CMI GRADUATE COURSE NOTES (LAZARSFELD I)

T.R. RAMADAS

By a variety, we will mean (unless otherwise stated) a complex quasi-projective

variety, usually reduced and irreducible, and for the most part normal. We
will work in the Zariski topology and with algebraic functions. Remember
that Zariski open sets are very big – for example, on a compact Riemann
surface (=smooth projective curve) such an open set is the complement of
finitely many points. Any closed set is a finite union of irreducible closed
sets. Given an irreducible closed subset Y , the subset Y reg of its smooth
points is open and dense in Y . If Y has codimension ≥ 2, any regular func-
tion on X \ Y extends across Y – this is an important consequence of the
normality of X; another is that X \ X reg is of codimension ≥ 2. Given
an irreducible codimension one closed subset D, Dreg ∩ X reg is dense in D.
(We can safely take the codimension of an irreducible closed set Y to mean
dim X − dim Y .)
For the reader’s convenience, whenever possible I adopt the notations
from Lazarsfeld’s book.

1. Line Bundles, Cartier/Weil Divisors, Linear Systems

1.1. Line Bundles and Invertible Sheaves. Let X be a variety. Let r
be a positive integer. By a locally free sheaf E on X of rank r, we mean a
coherent (i.e., finitely generated) sheaf of OX -modules which is locally free.
In other words, we are given a sheaf of complex vector spaces
U 7→ E(U )
where
• for each open set U , the complex vector space E(U ) is also a finitely
generated module over the ring of functions OX (U ),
• for U 0 ⊂ U , the restriction maps E(U ) → E(U 0 ) are morphisms of
OX (U )-modules,
• and every point of X has a neighbourhood U such that E(U ) is freely
generated by r elements of E(U ), which (after restriction) also freely
generate E(U 0 ) for every open U 0 ⊂ U . (A more elegant way to say
this is that the r sections on U give a sheaf isomorphism OUr → E|U .
We denote by OUr the direct sum of r copies of OU .)

Date: November 24, 2016.

These notes are supplied on an as-is-where-is basis and will be periodically corrected
and updated without warning.
1
2 T.R. RAMADAS

By a vector bundle on X of rank r, we mean a variety E together with

some additional structures. What follows may not be a minimal character-
isation, but we should keep the complete package in mind. This consists
of
(1) a surjective morphism π : E → X,
(2) a distinguished section X → E (“the zero section”)
(3) a morphism (of varieties) C × E → E (λ, v) 7→ λv which preserves
the fibres of π (i.e., π(λv) = π(v)),
(4) a morphism + : E ×X E → E (with ×X denoting the fibre product
of varieties) which preserves the fibres of π (i.e., π(v + v 0 ) = π(v) =
π(v 0 )),
with the further proviso of local triviality and “freeness”, i.e., such that every
point of X has a neighbourhood U such that E|U ≡ π −1 (U ) is isomorphic to
the product variety U × Cr , in a way compatible with all the structures in
the above list. (For this last compatibility to make sense, one has to check
that the structures given on E localise to every open set U : we leave this as
an exercise.)
Given a vector bundle E, we associate to it the sheaf of modules:

U 7→ E(U ) = sections of E|U

and one can check that E is indeed coherent, locally free, and of rank r. On
the other hand, given a coherent sheaf E, one can ask if there is a variety E
that “represents its sections” as above. This is certainly true if E is locally
free.
We outline the construction in the case that is most important to us
– when the rank r = 1. In this case we will say that E is “invertible”
and E is a “line bundle”. We will signal our shift in emphasis by using the
notation L and L respectively. (By considering line bundles we avoid having
to deal with matrix-valued functions and having to keep track of order of
multiplication.) First a preliminary

Remark: We say a line bundle L is trivial on an open set U if it is isomorphic

to the trivial line bundle U × C. A “trivialisation” is an actual isomorphism
ΦU : U × C → L|U commuting with the projection to U and respecting
the line bundle structures. Giving a trivialisation is equivalent to giving a
(regular) section σ0 of L|U which is nowhere-vanishing. The morphism ΦU
and the section σ0 determine one another via:

Φ(x, z) = σ0 (x), x ∈ U

Suppose then that L is an invertible sheaf. Let us first deal with the case
when L is globally free. That is, there is an element σX ∈ L(X) such that
the maps
OX (U ) → L(U ), f 7→ f σX |U
 LINE BUNDLES 3

are isomorphisms. Consider the trivial line bundle X×C. The corresponding
sheaf of sections is clearly OX , and the sheaf map
OX (U ) → L(U ), f 7→ f σX |U
is an isomorphism. If L is only locally free, there exists an open cover Uα of
X and σα ∈ L(Uα ) such that for each i the map
OUα → L|Uα f 7→ f σα
is an isomorphism of OUα -modules. On the overlaps Uα ∩ Uβ we have
σα = χα,β σβ
where χα,β is a function, regular and nowhere-vanishing on Uα ∩ Uβ . The
collection of functions χα,β satisfy
χα,β χβ,α = 1
χα,β χβ,γ χγ,α = 1
(In other words, w.r.to the cover {Uα } we have a 1-cocycle with values in
the sheaf OX ∗ .)

We now seek a line bundle L̃, trivialised by nowhere-vanishing sections

σ̃α on Uα , such that L̃ is isomorphic to L. One checks that this will be the
case if on overlaps Uα ∩ Uβ ,
σ̃α = χα,β σ̃β
for, in this case,
This tells us how the variety L̃ is to be constructed. Take the disjoint
union
tα Uα × C
and make the identifications
(x, z) ∈ (Uα ∩ Uβ ) × C ⊂ Uα × C = (x, χα,β (x)z) ∈ (Uα ∩ Uβ ) × C ⊂ Uβ × C
This gives us an “abstract variety” and the rest of the vector bundle package
can also be implemented. What is not clear is that this is in fact quasipro-
jective.
Before we go on, we note two important binary operations we can perform
on vector bundles. Given two bundles E and E 0 , we can define their direct
sum E ⊕ E 0 and tensor product E ⊗ E 0 . As a variety, the direct sum can be
identified with the fibre product E ⊗X E 0 ; in terms of the sheaf of sections,
E ⊕ E 0 = E ×X E 0 = E ⊕ E 0
and
E ⊗ E 0 = E ⊗OX E 0

Remark: The tensor product of two coherent sheaves F1 and F2 is defined

as follows. The assignation
U 7→ F1 (U ) ⊗OX F2 (U )
4 T.R. RAMADAS

defines a presheaf. One has to sheafify this to get F1 ⊗OX F2 . This corrects
a statement I made in the lecture.
We can also define the dual Ě of a vector bundle; we have
Ě = HomOX (E, OX )
Under direct sums ranks add, and under tensor products they multiply. In
particular, the tensor product of two line bundles is a line bundle. The
natural map
L ⊗OX Ľ → OX
is an isomorphism, and it is easy to check that the set of isomorphism classes
of line bundles on X is an abelian group. This is why a line bundle is called
an “invertible sheaf”. From now on we will set L−1 = Ľ.

1.2. Weil and Cartier Divisors. We will suppose that X is a (reduced,

irreduciblePand) normal variety. By a Weil divisor on X we mean a formal
sum D = k ak Dk where each Dk is an reducible irreducible subvariety of
codimension 1 and ak ∈ Z. Thus a Weil divisor is an element of the free
abelian group generated by the set of irreducible codimension 1 subvarieties
of X.
By a meromorphic function on X we mean an element of the function field
C(X). (We will denote the algebra of regular functions on X by C[X]; if
X is projective, C[X] = C.) Concretely, this is a function φ defined outside
some closed set Y 0 and regular on X \ Y 0 (and we identify two such if they
agree on a common open subset.) We can take Y 0 to be minimal, in which
case (by normality of X) it is a finite union of irreducible codimension 1
subvarieties Dj0 and φ has a pole of order a0j > 0 along Dj0 . Suppose φ is not
identically zero, and that the zero set Y ⊂ X \ Y 0 is the union of irreducible
subvarieties Di0 (necessarily of codimension 1 by Krull’s Hauptidealsatz),
and the order of vanishing of φ along Di0 is ai > 0. Let Di be the Zariski
closure of Di0 in X. We set
X X
(φ) = ai Di − a0j Dj0
i j

Thus we have associated to each nonzero meromorphic function a Weil di-

visor.
Note that given two nonzero meromorphic functions φ, φ0 , we have (φ ×
φ0 )= (φ) + (φ0 ), and (φ) = (φ0 ) iff the two differ by a nowhere vanishing
regular function. So we have an injective homomorphism
C(X)∗ /{units in C[X]} ,→ {W eil divisors}
The image of this map is (by definition) the subgroup of principal divisors.
Suppose now that L is a line bundle on X. By a meromorphic section
of L, we mean a section that is regular outside some closed Y 0 . Two such
sections are regarded as equal if they agree on a nonempty open subset. We
 LINE BUNDLES 5

will denote by M er(L) the space of such sections. This is a one-dimensional

vector space over C(X).
Remark: As we remarked earlier, a “trivialisation” of a line bundle L on an
open U is equivalent to giving a (regular) section σ0 of L|U which is nowhere-
vanishing. Since we are dealing with irreducible varieties and working in
the Zariski topology, we can regard σ0 as a meromorphic section of L. (In
particular there exist nonzero meromorphic sections!) Conversely, a nonzero
meromorphic section of L defines a trivialisation of L away from its poles
and zeroes.
As before, given a nonzero meromorphic section σ of L, we associate to
it a Weil divisor: X X
(σ) = ai Di − a0j Dj0
i j
(What does it mean to say that σ has a zero or pole along a codimension one
irreducible variety D? At any smooth point of D ∩ X reg – and such points
are Zariski dense in D – choose a local generator of L(U ), that is, a section
σ0 that is regular and nowhere-vanishing in a neighbourhood U ⊂ X. Then
σ = f σ0 for some meromorphic function f . The function f has a zero or
pole along D, and the order of this zero or pole is independent of all choices.)
Suppose given nonzero meromorphic sections σ, σ 0 (respectively) of two
line bundles L, L0 . then σ ⊗ σ 0 is a meromorphic section of L ⊗ L0 . A
meromorphic φ function is nothing but a section of the trivial bundle, so
one can also multiply φ and σ. (Of course one can see this directly.) This
was implicit when we said that M er(L) is a one-dimensional vector space
over the function field C(X). Clearly,
(σ ⊗ σ 0 ) = (σ) + (σ 0 )
(φσ 0 ) = (φ) + (σ 0 )
Given a meromorphic section σ of L, we define σ −1 to be the meromorphic
section of L−1 such that σ −1 [σ] = 1. Here the brackets [ ] mean “evaluate
σ −1 on σ”. When do σ and σ 0 define the same Weil divisor? When σ −1 [σ 0 ]
is a unit in C[X]. In other words, there is an isomorphism L → L0 taking σ
to σ 0 .
Summarising, we have:
Proposition 1.1. The map (L, σ) 7→ (σ) induces an injective homomor-
phism of abelian groups

pairs (L, σ) with L a line bundle and σ a nonzero meromorphic section / ∼
→ {Weil divisors}
where ∼ signals that we take isomorphism classes.
We define a Cartier divisor to be the divisor associated to a nonzero
meromorphic section of a line bundle. This is not the conventional definition
because it works only on normal varieties. We will shortly give another,
6 T.R. RAMADAS

equivalent description, which generalises to the non-normal case. But first

we ask: how do we characterise Cartier divisors among Weil divisors? Let σ
be a meromorphic section of a line bundle L and consider the Cartier divisor
(σ). Suppose given an open set U on which L is trivial. Then there exists a
meromorphic section σ0 which is regular on U and nowhere vanishing there.
Then φ = σσ0−1 is a meromorphic function such that (φ|U ) = (σ|U ) as Weil
divisors on U . Thus every Cartier divisor isPlocally principal.
Conversely, suppose given a Weil divisor i ak Dk which is locally princi-
pal. That is, suppose we can cover X by open sets Uα and find meromorphic
functions φα satisfying
X
(φα |Uα ) = ai {Di ∩ Uα }
i
as Weil divisors on Uα . Then for each pair of indices α, β, we have a mero-
morphic function
χα,β = φ−1α φβ
which is regular and invertible on Uα ∩ Uβ . Further, on Uα ∩ Uβ ∩ Uγ , we
have
χα,β χβ,γ χγ,α = 1
So there exists a line bundle L with meromorphic sections σα such that σα
is regular and nowhere-vanishing on Uα and such that on Uα ∩ Uβ we have
σα = χα,β σβ
For each α define the meromorphic section τα by
τα = φα σα
On Uα ∩ Uβ we have
τα = φα σα = φα χα,β σβ = φβ σβ = τβ
So the the meromorphic sections τα are all equal, and define a meromorphic
section τ of L. Clearly X
(τ ) = ak Dk
k
In conclusion: a Weil divisor is Cartier iff it is locally principal. In
particular, if X is smooth the notions of Cartier and Weil divisors coincide.
More generally, this holds when X is locally factorial, i.e., when all local
rings are UFDs. (Warning: this is an algebraic property, and not verifiable
analytically or at the level of completions.) We can summarise the story so
far with the diagram of inclusions of sets (in fact injective homomorphisms
of abelian groups):
C(X)∗ ∗

C[X]∗ tL∈P ix(X) MC[X]

er(L)
∗

∼ ∼

principal divisors Cartier divisors W eil divisors

 LINE BUNDLES 7

where the disjoint union tL is over the group of isomorphism classes of line
bundles, the Picard group of X. (The Picard variety, which we shall mostly
avoid, is a subtler object.)
The “traditional” definition of Cartier divisor is essentially as isomor-
phism class of pairs (L, σ) with L a line bundle and σ a meromorphic section,
but reformulated to avoid talking of line bundles, as follows.
Consider the constant sheaf M∗ of nonzero meromorphic functions. This
is the sheaf that associates to any (nonempty) open U the abelian group of
nonzero meromorphic functions on X (and associates to the empty set the
final object in the category of abelian groups, namely the singleton group).
This contains as subsheaf the sheaf OX ∗ of units in the sheaf of algebras O .
X
I claim that there is a bijective homomorphism between global sections of
the quotient sheaf M∗ /OX ∗ and isomorphism class of pairs (L, σ) as above.

