a polarization. An abelian surface is
principally polarized if it admits a principal polarization. An interesting result
is the following:

Theorem 5.6.6. Let X be an abelian surface, and let L be a polarization of

type (d1 , d2 ) on X. If d1 ≥ 3, the map ϕL : X −→ PN is an embedding.

5.6.2 The Torelli Theorem for abelian surfaces

Abelian surface, and more generally complex tori, are very special for several
reasons. First of all, the theory of abelian surfaces and 2−dimensional complex
torus can almost completely be generalized to higher dimension. Moreover, as
for K3 and Enriques surfaces we have a Torelli Theorem for 2−dimensional
complex tori: the main difference is that it does not involve H 2 (X, Z) but
H 1 (X, Z).
182 Further properties and classifications

Theorem 5.6.7. (Torelli’s Theorem for complex tori). Let X and X 0 be

two complex tori of dimension 2. A Hodge morphism

φ : H 1 (X 0 , Z) −→ H 1 (X, Z)

is a Hodge isomorphism if and only if there is an isomorphism f : X −→ X 0

such that f ∗ = φ.

Proof. The proof uses properties of the Albanese map. Recall that aX : X −→
Alb(X) is an isomorphism by the universal property of the Albanese torus, hence
X and X 0 are isomorphic if and only if Alb(X) and Alb(X 0 ) are isomorphic.
Recall now the definition of Alb(X): we have

Alb(X) := H 0 (X, ΩX )∗ /im(H1 (X, Z)).

The morphism φ induces a morphism

φ1,0 : H 0 (X 0 , ΩX 0 ) −→ H 0 (X, ΩX )

such that φ1,0 (im(H1 (X 0 , Z))) ⊆ im(H1 (X, Z)). Hence φ is a Hodge isomor-
phism if and only if there is an isomorphism h : Alb(X 0 ) −→ Alb(X). Hence we
get an isomorphism

f := a−1
X0 ◦ h
−1
◦ aX : X −→ X 0 ,

and it is easy to verify that f ∗ = φ.

This provides a complete classification of 2−dimensional complex tori. One

can read this in terms of H 2 (X, Z) and its Hodge structure, as H 2 (X, Z) '
V2 1
H (X, Z). Indeed, one has the following:

Theorem 5.6.8. Let X and X 0 be two 2−dimensional complex tori. A Hodge

morphism respecting the intersection forms

φ : H 2 (X 0 , Z) −→ H 2 (X, Z)

is a Hodge isometry if and only if there is an isomorphism f : X −→ X 0 such

that f ∗ = φ.

V2 1
Proof. The morphism φ is of the form φ = ψ ∧ ψ as H 2 (X, Z) ' H (X, Z)
for every complex torus X. One can show (using lattice theory) that φ is a
Hodge isometry if and only if ψ is a Hodge isomorphism. But using the Torelli
Theorem for complex tori, we have the statement.
Elliptic surfaces 183

5.7 Elliptic surfaces

The next class of surfaces we want to study is the one of elliptic surfaces. Recall
that an elliptic surface is simply a surface X together with a holomorphic map
π : X −→ C, where C is a smooth curve, and the generic fiber of π is a smooth
elliptic curve. By the Enriques-Castelnuovo classification of surfaces, we have
seen that if a surface X is elliptic, then κ(X) ≤ 1, and we have seen examples
of elliptic surfaces of Kodaira dimension −∞, 0 and 1. It is then a natural
question if it is possible to classify ellitpic surfaces up to birationality. In order
to do so, let me introduce the following definition

Definition 5.7.1. An elliptic surface X is relatively minimal if it has no

(−1)−curves in the fibers. Two elliptic surfaces π1 : X1 −→ C and π2 : X2 −→
C are birational as elliptic surfaces if there is a birational map f : X1 99K X2
such that π2 ◦ f = π1 .

In general, as the generic fiber of π is an elliptic surface, we suppose that

there is a section s : C −→ X of π: simply send a point p ∈ C over which
the fiber is smooth to s(p) := O, where O is the zero point of π −1 (p). More
generally one can consider X without sections, if there is no choice of origin of
the fibers.

5.7.1 Classification of singular fibers

In this first section I want to present the classification of singular fibers of an
elliptic fibration given by Kodaira. To prove this, I need to recall some basic
facts about graphs and Dynkin diagrams. First of all, let me give the list of the
Dynkin diagrams. These are An for n > 0 which is of the form is a cycle with
n + 1 vertices; Dn , for n ≥ 4; E6 , E7 and E8 .
Let me recall some basic facts about graph theory: let G be a graph with
possible loops and multiple edges, and consider the vector space VG whose basis
is in bijection with the vertices of G. We can define a symmetryc bilinear form
(., .) on VG as follows: (v, v) = −2 + 2l(v), where l(v) is the number of loops of
v, for every vertex v; and (v, w) = e(v, w) for every v 6= w, where e(v, w) is the
number of edges joining v and w. The vector space VG together with the form
((., .) is called the associated form to G.

Lemma 5.7.1. We have the following:

1. The associated form to any Dynkin diagram is negative semi-definite with

a 1−dimensional kernel spanned by the vector of multiplicites.

2. Every graph contains or is contained in some Dynking diagram; every

connected graph without loops or multiple edges cointains or is contained
in some Dynkin diagram without loops or multiple edges.
184 Further properties and classifications

3. If a graph has associated form which is negative semi-definite and has

1−dimensional kernel, then it is a Dynkin diagram.

Using this Lemma, we are ablo to show the following:

Theorem 5.7.2. (Kodaira’s classification of singular fibers). Let π :

X −→ C be a minimal elliptic surface. Then the only possible fibers of π can
be:

1. Type I0 : a smooth elliptic curve.

2. Type I1 : a nodal rational curve.

3. Type In for n ≥ 3: n smooth rational curves meeting in a cycle, i. e. with

dual graph An .

4. Type In∗ for n ≥ 0: n + 5 smooth rational curves meeting with dual graph
Dn+4 .

5. Type II: a cuspidal rational curve.

6. Type III: two smooth rational curves meeting at one point with order 2.

7. Type IV : three smooth rational curves meeting at one point.

8. Type IV ∗: seven smooth rational curves meeting with dual graph E6 .

9. Type III∗: eight smooth rational curves meeting with dual graph E7 .

10. Type II∗: nine smooth rational curves meeting with dual graph E8 .

11. Type m In : for n ≥ 0 and m ≥ 2: topologically In , but with multiplicity m.

Moreover, every component of a fiber has self-intersection −2.

Proof. Let π : X −→ C have a singular fiber over 0 ∈ C, and let it be X0 . It can

P
be written in the form X0 = i ni Ci , where ni ∈ N and Ci is irreducible. Let
m := mult(X0 ) be the multiplicity of the fiber. Write then X0 = mF , where
P
F = i ri Ci , where ni = mri .
If F is irreducible, then g(F ) = 1 by the genus formula, so it can either be
a smooth elliptic curve (i. e. type I0 ), a nodal rational curve (i. e. type I1 ) or
a cuspidal rational curve (i. e. type II). Assume now that F is reducible, and
let E be the generic fiber of π. As KX · E = 0, we have KX · F = 0, so that
X X
0 = KX · F = ri Ci · KX = ri (2g(Ci ) − 2 − Ci2 ).
i i

If 2g(Ci ) − 2 − Ci2 < 0 for some i, then we get

−2 ≤ 2g(Ci ) − 2 < Ci2 < 0,

Elliptic surfaces 185

where the last inequality follows from the Zariski Lemma. Hence we have
g(Ci ) = 0, so that Ci is a smooth rational curve, and Ci2 = −1, which is
not possible as X is relatively minimal. Hence we need 2g(Ci ) − 2 ≥ Ci2 for
every i. As ri > 0, we then need 2g(Ci ) − 2 − Ci2 = 0 for every i. Now Ci2 < 0,
so that 2g(Ci ) − 2 < 0, hence g(Ci ) = 0 and Ci is a smooth irreducible curve
such that Ci2 = −2.
Now, consider the dual graph G of F : for every component Ci we have a
vertex, for every i, j we have Ci · Cj vertices joining vi to vj and multiplicity Ci2
to the vertex vi . As the intersection form is negative semi-definite, we need the
graph G to be one of the Dynkin diagrams by Lemma 5.7.1. If G = A1 , then
F is either I2 if the two components meet in two different points, or III if the
two components meet at one point. If G = A3 , then F is either I3 if the three
components meet in a cycle, or IV is they meet in a single point. In any other
case we obtain the other possible diagrams.
Now, let us complete the analysis for m > 1. In this case F must be simply
connected because of Corollary 4.3.9, so that the topological type of X0 is In ,
and we are done.

