MAT1360: Complex Manifolds and

Hermitian Differential Geometry

University of Toronto, Spring Term, 1997
Lecturer: Andrew D. Hwang

Contents
1 Holomorphic Functions and Atlases
1.1 Functions of Several Complex Variables . . . . . . . . . . . . . . . . . . . . .
1.2 Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
4

2 Almost-Complex Structures and Integrability

2.1 Complex Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Almost-Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Integrability Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9
10
11
15

3 Sheaves and Vector Bundles

3.1 Presheaves and Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20
21
24

4 Cohomology

4.1 Cech
Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Dolbeault Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Elementary Deformation Theory . . . . . . . . . . . . . . . . . . . . . . . . .

30
31
35
39

5 Analytic and Algebraic Varieties

5.1 The Local Structure of Analytic Hypersurfaces . . . . . . . . . . . . . . . . .
5.2 Singularities of Algebraic Varieties . . . . . . . . . . . . . . . . . . . . . . . .

40
43
45

6 Divisors, Meromorphic Functions,

6.1 Divisors and Line Bundles . . . .
6.2 Meromorphic Functions . . . . . .
6.3 Sections of Line Bundles . . . . .

52
52
53
55

and Line
. . . . . .
. . . . . .
. . . . . .
i

Bundles
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .

6.4

Chows Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

56

7 Metrics, Connections, and Curvature

7.1 Hermitian and Kahler Metrics . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Connections in Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . .

60
61
64

8 Hodge Theory and Applications

8.1 The Hodge Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 The Hodge Decomposition Theorem . . . . . . . . . . . . . . . . . . . . . . .

71
71
77

9 Chern Classes
9.1 Chern Forms of a Connection . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Alternate Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82
84
85

10 Vanishing Theorems and Applications

10.1 Ampleness and Positivity of Line Bundles and Divisors
10.2 The Kodaira-Nakano Vanishing Theorem . . . . . . . .
10.3 Cohomology of Projective Manifolds . . . . . . . . . .
10.4 The Kodaira Embedding Theorem . . . . . . . . . . . .
10.5 The Hodge Conjecture . . . . . . . . . . . . . . . . . .

90
90
91
93
96
97

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

11 Curvature and Holomorphic Vector Fields

100
11.1 Ricci Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
11.2 Holomorphic Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Metrics With Special Curvature

105

12 Einstein-K
ahler Metrics
105
12.1 The Calabi Conjectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.2 Positive Einstein-Kahler Metrics . . . . . . . . . . . . . . . . . . . . . . . . . 109

ii

Preface
These notes grew out of a course called Complex Manifolds and Hermitian Differential
Geometry given during the Spring Term, 1997, at the University of Toronto. The intent
is not to give a thorough treatment of the algebraic and differential geometry of complex
manifolds, but to introduce the reader to material of current interest as quickly as possible.
As a glance at the table of contents indicates, Part I treats standard introductory analytic material on complex manifolds, sheaf cohomology and deformation theory, differential
geometry of vector bundles (Hodge theory, and Chern classes via curvature), and some applications to the topology and projective embeddability of Kahlerian manifolds. The intent
is to provide a number of interesting and non-trivial examples, both in the text and in the
exercises. Some details have been skipped, such as the a priori estimates in the proof of the
Hodge Theorem. When details are omitted, I have tried to provide ideas of proofs, particularly when there is geometric intuition available, and to indicate what needs to be proven
but has not been.
Part II is a fairly detailed survey of results on Einstein and extremal Kahler metrics from
the early 1980s to the present. It is hoped that this exposition will be of use to young
researchers and other interested mathematicians and physicists by collecting results and
references in one place, and by pointing out open questions. The results described in Part II
are due to T. Aubin, S. Bando, E. Calabi, S. Donaldson, A. Futaki, Z. D. Guan, N. Hitchin,
S. Kobayashi, N. Koiso, C. Lebrun, T. Mabuchi, A. M. Nadel, H. Pedersen, Y.-S. Poon,
Y. Sakane, S. R. Simanca, M. F. Singer, Y.-T. Siu, G. Tian, K. Uhlenbeck, S.-T. Yau. I
offer my sincere apologies to authors whose work I have overlooked.
In Part I, my debt to the book of Griffiths-Harris is great, and to books of several other
authors is substantial. The bibliography lists, among other works, the books from which the
course packet was drawn. I hope readers find the exercises useful; while there are texts at
this level which contain exercises, it seems there are few which deal with the specific but
colourful examples scattered though folklore and the literature.
I have taken some care to ensure that the notationincluding signs and other constant
factorsis internally consistent, and maximally consistent with other works. Occasionally
a concept is introduced informally, in which case the term being defined in enclosed in
quotation marks. The subsequent formal definition contains the term in italics. The
following lists the end-of symbols: occurs at the end of proofs, 2 denotes the end of an
example or remark, and  signifies the end of an exercise.

iii

Holomorphic Functions and Atlases

A function f : D C of one complex variable is (complex) differentiable in a domain D if

the ordinary Newton quotient
f (w) f (z)
wz
wz

f 0 (z) := lim

exists for every point z D. For present purposes, there are two other useful characterizations of this condition. The first is to identify the complex line C with the real plane R2 .
The function f is complex differentiable if and only if the associated function f : D R2
has complex-linear derivative at every point, in which case f is said to be holomorphic.
Concretely, there is a ring homorphism


a b
a + bi C ,
R22 ,
b a
so f 0 = Df is complex-linear if and only if u = Re f and v = Im f satisfy the CauchyRiemann equations.
On the other hand, if f is holomorphic in a disk of radius > r centered at z0 , then for all
z with |z z0 | < r, the Cauchy integral formula gives
Z
1
f (w) dw
f (z) =
.
2i |wz0|=r w z
Writing 1/(w z) as a geometric series in z z0 and integrating term-by-term shows that
a holomorphic function may be expressed locally as a convergent power series. In words, a
holomorphic function is complex-analytic. Intuitively, the averaging process effected by the
contour integral makes the integrand smoother; if f is of class C k , then the expression on
the right is of class C k+1 . Since f (times a smooth function) is the integrand, f itself must
be smooth. This is the prototypical bootstrap argument, and perhaps the most elementary
example of elliptic regularity.

1.1

Functions of Several Complex Variables

For functions of more than one variable, much of this philosophy carries over by the same
reasoning. Let D Cn be an open set. A function f : D C is holomorphic if the
Cauchy-Riemann equations hold on D. More precisely, write z = x + iy and f = u + iv
with u and v real-valued. Then u and v may be regarded as functions on a subset of R2n ,
and f is holomorphic if f is of class C 1 and
(1.1)

u
v
=
,
x
y

u
v
=

y
x
1

at each point of D. Holomorphicity is related to separate holomorphicity (Osgoods

Lemma, Proposition 1.1 below), that is, holomorphicity of the functions obtained by fixing n 1 of the variables and varying the remaining one. The continuity hypothesis may be
dropped (Hartogs Theorem), though the proof becomes substantially more difficult.
Proposition 1.1 Let D Cn be a non-empty open set. If f : D C is continuous and
separately holomorphic, then f is holomorphic.
Let r = (r1 , . . . , rn ) be a radius, that is, an n-tuple of positive real numbers, and let
z0 = (z01 , . . . , z0n ) Cn . If r and r 0 are radii, then r 0 < r is taken to mean r0 < r for
= 1, . . . , n. The polydisk of radius r centered at z0 is, by definition,
r (z0 ) = {z Cn : |z z0 | < r

for = 1, . . . , n} = {z Cn : |z z0 | < r}.

Thus a polydisk is exactly a Cartesian product of ordinary disks. While polydisks are not
generally domains of convergence for power series of several variables, they are nonetheless
the most convenient sets to use for local purposes.
Let D be a non-empty open set in Cn . A function f : D C is complex analytic if,
for every z0 D, there is a complex power series centered at z0 which converges and is
equal to f on some polydisk r (z0 ). In order to avoid purely notational complications, it is
convenient to use multi-indices. If I = (i1 , . . . , in ) is a multi-index, then set
|I| = i1 + + in ,

z I = (z 1 )i1 (z n )in ,

fI =

f
k f
.
=
z I
(z 1 )i1 (z n )in

Analyticity means there is a polydisk r (z0 ) such that

(1.2)

f (z) =

X
X
1
fI (z0 )(z z0 )I
k!
k=0
|I|=k

for all z r (z0 ). As in the case of one variable, holomorphicity and analyticity are
equivalent, as is seen by using (an obvious generalization of) the Cauchy integral formula.
The concepts of holomorphicity and analyticity extend in the obvious way to functions
with values in Cm , which are usually called holomorphic maps. A holomorphic map
between open subsets of Cn which possesses a holomorphic inverse is a biholomorphism.
Remarkably, a one-to-one holomorphic map between open subsets of Cn is a biholomorphism,
see Theorem 5.3; this result has no analogue in the smooth category, even for real polynomial
maps, as is shown by x 7 x3 . The set of biholomorphisms between open subsets of Cn is
a pseudogroup: The composite of two biholomorphisms is a biholomorphism wherever it is
defined. It is sometimes useful to consider anti-holomorphic maps. These are exactly complex
conjugates of holomorphic maps. The set of anti-holomorphic maps is not a pseudogroup,
since a composite of two anti-holomorphic maps is holomorphic.
2

Holomorphic functions of n > 1 variables satisfy a version of the identity theorem, see
Proposition 1.2 below. However, it is important to emphasize that a holomorphic function
may vanish on an infinite set without being identically zero. For example, let be a polydisk
centered at the origin in C2 . The function z 1 vanishes along the z 2 -axis, which is real-linearly
isomorphic to R2 . On the other hand, if f : C vanishes on R2 , then f 0 on , as
is easily verified by inspecting the coefficients in the series expansion of f . The statement of
Proposition 1.2 is certainly not the strongest possible, but is adequate for present purposes.
Proposition 1.2 Let f : C be a holomorphic function on a polydisk, and suppose
f |U 0 for some non-empty open set U . Then f 0.
Holomorphic functions satisfy a maximum principle, as may be seen by restricting to
lines and applying the one-variable version.
Proposition 1.3 Let be a polydisk. If f : C is holomorphic and if |f | has a local
maximum at p , then f is constant on .
A remarkable extension result for holomorphic functions of n > 1 variables is Hartogs
Phenomenon. There is no analogous result for holomorphic functions of one variable.
Proposition 1.4 Let be a polydisk in Cn , n 2, and let K be a compactly contained
subset. If f : \K C is holomorphic, then there exists a holomorphic function f : C
which extends f .
proof (Sketch) There exists a polydisk 0 = r (z0 ) which compactly contains K and is
compactly contained in . The function

n Z
1
f (w) dw

f (z) =
2 1
|wz0 |=r w z
is defined for all z r (z0 ), holomorphic, and is equal to f whenever both are defined.
In words, the idea is to use the Cauchy integral formula to extend f formally, then to
argue that the expression obtained is in fact a holomorphic extension of f . The latter is
accomplished by picking a slice of parallel to a coordinate hyperplane such that the slice
misses the closure of K. (It is in choosing such a slice that the hypothesis n > 1 is used.)
The formal extension of f coming from the Cauchy integral formula therefore agrees with f
on a neighborhood of the slice.
Topologically, Hartogs phenomenon is a consequence of the following observation: If 0
and are polydisks of dimension n with 0 , then \ 0 has the homotopy type of a
(2n 1)-sphere, and is therefore n-connected when n > 1. In this event, for every z 0
the closed n-form
z (w) :=

f (w) dw
dw 1
dw n
= f (w 1, . . . , w n ) 1

,
wz
w z1
wn zn
3

w \ 0 ,

has a primitive, so the integral of z over a real n-torus in \ 0 is independent of the choice
of torus, and defines a holomorphic function of z which extends f . By contrast, if n = 1,
then existence of a primitive is not automatic.
Let f : C be holomorphic on a polydisk in Cn , n 2. Proposition 1.4 implies, in
particular, that f cannot have an isolated singularity, nor can it have an isolated zero since
then 1/f would have an isolated singularity. More generally, the zero set of f cannot cannot
be compact, and cannot have complex codimension greater than one.

1.2

Complex Manifolds

A complex manifold is a smooth manifold, locally modelled on the complex Euclidean space
Cn and whose transition functions are holomorphic. More precisely, a complex manifold is a
pair (M, J) consisting of a smooth, real manifold of real dimension 2n and a maximal atlas
whose overlap maps lie in the pseudogroup of biholomorphic maps between open subsets of
Cn briefly, a holomorphic atlas. There are various other ways of specifying the same data,
which are investigated below.
Not every 2n-dimensional manifold admits a holomorphic atlas, and a single smooth
manifold may admit many inequivalent holomorphic atlases. Generally, determination of
the set of holomorphic atlases up to equivalence on a particular smooth manifold is extremely
difficult, even if the manifold is compact. The most famous open question along these lines
concerns (non-)existence of a holomorphic atlas on the six-dimensional sphere, but there are
other open questions of greater interest which are almost as easily stated. Further details
are deferred until more tools and terminology are available.
A map between complex manifolds is holomorphic if, with respect to arbitrary charts,
the induced map is holomorphic. More precisely, f : M M 0 is holomorphic at p M if
there exists a chart (, U) near p and a chart (, V ) near f (p) M 0 such that f 1
is a holomorphic map between open subsets of complex Euclidean spaces. This condition is
independent of the choice of charts because overlap maps are biholomorphic.
A basic consequence of the maximum principle Proposition 1.3 is that every holomorphic
function on a connected, compact complex manifold is constant; the absolute value must
have a maximum value by compactness, so the function is locally constant by the maximum
principle, hence globally constant since the manifold is connected. If M CN is a complex
submanifold, then each coordinate function on CN restricts to a global holomorphic function
on M. In particular, there is no holomorphic analogue of the Whitney embedding theorem;
the only connected, compact complex manifold which embeds holomorphically in CN is a
point.
A complex manifold which embeds as a closed submanifold in a complex Euclidean space
is called a Stein manifold. The study of Stein manifolds falls most naturally into the realm
of several complex variables, though affine varieties are of interest in algebraic geometry
as well.
4

There are three commonly considered equivalence relations between complex manifolds,
each of which is strictly weaker than the previous one.
Complex manifolds M0 and M are said to be biholomorphic if there exists a holomorphic
map f : M0 M with holomorphic inverse. Assertions regarding uniqueness of
complex structure on a fixed manifold M are always meant up to biholomorphism
unless otherwise specified. As noted above, a single smooth manifold may admit many
non-biholomorphic complex structures, and a topic of intense current research is the
study of moduli spaces of complex structures on fixed smooth manifolds.
Complex manifolds M0 and M are deformation equivalent if there exists a complex
manifold X and a holomorphic submersion : X , C the unit disk, with
1 (0) = M0 and 1 (t) = M for some t .
If the underlying smooth manifolds of M0 and M are diffeomorphic, then the complex
manifolds themselves are said to be diffeomorphic.
Biholomorphic manifolds are obviously deformation equivalent; take X = M . It is
not difficult to see that deformation equivalent manifolds are diffeomorphic, but the proof is
deferred to the systematic introduction to deformation theory. Examples below show that
neither of these implications is reversible in general.

Examples
Example 1.5 Euclidean space Cn is a complex manifold. More interesting examples are
gotten by dividing by a lattice (i.e. a finitely generated discrete subgroup) Cn . Since
acts on Cn by translation and this action is properly discontinuous and holomorphic, the
quotient space Cn / inherits the structure of a complex manifold from the standard atlas on
Cn . If is generated by an R-basis of Cn , then the quotient is a compact manifold, called
a compact complex torus. Although all compact tori are diffeomorphic to the real 2n-torus,
their complex-analytic properties depend on arithmetic properties of the lattice.
Generally, if a group acts properly discontinuously by biholomorphisms on a complex
manifold M, then the quotient M/ inherits a complex structure from M. Another class of
examples is the family of Hopf manifolds: Let n > 1, and let be a complex number with
|| > 1. Consider the action of ' Z on Cn \ 0 generated by the map z 7 z. The quotient
is a compact complex manifold diffeomorphic to S 1 S 2n1 . The complex analytic properties
of general Hopf surfaces are investigated in Exercise 2.3.
2
Example 1.6 Open subsets of Cn are of course complex manifolds, and some of them are
important or otherwise remarkable. A complex torus is a manifold biholomorphic to (C )n
(cf. Example 1.5). A complex torus has the structure of a complex Lie group; equivariant
compactifications form the intensively-studied class of toric manifolds.
5

The general linear group GL(n, C) Cnn is a complex Lie group under matrix multiplication. This manifold has various closed complex subgroups, such as SL(n, C) (matrices
of unit determininant) and O(n, C) (complex orthogonal matrices). Compact groups such as
U(n) and SU(n) are not complex Lie groups, nor are they complex submanifolds of GL(n, C).
In fact, a compact, connected, complex Lie group is a compact torus. This is not trivial,
though it is easy to see that such a group is Abelian: the adjoint representation must be
trivial, since it may be regarded as a map from a compact complex manifold into a complex
Euclidean space.
Convex open sets in Cn , n 2, exhibit subtle analytic behaviour; slightly deforming
the boundary of the unit ball gives an uncountable family of mutually non-biholomorphic
complex structures, for example.
2
Example 1.7 One of the most important compact complex n-manifolds is the complex
projective space Pn . Intuitively, a point of Pn is a line through the origin in Cn+1 . More
precisely, the group C acts on Cn+1 \ 0 by scalar multiplication. If the orbit space is given
the quotient topology, then the complex structure of Cn+1 descends. The equivalence class
of a point Z = (Z 0 , . . . , Z n ) Cn+1 \ 0 is denoted [Z] = [Z 0 : : Z n ], and the Euclidean
coordinates of Z constitute so-called homogeneous coordinates of [Z]. While Z is not a welldefined holomorphic function on Pn , the equation Z = 0 is unambiguous. Furthermore,
every quotient Z /Z is well-defined, and holomorphic except where Z = 0. There is an
atlas consisting of n + 1 charts. For each = 0, . . . , n, let U = {[Z] Pn : Z 6= 0}, and
use local coordinates
Z0
Zn
z0 = , . . . , zb , . . . , zn = .
Z
Z
On U := U U , the overlap mapessentially multiplication by Z /Z is holomorphic,
so Pn admits the structure of a complex manifold. To see that Pn is compact, observe that
the unit sphere in Cn+1 is mapped onto Pn by the quotient map.
If V is a finite-dimensional complex vector space, then the projectivization of V , denoted
P(V ), is formed as above by removing the origin and dividing by the action of C . This
construction, while less concrete than the construction of Pn = P(Cn+1 ), captures functorial
properties of V , and is ultimately a better way to view projective space.
Many concepts from linear algebra (linear subspaces and spans, intersections, and the
language of points, lines, and planes) carry over in the obvious way to projective space; for
example, the line xy determined by a pair of points in Pn is the image of the plane in Cn+1
spanned by the lines representing x and y. Disjoint linear subspaces of Pn are said to be
skew. For example, there exist pairs of skew lines in P3 , while every line in P3 intersects
every plane in P3 in at least one point. A pair of skew linear subspaces of Pn is maximal
if the respective inverse images in Cn+1 are of complementary dimension in the usual sense.
The prototypical maximal skew pairs in Pn+1 are indexed by an integer k = 0, . . . , n, and

are of the form

{[X 0 : : X k ], [Y 0 : : Y nk ]} Pk t Pnk 7 [X 0 : : X k : Y 0 : : Y nk ] Pn+1 .
(Linear Maps of Projective Spaces) Every linear automorphism of Cn+1 induces a biholomorphism from Pn to Pn ; Proposition 6.12 below asserts, conversely, that every automorphism of Pn is induced by a linear automorphism of Cn+1 . Other linear transformations on
Cn+1 descend to interesting holomorphic maps defined on subsets of Pn . Simple but geometrically important examples are furnished by projection maps. Let {P1 , P2 } be a maximal
skew pair of linear subspaces. There is a holomorphic map : Pn \P1 P2 , called projection
away from P1 onto P2 , with the following geometric description. For each point x 6 P1 , the
linear span xP1 intersects P2 in a unique point (x), which is by definition the image of x
under projection. There is an algebraic description, namely that such a projection is exactly
induced by a linear projection (in the usual sense) on Cn+1 ; see Exercise 1.1.
(Submanifolds of Projective Space) A closed complex submanifold of Pn is called a projective manifold. There is an intrinsic necessary and sufficient criteriongiven by the Kodaira
Embedding Theoremfor a compact complex manifold to be projective; see Theorem 10.10
below. Hopf manifolds do not satisfy this criterion, while compact complex tori are projective if and only if the lattice satisfies certain (explicit) arithmetic properties. A compact,
projective torus is an Abelian variety.
A projective algebraic variety is the image in Pn of the common zero set of a finite
set of homogeneous polynomials on Cn+1 . Every smooth projective algebraic variety is a
compact complex manifold. Remarkably (Chows Theorem, 6.13 below), the converse is true:
every compact complex submanifold of Pn is the zero locus of a finite set of homogeneous
polynomials in the homogeneous coordinates.
2
Example 1.8 A Riemann surface or complex curve is a one-dimensional complex manifold.
Apart from the rational curve P1 , the simplest curves are elliptic curves, namely quotients
of C by a lattice of rank two. Let 1 and 2 be generators of the lattice ; thus 2 /1 =
is non-real, and without loss of generality has positive imaginary part.
Suppose E1 and E2 are elliptic curves, and write Ei = C/i. Assume there is a nonconstant holomorphic map f : E1 E2 . Then f lifts to an entire function, denoted f. The
derivative f0 : C C is doubly-periodic, hence constant by Liouvilles Theorem. Thus f is
affine: There exist complex numbers 6= 0 and such that f(z) = z+, and by translating
if necessary, = 0 without loss of generality. Multiplication by is a homomorphism of
additive groups from the lattice 1 to the lattice 2 , and is onto (i.e. is an isomorphism of
lattices) if and only if the map f : E1 E2 is a biholomorphism. The curve associated to
the lattice generated by 1 and 2 is therefore biholomorphic to the curve associated to the
lattice generated by 1 and = 2 /1 .
Let i be the lattice generated by 1 and i . Subject to this normalization, 1 = 2 if
and only if there is a modular transformation of the upper half plane h carrying 1 to 2 .
7

More concretely, this is the case exactly when there exist integers a, b, c, and d with
(1.3)

a1 + b
= 2 ,
c1 + d

ad bc = 1.

The orbit space of h under this action of SL(2, Z) has the structure of a one-dimensional
complex manifold except at the two points stabilized by non-trivial elements of SL(2, Z)
so-called orbifold pointswhere the local structure is that of a disk divided by the action
of a finite cyclic group of rotations. The orbits are in one-to-one correspondance with biholomorphism classes of elliptic curves, and the orbit space is the moduli space of elliptic
curves.
To each lattice of rank two is associated a Weierstrass -function, defined by

X
1
1
1
(1.4)
(z) = 2 +

.
z
(z )2 2

The following facts are not difficult toPestablish. (See, for example, L. Ahlfors, Complex
2k
(the sum being taken over ), the
Analysis, pp. 272 ff.) Setting Gk =
6=0
-function satisfies the first-order differential equation
(1.5)

0 (z)2 = 4(z)3 60G2 (z) 140G3 .

Consequently, the elliptic curve C/ embeds as a cubic curve in P2 via the mapping
z 6 7 [1 : (z) : 0 (z)],

z 7 [0 : 0 : 1].

Consider the action of = Z Z on h C defined by m,n (, z) = (, z + m + n ).

The quotient is a (non-compact) complex surface S equipped with a holomorphic projection
map : S h whose fibres are elliptic curves. Indeed, the fibre over h is the elliptic
curve associated to the lattice generated by 1 and . Observe that while distinct fibres are
diffeomorphicin fact, are deformation equivalentthey are not necessarily biholomorphic.
An algebro-geometric version of this picture is easily constructed from the modular invariant : h C \ {0, 1}; the surface S is thereby realized as the zero locus in P2 h of a
cubic polynomial whose coefficients are analytic functions of .
2
To avoid trivial exceptions to assertions about existence of holomorphic maps, all manifolds are assumed from now on to have dimension greater than zero unless otherwise specified.
In the study of real manifolds, a basic tool is existence of smooth submanifolds passing
through an arbitrary point, and having arbitrary tangent space. The rigidity of the holomorphic category makes this tool available only to a limited extent for complex manifolds.
If C is the unit disk, M is a complex manifold, and (p, v) T M is an arbitrary one-jet,
then there need not exist a holomorphic map f : M with f (0) = p and f 0 (0) = v,
8

though it is always possible to arrange that f 0 (0) = v for  1. It is usually difficult to determine whether or not a complex manifold M 0 embeds holomorphically in another complex
manifold M. Even if M 0 is one-dimensional, existence of an embedding depends in a global
way on the complex structure of M; the prototypical result is Liouvilles theorem, which
asserts that every bounded, entire function (a.k.a. holomorphic map f : C ) is constant.
Existence of compact holomorphic curves in M has been an area of active interest since the
mid-1980s, following the work of Mori in complex geometry and the work of Gromov and
McDuff in symplectic geometry. It is also of interest to determine whether or not there exist
embeddings of C into M; this is related to the study of hyperbolic complex manifolds and
value distribution theory.

Exercises
Exercise 1.1 A projection on Cn+1 is a linear transformation with 2 = . Prove
that every such linear transformation induces a holomorphic mapprojection away from
P1 = P(ker ) onto P2 = P(im )as described in Example 1.7. In particular, if has
rank ` + 1 as a linear transformation, then after a linear change of coordinates projection
away from P(ker ) has the form
[Z] = [Z 0 : : Z n ] 7 [Z 0 : : Z ` : 0 : : 0],
and the image is P` Pn .