Suppose first that a pair (L, σ) is given. Cover X by open sets Uα over
which L is trivialised by meromorphic sections σα . (That is, σα is regular
and non-vanishing on Uα .) Then, for each α, there exists a meromorphic
function φα such that
σ = φα σ α
On overlaps Uα ∩ Uβ , the meromorphic functions φα and φβ differ by a
regular invertible function. In other words, the φα define a global section of
the quotient sheaf M∗ /OX ∗ . Changing the local trivialisations σ does not
α
change this global section, which therefore depends only on (the isomorphism
class of) the pair (L, σ).
Conversely, a global section of M∗ /OX ∗ is, by definition, given by an open

cover Uα and meromorphic section functions φα which on overlaps differ by

a regular invertible function χα,β . These functions define a 1-cocycle which
can be used to construct a line bundle L and a meromorphic section σ.
We have described the bijective maps
(1) from isomorphism class of pairs (L, σ) to Cartier (=locally principal
Weil) divisors
(2) between isomorphism class of pairs (L, σ) and global sections of
M∗ /OX ∗ .

We can in fact give a bijection between isomorphism classes of pairs (L, σ)

and Cartier divisors in the language of locally free rank one sheaves (and
bypassing line bundles) as follows. The map to Cartier divisors isPgiven as
before,
P 0 and the converse map as follows: given a Weil divisor D = i ai Di −
a D 0 (with a , a0 positive), define the subsheaf O (D) of the constant
j j j i j X
sheaf MX of meromorphic functions (this latter is a quasi-coherent sheaf)
by: X X
open U ⊂ X 7→ OX ( ai Di − a0j Dj0 )(U )
i j
0 0
P P
where OX ( i ai Di − j aj Dj )(U )
is the space of regular φ functions on
U \ (∪i Di ) and such that
(1) φ has poles of order at most ai along Di , and
8 T.R. RAMADAS

(2) φ vanishes to order at least a0j along Dj0 .

One checks
P that the Psheaf so defined is locally free of rank one if (iff?) the
divisor i ai Di − j a0j Dj0 is Cartier. The constant function with value 1
is a meromorphic section of the corresponding line bundle, and the corre-
sponding divisor the one we started with. If you check this claim you will
see the reason for the apparent reversal of roles between the ai and a0j .
Note that if D, D0 are Cartier,
OX (D) ⊗OX OX (D0 ) = OX (D + D0 )
When do two Cartier divisors D and D0 define the “same” (i.e., isomor-
phic) line bundles? If this is the case, there is a line bundle L and two
meromorphic sections σ, σ 0 such that (σ) = D and (σ 0 ) = D0 . Then σ = φσ 0
for some meromorphic function φ, and D = (φ) + D0 , or to put it another
way:
D − D0 = (φ) is principal
Terminology: We say that two Cartier divisors are linearly equivalent if their
difference is a principal divisor. This is clearly an equivalence relation.
Summarising the above discussion:
(1) P icard group ∼ {Cartier divisors modulo linear equivalence}

Remark: Consider the exact sequence of sheaves

∗
1 → OX → M∗X → M∗X /OX
∗
→1
Noting that H 1 (M∗X ) vanishes because M∗X is flasque1, the long exact se-
quence in cohomology yields the exact sequence of abelian groups:
1 → {C(X)∗ /C[X]∗ } → H 0 (M∗X /OX
∗ ∗
) → H 1 (X, OX )→1
Now, H 0 (M∗X /OX ∗ ) is the group of Cartier divisors, going modulo {C(X)∗ /C[X]∗ }

is rational equivalence, and H 1 (X, OX ∗ ) is the Picard group, so we have re-

covered (1).
Notation/Definition: If D is a Cartier divisor, we mean by OX (D) the in-
vertible sheaf defined above. Less precisely, we will mean a line bundle L,
together with a meromorphic section σ such that (σ) = D. As noted earlier,
the pair (L, σ) is unique up to isomorphism. Even less precisely, we will
mean just the line bundle L. We will often confound L and L and OX (D).
We end by giving the canonical example of a Weil divisor that is not
Cartier. Consider the cone in C3 :
{(x, y, z)|x2 + y 2 = z 2 )}
This is an affine variety of dimension 2. Let D be the dimension 1 subvariety
defined by the ideal generated by x and z − y. One checks that D is reduced
and irreducible, but D is not defined by the vanishing of a single regular
function.
1http://www.math.umn.edu/ garrett/m/algebra/cech.pdf
 LINE BUNDLES 9
P
1.3.
P Line bundles and divisors: some easy remarks. Let D = i ai Di −
a0 D 0 be a Cartier divisor, with a , a0 > 0.
j j j i j
(1) The line bundle O(D) is trivial iff D is a principal divisor.
(2) What are the regular sections of OX (D) on (all of) X? By definition,
these are the meromorphic functions, regular outside ∪i Di and with
poles of order at most ai along Di and vanishing to order at least
order a0j along Dj0 .
(3) If D is an effective
P divisor, that is, a positive linear combination of
divisors (D = i ai Di , with ai > 0), then H 0 (X, O(D)) 6= 0, for
then any constant function is a global section.
(4) On a complete (which in our context means projective) variety, if D
is effective and nonzero, then O(D) is a nontrivial line bundle.
(5) Suppose D is effective as above, and σ ∈ H 0 (X, O(D)) is any nonzero
P
(regular) section, then (σ) = D̃ = ĩ aĩ Dĩ with aĩ > 0 and
X X
aĩ Dĩ − ai Di
ĩ i

is principal. That is, there exists a meromorphic function φ with

polar divisor D̃ and zero divisor D.

1.4. Line bundles on smooth projective curves. Let us see how the
theory plays out on a smooth projective curve X of genus g. Since X
is smooth, Weil divisors are locally principal, and we will drop the prefix
“Cartier”. A divisor is just a formal sum with integer coefficients:
X
D= ax x
x∈X

where the ax are integers and nonzero only for finitely many x. Given a
divisor D, we define its degree by
X
degree D = ax
x∈X

Given a meromorphic function φ, the meromorphic form φ−1 dφ has simple

poles at the points x such that ax 6= 0, and the residue at such an x is ax .
OTOH, the sum of residues of a meromorphic form is zero, which proves
that
degree(φ) = 0
Thus the degree is an integer-valued function on the Picard group (in fact,
a homomorphism onto (why?) Z.)
Clearly, if a line bundle has a nonzero regular section, its degree is non-
negative. If such a line bundle has degree zero, it has a nowhere vanishing
regular section and is therefore trivial.
For curves of arbitrary genus it is difficult to proceed without more ma-
chinery. Let us consider line bundles on the projective line P1 . Fix a point
10 T.R. RAMADAS

x∞ ; its complement is isomorphic to the affine line C. Fixing this isomor-

phism amounts to giving a meromorphic P functionP z with one simple pole at
x∞ . Suppose given a divisor D = i ai xi − j aj xj + a∞ x∞ , where the
xi , xj are distinct from each other and from x∞ , and ai > 0, aj > 0. Set
Y Y
φ= (z − z(xi ))ai (z − z(xj ))−aj
i j
P P P P
Clearly (φ) = i ai xi − j aj xj + ( j aj − i ai )x∞ , which shows that D
is principal iff its degree is zero. We can now conclude that
Picard group of P1 → Z
degree

is an isomorphism. Given l ∈ Z, there is up to isomorphism one line bundle

of degree l, which we can take to be O(lx∞ ). Regular sections of O(lx∞ )
are polynomial functions on C with
(1) a zero of order at least −l at ∞ if l < 0,
(2) regular at ∞ if l = 0, and
(3) a pole of order at most l at ∞ if l > 0,
It follows that H 0 (O(lx∞ )) = 0 if l < 0, H 0 (O(lx∞ )) = C if l = 0, and if
l>0
(2) H 0 (O(lx∞ )) = polynomials of degree ≤ l if l > 0.
In particular if L is a (‘the’) line bundle of degree l,
dim H 0 (P1 , L) = l + 1

2. Linear series, Maps to projective space

Abuse of notation ahead: From now on we shall cease differentiating

between a vector bundle E and the corresponding sheaf of sections E, except
when the occasion seems to call for it.

2.1. Preliminaries on Grassmannians. Given a finite-dimensional vec-

tor space V , we will mean by the projective space P(V ) the variety which
parametrises one-dimensional quotients of V . More generally, consider the
Grassmannian Grk (V ) of k dimensional quotients of V , where k is an inte-
ger, 0 < k < dim V . Any point V → Q → 0 with dim Q = k of Grk (V )
determines an exact sequence of vector spaces 0 → S → V → Q → 0, with
dim Q = k. Two such quotients are identified if there is a commutative
diagram:
0 −−−−→ S −−−−→ V −−−−→ Q −−−−→ 0
  
=y =y ∼y
  

0 −−−−→ S 0 −−−−→ V −−−−→ Q0 −−−−→ 0,

in other words, if the kernels S, S 0 are the same. Thus Grk (V ) can be iden-
tified with the set of (dim V − k)-dimensional subspaces of V . Grothendieck
 LINE BUNDLES 11

adopted the quotient definition (which requires the above fuss with identify-
ing quotents) for good reason, and (following Lazarsfeld) we will stick to it.
On Grk (V ) the vector spaces Q and S together build up tautological vector
bundles Q of rank k and S of rank (dim V −k) as quotient and (respectively)
sub of the trivial bundle with fibre V . In terms of the corresponding locally
free sheaves, we write:
0 → S → V ⊗C OGrk (V ) → Q → 0
Allowing for abuse of notation, the above sequence could be written:
0 → S → V ⊗C OGrk (V ) → Q → 0

Remark: If V is a finite-dimensional vector space, the trivial vector bundle

X × V → X, which we could denote by (say) VX , has the associated locally
free sheaf V ⊗C OX . The latter notation is standard, but there is no standard
notation for the geometric vector bundle. So you can think of the above
sequence as standing in for the exact sequence of vector bundles 0 → S →
VX → Q → 0.
It is a fact that the tangent bundle of Grk (V ) is Hom(S, Q) = Š ⊗C Q.
The variety Grk (V ) has the following important property. Given a rank
r vector bundle QX on X, expressed as a quotient
V ⊗C OX → QX → 0
there is a unique morphism ψ : X → Grk (V ) and an unique isomorphism
ψ ∗ Q → QX which makes the following commutative:
V ⊗C OX −−−−→ ψ ∗ Q −−−−→ 0
 
=y ∼y
 

V ⊗C OX −−−−→ QX −−−−→ 0
Set-theoretically the map ψ is described as follows. Let x ∈ X; the surjective
map of bundles V ⊗C OX → QX → 0 yields the surjective map of vector
spaces:
V → (QX )x → 0
which represents the point ψ(x) ∈ Grk (V ). We will prove below that ψ is a
morphism in the case that interests us, that of projective spaces.