One can actually give examples of elliptic surfaces carrying at least one of
these fibers.
Example 5.7.1. Let us consider C1 a smooth cubic in P2 , and C2 be another
cubic. The pencil generated by C1 and C2 has 9 base points p1 , ..., p9 counted
with multiplicity. Consider X := Blp1 ,...,p9 P2 , which has a natural map

π : X −→ P1 ,

whose fiber is an element in the pencil generated by C1 and C2 . As the generic

element is a smooth cubic in P2 , it is an elliptic curve. Notice that there are no
(−1)−curves on X, so X is minimal.
Now, let us suppose that C2 is a reduced curve meeting C1 transversally, i.
e. in 9 distinct points. The fiber of π corresponding to C2 is then C2 itself.
Hence if C2 is a nodal cubic we obtain a fiber of type I1 , if C2 is a cuspidal cubic
we obtain a fiber of type II. If C2 = Q + L is a conic plus a line, then the fiber
is of type I2 if L is not tangent to Q, of type III if the line is tangent. If C2 is
given by three distinct lines, the fiber is of type III if they are not cuncurrent,
or of type IV is they are cuncurrent. If we chose C2 = 2L + M , for L and M
distinct such that C2 meets transversally M and L in three points each, the
fiber is of type I0∗ (here the proper transform of L occurs with multiplicity 2 in
the fiber). If C1 is tangent to L, we obtain a fiber of type I1∗ . If C2 = 3L and
C1 meets L in three distinct points, then we obtain a fiber of type IV ∗ . If C1 is
tangent to L, then we get a fiber of type III ∗ . If C1 has a flex on L, then the
fiber is of type II ∗ .
186 Further properties and classifications

To get multiple fibers, we consider C a smooth cubic, p1 , ..., p9 ∈ C so that

P
i pi is not a divisor in 3H (where H is the hyperplane section of C), but
2 i pi is a divisor in 6H. Consider the set of sextics of P2 having 9 double
P

points at the pi ’s. There is at least one of such sextics, for example 2C. One
can show that there are 2 such sextics, one is 2C and the other is S. The generic
member of this pencil has genus 1 because a smooth sextic has arithmetic genus
10. Blowing up P2 at p1 , ..., p9 be obtain a smooth minimal surface having an
elliptic fibration over P1 , whose singular fiber is the proper transform of 2C,
which is of type 2 I0 . If one considers curves of degree 3m having 9 m−fold
points as base locus, and cutting out a divisor D on a smooth cubic sucht that
mD is a divisor in 3mH, we get an elliptic surface with singular fiber m I0 .

5.7.2 Classification of elliptic surfaces

In this section we want to give a complete classification of relatively minimal
elliptic surfaces according to the Enriques-Castelnuovo classification. The first
property we need to prove is the following:

Proposition 5.7.3. Let π1 : X1 −→ C and π2 : X2 −→ C be two minimal

elliptic surfaces which are birational as elliptic surfaces aver C. Then they are
isomorphic as elliptic surfaces over C. In particular, every elliptic surface over
C has a unique minimal model as elliptic surface over C

Proof. Let f : X1 99K X2 be a birational map such that π2 ◦ f = π1 . As f

preserves the elliptic fibrations, it can be resolved only by blowing-up points
and blowing-down curves in the fibers of πi : hence we can find a surface X
together with two morphisms g1 : X −→ X1 and g2 : X −→ X2 such that
f ◦ g1 = g2 . Consider it to be such that the number of blow-ups to resolve f is
minimal. Hence g2 is an isomorphism: it if is not, we can factor it by a sequence
of blow-ups, and let E be the first exceptional curve which g2 contracts. If E is
exceptional for g1 , then contracting it by g1 we get a new surface X 0 resolving
f , which is not possible as X is obtained by the minimal number of blow-ups
resolving f , so that E is not exceptional for g1 .
Hence E is the proper transform of a component of a fiber of π1 . The image
Ee of E in X1 is either a smooth rational curve with self-intersection −2, a nodal
or cuspidal rational curve with self-intersection 0, or an elliptic curve. As E is
smooth and rational, E e cannot be an elliptic curve. Moreover, E 2 = −1 as E is
e 2 6= −2. Hence E
exceptional for g2 , so that E e is a rational curve with a singular
double point with self intersection 0. In the sequence of blow-ups to obtain X,
this double point has to be blown-up. But in this case the self-intersection drops
by 4, so E 2 ≤ −4, which is not possible. Hence such an E cannot exists, so that
g2 is an isomorphism. Similarily one proves that g1 is an isomorphism, so f is
an isomorphism and we are done.
Elliptic surfaces 187

Now, let me describe an interesting property of elliptic surfaces. Let C be

a smooth curve, π : X −→ C an elliptic surface. The generic fiber of π is
an elliptic curve, so that if can be written in its Weierstrass form: there are
A, B ∈ C such that the equation of E in P2 is

zy 2 = x3 + Axz 2 + Bz 3 .

Moreover, the pair (A, B) is unique up to the action of C∗ defined as λ(A, B) =

(λ4 A, λ6 B).

Definition 5.7.2. The discriminant of the Weierstrass equation of E is

∆ := 4A3 + 27B 2 .

The Weierstrass coefficients are A and B.

It is an important property of the Weierstrass equation that E is singular if

and only if ∆ = 0.
One can consider the inverse: let C be a smooth curve, A, B ∈ k(C) be two
non-trivial functions on C, and let U ⊆ C be the open subset of C where A and
B are defined. Then we can consider

X := {((x : y : z), t) ∈ P2 × U | zy 2 = x3 + A(t)xz 2 + B(t)z 3 },

which has a ntural projection π : X −→ C. The fiber π −1 (t0 ) on t0 ∈ C is the

the curve
zy 2 = x3 + A(t0 )xz 2 + B(t0 )z 3 ,

which is a smooth elliptic curve for the generic choice of t0 . Hence π : X −→ Cis
an elliptic surface. Notice that π has a natural section

s : C −→ X, s(t) := ((0 : 1 : 0), t) ∈ X,

and the singular fibers are only over those t ∈ C such that ∆(t) = 0. This can
only be then of the form I1 or II, as they are only nodal or cuspidal rational
curves. Let me define the following:

Definition 5.7.3. Let π : X −→ C be an elliptic surface. We say that it is a

Weierstrass fibration if π has a section S and every fiber of π is either of type
I0 , I1 or II.