Exercise 1.2 Let p : Cn+1 \ 0 Pn be the natural projection. Prove that there is no
holomorphic map s : Pn Cn+1 \ 0 with p s = identity. (In fact, there is no continuous
map with this property, but the latter requires some algebraic topology.)


Exercise 1.3 Give an example of a non-compact complex manifold M such that every
holomorphic function on M is constant.

Exercise 1.4 Let fe : Cn+1 be a non-constant holomorphic map of the unit disk into
Cn+1 with fe(0) = 0. Prove that the induced map f : Pn on the punctured unit disk
extends to the origin.


Exercise 1.5 Fix h, and let E = E be the elliptic curve associated to the lattice
generated by 1 and . Find all biholomorphisms f : E E, that is, all automorphisms
of E. Observe that the curves corresponding to orbifold points of the moduli space have
automorphisms arising from complex multiplication.


Almost-Complex Structures and Integrability

In order to apply the machinery of differential geometry and bundle theory to the study of
complex manifolds, it is useful to express holomorphic atlases in bundle-theoretic terms. The
9

first task is to study pointwise objects, that is, to construct complex linear algebra from
real linear algebra. The constructions obtained are then applied fibrewise to tensor bundles
over smooth manifolds equipped with some additional structure.

2.1

Complex Linear Algebra

Let V be an m-dimensional real vector space. An almost-complex structure on V is an

operator
J : V V with J 2 = I. Complex scalar multiplication is defined in terms of J
by (a + b 1)v = av + b Jv. The operator J is also an almost-complex structure on V ,
called the conjugate structure, and the space (V, J) is often denoted V for brevity.
The

standard complex vector space is V = Cn with J induced by multiplication by 1.

Lemma 2.1 If V admits an almost-complex structure, then V is even-dimensional and has
an induced orientation.
proof Taking the determinant of J 2 = I gives (det J)2 = (1)m , from which it follows
m = 2n is even. If {ei }ni=1 is chosen so that {ei , Jei }ni=1 is a basis of V , then the sign of the
volume element
e1 Je1 en Jen
is independent of {ei }ni=1 . (See also Remark 2.2 and the remarks following Proposition 2.5
below.)
The complexification of a real vector space V is VC := V C, with C regarded as a real
two-dimensional vector space. It is customary to write v 1 = v and v i = iv. If V has an
almost-complex structure J, then J extends to VC by J(v ) = Jv , and VC decomposes
into i-eigenspaces:
V 1,0 = {Z VC : JZ = iZ} = {X iJX : X V },
V 0,1 = {Z VC : JZ = iZ} = {X + iJX : X V }.
The complex vector space (V, J) is C-linearly isomorphic to (V 1,0 , i) via the map
(2.1)

1
X = 2 Re Z 7 (X iJX) = Z =: X 1,0 .
2

Similarly, V = (V, J) is C-linearly isomorphic to (V 0,1 , i). Complex conjugation induces

a real-linear isomorphism of VC which exchanges V 1,0 and V 0,1 . The fixed point set is
exactly the maximal totally real subspace V = V 1. Topologists sometimes define the
complexification of (V, J) to be V V together with the real structure which exchanges the
two factors.

10

If V is equipped with an almost-complex structure J, then the dual pairing induces an

almost-complex structure on V also denoted by Jvia
(2.2)

hJv, i = hv, Ji.

The associated eigenspace decomposition of VC = V C is

V1,0
= { VC : J = i} = { + iJ : V },

V0,1
= { VC : J = i} = { iJ : V }.

0,1

1,0
By equation (2.2), the space
V V1,0 is the annihilator of V ; similarly V0,1 annihilates V .
The exterior algebra VV has a decomposition
into tensors of type (p, q), namely, fully
V

skew-symmetric elementsVof p V1,0

q V0,1
. For convenience, the space of skew-symmetric
(p, q)-tensors is denoted p,q V .
Remark 2.2 The standard complex vector space Cn has coordinates
z = (z 1 , . . . , z n ).

There are two useful (real-linear) isomorphisms with R2n ; if z = x + 1y with x and
y real, then z Cn is associated with either

(2.3)

(x, y) = (x1 , . . . , xn , y 1, . . . , y n )

or

(x1 , y 1 , . . . , xn , y n)

in R2n . With respect to the first set of real coordinates, a complex-linear transformation is
represented by a 2 2 block matrix with n n blocks (see the remarks following Proposition 2.5 below). With respect to the second set of real coordinates, a complex-linear
transformation is represented by an n n block matrix of 2 2 blocks. This representation
is preferable when working with metrics and volume forms.
2

2.2

Almost-Complex Manifolds

An almost-complex manifold is a smooth manifold M equipped with a smooth endomorphism

field J : T M T M satisfying Jx2 = Ix for all x M. The linear algebra introduced above
may be applied pointwise to the tangent bundle of M. The complexified tangent bundle is
TC M = T M C, where C is regarded as a trivial vector bundle. The tensor field J splits
the complexified tangent bundle into bundles of eigenspaces
TC M = T 1,0 M T 0,1 M,
and the smooth complex vector bundles (T M, J) and (T 1,0 M, i) are complex-linearly isomorphic. Complex conjugation induces an involution of TC M which exchanges the bundles
of eigenspaces. A (local) section Z of T 1,0 M is called a vector field of type (1, 0), though Z
is not a vector field on M in the sense of being tangent to a curve in M. If ordinary tangent
vectors are regarded as real differential operators, then (1, 0) vectors are complex-valued
differential operators.
11

The splitting of the set of complex-valued skew-symmetric r-tensors into skew-symmetric

(p, q)-tensors gives rise to spaces of (p, q)-forms. If Ar and Ap,q denote the space of smooth
r-forms and the space of smooth (p, q)-forms respectively, then
M
Ar =
Ap,q .
p+q=r

In local coordinates, Ap,q is generated by the forms dz I1 d

z I2 with |I1 | = p and |I2 | = q.
Example
2.3 Complex Euclidean space Cn is an almost-complex manifold. Explicitly, let

z = x + 1y be the usual coordinates on Cn , identified with coordinates (x, y) on R2n .

The real tangent bundle and its complexification have the standard frames








1
1

,
,
,
=
1 ,
=
+ 1
x y
z
2 x
y
z
2 x
y
while the real cotangent bundle and its complexification have coframes

z = dx 1dy }.
{dx , dy },
{dz = dx + 1dy , d

Multiplication by 1 acts only on tangent spaces, not on the actual coordinates. Thus
(2.4)

= ,

x
y

= .

y
x

The tensor field J has constant components with respect to a holomorphic coordinate system.
The exterior derivative operator d : Ar Ar+1 maps Ap,q to Ap+1,q Ap,q+1 , and the
On functions,
corresponding operators are denoted and .
df =

f
f

dz +
d
z =: f + f.
z
z

= 0. More generally, a holomorphic p-form

Observe that f is holomorphic if and only if f
= 0.
is a (p, 0)-form with
2
Example 2.4 It is natural to ask which spheres admit almost-complex structures. It is
not difficult to show that if S n admits an almost-complex structure, then T S n+1 is trivial.
Deep results from algebraic topology imply that n = 2 or n = 6 (the case n = 0 being
uninteresting).
Conversely, the spheres S 2 and S 6 actually admit almost-complex structures. For S 2 ,
rotate each tangent plane is through an angle /2, in a fashion consistent with a choice
of orientation. The following algebraic description of this almost-complex structure considerably illuminates the corresponding construction for S 6 . Consider the usual embedding
S 2 R3 ; a point x S 2 may be regarded as a unit vector in R3 , and via the cross product
12

induces a linear transformation Jx (y) = x y on R3 . If y x (i.e. if y Tx S 2 ), then

(x y) x as well, so this linear transformation is an endomorphism Jx of the tangent
space Tx S 2 . It is easy to verify that Jx2 = I, so that this field of endomorphisms is an
almost-complex structure on S 2 . Regarding R3 as the space of pure imaginary quaternions,
the cross product arises as the imaginary part of quaternion multiplication. Thus the almostcomplex structure on S 2 is intimately tied to existence of a division algebra structure on R4 ,
which in turn leads to triviality of T S 3 .
Similarly, multiplication by a pure imaginary Cayley number defines a cross product
on R7 , and the analogous construction defines an almost-complex structure on S 6 . The
set of almost-complex structures on S 6 is very large, since small perturbations of a given
structure give rise to another (usually non-equivalent) structure. By contrast, S 2 has only
one almost-complex structure up to equivalence.
2
0
0
A map f : (M, J) (M , J ) between almost-complex manifolds is almost-complex or
pseudoholomorphic if (f )J = J 0 (f ). The following is a straightforward consequence of the
Chain Rule and the Cauchy-Riemann equations.
Proposition 2.5 Let Cn be a polydisk. A map f : Cm is pseudoholomorphic if
and only if f is holomorphic.
There are several useful applications, of which two deserve immediate mention:
An almost-complex manifold has a natural orientation.
A complex manifold has a natural almost-complex structure.
To prove the first assertion, suppose m = n and f is pseudoholomorphic. Using the isomorphism z Cn (x, y) R2n , and writing f = u + iv with u and v real-valued, the
Jacobian f has matrix


 



0
ux uy
A B
A + 1B

,
=
C
vx vy
B
A
0
A 1B
with A, B Rnn . This matrix has positive determinant, proving that an almost-complex
manifold is orientable. The natural orientation is by definition the orientation compatible
with an ordered basis {e1 , Je1 , . . . , en , Jen }, cf. the proof of Lemma 2.1.
To prove the second assertion, let M be a complex manifold, and let (, U) be a chart
near x M. The standard almost-complex structure of Cn induces an almost-complex
structure on U, and by Proposition 2.5 this almost-complex structure is well-defined.
Combining these observations, a mapping f : M M 0 between complex manifolds
is holomorphic if and only if f is pseudoholomorphic with respect to the induced almostcomplex structures.
13

Almost-Complex Structures, Orientations, and Conjugate Atlases

When (M, J) is a complex manifold, J is tacitly taken to be the induced almost-complex
structure and the orientation is taken to be that induced by J. In the absence of such
a convention, there are potentially confusing relationships between almost-complex structures, choices of orientation, and pseudoholomorphic maps, which the following remarks are
intended to clarify. The distinction is not merely pedantic; increasingly in symplectic geometry and geometric field theory it is useful to fix an orientation on M, then seek compatible
almost-complex structures.
Let M be a real 2n-dimensional smooth manifold admitting an almost-complex structure
J : T M T M. There is an induced orientation as just observed; this may be realized
concretely as a non-vanishing smooth 2n-form on M. Let M+ denote the corresponding
oriented 2n-manifold, and M denote the oppositely oriented manifold. There is also a
conjugate almost-complex structure J : T M T M, and the pair (M, J) equipped with
the induced orientation is denoted M.
There are four choices of orientation and almost-complex structure on the smooth manifold M, which will be denoted (see also Remark 2.6 below)
M := (M+ , J),

M := (M , J),

M + := (M+ , J),

and M := (M , J).

The pair (M+ , J) is compatible in the sense that the orientation is induced by the almostcomplex structure. Since the orientation induced by J is (1)n times the orientation
induced by J, either the third or fourth pair is compatiblei.e. is equal to M depending
on whether the complex dimension n is even or odd.
If (M, J) is a complex manifold, then there is a holomorphic atlas J on the smooth
manifold M whose charts are complex conjugates of the charts in J. More precisely, if (U, )
is a chart in J, then (U, ) is, by fiat, a chart in J. Observe that J is a holomorphic atlas
because the overlap maps are holomorphic. Further, it is clear that if J is the almost-complex
structure induced by J, then J is induced by J. Suppressing almost complex structures,
if M is a complex manifold, then M is a complex manifold having the same underlying
smooth manifold. The identity map is an antiholomorphic diffeomorphism from M to M .
As complex manifolds, M and M may or may not be biholomorphic, see Exercises 2.4 and
2.5. An overkill application of Theorem 2.10 below gives an alternate, less constructive proof
that M is a complex manifold.
Remark 2.6 Let P n be the smooth, real 2n-manifold underlying the complex projective
space Pn , and let J be the usual almost-complex structure. By Exercise 2.4, the complex
manifolds (P n , J) and (P n , J) are biholomorphic. Thus when n is odd both orientations
on P n are compatible with a holomorphic atlas. When n is even this need not be the case;
there is no almost-complex structure on P 2 which induces the anti-standard orientation.
Unfortunately, it is nearly universal in the literature to write P2 for the oriented manifold
here denoted P2 . Forming the oriented connected sum of a four-manifold N with M = P2
14

is the topological analogue of blowing up a point of N. This analytic operation is defined

precisely in Exercise 3.5 when N = C2 ; other exercises in that section justify the topological
claim just made.
2
In general, nothing can be said about the oriented manifold M . There may or may
not exist a compatible almost-complex structure, to say nothing of a compatible holomorphic atlas. Further, there is no general relationship between complex manifolds M and M ;
they may or may not be biholomorphic, regardless of the respective induced orientations.
Exercise 2.5 investigates the case when M is an elliptic curve.

2.3

Integrability Conditions

Every complex manifold comes equipped with an induced almost-complex structure. Conversely, it is of interest to characterize almost-complex structures which arise from a holomorphic atlas. It is to be expected that some differential condition on J is necessary, since
J is an algebraic object while a holomorphic atlas contains analytic information.
On an arbitrary almost-complex manifold, the exterior derivative has four type components, namely d : Ap,q Ap1,q+2 Ap,q+1 Ap+1,q Ap+2,q1 Ap+q+1 . This is easily seen
from dA1,0 A2,0 A1,1 A0,2 and induction on the total degree. Under a suitable firstorder differential condition, the unexpected components are equal to zero. To introduce
this condition, first define the Nijenhuis1 (or torsion) tensor NJ of J by


(2.5)
NJ (X, Y ) = 2 [JX, JY ] [X, Y ] J[JX, Y ] J[X, JY ]
for local vector fields X and Y . The torsion tensor measures involutivity of the i-eigenspace
bundle T 1,0 M in the following sense.

Lemma 2.7 Let X and Y be local vector fields, and set Z = [X iJX, Y iJY ]. Then
2(Z + iJZ) = NJ (X, Y ) iJNJ (X, Y ).
In words, the torsion tensor is gotten by sending a pair of real vector fields to the corresponding (1, 0) fields and taking minus the real part of the (0, 1)-component of their Lie
bracket.
Theorem 2.8 The following are equivalent:
(a) If Z and W are (1, 0) vector fields, then so is [Z, W ], i.e. T 1,0 M is involutive.
(b) T 0,1 M is involutive.
(c) dA1,0 A2,0 A1,1 and dA0,1 A1,1 A0,2 .
1

Pronounced N yen haus.

15

(d) dAp,q Ap+1,q Ap,q+1.

(e) If X and Y are local vector fields, then NJ (X, Y ) = 0.
proof (a) is equivalent to (b) by taking complex conjugates. To show (a) and (b) are
equivalent to (c), recall that for an arbitrary one-form ,






(2.6)
2d(Z, W ) = Z (W ) W (Z) [Z, W ] .

If (a) and (b) hold, and if A1,0 , then the right side of (2.6) vanishes for all Z, W of type
(0, 1), proving d has no component of type (0, 2), i.e. (c) holds.
Conversely, suppose (c) holds and let A0,1 . For all Z, W of type (1, 0), (2.6) implies
([Z, W ]) = 0. Thus [Z, W ] is of type (1, 0) and (a) holds.
(c) implies (d) by induction, while (d) implies (c) trivially. Finally, (a) and (e) are
equivalent by Lemma 2.7.
Consequently, if NJ vanishes identically then there is a decomposition d = + as in
Example 2.3 above. Considering types and using d2 = 0, it follows that
(2.7)

2 = 0,

= ,

2 = 0.

Theorem 2.9 Let f : (M, J) (M 0 , J 0 ) be a mapping of almost-complex manifolds. The

following are equivalent.
(a) For every local vector field Z, if Z is of type (1, 0), then f Z is of type (1, 0).
(b) For every local vector field Z, if Z is of type (0, 1), then f Z is of type (0, 1).
(c) Pullback by f preserves types, i.e. f : Ap,q (M 0 ) Ap,q (M).
(d) f is pseudoholomorphic.
proof Since f commutes with conjugation, (a) and (b) are equivalent. Elements of Ap,q
are locally generated by elements of A1,0 and A0,1 , so (a)/(b) are equivalent to (c). Finally,
f (X iJX) = f X if JX is of type (1, 0) if and only if f JX = J 0 f X, i.e. if and only
if f is almost-complex.
Vanishing of the torsion tensor is a necessary condition for an almost-complex structure
J to be induced by a holomorphic atlas; since the components of J are constant in a holomorphic coordinate system by equation (2.4), the torsion of the induced almost-complex
structure vanishes identically. More interestingly, vanishing of the torsion (together with a
mild regularity condition) is sufficient for an almost-complex structure to be induced by a
holomorphic atlas. When (M, J) is real-analytic, this amounts to the Frobenius theorem.
When (M, J) satisfies less stringent regularity conditions, the theorem is a difficult result
in partial differential equations, and is loosely known as the Newlander-Nirenberg Theorem.
The weakest hypothesis is that (M, J) be of Holder class C 1, for some > 0.
16

Theorem 2.10 Let (M, J) be a real-analytic almost-complex manifold, and assume NJ vanishes identically. Then there exists a holomorphic atlas on M whose induced almost-complex
structure coincides with J.
proof (Sketch) The new idea required is to complexify M. Then the Frobenius theorem
may be applied.
Suppose : R2n R is real-analytic. Thenlocallythere is a holomorphic extension
C : C2n C obtained by expressing as a convergent power series and regarding
the variables as complex numbers.
Complexifying the charts of M gives the following: For each x M, there exists a
coordinate neighborhood U of x and a complex manifold UC isomorphic to U B, B a ball
in R2n , whose charts extend the charts of M. The tangent bundle T UC , when restricted to
U, is the complexification of T U, and in particular contains the involutive (by hypothesis)
subbundle T 1,0 U. By the Frobenius theorem, there is an integral manifold through x, and
local holomorphic coordinates on this leaf induce holomorphic coordinates on U.
Example 2.11 The almost-complex structure on S 6 induced by Cayley multiplication has
non-zero torsion tensor. It is not known at present whether or not S 6 admits a holomorphic atlas, though there is circumstantial and heuristic evidence against, and the answer is
generally believed to be negative.
2
The Newlander-Nirenberg theorem characterizes holomorphic atlases in terms of real
data, namely an almost-complex structure with vanishing torsion. It is desirable to have a
similar description of holomorphic vector fields on a complex manifold. A holomorphic
vector field on a complex manifold is a vector field Z of type (1, 0) such that Zf is holomorphic
for every local holomorphic function f . In local coordinates,
Z=

n
X

j=1

with local holomorphic functions. There are global characterizations of holomorphy,

discussed later.
A holomorphic vector field is not, in the sense of real manifolds, a vector field on M. In
spite of this it is often convenient to identify holomorphic vector fields and certain real vector
fields; the next two propositions make this identification precise. The real counterpart of a
holomorphic vector field is an infinitesimal automorphism of an almost-complex structure J,
namely, a vector field X for which the Lie derivative LX J vanishes.
Proposition 2.12 A vector field X is an infinitesimal automorphism of J if and only if
[X, JY ] = J[X, Y ] for all Y . If X is an infinitesimal automorphism of J, then JX is an
infinitesimal automorphism if and only if NJ (X, Y ) = 0 for all Y .
17

proof By the Leibnitz rule, [X, JY ] = LX (JY ) = (LX J)(Y ) + J[X, Y ] for all Y , proving the first assertion. To prove the second assertion, note
that if X is an infinitesimal
automorphism of J, then N(X, Y ) = 2 [JX, JY ] J[JX, Y ] ,

Proposition 2.13 Let M be a complex manifold with induced almost-complex structure J.

Then X is an infinitesimal automorphism of J if and only if Z = X 1,0 := (1/2)(X iJX) is
a holomorphic vector field. Furthermore, the map X 7 X 1,0 is an isomorphism of complex
Lie algebras.

Exercises
Exercise 2.1 Consider the following descriptions of the complex projective line P1 :
The unit sphere {(u, v, w) R3 | u2 + v 2 + w 2 = 1}, which is identified with the
Riemann sphere C by stereographic projection.
Two copies of the complex line C suitably glued together. More precisely, let z 0 and
z 1 be complex coordinates in the two copies of C, and identify z 0 with 1/z 1 . In this
picture, the origin in each copy of C is the point at infinity in the other copy.
The set of non-zero pairs of complex numbers (Z 0 , Z 1 ) C2 \(0, 0), with the equivalence
relation (Z 0 , Z 1 ) (W 0 , W 1 ) if and only if Z 0 W 1 = Z 1 W 0 , i.e. the points (Z 0 , Z 1 )
and (W 0 , W 1) lie on the same complex line through (0, 0). In other words, a point of
P1 is a line through the origin in C2 .
Show that these three descriptions are equivalent by using the identification z 0 = Z 0 /Z 1 . (A
sketch of the real points may be helpful.) Describe the space of holomorphic vector fields on
P1 in terms of each presentation; determine, in particular, which vector fields correspond to
rotations of the sphere. Describe the space of holomorphic 1-forms on P1 and the space of
meromorphic 1-forms dual to holomorphic vector fields.

Exercise 2.2 For a complex manifold N, denote by H 0 (N, T N) the space of holomorphic
vector fields on N. Let Mi , i = 1, 2 be compact complex manifolds, M = M1 M2 their
product (as a complex manifold). Prove that
H 0 (M, T M) ' H 0 (M1 , T M1 ) H 0 (M2 , T M2 ).
Roughly, every vector field on M is uniquely the sum of a vector field on M1 and a vector
field on M2 . Give examples to show that if compactness is dropped then the result may or
may not hold.


18

Exercise 2.3 Fix a complex number and a complex number with || > 1. The linear
transformation
 
 

 
z1
z1 + z2
z1
=
7
0 z2
z2
z2
generates a properly discontinuous action of Z on C2 \ (0, 0) whose quotient X ( ) is a
complex surface, called a Hopf surface. Show that the Hopf surface X ( ) is diffeomorphic
to S 1 S 3 for every (, ). Find necessary and sufficient conditions under which X ( )
and X0 ( 0 ) are biholomorphic. Calculate the dimension d(, ) of the space of holomorphic
vector fields on X ( ). Show that d is upper semicontinuous, i.e. {(, ) | d(, ) c} is
closed for every c R. Describe the space of holomorphic 1-forms on X ( ). Essentially by
construction, two Hopf surfaces are deformation equivalent, but the dimension of the space
of holomorphic vector fields jumps as the parameters and vary.


Exercise 2.4 Let Pn denote the space of complex lines through the origin in Cn+1 , with
the holomorphic atlas J described in Example 1.7, and let M be the underlying real 2ndimensional smooth manifold. Prove that complex conjugation on Cn+1 induces a diffeomorphism f : M M such that J f = J; describe the fixed points of f . In other words,
n
if P denotes the complex manifold whose charts are complex conjugates of the standard
n
charts, then Pn and P are biholomorphic.
The diffeomorphism f preserves orientation if n = 2; in fact, it is known that there is
no holomorphic atlas on the smooth manifold underlying P2 which induces the orientation
opposite to the standard orientation. Briefly, P2 is not a complex manifold.


Exercise 2.5 Let E be the elliptic curve associated to h, as in Exercise 1.5. Prove
that E is biholomorphic to E if and only if 2 Re is an integer. This involves several
easy arguments very much in the spirit of Example 1.8. Products of suitable elliptic curves
give examples of compact tori for which M and M have the same orientation but are not
biholomorphic.


Exercise 2.6 Let C1 and C2 be compact, connected Riemann surfaces, and let f : C1 C2
be a non-constant holomorphic map. Prove that f is onto, and that if f is one-to-one,
then f is a biholomorphism. Prove that there is a finite set R C2 such that f restricted
to f 1 (C2 \ R) is a covering map. Conclude that there is a positive integer dthe degree
of f such that f is d-to-one except for a finite (possibly empty) subset of C1 .
Suggestion: Let R be the image of the set of critical points of f .

A branch point of a (non-constant holomorphic) map between Riemann surfaces is a point
p C1 at which f is not locally one-to-one (i.e. there does not exist a neighborhood V of p
with f |V one-to-one). The prototypical branch point is the origin for the mapping z 7 z n ;
if p is a branch point of f and if f (p) = q, then there exist charts near p and near q such
that f 1 (z) = z n for some integer n > 1 independent of and . The number n 1
is called the ramification number or branching order of f at p, and is zero except possibly
19

at finitely many points. Consequently, the total branching order b (the sum of all branching
orders) is well-defined when C1 is compact.
Exercise 2.7 Let f : C1 C2 be a holomorphic map of Riemann surfaces. Prove that
branch points of f are exactly critical points, i.e. points where df = 0. Assume further that
C1 and C2 are compact Riemann surfaces of respective topological genera g1 and g2 , and
that f has degree d. Prove the Riemann-Hurwitz formula:
(2.8)

b
g1 = d(g2 1) + 1 + .
2

Conclude that g2 g1 . Calculate the total branching order of a rational function of degree d,
regarded as a map P1 P1 , and verify the Riemann-Hurwitz formula in this case.
Suggestion: Triangulate C2 so that every image of a critical point is a vertex, then lift to C1
and compute the Euler characteristic.


Exercise 2.8 Let d be a positive integer, and let f : C3 C be a the Fermat polynomial
f (Z 0 , Z 1 , Z 2 ) = (Z 0 )d + (Z 1 )d + (Z 2 )d of degree d. The zero locus C P2 is a smooth curve.
Find the Euler characteristic and genus of C in terms of d.
Suggestion: Project away from [0 : 0 : 1] 6 C to get a branched cover of P1 , then use the
Riemann-Hurwitz Formula.