2.2. Projective spaces. Specialise now to k = 1. On P(V ) the tautological

quotient line bundle is conventionally denoted by O(1) (and its dual by
O(−1)). From the above remarks, we see that it sits in an exact sequence
0 → ΩP(V ) ⊗C O(1) → V ⊗C OP(V ) → O(1) → 0
where, for a smooth variety X, ΩX denotes cotangent bundle. (We have
used fact from above.) Note that this gives a formula for the canonical
bundle: κ = (det V ) ⊗C O(−dim V ). It also follows (using the vanishing
12 T.R. RAMADAS

of H 0 (ΩP(V ) ⊗C O(1)) and H 0 (ΩP(V ) ⊗C O(1)), special cases of a vanishing

theorem due to Bott) that the map
V = H 0 (P(V ), V ⊗C OP(V ) ) → H 0 (P(V ), O(1))
is an isomorphism. Of course, the injectivity of the above map is clear (given
a nonzero vector v ∈ V , its image under a surjection V → Q → 0 is nonzero
outside a linear hyperplane), and we will outline below an argument for the
surjectivity.
On occasion, we will need to deal with one-dimensional subspaces of a
vector space. For the moment consider V̌ , the dual of V . The corresponding
variety will be denoted Psub (V̌ ); clearly Psub (V̌ ) = P(V ). This avatar of
projective space is easier to visualise. The one-dimensional subspaces fit
together into a “tautological” line sub-bundle of the trivial vector bundle
with fibre V̌ ; let us denote this line bundle by Õ(−1) and its dual by Õ(1).
The total space of Õ(−1) is the canonical example of a blow-up (of V̌ at the
origin). We have
0 → Õ(−1) → V̌ ⊗C OPsub (V̌ ) → ΘPsub (V̌ ) ⊗ Õ(−1) → 0
where Θ defines the tangent bundle. Comparing exact sequences we see
that under the identification Psub (V̌ ) = P(V ) the line bundles Õ(±1) co-
incide with O(±1). (This gives a direct proof of the surjectivity of V →
H 0 (P(V ), O(1)). For, a section of O(1) gives a regular function on the total
space of O(−1) = Õ(−1), homogeneous of degree 1, which in turn gives a
regular function on V̌ \ {0} homogeneous of degree 1. This has to extend
across the origin, and determines an element of V .)
Homogeneous coordinates: Given a point V → Q → 0 of P(V ), choosing
an ismorphism Q ∼ C, we get a nonzero element v̌ ∈ V̌ . Clearly v̌ spans
the corresponding one-dimensional subspace of Psub (V̌ ). Given any such
nonzero v̌ we say that the homogeneous coordinate of the point is [v̌].
Suppose given a line bundle bundle QX on X, expressed as a quotient
V ⊗C OX → QX → 0
let us prove that the corresponding map ψ : X → P(V ) is a morphism. Let
V = ` ⊕ V 0 where ` is a one-dimensional subspace. This defines an open set
U` in P(V ) such that the composite map
` ⊗C OU` (,→ V ⊗C OU` ) → O(1)
is an isomorphism. (Note that the open set U` depends only on ` and not
on the choice of supplement V 0 .) This gives a map of bundles on U` :
V 0 ⊗C OU` → O(1) ∼ ` ⊗C OU ` .
The structure of variety on P(V ) is defined so that the induced (bijective)
map φ` : U` → Hom(V 0 , `) is an isomorphism. The inverse image of U` by
 LINE BUNDLES 13

ψ is clearly (the open set) Ũ` ⊂ X such that the induced map
` ⊗C OŨ` (,→ V ⊗C OŨ` ) → QX → 0

is an isomorphism. This gives a map of bundles on Ũ` :

V 0 ⊗C OŨ` → QX ∼ ` ⊗C OŨ ` .
which in turn yields a morphism
Ũ` → Hom(V 0 , `)
ψ̃`

Clearly ψ̃` = φ` ◦ ψ|Ũ` , or equivalently

ψ|Ũ` = φ−1
` ◦ ψ̃`

which shows that ψ|Ũ` is a morphism. Since the sets U` cover P(V ) we are
done.
From now on, we will (unless otherwise flagged) consider projective vari-
eties.
If X is projective and L is a line bundle then H 0 (X, L) is a finite-
dimensional vector space. Any nonzero section defines an effective divisor,
and two sections define the same divisor iff one is a nonzero scalar multiple
of the other. This has the following consequence: given a Cartier divisor D
on a projective variety, the set
{D̃|D̃ is effective and linearly equivalent to D}
can naturally be identified with the projective space Psub (H 0 (X, OX (D))) of
one-dimensional subspaces of H 0 (X, OX (D)). (Of course, it could happen
that OX (D) has no nonzero regular sections, in which case |D| is empty. If
D itself is effective, it will represent a point of the projective space.)
By a linear series, we will mean (interchangeably) a nonzero subspace V ⊂
0
H (X, L) or the set |V | of effective divisors corresponding to such a subspace
V . Note that the set |V | can be identified with Psub (V ) ⊂ Psub (H 0 (X, L)).
If V = H 0 (X, L) we will say that the linear series is complete. If D is a
Cartier divisor, we denote by |D| the complete linear series associated to
OX (D).
Given a linear series V , its base locus Bs(|V |) is the (Zariski)-closed set
of points where all the nonzero sections in V vanish. If L = O(D) and
V is the corresponding complete linear series, we will also denote this by
Bs(|L|) = Bs(|D|). The notation signifies that the base locus comes with a
natural scheme structure, and one can in fact talk of the base scheme defined
by the sheaf of ideals which is the image of the sheaf map (“evaluate σ ∈ V
at x ∈ X to get σ(x) ∈ Lx , compose with the map Lx ⊗ L−1 x → C”):

V ⊗C L−1 → OX
We will not deal with the base scheme.
14 T.R. RAMADAS

We say that the linear series |V | is (base-point) free if its base locus is
empty. In other words, |V | is free if for every x ∈ X, there is a section
belonging to V which is nonvanishing at x. Given a line bundle L (or a
divisor D), we say that it is free (“generated by global sections”) if the
corresponding complete series is free.
Outside the base locus B of V we have the evaluation morphism e(“evaluate
σ ∈ V at x ∈ X to get σ(x) ∈ Lx ”):
V ⊗C OX → L → 0
e
This yields a morphism φ|V | : X \ B → P(V ) and a commutative diagram:
V ⊗C OX −−−−→ ψ ∗ O(1) −−−−→ 0
 
=y ∼y
 

V ⊗C OX −−−−→ L −−−−→ 0
Often we think of φ|V | as a rational map φ|V | : X 99K P(V ).
Let us make this explicit. Let dim V = n. Choose a basis of sections
σ1 , . . . , σn ∈ V , and let σ̌i ∈ V̌ be the dual basis. Let Ui ⊂ X be the open
set
Ui = {x ∈ X|σi (x) 6= 0}
On U1 we have regular functions x1 ≡ 1, x2 , . . . , xn such that
X X
e( ai σi ) = ( ai xi )σ1 (x)
i i

which shows that the corresponding map U1 → Psub (V̌ ) is given by

X
x 7→ [ xi σ̌i ]
i
Suppose that X is a normal projective variety, L a line bundle, and
V ⊂ H 0 (X, L) a linear series, For simplicity we take |L| to be base-point
free. How are the morphisms induced by V and the complete linear series
H 0 (X, L) related? Given a one-dimensional quotient H 0 (X, L) → Q → 0,
the composite map V ⊂ H 0 (X, L) → Q will give a point of P(V ) iff this map
is nonzero. So let us identify the set of quotients H 0 (X, L) → Q → 0 such
that the composite map V ⊂ H 0 (X, L) → Q is zero. This happens iff the
dual map Q̌ ⊂ H 0 (X, L)∨ → V̌ is zero. That is iff, under the identification
P(H 0 (X, L)) = Psub (H 0 (X, L)∨ )
the point H 0 (X, L) → Q → 0 belongs to Psub (V ⊥ ), where V ⊥ ⊂ H 0 (X, L)∨
is the annihilator of V . Thus we have a morphism (“projection”)
π : P(H 0 (X, L)) \ Psub (V ⊥ ) → P(V )
and
(1) |V | is free iff Psub (V ⊥ ) ∩ image(φL ) = ∅,
(2) in general Bs(|V |) = φ−1 L (Psub (V ) ∩ image(φL )), and
(3) on X \ Bs(|V |), we have φ|V | = π ◦ φ|L| .
 LINE BUNDLES 15

Let X be a projective variety, and suppose given a map φ : X → P(V )

for some vector space V ; set L ≡ φ∗ O(1). This yields a map
V = H 0 (P(V ), O(1)) → H 0 (X, φ∗ O(1)) = H 0 (X, L)
Since X is projective (and irreducible), its image in P(V ) is a closed (and
irreducible) set and hence a (reduced) subvariety X 0 ⊂ P(V ). In fact, since
X is reduced, the morphism φ factors through X 0 (exercise). We write:
π = ι ◦ π 0 where
φ0 : X → X 0 and ι : X 0 → P(V )
The pull-back map on sections also factors:
V = H 0 (P(V ), O(1)) →
∗
H 0 (X 0 , O(1)|X 0 ) →
0∗
H 0 (X, φ0∗ O(1)|X 0 ) = H 0 (X, L)
ι φ

Since the map φ0 is surjective, the map φ0∗ is injective (locally on X 0 ) on

functions, and therefore on sections of O(1)|X 0 (by local triviality). The map
ι∗ is injective iff X 0 is not contained in any hyperplane in P(V ). (In this
case one sometimes says that X 0 s a nondegenerate subvariety of projective
space.) To summarise the discussion so far:
Let X be a projective variety, and suppose given a map φ : X → P(V ) for
some vector space V , with nondegenerate image. Then V ⊂ H 0 (X, L) is a
base-point free linear series and φ = φ|V | .
For V to be the complete linear series H 0 (X, L), we need both ι∗ and φ∗
to be surjective. More on this later.
Finally, we note the following. If X is nonsingular, B̃ a closed subset
of codimension ≥ 2 and φ : X \ B̃ → P(V ) a morphism whose image is
nondegenerate, then
(1) φ∗ O(1) extends uniquely as a line bundle L on X, and
V = H 0 (P, O(1)) ⊂ H 0 (X \ B̃, O(1)) = H 0 (X, L),
(2) B ≡ Bs(|V |) ⊂ B̃, and φ|V | : X \ B extends φ.

2.3. Example: The projective line as conic. First some preliminary

linear algebra. Given an exact sequence of vector spaces:
0 → V1 → V2 → V3 → 0
then, for any k ≥ 2, the induced map S k V2 → S k V3 is onto, and has kernel
(isomorphic to) V1 ⊗C S k−1 V2 .
Let now W be a two dimensional vector space, and set X = P(W ), the
corresponding projective space. We have the tautological sequence
0 → Ω ⊗OX O(1) → W ⊗C OX → O(1) → 0
This induces:
0 → S k−1 W ⊗C Ω ⊗OX O(1) → S k W ⊗C OX → O(k) → 0
16 T.R. RAMADAS

which yields (using vanishing theorems as we did earlier, or using (2)):

S k W = H 0 (X, O(k)), k ≥ 1
Consider the complete free linear series H 0 (X, O(k)). This yields maps:
φk ≡ φ|O(k)| : X → P(H 0 (P(W ), O(k))) = P(S k W )
In terms of homogeneous coordinates
φk ([w̃]) = [w̃k ]
In the case k = 2, the image is a curve in the projective plane P(S 2 W ). By
definition OP(S 2 W ) (1) pulls back to OX (2), so the image is a curve of degree
2. (We are anticipating some interesection theory, which is yet to come.)
Consider the map B : S 2 W × S 2 W → (det W )2 given by
X X X
B( ui vi , u0j vj0 ) = (ui ∧ u0j )(vi ∧ vj0 ) + (ui ∧ vj0 )(vi ∧ u0j )
i j i,j
This is clearly bilinear and symmetric, an so defines a homogeneous poly-
nomial of degree 2 on S 2 W with values in the line (i.e., one-dimensional
complex vector space) (det W )2 . This can be thought of as a section of
H 0 (P(S 2 W ), OP(S 2 W ) (2) ⊗C (det W )2 ) = S 2 (S 2 W ) ⊗C (det W )2
whose zero set is precisely the image of X under φ2 .

3. Intersection classes; intersection numbers

A comprehensive development of the theory is in Fulton’s book. The
beginner should read the treatment of intersection theory on surfaces in
Hartshorne’s Algebraic Geometry. Both books treat intersection theory as
defining a product on the Chow ring of algebraic cycles (formal linear com-
bination of irreducible subvarieties, modulo rational equivalence). In codi-
mension one, this is the group of Weil divisors modulo principal divisors.
Chern classes (see below) are defined to take values in the Chow ring. For a
nice summary (of the theory on smooth varieties), see Hartshorne’s Appen-
dix. See also 3264 & All That: Intersection Theory in Algebraic Geometry,
Eisenbud and Harris.2)
We follow Lazarsfeld, who has a more topological approach that suffices
for complex algebraic varieties.
Topologically, a line bundle Ltop is classified up to isomorphism by its
Chern class, an element c1 (Ltop ) ∈ H 2 (X, Z). This is best defined in terms
of Čech cohomology. Consider the exact sequence of sheaves in the analytic
(not Zariski) topology:
∗
0 → Z → Ocont → Ocont →0
f 7→e2πif

where Ocont is the (additive) sheaf of continuous complex-valued functions

∗
and Ocont the (mutliplicative) sheaf of invertible continuous complex-valued
2http://scholar.harvard.edu/files/joeharris/files/000-final-3264.pdf
 LINE BUNDLES 17

functions. If X is connected and paracompact (as irreducible varieties cer-

tainly are in the analytic topology), the higher cohomologies of Ocont vanish,
and we get
∗
H 1 (X, Ocont ) ∼ H 2 (X, Z)
The cohomology group on the left classifies isomorphism classes of continu-
ous line bundles Ltop , and c1 (Ltop ) is (up to a sign change) the corresponding
element of H 2 (X, Z).
Instead of working with continuous functions, let us work with the sheaf
of analytic functions, again in the analytic topology:
∗
0 → Z → Oan → Oan →0
f 7→e2πif