If π has a section S, then S · F = 1 for every fiber F . Notice that every

singular primitive fiber has at least one component of multiplicity 1, so that
the section has to intersect it on that component, and only in it. Moreover,
the existence of a section forces the elliptic surface to have no multiple fibers.
Moreover, starting from every elliptic surface with a section, one can produce
a Weierstrass fibration simply contracting every component of the fibers which
188 Further properties and classifications

do not intersect the section. Contracting curves one may get singularities. An
important result is to show that every Weierstrass fibration has a Weierstrass
equation. To do that, consider the following: let π : X −→ C be a Weierstrass
fibration with section S, and consider the exact sequence

0 −→ OX −→ OX (S) −→ NS/X −→ 0.

As NS/X is supported on S and the fibers of π restricted to S are of dimension

0, we get that R1 π∗ NS/X = 0. Moreover (π∗ OX (nS))p ' H 0 (Xp , OXp (nS)) and
(R1 π∗ OX (nS))p ' H 1 (Xp , OXp (nS)), where Xp := π −1 (p). By Riemann-Roch
on an elliptic curve, we see that π∗ OX ' OC and R1 π∗ OX are line bundle;
moreover, for every n ≥ 1 the sheaf π∗ OX (nS) is locally free of rank n and
R1 π∗ OX (nS) = 0. Hence we get that the natural map

π∗ NS/X −→ R1 π∗ OX

is an isomorphism.

Definition 5.7.4. The line bundle L ∈ P ic(C) defined as

L := (π∗ NS/X )−1 ' (R1 π∗ OX )−1

is called the fundamental line bundle of the elliptic surface π : X −→ C.

Now let me define

Definition 5.7.5. A triple (L , A, B) gives Weierstrass data over C if L ∈

P ic(X), A ∈ H 0 (C, L 4 ), B ∈ H 0 (C, L 6 ) and the discriminant section ∆ :=
4A3 + 27B 3 ∈ H 0 (C, L 12 ) is not identically 0.

Theorem 5.7.4. We have the following:

1. Let π : X −→ C be a Weierstrass fibration. Then there are A ∈ H 0 (C, L4 )

and B ∈ H 0 (C, L6 ) such that (L, A, B) are Weierstrass data over C.

2. Let C be a smooth curve and (L , A, B) be Weierstrass data over C. Then

they define a Weierstrass fibration π : X −→ C such that L = L.

3. Any Weierstrass fibration can be seen as a section of a P2 −bundle P over

C, and KX = π ∗ (KC × L).

Proof. We begin by proving the following:

Lemma 5.7.5. Le π : X −→ C be a Weierstrass fibration with a section S,

and let n ∈ N, n ≥ 2. Then there is an exact sequence

0 −→ π∗ OX ((n − 1)S) −→ π∗ OX (nS) −→ L−n −→ 0.

Moreover, we have a splitting

π∗ OX (nS) = OC ⊕ L−2 ⊕ ... ⊕ L−n .

Elliptic surfaces 189

Proof. We have the natural exact sequence

0 −→ OX ((n − 1)S) −→ OX (nS) −→ OS (nS) −→ 0.

If n ≥ 2 we have R1 π∗ OX ((n − 1)S) = 0, so that applying the functor π∗ to this

exact sequence we get the statement, as L−1 ' π∗ NS/X , and NS/X ' OS (S).
The last part of the statement we need to prove that the sequence is split,
and then apply induction. For the splitting of the sequence, this comes from
properties of elliptic curves.

Now, let us begin with the first item, so let π : X −→ C be a Weierstrass

fibration. Fix a sufficiently fine open cover {Ui }i∈I of X trivializing L, and fix a
basis section ei of L|Ui for every i ∈ I. Then e−n i is a basis for L−n
|Ui . Moreover,
let αij be transition functions, i. e. ei = αij ej .
Choose fi ∈ π∗ OX (2S)(Ui ) for every i ∈ I, such that fi projects onto e−2 i ∈
−3
L (Ui ), and choose gi ∈ π∗ OX (3S)(Ui ) which projects onto ei ∈ L (Ui ).
−2 −3

As the generic fiber of π is an elliptic curve, there are a1 , ..., a6 ∈ OC (Ui ) such
that
gi2 = a6 fi3 + a5 fi gi + a4 fi2 + a3 gi + a2 fi + a1 ,

which is an equality of section of π∗ OX (6S)(Ui ).Taking the projections of the

two members to L−6 (Ui ) we get a relation

e−6
i = a6 e−6 −5 −4 −3 −2
i + a5 ei + a4 ei + a3 ei + a2 ei + a1 ,

so that we need a6 = 1. To get a Weierstrass equation for the generic fiber we

need to complete the square in gi , getting a section yi ∈ π∗ OX (3S), and the
completing the cube in fi , getting a section xi ∈ π∗ OX (2S) such that

yi2 = x3i + Ai xi + Bi ,

−2 −3
where Ai , Bi ∈ OC (Ui ). Moreover, we have xi = αij xj and yi = αij yj , so that
4 6 4
we finally get Ai = αij Aj and Bi = αij Bj , so that Ai ei patch together to form
a global section of L4 , and Bi e6i patch together to form a global section of L6 .
Notice that a point of x ∈ X is singular if and only if ∆(π(x)) = 0, and we get
that (L, A, B) are Weiestrass data.
The second item is easier: given any (L , A, B) Weierstrass data over C, the
Weierstrass fibration is simply built patching together local surfaces determined
by local Weierstrass equations

yi2 = x3i + Ai xi + Bi ,

where Ai and Bi are local representant of A and B over some open cover of C.
The third item goes as follows. Let π : X −→ C be a Weierstrass fibration
with section S. By Lemma 5.7.5 there is a splitting π∗ OX (3S) ' OC ⊕L−2 ⊕L3 ,
190 Further properties and classifications

which is a rank 3 vector bundle on C. Let P := P(π∗ OX (3S)), which has a

structure of P−2 vector bundle pP −→ C over C. Clearly, we have a map
f : X −→ P such that p ◦ f = π, induced by the fact that OX (3S) is a line
bundle on X. Moreover, the splitting gives generators of π∗ OX (3S) as 1, x and
y satisfying a Weierstrass equation: as they generate the homogenous ring of
a fiber of π, f is an embedding. An equation of X inside P is the Weierstrass
equation with Weierstrass coefficients A and B.
Now, X is a divisor in P, and its class in P is (p∗ L6 )(3), as the individual
terms of the equation are formally sections of L6 , and the equation is of degree
3. Recall that the canonical class of P is p∗ (OC (KC ) ⊗ L−5 )(−3). Hence

OX (KX ) = OX (KP + X) = p∗ (OC (KC ) ⊗ L),

and we are done.

As a consequence we have the following:

Lemma 5.7.6. Let π : X −→ C be a Weierstrass fibration, and let L be its

fundamental line bundle. Then deg(L) ≥ 0. If deg(L) = 0, then either L4 = 0
or L6 = 0, so that L is torsion of period 1,2,3,4 or 6 in P ic(C). Moreover, X is
a product of C with an elliptic curve if and only if L ' OC . In particular if X is
a product, then H 0 (C, L−1 ) = 1, and if X is not a product then H 0 (C, L−1 ) = 0.

Proof. As we have a Weierstrass fibration, we have Weierstrass data (L, A, B).

Hence ∆ is a non vanishing global section of L12 , so that h0 (C, L) > 0. But
this implies that deg(L) ≥ 0. If deg(L) = 0, the at least one between A and B
is non-zero, hence one between L4 and L6 is trivial.
If L ' OC , then P ' P2 × C, and A, B have to be constant. Hence X is a
product. For the converse, let X = C × E, where E is an elliptic curve. As X
is a Weierstrass fibration, we have a section s : C −→ C × E, i. e. a morphism
f : C −→ E. Consider the automorphism

σf : X −→ X, σf (c, e) := (c, e − f (c)).