Exercise 2.9 (A first attempt at meromorphic functions) Let M be a complex manifold x
a point of M. A meromorphic function is given in a neighborhood U of x as the quotient
of a pair of holomorphic functions 0 , : U C with not identically zero. Show that
a meromorphic function on a Riemann surface may be regarded as a holomorphic map to
P1 ; Exercise 1.4 may be helpful. There is no general analogue of this assertion if M has
dimension at least two; on C2 , it is impossible to extend z 1 /z 2 to the origin even if is
allowed as a value. Meromorphic functions are defined precisely in Section 6.

Exercise 2.10 The great dodecahedron is the regular (Keplerian) polyhedronnot embedded in R3 whose 1-skeleton is that of a regular icosahedron and whose twelve regular
pentagonal faces intersect five at a vertex. Thus the edges of a single face are the link of
a vertex of the 1-skeleton. Realize the great dodecahedron as a three-sheeted cover of the
sphere (P1 ) branched at twelve points, and use the Riemann-Hurwitz formula to compute
the genus.


Sheaves and Vector Bundles

Sheaves are a powerful bookkeeping device for patching together global data from local data.
This is due principally to existence of cohomology theories with coefficients in a sheaf and the
attendant homological machinery. Many geometric objects (line bundles and infinitesimal
deformations of pseudogroup structures, for example) can be expressed in terms of higher
20

sheaf cohomology. Finally, there are theorems which relate sheaf cohomology to de Rham
or singular cohomology, and theorems which guarantee vanishing of sheaf cohomology under
various geometric hypotheses.

3.1

Presheaves and Morphisms

Let X be a paracompact Hausdorff space. A presheaf F of Abelian groups over X is an

association of an Abelian group F (U) to each open set U X and of a restriction
homomorphism U V : F (U) F (V ) for each pair V U of nested open sets, subject to
the compatibility conditions
U U = Identity,

V W U V = U W

if W V U.

The notation F follows the French faisceau. An element s F (U) is a section of F over U;
elements of F (X) are global sections of F . It is useful to allow the group F (U) to be empty
(as opposed to trivial), as in Example 3.4 below.
Remark 3.1 In abstract terms, associated to a topological space X is a category whose
objects are open sets and whose morphisms are inclusions of open sets. A presheaf of Abelian
groups on X is then a functor from this category to the category of Abelian groups and group
homomorphisms.
2
It is often useful to consider presheaves of commutative rings or algebras; these are defined
in the obvious way. A presheaf M of modules over a presheaf R of rings is an association, to
each open set U, of an R(U)-module M(U) in a manner compatible with restriction maps.
There is nothing in the definition to force the restriction maps to be actual restriction
maps. However, in all subsequent examples, restriction may be taken literally, and restriction
maps will be denoted with more standard notation.
Example 3.2 The simplest presheaves over X are the constant presheaves G, whose sections
are locally constant G-valued functions.
0
0
The presheaf of continuous functions CX
is defined by letting CX
(U) be the space of
continuous, complex-valued functions on U. If M is a smooth manifold, then AMr denotes
the presheaf of smooth r-forms; it is customary to write AM instead of AM0 . These are all
presheaves of rings. A surprisingly interesting example is the presheaf AM of non-vanishing
smooth functions. It is a presheaf of Abelian groups under pointwise multiplication of
functions.
2
Example 3.3 Let (M, J) be an almost-complex manifold. The presheaf of smooth (p, q)forms on M is denoted AMp,q . If M is complex, then the operator is a morphism of

presheaves, : AMp,q AMp,q+1, and the kernelconsisting of -closed

(p, q)-formsis dep,q
noted Z . The presheaf of local holomorphic functions is denoted OM , and the presheaf of
local holomorphic p-forms is denoted pM .
2
21

Example 3.4 Let : Y X be a surjective mapping of topological spaces. The presheaf

of continuous sections of associates to each open set U X the (possibly empty) set (U)
of continuous maps s : U Y with s = id. The presheaf restrictions are the ordinary
restriction maps. Generally there is no algebraic structure on (U).
2
A morphism of presheaves is a collection of group homomorphisms (U) : F (U) E(U)
which are compatible with the restriction maps. A morphism is injective if each map (U)
is injective. The kernel of a presheaf morphism is defined in the obvious way. Surjectivity
and cokernels are more conveniently phrased with additional terminology, and are discussed
later, see also Exercise 3.1.
Example 3.5 Suppose f : X Y is a continuous map and E is a presheaf on X. The
pushforward of E by f is the presheaf on Y defined by f E(V ) = E(f 1 (V )). If F is a presheaf
on Y , the pullback is the presheaf on X defined by f F (U) = F (f (U)).
2
Let U = {U }I be a cover of X by open sets. It is convenient to denote intersections of
sets by multiple subscripts; thus U = U U , for example. A presheaf over X is complete
if two additional properties are satisfied:
i. If s, t F (U) and s|V = t|V F (V ) for all proper open sets V U, then s = t F (U).
In words, sections are determined by their values locally.
ii. For every open set U X, if U = {U } is a cover of U by open sets, and if there exist
compatible sections s F (U ), i.e. such that
s | U

= s |U

whenever U U is non-empty,

then there is a section s F (U) with s|U = s for all . In words, compatible local

data may be patched together.

The presheaves described in Examples 3.23.4 are complete.
There is a natural way of completing an arbitrary presheaf. Associated to a presheaf F
over X is a sheaf of germs of sections, which is a topological space equipped with a surjective
local homeomorphism to X. The inverse image of a point x X is called the stalk at x, and
is defined to be the direct limit
(3.1)

Fx = lim
F (U).
xU

To elaborate on this definition, note that the set of neighborhoods of x forms a directed
system under inclusion of sets, and because restriction maps satisfy a compatibility condition,
it makes sense to declare s F (U) to be equivalent to t F (V ) if there is a neighborhood
W U V of x with s|W = t|W . The stalk Fx is the set of equivalence classesor germs.
Any algebraic structure possessed by the spaces F (U) will be inherited by the stalks.
22

S
as
Let F = xX Fx be the union of the stalks. A basis for a topology on F is defined
S
follows: For an open set U X and a section s F (U), there is a natural map to xU Fx
which sends s to its germ sx Fx . The image of such a map is a basic open set. The
projection map : F X sending each stalk Fx to x X is a local homeomorphism.
A sheaf is a complete presheaf. The completion process just described associates a sheaf
to an arbitrary presheaf. The presheaf of continuous sections of the completion is isomorphic
to the original presheaf, so it usually unnecessary to be scrupulous in distinguishing a presheaf
and its completion.
Example 3.6 There are presheaves which fail to satisfy each of the completeness axioms.
Let X be an infinite set with the discrete topology, and let C(U) denote the set of complexvalued functions on U. Define U V to be the zero map if V is a proper subset of U. This
defines a presheaf C which fails to satisfy axiom i. Intuitively, C has bad restriction maps.
If B is the presheaf of bounded holomorphic functions over C, then B fails to satisfy
axiom ii. If Un denotes the disk of radius n centered at 0 C, then the function z lies in
B(Un ) for every n, but these local functions do not give a globally defined bounded function.
Intuitively, this presheaf is defined by non-local information. The completion of B is OC ,
whose stalk at z0 is the ring OC,z0 of locally convergent power series centered at z0 .
2

Remark 3.7 Some authors define a sheaf to be a topological space F together with a
surjective local homeomorphism to X satisfying additional conditions. This is an especially
useful point of view in algebraic geometry, but for the relatively naive purposes below, the
concrete definition is simpler.
2
A morphism of sheaves of Abelian groups over X is a continuous map : E F of
topological spaces which maps stalks homomorphically to stalks. A morphism of sheaves of
Abelian groups is injective/surjective (by definition) if and only if is injective/surjective
on stalks. The quotient of a sheaf by a subsheaf is similarly defined on stalks. Sheaves of
appropriate algebraic structures may be direct summed, tensored, dualized, and so on. A

sequence S F Q is exact at F if the image of is equal to the kernel of .

Example 3.8 For a complex-valued function f , set Exp(f ) = e2 1f . If M is a smooth

manifold, then the sequence
(3.2)

Exp

0 Z A A 0

is exact. If M is a complex manifold, there is a short exact sequence

(3.3)

Exp

0 Z O O 0

called the exponential sheaf sequence. These sequences are geometrically important in view
of the cohomological interpretation of line bundles discussed later.
2
Let M be a complex manifold. An analytic sheaf on M is a sheaf
L of OM -modules. An
analytic sheaf F is finitely generated if there is an exact sequence k OM F 0, and is
23

coherent if, in addition,

theL
kernel of the first map is finitely generated, that is, there is an
L`
exact sequence
OM k OM F 0. It is perhaps worth emphasizing that these
sequences are essentially local, in the sense that exactness holds only at the level of stalks.
Thus a finitely generated analyticL
sheaf has the property that for every x M, there is an
open neighbourhood U such that k OM (U) F (U) 0 is an exact sequence of Abelian
groups.
Example 3.9 Let Y X be a closed subspace. If F is a sheaf of Abelian groups on Y ,
then there is a sheaf on X obtained by extending by zero; to each open set U X, associate
the Abelian group F (U Y ) when this intersection is non-empty, and associate the trivial
Abelian group otherwise.
An important special case is when Y is a closed complex submanifold of a complex
manifold M. The ideal sheaf IY is the subsheaf of OM consisting of germs of holomorphic
functions which vanish on Y . The quotient OM /IY is the extension by zero of OY . Both of
these sheaves are coherent.
2
A coherent analytic sheaf E is locally free if each x M has a neighborhood U such that
E(U) is a free OM (U)-module. In other words, each stalk Ex is isomorphic to a direct sum of
finitely many copies of OM,x . Locally free sheaves are closely related to holomorphic vector
bundles, see Example 3.10.
Coherent analytic sheaves are basic tools in algebraic geometry, several complex variables,
andto an increasing extentdifferential geometry. As a category, coherent sheaves behave
better than locally free sheaves. Kernels, images, quotients, and pushforwards of coherent
sheaves are coherent, while analogous assertions about locally free sheaves are generally false.

3.2

Vector Bundles

Intuitively, a vector bundle over a smooth manifold M may be regarded as a family of vector
spaces (fibres) parametrized by points of the manifold. The family is required to satisfy a
local triviality condition which, among other things, implies that over each component of
M the fibres are all isomorphic. According to the nature of the fibres, one speaks of real or
complex vector bundles of rank k.
Precisely, a complex vector bundle of rank k over a smooth manifold M is a smooth
submersion p : E M of smooth manifolds with the following properties:
i. For each x M, Ex := p1 (x) ' Ck .

ii. For each x M, there exists a neighborhood U of x and a diffeomorphismcalled a

vector bundle chart : p1 (U) U Ck such that p 1 is projection on the first
factor.
iii. If ( , U ) and ( , U ) are vector bundle charts at x, then the overlap map
1
|U

: U Ck U Ck
24



is given by (x, v) 7 x, g (x)v for a smooth map g : U GL(k, C), called the
transition function from U to U .
The simplest vector bundles may be presented as collections of charts and transition functions. Examples are given in the exercises. In many situations of interest, such explicit
information is impossible to obtain.
The manifold M is the base space of the vector bundle, and E is the total space. Though it
is not unusual to speak of the vector bundle E, this abuse of language can be substantially
imprecise. If p : E M is a complex vector bundle over a complex manifold and the
transition functions are holomorphic maps into the complex manifold GL(k, C), then p :
E M is a holomorphic vector bundle. In this case, the total space E is a complex
manifold, and the projection p is a holomorphic submersion.
Example 3.10 Let p : E M be a holomorphic vector bundle. The presheaf of local
holomorphic sections gives rise to a locally free analytic sheaf. Conversely, a locally free
analytic sheaf E determines a holomorphic vector bundle whose presheaf of sections induces
the original sheaf. To accomplish the latter, choose sections ej , j = 1, . . . , k for E(U) which
generate E(U) as an OM (U)-module, and use these to define a vector bundle chart in the
obvious way.
Roughly, coherent sheaves may be regarded as families of vector spaces whose fibre dimension jumps from point to point, as in Example 3.9.
2
Example 3.11 Let M be a complex manifold, and let T M denote the (real) tangent
bundle of the underlying smooth manifold. The bundle T 1,0 M T M C is a holomorphic
vector bundle, called the holomorphic tangent bundle of M, whose underlying (smooth)
complex vector bundle is complex-linearly isomorphic to T M endowed with the natural
almost-complex structure.
N
V
V
The tensor bundles p T 1,0 M and p T 1,0 M are holomorphic, while q T 0,1 M is merely
smooth. Since the composite of anti-holomorphic maps is not anti-holomorphic, there is no
notion of an anti-holomorphic vector bundle.
2
A morphism of vector bundles over M (or a vector bundle map) is a map : E F of
smooth manifolds which is linear on fibres and is compatible with the projection maps. The
kernel, cokernel, and image of a vector bundle map are defined in the obvious way. However,
these families of vector spaces are not generally vector bundles; the fibre dimension may vary
from point to point. Two vector bundles E and E 0 over M are isomorphic if there exists a
vector bundle map from E to E 0 which is an isomorphism on fibres.
It is sometimes useful to consider maps between vector bundles over different base spaces.
In this case, it is necessary to have compatible mappings of the total spaces and the base
spaces, of course. The prototypical example is the pullback of a vector bundle under a
smooth map. Let p : E M be a vector bundle, and let f : M 0 M be a smooth map.

25

There is a pullback vector bundle f E M 0 whose total space is defined to be the set
(3.4)

f E = {(x, v) M 0 E|f (x) = p(v)} M 0 E,

with projection induced by projection on the first factor. The fibre of f E over x M 0 is the
fibre Ef (x) of E over f (x), and the isomorphism class of f E is determined by the homotopy
class of the map f .
Remark 3.12 An important result of algebraic topologyof which only a very special
case is given hereis that for each positive integer k, there exists a classifying space Gk
and a universal bundle Uk Gk with the following property: For every smooth, rank k
vector bundle p : E M over a finite-dimensional smooth manifold, there is a classifying
map : M Gk such that E = Uk , and the classifying map is unique up to smooth
homotopy. Roughly, rank-k vector bundles over M are in natural one-one correspondance
with homotopy classes of smooth maps M Gk . One way of constructing the classifying
space is via infinite-dimensional Grassmannians (see Exercise 3.12 and Section 9).
2
For i = 1, 2, let pi : Ei M be smooth complex vector bundles of rank ki over a smooth
manifold. A differential operator is a linear map D : A0 (E1 ) A0 (E2 ) which, when expressed
with respect to arbitrary charts, is a differential operator in the usual sense between k1 -tuples
of functions and k2 -tuples of functions. The order of the local representative is independent
of the choice of vector bundle charts.
Two elementary examples warrant immediate mention. Let M be a smooth manifold.
The exterior derivative d is a first-order differential
operator, V
which in local coordinates is
V
expressed as d(f dxI ) = df dxI . Here E1 = r T M and E2 = r+1 T M. If M is a complex
manifold, then the operators and are similarly first-order differential operators. As in the
example of d, differential operators are often specified by their action on local sections. It is
necessary in such an event to check that local expressions of sections transform correctly,
i.e. are independent of chart. This is illustrated by the second example, which is elementary,
but of sufficient importance to deserve special mention.
Proposition 3.13 Let p : E M be a holomorphic vector bundle. Then there is a differential operator : Ap,q (E) Ap,q+1(E) acting on (p, q)-forms with values in E.
proof Let U M be a trivializing neighborhood for E, and let {ej }kj=1 be a local holomorphic frame over U. Every holomorphic frame {ej 0 }kj=1 over U is of the form e
ei = gij ej for
a non-singular matrix (gij ) of holomorphic functions on U. Let
s=

k
X
j=1

s ej =

k
X

i,j=1

26

sei gij ej

k
X
i=1

seie
ei

V
= Pk s
j ej , which is a
be a local smooth section of p,q E. Then the expression s
j=1
Vp,q+1
local section of
E, is independent of the choice of frame. Indeed, gij is holomorphic
j = 0, so
in U, i.e. g
i

k
X
i=1

si e
e
ei =

k
X

i,j=1

si g j ej =
e
i

k
X

si )g j ej =
(e
i

i,j=1

k
X

i,j=1

si g j ) ej =
(e
i

k
X
j=1

j ej .
s

is independent of vector bundle charts.

It follows that s
The philosophical content of Proposition 3.13 is that holomorphic functions are constant
with respect to the ordinary operator, so if E M is a holomorphic vector bundle, then
there is an induced map on forms with values in E. There is no natural analogue of the
exterior derivative d for forms with values in E because the only d-constant functions
are genuine constants. It is clear that 2 = 0 as a differential operator acting on E-valued
forms. This gives rise to a cohomology theory, studied in Sections 4 and 8. There is no
analogue for the operator because -constant functions are anti-holomorphic, and the
set of anti-holomorphic functions does not form a pseudogroup.
The theory of differential operators on manifolds is vast and deep. A cornerstone is the
Atiyah-Singer Index Theorem for elliptic operators, which expresses the index (dimension
of kernel minus dimension of cokernel) of certain differential operators in terms of integrals
over M of cohomology classes depending on the operator. In many cases of interest, these
classes may be effectively computed. Hodge Theory is the prototypical application of the
theory of differential operators, see Section 8.

Exercises
Exercise 3.1 Consider the condition: For every open set U X, (U) : S(U) F (U) is
a surjective homomorphism of Abelian groups. Explain carefully how this condition differs
from the definition of surjectivity for a presheaf morphism. It may be helpful to consider the
map Exp in the exponential sheaf sequence, which is surjective as a presheaf morphism but
does not generally satisfy the condition above.

1
Holomorphic Line Bundles Over P
Exercises 3.23.7 introduce several holomorphic line bundles over P1 . It is shown later
that holomorphic line bundles over P1 are completely determined by an integer called their
degree.
The holomorphic line bundle of degree k Z over P1 is constructed explicitly as follows.
Using the standard atlas, take two trivial families over U0 and U1 with coordinates (z 0 , 0)
and (z 1 , 1)z i is the base coordinate and i is the fibre coordinate over Ui and glue them
together over the set C = P1 \ {0, } by the identification
(3.5)

z0 =

1
,
z1

0 =
27

1
.
(z 1 )k

This line bundle is usually denoted OP1 (k), or simply O(k) for brevity. The total space H k
of O(k) is the identification space; thus H k consists of the disjoint union U0 t U1 of two
copies of C2 = C C, with respective coordinates (z 0 , 0) and (z 1 , 1), which have been
glued together along C C according to (3.5). The transition function from z 0 to z 1 is
g01 (z 0 ) =

1
;
(z 0 )k

the other is g10 (z 1 ) = 1/(z 1 )k .

A holomorphic section of O(k) is represented by holomorphic functions 0 = f0 (z 0 ) and
1 = f1 (z 1 ) with
(3.6)

g01 (z 0 )f0 (z 0 ) = f1 (z 1 ).

Every bundle has a trivial section, given by i = 0; the graph of this section is often called
the zero section. If there are no other sections, the bundle is said to have no sections.
Exercise 3.2 Find all holomorphic sections of O(k). The projection p : H k P1 is defined
by sending (z i , i) to z i . Check that this map is well-defined, that H k is a complex manifold,
and that a holomorphic section of O(k) is a holomorphic map : P1 H k which satisfies
p = identity on P1 . In other words, a section is a (holomorphic) choice of point in each
fibre.


Exercise 3.3 Find the value of k for which O(k) is the bundle of holomorphic 1-forms on
P1 . (Suggestion: Use the identifications 0 dz 0 (z 0 , 0) and 1 dz 1 (z 1 , 1).) Answer
the same question for the tangent bundle of P1 .


Exercise 3.4 For each point [Z] in P1 , let `[Z] be the line in C2 represented by this point.
Show that the family {`[Z] | [Z] P1 } may be regarded as a holomorphic line bundle in the
following manner: Set L = {([Z], ) P1 C2 | `[Z] }, with projection to the first factor.
This holomorphic line bundle is called the tautological bundle. Find k such that L = O(k).
Use the fact that a holomorphic map s : P1 C2 must be constant to show that L has no
sections (cf. Exercises 3.2 and 1.2).

Exercise 3.5 (Blowing up) In the notation of Exercise 3.4, let : L C2 be the projection
on the second factor. Show that maps the zero section of L to the origin (0, 0), and that
maps the complement of the zero section biholomorphiocally to C2 \ (0, 0). Thus, the total
space of L is obtained from C2 by removing the origin and gluing in a P1 ; the geometric
effect is to give each line through (0, 0) a distinct origin. The total space of L is called the
blow-up of C2 at the origin.

Exercise 3.6 Let [Z 0 : Z 1 : Z 2 ] be homogeneous coordinates on P2 . The map
p : [Z 0 : Z 1 : Z 2 ] 7 [Z 0 : Z 1 : 0]
28

is defined everywhere except at the point [0 : 0 : 1], and the image is the P1 in P2 with
equation Z 2 = 0. Let H = P2 \ [0 : 0 : 1]. Show that p : H P1 is a holomorphic line
bundle, and find the corresponding value of k. Show that the space of holomorphic sections
is the set of linear functions {a0 Z 0 +a1 Z 1 | ai C}.


Exercise 3.7 Let L be the tautological bundle (Exercise 3.4), and H the hyperplane bundle
(Exercise 3.6). Show that H the complement of the zero section of His biholomorphic
to C2 \ (0, 0), which is biholomorphic to L by Exercise 3.5. Find an analogous statement
for the bundles O(k) and O(k). Prove that O (k) is the quotient of H by the action of
a cyclic group of order k.

n
n
Let U = {U }=0 be the standard atlas on P . The line bundle OPn (k) is defined by the
transition functions g ([z]) = (z /z )k .
Exercise 3.8 Show that the line bundle L = OPn (1) is the tautological bundle, that is,
the subbundle of L Pn Cn+1 whose fibre at [Z] Pn is the line `Z Cn+1 through 0
and Z. Let Symdn denote the space of homogeneous polynomials of degree d on Cn . Verify
that the space of holomorphic sections of OPn (k) is equal to Symkn+1 if k 0 and is trivial
otherwise. In particular, the homogeneneous coordinate functions are sections of the line
bundle H = L .


Projectivization, Tautological Bundles, and Blowing Up

Let p : E M be a holomorphic vector bundle of rank r > 1 over a compact complex
manifold. The group C acts by scalar multiplication in the fibres on the total space of E,
and the action is free on the complement E of the zero section.
Exercise 3.9 Prove that the quotient space P(E) = E /C called the projectivization of
Eis a compact complex manifold, and that the projection p : E M factors through the
quotient map. The induced map : P(E) M is a fibre bundle projection whose fibres are
projective spaces. Show that if L is a line bundle over M, then P(E L) is biholomorphic
to P(E).

With notation as in Exercise 3.9, consider the rank r vector bundle E P(E). A
point of P(E) is represented by a point x M and a line `x in the fibre Ex . Let E E
be the line subbundle whose fibre is the line `x . The bundle E is called the tautological
bundle of E (cf. Exercise 3.8, where M is a point).
Exercise 3.10 Prove that the restriction of E to a fibre of P(E) is a tautological bundle
over a projective space. Let $ : L M be a line bundle. Prove that the tautological bundle
of E L is E $ L. (This equation involves a minor abuse of notation.) Prove that E
and E are biholomorphic; in other words, the complement of the zero section in E is the
same as the complement of the zero section of E . The total space of E is obtained from the
total space of E by blowing up the zero section.

Exercise 3.11 Let O denote the trivial line bundle over M. Prove that the total space of
E embeds in P(E O)the completion of Eas an open, dense submanifold. Describe the
29

complement of E in P(E O).

Grassmannian Manifolds
Let 0 < k < n be integers. The set of k-dimensional subspaces in Cn can be made into a
compact complex manifold, called the (complex) Grassmannian Gk (Cn ) = Gk,n . Consider
first the Stiefel manifold Sk,n of k-frames in Cn , which may be realized as the open set in
Cnk consisting of matrices of rank k. The group GL(k, C) acts by right multipication,
and the orbits of this action are sets of frames which span the same linear subspace in Cn .
The Grassmannian is defined to be the quotient Sk,n /GL(k, C) with the quotient topology.
Observe that G1,n+1 = Pn .
Exercise 3.12 Prove that the Grassmannian Gk,n has the structure of a compact complex
manifold, and that the tangent space at W Cn is Hom(W, Cn /W ). Let Uk,n Gk,n be
the universal (k-plane) bundle, that is, the subbundle of Gk,n Cn whose fibre at W Gk,n
is W . Prove that the quotient bundle Q in the exact sequence
0 Uk,n Gk,n Cn Q 0

satisfies T Gk,n = Hom(Uk,n , Q). Prove there is a surjection OPn (1) Cn+1 T Pn of
holomorphic vector bundles. Describe the kernel.


Exercise 3.13 Show that the Euclidean inner product on Cn induces a diffeomorphism
Gk,n ' Gnk,n which associates a k-plane to its orthogonal complement, but that this map
is not a biholomorphism.
Let V be a complex vector space of dimension n, and let f : V V0 be a vector space
isomorphism. Prove that Gk (V ) and Gnk (V ) are biholomorphic, but that the maps on
Grassmannians induced by f go in opposite directions.

The construction of the Grassmannian and universal bundle is functorial (see Exercise 3.12); it therefore makes sense to form bundles of Grassmannians from vector bundles,
and to take universal bundles over bundles of Grassmannians. There are useful notions of
blowing up which are defined in terms of bundles of Grassmannians.