This yields:
H 1 (X, Oan ) ∗
{1} → → H 1 (X, Oan ) → H 2 (X, Z) → {1}
H 1 (X, Z)
∗ ) classifies isomorphism classes of holo-
The cohomology group H 1 (X, Oan
morphic line bundles. The map H 1 (X, Oan ∗ ) → H 2 (X, Z) clearly factors

as:
∗ ∗
H 1 (X, Oan ) → H 1 (X, Ocont ) → H 2 (X, Z)
which shows that topologically trivial line bundles are parametrised by (the
Jacobian)
H 1 (X, Oan )
H 1 (X, Z)
Finally, on an algebraic variety, algebraic line bundles are classified by
H 1 (X, O∗ ) where now O∗ is the Zariski sheaf of invertible regular functions.
It is not entirely obvious how to compare the cohomology groups H 1 (X, O∗ )
and H 1 (X, Oan∗ ), but GAGA assures that on a projective variety analytic and

algebraic vector bundles “are the same”. To summarise, to an (algebraic)

line bundle on a projective variety, one can associate the Chern class c1 (L) ∈
H 2 (X, Z).
We have, given line bundles L, L0 ,
c1 (L ⊗C L0 ) = c1 (L) ∧ c1 (L0 )
where the product is taken in the cohomology ring. In fact, these (and
higher Chern classes which are relevant for vector bundles of arbitrary rank)
are even degreePclasses, and products are taken in the commutative ring
∗
Heven (X, Z) ≡ i H 2i (X, Z).
Not every topological line bundle admits an algebraic structure. If X
is smooth and projective, the Lefshetz Theorem on (1, 1)-classes gives a
necessary and sufficient condition: given an algebraic L the image of c1 (L)
in H 2 (X, R) ⊂ H 2 (X, C) = H (2,0) (X, C) ⊕ H (1,1) (X, C) ⊕ H (0,2) (X, C) lies
in the factor H (1,1) (X, C), and every such (integral) class arises this way.
From now on, varieties are projective unless otherwise flagged.
18 T.R. RAMADAS

Let Y be a complex projective variety of dimension k. “By a theorem

of Lojasiewicz [1964] (see Hironaka [1975]), Y admits a finite triangulation
in which the singular locus is a subcomplex. Since the singularities of Y
occur in real codimension 2, the sum of the simplices of dimension 2k in
the triangulation is a cycle, called the fundamental cycle of Y ; the class
of this cycle is called the fundamental class of Y and is again denoted by
[Y ] ∈ H2k (Y ; Z)” (from 3264 & All That: Intersection Theory in Algebraic
Geometry.)
Having finished with our quote, let us revert to considering a projective
variety X. Given a subvariety V , consider the image of the fundamental class
[V ] under the map H2k (V ; Z) → H2k (X; Z) induced by the inclusion of V in
X; by abuse of notation, we will continue to denote this by [V ] ∈ H2k (X; Z).
Definition: More generally, given a subscheme V of pure dimension k (a term
we will explain in a minute), we will set
X
[V ] = li [Vi ]
i
Here the sum is taken over the components Vi of the reduced subscheme
Vred . To say that V has pure dimension k means that each component
Vi has dimension k. (Warning: if V has embedded components, the def-
inition is disputed. See https://mathoverflow.net/questions/30495/pure-
dimensional-and-embedded-components.) The integers li are defined by
li = lengthOVi OV
If V is a subvariety of dimension k and Li , i = 1, . . . , k are line bundles,
we use the pairing H 2k × H2k → Z to define the intersection number
< c1 (L1 ) ∧ · · · ∧ ck (Lk ), [V ] > ∈ Z
We extend this by linearity to the case when V is a subscheme. If Li =
O(Di ), we also use the notations:
Z
(3) D1  · · ·  Dk
V
or
(4) (D1  · · ·  Dk  V )
for the above intersection number. Clearly (D1  · · ·  Dk  V ) depends only on
the rational equivalence classes of the Di ’s, and is mutilinear and symmetric
in the Di s.
The last two notations are very suggestive and indicate two other ways of
viewing these numbers. The expression (3) says that the intersection number
can be computed as an integral. Each of the Chern classes is represented
by a closed (1,1)-form on X reg , their product is a (k, k)-form, which can
be integrated over V reg to yield the intersection number. (Since V reg is
noncompact if V is singular, this is not obvious.) The expression (4) signals
a way of computing intersection numbers which is very important:
 LINE BUNDLES 19

The intersection theory black box.

Let D1 , . . . , Dn be Cartier divisors:
X
Di = ai,ji Di,ji
ji

with ai,ji ∈ Z and the Di,ji reduced, irreducible codimension one subvari-
eties. Then
X
(D1  · · ·  Dn  X) = a1,j1 . . . an,jn #{D1,j1 ∩ · · · ∩ Dn,jn }
j1 ,...,jn

provided that at each of the intersection points p in the sum,

• each of the the relevant codimension one subvarieties is regular, as
is X, and
• one can find analytic local coordinates z1 , . . . , zn on a neighbourhood
U ⊂ X reg such that for all i, zi (p) = 0 and Di,ji ∩ U = {zi = 0}.
Given a subvariety V of dimension k, then (D1  · · ·  Dk  V ) can be
computed by replacing each Di by a linearly equivalent divisor Di0 such its
support does not contain V , and intersecting the restrictions Di0 |V . If Dn is
an irreducible, reduced, codimension one subvariety that is locally principal,
(D1  · · ·  Dn  X) = (D1  · · ·  Dn−1  Dn )
where on the right-hand side we are intersecting n − 1 divisors on the sub-
variety Dn .
Note:
(1) In particular, if L is a line bundle on X and C ⊂ X is a curve,
deg L|C ≡ c1 (L)C ≡< c1 (L), [C] > is the pairing of the cohomology
class c1 (L) and the homology class [C]. If C is smooth, this coincides
with the degree of L restricted to C as defined earlier.
(2) If X is a projective surface, and C, C 0 irreducible Cartier divisors
intersecting transversally (only) at regular points of X,
#{C ∩ C 0 } = deg O(C)|C 0 = deg O(C 0 )|C
(If C and C 0 are not assumed locally principal, is this still true? I
am not sure.)
3.1. Numerical equivalence. Let X be a projective variety. We have seen
that
H 1 (X, OX )
< P ic(X)
H 1 (X, Z)
with the quotient mapping (via the Chern class map) injectively into H 2 (X, Z);
let us denote by H 2 (X, Z)alg the image. The image H 2 (X, Z)alg contains
1
the torsion classes in H 2 (X, Z). If X is smooth, the subgroup HH (X,O X)
1 (X,Z)

is naturally a complex torus and a projective variety (in other words, an

abelian variety).
On a general variety X, we say that two line bundles L1 and L2 are
20 T.R. RAMADAS

(1) algebraically equivalent if there is a connected variety T , a line bundle

L on T × X, and points t1 , t2 ∈ T such that L|{tj }×X = Lj , j = 1, 2.
(2) homologically equivalent if c1 (L1 ) = c1 (L2 )
(3) numerically equivalent if < c1 (L1 ), C >=< c1 (L2 ), C > for every
irreducible curve C ⊂ X.
These are all equivalence relations3. We have
Algebraic equivalence =⇒ homological equivalence =⇒ numerical equivalence.
A B

At least for smooth projective varieties, the implication A can be reversed.

Important fact: going modulo numerical equivalence kills only the torsion
in H 2 (X, Z)alg .
Summarising:
P ic(X)/{hom. eq. (= alg. eq. in smooth projective X)} ∼ H 2 (X, Z)alg
L7→c1 (L)

→ H 2 (X, Z)alg /H 2 (X, Z)tor ≡ N 1 (X)

num. eq.

where, following Lazarsfeld, we define4 the Neron-Severi group N 1 (X) to be

P ic(X)/{numerical equivalence}.
Note that the Neron-Severi group is a free abelian group of finite rank less
than or equal to the second Betti number of X. The rank of the Neron-Severi
group is called the Picard number; we denote it by ρ(X).
Intersection numbers are unchanged under numerical equivalence. To see
this note that by definition these depend only on the Chern classes of the
line bundles in question; note next that a torsion class must have zero in-
tersection number with anything else.

4. Asymptotic Riemann-Roch

The (Hirzebruch-)Riemann-Roch Theorem computes the Euler character-

istic of a coherent sheaf in terms of its Chern Classes. A consequence of this
theorem (or, rather, its analogues for singular varieties) is the following,
which we will use:
Theorem 4.1. Let X be an n-dimensional projective variety, F a coherent
sheaf, D a Cartier divisor, L = O(D). Then there is a polynomial with
rational coefficients which gives at integral values the function m 7→ χ(F ⊗
Lm ), and
Dn n
χ(F ⊗ Lm ) = (rank F) m + terms of order mn−1 or lower
n!
3See https://math.stackexchange.com/questions/1057281/algebraic-equivalence-of-
line-bundles for a clever proof in the first case.
4This is slightly unconventional; the usual definition identifies N S(X) with
2
H (X, Z)alg .
 LINE BUNDLES 21

A coherent sheaf F on an irreducible (not necessarily projective) scheme is

free on a (Zariski) open, dense set. By rank F we mean the rank r of this
free sheaf; the rank is zero iff the support of F is a proper (closed) subset.
The term Dn refers to the self-intersection
Dn = (D
|  ·{z
· ·  D})
n times

We will give a proof of the Theorem, stopping short of pinning down the
dependence on the self-intersection of D. We follow the treatment of O.
Debarre’s “Higher-Dimensional Algebraic Geometry” where a more general
result is proved.

Proof. Recall that associated to any coherent sheaf F on an (irreducible,

reduced) variety is its torsion subsheaf Ftor , whose support Y is a proper
closed subscheme (which need not be reduced or irreducible.) We have
0 → Ftor → F → Ftf → 0
with Ftf torsion-free. Clearly, rank F = rank Ftf . We will now use a
very useful result from EGA III.3.1 which implies: a coherent sheaf G on
a projective scheme Y has a filtration with successive quotients torsion-free
on reduced, irreducible sub-varieties of Y . Applying this to Ftor , and by
induction on n, we see that χ(Ftor ⊗ Lm ) has order at most mn−1 , so we
can assume that F itself is torsion-free.
Let F → Or be an isomorphism on an open U ⊂ X. We claim that
the morphism F → Or extends to an injection F → KX r on X where K

is the quasi-coherent (constant) sheaf of meromorphic functions on X. Let

G = F ∩ Or . The quotients F/G and Or /G are both supported on lower-
dimensional subschemes of X, so as above we argue that
χ(F ⊗ Lm ) ∼ χ(G ⊗ Lm ) ∼ χ(Or ⊗ Lm )
where ∼ indicates an equality up to terms of order mn−1 or lower.
So it suffices to treat the case F = O.
Let D = D1 − D2 where D1 and D2 are positive integral combinations of
irreducible codimension one subvarieties. Then O(−Dj ), j = 1, 2, defined
earlier in these notes, are sheaves of ideals (not necessarily invertible, unless
both Dj are Cartier); to emphasize this, let us write O(−Dj ) = Ij . Let Dj
denote the corresponding subschemes. We have exact sequences:
0 → Ij → O → ODi → 0, i = 1, 2
Tensoring by the locally free sheaf O(lD) these remain exact, so we get:
0 → O(lD) ⊗ Ij → O(lD) → O(lD)|Dj → 0
Take l = m and j = 2, we get
0 → O(mD) ⊗ I2 → O(mD) → O(mD)|D2 → 0
22 T.R. RAMADAS

More subtly, if we take l = m+1 and j = 1; we get (since O((m+1)D)⊗I1 =

O(mD) ⊗ I2 ),
0 → O(mD) ⊗ I2 → O((m + 1)D) → O((m + 1)D)|D1 → 0
This shows that
χ(X, O((m+1)D))−χ(X, O(mD)) = χ(X, O((m+1)D)|D1 )−χ(X, O(mD)D2 )
Now Dj need not be irreducible or reduced, but one can again use filtering
as above and induction to conclude that the RHS is a polynomial of order
at most mn−1 . The claim of the Theorem now follows. 