Then the section S is carried into the section (c, e0 ), which has trivial normal
bundle as it is a fiber of the projection to E. Hence L ' OC .
The remaining part is clear if X is a product, otherwise it follows from the
fact that deg(L) ≤ 0.

We can finally pass to the classification of elliptic surfaces.

Lemma 5.7.7. Let π : X −→ C be a Weierstrass fibration. Then we have:

1. if X is a product, then q(X) = g(C) + 1;

2. if X is not a product, then q(X) = g(C).

Elliptic surfaces 191

Proof. We need to use the Leray spectral sequence: let E2p,q := H p (C, Rq π∗ OX ).
The Leray spectral sequence gives an exact sequence

0 −→ E21,0 −→ H 1 (X, OX ) −→ E20,1 −→ E22,0 .

Since π∗ OX ' OC we have E22,0 = 0 and dim(E21,0 ) = g(C). Hence the exact
sequence gives
q(X) = g(C) + h0 (C, L−1 ),
so the result follows from Lemma 5.7.6.

Lemma 5.7.8. Let π : X −→ C be a Weierstrass fibration. Then we have:

1. if X is a product, then pg (X) = g(C) + deg(L);

2. if X is not a product, then pg (X) = g(C) + deg(L) − 1.

Proof. We have the following chain of equalities:

pg (X) = h2 (X, OX ) = h0 (X, OX (KX )) = h0 (X, π ∗ OC (KC ) ⊗ L) =

= h0 (C, OC (KC ⊗ L)) = h1 (C, L−1 ) = h0 (C, L−1 ) − χ(L−1 ) =

h0 (C, L−1 ) + deg(L) − 1 + g(C),
and we are done by Lemma 5.7.6.

As a consequence we have

χ(OX ) = deg(L), e(X) = 12deg(L).

Lemma 5.7.9. Let π : X −→ C be a Weierstrass fibration.

1. If g(C) = 0 and deg(L) ≤ 1, then Pm (X) = 0.

2. If g(C) = 0 and deg(L) > 1, then Pm (X) = 1 + m(deg(L) − 2).

3. If g(C) = 1 and deg(L) ≥ 1, then Pm (X) = mdeg(L).

4. If g(C) = 1 and deg(L) = 0, then Pm (X) = 1 if the period of L divides

m, and Pm (X) = 0 otherwise.

5. If g(C) ≥ 2, then Pm (X) = m(2g(C) − 2 + deg(L)) + 1 − g(C).

Proof. We have the following chain of equalities

Pm (X) = h0 (X, OX (mKX )) = h0 (C, OC (mKC ) ⊗ Lm ) =

= χ(OC (mKC ) ⊗ Lm ) + h1 (C, OC (mKC ) ⊗ Lm ) =

m(2g(C) − 2 + deg(L)) + 1 − g(C) + h0 (C, OC ((1 − m)KC ) ⊗ L−m ).
One can then conclude with the statement.
192 Further properties and classifications

Theorem 5.7.10. (Classification of elliptic surfaces). Let π : X −→ C

be an elliptic surface.

1. If g(C) = 0 and deg(L) = 0, then X ' E × P1 , i. e. X is a ruled surface

with q = 1.

2. If g(C) = 0 and deg(L) = 1, then X is a rational surface.

3. If g(C) = 0 and deg(L) = 2, then X is a K3 surface.

4. If g(C) = 0 and deg(L) > 2, then X is a proper elliptic surface, i. e.

κ(X) = 1.

5. If g(C) = 1 and L ' OC , then X is an abelian surface.

6. If g(C) = 1 and L is torsion of period 2, 3, 4 or 6, then X is a bielliptic

surface, and KX has the same order of L.

7. If g(C) = 1 and deg(L) ≥ 1, then X is a proper elliptic surface, i. e.

κ(X) = 1.

8. If g(C) ≥ 2, then X is a proper elliptic surface, i. e. κ(X) = 1.

Proof. This is simply an application of the previous Lemmas.

As a consequence of this, we end up with the following:

Theorem 5.7.11. Let X be a smooth minimal projective surface. Then

1. κ(X) = −∞ if and only if P12 = 0;

2. κ(X) = 0 if and only if P12 = 1;

2
3. κ(X) = 1 if and only if P12 > 1 and KX = 0;
2
4. κ(X) = 2 if and only if P12 > 1 and KX > 0.

Proof. The proof is very easy: we have already proven that κ(X) = −∞ if
and only if P12 (X) = 0. If κ(X) = 1, then we know that X admits an elliptic
fibration, so that by the previous Theorem we have that P12 (X) > 1. Moreover,
2
we know that KX = 0. If κ(X) = 0, then we have already shown that P12 (X) =
1. If P12 (X) = 1, then κ(X) ≥ 0, and κ(X) ≤ 1 by Proposition 4.2.13, so that
κ(X) = 0, 1. But if κ(X) = 1 we have P12 (X) > 1, hence κ(X) = 0. Finally if
2
κ(X) = 2 we have P12 (X) > 1 and KX > 0 by Proposition 4.2.13. If P12 (X) >
2
1, then κ(X) ≥ 1, but if κ(X) = 1 then KX = 0, so that κ(X) = 2.

To conclude with this brief account on elliptic surfaces, I just want to mention
that the Weierstrass equation of an elliptic surface is a very important tool for
studying and classifying it. Recall that any Weierstrass fibration πX −→ C can
Elliptic surfaces 193

be seen as a divisor in some P−2 −bundle on a smooth curve C, whose equation

is
Y 2 Z = X 3 + AXZ 2 + BZ 3 ,

where A and B for Weierstrass data over C, together with L. If one considers
a sufficiently fine open cover {Ui }i∈I of C, one can see it as an equation y =
x3 + Ax + B, where A, B are functions on some open disc. A point x ∈ X
is singular if and only if the discriminant ∆(π(x)) = 0, and the type of the
singularity is completely determined by A, B and ∆. In order to resolve the
singularity, we need to blow-up. If x is not singular for X, then we do not
need to blow-up, and the fiber can either be a nodal rational curve (i. e. a
fiber of type I1 ) or a cuspidal rational curve (i. e. a fiber of type II). If x is
singular, we need to blow-up, and the resulting fiber is completely determined
by the singularity, hence by A, B and ∆. As instance, if the fiber has a nodal
singularity which we can solve by blowing up only once, the fiber is of type I2 .
Let me now introduce some notations: let c := π(x), and let a be the order
of vanishing of A at c, b the order of vanishing of B at c, δ the order of vanishing
of ∆ at c. Moreover, let e be the topological Euler characteristic of π −1 (c), r be
the number of the components of π −1 (c) not meeting the section and d be the
number of the components of π −1 (c) with multiplicity 1. The following is just
an application of blow-ups, and the proof is then omitted:

Theorem 5.7.12. (Tate’s Algorithm). Let π : X −→ C, and consider

the notations introduced above for some point c ∈ C. Then we have only the
following possibilites:

1. The fiber is of type I0 if and only if δ = 0 and either a ≥ 0 and b = 0, or

a = 0 and b ≥ 0. In this case e = 0, r = 0 and d = 1.

2. The fiber is of type In for n > 0 if and only if a = b = 0 and δ = n. In

this case e = n, r = n − 1 and d = n.

3. The fiber is of type I0∗ if and only if δ = 6 and either a = 2 and b ≥ 3, or

a ≥ 3 and b = 3. In this case e = 6, r = 4 and d = 4.

4. The fiber is of type In∗ for n > 0 if and only if a = 2, b = 3 and δ = n + 6.

In this case e = n + 6, r = n + 4 and d = 4.