Cohomology

Sheaves are a natural tool for piecing together global data (e.g. holomorphic functions) from
local data (e.g. germs). Potential obstructions to the possibility of patching are measured by

Cech
cohomology, which may also be regarded as the simplicial cohomology of the nerve
of a locally finite open cover. Dolbeault cohomology, by contrast, is the complex analogue of
de Rham cohomology on a complex manifold. The Dolbeault Theorem (Theorem 4.11 below)

asserts the isomorphism of certain Cech

and Dolbeault spaces. Coupled with the Hodge Decomposition, the Dolbeault isomorphism allows calculation of sheaf cohomology, and forges a
powerful link between complex analysis (sheaves), differential geometry (Dolbeault theory),
and topology (singular theory).
30

4.1

Cech
Cohomology

Let X be a paracompact Hausdorff space. To a locally finite open cover U = {U }I is

associated a simplicial complex, the nerve of U, whose r-simplices are non-empty (r + 1)-fold
intersections of sets in U. As usual, multiple subscripts denote intersections of sets. The
typical r-simplex is denoted U0 r .

Let F be a sheaf of Abelian groups (written additively) over X. A Cech

r-cochain with
values in F is a collection of sections
f0 r F (U0 r ),
one for each ordered (r + 1)-tuple of indices. The (r + 1)! sections associated to a fixed
r-simplex are assumed to be completely skew-symmetric in their indices. The group of
r-cochains is denoted C r (U, F ).
The coboundary operator r : C r (U, F ) C r+1 (U, F ) is defined by
(r f )0 r+1 =

r+1
X
(1)j j f0 c
j r+1 ;
j=0

here j denotes the presheaf restriction from the r-simplex U0 c

j r+1 to the (r +1)-simplex
r

U0 r+1 . The group of Cech cocycles Z (U, F ) is defined to be the kernel of r , and the

group of Cech
coboundaries Br (U, F ) is the image of r1 . The usual calculation shows that
r r1 = 0, so every coboundary is a cocycle, and the r-dimensional cohomology of U with
coefficients in F is defined to be H r (U, F ) = Z r (U, F )/Br (U, F ).
The term higher cohomology refers to cohomology in dimension greater than zero. The

zero-dimensional Cech
cohomology of an open cover is independent of the cover, a simple
fact of recurring importance.
Proposition 4.1 Let F X be a sheaf of Abelian groups over a paracompact Hausdorff
space. For every cover U of X by open sets, the zero-dimensional cohomology H 0 (X, F ) is
naturally isomorphic to the group of global sections of F .
proof

The coboundary operator 0 sends a 0-cochain f = {f } to the 1-cochain

0 f = {(0 f ) },

(0 f ) = f |U f |U .

In other words, 0-cocycles are exactly compatible collections of local sections.

A refinement of U is an open cover V = {V }J together with a map : J I of index
sets such that V U() for all J. Every refinement induces a map r on r-cochains
by including open sets into U, evaluating cochains, and restricting. These maps commute
with the respective coboundary maps, so there is an induced map r in cohomology. The

hope is to define the r-dimensional Cech

cohomology of X with coefficients in F to be the
direct limit over successive refinements, H r (X, F ) = limH r (U, F ). This is sensible because
different refinement maps between V and U induce the same map in cohomology:
31

Proposition 4.2 Let , : J I be refinements. Then r = r for all r 0.

proof (Sketch) Let = V0 r1 be an (r 1)-simplex of V. For j = 0, . . . , r 1, let
ej
be the r-simplex

ej = V(0 ) V(j ) V (j ) V (r1 ) ,

and define hr : C r (U, F ) C r1 (V, F ) by refining, evalutaing cochains on

ej , and taking
the alternating sum. An involved but straightforward calculation shows that {hr } is a
homotopy operator, that is,
hr+1 r + r1 hr = r r

as maps on cochains. As usual, this shows the induced maps on cohomology coincide.
In order to have a computable theory, it is necessary to have criteria for an open cover
under which the map to the direct limit is an isomorphism, or to have theorems which relate

Cech
cohomology to other cohomology theories (e.g. singular or de Rham). Such a criterion
can be given immediately, and useful isomorphism theorems will be developed presently.
A locally finite open cover U is acyclic or Leray for the sheaf F if every simplex in the
nerve of U has trivial higher cohomology with coefficients in F . The following general result,
due to Leray, is proven after Theorem 4.11 below in the special case F = p , the sheaf of
germs of holomorphic p-forms.
Theorem 4.3 If U is an acyclic cover of X for the sheaf F , then the map to the direct limit

induces an isomorphism from H r (U, F ) to the Cech

cohomology H r (X, F ).

Let X be a simplicial complex. For every Abelian group G, the Cech

cohomology
r
H (X, G) of X with coefficients in the constant sheaf G is isomorphic to the r-dimensional
simplicial cohomology of X. The idea is to associate a locally finite open cover to the simplicial complex, in an inverse procedure to forming the nerve of a cover. This is accomplished
by taking, for each simplex , the interior of the union of all simplices incident on . The
combinatorics of the simplicial complex are the same as that of the resulting cover.
The topological space underlying an arbitrary smooth manifold admits the structure of a
simplicial complex (every smooth manifold is triangulable). Further, the simplicial cohomology of a simplicial complex is isomorphic to the singular cohomology of the underlying
topological space. The observation of the previous paragraph implies the following.
Proposition 4.4 Let G be an Abelian group, and let G X be the associated constant

sheaf over a compact smooth manifold. Then the Cech

cohomology H r (X, G) is isomorphic
to the r-dimensional singular cohomology of X with coefficients in G.
Let F be a sheaf of Abelian groups on X. A partition of unity of F subordinate to a
locally finite open cover U is a collection of sheaf morphisms { }I such that
32

i. is the zero map on a neighborhood of X \ U (i.e. is supported in U ), and

P
ii. I is the identity map of F .

A sheaf which admits a partition of unity subordinate to every locally finite open cover of
X is fine.
Theorem 4.5 If F is a fine sheaf over X, then H r (X, F ) = 0 for r 1.
proof (Sketch) It suffices to prove that H r (U, F ) = 0 for every locally finite open cover.
Take a subordinate partition of unity { }. If f is an r-cochain, then set
X
(hr f )0 r1 =
(f0 r1 ),
I

where (f0 r1 ) has been extended to U0 r1 by zero off the support of . It is

illustrative to write h1 f explicitly. A straightforward calculation shows that hr+1 r +r1 hr
is the identity map on r-cochains. If f is an r-cocycle, r 1, then g = hr f is an (r 1)cochain with r1 g = f , so f is a coboundary.
If X is a smooth manifold, the sheaves A and A r (smooth complex-valued functions
and r-forms, respectively) are fine. Similarly, on a complex manifold, the sheaves A p,q are
fine. In each case, choose an ordinary smooth partition of unity and define sheaf maps by
pointwise multiplication. By contrast, the sheaves G, O, and p (constants, holomorphic
functions, and holomorphic p-forms) are not fine.
Let M be a smooth manifold. The sheaf A of germs of smooth, non-vanishing, complexvalued functions on M is a sheaf of Abelian groups under pointwise multiplication. By
Proposition 4.6 below, A does not generally have vanishing cohomology in dimension one,
hence cannot be fine. Intuitively, an attempt to define a partition of unity by exponentiating
to the power of an ordinary partition of unity will be met with failure of the complex
logarithm to be single-valued.
Proposition 4.6 Let M be a smooth manifold. There is a natural one-one correspondance
between equivalence classes of smooth complex line bundles over a complex manifold M and
elements of H 1 (M, A ).
proof The idea is extremely simple: Transition functions for a line bundle constitute a
one-cocycle with coefficients in A , and isomorphic bundles differ by a coboundary.
More precisely, let p : L M be a smooth complex line bundle, and let U be a locally
finite trivializing cover. The transition functions for L take values in the Abelian group

GL(1, C) = C , so {g } may be interpreted as a Cech

1-cochain L C 1 (U, A ). The
compatibility condition g g g = 1 in U is exactly the cocycle condition for L .
33

Suppose L0 is a line bundle isomorphic to L, and let U be a locally finite cover of M

which simultaneously trivializes L and L0 . Over each open set U U, an isomorphism
: L L0 is given by multiplication by a smooth, non-vanishing function ; the collection
{ } constitutes a zero-cochain , whose coboundary is immediately verified to satisfy L0 =
L + 0 .
In summary, these observations prove that for every locally finite cover U of M by open
sets, there is a natural one-one correspondance between equivalence classes of complex line

bundles trivialized over U and elements of the Cech

cohomology group H 1 (U, A ). Passing
to direct limits finishes the proof.
The cohomology space H 1(M, A ) comes equipped with a group structure. The induced
group operation on line bundles is the tensor product, since the tensor product of two
numbers (1 1 matrices) is their ordinary product. In particular, every complex line bundle
L has an inverse L1 whose transition functions are reciprocals of the transition functions of
L. Completely analogous remarks are true for holomorphic line bundles, which correspond

to elements of H 1 (M, OM
). Algebraic geometers often speak of invertible sheaves instead of
line bundles.
The standard homological tool for a cohomology theory is the long exact sequence associated to a short exact sequence of cochain complexes. As in singular or simplicial theory,
the connecting homomorphism is defined by a diagram chase, and exactness is verified by
standard arguments.
Proposition 4.7 Let 0 S F Q 0 be a short exact sequence of sheaves over X.
Then there is a long exact sequence
0 H 0 (X, S) H 0 (X, F ) H 0 (X, Q)
H 1 (X, S) H 1 (X, F ) H 1(X, Q)

(4.1)
in sheaf cohomology.

Example 4.8 (Chern classes of line bundles) Let M be a compact, smooth manifold, and
let 0 Z A A 0 be the smooth exponential sheaf sequence. The associated long
exact sequence contains the terms
c

1
H 1 (M, A ) H 1 (M, A )
H 2 (M, Z) H 2 (M, A ).

Because A is a fine sheaf, the first and last terms vanish. Thus the coboundary operator c1
is an isomorphism between the group of smooth complex line bundles on M and H 2(M, Z),
which by Proposition 4.4 is the ordinary integral singular cohomology of M. The image of
a line bundle L in H 2 (M, Z) is called the first Chern class c1 (L).
Similar considerations for holomorphic line bundles imply there is an exact sequence
c

H 1 (M, OM ) H 1 (M, OM
)
H 2 (M, Z) H 2 (M, OM ).

34

However, the groups H i (M, OM ) are not generally trivial, which is a reflection of the fact
that a fixed complex line bundle may admit many inequivalent holomorphic structures. This
sequence is investigated further in Example 4.13 when M is a complex curve.
2

4.2

Dolbeault Cohomology

Let M be a complex manifold. The complex structure splits the exterior derivative d into an
operator of type (1, 0) and an operator of type (0, 1). As noted previously, the latter is
The
of greater importance because holomorphic functions are constant with respect to .
p,q
p,q+1
cohomology theory of the chain complex : A A
is called the Dolbeault cohomology
of M, and is a complex analogue of the de Rham cohomology of a smooth manifold. The

space Hp,q
(M, C) of -closed (p, q)-forms modulo exact forms is a complex vector space.
It is not uncommon to blur the distinction between a vector bundle, its sheaf of germs
of sections, and the space of
global sections, especially when the bundle is not holomorphic.
Vp,q
When precision demands,
denotes the bundle of smooth skew-symmetric (p, q)-tensors,
A p,q is the sheaf of germs of smooth sections, and Ap,q is the space of smooth global sections.
If : E M is a smooth
V complex vector bundle, then the bundle of E-valued (p,
Vq)-forms
is the tensor product p,q E of AM -modules. The sheaf of germs of sections of p,q E is
denoted AMp,q (E) and the space of global sections is denoted Ap,q (M, E) or simply Ap,q (E)
if no confusion is likely. Recall Proposition 3.13, which asserts that if : E M is a

holomorphic vector bundle, then there is a -operator

acting on (p, q)-forms with values

in E. These -operators are coboundary operators, and the resulting cohomology theory is
called E-valued Dolbeault cohomology.
The basic property of Dolbeault cohomologythe Dolbeault-Grothendieck Lemmais
analogous to the Poincare Lemma for de Rham theory; polydisks have no higher cohomology.

Theorem 4.9 Let Cn be a polydisk, and let be a smooth -closed

(p, q) form on ,

q 1. Then there is a smooth (p, q 1)-form on with = .

proof (Outline) It suffices to assume is centered at the origin. The result is first shown
using an integral formula for a slightly smaller polydisk. This is accomplished by an explicit
calculation in one variable, together with a slightly delicate induction on dimension. Next
choose an exhaustion of by smaller polydisks. The requisite detail is to ensure that the
solutions found on the smaller polydisks patch together. This is not automatic since the
= is nowhere close to having unique solutions, but conversely is facilitated by
equation
the extreme latitude in choosing solutions. The sketched details follow.
Smaller polydisk, q = 1: Let = f (z) d
z on = r , and let r 0 < r. Motivated by
Stokes Theorem, set
Z
1
f ()

(z) =
d d.
2i 0r z
35

A short calculationessentially
application of Stokes Theorem to the differential form


2
= f (z)d
d f () log | z| d proves that
z , and this solution is unique up to an added
n1
holomorphic function. If D is a domain in C , is a holomorphic p-form on D, and if

= f (z) d
z , then with chosen as before, ((z)
) = .
p,q
Smaller polydisk, q > 1: Let Aj denote the space of (p, q)-forms on r0 which do not
contain d
z j+1 , . . . , d
z n . Consider the induction hypothesis:
(H)j :

p,q1

= .
If Ap,q
with
j and = 0, then there is a Aj

(H)1 has just been established. Assume (H)j1 , and let Ap,q
j be -closed.
p,q1

Write =
e + d
z j with Aj1
,
e Ap,q
j1 . Since = 0, the coefficients of
and
e are holomorphic in z j+1 , . . . , z n . By the q = 1 step, for each pair of multi-indices I
and J (with |I| = p, |J| = q 1, and each index of J at most j 1), there is a function
0
0
0 d
) = (
+
0 ) = 0. By
IJ
with IJ
/ zj = IJ . Set =
z j Ap,q
j1 , so (e
p,q1
=
the induction hypothesis, there is a Aj1
with
e . Thus = 0 + satisfies
= .

Enlarging the polydisk for q 2: The above argument shows that if is a -closed
(p, q)-form on , then for each r compactly contained in , there exists a (p, q 1)-form
= on r .
on with
Choose an exhaustion sequence rk % r, and let k denote the corresponding polydisk.
k = on k . There is a k with
k = in k+1 ,
Choose k with supp k k+1 and
k k ) = 0 in k . Again by the first part, there is a k of type (p, q 2) with
so (

k , so that
k+1 = on k+1 and k = k+1 on
k = k k on k1 . Set k+1 = k +
k1 . The sequence {k } thereby constructed converges uniformly on compact sets, to by
declaration.
Enlarging the polydisk for q = 1: Proceed as above to choose k and k . Use holomorphicity of k k in k to approximate uniformly by a polynomial k in k1 , and set
k+1 = k + k . The only difference between this and the previous case is that k+1 k does
not vanish on k1 , but is merely uniformly small; this however is sufficient to guarantee
convergence of the sequence {k }.
Theorem 4.9 holds when r = , i.e. when = Cn . The theorem also holds when the
polydisk is replaced by a product of disks and punctured (one-dimensional) disks, as is seen
with easy modification of the argument by using annuli in the q = 1 step, or by using Laurent

series to compute the Cech

cohomology.
Theorem 4.10 Let denote a punctured disk, and let k and ` be non-negative integers.
Then


()k ( )` = 0
Hp,q

for q 1.

36

The analogue of Theorem 4.9 does not hold for punctured polydisks; as a complex vector
2
1
2
space, H0,1
(C \ 0) ' H (C \ 0, O) is infinite-dimensional by Exercise 4.1 below. This

example highlights the analytic (as opposed to topological) nature of Cech

and Dolbeault
2
cohomologies on non-compact manifolds, since C \ 0 has the homotopy type of S 3 , but the
one-dimensional cohomology is infinite-dimensional.
A basic consequence of the Dolbeault-Grothendieck Lemma is the isomorphism between

Dolbeault and Cech

cohomology, usually called the Dolbeault Theorem.
Theorem 4.11 Let M be a complex manifold. For every pair of non-negative integers p
q
p

and q, the Dolbeault cohomology Hp,q

(M) is isomorphic to the Cech cohomology H (M, ).
proof
(4.2)

For an arbitrary complex manifold, the sheaf sequence

p,q
0 Zp,q
Zp,q+1
0
, A

is exact; surjectivity of is the only non-obvious assertion, and is exactly the content of
Theorem 4.9. Taking q = 0, the long exact sequence contains, for r 1,
(4.3)

p,0
p,0
r
r
H r1 (M, AMp,0 ) H r1 (M, Zp,1
) H (M, Z ) H (M, AM ).

When r > 1, the first and last terms vanish, and the middle spaces are isomorphic. Continuing inductively,
(4.4)

p,r1
r1
1
H r (M, Zp,0
(M, Zp,1
).
) ' H
) ' ' H (M, Z

p,r1
0
When r = 1, (4.3) yields the isomorphism H 1 (M, Zp,r1
) ' H 0 (M, Zp,r
),

)/H (M, AM
p,r
which is exactly the Dolbeault cohomology space H (M). Replacing r by q and using the

fact that -closed

(p, 0) forms are exactly holomorphic p-forms completes the proof.
Using the argument in the proof of the Dolbeault Theorem, it is possible to establish a
version of the Leray Theorem (Theorem 4.3 above) for the sheaf of germs of holomorphic

p-forms. The idea is to prove a higher Cech

cohomology is isomorphic to a quotient of

two other zero-dimensional Cech cohomology spaces, which are identied with global sections
independently of the cover.

Proposition 4.12 Let U be a Leray cover for the sheaf p . Then H q (U, p ) ' H q (M, p ).
proof By hypothesis, each r-simplex U0 r in the nerve of U has trivial higher coho p,q (U r ). Thus there is a short exact sequence of chain
(U0 r ) = A
mology, i.e. Zp,q+1

0
complexes analogous to (4.2) but with cochains on the nerve of U. By the same long exact
sequence argument that yields (4.4), and which relies only on fineness of the sheaf AMp,q ,
p,q1
q
p
0
H q (U, p ) ' H 0(U, Zp,q
) ' Hp,q
)/H (U, AM
(M) ' H (M, )

37

as claimed.
Example 4.13 If M is a compact Riemann surface, then the dimension h0 (M, 1 ) of the
space H 0 (M, 1 ) of holomorphic one-forms on M is called the genus of M and is usually
0,1
denoted by g. There is a real-linear isomorphism H1,0
(M) 'R H (M) induced by complex
conjugation on forms. By the Dolbeault Theorem, H 0 (M, 1 ) 'R H 1 (M, O).
On a compact manifold, holomorphic functions are constant, so the sheaf morphism
Exp : O O induces a surjection H 0 (M, O) H 0 (M, O ). If M is a curve, then there
are no non-zero (0, 2)-forms on M, so H 2 (M, O) = 0 by the Dolbeualt Theorem. The
interesting portion of the long exact sequence associated to the exponential sheaf sequence
on a Riemann surface M is therefore
(4.5)

1
0 H 1 (M, Z) H 1 (M, O) H 1 (M, O )
H 2 (M, Z) 0.

As seen in Example 4.8, the group of smooth complex line bundles on M is isomorphic to
Z via the Chern class map c1 . More precisely, if L M is a complex line bundle, then the
value c1 (L) of the Chern class map is a 2-dimensional cohomology class, whose pairing with
the fundamental class of M is the degree of L:
deg L = hc1 (L), [M]i Z.
Consider the Chern class map on the space H 1 (M, O ) of holomorphic line bundles. The
kernel of c1 is a subgroup, consisting of degree zero line bundles. These may be regarded
as non-trivial holomorphic structures on the topologically trivial complex line bundle; while
the total space of a degree zero line bundle is diffeomorphic to M C, a non-trivial degree
zero bundle has no holomorphic sections other than the zero section. Exactness of the
sequence (4.5) implies that the kernel of c1 is isomorphic to
(4.6)

H 1 (M, O)/H 1(M, Z) =: J0 (M),

called the degree zero Jacobian of M. It is not difficult to see that the Jacobian is a compact complex torus of dimension g. Fix a basis for the one-dimensional singular homology
H1 (M, Z); it is customary to choose curves ai and bj , 1 i, j g, whose oriented intersection
numbers are given by
ai aj = bi bj = 0, ai bj = ij .

Topologically the a-curves are latitudes around the handles of M and the b-curves are
longitudes with suitable orientations. The space H 1 (M, O) is real-linearly isomorphic to
the space of holomorphic one-forms, which has complex dimension g, and the space H 1 (M, Z)
may be regarded as the rank-2g lattice of holomorphic one-forms whose integrals over the
curves ai and bj are integers. Such forms are said to have integral periods.
When M = P1 (g = 0), the Jacobian is trivial; holomorphic line bundles are classified
by their degree. When M is an elliptic curve (g = 1), it turns out that M and J0 (M) are
isomorphic, not only as complex curves but also as Abelian groups, see Example 6.8 below.
38

There is a holomorphic embedding of a compact Riemann surface of genus g 1 into

its Jacobian, given by integrating holomorphic 1-forms over (real) curves. There is also
an algebro-geometric interpretation in terms of divisors on M. This mapping generalizes
the isomorphism between an elliptic curve and its degree zero Jacobian. The definition of
this embedding by period integrals originated in the late 19th Century, while the algebrogeometric interpretation arose only in the mid-20th Century.
2

4.3

Elementary Deformation Theory

In order to study the set of distinct complex structures on a fixed smooth manifold, it is
perhaps most sensible to attempt to define infinitesimal deformations of a holomorphic
atlas. The idea and basic results are due to Kodaira and Kodaira-Spencer.
Kodairas idea was to view a complex manifold as being obtained from small open sets
by gluing via biholomorphisms. Let (M, J) be a fixed complex manifold, and let U be a
locally finite cover of M which is acyclic for the structure and tangent sheaves OM and M .
If U and U are open sets with non-empty intersection, and if , , are the respective
charts to polydisks in Cn , then the transition function is a biholomorphism
(4.7)

g : (U ) (U )

which specifies the manner in which U and U are to be glued. A deformation of the complex
structure should be regarded as a one-parameter family {g (t)} of transition functions.
The differentials g represent an infinitesimal deformation of the complex structure, and

constitute a Cech
one-cocycle with values in the tangent sheaf M . The space of infinitesimal
deformations of the complex structure of M is therefore identified with the cohomology
space H 1 (M, M ). In the most optimistic circumstance, the set of complex structures on M
which are deformation equivalent to (M, J) will be a complex manifoldthe moduli space
of complex structures on M, and the tangent space at a complex structure will be exactly
the cohomology H 1 (M, ).
There are several technical problems which may arise. First the moduli space may not be
smooth. There is a purely cohomological criterion which guarantees smoothness. It may also
occur that some infinitesimal deformation may not arise from an actual one-parameter family
of deformations. In this event, the infinitesimal deformation is said to be non-integrable.

Exercises
Exercise 4.1 Let M = C2 \ 0. By considering the open cover U0 = C C , U1 = C C,
which is acyclic for the sheaf O, show that H 1 (M, O) is infinite-dimensional. Observe that
H q (M, O) is trivial for q > 1.

39


Exercise 4.2 Compute the Cech
cohomology H q (P1 , OP1 ), q 1, using the standard

q
1
atlas (which is a Leray cover for O). Compute H q (P1 , OP
1 ). (Use the fact that H (P , Z)
is isomorphic to the singular cohomology of P1 ; the standard cover is not acyclic for the
constant sheaf Z.) Finally, show that the first Chern class of OP1 (1) is the positive generator
of H 2 (P1 , Z), and conclude that Every holomorphic line bundle over P1 is of the form O(k)
for some integer k.

n
A similar result is true for holomorphic line bundles over P , but the bookkeeping makes
n
direct calculation of H q (Pn , OPn ) = H0,q
(P , C) tedious. However, the following result (see
also Corollary 8.19 below), which gives the Dolbeault cohomology of Pn , is plausible and is
so useful that it is convenient to state it now for future use.
Proposition 4.13 The Dolbeault cohomology of complex projective space is
(
C if 0 p = q n
p,q
n
H (P , C) =
0
otherwise
proof (Idea) The singular cohomology of Pn is one-dimensional in even dimensions up
to 2n and is zero otherwise. But for general reasons, the singular cohomology H r (Pn , C) is
n
isomorphic to the direct sum over p + q = r of the Dolbeault cohomology spaces Hp,q
(P , C),
q,p
and further Hp,q
and H are real-linearly isomorphic. Proposition 4.13 follows at once.
Exercise 4.2 Let U = {U }n=0 be the standard atlas on Pn . Recall that the line bundle
OPn (k) has transition functions from U to U given by g ([Z]) = (z /z )k . Use Proposition 4.13 to prove that every holomorphic line bundle on Pn is isomorphic to OPn (k) for
some integer k.


Analytic and Algebraic Varieties

Many examples of smooth real manifolds arise as level sets of mappings; indeed, this is the
historical source of interest in manifolds. Similarly, examples of complex manifolds arise as
level sets of holomorphic mappings; the Inverse and Implicit Function Theorems have direct
holomorphic analogues.
Theorem 5.1 Let Cn be a polydisk containing the origin, and let f : Cn be a
holomorphic map with Df (0) non-singular. Then there exists a polydisk 0 such that
f |0 is a biholomorphism onto its image.
proof By the ordinary Inverse Function Theorem in R2n , there is a polydisk 0 such
that f |0 has a smooth local inverse g and Df (z) is non-singular for z 0 . Holomorphicity
40

is a simple consequence of the Chain Rule. Indeed, let w = f (z) with an obvious use of
multi-index notation. Then z = g(w) = g(f (z)) on 0 , and
0=

g f
g f
+
.
w z w z

The first term vanishes since f is holomorphic, while the matrix ( f/ z) in the second term
is non-singular on 0 . Thus g/ w = 0, i.e. g is holomorphic.
Theorem 5.2 Let Cn be a polydisk containing the origin, and let f : Ck be a
holomorphic map with rk Df (0) = k. Then, after permuting coordinates if necessary, there
exists a polydisk 0 Cnk and a holomorphic map : 0 Ck such that f (z, (z)) = 0
for all z 0 .
One striking difference between the smooth and holomorphic categories is a strengthening of the Inverse Function Theorem which says that a one-one holomorphic map is a
biholomorphism. In other words, there is no analogue of the smooth map x R 7 x3 ,
which is one-one but not a diffeomorphism.
Theorem 5.3 If f : U V is a one-one holomorphic map between open subsets of Cn ,
then f 1 : V U is holomorphic, i.e. det Df 6= 0.
proof It suffices to assume f (0) = 0 and to prove f 1 is a biholomorphism on a neighborhood of 0. The strategy is to induct on n as follows:
(Hn ) :

If f : Cn Cn is holomorphic, and if Df (0) is singular, then Df (0) = 0.