5. Ample line bundles

Definition: A line bundle L on a projective variety X is said to be very ample
if the complete linear system H 0 (X, L) is base-point free and the induced
map φ|L| : X → P(H 0 (X, L)) is a closed embedding. (This means that the
map is injective, and a isomorphism of varieties onto the image Y ; that is,
(φ|L| )∗ OX = OY . In other words, X is a subvariety of a projective space
and O(1) restricts to L.) A line bundle L is ample if some positive power
is very ample. A Cartier divisor D is ample (resp., very ample) if O(D) is
ample (resp., very ample).
The basic result (“Cartan-Serre-Grothendieck” in Lazarsfeld) about am-
ple line bundles is:
Theorem 5.1. Let L be a line bundle on a projective variety X. Then
following are equivalent:
(1) L is ample.
(2) Given any coherent sheaf F on X, there is an integer m1 (F) such
that for i > 0,
H i (X, F ⊗ Lm ) = 0 provided m ≥ m1 (F) .
(3) Given any coherent sheaf F on X, there is an integer m2 (F) such
that F ⊗ Lm is globally generated for m ≥ m2 (F).
(4) There is a positive integer m3 such that Lm is very ample for m ≥
m3 .
Now, this is not the natural context for the result. One certainly needs
compactness, which forces us to restrict to projective varieties. In the
complex-analytic (Cartan’s) context, the theorem would apply to compact
complex manfolds; in the algebro-geometric context, the natural context is
that of a possibly nonreduced and reducible abstract algebraic variety that
is complete. For the definition of “completeness” see, eg., Mumford.
Before turning to the proof of the above Theorem, we recall some basics.
Given a sheaf F on a topological space, denote by Fx its stalk at a point
x ∈ X. By definition this is the direct limit
lim Γ(U, F)
→
 LINE BUNDLES 23

where the limit is taken over open sets containing x, partially ordered by
inclusion. If F is a coherent sheaf on a variety X,
Fx = Γ(U, F) ⊗C[U ] OU,x
for any affine U containing x, where OU,x = OX,x is the local ring of X at
x. The local ring OX,x is itself the stalk of OX .
Let I(x) ⊂ OX denote the ideal sheaf of x, the sheaf of regular functions
vanishing at x. (In the lectures I used the notation mX,x , but this is better
reserved for the maximal ideal in the local ring OX,x .) We have the defining
exact sequence
0 → I(x) → OX → C[x] → 0
where C[x] is the skyscraper sheaf at x:
H 0 (U, C[x] ) = C if x ∈ U and {0} otherwise
and the map OX → C[x] is the evaluation map at z. (Thus C[x] = OX /I(x).)
If F is a coherent sheaf, tensoring by it yields
I(x) ⊗OX F → F → C[x] ⊗OX F → 0
(In general left-exactness is not preserved.) Denoting by I(x)F the image
of the sheaf morphism I(x) ⊗OX F → F, we get the exact sequence
0 → I(x)F → F → C[x] ⊗OX F → 0
Here C[x] ⊗OX F is a skyscraper sheaf at x with “fibre” the finite-dimensional
vector space H 0 (X, C[x] ⊗OX F). (The term “fibre” is not standard except if
F is locally free in which case this is indeed the fibre at x of the correspond-
ing vector bundle5.) A sequence of sheaves is exact iff the corresponding
sequence of stalks is exact at every point, we have the exact sequence of
modules over the local ring OX,x
0 → (I(x)F)x → Fx → H 0 (C[x] ⊗OX F) → 0
Let mx denote the maximal ideal in OX,x ; then (I(x)F)x = mx Fx . We see
that Nakayama’s Lemma (“if M is a finitely generated module over a local
ring with maximal ideal m, then mM = M iff M = 0”) implies: if F is
coherent, then its fibre at x vanishes iff its stalk FX,x vanishes, that is, iff
F itself restricts to zero on some open neighbourhood of x.
As a consequence:
Lemma 5.2. Let F → G be a morphism of coherent sheaves on a variety
X, and suppose that for some x ∈ X the induced map (of finite-dimensional
vector spaces)
F ⊗OX C[x] → G ⊗OX C[x]
is onto. Then there is an open neighbourhood U of x on which the map of
sheaves is onto.
5If E is a vector bundle, and E the corresponding sheaf, we reserve the right to use
the notation Ex for the fibre and E x for the stalk. In such contexts, we will stick to the
notational distinction between E and E
24 T.R. RAMADAS

Proof. Let K denote the cokernel sheaf, so that we have the exact sequence:
F →G→K→0
Tensoring with C[x] we get the exact equence
F ⊗OX C[x] → G ⊗OX C[x] → K ⊗OX C[x] → 0
which shows that K ⊗OX C[x] = 0. The previous Lemma applies and we see
that K is zero on some neighbourhood of x. 
We will need another consequence of Nakayama’s Lemma:
Lemma 5.3. Let f : (A, m) → (A0 , m0 ) be a local homomorphism of Noe-
therian local rings, with A0 finitely generated as an A-module and suppose
that f induces an isomorphism of residue fields A/m = A0 /m0 ≡ k. For
f to be surjective, it is (necessary and) sufficient that the induced map of
“Zariski cotangent spaces”
m/m2 → m0 /m02
is surjective.
Proof. We prove sufficiency. (My earlier proof had an error, as pointed out
by Akashdeep; this one is adapted from Joe Harris: Algebraic Geometry:
A First Course.) We will use the follwing corollary of Nakayama’s Lemma
repeatedly: if A is a Noetherian local ring with maximal ideal m, M a finite
A-module, and N a submodule such that N + mM = M , then N = M .
By replacing A by its image in A0 , we can assume that the f is an injection.
By assumption, the inclusion m → m0 induces a surjection m/m2 → m0 /m02 ,
so
mA0 + m02 = m0
Applying the above Corollary to the inclusion of A0 -modules mA0 ⊂ m0
yields mA0 = m. Consider now the inclusion of A-modules A ⊂ A0 . Since
mA0 + A = m0 + A = A0 , we see that A = A0 . 
We can now turn to the proof of the Theorem.
Proof. (1) implies (2): Assume first that L is very ample. Then
X ,→ P(V ),
ι

such that ι∗ O(1)= L. (We will need to talk about the fibre of L, so for a
while we will keep the distinction between L and L.) Then H i (X, F ⊗OX
Lm ) = H i (P(V ), ι∗ (F ⊗OX Lm )) = H i (P(V ), ι∗ F ⊗OP (V ) O(m)). We have
used the extension by zero operation ι∗ ; this takes a coherent sheaf on a
closed subvariety to a coherent sheaf on the ambient variety. This is a
special case of a direct image by a morphism. We have also used a special
case of the projection formula: given a morphism π : X → Y ,
π∗ (F ⊗OX π ∗ V) = π∗ F ⊗OY V
 LINE BUNDLES 25

provided V is locally free6. Now, given any coherent sheaf G on P(V ), there
exists a m(G) such that for i > 0,
H i (P(V ), G ⊗OX O(m)) = 0 provided m ≥ m(G)
Applying this to ι∗ F gives the desired result, with m1 (F) = m(ι∗ F). If L
is only ample, with LM very ample, let ι be the corresponding projective
embedding. We have, for 0 ≤ l ≤ M − 1 and i > 0,
H i (X, F ⊗OX Ll ⊗OX LmM ) = 0 provided m ≥ m(ι∗ (F ⊗OX Ll )),
With a little book-keeping, this gives the desired result.
(2) implies (3): For x ∈ X, let m1 (x) be such that for i > 0,
H i (X, I(x)F ⊗OX Lm ) = 0 provided m ≥ m1 (x)
As a consequence, for m ≥ m1 (x), the evaluation map
H 0 (F ⊗OX Lm ) ⊗C OX → F ⊗OX Lm
is onto at x and therefore in an open neighbourhood Ux . By (quasi-)compactness
we can cover X by finitely many of these open sets. Taking m1 to be the
supremum of the corresponding m1 (x)’s we are done.
(3) implies (4): For x ∈ X, consider the sheaf I(x)L; by (3) there exists
m(x) such that I(x)L ⊗OX Lm is globally generated for m ≥ m(x). By
quasi-compactness we can m̃3 such that for m ≥ m̃3 + 1, the evaluation map
(5) H 0 (I(x)Lm ) ⊗C OX → I(x)Lm
is onto for every x ∈ X. (Since L is invertible, I(x)L ⊗OX Ll = I(x)Ll+1 .)
Now I(x)Lm is the sheaf of sections of Lm vanishing at x. So, given distinct
x, x0 ∈ X, there is a section of Lm that is nonvanishing at x0 and vanishing
at x. In other words, the linear system of Lm is base-point free and the
corresponding map φ|Lm | : X → P(H 0 (X, Lm )) is injective. Since X is
compact, the image is closed.
All that is left is to make sure that every (local) regular function on X is
the pull back of a (local) regular function. Explicitly, let f ∈ OX,x ; this is
a regular function on U ∩ X, with U open in P(V ). We need to exhibit a f˜
regular on U , possibly at the cost of shrinking U . In other words, we have to
ensure that for every x ∈ X, the map of local rings OP(V ),x → OX,x is onto.
In turn, it suffices to show (Lemma 5.3) that mP(V ),x /m2P(V ),x → mX,x /m2X,x
is onto. Consider the map
H 0 (X, Lm ) ⊗OX L−m → O .
This restricts to
H 0 (X, I(x)Lm ) ⊗OX L−m → I(x) .

6As Pramath pointed out, if π is an affine morphism (as ι is) this holds even if V is
not locally free. Proof: given a ring homomorphism A → B, an A-module MA and a
B-module MB , we have an isomorphism of A-modules MB ⊗B B ⊗A MA → MB ⊗A MA .
26 T.R. RAMADAS

That this is onto for m ≥ m3 ≡ m̃3 + 1 can be seen by tensoring (5) by

L−m . This induces a surjective map of vector spaces
H 0 (X, I(x)Lm ) ⊗C L−m
x → mX,x /m2X,x
where Lx is a one-dimensional vector space, namely, the fibre of the line
bundle L at x. For y ∈ P(V ), we also have the analogous map
H 0 (P(V ), IP(V ) (y)O(1)) ⊗C O(−1)x → mP(V ),y /m2P(V ),y

which is also onto, although we seem not to need this (see below). But we
have a commutative diagram:
H 0 (P(V ), IP(V ) (x)O(−1)) ⊗C O(−1)x −−−−→ mP(V ),x /m2P(V ),y
 
=y
 
y
onto
H 0 (X, I(x)Lm ) ⊗C L−m
x −−−−→ mX,x /m2X,x

which shows that the map mP(V ),x /m2P(V ),x → mX,x /m2X,x is onto, provided
we justify the equality H 0 (P(V ), IP(V ) (x)O(1)) = H 0 (X, I(x)Lm ). To see
this, consider the commutative diagram
0 −−−−→ H 0 (P(V ), IP(V ) (x)O(1)) −−−−→ H 0 (P(V ), O(1))
 
=
 
y y
0 −−−−→ H 0 (X, I(x)Lm ) −−−−→ H 0 (X, Lm )
Finally, (4) implies (1) by definition.


We will join Lazarsfeld’s losing battle and refer to the property of ample-
ness as “amplitude”.

5.1. Basic properties of amplitude. (I) If L is ample and L0 is any line

bundle L0 ⊗ Lm is very ample for m ≥ m(L0 ) large enough. Proof: Let
m1 be such that Lm is very ample for m ≥ m1 , and let m2 be such that
L0 ⊗ Lm is globally generated for m ≥ m2 . Then L0 ⊗ Lm is very ample for
m ≥ m1 + m2
(II) If L and M are ample line bundles on projective varieties X and Y
respectively, L  M is ample on X × Y . Proof: We can suppose that L and
M are both very ample. By Kunneth, H 0 (X × Y, L  M ) = H 0 (X, L) ⊗C
H 0 (Y, M ), so L  M is free. The corresponding map φ|LM | is the compo-
sition
X ×Y → P(H 0 (X, L)) × P(H 0 (Y, M ))
φ|L| ×φ|M |

→ P(H 0 (X, L) ⊗C H 0 (Y, M )) = P(H 0 (X × Y, L  M ))

V
 LINE BUNDLES 27

where V is the “Veronese” embedding7. This shows that L  M is very

ample.
(III) If f : Y → X is a finite map of projective varieties and L is an ample
line bundle on X, then f ∗ L is ample on Y . Proof: Let F be a coherent
sheaf on Y . We have for any m and i
H i (F ⊗ f ∗ Lm ) = H i (f∗ (F ⊗ f ∗ Lm ))
= H i (f∗ F ⊗ Lm ))
where the first equality holds because f is affine, and the second is the
projection formula. Now use part (2) of the Theorem. (Note that f ∗ L is
trivial when restricted to any fibre of f . So if f is not finite, which in our
context means if f has a positive-dimensional fibre, then f ∗ L is not ample.)
(IV) If L is free, then it is ample iff the corresponding map φ|L| : X →
P(H 0 (X, L)) is finite iff deg L|C > 0 for every irreducible curve C in X.
(V) Asymptotic Riemann-Roch II: Let X be an n-dimensional projective
variety, F a coherent sheaf, D an ample Cartier divisor, L = O(D). Then
there is a polynomial with rational coefficients which gives at integral values
the function m 7→ χ(F ⊗ Lm ), and
Dn n
h0 (F ⊗ Lm ) = (rank F) m + terms of order mn−1 or lower
n!
where hi ≡ dim H i . Proof: Since L is ample, h0 (F ⊗ Lm ) = χ(F ⊗ Lm )
for m large enough.
(VI) Amplitude in families: If f : X → T is a proper morphism of quasipro-
jective varieties, and L a line bundle on X. For t ∈ T , set Xt = f −1 (t). If
L|Xt0 is ample for some t0 ∈ T , then there is a neighbourhood U of t0 such
that L|Xt) is ample for t ∈ U .

5.2. Projective schemes. For the most part we want to stick to varieties,
but we will have to deal with more general schemes occasionally. Instead of
biting the bullet and defining amplitude on schemes, we will manage with
the following
Lemma 5.4. Let Y ⊂ P(V ) be a subscheme, F a coherent sheaf on Y ,
and L an invertible sheaf on Y such that L|Y1 is ample for every reduced,
irreducible subscheme Y1 ⊂ Y . Then for large enough m we have
H i (Y, F ⊗OX Lm ) = 0, i > 0 .
Proof. We use the fact that F has a filtration
F = FN ⊃ FN −1 ⊃ · · · ⊃ F0 = {0}
7Given finite-dimensional vector spaces V, W , the bilinear map (v̌, w̌) 7→ v̌ ⊗ w̌ induces
an embedding Psub (V̌ ) × Psub (W̌ ) → Psub (V̌ ⊗ W̌ ), and dually P(V ) × P(W ) → P(V ⊗ W ).
28 T.R. RAMADAS

which gives us exact sequences:

0 → Fl−1 → Fl → Fl /Fl−1 → 0
with successive quotients Fl /Fl−1 torsion-free on reduced, irreducible sub-
varieties Yl of Y ,. By hypothesis L is ample on all these subvarieties. Tensor
with Lm , and choose m large enough that for i > 0
H i (Yl , Fl /Fl−1 ⊗OX Lm ) = 0, l = . . . N
An upward induction on l now yields the desired vanishing of H 1 (Y, F ⊗OX
Lm ). 