5. The fiber is of type II if and only if a ≥ 1, b = 1 and δ = 2. In this case

e = 2, r = 0 and d = 1.

6. The fiber is of type III if and only if a = 1, b ≥ 2 and δ = 3. In this case

e = 3, r = 1 and d = 2.

7. The fiber is of type IV if and only if a ≥ 2, b = 2 and δ = 4. In this case

e = 4, r = 2 and d = 3.
194 Further properties and classification

8. The fiber is of type IV ∗ if and only if a ≥ 3, b = 4 and δ = 8. In this case

e = 8, r = 6 and d = 3.

9. The fiber is of type III ∗ if and only if a = 3, b ≥ 5 and δ = 9. In this

case e = 9, r = 7 and d = 2.

10. The fiber is of type II ∗ if and only if a ≥ 4, b = 5 and δ = 10. In this

case e = 10, r = 8 and d = 1.

In particular, in every case we have e = δ and 0 ≤ e − r ≤ 2, where e = r if

and only if the fiber is of type I0 , e = r + 1 if and only if the fiber is of type In
for n > 0, and e = r + 2 in all other cases.

An important consequence is the following: let W be the set of points of C

such that π −1 (c) is singular. It is a finite set, possibly empty, and we have
X
e(X) = e(π −1 (C \ W )) + e(π −1 (c)) =
c∈W
X X
= e(F )e(C \ W ) + e(π −1 (c)) = e(π −1 (c)),
c∈W c∈W

where F is the fiber over a point of C \ W , which is an elliptic curve, and hence
e(F ) = 0. Hence e(X) (and then deg(L)) is completely determined by the type
of the fibers, i. e. by a, b and δ by the Tate’s algorithm. Using now Theorem
5.7.10, we are able to classify an elliptic surface by knowing its singular fibers.

5.8 Surfaces of general type

The last remaining class of surfaces in the Enriques-Kodaira classification is
the one of surfaces of general type. These form a very wild famil which is still
to investigate, even if there are several interesting results and examples. The
variety of examples renders quite impossible a further classification of surfaces
of general type, but still we have different methods of investigation. A first one
is to determine which possible invariants surfaces of general type can have, and
in this directions we have several results; another way is to study pluricanonical
maps and canonical models of surfaces of general type.
A first result is the following:
2
Proposition 5.8.1. Let X be a minimal surface of general type. Then KX >0
and e(X) > 0.

Proof. Let H be a hyperplane section on X, and consider the exact sequence

0 −→ OX (nKX − H) −→ OX (nKX ) −→ OH (nKH ) −→ 0.

As κ(X) = 2, tehere is α ∈ R>0 such that h0 (X, OX (nKX )) > cn2 for n  0,
whereas h0 (H, OH (nKX )) grows linearily with n at most. Hence there is n0  0
Surfaces of general type 195

such that H 0 (X, OX (n0 KX − H)) 6= ∅, so that there is an effective divisor D

linearily equivalent to n0 KX − H. Then KX · D ≥ 0 as KX is nef, so that

2
n0 KX = n0 KX ·(D +H) = n0 KX ·D +n0 KX ·H ≥ n0 KX ·H 2 +H ·D ≥ H 2 > 0.

The proof of the remaining part is omitted.

Another important result is the following:

Theorem 5.8.2. (Noether’s Inequality). Let X be a minimal surface of

general type. Then
K2
pg (X) ≤ X + 2.
2
Proof. As X is a minimal surface of general type, by Proposition 5.8.1 we have
2
KX ≥ 1, so that pg (X) ≥ 3. Let |KX | = |C| + V , where V is the fixed part, and
|C| is the moving part. Then either |C| is composed with a pencil of curves,
either its general member is irreducible.
Let us suppose we have a pencil, i. e. there is a 1−dimensional family of
curves whose generic memeber F is irreducible, such that there is n ∈ N such
that KX = nF + V , and we have pg (X) = dim|C| + 1 ≤ n + 1. If KX · F ≤ 1,
then
1 ≥ KX · F = (nF + V ) · F = nF 2 + F · V,

so that F 2 ≤ 1. As F 2 ≥ 0 and KX · F ≡2 F 2 , we have only F 2 = KX · F = 1

or F 2 = KX · F = 0. In the first case we get n = 1, so that pg ≤, whereas we
assumed pg (X) ≥ 3, so that we need F 2 = KX · F = 0. It is easy to see that
this implies that F is a (−2)−curve, which is not possible, so that |C| cannot
have a pencil.
Hence C 2 > 0, and take an irreducible member C. The exact sequence

0 −→ OX (KX ) −→ OX (KX + C) −→ OC (KX + C) −→ 0

gives
h0 (C, OC (KC )) = h0 (X, OX (KX + C)) − pg (X) + q(X) =
1 1 2
= 1 + (C 2 + C · KX ) ≤ KX + 1.
2 2
Then

h0 (C, OX (KX + C)) = h0 (C, OX (2C + V )) ≥ h0 (C, OX (2C)) ≥

≥ 2h0 (X, OX (C)) + 1 ≥ 2h0 (X, OX (C)) − 3 =

= 2h0 (X, OX (C + V )) − 3 = 2pg (X) − 3.

In conclusion we get the statement.

As a corollary we have the following:

196 Further properties and classification

Corollary 5.8.3. Let X be a minimal surface of general type.

2 2
1. If KX is even, then 5KX ≥ e(X) − 36, and if the equality holds we have
q(X) = 0.
2 2
2. If KX is odd, then 5KX ≥ e(X) − 30, and if the equality holds we have
q(X) = 0.
2
Proof. Suppose KX to be even. Then

1 2
χ(OX ) ≤ pg (X) + 1 ≤ K + 3.
2 X
2
By Noether’s Formuls we then get the statement. Similarily for KX odd.

Another important result, whose proof is omitted, is the following:

Theorem 5.8.4. (Bogomolov-Miyaoka). Let X be a minimal surface of

general type. Then
2
KX ≤ 3e(X).

Let us now look at pluricanonical maps, and consider the following:

Definition 5.8.1. Let X be a surface, and L ∈ P ic(X). The rational mor-

phism ϕL is a C-isomorphism if

1. the linear system |L | has no fixed components;

2. ϕL (X) is a normal surface with at most a finite number of rational double

points as singularities.

3. the curves Ci contracted by ϕL are the (−2)−curves on X.

4. the morphism ϕL is an isomorphism outside the Ci ’s.

We have the following basic result:

Theorem 5.8.5. Let X be a minimal surface of general type, and let fn :=

ϕnKX . Then fn is a C−isomorphism if n ≥ 5. In particular, the canonical ring
R(X) := n H 0 (X, OX (nKX )) is a finitely generated Noetherian ring.
L

Notice then that every surface of general type has to be projective. The
most important consequence of this fact is that one can define the canonical
model of every minimal surface of general type X, which is

Xcan := P roj(R(X)),

which is a normal projective surface having at most a finite number of singular

points, all of which are rational double points.
Let me now describe some example of surfaces of general type:
Surfaces of general type 197

Example 5.8.1. In Pn+2 take Hd1 , ..., Hdn general hypersurfaces of whose degrees
Pn
are d1 , ..., dn . Let X be the complete intersection of the Hdi ’s. If i=1 di > n+3,
then X is a surface of general type.
Example 5.8.2. Let C1 and C2 be two smooth curves, g(Ci ) ≥ 2. Then X =
C1 × C2 is of general type. If G is a finite group of order (g(C1 ) − 1)(g(C2 ) − 1)
acting freely on C1 and C2 , the quotient Y := X/G is a surface of general type.
This is called Beauville surface.
Example 5.8.3. If X is a surface of general type and f : Y −→ X is surjective,
where Y is a surface, then Y is of general type.
Example 5.8.4. Let S ⊆ P3 be the quintic of equation X 5 + Y 5 + Z 5 + T 5 = 0.
Let ζ be a primitive root of 1, ζ 6= 1, and consider