The assertion (H1 ) is clear. Suppose f : U Cn is a holomorphic map on a neighborhood

of 0 Cn and Df (0) has rank k < n. By permuting coordinates if necessary, f followed
by projection to Ck is a biholomorphism of neighborhoods of the origin in Ck . However,
restricted to the orthogonal complement Cnk , f has singular Jacobian at the origin. By
the inductive hypothesis k = 0, i.e. the Jacobian vanishes everywhere it is singular. In this
case, f maps each component of the zero locus of Df to a point. But the function det Df
cannot have an isolated zero, so if the Jacobian vanishes at one point, then it vanishes at
infinitely many points, so f is not one-one.
A simple consequence of Theorem 5.2 is a criterion for determining smoothness of the
zero locus in Pn of a single homogeneous polynomial f of degree d. It is useful to introduce,
for each coordinate chart U on Pn , the polynomial f defined by
f (Z 0 , . . . , Z n ) = (Z )d f (z0 , . . . , zb , . . . , zn );

f is obtained by setting Z = 1, and defines the same zero set as f in U .

41

Proposition 5.4 If f : Cn+1 C is a homogeneous polynomial, and if f (Z) 6= 0 when

Z 6= 0, then the zero locus of f in Cn+1 \ 0 is a smooth submanifold and projects to a smooth,
compact complex submanifold Vf Pn .
proof The first assertion is immediate from Theorem 5.2. To prove the second, consider
the coordinate neighborhood U0 = {Z 0 6= 0}, with coordinates z = Z /Z 0 , and write
f (Z) = (Z 0 )d f0 (z) on U0 . By the Chain Rule,


f (Z) = d(Z 0 )d1 f0 (z), (Z 0 )d f0 (z) 6= 0.

But f0 (z) = 0 exactly when z Vf U0 , so f0 (z) cannot vanish on Vf0 = Vf U0 . By

Theorem 5.2, the set Vf U0 is a smooth complex manifold. A similar argument proves
Vf U is smooth for = 1, . . . , n. Finally, Vf is a closed subset of Pn , hence is compact.
The set Vf is an algebraic hypersurface of degree d. More generally, if f : Cn+1 Ck has
component functions which are homogeneous polynomials of degree d = (d1 , . . . , dk ), and if
rk Df (Z) = k except at Z = 0, then the zero locus of f projects to a closed submanifold of
Pn , called a complete intersection of (multi-)degree d.
Example 5.5 The Fermat hypersurface of degree d in Pn is the zero locus of the polynomial
f (Z) = (Z 0 )d + + (Z n )d . The Fermat quadric in P3 is equivalent, after a linear change
of coordinates, to the quadric surface Q P3 defined by Z 0 Z 3 Z 1 Z 2 = 0. The Segre
mapping
[X 0 : X 1 ] [Y 0 : Y 1 ] P1 P1 7 [X 0 Y 0 : X 0 Y 1 : X 1 Y 0 : X 1 Y 1 ] P3

is a biholomorphism from P1 P1 to Q.
The twisted cubic curve C P3 , defined as the image of the map
[s : t] P1 7 [s3 : s2 t : st2 : t3 ] P3

is locally a complete intersection, but is not a complete intersection. A proof of this fact
requires examination of commutative-algebraic local data near the point [0 : 0 : 0 : 1] and is
not carried out here.
2
It is often the case in algebraic geometry that families of objects are parametrized by
points of a projective variety. The simplest example is the set of linear subspaces of dimension
k < n in Pn , which is the Grassmannian of (k + 1)-planes in Cn+1 . A property possessed by
some members of such a family is generic if the set of members failing to satisfy the property
lies in a proper subvariety of the family. For example, if ` P3 is a line, then the generic
line in P3 is disjoint from `.
The set of algebraic hypersurfaces of degree d in Pn may be regarded as a projective
space:

The set of monomials of degree d in n + 1 variables is a vector space of dimension
n+d
=: N + 1, and two monomials define the same projective variety if and only if they
d
differ by a non-zero multiplied constant. The set of degree d hypersurfaces in Pn is therefore
PN .
42

Proposition 5.6 A generic hypersurface of degree d in Pn is smooth.

5.1

The Local Structure of Analytic Hypersurfaces

By definition, an analytic variety is a subset V of a complex manifold M which is locally

the common zero locus of a finite set of analytic functions. A hypersurface is locally the zero
locus of a single function.
Locally, an analytic hypersurface is either a complex manifold, or else is a finite-sheeted
branched cover of a polydisk, branched along an analytic subvariety. The prototypical behavior is exhibited by Weierstrass polynomials, which are holomorphic functions of the form
(5.1)

f (z, w) = w d + ad1 (z)w d1 + + a1 (z)w + a0 (z),

with z Cn1 , w C, and ai : Cn1 C holomorphic.

Theorem 5.7 Let f : U Cn C be a holomorphic function on a neighborhood of the

origin, not identically zero, but with f (0) = 0. Then there exists a polydisk containing 0,
a Weierstrass polynomial g : C, and a holomorphic function h : C with h(0) 6= 0
and f = gh on .
proof After a linear change of coordinates, one may assume f (0, w) is not the zero function,
i.e. that the w-axis is not contained in the zero locus of f . There is a non-zero constant a
with f (0, w) = aw d + o(w d ) near w = 0. Choose > 0 so that f (0, w) 6= 0 for |w| = ,
and choose a multi-radius r > 0 such that f (z, w) 6= 0 for |w| = and |z| < r. For each z
with |z| < r, there are exactly d zeros of f in the disc {z} {|w| < }. (This is clear for
z = 0, and follows for |z| < r by continuity, since f (z, w) 6= 0 on |w| = .) Let these zeros be
denoted by b1 , . . . , bd . For each fixed z, and for an arbitrary positive integer k, the Residue
Theorem gives
Z
d
X
1
1
f
k
wk
bj =
(z, w) dw.
2i |w|=
f (z, w) w
j=1

Thus while the roots bj are not necessarily analytic functions of z, their power sums are. Let
1 , . . . , d be the elementary symmetric polynomials and set
g(z, w) = w d 1 (z)w d1 + + (1)d d (z).

The function g is holomorphic on the polydisc where |z| < r and |w| < , and vanishes
exactly where f vanishes.
It remains to check that the function h = f /g extends to a holomorphic function on
. But for each z with |z| < r, the function h(w) = f (z, w)/g(z, w) has only removable
singularities, hence is defined everywhere. The integral representation
Z
h(z, )
1
d
h(z, w) =
2i ||=r z w
43

proves that h is holomorphic in z.

Once the coordinates (z, w) have been chosen in the proof of Theorem 5.7, it is clear that
the decomposition f = gh is unique. In addition, the proof establishes a Riemann extension
theorem for functions of several variables:
Proposition 5.8 Let f : C be a holomorphic function on a polydisc, and assume
g : \ {f = 0} C is a bounded holomorphic function. Then there is an extension
g : C of g.
e
Let On,z0 denote the ring of germs of local holomorphic functions at z0 Cn . Understanding the local structure of an analytic variety V Cn is tantamount to understanding
the zero locus of a germ f On , the ring of germs at the origin. Units in On are exactly
germs which do not vanish at 0, i.e. locally convergent power series with non-zero constant
term.
A (non-zero) non-unit in On is sometimes said to be the germ of an analytic variety at
0 Cn . An analytic variety V is reduced if the local defining functions for V vanish to first
order, is irreducible if V is not the union of (non-empty) proper, closed subvarieties, and is
locally irreducible at p V if there is a neighborhood of p on which V is irreducible. Thus,
an analytic variety V Cn is reduced and irreducible at 0 Cn if and only if the germ of
V at 0 is an irreducible element of On . Corollary 5.12 below implies that locally, an analytic
hypersurface is a finite union of irreducible hypersurfaces.
Proposition 5.9 The ring On is a unique factorization domain.
proof (Sketch) The assertion is true when n = 1, and the Weierstrass Preparation Theorem says every f On is the product of a unit h and a Weierstrass polynomial g On1 [w].
By Gauss lemma, the proposition follows by induction on n.
Proposition 5.10 If f and g are relatively prime in On , then there is a polydisk on which
f and g define holomorphic functions whose germs are relatively prime in On,z for all z .
proof (Sketch) It suffices to take f and g to be Weierstrass polynomials in On1 [w]; let
On1 be their resultant, and let be a polydisk on which f , g, and define holomorphic
functions. If the germs of f and g have a common factor h in On,z for some z , then
h|, implying h On1 . Since f does not vanish identically along the w-axis, h must be the
zero element in On1 .
The next result, the Weierstrass Division Theorem, gives as a corollary the divisibility
criterion for a germ to vanish on the zero locus of another germ. This proves the assertion
made above: locally an analytic hypersurface is a union of irreducible hypersurfaces.

44

Theorem 5.11 Let g On1 [w] be a Weierstrass polynomial of degree d in w. If f On ,

then there is an h On and a polynomial r On1 [w] of degree < d in w with f = gh + r.
proof (Sketch) Choose > 0 so that g(0, w) 6= 0 for |w| = , and choose a multi-radius
r > 0 such that g(z, w) 6= 0 for |w| = and |z| < r. For each z Cn1 with |z| < r, define
Z
1
f (z, ) d
.
(5.2)
h(z, w) =
2i ||= g(z, ) w
Then h is holomorphic (cf. proof of Theorem 5.7), and so is r = f gh. But
Z
Z
1
f (z, ) g(z, ) g(z, w)
1
f (z, )
r(z, w) =
d =:
p(z, , w) d,
2i ||= g(z, )
w
2i ||= g(z, )
with p(z, , w) a polynomial of degree < d in w.
Corollary 5.12 If f On is irreducible and if h On vanishes on Vf , then f |h in On .
proof Assume without loss of generality that f On1 [w] is a Weierstrass polynomial
of degree d in w. Since f is irreducible, f and f /w are relatively prime in On1 [w]; let
On1 be their resultant. Since is non-vanishing on some polydisk about the origin
by Proposition 5.10, f (z, ) has distinct roots for z Cn1 .
Using Theorem 5.11, write f = gh + r with r On1 [w] of degree < d in w. On slices
of where z is fixed, the function h has the same roots as f by equation (5.2), and in
particular has d distinct roots. Since r has degree < d in w, r 0.

5.2

Singularities of Algebraic Varieties

(Lecture by E. Hironaka.) Let M be a complex manifold, with atlas {U , }. Recall that

a subset V M is an analytic subvariety if, for every p M, there is a neighborhood U
of p and a finite collection of holomorphic functions f i on Ucalled local defining functions
for V in Usuch that the common zero locus {f i = 0} U is exactly V U. An analytic
subvariety is necessarily a closed subset of M, and if p 6 V , then it is conventional to take
U = M \ V , with local defining function f = 1.
These definitions may be phrased globally in terms of the structure sheaf OM ; define the
ideal sheaf of V to be the subsheaf IV OM whose stalk at p M consists of germs of
holomorphic functions which vanish on V . The structure sheaf of V is defined by the exact
sequence
0 IV OM |V OV 0.
The ideal and structure sheaves of an analytic subvariety are coherent.
45

In order to study locally the singularities of a variety V , it suffices to work near the
origin in Cn . Let {f i }ki=1 be local defining functions for V at 0, chosen so that there is a
neighborhood U of 0 on which the differentials df i are not everywhere linearly dependent.
The origin is a smooth point of V if the map Df (0) : Cn Ck has rank k. By the Implicit
Function Theorem, this is equivalent to V being a manifold at 0. Otherwise 0 is a singular
point of V ; the set of all singular points is denoted Sing V , and the smooth locus V V
is the complement of the singular set. The smooth locus is a complex manifold and the
dimension of V is defined to be the dimension of V .
Proposition 5.13 Let V be an analytic subvariety of a complex manifold M. Then the set
of singular points Sing V is an analytic subvariety of M.
proof It suffices to work locally. Suppose 0 V is a singular point, and let U Cn
be a neighborhood of 0 on which {f i }ki=1 is a set of local defining functions for V . The
intersection U Sing V is cut out by {f i } together with the set of determinants of k k
minors of Df : U T Cn . The latter are local holomorphic functions in U.
Proposition 5.14 An analytic variety V is irreducible if and only if the smooth locus V is
connected.
proof (Sketch) If V = W1 W2 is a union of non-empty, proper closed subvarieties, then
W1 W2 Sing(W1 W2 ), so the sets Wi \ (Sing V ) separate V . Conversely, the closure of
a component of V is a proper, non-empty, closed subvariety of V .
Locally, a hypersurface V = Vf is cut out by a single holomorphic function f . The
multiplicity of V at p is defined to be the order of vanishing of f at p:

(
)
 If

(5.3)
multp (V ) = max m N  I (p) = 0 for all multi-indices I with |I| < m .
 z

If f is expanded as a power series centered at p with homogeneous terms fi of weight i in z,

then m = multp (V ) is the degree of the first non-vanishing term: f = fm + fm+1 + . The
tangent cone of V at p is the zero locus of fm in Tp Cn , which is alternately characterized as
the set of tangent vectors

( n
)

X


Tp V =
X i i  fm (X 1 , . . . , X n ) = 0 .
z 
i=1

For example, if V is smooth at p, then Tp V is the usual tangent space, while the tangent
cone at the origin of the zero locus of f (x, y) = y 2 x2 x3 is the union of the lines y = x.
46

Affine Varieties and Coordinate Rings

Let R be a commutative ring with unit element. A radical ideal r R is an ideal with the
property that if xn r for some n > 1, then x r. In words, r contains the nth roots in R
of all its elements.
Let V Cn be an algebraic subvariety, that is, an analytic variety whose local defining
functions are polynomials. By Proposition 6.4 below, V is the common zero set of finitely
many polynomials on Cn . The ideal r(V ) = {f C[z] : f |V = 0} of polynomials which
vanish along V is the radical of I, the ideal of global sections of the ideal sheaf of V , and
the affine coordinate ring of V is the quotient O(V ) = C[z]/r(V ). Conversely, given a
radical ideal r C[z], the associated affine algebraic variety is V (r) = {z Cn | f (z) =
0 for all f r}. There is a one-to-one correspondence between radical ideals in C[z] and
affine algebraic varieties in Cn . Under this correspondance, prime ideals correspond to
irreducible varieties, and maximal ideals correspond to points.
A projective algebraic variety is a (necessarily closed) subset V Pn whose intersection
V with the affine coordinate chart U is an affine algebraic variety for each j = 0, . . . , n.
Equivalently, a projective algebraic variety is the common zero locus of finitely many homogeneous polynomials in the homogenerous coordinates on Pn .
In algebraic geometry, one often works with the Zariski topology on an affine algebraic
variety. A Zariski-open set in an algebraic variety V Cn is the complement of an algebraic
subvariety of V . Regarding C as an affine variety, Zariski-open sets are exactly sets with
finite complement. The Zariski topology on a complex variety of positive dimension is never
Hausdorff. One advantage of the Zariski topology is that algebraic objects (e.g. meromorphic
functions) are determined by their restriction to an arbitrary non-empty open set. A striking
example is that with respect to their Zariski topologies, P1 and an elliptic curve E are not
even locally isomorphic.
Example 5.15 The basic objects of algebraic geometry are schemes. Under the correspondance just outlined, affine varieties correspond to radical ideals r C[z], via their coordinate
ring O(V ). The points of V correspond to maximal ideals in O(V ), and all questions about
V may be phrased in terms of O(V ).
It is natural to ask what sort of geometric object can be associated to an arbitrary
commutative ring R with unity. To this end, let spec R denote the set of prime ideals of R.
For each ideal I R, the associated basic closed set is defined to be the set of prime ideals
containing I; the complements of basic closed sets form a basis for the Zariski topology on
spec R. Finally, the topological space spec R is endowed with a sheaf O of rings; the stalk
of O at a point (i.e. prime ideal) p spec R is the localization Rp. The topological space
spec R together with the sheaf of rings is called an affine scheme.
Generally, a topological space X together with a sheaf of rings OX is a scheme if every
point p X has an open neighborhood U such that (U, OX (U)) is isomorphic to some affine
scheme spec R. Various geometric constructions (intersection theory, for example) work
47

better in the category of schemes than in the category of varieties.

Maximal ideals in R correspond to closed points of spec R. The additional points
corresponding to non-maximal prime ideals p are generic points, whose closures contain,
among other things, the closed points of the subscheme cut out by p. An illustrative example
is furnished by R = C[x, y], whose spectrum is the affine plane. Points (x0 , y0 ) C2 are
in one-to-one correspondance with maximal ideals (x x0 , y y0 ) R. Every irreducible
polynomial, e.g. x2 y, generates a prime ideal; the closure of the point p = (x2 y) contains
all the closed points lying on the parabola y = x2 .
2
An Introduction to Desingularization
One general goal in algebraic geometry is to find, for a given variety V , a variety Ve which
is isomorphic to V away from a Zariski-closed set, and which has nicer singularities. A
prototypical example is the following desingularization result.
Theorem 5.16 Let V Cn be an algebraic variety. Then there exists a smooth algebraic
variety Ve CN and a proper, surjective, holomorphic map : Ve V whichon the
complement of a divisoris a biholomorphism onto its image.
This desingularization can in fact be accomplished for arbitrary varieties defined over
fields of characteristic zero, as shown by H. Hironaka in the late 1950s. Bierstone and Milman
have recently given an explicit procedure for desingularization in characteristic zero. A more
elementarybut already non-trivialresult can be obtained from commutative algebra via
the correspondance between varieties and ideals, see Proposition 5.21 below.
Let V Cn be an affine (algebraic) variety. To each p V , let mp denote the corresponding maximal ideal in the coordinate ring O(V ). The local ring of V at p is the localization
Op (V ) of O(V ) at mp . An element of the local ring should be regarded as a germ of a regular
function at p; the image of f O(V ) in O(V )/mp ' C is the value of f at p.
Generally, a local ring is a Noetherian, commutative ring with unity having a unique
maximal ideal. The local ring of a variety is indeed a local ring in this sense. The Krull
dimension of a commutative ring R is one less than the maximum length of a chain of nested
prime ideals in R. For example, the polynomial ring C[z] in n variables has Krull dimension
n. A local ring O with maximal ideal m is regular if dim O = dim(m/m2 ).

Proposition 5.17 A variety V is non-singular at p if and only if Op (V ) is a regular local

ring.
Example 5.18 Let V be the cusp curve x3 = y 2 in C2 . The affine coordinate ring of C2 is
C[x, y], and the local ring at the origin p is ap = (x, y), the maximal ideal of C[x, y]. The
ideal a2p is (x2 , xy, y 2), and the quotient is ap /a2p = C[x, y], so C2 is smooth at the origin, as
expected.
48

By contrast, let mp = (x, y) IV be the maximal ideal of the affine coordinate ring
C[x, y]/(x2 y 3). Then
mp /m2p = (x, y)/(x2 , xy, y 2, x2 y 3) = (x, y)/(x2, xy, y 2),
which has Krull dimension two, while V has dimension one. Thus the local ring Op (V ) is
not regular, so the cusp curve V is not smooth at the origin.
2
Proposition 5.19 Let ap be the maximal ideal of a regular local ring Op , let V be a variety
containing p, and let mp be the maximal ideal in the coordinate ring of V at p. Then
m/m2 ' a/(a2 + IV ).
proof Let {f i}ki=1 be local defining functions for V at p. Then the rank of the Jacobian
Df (p) is equal to dim(ap /a2p ) dim(mp /m2p ). But
dim(ap /a2p ) dim(mp /m2p ) = dim(a2p + IV )/a2p ,
so V is non-singular in the analytic sense if and only if the dimension is as expected.
Normal Varieties
Let X be an affine algebraic variety. The quotient field K(X) of the integral domain O(X)
is the function field of X; elements of K(X) are meromorphic functions on X. For each
p X, there are inclusions O(X) Op (X) K(X).
If A and B are integral domains, then a ring homomorphism : A B induces a
morphism spec B spec A. Conversely, if X and Y are affine varieties, then a regular
morphism f : X Y induces a ring homomorphism O(Y ) O(X). If f is surjective, then
the induced map on coordinate rings is injective.
An element x B is integral over A if there exists a monic polynomial q A[] with
q(x) = 0. If every x B which is integral over A actually lies in A, then A is integrally
closed in B.
Lemma 5.20 An integral domain R is integrally closed in its quotient field Q if and only if
the localization Rm is integrally closed in Q for every maximal ideal m.
If X is irreducible, then K(X)the quotient field of O(X)is the quotient field of Op (X)
for every p X. A variety X is normal at p if Op (X) is integrally closed in K(X), and is
normal if normal at p for every p X. By Lemma 5.20, X is normal if and only if O(X) is
integrally closed in K(X).
A birational map f : X Y is a map which induces an isomorphism K(Y ) K(X) of
function fields. In other words, a birational map is an isomorphism off a Zariski-closed set.
49

A finite morphism is a morphism with finite fibres; more concretely, the preimage of each
point is a finite set. Branched covers are finite, while quadratic transformations (blowups) are not. A desingularization of Y is a surjective, birational mapping f : X Y with
X a smooth variety. A weaker but more elementary construction gives a birational map
f : X Y with X normal: Let X be an irreducible affine variety, and let O(X) be the
integral closure of O(X) in K(X). The associated variety, together with the induced map
to X, is the normalization of X.
e of closed points of spec O(X) is a normal variety endowed
Proposition 5.21 The variety X
e X.
with a finite, surjective, birational morphism : X
Theorem 5.22 A regular local ring is integrally closed in its quotient field.

By definition, X is smooth at p if and only if Op (X) is a regular local ring, while X is

normal at p if and only if Op (X) is an integrally closed local ring. By Theorem 5.22, if X is
smooth at p, then X is normal at p.
Proposition 5.23 If p X is a normal point, and if f K(X) is regular on a deleted
neighborhood of p, then f extends to p, i.e. f Op (X).
Example 5.24 Let X = {x3 = y 2 } C2 be the cusp curve. The function t = f (x, y) = y/x
restricts to a meromorphic function on X \ 0, but t2 x = 0 O(X)[t], so t cannot be
extended to a regular function on X. Thus the origin is not a normal point of X.
Intuitively, the normalization described in Proposition 5.21 adds in meromorphic funce is
tions which are defined away from a point. More precisely, the coordinate ring of X


2
2
3
O(X)[t]/(t x) = C[x, y]/(y x ) [t]/(t2 x),
and the normalization is accomplished by mapping the (x, t)-plane to the (x, y)-plane by
(x, t) 7 (x, xt). Under this mapping, the equation x3 y 2 = 0 pulls back to x t2 = 0,
whose zero locus is a smooth curve. This mapping is exactly the blow up at the origin
expressed in an affine chart.
2
Proposition 5.25 Let X be a normal variety. Then Sing X has codimension 2 in X.
Consequently, if a curve is normal at p, then p is a smooth point. A normal variety of
arbitrary (positive) dimension is locally irreducible.
Let X be a normal variety, K = K(X) the function field, and let L K be a finite field
extension. The coordinate ring O(X) has an integral closure O(X) in L, whose spectrum
e is equipped with a finite, surjective morphism to X. Conversely, if f : X
e X is a finite,
X
50

e K(X)] is finite,
surjective morphism to a normal variety X, then the field extension [K(X),
e is the coordinate ring of X.
e
and the integral closure of O(X) in K(X)
The degree of f is by definition the degree [L : K] of the field extension. For every
x X, #{f 1 (x)} deg f , and the maximum is achieved on a Zariski-open set. The
algebraic subvariety
B = {x X | #{f 1 (x)} < deg f }
e \f 1 (B) X \B is an unbranched
is called the ramification locus of f . The restriction f : X
covering, and the covering transformations permute points in the fibres of f .

Proposition 5.26 The group of deck transformations acts transitively on fibres if and only
e
1 (X), if and only if the extension L K is Galois.
if f 1 (X)
If B Y is normal and of pure codimension one, then to each subgroup of 1 (Y \ B) is
associated a covering space X which inherits the structure of an algebraic subvariety. The
following is due to Enriques, Grauert-Remmert, and Grothendieck.

Theorem 5.27 With the above notation, there is a unique normal algebraic variety Ye which
completes X and which admits a finite map to Y such that the diagram
X
Ye

Y \ B , Y

commutes.
e f 1 B X
e on which Df is not of maximal rank is the branch locus of f .
The set B
The following remarkable result on the structure of the branch locus is due to Zariski.
Theorem 5.28 The branch locus of f is of pure codimension one, that is, every irreducible
e has codimension one.
component of B
A variety X is projectively normal if there is a surjection K(Pn ) K(X). The following
is known as the Stein Factorization Theorem.

Theorem 5.29 Let f : X Y be a surjective, finite map of normal varieties of the same
dimension, and let Ye be the normalization of Y in K(X). Then the natural map X Ye is
a surjective, birational map with connected fibres, and the natural map Ye Y is a finite,
surjective morphism.