6. Numerical criteria for amplitude

We begin with the criterion of Nakai-Moishezon-Kleiman:
Theorem 6.1. A line bundle L on a projective variety X is ample iff for
every irreducible Y ⊂ X
c1 (L)(dim Y)
[Y ] > 0
Proof. We prove first that if L is ample, then the inequality holds. Then
some positive power Lm is very ample, and since c1 (Lm ) = mc1 (L), we have
c1 (Lm )(dim Y ) [Y ] > 0 =⇒ c1 (L)(dim Y ) [Y ] > 0. So we can assume that
Y ⊂ (X ⊂) P(V ), and L is the hyperplane bundle O(1). Now, the Bertini
Theorem assures us that for a generic hyperplane H1 , the intersection H1 ∩Y
is an irreducible variety with (H1 ∩ Y )sing ⊂ Y sing . By induction, we get
hyperplanes H1 , H2 , . . . , H(dim Y −1) such that C ≡ Y ∩ H1 ∩ H2 ∩ · · · ∩
H(dim Y −1) is a (reduced, irreducible) curve with C sing ⊂ Y sing . One can
now find a hyperplane H(dim Y ) which avoids the singularities of C and
intersects it transversally at its regular points. Then using our intersection
theory black box, we get
c1 (L)(dim Y)
[Y ] = #{Y ∩ H1 ∩ H2 ∩ · · · ∩ H(dim Y )} >0
(This integer is, by definition, the degree of Y with respect to the given
projective embedding.)
Suppose now that the inequality c1 (L)(dim Y ) [Y ] > 0 holds for every sub-
variety Y ⊂ X. We will prove that L is ample, by induction on n = dim X.
Step 1: Write L = L ⊗ M ⊗ M −1 with both L ⊗ M and M very ample.
By Bertini we can assume that L ⊗ M = O(D) and M = O(D0 ) with
D and D0 reduced and irreducible divisors. We have an exact sequence
0 → O(−D) → OX → OD → 0, and a similar one involving D0 . These
yield, for any integer m:
0 → Lm+1 ⊗O O(−D) → Lm+1 → Lm+1 |D → 0
and
0 → Lm ⊗O O(−D0 ) → Lm → Lm |D0 → 0
 LINE BUNDLES 29

Noting that L(−D) = O(−D0 ), we get

0 →Lm ⊗O O(−D0 ) → Lm+1 → Lm+1 |D → 0
0 →Lm ⊗O O(−D0 ) → Lm → Lm |D0 → 0
By our inductive hypothesis, the restriction of L to D is ample, as is its
restriction to D0 . There exists a m3 such that for m ≥ m3 ,
H i (D, Lm ) = 0, i > 0
H i (D0 , Lm ) = 0, i > 0
This implies that for i ≥ 2
H i (X, Lm ) = H i (X, Lm ⊗O O(−D0 )) = H i+1 (X, Lm )
which shows that the function m 7→ H i (X, Lm ) is eventually constant. On
the other hand, the Asymptotic Riemann-Roch (I) asserts
c1 (L)n [X] n
χ(X, Lm ) = m + terms of order mn−1 or lower
n!
where n = dim X. We have thus
c1 (L)n [X] n
h0 (X, Lm ) = m + h1 (X, Lm )
n!
+ alternating sum of hi , i ≥ 2
+ terms of order mn−1 or lower
Now h1 is nonnegative, the hi , i ≥ 2 bounded, and c1 (L)n [X] > 0 by
assumption, so we see that Lm is effective for large enough m.
Step 2: We will show that some power of L is globally generated. By re-
placing L by a suitable power, we can suppose that it has a nonzero section
σ, with (σ) = D. In general D will have many components (all with codi-
mension one) with multiplicities, so it need not be a variety. Consider:
0 → O(−D) → O → O|D → 0
where O|D is the structure sheaf of the scheme D, pushed forward to X.
Tensoring by Lm , we get
0 → Lm (−D) → Lm → Lm |D → 0
By the inductive hypothesis L is ample restricted to each (reduced, irre-
ducible) component of D, so by Lemma 5.4 H 1 (X, Lm |D ) vanishes for large
enough m, so that in that case we have the exact sequence:
H 1 (X, Lm−1 ) → H 1 (X, Lm ) → 0
This shows that the h1 (X, Lm ) is eventually constant and the maps eventu-
ally isomorphisms. In turn this shows that the maps
H 0 (X, Lm−1 ) → H 0 (X, Lm |D )
30 T.R. RAMADAS

are eventually surjective. Let D0 be any one of the irreducible components

of Dred ., and let ID0 be the ideal sheaf of D0 in D. We have
0 → Lm ⊗ ID0 → Lm ⊗ OD → Lm |D0 → 0
Again using Lemma 5.4, H 1 (Lm ⊗ ID0 ) vanishes for large enough m, so that
eventually
H 0 (Lm ⊗ OD ) → H 0 (Lm ⊗ OD0 )
is onto. By the inductive hypothesis, L|D0 is ample, so Lm
D0 is eventually
globally generated. Now we can finish the proof of global generation as
follows:
(1) L is globally generated outside D (by the section σ), and hence so
is any positive power, and
(2) for a large enough m, on each component Dl of D, Lm is globally
generated and every section of Lm on Dl extends to X.

Step 3: Let m be large enough that Lm is globally generated, so that we

have a morphism
φLm : X → P(H 0 (X, Lm ))
Note that O(1) pulls back to Lm , so that Lm is trivial on the fibres. This
ensures that the fibres are zero-dimensional, since else on a curve C ∈ X
mapping to a point, we would have
c1 (L)[C] = 0
Thus φLm is finite. Consequently Lm is ample, and so also L. 

The Nakai criterion has important corollaries.

Corollary 6.2. Amplitude is stable under numerical equivalence. (This lets
us talk of the ample cone in the Neron-Severi group.)
Proof. This is a consequence of the fact, important in itself, that intersection
numbers are defined modulo numerical equivalence. In other words, if Y is
a k-dimensional variety and L1 , . . . , Lk line bundles, the intersection
c1 (L1 ) ∧ · · · ∧ c1 (Lk )[Y ]
depends only on the numerical equivalence class of the Lj . To see this,
suppose Lj ∼ne L0j where ∼ne indicates numerical equivalence. Then

c1 (L1 ) ∧ · · · ∧ c1 (Lk )[Y ] − c1 (L01 ) ∧ · · · ∧ c1 (L0k )[Y ]

= {c1 (L1 ) − c1 (L01 )} ∧ · · · ∧ c1 (Lk )[Y ]
+ c1 (L01 ){c1 (L2 ) − c1 (L02 )} ∧ · · · ∧ c1 (Lk )[Y ]
+ c1 (L01 ) ∧ · · · ∧ {c1 (Lk ) − c1 (L0k )}[Y ]
 LINE BUNDLES 31

so it suffices to consider the case when Lj = L0j = O(Dj ), j ≥ 2. Then

c1 (L1 ) ∧ · · · ∧ c1 (Lk )[Y ] − c1 (L01 ) ∧ · · · ∧ c1 (Lk )[Y ]
= {c1 (L1 ) − c1 (L01 )}[D2 ∩ . . . Dk ]
= c1 (L1 )[D2 ∩ . . . Dk ] − c1 (L01 )[D2 ∩ . . . Dk ]
Now L1 and L01 being numerically equivalent, their intersections with the
curve D2 ∩ . . . Dk are equal. 
Corollary 6.3. Let f : Y → X be a finite map of projective varieties, and
suppose L is a line bundle on X such that f ∗ L is ample on Y . Then L is
ample on X.
Proof. Let Y ⊂ X be an irreducible variety of dimension k, and let W be
an irreducible component of the inverse image that maps onto Y . Then
c1 (f ∗ L)k [W ] = (degree f |W ) × c1 (L)k [Y ] > 0. 

7. Nef line bundles; Theorem of Kleiman

Consider the Neron-Severi group N 1 (X) of line bundles modulo numerical
equivalence. This is finitely generated free abelian group of rank ρ(X).
Tensoring with Q (resp., R) yields rational (resp., real) vector spaces of
dimension ρ(X), which we will denote N 1 (X)Q (resp., N 1 (X)R ).
(In case X is a smooth projective variety, the Lefshetz Theorem on (1, 1)-
classes identifies N 1 (X)Q with H 2 (X, Q) ∩ H (1,1) (X, C).)
Recall that amplitude is preserved under numerical equivalence. Consider
the subset Amp(X)Z (this notation is not standard) of classes in N 1 (X) that
correspond to ample line bundles. This is closed under addition because a
tensor product of ample line bundles is ample. (It fails to be a semigroup
because the trivial line bundle is not ample.)
Definition 7.1. A line bundle L on a projective variety X is numerically
effective (nef for short) if deg L|C ≥ 0 for every (reduced, irreducible) curve
C ⊂ X.
The subset N ef (X)Z (this notation is not standard either) of nef classes
in N 1 (X) is clearly a semigroup.
By a cone in a real vector space, we will mean a subset closed under
multiplication by strictly positive scalars. We will now define two convex
cones in the vector space N 1 (X)R , the ample cone Amp(X) and the nef cone
N ef (X). The ample cone is the convex cone generated by ample classes.
Explicitly,
X X
Amp(X) = { bj [Lj ]|bj ≥ 0, bj > 0, Lj ample}
j j

Note that if C is a (reduced, irreducible) curve, [L] 7→ c1 (L)[C] is a

homomorphism N 1 (X) → Z; it extends to a linear map N 1 (X)R → R. The
32 T.R. RAMADAS

nef cone is defined by8

X X
N ef (X) = { bj [Lj ]| bj c1 (Lj )[C] ≥ 0 ∀ C}
j j

We claim that
(1) ∅ =
6 Amp(X) ⊂ N ef (X)
(2) Amp(X) is open, and
(3) N ef (X) is closed.
In fact, we will see later that the nef cone is the closure of the ample cone
and the ample cone is the interior of the nef cone.
Let us pause to justify the above claims. Since X is projective, the pull-
back of the hyperplane bundle by any projective embedding is very ample,
so ample classes exist. An ample class restricted to any curve C is ample
and hence has positive degree. Thus
∅=
6 Amp(X)Z ⊂ N ef (X)Z
and (1) follows. That N ef (X) is closed is clear because it is the intersection
of closed “half-spaces”. It remains to show:
Proposition 7.2. Amp(X) is open.
Proof. Before going further, we fix a basis of generators {[Li ]|i = 1, . . . , ρ(X)}
for the abelian group N 1 (X). This defines an L∞ -norm on the vector space
N 1 (X)R which we will use to topologise it from now on.
We prove first: let [L] be an integral ample class. Then there exists (L) >
0 such that every rational class [L0 ] with ||[L0 ] − [L]|| < (L) is ample. In
fact, there exists m0 > 0 such that Lm Li is ample for all i and m ≥ m0 . Let
[L0 ] = [L] + i ai [Li ], with ai rational numbers. Then
P

1 X
[L0 ] = {[L] + ai ρ(X)[Li ]}
ρ(X)
i

is ample provided |ai ρ(X)| ≤ m10 , i.e., ||[L0 ] − [L]|| < (L) ≡ ρ(X)m
1
0
.
We prove next: let [L], [M ] be integral classes, with L ample. Suppose give
δ > 0 such that [L] + d[M ] is ample for every rational d such that |d| < δ.
Then [L] + dM ] is ample for every real d such that |d| < δ. Proof: Pick
rational numbers d1 , d2 such that −δ < d2 < d < d1 < δ. Then
d − d2 d1 − d
[L] + t[M ] = ([L] + d1 [M ]) + ([L] + d2 [M ])
d1 − d2 d1 − d2
which expresses [L] + d[M ] as a convex (real) linear combination of ample
rational classes.

8Note that {P b [L ]|b ≥ 0, P b > 0, L nef } ⊂ N ef (X), but apparently this can
j j j j j j j
be a strict inclusion.
 LINE BUNDLES 33
P
Finally suppose [L] = j tj [Hj ] with tj real and positive and [Hj ] integral
and ample. Let 0 < t < t1 be rational. Then
X X X
[L] + ai [Li ] = t[H1 ] + ai [Li ] + (t1 − t)[H1 ] + tj [Hj ]
i i j>1
P
It is enough to show that t[H1 ]+ i ai [Li ] is ample for |ai | small enough. By
the preceding paragraphs, there exists 1 > 0 such that for each i, [H1 ]+bi [Li ]
for bi real and |bi | < 1 . Now note that
X t X ρ(X)
t[H1 ] + ai [Li ] = {[H1 ] + ai [Li ]}
ρ(X) t
i i

and [H1 ] + ai ρ(X)

t [Li ] is ample for |ai | <
t
ρ(X) 1 . 