σ : S −→ S, σ(X : Y : Z : T ) := (X : ζY : ζ 2 Z : ζ 3 T ),

which is an automorphism on S of order 5. The group G generated by σ acts

freely on S, and the quotient X := S/G is smooth. As q(S) = 0 and KS = H,
we have pg (S) = 4 and KS2 = 5. Then q(X) = 0,

1
χ(OX ) = χ(OS ) = 1,
5
so that pg (X) = 0. Moreover we have

2 1 2
KX = K = 1,
5 S
hence X is a surface of general type with pg (X) = q(X) = 0. This is called
Godeaux surface.
198 Further properties and classification
Chapter 6

Generalizations of the
Enriques-Castelnuovo
Classification

This last chapter can be viewed more as an appendix than a true chapter. In the
previous chapters we have proved the existence of a beautiful classification of
smooth projective complex surfaces, which is called the Enriques-Castelnuovo
Birational Classification. The main problem is if it is possible to generalize
this classification theorem, i. e. to remove hypothesis on the objects we are
classifying.
The first generalization is to consider projective complex surfaces, without
any informations about singularities. Next, we want to classify complex sur-
faces, i. e. we get rid of the hypothesis of projectivity: this has been done
by Kodaira, getting the so-called Enriques-Kodaira Birational Classification of
surfaces. Another important step is to consider surfaces, i. e. not necessarily
defined over the field C: this has been done by Mumford and Kodaira, who pro-
vided a classification of surfaces defined over an arbitrary algebraicailly closed
field k.
A final step is to consider a classification of higher dimensional analogues of
surfaces, i. e. varieties or manifolds. This starts the nowadays called MMP, or
Minimal Model Program, which is still open at several steps.
In this brief chapter I just want to give the basic results without any proof.

6.1 Singular projective surfaces

The first hypothesis we need to get rid of, is the smoothness of our surfaces. The
result we need is the following, which was first proven by Hironaka in complete

199
200 Generalizations of the Enriques-Castelnuovo Classification

generality:

Theorem 6.1.1. (Hironaka). Let X be a projective variety defined over an al-

gebraically closed field of characteristic 0. Then there exists a smooth projective
variety X
e together with a birational proper surjective morphism

e −→ X,
π:X

which is an isomorphism over X \ Xsing .

In the case of surface, this is quite explicit: if X is a projective surface,

consider X0 its normalization, and ν : X0 −→ X, the normalization map, which
is birational. Now, as X0 is normal, its singular locus is given by a finite number
of points. Hence we can blow them up, getting a new surface π1 : X1 −→ X0 .
If X1 is smooth, then we are done. If X1 is not smooth, we can blow it up in
its singular points. We then get a sequence
πn+1 n π πn−1
2 π
1 π ν
... −→ Xn −→ Xn−1 −→ ... −→ X1 −→ X0 −→ X.

Zariski showed that this sequence must end, i. e. there is some n0 in this
sequence such that Xn0 is smooth. If for every n < n0 the surface Xn is still
singular, we put X
e := Xn . In conclusion, every projective surface is birational
0

to a smooth projective one, and as our classification is up to birationality, we

are done.
Moreover, blowing up singular points one adds a finite number of smooth
rational curves, none of which can be a (−1)−curve: the Castelnuovo Contrac-
tion Theorem shows that any (−1)−curve is contracted to a smooth point. If
one supposes that KXe = 0, then by the genus formula we get that every curve
of this sort is a (−2)−curve.

6.2 The Enriques-Kodaira Classification

The next topic is to extend the classification of projective surfaces to all com-
plex surfaces. We have already seen two examples: there are surfaces which
are not Kähler, as the Hopf surface; and there surfaces which are Kähler, but
not projective, as several K3 surfaces and 2−dimensional complex tori. In the
Enriques-Castelnuovo classification appear several other types of surfaces, which
are all projective: ruled surfaces are easily seen to be all projective. Bielliptic
surfaces are all projective by construction, and Enriques surfaces are all projec-
tive by Proposition 5.4.3. Finally, every surface of general type is projective.
Kodaira extended the Enriques-Castelnuovo classification to all complex sur-
faces, getting what is nowaday known as the Enriques-Kodaira Classification.
One of the main difficulties is that for a generic complex surface there is no
Hodge decomposition. We can still define the main objects we need, like line
The Enriques-Kodaira Classification 201

bundles, intersection product, Kodaira dimension, irregularity, plurigenera, bi-

rational morphisms (which is more convenient to call bimeromorphic). But it
is no longer true, for example, that b1 (X) = 2q(X): we have two main cases.
Either b1 (X) is even, or it is odd.

Definition 6.2.1. Let X be a complex surface. We define q(X) := h1 (X, OX )

and h(X) := h0 (X, ΩX ). We define b+ (X) to be the number of positive eigen-
values of the intersection form on H 2 (X, R), and b− (X) to be the number of
negative eigenvalues.

The first result we need is the following:

Proposition 6.2.1. Let X be a compact complex surface. Then one of the two
possibilities is verified:

1. we have that b1 (X) is even, and in this case q(X) = h(X) = 21 b1 (X) and
b+ (X) = 2pg (X) + 1;

2. we have that b1 (X) is odd, and in this case b1 (X) = 2q(X) − 1, h(X) =
q(X) − 1 and b+ (X) = 2pg (X).

Proof. We just give a sketch for of the proof. One can prove that we always
have
1 2
b+ (X) − b− (X) = (KX − 2e(X)).
3
Comparing with Noether’s Formula, which is true for every complex surface, we
have

(4 − 4q(X) + 4pg (X)) + (b+ (X) − b− (X)) = e(X) = 2 − 2b1 (X) + b2 (X).

As b2 (X) = b+ (X) + b− (X), we then get

(b+ (X) − 2pg (X)) + (2q(X) − b1 (X)) = 1.

The two parenthesis of the first member are positive: indeed, for the first we have
that H 0 (X, OX (KX )) ⊕ H 0 (X, OX (KX )), which has dimension 2pg (X), injects
by the de Rham cohomology, into H 2 (X, C), which has dimension b2 (X). But
one easily verifies that the restriction of the intersectionn form to the real part
of this injection is positive, so that b+ (X) ≥ pg (X). For the second part, we
have an exact sequence of sheaves

0 −→ C −→ OX −→ ZΩX −→ 0,

where ZΩX is the sheaf of closed holomorphic 1−forms. As every global 1−form
is closed, we get an exact sequence

0 −→ H 0 (X, ΩX ) −→ H 1 (X, C) −→ H 1 (X, OX ),

202 Generalizations of the Enriques-Castelnuovo Classification

so that b1 (X) ≤ q(X) + h(X). Again, H 0 (X, ΩX ) ⊕ H 0 (X, ΩX ) injects into

H 1 (X, C), so that 2h(X) ≤ b1 (X). Hence h(X) ≤ q(X) and

2h(X) ≤ b1 (X) ≤ h(X) + q(X) ≤ 2q(X),

which implies that 2q(X) − b1 (X) ≥ 0.

We have now only two possibilities: either b+ (X) = 2pg (X), so that b1 (X) =
2q(X) − 1, getting then h(X) = q(X) − 1; or 2q(X) = b1 (X), so that h(X) =
q(X) and b+ (X) = 2pg (X) + 1.

For the first case, we have this important result:

Theorem 6.2.2. Let X be a complex surface. If b1 (X) is even, then X is

Kähler.