51

Divisors, Meromorphic Functions, and Line Bundles

Let M PN be a complex manifold embedded in a projective space. A hyperplane section

of M is the intersection of M with a hyperplane in PN . The hyperplane sections of M carry
information about the embedding. Conversely, given an arbitrary complex manifold it is of
interest to describe and classify hyperplane sections intrinsically. The relevant objects are
called divisors; divisors, meromorphic functions, and line bundles are three ways of regarding
essentially the same data from geometric, algebraic, and topological points of view.

6.1

Divisors and Line Bundles

Let M be a complex manifold. A divisor D on M is a locally finite sum of closed, reduced,

irreducible analytic hypersurfaces (the components of D) with non-zero integer coefficients.
Closed means closed as subsets in the complex topology, sum with integer coefficients
should be taken in the spirit of free Abelian groups, with the distinction that the sum here
may be infinite (meaning there may be infinitely many components), and locally finite
means that every x M has a neighborhood U which intersects only finitely many components. A divisor is effective if every component has positive coefficient. An effective divisor
is locally cut out by a holomorphic function ; the function vanishes along a union of
irreducible analytic hypersurfaces, and the integer attached is the order of vanishing.
If M itself is compact, then a divisor is exactly an element of the free Abelian group
on the set of closed, irreducible analytic hypersurfaces. The group of divisors is denoted
div (M). The support of a divisor D is the union of the components of D.
Example 6.1 Let M be a compact Riemann surface. A divisor on M is a finitePset of points
weighted with integers. Alternately, a divisor may be regarded as a sum D = xM mx x,
where the coefficient mx Z is understood
Pto be zero except for at most finitely many x M.
The degree of D is defined to be deg D = mx .
2

Example 6.2 A divisor on Pn is a finite sum of algebraic hypersurfaces with integer

coefficients. (Certainly, every such sum is a divisor; the converse is essentially Theorem 6.13
below.)
If N M is a closed complex submanifold and D is a divisor on M, then N D (taken
in the obvious sense) is a divisor on N. Thus, projective manifolds have many divisors.
At the other end of the spectrum, there exist compact complex manifolds having no divisors.
Generic compact tori provide examples of this counterintuitive behaviour.
2
Assume D is an effective divisor on M, and let U = {U } be a locally finite open cover
such that there exist holomorphic functions O(U ) which locally cut out D. If U is

52

non-empty, then there is a non-vanishing holomorphic function on U such that



=
 .

U

The collection { } is therefore but a 1-cocycle of the nerve of U with coefficients in OM

,

1
so an effective divisor D represents an element [D] H (M, OM ). As noted previously,
this cohomology group may be interpreted as the group of holomorphic line bundles on M;
addition of divisors corresponds to taking tensor products of line bundles. If D is effective,
then [D] may be interpreted as the line bundle [D] = [D]1 dual to [D]. Finally, every
divisor D is uniquely a difference of effective divisors, say D = D+ D , so an arbitrary
divisor D represents the line bundle

[D] = [D+ ] [D ] .
This provides the following partial dictionary between divisors and holomorphic line bundles.

Proposition 6.3 The map D 7 [D] is a group homomorphism from div M to H 1 (M, OM
).

This homomorphism is neither injective nor surjective in general; distinct divisors may
give equivalent bundles, and not every holomorphic line bundle arises from a divisor. Happily,
it will presently be possible to give a simple characterization of when these possibilities occur.

6.2

Meromorphic Functions

A meromorphic function on a Riemann surface is a holomorphic map to P1 . Unfortunately,

there is no analogous simple criterion which gives meromorphic functions on complex manifolds of dimension greater than one. A better approach is to regard meromorphic functions
as functions which are locally a quotient of holomorphic functions. Making this idea precise is not difficult, and the corresponding notion reduces to the expected one for Riemann
surfaces. Explicitly, the difference between Riemann surfaces and higher-dimensional manifolds is that functions of one variable which vanish simultaneously have a common factor,
while in two or more variables this is no longer automatic.
Let Mpre be the presheaf on M whose field of sections over an open set U M is the
quotient field of the integral domain OM (U), and let M denote the completion. A meromorphic function on M is a global section of the sheaf M. A non-zero meromorphic function f
is therefore locally the quotient of holomorphic functions in the following sense: For every
x M, there exists a neighborhood U of x and a pair of relatively prime holomorphic functions 0 and (in particular, neither is identically zero) such that f is represented on U
by 0 / . The field of global sections of M is called the function field of M, and is often
denoted K(M).
53

Up to multiplication by a non-vanishing local holomorphic function, 0 and depend

only on f M. Thus there is a zero divisor (f )0 , locally defined by 0 , and a polar divisor
(f ) , locally cut out by . The divisor (f ) = (f )0 (f ) is the principal divisor associated
to f . The indeterminacy set of f is the intersection of the zero and polar divisors of f , and is
therefore an analytic variety of codimension two. Away from the indeterminacy set, f may
be regarded as a holomorphic map to P1 .
Proposition 6.4 Every analytic hypersurface D Cn is the zero locus of an entire function.
Every meromorphic function f on Cn is a quotient of entire functions.
proof By Theorem 4.9, H q (Cn , O) = 0 for q 1, while H q+1 (Cn , Z) = 0 for q 0
since Cn is contractible. The long exact sequence induced by the exponential sheaf sequence
contains
H q (Cn , O) H q (Cn , O ) H q+1(Cn , Z),

which for q = 1 implies H 1 (Cn , O ) = 0. In particular the cocycle [D], represented by a

locally finite cover U = {U } and local holomorphic functions on U , is a coboundary.
After refining U if necessary, there exist local non-vanishing holomorphic functions h such
that (omitting restrictions to U )
h

= =
.

h
The local functions {h } agree on overlaps, and therefore define an entire function.
The second assertion is now immediate: Choose entire functions cutting out the zero and
polar divisors of f .
Divisors D and D 0 on M are linearly equivalent if D D 0 is principle, that is, if there
is a meromorphic function f on M with D D 0 = (f ). On Pn , a quotient of homogeneous
polynomials of the same degree (in the homogeneous coordinates) defines a meromorphic
function. Therefore, two algebraic hypersurfaces of the same degree define linearly equivalent
divisors.
It is sometimes useful to regard a divisor as a global section of the quotient sheaf M /O ;
this highlights the group structure and functorial properties of div M. The subgroup consisting of principal divisors is often denoted div0 (M), and the quotient
Pic (M) := div (M)/div0 (M)
is called the Picard group of M.

54

6.3

Sections of Line Bundles

Let p : L M be a holomorphic line bundle. A holomorphic section of L is a holomorphic

map : M L with p = identity. When a line bundle is presented as a 1-cocycle
{ }, a section is realized concretely as a collection of holomorphic functions { } satisfying
= . For this reason, holomorphic sections are sometimes called twisted functions.
If L = [D] comes from a divisor, then OM (D) = O(L) is used to denote the sheaf of germs
of sections of L. The following is immediate:
Lemma 6.5 If L = [D] is induced by an effective divisor, then each collection { } of local
defining functions for D gives a holomorphic section of L.
A meromorphic section of a line bundle is defined to be a collection of local meromorphic
functions {f } satisfying the compatibility condition f = f . In other words, a meromorphic section of L is a global section of the sheaf O(L) M. From Lemma 6.5, it is easy
to derive a satisfactory supplement to Proposition 6.3.

Proposition 6.6 A line bundle L H 1 (M, OM

) is of the form [D] if and only if L admits a
meromorphic section. Two line bundles L = [D] and L0 = [D 0 ] are holomorphically equivalent
if and only if D D 0 is principal.

Example 6.7 Let H denote the hyperplane class in H2n2 (Pn , Z). Every line bundle on
Pn comes from a divisor D = dH for some integer d, called the degree of the bundle. By
contrast, on a generic compact torus, only the trivial bundle comes from a divisorthe empty
divisorsince there are no other divisors. The meromorphic sections are exactly constant
functions.
2
Example 6.8 Let E be an elliptic curve, and fix a point 0 E. For each x E, the divisor
(0) (x) has degree zero, but is principal if and only if x = 0. This may be seen with a bit
of elementary complex analysis. If f : E P1 were a meromorphic function with principal
divisor (0) (x), then f would have a simple pole at x and no other poles. The meromorphic
1-form f (z) dz would therefore have non-zero total residue, which is impossible.
A choice of 0 E defines a group law with identity element 0 (coming from a realization
of E as a quotient C/). The following illustrates the beautiful interplay between complex
analysis, projective geometry, and algebraic geometry.
Proposition 6.9 A divisor (x1 )+(x2 )+(x3 ) on E is principal if and only if x1 +x2 +x3 = 0
as elements of E, if and only if the images of the points xi under the embedding by the function and its derivative (see Example 1.8) are collinear in P2 .

55

proof (Sketch) A choice of 0 E determines a representation E = C/. The embedding

E , P2 defined by the -function has a flex at 0 because 0 has a triple pole; in other words,
the tangent line to E at 0 (which is a hyperplane in P2 ) intersects E to order three, so the
divisor 3(0) is a hyperplane section of E. Every other hyperplane section consists of three
(possibly non-distinct) collinear points, and these are exactly the effective divisors linearly
equivalent to 3(0). The fact that a sum of collinear points is zero under the group law of E
is related to addition theorems for the -function, see Ahlfors, Complex Analysis, p. 277.
A restatement of the first assertion in Proposition 6.9 is worth mentioning. As in equation (4.6) above, let J0 (E) denote the group of degree-zero line bundles on E. The map
E J0 (E) which sends x E to the line bundle associated to the divisor (0) (x) is
an isomorphism of Abelian groups. Observe that when x 6= 0, the associated line bundle is
topologicallybut not holomorphicallytrivial.
2

6.4

Chows Theorem

The theory of complex manifolds is analogous in many ways to complex algebraic geometry.
The rough principle, known as the GAGA Principle from Serres Geometrie algebrique et
geometrie analytique, is that compact analytic objects in projective space and morphisms
between them are in fact algebraic. Chows Theorem (Theorem 6.13 below) is the prototypical example of the GAGA Principle. As will become apparent, the underlying reason for
GAGA is the simplicity of the analytic cohomology of Pn .
Let M be a compact complex manifold of (complex) dimension n. Because M is oriented,
there is a fundamental class [M] H2n (M, Z) and a Poincare duality isomorphism
Hk (M, Z) ' H 2nk (M, Z).
The induced ring structure on homology is the intersection pairing, which is described in
detail below for Pn . If V is a k-cycle Poincare dual to H 2nk (M, Z), then
R
R
i. For all H k (M, Z), V = M = ( )[M];
R
ii. For all (2n k)-cycles W , V W = W .

Example 6.10 (Intersection theory in projective space.) Let M = Pn . The cohomology

ring is a truncated polynomial ring in one variable. More precisely, let H 2 (Pn , Z) denote
the cohomology class Poincare dual to a hyperplane class H H2n2 (Pn , Z). Then
H (Pn , Z) ' Z[]/hn+1 i '

n
M

Zk .

k=0

Since Pn is oriented, H2n (Pn , Z) is naturally isomorphic to Z, and this isomorphism is

compatible with Poincare duality and the description of the cohomology ring just given.
56

Let V Pn be an algebraic hypersurface of degree d. A line in Pn intersects V in

at most d points, and if intersections are counted with multiplicities, then every line intersects V exactly d times. From Property i. above it follows that V is Poincare dual to d.
By intersecting further with hyperplanes, it is easy to see that every positive cohomology
classnamely, every dnk with d > 0, 0 k nis Poincare dual to a homology class
represented by an analytic cycle. More generally, if V is a complete intersection
Q of multidegree (d1 , . . . , dnk ), then the homology class [V ] is Poincare dual to V = ( i di )nk .
The coefficient is the number of points of intersection when V is intersected with a generic
linear subspace of complementary dimension.
For arbitrary homology classes of complementary dimension, the intersection product
[V ][W ] may be interpreted as an oriented intersection number, giving geometric significance
to condition ii. Indeed, every non-zero homology class in Pn is Poincare dual to dnk for
some d Z, and may therefore be represented by V for a smooth complete intersection of
complex dimension k. Two generic smooth, complete intersection varieties of complementary
dimension intersect transversally (in finitely many points).
2
In an arbitrary complex manifold, not every homology class is representable by a smooth
submanifold (even up to sign). However, it is always possible to represent a homology class
by a cycle whose support is an embedded manifold except for a singular set of smaller dimension. Further, for each pair of homology classes of complementary dimension, representative
cycles may be chosen which intersect transversally in finitely many manifold points. Let V
and W be such representative cycles, and let p be a point of (transverse) intersection of their
supports. If these cycles are oriented, then each tangent space Tp V , Tp W has an induced
orientation, determined by ordered bases {v1 , . . . , vk } and {w1 , . . . , wnk }. The intersection
number Ip (V, W ) of V and W at p is defined to be 1, depending on whether the orientation
on Tp M determined by the ordered basis {v1 , . . . , vk , w1 , . . . , wnk } agrees with the orientation of Tp M or not. The oriented intersection number of the homology classes [V ] and [W ]
is defined to be
X
I(V, W ) = [V ] [W ] =
Ip (V, W ).
pV W

If [V ] and [W ] are represented by transversely intersecting analytic cycles (i.e. embedded

subvarieties of M), then [V ] [W ] 0. This follows immediately from the observation that
the usual orientation on Cn agrees with the orientation induced from a pair of complex
subspaces of complementary dimension.
Example 6.11 Fix an integer k and let H k = OP1 (k). The Hirzebruch surface Fk is the
P1 -bundle P(H k O) obtained by adding an infinity section. The two-dimensional homology
of Fk is generated by a fibre class [F ] and the class [S0 ] of the zero section. The intersection
pairing is given by
[F ] [F ] = 0,

[F ] [S0 ] = [S0 ] [F ] = 1,
57

[S0 ] [S0 ] = k.

The homology class of the infinity section S is equal to [S0 kF ], as is readily checked by
solving [S ] = [aF +bS0 ] for a and b. In particular, [S ][S ] = k.
2
Using intersection theory in projective space, it is easy to prove two basic but useful
GAGA-type results.
Proposition 6.12 Let f : Pn Pn be a biholomorphism. Then there is a linear transformation fe : Cn+1 Cn+1 which induces f . Briefly, every automorphism of Pn is linear.

proof Let H Pn be a hyperplane, and let f be a biholomorphism of Pn . Then f (H)

is an analytic subvariety which is homologous to a hyperplane. Let p1 and p2 be points in
f (H), and let ` ' P1 be the line joining them. Then {p1 , p2 } ` H, so if there were
only finitely many points of intersection then [`] [H] 2. But [`] [H] = 1 for homological
reasons, so ` H is infinite, hence must be all of ` since ` and H are analytic varieties. It
follows that ` = p1 p2 f (H) for every pair of points p1 , p2 f (H), so that f (H) is actually
a hyperplane. It is now a matter of linear algebra to verify that f must be covered by a
linear isomorphism of Cn+1 .
A surprisingly easy generalization of Proposition 6.12 is the prototypical GAGA theorem,
usually known as Chows Theorem.
Theorem 6.13 Let V Pn be an analytic subvariety. Then V is an algebraic subvariety.
proof The long exact sequence associated to the exponential sheaf sequence on Pn gives
an isomorphism
H 1 (Pn , O ) ' H 2(Pn , Z) ' Z;

in words, holomorphic line bundles on Pn are classified by their first Chern class. Let V Pn
be an analytic hypersurface, and consider the holomorphic line bundle L = O(V ). There is
a positive integer d such that L = OPn (d). The space of sections of OPn (d) is exactly the
space of homogeneous, degree d polynomials in the homogeneous coordinates. It follows that
V is the zero locus of a single homogeneous polynomial, i.e. is an algebraic hypersurface.
(This gives an alternate proof that the image of a hyperplane under an automorphism is a
hyperplane.)
Now let V be an arbitrary analytic variety of dimension k in Pn . The strategy is to show
that for every p 6 V , there is a homogeneous polynomial F which vanishes along V , but with
F (p) 6= 0. It then follows that the common zero locus of the set of polynomials vanishing
along V which is a priori larger than V is equal to V , so that V is an algebraic variety.
Pick a hyperplane H not containing p, and project away from p. The image of V is
an analytic subvariety in H. Proceed inductively until the image is a hypersurface in a
linear space Pk+1, which after linear change of coordinates may be taken to be the set of
points of the form [z 0 : : z k+1 ]. By the hypersurface case proven above, there is a
homogeneous polynomial f of degree d in the variables z 0 , . . . , z k+1 whose zero locus is the
58

image of V in Pk+1. The sequence of projections may actually be accomplished in a single

step, by projecting away from a generic linear space of dimension n k 2 not containing
p. The polynomial F defined by F (z 0 , . . . , z n ) = f (z 0 , . . . , z k+1 ) is the desired polynomial.
Geometrically, the zero locus of F is the cone on V with vertex Pnk2.

Exercises
Let i : V , M be a smooth complex submanifold of a complex manifold. There is an exact
sequence
(6.1)

0 T V i T M V /M 0

of vector bundles over V , whose quotient term is called the normal bundle of V in M. The
dual of the normal bundle is the conormal bundle.
Exercise 6.1 Let V M be a smooth, irreducible hypersurface. Prove that the conormal
bundle of V in M is isomorphic to the line bundle i [V ]; in words, this is the restriction
to V of the line bundle on M induced by the divisor V . This fact is sometimes called the
first adjunction formula.
Suggestion: Let {f } be local defining functions for V ; show that after restricting to V , the
one-forms {df } constitute a global non-vanishing section of a certain line bundle.

Exercise 6.2 Find the normal bundle of Pk Pn , and prove that the exact sequence
0 T Pk i T Pn Pk /Pn 0
splits holomorphically. In fact, the total space of Pk /Pn embeds as a Zariski-open subset of
Pn ; identify this set and its complement geometrically.

If p : E M is a holomorphic vector V
bundle of rank k, then the determinant bundle
det E is defined to be the top exterior power k E. Thus det E is the holomorphic line bundle
whose transition functions are determinants of the transition functions of E. Let M be a
complex manifold, and let E = T 1,0 M denote the holomorphic tangent bundle of M. The
anticanonical bundle of M is KM = det E, and the canonical bundle of M is KM = det E .
Thus the canonical bundle of M is the bundle of holomorphic n-forms.
Exercise 6.3 Prove that KPn = OPn (n+ 1).

Exercise 6.4 Let 0 S E Q 0 be an exact sequence of holomorphic vector bundles
over a complex manifold M. Prove that det E ' det S det Q.


Exercise 6.5 Let V M be a smooth hypersurface. Prove the second adjunction formula:
(6.2)

KV = i (KM [V ]).

Suggestion: Use Exercises 6.1 and 6.4.


59

When V M is a smooth hypersurface, the induced map KM KV on global sections

is called the Poincare residue map, and is extremely useful for describing holomorphic forms
on hypersurfaces in Pn . Take M = Pn , n > 1, and let V = {f = 0} be a smooth, irreducible
hypersurface of degree d. By the second adjunction formula, KV = i OPn (d n 1). A
(local) section of the bundle nPn [V ] is a (local) meromorphic n-form on Pn which has a
single pole along V and is elsewhere holomorphic. (A global holomorphic section of nPn [V ]
exists if and only if d = deg f n + 1.) The Poincare residue of is the unique holomorphic
form 0 on V such that
df
=
0 .
f
The kernel of the Poincare residue map is the sheaf of germs of holomorphic n-forms; in
other words, there is an exact sheaf sequence 0 nPn nPn [V ] Vn1 0, whose
long exact cohomology sequence includes the terms
H 0 (Pn , nPn [V ]) H 0 (V, Vn1 ) H 1 (Pn , nPn );
the last term is trivial by Proposition 4.13. Thus, every holomorphic (n 1)-form on V is
the Poincare residue of a global meromorphic form on Pn .
Exercise 6.6 (Algebraic K3 Surfaces) Let V be a smooth quartic hypersurface in P3 . Show
that the canonical bundle of V is trivial, and find a non-vanishing holomorphic two-form
on V . (By Theorem 10.5 below, the first homology of V is trivial. In fact, V is simplyconnected. A simply-connected complex surface with trivial canonical bundle is called a
K3 surface, after Klein, Kummer, and Kodaira.)
Let M = P1 P1 P1 , and let V M be a smooth hypersurface of degree (2, 2, 2);
this means V is the zero locus of an irreducible sextic polynomial which is quadratic in the
homogeneous coordinates on each P1 factor. Prove that the canonical bundle of V is trivial,
and find eight automorphisms of V .

Exercise 6.7 Let C P2 be a smooth curve of degree d. Calculate the dimension of
the space of holomorphic one-forms on C using the Poincare residue map, thereby giving an
alternate proof of the degree formula (Exercise 2.8) for smooth plane curves.


Metrics, Connections, and Curvature

In differential geometry, a metric is a symmetric two-tensor that by declaration measures

lengths of tangent vectors. The length of a path of class C 1 is defined to be the integral
of kk,
and the distance between two points is the infimum of the lengths of paths joining
the points. A connection is a piece of data that allows tensor fields to be differentiated in
a coordinate-independent manner; this gives rise to a means of comparing tensors at different points (hence the term connection) by parallel transport. Unlike partial derivatives,
60

covariant derivatives in different directions do not generally commute. The failure of commutativity is measured by the curvature tensor of the connection. A fundamental principle
of differential geometry is that a metric determines a unique connection that is compatible
with the metric and satisfies an additional property (usually either a symmetry condition
or a compatibility with a holomorphic structure).

7.1

Hermitian and K
ahler Metrics

Let (V, J) be an almost-complex vector space. An inner product g on V is Hermitian if the

endomorphism J is g-orthogonal, that is, if g(X, Y ) = g(JX, JY ) for all X, Y V . Every
inner product may be averaged to yield an Hermitian inner product; if g is an arbitrary inner
product, then

1
(7.1)
g(X, Y ) =
g(X, Y ) + g(JX, JY )
2

defines an Hermitian inner product.

An Hermitian inner product has a C-bilinear extension gC on VC , and it is immediate
to verify that the eigenspaces V 1,0 and V 0,1 are isotropic with respect to gC . However, the
complex-valued inner product h on V defined by

1
1,0
0,1
g(X, Y ) + ig(X, JY )
(7.2)
h(X, Y ) = gC (X , Y ) =
2

is easily verified to be C-linear in the first variable, C-antilinear in the second, and to be
conjugate symmetric. Thus, h is an Hermitian form on V in the usual sense. The imaginary
part of h is denoted /2, and is skew-symmetric.
Lemma 7.1 If two of the tensors g, J, and are known, then the third is determined
uniquely.
proof
(7.3)

If J is known, then passing from g to is trivial: If X, Y V , then

(X, Y ) = g(X, JY ),

and since g is J-Hermitian, g(X, Y ) = (JX, Y ).