We will prove below that Amp(X) is the interior of N ef (X) and N ef (X)
the closure of Amp(X). This will be one of the corollaries of
Theorem 7.3. (Kleiman) Let L ne a nef line bundle on a projective variety
X. Then
c1 (L)dim V [V ] ≥ 0
for any reduced, irreducible subvariety V ⊂ X.
Proof. The proof will be by induction on n = dim X. The claim is true for
n = 1. By induction, it suffices to prove the above inequality for V = X.
We will suppose that the equality fails. i.e., that
c1 (L)n [X] < 0
and derive a contradiction.
Let M be a very ample line bundle, and consider, for t ∈ R,
P (t) = (c1 (L) + tc1 (M ))n [X]
Expanding the expression on the right we get
P (t) = c1 (L)n [X] + tc1 (L)n−1 c1 (M )[X] + · · · + tn c1 (M )n [X]
The coefficient of the tn−k term is c1 (L)k c1 (M )n−k [X]; provided k < n, we
can (by Bertini) compute this by taking n − k generic hyperplane sections
H1 , . . . , Hn−k such that their intersection is a reduced, irreducible variety of
dimension k and evaluating c1 (L)k on this intersection:
c1 (L)k c1 (M )n−k [X] = c1 (L)k [H1 ∩ · · · ∩ Hn−k ]
By our inductive hypothesis these coefficients are all non-negative, and the
coefficient of tn (when k = 0) is positive (being the degree of X w.r.to the
embedding given by M ). If the constant term is negative, the polynomial
P will have precisely one real root t0 > 0.
Now write P (t) = Q(t) + R(t) where
Q(t) = c1 (L)(c1 (L) + tc1 (M ))n−1 [X]
34 T.R. RAMADAS

and
R(t) = tc1 (M )(c1 (L) + tc1 (M ))n−1 [X]
Now R(t) is a polynomial with no constant term, and non-negative coeffi-
cients, and with the coefficient of the top degree term strictly positive. So
R(t0 ) > 0.
We will now show next that Q(t0 ) ≥ 0, which yields the desired contra-
diction since Q(t0 ) + R(t0 ) = P (t0 ) = 0.
To prove that Q(t0 ) ≥ 0 it suffices to show that Q(t) ≥ 0 for any rational
number ab ≡ t > t0 (where a, b are coprime positive integers, of course).
Note that
1 1
c1 (L) + tc1 (M ) = {bc1 (L) + ac1 (M )} = c1 (Lb M a )
b b
so that Now given any k-dimensional subvariety V ⊂ X, with k < n,
c1 (Lb M a )k [V ] = (bc1 (L)+ac1 (M ))k [V ] = ak c1 (M )k [V ]+non − negativeterms > 0
since M is ample and therefore the first term is positive. On the other hand,
a
c1 (Lb M a )n [X] = (bc1 (L) + ac1 (M ))n [X] = bk P ( ) > 0
b
b a
Hence, by Nakai’s criterion, L M is ample. Then
1
Q(t) = ( )n−1 c1 (L)c1 (Lb M a )n−1 [X] ≥ 0
b
because L is nef. 

8. Consequences of Kleiman’s Theorem

8.1. Multiplicity. We will need the notion of the multiplicity of a (reduced,
irreducible) curve C at a point x ∈ C. For these matters, Fulton’s book(s)
on intersection theory are the standard reference. I also found Mumford’s
Algebraic Geometry I useful, as also 3264 & All That Intersection Theory
in Algebraic Geometry.
Let X be a variety and x a (closed) point. Let Ox be the local ring at x,
and mx the maximal ideal. Consider the graded ring
C[CTx X] = C ⊕ mx /m2x ⊕ · · · ⊕ mm m+1
x /mx
=
Ox /mx

This is generated by mx /m2x and therefore defines a cone in the Zariski tan-
gent space Tx X ≡ HomC (mx /m2 ). This affine scheme (“tangent cone”) need
not be reduced or irreducible (even if V itself is), but its (Krull) dimension
is equal to the dimension of X at x. In the projective space P(mx /m2x ) this
defines a sub-scheme (“projectivised tangent cone”) of dimension dim X −1.
The multiplicity multx (X) of X at x is defined to be its degree:
multx (X) = c1 (O(1))dimx X−1
[projectivised tangent cone]
Even if X itself is reduced and irreducible, the tangent cone is likely to be
neither, so our definition of intersection numbers has to be suitably extended.
 LINE BUNDLES 35

If X ⊂ CN is a hypersurface, then multx (X) is characterised by

X
(∂/∂z1 )α1 . . . (∂/∂zN )αN f (x) = 0 iff αi < multx (X)
i
In other words multx (X) is the degree of the lowest degree term in the
Taylor expansion at x of a defining equation for X.
We will need three facts about multiplicities:
(1) If x is a regular point, the multiplicity is one. This is because
CTx X = Tx X.
(2) If x is a regular point of X, and C is a (reduced, irreducible) curve
passing through x,
multx (C) = deg O(E)|C 0
where E is the exceptional divisor of the blow-up µ : Blx (X) → X
of X at x and C 0 is the proper transform of C, i.e., the closure in
Blx (X) of µ−1 (C \ {x}).
(3) if an effective divisor C and a curve D intersect at a singular point
of either C or D, the “’intersection multiplicity” is bounded by the
product of the multiplicities of C and D at that point. (See below.)
8.2. Intersection Multiplicity. Let X be projective variety C ⊂ X an
(irreducible, reduced) curve, and L a line bundle. According to our inter-
section theory black box,
P
(1) if σ is a section of L such that (σ) = D = i Di , with each Di an
irreducible Weil divisor,
(2) C is not contained in any of the Di ,
(3) C is disjoint from the intersections Di ∩ Dj , for i 6= j,
(4) x ∈ C ∩ D iff x is a regular point of X, D and C, and
(5) the intersections C ∩ D are transverse,
P
then c1 (L)[C] = D.C = i #{C ∩ D}. What happens when conditions (4)
and (5) are dropped? To cut a long story short,
X X
c1 (L)[C] = D.C = I(C, Di , x)
i x∈C∩Di

where the I(C, D, x) are intersection multiplicities which we will not define.
The key fact which we will use is:
I(C, D, x) ≥ multx (C)multx (D)
8.3. Seshadri’s Criterion.
Theorem 8.1. Let X be a projective variety, and L a line bundle on X.
Then L is ample iff there exists (L) > 0 such that for every (reduced,
irreducible) curve C and every point x ∈ C, we have
(6) deg L|C ≥ (L) multx C
(In other words, the degree of L restricted to any curve C should be bounded
below uniformly in terms of the “maximum singularity” of C.)
36 T.R. RAMADAS

Proof. Suppose first that L is ample. Then some power Lm has a section
such that E ≡ (σ) such that σ|C 6= 0 but σ(x) = o. Thus E is effective,
passes through x, and “meets C properly” (i.e., C * E.) Then
X
m deg L|M = i(E, C, y) ≥ i(E, C, x) ≥ multx C
y∈C∩E

Here i(E, C, y) is the intersection multiplicity of E and C at y. So (6) holds

with (L) = 1/m.
Conversely, suppose (6) holds for some positive (L). By induction9 on
dimension and using Nakai’s criterion, it suffices to show that
c1 (L)n [X] > 0
where n = dim X. Fix a smooth point x ∈ X, and let µ : X 0 → X be
the blow-up of X at x; let E be the exceptional divisor. Note that E is
isomorphic to Pn−1 and O(−E)|E is the hyperplane bundle.
We claim that Lm (−E) is nef on X 0 provided (L) < 1/m. Granting this,
we have by Kleiman’s Theorem:
0 < (mc1 (L)−c1 (O(E)))n [X 0 ] = mn c1 (L)n [X]+(−1)n c1 (O(E)))n [X 0 ] = mn c1 (L)n [X]−1
which yields the desired inequality.
Turning now to the claim, let C 0 be any (reduced, irreducible) curve in
0
X . We need to show that
1
deg µ∗ L|C 0 ≥ deg O(E)|C 0
m
If C 0 ⊂ E, then L is trivial on C 0 and O(E)) has negative degree on C 0 . If
C 0 * E, let C = µ(C 0 ), so that C 0 is the proper transform of C. Now
deg µ∗ L|C 0 = deg L|C
and
deg O(E)|C 0 = multx (C)
so that by hypothesis, the claim is proved. 
8.4. Other consequences of Kleiman’s Theorem.
Proposition 8.2. Let X be a projective variety. Let z be a nef class and w
an ample class. Then z + w is ample.
Proof. Let  > 0 be such that Amp(X) contains an open -ball around w.
Let z1 be a positive rational linear combination of (integral) nef classes such
that ||z − z1 || < /2. Let w1 be a rational ample class such that ||w1 || < /2.
It suffices to prove that z1 + w1 is ample. To see this note that
z + w = z1 + w1 + {w + (z − z1 − w1 )}
9This induction takes care of positivity on subvarieties. These subvarieties could be
contained in the singular locus of X, in which case we need to consider what happens at
singular points x ∈ X. Thanks to Pramath for spotting this. For positivity on X itself
we get by by considering a regular point x.
 LINE BUNDLES 37

and (since ||(z − z1 − w1 )|| < ) the term {w + (z − z1 − w1 )} is ample.

By multiplying by z1 and w1 by an integer, we can assume that z1 and
w1 are integral classes, with w1 being the class of a very ample line bundle
L. Now we check ampiltude of z1 + w1 using the Nakai criterion. Let V be
any irreducible subvariety; then
(w1 + z1 )k [V ] = w1k [V ] + rest
The first term is positive since w1 is ample; each of the other terms is (up
to a positive constant factor) of the form
z1k−l [H1 ∩ . . . Hl ∩ V ]
where l < k and Hi are irreducible divisors such that O(Hi ) = L. Each of
the above intersection numbers is non-negative by Kleiman’s Theorem. 
Corollary 8.3. Fix an ample class w in N 1 (X)R . A class z ∈ N 1 (X)R is
ample if for some  > 0 such that for any curve C ⊂ X,
z[C] ≥ w[C]
i.e., if z − w is nef.
Proof. z = z − w + w. 
As another corollary, we get:
Theorem 8.4. Let X be a projective variety. Then
(1) Amp(X) = N ef (X).
(2) Amp(X) = N ef (X)interior .
Proof. The statement (1) is immediate: let z be a nef class, w any ample
class, and for t > 0 consider the family of ample classes z + tw; as t → 0
this tends to z. As for (2), let z be a nef class such that an open open -ball
around z consists of nef classes. We wish to show that z is ample. Let w be
ample with ||w|| < /2 and write = z − w + w. 
8.5. The cone of curves. Define N1 (X)R to be the quotient of the real
vector space generated by reduced, irreducible curves in X modulo numerical
equivalence:
X
N1 (X)R = { aj Cj |red., irred., curve Cj ,→X, aj real}/{num. equiv.}
j

By definition, there is a non-degenerate pairing

N1 (X)R × N 1 (X)R → R
induced by (C, [L]) 7→ deg L|C . In particular N1 (X)R is finite-dimensional
of dimension ρ(X). The cone of curves N E(X) is the convex cone
X
N E(X) = { aj Cj |red., irred., curve Cj ,→X, aj real, ≥ 0}/{num. equiv.}
j

Its closure N E(X) ⊂ N1 (X)R is the closed cone of curves.

38 T.R. RAMADAS

Clearly,
N ef (X) = {z ∈ N 1 (X)R |z(C) ≥ 0, ∀ C ⊂ X}
curve
1
= {z ∈ N (X)R |z(α) ≥ 0, ∀ α ∈ N E(X)}
= {z ∈ N 1 (X)R |z(α) ≥ 0, ∀ α ∈ N E(X)
The theory of dual cones yields
N E(X) = {α ∈ N1 (X)R |z(α) ≥ 0 ∀ nef z}
We finish with Kleiman’s criterion for amplitude via cones.
Proposition 8.5. A class w ∈ N 1 (X)R is ample iff w(α) > 0 for all
nonzero classes in N E(X). That is, iff w 6= 0 and N E(X) \ 0 is contained
in the open half-space defined by w:
N E(X) \ 0 ⊂ {α|w(α) > 0} ⊂ N1 (X)R
Proof. If w is ample, clearly w(α) ≥ 0 for α ∈ N E(X). Suppose w(α) = 0
for some nonzero α ∈ N E(X). Let z ∈ N 1 (X)R such that z(α) < 0.
Then z + tw is ample for large t, so z(α) = {z + tw}(α) ≥ 0, yielding a
contradiction.
Conversely, suppose w is a class such that
N E(X) \ 0 ⊂ {α|w(α) > 0}
Consider the function α 7→ w(α) on the closed compact set:
N E(X) ∩ {α|||α|| = 1}
Let  be the minimum value. By assumption  > 0, which yields, for any
curve C
w(C) > ||[C]||
where [C] is the class of C in N1 (X)R . Suppose that the norm in question
is X
||α|| = |wi (α)|
i

where wi is a basis of N 1 (X)

consisting of integral ample classes. (Such a
basis exists because of the open-ness of Amp(X).) This yields
w(C) > w1 (C)
Now we appeal to Corollary 8.3. 