Hence we can apply Hodge Theory, and almost every step in the classification
goes through to this generalization. Hence we have a classification of Kähler
surfaces, which is the same as the Enriques-Castelnuovo one.
The remaining case is then when b1 (X) is odd. Let me start with the fol-
lowing definition:

Definition 6.2.2. A complex surface X is called surface of Kodaira type V II

if b1 (X) = 1 and κ(X) = −∞. A minimial surface of Kodaira type VII is
sometimes called surface of Kodaira type V II0 .

Here is one of the main steps in the Enriques-Kodaira classification:

Theorem 6.2.3. (Kodaira). Let X be a minimal comple surface of Kodaira

dimension −∞. Then either X is projective, or it is a surface of Kodaira type
V II0 .

We have several examples of surfaces of Kodaira type VII. The first one is
given by the Hopf surfaces. We have already seen an example of Hopf surface:
consider the surface Y obtained as the quotient of C2 under the action of the
homothety z 7→ 2z. One can generalize this contruction taking the quotient not
by z 7→ 2z but by the map (z1 , z2 ) 7→ (α1 z1 , α2 z2 ), where 0 < |α1 | ≤ |α2 | < 1.
Such a surface is denoted Hα1 ,α2 , and the original Hopf surface is H1/2,1/2 . The
more general definition is the following:

Definition 6.2.3. A Hopf surface is a compact complex surface X whose uni-

versal cover is C2 \ {(0, 0)}.

If X is a Hopf surface, then one can show that b1 (X) = 1, so that q(X) = 1;
moreover pg (X) = 0 and b2 (X) = 0. In general, a Hopf surface is not of the form
Hα1 ,α2 . Finally, every Hopf surface has Kodaira dimension −∞. Moreover, one
can prove that on every Hopf surface there is always a curve.
The Enriques-Kodaira Classification 203

Another family of examples is the following. Let M = [mij ] ∈ SL(3, Z) be a

matrix having one real positive eigenvalue α ∈ R>0 , and two complex conjugate
eigenvalues β, β such that β 6= β. A typical example is the matrix
 
0 1 0
Mn :=  n 0 1 ,
 

1 1−n 0

where n ∈ Z. Let now (a1 , a2 , a3 ) be a real eigenvectors corresponding to α,

and (b1 , b2 , b3 ) be an eigenvector corresponding to β. As (a1 , a2 , a3 ), (b1 , b2 , b3 )
and (b1 , b2 , b3 ) are linearily independent over C, the vectors ab11 , ab22 and ab33

are linearily independent over R. Now, let

H := {z ∈ C | Im(z) > 0},

and let
g0 : H × C −→ H × C, g0 (z, w) := (αz, βw),

and for every i = 1, 2, 3 let

gi : H × C −→ H × C, g0 (z, w) := (z + ai , w + bi ).

Let GM be the group of automorphism of H × C generated by g0 , g1 , g2 , g3 . One

can show that the action of GM on H × C is free, proper and discontinuous, and
the quotient XM := H × C/GM is a compact complex surface. We can even
consider the group generated by the same gi , where instead of β and bi one uses
β and bi .

Definition 6.2.4. The surface XM is called Inoue surface associated to M .

If X is an Inoue surface, then one can show that b1 (X) = 1, so that q(X) = 1;
moreover pg (X) = 0 and b2 (X) = 0. Any Inoue surface has Kodaira dimension
−∞, and one can show that on an Inoue surface there are no curves.

Theorem 6.2.4. (Bogomolov). Let X be a minimal surface of type V II0 . If

b2 (X) = 0, then X is either a Hopf surface or an Inoue surface.

For b2 (X) > 0, the situation is much less clear: we have several examples,
like Inoue-Hirzebruch, Enoki surfaces and Kato surfaces. This last class of
surfaces covers even Inoue-Hirzebruch surfaces and Enoki surfaces, and we have
the following conjecture:

Conjecture 6.2.5. Let X be a minimal surface of type V II0 . If b2 (X) > 0,

then X is a Kato surface.

The proof of this conjecture would conclude the classification of comples

surfaces of Kodaira dimension −∞. For non-negative Kodaira dimension, the
situation is the following: we have already seen that if X is a complex surface
204 Generalizations of the Enriques-Castelnuovo Classification

such that κ(X) = 2, then X is projective, and if κ(X) = 1 then X carries an

elliptic fibration. It then remains to study the case of Kodaira dimension 0.

Definition 6.2.5. A complex surface X is called surface of primary Kodaira

type if b1 (X) = 3, and X amits an elliptic fibration over an elliptic curve. It
is called surface of secondary Kodaira type if it admits a surface of primary
Kodaira type as a proper unramified cover.

Every surface X of primary Kodaira type has trivial canonical bundle. The
invariants are

q(X) = 2 h(X) = 1, pg (X) = 1, χ(OX ) = 0,

e(X) = 0, b1 (X) = 3, b2 (X) = 4.

Every surface X of secondary Kodaira type is the quotient of a primary Kodaira

type surface by the action of a finite group, and it is possible to show that the
possible orders of such a group are 2, 3, 4 or 6. In any case, the canonical bundle
of X is not trivial, but has period 2, 3, 4 or 6. Moreover, X admits an elliptic
fibration over P1 . The invariants are

q(X) = 1 h(X) = 0, pg (X) = 0, χ(OX ) = 0,

e(X) = 0, b1 (X) = 1, b2 (X) = 0.

We have the following result:

Theorem 6.2.6. (Kodaira). Let X be a minimal surface of Kodaira dimension

0. Then either X is Kähler, or it is of primary or secondary Kodaira type.

Finally, the complete statement of the classification is:

Theorem 6.2.7. (Enriques-Kodaira Classification). Let X be a minimal

complex surface. Then X is one of the following:

1. Kodaira dimension −∞:

• rational surface (all projective);

• non-rational geometrically ruled surface (all projective);
• surface of Kodaira type V II0 (all non-Kähler);

2. Kodaira dimension 0:

• Enriques surface (all projective);

• bielliptic surface (all projective);
• K3 surface (all Kähler);
• 2−dimensional complex torus (all Kähler);
Characteristic p 205

• surface of primary Kodaira type (all non-Kähler)

• surface of secondary Kodaira type (all non-Kähler)

3. Kodaira dimension 1: proper elliptic surface;

4. Kodaira dimension 2: surface of general type (all projective).

6.3 Characteristic p
Zariski was the first to study surfaces defined over algebraically closed fields of
any characteristic. Anyway, the first attempt in the classification were made by
Mumford, at the end of the 1960s. The classification was then completed by
Mumford and Bombieri in the late 1970s. To accomplish such a classification
one needs all the machinery of scheme theory: let k be an algebraically closed
field.

Definition 6.3.1. A k−surface is a a 2−dimensional smooth projective scheme

defined over the field k.

All the machinery we used in the Enriques-Castelnuovo classification can be

extended to this more general setting, and all the definitions remains: hence we
can talk about Kodaira dimension, ruled surfaces, elliptic fibrations, K3 surfaces
and so on.
If X is a k−surface of Kodaira dimension −∞, one can show that all the
results we had in characteristic 0 are true even in characteristic p > 0. Hence
the classification of k−surfaces of Kodaira dimension −∞ is equal to the corre-
sponding one over C.
Anyway, things are different once we move to Kodaira dimension 0. Again,
one has K3 surfaces, abelian surfaces, Enriques surfaces and bielliptic surfacea,
but there are some differences.
For K3 surfaces, every property of K3 surfaces which is true over C is true
even over any k algebraically closed. Moreover, Deligne shows that every K3
surface defined over an algebraically closed field k of characteristic p is obtained
from a K3 surface defined over a field of characteristic 0 via reduction modulo
p. Even for abelian surfaces the situation is the same as the one we have over
C.
For Enriques surfaces, the situation is more complicated: if char(k) 6= 2,
these are all quotient of a K3 surface by an involution without fixed points, and
they all carry an elliptic fibration. If char(k) = 2, this is no longer true: the
canonical double cover is inseparable, so it is no longer a K3 surface. Moreover,
they are not necessarily elliptic:
206 Generalizations of the Enriques-Castelnuovo Classification

Definition 6.3.2. A k−surface X is quasi-elliptic if it admits a proper surjec-

tive morphism π : X −→ C, where C is a smooth k−curve, and the generic
fiber of π is a cuspidal cubic.