Suppose that g is an inner product on V and that is a skew-symmetric two-tensor
which in addition is non-degenerate. The latter means n 6= 0, or equivalently that for
every X V , there is a Y V with (X, Y ) 6= 0. There is a unique linear transformation
S : V V such that for all X, Y V , g(X, SY ) = (X, Y ). This mapping is invertible
because is non-degenerate, and is skew-symmetric with respect to g because is skewsymmetric. The transformation S 2 is (symmetric and) positive-definite with respect to g,
61

hence has a square root T which commutes with S. Thus J := ST 1 is an almost-complex

structure on V .
When (V, J) is an almost-complex vector space, the relation (7.3) generally gives a correspondance between real, symmetric, J-invariant 2-tensors and real, skew-symmetric (1, 1)tensors (in the sense of complex bigrading).
This algebraic construction may be made pointwise on an almost-complex manifold; the
resulting structure is called an almost-Hermitian structure. However, the interaction between
g and J is richer if at least one of the tensor fields is required to satisfy an integrability
condition.
If g is integrable in the sense that the two-form is closed, then the pair (M, ) is a
symplectic manifold. Symplectic manifolds were originally studied by virtue of their appearance in Hamiltonian mechanics; the cotangent bundle of a configuration spacethe so-called
phase space of the physical systemcarries a natural symplectic form, and the time evolution of the system is elegantly described in terms of geometry of the symplectic structure.
Interest in symplectic manifolds was rekindled in the mid-1980s, in relation to topological
applications of Yang-Mills gauge theory, and quantum cohomology.
If J is integrable in the sense of having vanishing Nijenhuis tensor, then M has a complexanalytic atlas for which J is the induced almost-complex structure on T M. A triple (M, J, g)
constitutes an Hermitian manifold if the Riemannian metric g is Hermitian as an inner
product in each tangent space. The two-form is the fundamental form of g, and has the
following useful interpretation.
Lemma 7.2 Let (M, J, g) be an Hermitian manifold of complex dimension n. Then the
volume form dvolg of g is ( n /n!).
1 + + en e
n and take the
proof Write in terms of a unitary coframe = e1 e
top exterior power.
An Hermitian metric is Kahler if the fundamental two-form is closed. In this case, is
ahler
called the Kahler form of g, and the de Rham class = [] Hd2 (M, R) is called the K
class of g. When a complex structure is fixed, it is convenient to identify g and , and to
speak of a Kahler metric or of a metric g in . A complex manifold which admits a
Kahler metric is Kahlerian or of Kahler type.
A two-form is positive if (JX, X) > 0 for every non-zero tangent vector X, cf. equation
(7.3) above. A de Rham cohomology class containing a positive representative is a positive
class. The fundamental two-form of an Hermitian metric is positive, and it is common to
refer to a positive (1, 1)-class on a compact complex manifold as a Kahler class.
Every complex manifold admits an Hermitian structure; indeed, an arbitrary Riemannian
metric may be averaged as in (7.1). By contrast, there are topological obstructions to
existence of a Kahler metric on a compact complex manifold. The most elementary of these
is a simple consequence of Lemma 7.2. Since the volume of M with respect to g may be
62

computed as

1
n
= hn , [M]i = he , [M]i,
n!
n!

the de Rham class is non-zero, so a fortiori is not zero, nor are its k-fold exterior
powers for k n. There are additional, much more subtle, necessary conditions for a
compact complex manifold to be Kahlerian. Some of these will appear later as consequences
of the Hodge Theorem.
Example 7.3 Every Hermitian metric on a Riemann surface is Kahler, simply because
d is a 3-form, hence vanishes identically. The Hopf manifolds, which are diffeomorphic to
S 1 S 2n1 , n > 1, are not Kahlerian. Nor is the sphere S 6 , leaving aside the question of
whether or not S 6 has a holomorphic atlas.
2
Example 7.4 The flat metric on Cn is Kahler. The Kahler form

= 1(dz 1 d
z 1 + + dz n d
zn )
is, indeed, exact. Since the flat metric is invariant under translation, every compact complex
torus admits a Kahler metric.
2
Example 7.5 Complex projective space Pn admits a U(n + 1)-invariant Kahler metric, the
Fubini-Study metric. Let Z be standard coordinates on Cn+1 , put = kZk2 , and set



(7.4)
2e
=
1 log =
1

2
P
P
P



kZk2 dZ j dZ j ( Z j dZ j ) ( Z j dZ j )
=
1
.
kZk4
The form
e is clearly U(n + 1)-invariant (a function of kZk2 alone), and is equally clearly
invariant under scalar multiplication by non-zero complex numbers (numerator and denominator are homogeneous of weight four). There is consequently a well-defined push-forward
2 on Pn . To see that is positive-definite, evaluate at [1 : 0 : : 0] where definiteness is obvious, then use invariance under the unitary group U(n + 1). The de Rham
class of generates the integral cohomology of Pn . To see this, consider the embedding
i : [Z 0 : Z 1 ] P1 7 [Z 0 : Z 1 : 0 : : 0] Pn . In the chart U0 with coordinate z = Z 1 /Z 0 ,

Z
1 dz d
z

i=
, so
=1
2 (1 + z
z )2
P1
by integrating in polar coordinates.
2
A largeand extremely importantclass of Kahler manifolds comes from the following
observation.
63

Proposition 7.6 Let (M, J, g) be a Kahler manifold, and let N be a complex submanifold.
Then the restriction of g to N is Kahler.
proof By hypothesis, the restriction of J to N preserves the tangent bundle of N and is
equal to the almost-complex structure of N. The fundamental two-form of g|N is therefore,
by equation (7.3), equal to |N , which is closed.
Example 7.7 Every smooth algebraic variety admits a Kahler metric. In fact, the restriction
of the Fubini-Study form to a subvariety V Pn is an integral form, that is, represents a
class in H 2(V, Z).
A compact Kahler manifold whose Kahler class is rational is called a Hodge manifold.
The Kodaira Embedding Theorem (Theorem 10.10 and Corollary 10.11 below) asserts that
every Hodge manifold may be embedded in some projective space. Among compact complex
tori, Hodge manifolds form a set of measure zero; such a torus admits a Hodge metric if and
only if its lattice satisfies certain rationality conditions. These conditions are automatic in
dimension one, but are non-trivial in dimension greater than one.
A similar assertion is true for smooth complex surfaces which are diffeomorphic to the
Fermat quartic in P3 , the so-called K3 surfaces (after Klein, Kummer, and Kodaira). A
generic K3 surface is not projective algebraic, but there is a 19-dimensional family of quartic
surfaces in P3 . See also Example 10.8.
2
There are a number of useful alternate characterizations of when a metric is Kahler, some
of which are given in Proposition 8.11 below. As a general point of philosophy, Kahler manifolds exhibit a large degree of interplay between their real structures (differential-geometric,
smooth topological) and complex-analytic structure (Dolbeault cohomology).

7.2

Connections in Vector Bundles

Let E M be a complex vector bundle of rank k over a complex manifold. A connection

in E is a C-linear map D : A0 (E) A1 (E) satisfying the Leibniz rule
(7.5)

D(f s) = df s + f Ds

for all smooth functions f and all smooth sections s. Every connection admits unique
extensions D : Ar (E) Ar+1 (E) satisfying D(s) = d s + (1)r Ds for Ar .
A connection in E gives a means of transporting frames along paths in M, thereby
comparing or connecting fibres of E. If : [0, 1] M is a smooth path, and if e0 is a
frame at (0), then there is a unique section et of E, the parallel transport of e0 along ,
satisfying
D(t)
for all t [0, 1].
et = 0

Indeed, this condition is a linear first-order system of ordinary differential equations on [0, 1],
which therefore has a unique solution for every choice of initial conditions.
64

Locally, a connection is specified by a connection matrix. Precisely, to a local frame e

of E is associated a k k matrixP
= (e) of one-forms satisfying De = e; this equation
is interpreted as meaning Dei = j ij ej . If e0 = ae is another local frame, with a a local
GL(k, C)-valued function, and if De0 = 0 e0 , then the connection matrices are related by
0 a = a + da,

(7.6)

or 0 = aa1 + (da)a1 .

The space of connections in E is an affine space, and every choice of connection D furnishes an isomorphism with the vector space A1 (End E); indeed, if D1 and D2 are connections
in E, then their difference D1 D2 is linear over A0 by the Leibniz rule, hence is given by
wedging with the endomorphism-valued one-form 1 2 A1 (End E).
The curvature operator R = D 2 : A0 (E) A2 (E) measures the extent to which parallel
transport around a closed loop is not the identity. Remarkably, R is an algebraic operator
in the sense that R(f s) = f R(s) for all smooth f ; the value of Rs at x M depends only
on the value of s at x. With respect to a local frame, there is a k k matrix = (e) of
two-formsthe curvature matrix of Dsuch that R(e) = e. The following facts are easily
checked.
Under a change of frame e0 = ae, the curvature matrix transforms by a similarity:
0 = aa1 . In other words, the curvature operator is determined by an End E-valued
two-form A2 (End E). If E is a line bundle, then is an ordinary two-form on
M since in this case End E is canonically trivial, the identity endomorphism being a
natural choice of section.
With respect to an arbitrary frame, the Cartan Structure equation
(7.7)

= d

holds, the wedge product being taken as a matrix product with the entries wedged.
Taking the exterior derivative of and using (7.7) gives the differential Bianchi identity
DR = 0, or
(7.8)

d + = 0.

In words, the curvature tensor is parallel.

If E is a holomorphic vector bundle, then a connection D in E is compatible with the
holomorphic structure if D 0,1 = E . The curvature of a compatible connection has no (0, 2)component. Conversely, a connection D in a smooth complex vector bundle over a complex
manifold whose (0, 1) part satisfies (D 0,1 )2 = 0 determines a holomorphic structure; a local
section s of E is declared to be holomorphic if D 0,1 s = 0. This point of view is useful
65

for constructing moduli spaces of holomorphic structures on a fixed smooth complex vector
bundle.
An Hermitian structure in a smooth complex vector bundle E is a smooth field of Hermitian inner products in the fibres of E; alternately, an Hermitian structure is a positive-definite

section h A0 (E E ). With respect to a local frame, an Hermitian structure is given by

an Hermitian matrix-valued function H, with Hij = H(ei , ej ) which transforms according
to H 0 = aH
at . An Hermitian structure in T M is exactly an Hermitian metric on M.
A connection is compatible with an Hermitian structure h if


d h(s1 , s2 ) = h(Ds1 , s2 ) + h(s1 , Ds2 ) for all smooth, local sections s1 , s2 .

Geometrically, this means that under parallel transport a unitary frame remains unitary. In
terms of a local frame, the connection and curvature matrices satisfy
t

dH = H + H ,

H + H = 0,

i.e. the matrix H is skew-Hermitian. In particular, with respect to a local unitary frame
of E, the curvature matrix is skew-Hermitian.
Proposition 7.8 Let (E, h) M be an Hermitian holomorphic vector bundle. Then there
is a unique connection D in E which is compatible with the metric and holomorphic structure.
proof With respect to a local holomorphic frame e, the connection form is of type (1, 0).
Comparing types in the equation
= H + H
d = +
gives = (H)H 1 . The curvature matrix is
1
1
= ( H)H

=
+ (H)H 1 (H)H
.

It is straightforward to check that the locally defined matrix of one-forms defines a connection matrix (i.e. transforms correctly) which is compatible with h, and is clearly the only
such connection matrix of type (1, 0).
The connection whose existence is asserted in Proposition 7.8 is called the Chern connection or the canonical connection of (E, h). The connection form is formally log H, the
formality being that H is matrix-valued. If E is a line bundle, then this equation is literally
true. The curvature form is related to the Chern form 1 (E, h) by

21(E, h) = 1 log H = 1.
The Chern form represents the first Chern class of the line bundle E.
66

Example 7.9 The tautological bundle OPn (1) is a subbundle of Pn Cn+1 , and has an
induced Hermitian structure given by the usual Hermitian structure on Cn+1 . The curvature
form is exactly minus the Fubini-Study form, see Example 7.5 and Exercise 7.1.
2
A connection D in E induces connections in bundles associated to E. The conjugate
bundle, whose transition functions are complex conjugates of the transition functions of E,
has induced connection D
s = Ds for s A0 (E). The dual bundle acquires a connection via
the Leibniz rule. If the dual pairing is denoted by h , i : A0 (E) A0 (E ) A0 , then for a
section A0 (E), D is defined by requiring
for every section s A0 (E).

dhs, i = hDs, i + hs, Di

If DE and DF are connections in vector bundles E and F over M, then DE DF is a

k
connection in E F . An adapted frame is a collection of local sections {ei }k+`
i=1 with {ei }i=1 a
frame for E and {ei }`i=k+1 a frame for F . With respect to an adapted frame, the connection
and curvature forms of DEF split as




E 0
E 0
.
,
EF =
(7.9)
EF =
0 F
0 F
The tensor
product E F has an induced connection DE IF + IE DF . The exterior
V
products p (E) acquire connections
as subbundles of E p . An important special case is the
Vk
top exterior power det E =
E; the connection and curvature forms with respect to the
frame e1 ek are
(7.10)

tr ,

tr .

Let S E be a holomorphic subbundle of an Hermitian vector bundle E M, and let

Q = E/S be the quotient. There is a short exact sequence
0SEQ0
of vector bundles. Remarkably, such a sequence does not usually split holomorphically. Of
course, the orthogonal complement S E is isomorphic to Q as a smooth complex vector
bundle, and via this isomorphism the bundle Q acquires an Hermitian structure and a canonical connection. In Riemannian geometry, the Levi-Civita connection of a submanifold is the
tangential component of the Levi-Civita connection of the ambient space. The holomorphic
analogue of this result is as follows.
Theorem 7.10 Let (E, h) be an Hermitian holomorphic vector bundle, S a subbundle with
the restricted metric, and Q = E/S the quotient with the induced metric. There is a oneform A with values in Hom(S, S ) such that the connection matrix and curvature operator
of E are




S A
RS A A
D 1,0 A
(7.11)
E =
RE =
.
A Q
D 0,1 A
RQ A A
67

proof There are several assertions to check, but all are straightforward. First, apply DE
to a local section s of S and decompose into S and S components to deduce that the S
component of DE |S is DS and the S component is A0 -linear in s, i.e. is a one-form A with
values in Hom(S, S ). This one-form is called the second fundamental form of S in E, and
measures extrinsic geometry of S via its embedding.
A similar argument shows that DE |S has S somponent A (minus the adjoint) and
S component DQ . The expression of the curvature operator in (7.11) is the vector bundle
version of the Gauss-Codazzi equations.
Corollary 7.11 Let 0 S E Q 0 be a short exact sequence of Hermitian holomorphic vector bundles over M. Then the sequence splits holomorphically if and only if the
second fundamental form of S in E vanishes identically.
Let (E, h) (M, g) be an Hermitian holomorphic vector bundle over a Kahler manifold.
The curvature is a two-form with values in End(E). In a local coordinate system z, with
local frame ej and dual coframe ei , the curvature is

k
k
n
X
X
X
j

=
ji ei ej =
Ri
dz d
z ei ej .

i,j=1

i,j=1

,=1

Letting (g ) be the inverse matrix of (g), the mean curvature of h is the endomorphism

k
n
X
X

j i

tr =
g Ri
e ej .

i,j=1

,=1

The bundle (E, h) is Einstein-Hermitian if the mean curvature is a constant multiple of the
identity endomorphism.
When E = T M, the trace of the curvature may be regarded (after lowering an index) as
a symmetric 2-tensor, called the Ricci tensor of g. A Riemannian metric is Einstein if the
Ricci tensor is a multiple of the metric tensor. A metric which is both Einstein and Kahler
is an Einstein-Kahler metric (or a Kahler-Einstein metric to a Riemannian geometer). An
Einstein-Kahler metric may be regarded as a (very special) Einstein-Hermitian metric on
T M; explicitly, in searching for an Einstein-Hermitian metric, a Kahler metric g on M is
fixed, and a metric (E, h) is sought. Even if E = T M, h and g are generally different. The
existence problem for Einstein-Kahler metrics is substantially more difficult than the corresponding question for Einstein-Hermitian metrics. Finally, an Einstein-Hermitian metric
should not be confused with an Hermitian metric g on M which happens to be Einstein in
addition.
68

Exercises
Hermitian Structures on Line Bundles over P1
Recall that the total space H k of the line bundle O(k) P1 is obtained from two copies
of C C, with coordinates (z 0 , 0) and (z 1 , 1 ), by identifying
1
z = 1,
z
0

1
= 1 k.
(z )
0

Thus H k is a family of lines parametrized by points of P1 , the family is trivial in each

coordinate system separately, and the fibres in the z 1 coordinates rotate k times with respect
to the z 0 coordinate system.
An Hermitian structure is a smooth assignment of an Hermitian inner product to each
fibre of O(k). In the present context, an Hermitian structure is specified by giving two
positive, real-valued functions h0 (z 0 , z0 ) and h1 (z 1 , z1 ), such that
()

h0 (z 0 , z0 ) = (z 1 z1 )k h1 (z 1 , z1 )

on the set C = P1 \ {0, } where z 0 and z 1 are non-zero. By condition (), the function
hi (z i , zi ) ii is well-defined; it is called the norm (squared) function of the Hermitian line
bundle and is denoted r = k k2 : H k [0, ).
Exercise 7.1 Show that the functions hi (z i , zi ) = (1+z i zi )k , i = 0, 1, define an Hermitian
structure on O(k). Recall that the total space of O(1) is a submanifold of P1 C2 ; let
(w 0 , w 1) be coordinates on C2 . Show that the Hermitian structure hi (z i , zi ) = (1 + z i zi )
has norm function equal to w 0 w 0 + w 1 w 1 . In other words, the Hermitian structure on the
tautological bundle is induced from the usual Hermitian structure on P1 C2 .


Exercise 7.2 Let be a local

section of H k which does not vanish on U P1 .
holomorphic
Show that the two-form 1 log kk2 is invariant
gauge transforma under holomorphic
i i

tions. In particular, the curvature form 1 (h) = 1 log hi (z , z ) is a globally defined

two-form on P1 . Prove the Gauss-Bonnet Theorem: With the previous notation,
Z
()
1 (h) = 2k.
P1

As an elementary sub-exercise, verify equation () for the standard Hermitian structure

given above.


Exercise 7.3 Write stereographic projection in local coordinates, find the Hermitian structure on T P1 gotten by restricting the Euclidean metric du2 + dv 2 + dw 2 on R3 , and use the
Gauss-Bonnet Theorem to show (yet again) that T P1 = O(2).

Flat Vector Bundles and Holonomy Representations
69

Exercise 7.4 Let p : (E, h) (M, g) be a holomorphic Hermitian vector bundle of rank k
over an Hermitian manifold. Prove that there is an exact sequence (generally non-split)
(7.12)

0 p E T E p T M 0

of holomorphic vector bundles over E. In particular, a non-zero section of E (as a vector

bundle over M) gives rise to a non-zero vector field on E (as a complex manifold).


Exercise 7.5 Let = h1 h A1 (T E) be the Chern connection form of the Hermitian

structure h. Prove that there is a unique Hermitian metric on E (regarded as a complex
manifold) such that
i. The Hermitian structure of E is induced by the inclusion p E T E.
ii. The kernel of , which is smoothly isomorphic to p T M by (), is orthogonal to p E
and acquires the Hermitian structure p g on p T M.

Intuitively, this Hermitian metric on the total space of E is h in the vertical directions and
g in the horizontal directions.


Exercise 7.6 With the notation of Exercise 7.4, let H T E denote the kernel of , i.e. H is
the bundle of horizontal vectors in T E; as smooth vector bundles, T E = p E H. Show that
the curvature = d vanishes identically if and only if H is involutive, if and only
if the sequence (7.12) splits holomorphically. Intuitively, curvature of a connection may be
regarded as an obstruction to involutivity of the horizontal distribution.

Exercise 7.7 (Holonomy representations) Let p : (E, h) M be an Hermitian holomorphic
vector bundle over a connected complex manifold, and assume the Chern connection of E
has vanishing curvature. Let : [0, 1] M be a loop based at x M, and let () Aut Ex
denote parallel transport around . Prove that () depends only on the homotopy class
of in 1 (M). It may be helpful to consider the leaves of the distribution H; the lift of
lies in a fixed leaf. Fixing a frame at x identifies Aut Ex and GL(k, C), and the map
: 1 (M) GL(k, C) defined by lifting loops based at x is a group homomorphism, called
the holonomy representation of 1 (M). (In particular, each leaf is the total space of a
principal bundle over M with structure group 1 (M).)
f denote the universal cover of M. Prove that E = M
f Ck . In summary, flat
Let M
vector bundles of rank k over M correspond (more or less naturally) to conjugacy classes of
representations : 1 (M) GL(k, C), or to reductions of the structure group of E from
GL(k, C) to 1 (M).


Exercise 7.8 Prove that if M is simply-connected, then a flat vector bundle is holomorphically trivial. (This is essentially trivial from Exercise 7.7.) Prove that every topologically
trivial line bundle over Pn (or P1 , though the general case is not much more difficult) admits
a flat Hermitian structure. Let x and y be distinct points in an elliptic curve C/. Prove
70

that the line bundle associated to the divisor (x) (y) is topologically trivial but does not
admit a flat Hermitian structure.


Hodge Theory and Applications

Suppose : E M is a holomorphic vector bundle over a compact complex manifold.

There is a Dolbeault cohomology theory of -closed

(p, q)-forms with values in E, and an
isomorphism of complex vector spaces


q
p
Hp,q
(M, E) ' H M, (E) .

Define the Hodge numbers of M with coefficients in E by hp,q (M, E) = dimC Hp,q
(M, E).
Because Dolbeault cohomology is an invariant of the complex structure of M (and not
merely of the topology of M), it is not a priori obvious whether or not hp,q (M, E) is finite.
By way of example, if M = C2 \ {0}which has the homotopy type of the three-sphere,
then h1,0 (M) = h1 (M, O) = .
Suppose that in the above situation both E and M are endowed with metric structures.
More precisely, assume (E, h) is an Hermitian holomorphic vector bundle (i.e. h is a metric
in the fibres of E) and that (M, g) is an Hermitian manifold (i.e. g is a fibre metric in T M).
It is then possible to introduce a norm on the space of (p, q)-forms with values in E. The
Hodge Theorem (Theorem 8.3 below) asserts that each Dolbeault cohomology class contains
a unique representative of smallest norm, and that the set of such representatives is indeed
finite-dimensional when M is compact. In the event that (M, g) is a Kahler manifold and
E = O (the trivial line bundle), the interplay between real and complex potential theory gives
extra information about the de Rham cohomology of M versus the Dolbeault cohomology,
see Theorem 8.17.

8.1

The Hodge Theorem

Let : (E, h) (M, g) be an Hermitian holomorphic vector bundle of rank k over a

(connected) compact Hermitian manifold of dimension n. The metric g naturally induces

q
an Hermitian structure
Vp,q in each of the bundles T1,0 (M) T0,1 (M) of (p, q)-tensors, and
in the subbundle
(M) of (p, q)-forms. Together with the Hermitian structure of E,
these Hermitian structures determine an Hermitian structure h , i in the bundle of E-valued
(p, q)-forms.
It is probably instructive to write these Hermitian structures in terms of local frames.
Let { , J } be a local orthonormal coframe for g, i.e. the one-forms are real-valued,
and
n 

X
g=
+ J J .
=1

71

Setting = + i J T1,0
M, the sesquilinear extension of g is g + 1 = ,
and k k2 = 2. If I and J are multi-indices of length p and q, then the (p, q)-forms
I J
V
p,q
have (squared) length 2p+q , and the set of such forms is an orthogonal frame for
(M).
Concretely, if {ei }ki=1 is a local frame for E, and
X 
X 


i
I
J
I J ei ,
Ij JI J ej ,
=
=
|I|=p,|J |=q
1ik

|I|=p,|J |=q
1j k

then the pointwise inner product h, i A0 (M) is equal to

X
p+q
iI J I J h(ei , ej ),
h, i = 2
|I|=p,|J |=q
1 i, j k

and the global inner product is defined to be

Z
Z
n
(8.1)
(, ) =
h, i dvolg =
h, i .
n!
M
M
The global inner product is an Hermitian inner product on Ap,q (E), the space of smooth
(p, q)-forms with values in E, but of course Ap,q (E) is not complete. It will presently be
convenient to introduce a family of Sobolev completions of Ap,q (E) which will include the
L2 -completion as a special case.
Let : Ap,q (E) Ap,q1(E) be a formal adjoint of with respect to the global inner
for all Ap,q1(E), Ap,q (E). The desired smallest
product; thus ( , ) = (, )
norm representative of a Dolbeault class is formally expressed very simply.

Proposition 8.1 Let be a -closed

(p, q)-form. Then has smallest L2 -norm among all
if and only if = 0.
forms +
proof

Suppose kk2 is minimal. Then for every Ap,q1(E),


d 
2 = 2 Re(, )

0 =
 k + tk
dt 
t=0
d 
2 = 2 Im(, ).

0 =
 k + itk
dt 
t=0

= 0 for all Ap,q1(E), so = 0.

Thus ( , ) = (, )
Conversely, if = 0, then for every Ap,q1 (E),

2 = kk2 + kk
2 + 2 Re(, )

k + k
2 + 2 Re( , ) = kk2 + kk
2 kk2
= kk2 + kk
72

 = 0.
with equality if and only if
Formally, Proposition 8.1 says the Dolbeault space Hp,q
(M, E) is isomorphic to the so
lution space of = 0 where ranges over the space of -closed
(p, q)-forms, or in other
words that
p,q

Hp,q
(M, E) ' (ker ) (ker ) A (E).
There is a convenient reformulation of the latter in terms of a single second-order diffential
p,q

operator, the -Laplacian

2 = 2 := + .
(ker ).
Lemma 8.2 As operators on Ap,q (E), ker 2 = (ker )
2 + k k2 .
proof This is an immediate consequence of (2, ) = kk

is the space of (-)harmonic

(p, q)-forms, and is denoted Hp,q (M, E).
The kernel of 2p,q

The following result, which justifies the formal arguments made above, is the Hodge Theorem.
Theorem 8.3 Let : (E, h) (M, g) be an Hermitian holomorphic vector bundle over a
compact Hermitian manifold. Then the space of harmonic (p, q)-forms is finite-dimensional
p,q
and is isomorphic to the Dolbeault space Hp,q
: Ap,q (E) Hp,q (M, E)
(M, E). Let H
denote orthogonal projection. The Laplacian 2p,q
is invertible on the orthogonal complement

of the space of harmonic forms, that is, there exists a unique operator Gp,q of degree 2 such
that
(8.2)

=I
Hp,q + Gp,q 2p,q

= 0, and [Gp,q , ] = 0.
on Ap,q (M, E), Hp,q Gp,q = 0, [Gp,q , ]
The quasi-inverse G of 2 is called the Greens operator. Equation (8.2) is sometimes
expressed by saying that every (p, q)-form has a unique decomposition
(8.3)

G) + (G)

= H + (

as the sum of a harmonic form, an exact form, and a co-exact form, or that the equation
2 = has a solution if and only if is orthogonal to the space of harmonic (p, q)-forms,
and in this case = G.
proof The technical preliminaries for the proof of the Hodge Theorem are construction of
a formal adjoint and of appropriate completions of Ap,q (E), so that techniques of Hilbert
space analysis may be brought to bear. Once the groundwork is laid, a weak solution is
constructed (functional analysis), and is shown to be of class C (regularity theory). These
steps are sketched below, mostly without proof.
The Hodge star operator : Ap,q (E) Anp,nq (E ) is defined by the requirement that
(8.4)

for all , Ap,q (E).

h, i dvolg =
73

In order to write the star operator in terms of a frame, let I 0 , J 0 {1, . . . , n} be multiindices complementary to I and J, and let IJ = sign(I I 0 )sign(J J 0 ), where sign denotes
the sign of a permutation of {1, . . . , n}. If {ei } denotes the dual coframe of {ei }, then

X 
X 


0
0
i
I
J
p+qn
IJ I JI J ei .
I J ei = 2
=
|I|=p,|J |=q
1ik

|I|=p,|J |=q
1ik

It is easy to verify that = (1)p+q acting on Ap,q (E). It is sometimes desirable to introduce
a complex-linear star operator, defined as above but without taking complex conjugates of
the components of ; in this event, the star operator as defined above is denoted .

Proposition 8.4 The formal adjoint of is = .

proof Let Ap,q (E) and Ap,q1 (E). By the definitions of the global inner product
(8.1) and the Hodge star operator (8.4),
Z
Z
Z
p+q

).