8.6. Cones: the case of a smooth projective surface. Let X be a

smooth projective surface. The map
X X
ai Ci → O( ai Ci ), ai integers, Ci curves in X
i i
 LINE BUNDLES 39

from divisors to line bundles yields (because it is clearly surjective) an

isomorphism N1 (X)Q → N 1 (X)Q . Consider the symmetric bilinear form
B : N1 (X)Q = N 1 (X)Q → Q:
X X X X
B([ ai Ci ], [ a0j Cj0 ]) = ai a0j Ci .Cj0 = ai a0j c1 (O(Ci ))[Cj0 ]
i j i,j i,j

Let Q be the corresponding quadratic form, so that

Q(z) = B(z, z)
One of the cornerstones of the theory of surfaces is the Hodge Index Theorem
which says that the signature of Q is (1, ρ(X) − 1). More precisely, if w ∈
N 1 (X) is the class of any ample line bundle, then
(1) Q(w) = w.w > 0 (this is clear) and
(2) Q is negative definite on the orthogonal complement of w.
We extend Q and B in an obvious way to N 1 (X)R . We let C˜ denote the
open one:
C˜ = {z ∈ N 1 (X)|Q(z, z) > 0}
This has two components, one of which contains Amp(X); we let C denote
the closure of this component.
Consider now the cones:
N E(X)
⊂a

⊂
Amp(X) N ef (X)
⊂

(That Amp(X) ⊂ N E(X) is clear, and the other inclusions follow by taking
closures.)
(1) When is N ef (X) ⊂aP N E(X) not an equality?
P Clearly iff there is an
real effective divisor i ai Ci such that i ai Ci .C < 0 for some curve
C. Since C.C 0 ≥ 0 for distinct irreducible curves, this can happen
iff C = Ci for some i and C 2 < 0.
(2) Suppose there is no curve C with negative self-intersection. Then
N ef (X) = N E(X) ⊂ C. Since the interior of N ef (X) consists of
ample classes, any curve with C with zero self-intersection yields
a nonzero class [C] ∈ ∂N E(X) ∩ {z ∈ C|z 2 = 0}. Conversely, if
∂N E(X) ∩ {z ∈ C|z 2 = 0} is not empty, there exists such a curve.
(3) Suppose there does exist an irreducible curve C with negative self-
intersection. Let C≥0 ⊂ N ( X) denote the closed half-space of classes
whose intersection with C is non-negative. Claim: N E(X) is the
cone spanned by N E(X) ∩ C≥0 and C, and the half-line spanned by
40 T.R. RAMADAS

C is an extremal ray in N E(X).P Proof: Let u ∈ N E(X) \ C≥0 ,

u 6= 0. Then u is a limit of cycles i ai Ci , with ai > 0 and such that
P
i ai Ci .C < 0. At least one of the curves Ci must equal C. Now
induct.
8.7. Ruled surfaces: preliminaries. Let C be an irreducible smooth pro-
jective curve (eventually of genus g > 1), E a rank 2 vector bundle on C,
and let π : X = P(E) → C be the corresponding projective bundle. The
bundle P(E) carries a line bundle O(1), which sits in the exact sequence
0 → π ∗ (det E) ⊗ O(−1) → π ∗ E → O(1) → 0 ,
Given p ∈ C, the fibre Fp of π is the projective space of the vector space
Ep , and O(1) restricts to P(Ep ) as the corresponding tautological quotient
line bundle. A quotient:
E→`→0
with ` a line bundle, determines a section σ : C → X and a unique isomor-
phism σ ∗ O(1) → ` such that the following diagram commutes:
σ ∗ π ∗ E −−−−→ σ ∗ O(1) −−−−→ 0
 
=y
 
y
E −−−−→ ` −−−−→ 0
The exact sequence of bundles on X
0 → Θπ → ΘX → π ∗ ΘC → 0
where Θ denotes the tangent bundle and Θπ = O(2) ⊗ π ∗ (det E)−1 is the
tangent bundle along the fibres, yields a formula for the canonical bundle of
X:
KX = π ∗ {KC ⊗ (det E)} ⊗ O(−2)
Fix p ∈ C. Since the Picard group of Fp is generated by O(1)|Fp , the
restriction map P ic(X) → P ic(Fp ) is onto. On the other hand, any line
bundle from C, pulled back to X, clearly restricts to the trivial line bundle
on Fp . We will show that in fact we have an exact sequence of abelian groups
0 → P ic(C) → P ic(X) → P ic(Fp ) → 0
with O(1) providing a splitting of the restriction map P ic(X) → P ic(Fp ).
In the following discussion, we will make use of cohomology and base-
change as beautifully explained in Section 5 of Mumford’s book Abelian
Varieties. Let L be any line bundle on X. Since π is locally (on C) a product,
OX is flat over OC . Since L is locally free on OX , the sheaf L is flat over
OC , so the base-change machinery is in place. If L|Fp = O(b)|Fp , then this
must hold for all fibres (with the same value of b since the topological type
of L cannot jump), and the map p 7→ hi (L|Fp ) is constant. By Corollary 2
of Mumford, the direct images Ri π∗ L are locally free and obey base-change.
Namely,
(Ri π∗ L)p = H i (L|Fp )
 LINE BUNDLES 41

where (Ri π∗ L)p is the fibre at p of the vector bundle Ri π∗ L.

Armed with all this, consider a line bundle L on X which is trivial on
Fp . Thus L has degree zero and is therefore trivial on each fibre (because
the fibre is a genus zero curve). Since h0 (L|Fp ) = 1, the direct image π∗ L is
a line bundle. Exercise: show that the tautological map π ∗ π∗ L → L is an
isomorphism. This shows that the above sequence of Picard groups is exact
in the middle.
Suppose now that a line bundle M is pulled up from C. Note that
H 0 (C, π∗ π ∗ M ) = H 0 (C, M ⊗OX π∗ OX ) = H 0 (C, M ⊗OC OC ) = H 0 (C, M )
So if M pulled up to X becomes trivial, H 0 (C, M ) is one-dimensional, as is
H 0 (C, M̌ ), where M̌ is the dual line bundle. This can happen only if M is
itself trivial. We have therefore shown that the arrow P ic(C) → P ic(X) is
injective.
We want to describe the Neron-Severi group of X. This will turn out to
be all of the second integral second cohomology of X, so let us first deal
with this.
For a moment suppose that C is an arbitrary variety, E a vector bundle
on C with rank E = r, and X = P(E) the corresponding projective bundle.
Since the integral cohomology of any fibre Fp is generated by ξ ≡ c1 (O(1)),
the Leray-Hirsch theorem applies, and the integral cohomology of X is gen-
erated by ξ over H ∗ (C, Z). More precisely,
(1) H ∗ (X, Z) = H ∗ (C, Z)⊕H ∗ (C, Z)ξ⊕H ∗ (C, Z)ξ 2 +· · ·+H ∗ (C, Z)ξ r−1 ,
(2) the class ξ obeys:
ξ r − c1 (E)ξ r−1 + · · · + (−1)cr (E) = 0
where cl (E) ∈ H 2l (X, Z) is the lth Chern class of E. (Note that
the equation is written in the ring H ∗ (X, Z), so by cl (E) we mean
π ∗ c1 (E).)
Let us revert to the case at hand. We will suppose henceforth that det E =
OC . Since C is a curve and c1 (E) = 0 by assumption, ξ 2 = 0. The group
H 2 (C, Z) is generated by c1 (OC (p)) for p any point in C. Pulling back to
X, this is f ≡ c1 (OX (Fp )). Thus we see that
H 2 (X, Z) = Zf ⊕ Zξ = H 2 (X, Z)alg = N 1 (X)
with f 2 = 0, ξ 2 = 0, and f.ξ = 1.
8.8. Ruled surfaces: the nef cone and the cone of curves. Before
proceeding we need to make a definition:
Definition 8.6. Given a vector bundle E of rank r > 0 and degree d its
slope is the ratio
deg E
µ(E) ≡
rank E
The bundle E said to be semistable if given any (nonzero) sub-bundle E 0 ,→
E, we have µ(E 0 ) ≤ µ(E), or equivalently given a quotient bundle E  E 00 ,
42 T.R. RAMADAS

we have µ(E 00 ) ≥ µ(E). In particular if E is of rank 2 and degree zero,

it is non-semistable if there exists a quotient line bundle E  L, with
deg L = −d, d > 0. (Exercise: In the latter case the bundle L is unique.)
Consider a class af + bξ ∈ N 1 (X)R . If this is nef, its restriction to
any fibre must have non-negative degree, which forces b ≥ 0. Note also that
(ad+bξ)2 = 2ab, so the nef cone has to be contained in the positive quadrant
C = {a ≥ 0, b ≥ 0}.
Case 1: E non-semistable. Suppose now that E is non-semistable. The
destabilising quotient L determines a section σ : C → X such that σ ∗ O(1) =
L. Let C̃ ⊂ X denote the image of C by σ. Since deg ξ|C̃ = deg L|C = −d
and C̃.Fp = 1, we have [C̃] = ξ − df in N 1 (X). In particular [C̃]2 = −2d.
(This can also be seen as follows. Note that ¸(C̃)|C̃ is the normal bundle to
C̃, which in turn identified with can be identified with Θπ |C̃ = O(2)|C̃, so
C̃ 2 = deg O(C̃)|C̃ = deg O(2)|C̃ = deg σ ∗ (O(2))|C = 2deg L|C = −2d. )
If a class af +bξ is nef, its restriction to C̃ must have non-negative degree,
which forces a − bd ≥ 0, or equivalently,
b/a ≤ 1/d
We claim that in fact
N ef (X) = {a ≥ 0, b ≥ 0, and b/a ≤ 1/d}
from which we can conclude that
Amp(X) = {a > 0, b > 0, and b/a < 1/d}
To justify the claim, consider the cone of curves N E(X) ⊂ N 1 (X). This
contains the closed half-f -axis {af |a ≥ 0} as well as the half-line generated
by the class [C̃] = ξ − df . Since C̃ has negative self-intersection, it spans an
extremal ray in N E(X) which therefore is bounded by the half lines spanned
by f and ξ − df . The dual cone is exactly {a ≥ 0, b ≥ 0, and b/a ≤ 1/d}.
Case 2: E semistable. We will use the fact that, as a consequence10 of
the Theorem of Narasimhan and Seshadri, the symmetric powers S m E are
also semistable. (Exercise: show by induction that det S m E ∼ OX .) If C̃ is
a (reduced irreducible) curve in X, and not a fibre Fp , then π : C̃ → C is a
finite surjective map of degree m (say). The curve C̃ is the divisor of zeros
of a (nonzero) section of O(m) ⊗ π ∗ A for some line bundle A on C. This can
be viewed as a (nonzero) morphism from the dual line bundle: Ǎ → S m E.
This induces an injective map of bundles Ǎ(D) → S m E, where D is some
effective divisor on C, possibly trivial. By semistability of S m E, we have
deg S m E
deg Ǎ + deg D ≤ 0 =
m+1
10There are also algebro-geometric proofs avoiding appeal to the Narasimhan-Seshadri
Theorem.
In particular deg Ǎ ≤ 0, or equivalently, a ≡ deg A ≥ 0. Thus O(C̃) =
af + mξ, with m > 0, a ≥ 0, so we have proved that
N E(X) ⊂ {af + bξ|a ≥ 0, m > 0}
Since N ef (X) ⊂ N E(X) and these cones are mutually dual, it follows that
N ef (X) = N E(X) = {af + bξ|a ≥ 0, m ≥ 0}.
Here is a very helpful figure copied and pasted from Lazarsfeld’s book
(His U is our E.):
44 T.R. RAMADAS

8.9. More on the cone of curves. Can it happen that N E(X) is not
closed? As we saw above, the closure N E(X) contains the ξ-axis, but we
will show that there is no curve C such that [C] lies on this axis, provided
E is suitably chosen. If this were the case, we would have
O(C) = O(m) ⊗ π ∗ A
for some m > 0 and a line bundle of degree zero on C. As we argued above
this implies that the bundle S m E contains the degree zero line sub-bundle
(Ǎ). If we can find E such that S m E is stable for every m > 0 (and such
bundles exist for genus C > 1), this cannot happen. The existence of such
an E follows from the Narasimhan-Seshadri Theorem.
Here is a proof supplied by Narasimhan. It is a fact that there exist two
elements A1 , B1 in SU (2) that generate a dense subgroup. (In fact this
is true for SU (N ), N ≥ 2.) Let A2 , B2 , . . . , Ag , Bg be elements of SU (2)
defined by A2 = B1 , B2 = A1 and Ai = Bi = I, i > 2. Then
Y
Ai Bi (Ai )−1 (Bi )−1 = I
i
so we have a representation of the fundamental group of C in SU (2). Since
the symmetric powers of the standard two-dimensional representation (on
C2 ) of SU (2) are irreducible representations of SU 2), the induced represen-
tations of the fundamental group in SU (m+1) = SU (S m C2 ) are irreducible.

Continuing with the above example, if X = P(E) with E stable of degree

zero, then O(1) has positive degree on every curve but is not ample. For,
given any curve C we have
[C] = aξ + bf
with b > 0. All this is mostly due to Mumford.