One can show that every Enriques surface defined over an algebraically closed
field k of characteristic 2 is either elliptic or quasi-elliptic.
For bielliptic surfaces, the situation is the same as over C if char(k) 6= 2, 3.
If char(k) = 3, then we obtain all the possibilities of Theorem 5.5.5 but number
6 (i. e. G = (Z/3Z)2 ). If char(k) = 2, we obtain all the possibilites of Theorem
5.5.5, but number 4 (i. e. G = (Z/4Z) ⊕ (Z/2Z)). Moreover, for number 2 (i.
e. G = (Z/2Z)2 ) we have to write G = (Z/2Z) ⊕ µ2 , where µ2 is the group of
square roots of 1 in k.
If char(k) 6= 2, 3, then this is all. Otherwise we have other types of surfaces:

Definition 6.3.3. A quasi-bielliptic surface is a surface X = (E × C)/G, where

E is an elliptic curve, C is a cuspidal cubic and G is a group of translations on
E acting freely on C.

Any quasi-bielliptic surface has Kodaira dimension 0, and carries a quasi-

elliptic fibration. Moreover, we can classify them like in Theorem 5.5.5, getting
in particular that KX 6= 0, but it is torsion of period 2, 3, 4 or 6. These surfaces
appear only if char(k) = 2, 3.

Definition 6.3.4. A non-classical Enriques surface is a surface X such that

KX = 0 and q(X) = 1.

Non-classical Enriques surfaces appear only if char(k) = 2, and they are of

two types: singular non-classical Enriques surfaces, for which the Frobenius map
on H 1 (X, OX ) is bijective; and supersingular non-classical Enriques surfaces, for
which the Frobenius morphism on H 1 (X, OX ) is zero. The singular non-classical
Enriques surfaces are all quotient of a K3 surface by an involution without fixed
points. The others have a canonical inseparable cover with covering group µ2 .
Moreover, they are all elliptic or quasi-elliptic.
For surfaces of Kodaira dimension 1, it is possible to show that if char(k) 6=
2, 3, then they are all elliptic surface. If char(k) = 2, 3, they are either elliptic
or quasi-elliptic.
Hence, we have the complete classification:

Theorem 6.3.1. (Mumford-Bombieri Classification). Let k be an alge-

braically closed field of any characteristic, and let X be a minimal k−surface.
Then X is one of the following:

1. Kodaira dimension −∞:

• rational surface;
• non-rational geometrically ruled surface;
Higher dimensional classification 207

2. Kodaira dimension 0:

• Enriques surface;
• non-classical Enriques surface (only if char(k) = 2);
• bielliptic surface;
• quasi-bielliptic surface (only if char(k) = 2, 3);
• K3 surface;
• 2−dimensional torus;

3. Kodaira dimension 1:

• proper elliptic surfaces;

• proper quasi-elliptic surfaces (only if char(k) = 2, 3);

4. Kodaira dimension 2: surface of general type.

6.4 Higher dimensional classification

The final step is to pass to a classification of complex projective varieties of
any dimension. In this case one can still use line bundles, curves, Kodaira
dimension an the machinery we introduced before, but there are some important
differencies. First of all, we cannot define an interesection product on the Picard
group: anyway, one can still define an intersection product between line bundles
and curves. If X is our complex projective variety of dimension n, then we can
define
H 2 (X, Z) × H2 (X, Z) −→ Z,

and we can define ample line bundles, nef line bundles and so on. The second,
and more important difference, is that the Castelnuovo Contraction Criterion
is no longer true: blow-up of points are not the only birational transformation
we have to consider.
We have defined a minimal surface as a smooth projective surface having
no (−1)−curves. If the Kodaira dimension is non-negative, this is equivalent
as to ask that the canonical bundle is nef. Hence, a minimal projective variety
is defined as a projective variety whose canonical bundle is nef. Notice that
we didn’t specify anything about smoothness: performing birational transfor-
mations like contraction of curves in this higher dimensional situation, we can
produce singularities, hence we need to take care of singular projective varieties.
Anyway, if X is a projective variety such that KX is not nef, then there
is an effective curve C such that KX · C < 0. One of the main results of the
MMP, which is shown using the Kawamata-Mori Rationality Theorem, is the
following: consider the effective cone Ef f (X), which is defined as for surfaces.
208 Generalizations of the Enriques-Castelnuovo Classification

For every line bundle L ∈ P ic(X) we can define Ef f (X)L ≥0 as the subcone of
effective curves intersecting non-negatively L . Then one can show that there
are a finite number of curves C1 , ..., Cm such that
m
X
Ef f (X) = Ef f (X)KX ≥0 + R≥0 [Ci ].
i=1

This result is known as the Cone Theorem. Hence to produce a nef line bundle
the idea is to contract these finitely many negative curves. Is exactly at this
point that one can produce singularities even starting from smooth varieties. It
is then one of the major efforts in the MMP to find the good type of singularities
to allow in order to let this process run.
If one contracts an extremal curve C on X, there are three possible conditions
that the obtained variety Y verifies:
Case 1 : dim(Y ) < dim(X). In this case we get what is called a Mori-Fano
fiber space. In the case of surfaces, this typically happens if κ(X) = −∞, so
that we have a P1 −fibration over a smooth curve.
Case 2 : dim(Y ) = dim(X), and there is a divisor of X which is contracted
to a point. This is called a divisorial contraction. In this case one produces a
nuew projective varieties with the same type of singularities of X, but whose
Picard number is strictly smaller that the one of X. One can then continue
with the MMP: if KY is nef, then Y is a minimal model for X, otherwise we
start again.
Case 3 : dim(Y ) = dim(X), and the exceptional locus has codimension at
least 2. This is called a small contraction, and the situation is much more
complicated. In this case, the canonical Weil divisor KY is no longer Cartier,
nor even Q−Cartier: indeed, we have KX · C < 0, where C is the contracted
curve, and if KY is Cartier, then KX is the pull-back of KY (as we have a
small contraction, we have no exceptional divisors), so that KX · C = 0. The
solution is the to replace Y by a birational projective variety Y + which has the
same type of singularities as those of Y , but which has KY + Cartier (or more
generally Q−Cartier). The existence of such a Y + , which is called a flip, is
granted by a deep theorem of Birkar-Cascini-Hacon-McKernan, Then, if KY +
is nef, we get a minimal model for X, otherwise we need to start again.
Notice that the first case is completely separated from the others. If one
meets only divisorial contractions, as the Picard number always decreases, we
need to stop at some point, so that we are able to produce a minimal model for
X. If one meets small contractions, the Picard number does not change, and a
priori we can continue without any stop. The main conjecture of the MMP is to
show that we terminate the flip procedure, i. e. that we get a minimal model.
Notice that even in the case this conjecture would be proven true, we are not
yet (an maybe we will never be) in the position to classify projective varieties
of dimension n in a way similar to the Enriques-Castelnuovo Classification.