= (1)
(
(, ) =
() +
M

) = d( ), so the last term vanishes by

Because is of type (n, n 1), (
Stokes Theorem. The remaining integral is
Z
),

( ) = (,
M

as claimed.
proving that =
Cover M by coordinate charts which are also trivializing neighborhoods for E. Each
chart is modelled on U Ck , with U Cn ' R2n an open set, and M is covered by finitely
many charts. (It is here that compactness of M is used in an essential way.) To define
suitable norms on Ap,q (E), first define norms on smooth, compactly supported sections of
R2n Ck , then use a partition of unity to express an element of Ap,q (E) as a finite sum of
smooth, compactly supported sections in coordinate charts.
It is sufficient to consider smooth, compactly supported functions f : R2n C; the
norm of a vector-valued function may then be computed with respect to a fixed norm on
Ck . Recall that the Fourier transform of f is defined to be
Z
1
b
f (x)eixy dx.
f (y) =
(2)2n R2n
For each real number s, define the Sobolev s-norm of f to be
Z

s
2
b 2 1 + |y|2 dy,
(8.5)
kf ks =
|f(y)|
R2n

74

and define the Sobolev space Hs (R2n ) = Hs to be the completion in the s-norm
of the space of
Vp,q
p,q
2n
smooth, compactly supported functions on R . Let Hs (M, E) = Hs (M,
E) denote
p,q
the completion of A (E) with respect to the Sobolev s-norm as described above. While
the norm depends on the choice ofVtrivializing cover and subordinate partition of unity, it is
easy to see that the space Hs (M, p,q E) and its topology do not depend on these choices;
as above, this is a consequence of compactness of M. There is an obvious nesting, namely if
r < s then Hr Hs . Set
\
[
H =
Hs ,
H =
Hs .
sR

sR

Intuitively, the rate at which fb(y) decays as |y| is related (proportionally) to the
smoothness of f . As s , the term (1 + |y|2)s forces fb to decay rapidly as |y| . More
precisely, a measurable function has finite Sobolev s-norm if and only if the derivatives of
order up to and including s are in L2 (R2n ); in particular, C s Hs :
Proposition 8.5 For s 0, the Sobolev s-norm is equivalent to the norm
X Z
|I|s

1/2
|D f (x)| dx
,
I

R2n

and Hs is the topological dual of Hs .

The norm appearing in Proposition 8.5 comes from an inner product, so the Sobolev
space Hs is a Hilbert space. The is following (partial) reverse inclusion is called the Sobolev
Lemma.
Theorem 8.6 Hs+n+1 (R2n ) C s (R2n ). Globally, Hs+n+1(M,

Vp,q

E) C s (M,

Vp,q

E).

N.B. In Rm , the n in Theorem 8.6 becomes [m/2]. The dimension n enters via finiteness
of the integral
Z 
s
1 + |y|2 dy.
R2n

As a consequence of Theorem 8.6, H (M, E) = A0 (E). The next result, called the Rellich
Lemma, is important for technical reasons; it is used below to extract a compact, self-adjoint
operator on a fixed Hilbert space. The Hodge Theorem is then essentially the Spectral
Theorem.

Theorem 8.7 If s > r, then the

r of Hilbert spaces is a compact embedding.
V inclusion Hs HV
Globally, the inclusion Hs (M, p,q E) Hr (M, p,q E) is compact.
75

The Sobolev norms described above may be given more globally in terms of connections.
Specifically, there are natural connections on E and T M (the respective Chern connections),
and the Sobolev norms may be computed from covariant derivatives of tensor fields rather
than by using a partition of unity. As above, Sobolev norms defined in this way are equivalent
to Sobolev norms defined using a partition of unity because M is compact.
Define the Dirichlet inner product on Ap,q (E) by
2 + k k2 .
D(, ) = (, + 2 );
D(, ) = kk2 + kk

By expressing 2 in local coordinates as the Euclidean Laplacian plus terms of lower degree
(using a so-called Weitzenbock formula), it is possible to deduce an a priori estimate for the
Dirichlet norm, called Gardings Inequality.
Theorem 8.8 For each (p, q), there exists a C > 0, depending only on (M, g) and (E, h),
such that
kk21 C D(, )
for all Ap,q (E).
From Gardings Inequality, it follows that the operator I + 2 on Ap,q (E) has a bounded
weak inverse
p,q
T : Hp,q
0 (M, E) H1 (M, E).

By the Rellich Lemma, the induced operator on Hp,q

0 (M, E) (T followed by inclusion) is
compact, and is clearly self-adjoint. By the Spectral Theorm, there is a decomposition of
Hp,q
0 (M, E) into finite-dimensional eigenspaces of T . Thus I + 2 is weakly invertible, hence
p,q
2 is weakly invertible on (ker 2) . Explicitly, given a form H0 (M, E), there is a form
p,q
Hp,q
0 (M, E) such that for all A (E),
(, 2) = (, ).
It remains to show that eigenforms of 2, which are eigenforms of I + 2, are smooth.
This follows from the following regularity result; intuitively, the inverse of the Laplacian
adds two derivatives.
p,q
Theorem 8.9 Suppose Hp,q
0 (M, E) is a weak solution of 2 = Hs (M, E). Then
p,q
Hs+2 (M, E) .

Smoothness of eigenforms of 2 follows immediately by a bootstrap argument, for if

p,q
2 = for some R, then by Theorem 8.9, is in H2 (M, E). Again by Theorem 8.9
p,q
p,q
is in Hp,q
4 (M, E). Inductively, H (M, E) = A (E). This completes the proof of the
Hodge Theorem.
As above, let : (E, h) (M, g) be an Hermitian holomorphic vector bundle over a
compact Hermitian manifold. If 6= 0 is a harmonic (p, q)-form with values in E, then is
a harmonic (n p, n q)-form with values in E , and the global inner product of these forms
is kk2 > 0. Since these forms are natural representatives of their Dolbeault cohomology
classes, the following duality, the Kodaira-Serre Duality Theorem, follows immediately.
76

Theorem 8.10 With the above notation, the pairing

H q M, p (E) H nq M, np (E ) H n (M, n ) ' C

is non-degenerate.

The top exterior power of the holomorphic cotangent bundle of M is called the canonical
bundle KM ; the sheaf of germs of sections is n , the sheaf of germs of holomorphic n-forms.
The Kodaira-Serre Duality Theorem for p = 0 becomes
dual

H q (M, E) ' H nq (M, KM E ).

8.2

The Hodge Decomposition Theorem

On a compact Kahler manifold (M, J, g), there is compatibility between the Dolbeault decomposition and Hodge theory for the de Rham complex. Consequently, the cohomology of
a compact Kahler manifold satisfies several non-trivial restrictions, see Theorem 8.17 and its
corollaries.
Let (M, J, g) be an Hermitian manifold. Recall that there is a two-form defined by
(X, Y ) = g(X, JY ) for tangent vectors X and Y . Since (M, g) is in particular a Riemannian
manifold, there is a unique Levi-Civita connection which is symmetric (or torsion-free)
and compatible with g; respectively, these conditions mean that for all vector fields X, Y ,
and Z on M,


(8.6)
X Y Y X [X, Y ] = 0,
Z g(X, Y ) = g(Z X, Y ) + g(X, Z Y ).
The former is sometimes expressed by saying that the torsion tensor T (X, Y ) vanishes,
while the latter is expressed succinctly as g = 0. On the other hand, T M is a holomorphic,
Hermitian vector bundle, so there is a unique Chern connection D which is compatible with
or (equivalently)
the metric and which is also almost-complex, namely for which D 0,1 = ,
for which DJ = 0. On a general Hermitian manifold these connections are unrelated. The
Kahler condition is equivalent to equality of these connections.
At each point p of a Riemannian manifold, there exist geodesic normal coordinates, in
which the metric satisfies gij (p) = ij and the connection satisfies /xi |p = /xi . On an
Hermitian manifold, it is generally impossible to find local holomorphic coordinates with this
property. The Kahler condition is equivalent to existence of a local holomorphic coordinate
system in which the metric approximates the Euclidean metric to second order. Each of these
properties has an expression in terms of components of tensor fields, Christoffel symbols, and
their derivatives, but for present purposes the following conditions suffice.
Proposition 8.11 Let (M, J, g) be an Hermitian manifold. The following are equivalent.
77

i. The metric is Kahler, i.e. d = 0.

ii. The Levi-Civita connection is almost-complex.
iii. The Chern connection is torsion-free.
iv. For each p M, there exist local holomorphic coordinates centered at p such that
gij = ij + O(|z|2 ).
v. For each p M, there is a neighborhood U of p and a smooth, real-valued function
on U.
f : U R, such that = 1 f
The function f appearing in property v. is called a Kahler potential (function). Each of
the properties above is essentially local, so there is no need to assume M is compact. By
contrast, the most striking global results about Kahler manifolds depend on existence of
adjoints to various operators, and therefore require compactness.
proof Each of ii. and iii. is equivalent to equality of the Levi-Civita and Chern connections,
so these conditions are equivalent. Closedness of may be expressed in terms of derivatives
of components of g as either of the conditions
()

gi
gk
=
k
z
z i

g
gi
= ijk .
k
z
z

The connection form = g 1g of the Chern connection, and the torsion tensor T , are
expressed in a coordinate coframe as


n
n
X
X

gj k
`
j
i
ik
ik
dz dz
,
j =
g gj k ,
T = dz =
g
`
i
z
z
k=1
i,j,k,`=1
from which equivalence of i. and iii. follows.
Condition iv. implies d = 0 at p for each p M. Conversely, if d = 0, that is, if ()
holds locally, then it is straightforward to check that the change of coordinates
n
1 X gj k i j
ww
w =z
2 i,j=1 z i
k

kills the linear terms in the Taylor expansion of gi about p.

Condition v. implies condition i. immediately. Conversely, assume d = 0 in a simplyconnected coordinate neighborhood U. By the Poincare lemma, there is a real one-form
with d = on U. The (1, 0) and (0, 1)-components of are conjugate since is real, and
0,1 = 0, and =
1,0 + 0,1 . By the Dolbeaultsince is of type (1, 1), 1,0 = 0,
Grothendieck lemma, there exists a function on U with = 1,0 and = 0,1 . The
function f = 2 Im = i( )
is the desired potential function.
78

On a compact Kahlerian manifold, non-trivial holomorphic forms are d-closed and nonexact (Proposition 8.13), and holomorphic forms are harmonic with respect to an arbitrary Kahler metric (Proposition 8.14). Each of these is a consequence of a simple typedecomposition argument.
Lemma 8.12 Let (M, J, g) be a compact Kahler manifold. If Zp,0
is an exact holomorphic p-form, then = 0.
Z
proof For every (p, 0)-form ,
np = (, ) = kk2 . If = d is d-exact, then
2

kk =

np



np
=0
d

by Stokes Theorem (both and are d-closed), from which it follows that = 0.
Proposition 8.13 Let (M, J, g) be a compact Kahler manifold. Then there is an injection
H 0 (M, pM ) , Hdp (M, C), that is, holomorphic forms are closed and non-exact.
proof Let 6= 0 be a holomorphic p-form. Then d = is an exact, holomorphic
(p + 1)-form, hence zero by Lemma 8.12.
Proposition 8.14 Let (M, J, g) be a compact Kahler manifold, and let be a holomorphic
p-form. Then is harmonic.
= 0, while is of type (n p, n), so for type
proof If is a holomorphic p-form, then

reasons = = 0.
Let
ahler manifold with Kahler form . Define the operator
(M, J, g) be a compact K
c
c

d = 1( ). Then d and d are real operators, as is

(8.7)
ddc = 2 1 .
Let d = d and (dc ) denote the formal adjoints, and let p,q : A (M) Ap,q (M) denote
the projection operator, so that
M
p,q = I : Ar (M) Ar (M).
p+q=r

Define L : Ap,q (M) Ap+1,q+1 (M) by L() = , and let = L be the adjoint with
respect to the global inner product. On (1, 1)-forms, is the trace with respect to . The
formula [L, d] = 0 is easily proven; if is an r-form, then
[L, d] = L(d) d(L) = d d( ) = 0

since d = 0. Taking adjoints, [, d ] = 0. The following commutator formulas are messy to

establish, even in flat Euclidean space. However, once verified there, they may be deduced
to hold on an arbitrary compact Kahler manifold by virtue of Proposition 8.11 iv.
79

Proposition 8.15 On a compact Kahler manifold, [L, d ] = dc , [, d] = (dc ) , and

M
[L, ] =
(p + q n)p,q .
p,q

It is worth remarking that the type decomposition of the second formula in Proposition 8.15 is

= 1 ,
(8.8)
[, ]
[, ] = 1 .
Define the de Rham Laplacian 2d to be the second-order operator dd + d d. On a Rieman
nian manifold, there is a Hodge theorem analogous to the -Hodge
theorem, to the effect
that every de Rham class contains a unique harmonic representative. Proposition 8.15 has
the following important consequence, which gives the compatibility between the de Rham
and Dolbeault harmonic forms.
Theorem 8.16 On a compact Kahler manifold, [L, 2d ] = 0, [, 2d ] = 0, and
(8.9)

2 = 2 .

2d = 2 + 2 ,

In particular, 2d maps Ap,q (M) to itself.

In words, Theorem 8.16 says that on a compact Kahler manifold, -harmonic

forms are
d-harmonic, and that 2d preserves bidegree. It therefore makes sense to define the de Rham
(p, q)-harmonic space to be the kernel of 2d acting on Ap,q (M), and this is nothing but the

space of -harmonic
(p, q)-forms. The importance is that while Dolbeault cohomologyand
the space of Dolbeault harmonic formsis defined in terms of the complex-analytic structure
of M, the de Rham spaces depend only on the smooth structure of M. But each of these
spaces is isomorphic, via the corresponding Hodge theory, to a space of harmonic forms,
and by Theorem 8.16 these harmonic spaces coincide. This result is known as the Hodge
Decomposition Theorem for compact Kahler manifolds:
Theorem 8.17 Let (M, J, g) be a compact Kahler manifold. Then there are isomorphisms
M p,q
q,p
Hdr (M, C) '
H (M),
Hp,q
(M) 'C H (M).
p+q=r

In particular, hp,q = hq,p = hnp,nq .

Corollary 8.18 Let M be a compact Kahlerian manifold. Then the odd Betti numbers b2k+1
of M are even.

80

proof

By the de Rham theorem, b2k+1 = dimC H 2k+1(M, C), which is equal to

X

p,q

=2

k
X

hp,2k+1p

p=0

p+q=2k+1

since hp,q = hq,p by the Hodge Decomposition Theorem.

n
Corollary 8.19 H q (Pn , p ) = Hp,q
(P ) = C if 0 p = q n, and is 0 otherwise.

proof The de Rham cohomology (with complex coefficients) is C in evenPdimensions 2n

i,2pi
and is zero otherwise. By the Hodge Decomposition Theorem, 1 = b2p = 2p
. Thus
i=0 h
p,p
h = 1 for 0 p n, and all other Hodge numbers vanish.
A very pretty consequence of the Hodge Decomposition Theorem is the so-called Hard
Lefschetz Theorem. The modern proof, due to Chern, is an extremely elegant application
of representation theory for sl(2, C). Let M be a compact Kahler manifold of complex
dimension n, and introduce the weighted projection operator

h : A (M) A (M),

h=

2n
X
r=0

(n r)r .

By Theorem 8.16, each of the operators L, , and h commutes with 2d = 22, so that each
operator acts on the space Hd (M, C) of de Rham harmonic forms. Chern made the following
observation.
Proposition 8.20 [, L] = h, [h, L] = 2L, and [h, ] = 2.
In words, the association


0 1
X=

0 0


0 1
Y =
L
0 0



1
0
H=
h
0 1

defines a representation of the Lie algebra sl(2, C) on the finite-dimensional complex vector
space Hd (M, C). The irreducible submodules give the Lefschetz decomposition of the cohomology of M. A primitive element in an sl(2, C)-module is an eigenvector v of H such that
Xv = 0.
Proposition 8.21 If V is a finite-dimensional sl(2, C)-module, then primitive elements exist. If V is an irreducible (N + 1)-dimensional representation space for sl(2, C) and if v V
is primitive, then V is generated as a vector space by the elements {v, Y v, Y 2 v, . . .}. The
eigenvalues of H are the integers N, N + 2, . . . , N, the -eigenspaces V are onedimensional, and V is the direct sum of the eigenspaces of H. The operators H, X, and Y
act by
H(V ) = V ,
X(V ) = V+2 ,
Y (V ) = V2 ,
with the convention that V = 0 if is not an eigenvalue of H.
81

Irreducible (N + 1)-dimensional representations of sl(2, C) exist and are unique up to

equivalence. In fact, the Nth symmetric power of the standard representation of sl(2, C) on
C2 is an irreducible representation of dimension (N + 1). Generally, the space of primitive
elements P V of an sl(2, C)-module V is ker X, and V is a direct sum
V = PV Y PV Y 2 PV
Let (M, J, g) be a compact Kahler manifold, and define the primitive cohomology to be
P nk (M) = ker(Lk+1 ) = (ker ) Hdnk = -traceless (n k)-forms.
Applying Proposition 8.21 gives the Hard Lefschetz Theorem:
Theorem 8.22 If (M, J, g) is a compact Kahler manifold of complex dimension n, then for
0 k n, the map Lk : Hdnk Hdn+k is an isomorphism, and
M
Hrd (M) =
Lk P r2k (M).
kZ

If M PN (equipped with the restriction of the Fubini-Study metric), then the map L
corresponds via Poincare duality to intersecting with a hyperplane class, and Lk corresponds
to intersecting with a linear subspace of dimension N k. Primitive classes in M correspond
to cycles which do not intersect a hyperplane, and which therefore lie in an affine coordinate
chart. Such cycles cannot be represented by embedded complex submanifolds, of course. If
M P2 is a smooth plane curve, then every real one-dimensional cycle is primitive, and
holomorphic one-forms represent primitive cohomology classes.

Chern Classes

A vector bundle of rank k is a locally trivial family of k-dimensional vector spaces parametrized by points of a manifold M. Characteristic classes quantify global non-triviality of
the family. The prototypical result expresses the first Chern class of a line bundle, defined
to be the image of L under the connecting homomorphism c1 : H 1 (M, O ) H 2 (M, Z), in
terms of curvature, and as the Poincare dual of the divisor of a meromorphic section.
Proposition 9.1 Let (L, h) M be an Hermitian holomorphic line bundle over a compact
complex manifold. Then c1 (L) is represented by the closed (1, 1)-form

1
1
(9.1)
1 (L, h) =
log h.
2
2
If s is a global meromorphic section of L, so that L = [(s)], then c1 (L) is Poincare dual to
(s) = (s)0 (s) .
82

In equation (9.1), h may be interpreted as the norm of a local holomorphic section; it

is easy to see the form 1 (L, h) does not depend on the choice of local section. More is in
fact true: Every smooth form representing 2c1 (L) is the curvature form of an Hermitian
structure conformal to h. The proof of this fact is an easy application of the Hodge theorem.
Chern classes may be introduced in several ways, from points of view ranging from functorial to differential-geometric to obstruction-theoretic. Because Chern classes are topological
objects, there is no reason to work in the holomorphic category, or even to require the base
space M to be a complex manifold. The present treatment is to specify axioms for Chern
classes, then to define certain closed forms using curvature forms of a connection in E, verifying that these forms satisfy the axioms. After this a few other definitions are given whose
equivalence is asserted without proof.
Axioms for Chern Classes Let : E M be a smooth complex vector bundle over a
smooth CW complex. For most purposes, M may be taken to be a smooth compact manifold, though occasionally it will be necessary to consider infinite-dimensional Grassmannian
manifolds.
i. There is an element c(E) = 1 + c1 (E) + c2 (E) + H (M, R), called the total Chern
class of E, with cj (E) H 2j (M, R).


ii. (Naturality) For every smooth map f : N M, c(f E) = f c(E) H (N, R).
L
iii. (Whitney
Sum Formula) If Li , 1 i k are line bundles, then then c( i Li ) =
Q
i c(Li ).
iv. (Normalization) If L P1 is the tautological bundle, and if denotes the positive
generator of H 2(P1 , Z), then c(L) = 1 .

These axioms characterize Chern classes. It may be helpful to remark that if E M is a

vector bundle, then there exists a space N and a map f : N M such that f E N splits
into a sum of line bundles, and such that the map f : H (M, R) H (N, R) is injective.
This is the basis of the so-called Splitting Principle, to the effect that a universal relation
among Chern classes (such as the Whitney Sum Formula) which holds for line bundles holds
for arbitrary vector bundles. Construction of a splitting space N is an inductive variant
on the construction of the tautological bundle E E P(E).
Example 9.2 Let H 2 (Pn , Z) denote the positive generator. There is an exact sequence
of vector bundles 0 C OPn (1) Cn+1 T P n 0 over Pn . From the Splitting
Principle and the Whitney Sum Formula,


n
n+1
n
c(T P ) = c OP (1) C
= (1 + )n+1 H (Pn , Z).
For example, c(T P1 ) = 1+2 and c(T P2 ) = 1+3 +3 2.
83

9.1

Chern Forms of a Connection

Curvature is an infinitesimal measure of the failure of E to be geometrically trivial. Certain

invariant functions of the curvature forms of a connection give closed 2j-forms on M which
quantify the twisting of E. If E is trivialized over an open set U, then it is possible
to choose a connection whose curvature vanishes identically in U. If this neighborhood is
enlarged as much as possible, then the global non-triviality of E will be concentrated along
a cycle which looks like M \ U. Geometrically, the curvature is concentrating along certain
cycles in M.
Let gl(k, C) = Ckk denote the Lie algebra of the general linear group. A symmetric
polynomial f S j (gl(k, C)) is invariant (or ad-invariant) if
f (a1 X1 a, . . . , a1 Xj a) = f (X1 , . . . , Xj )

for all a GL(k, C), Xi gl(k, C).

Infinitesimally this means

f ([Y, X1 ], X2 , . . . , Xj ) + + f (X1 , . . . , Xj1 , [Y, Xj ]) = 0

for all Xi , Y gl(k, C).

Proposition 9.3 The algebra of invariant symmetric polynomials on gl(k, C) is generated

by the elementary symmetric functions fj S j (gl(k, C)), which are defined by
 X

k

k
1
1
1
det X.
X =
fj (X) = 1
tr X + +
det I
2i
2i
2i
j=0

Let D : A0 (E) A1 (E) be a connection, for example, the Chern connection of an

Hermitian structure, and let D 2 A2 (End E) be the curvature operator. With respect to
a local frame for E, there exists a curvature matrix of two-forms, and under a change of
frame transforms by a similarity. Consequently it makes sense to define j (E, D), the jth
Chern form of E with respect to the connection D, by
(9.2)

 X

k
1
det I
=
j (E, D).
2i
j=0

In other words, j (E, D) = fj (, . . . , ).

Theorem 9.4 The forms j (E, D) are closed, real 2j-forms. The de Rham class represented
by j (E, D) does not depend on the connection D, and the classes cj (E) = [j (E, D)] satisfy
Axioms i.iv.
proof

By the Bianchi Identity, D = 0. It follows that the Chern forms are closed, since
d j (E, D) = D j (E, D) = fj (D, , . . . , ) + + fj (, . . . , , D) = 0.
84

To see that the cohomology class is independent of D, choose connections D0 and D1 , and
let Dt = (1 t)D0 + tD1 be the path of connections interpolating them. With respect to a
local frame, there is a matrix-valued one-form = 1 0 which transforms by a similarity
under a change of frame. The connection form of Dt is t = 0 + t, the curvature form is
t = dt t t , and
d
t = d t t = Dt .
dt
Since fj is ad-invariant, and and transform by a similarity under change of frame, the
expression fj (, t , . . . , t ) is a globally defined (2j 1)-form on M. Set
Z 1
fj (, t , . . . , t ) dt A2j1 (M).
=j
0

A short calculation gives jd[fj (, t , . . . , t )] = (d/dt)fj (t , . . . , t ). It follows that

Z 1 
Z 1

d
d = j
fj (t , . . . , t ) dt = j (E, D1 ) j (E, D0 ),
d fj (, t , . . . , t ) dt =
0
0 dt
proving that the de Rham class of j (E, D) is independent of D. It remains to verify
Axioms i.iv. This is straightforward, since connections and curvature are natural with
respect pullbacks by smooth maps, and since the connection in a sum of line bundles splits
into a diagonal matrix of one-forms.
Corollary 9.5 Let E M be a holomorphic vector bundle, and let E be the dual bundle.
The Chern classes of E are related to those of E by cj (E ) = (1)j cj (E).
proof If E is endowed with an Hermitian structure h, then the map v E 7 h(, v) E
is a conjugate-linear vector bundle isomorphism. Thus E is isomorphic as a smooth complex
vector bundle to the dual bundle E, so cj (E ) = cj (E). If D is a connection in E, then D
is a connection in E; since the curvature matrix is pure imaginary, the curvature matrix
of D differs by a sign from the curvature matrix of D. By equation (9.2) the Chern forms
satisfy j (E, D) = (1)j j (E, D).

9.2

Alternate Definitions

Let p : E M be a vector bundle of rank k over a compact manifold. The tautological

bundle E is defined as follows. Let E denote the complement of the zero section, and let
C act by scalar multiplication in the fibres of E . The quotient space is denoted P(E) and
is called the projectivization of E. There is an induced map p : P(E) M whose fibres are
(k 1)-dimensional projective spaces. Consider the bundle p E P(E). For x M, each
point of P(Ex ) represents a line through the origin in the fibre p Ex , and the tautological
85

bundle is defined to be the line subbundle of p E whose fibre at a point of P(E) is the line
represented by that point. It is not difficult to verify that the total spaces of E and E
are biholomorphic; indeed, the restriction of p : p E E to E is a biholomorphism. Along
the zero section, p collapses the fibres of P(E). The total space of E is said to be obtained
from the total space of E by blowing up the zero section.
Let E denote the dual bundle, and let = c1 (E ) H 2 (P(E ), R); since E is a line
bundle, the first Chern class has been defined. By the Leray-Hirsch Theorem from topology,
the cohomology ring H (P(E), R) is generated, as a p H (M, R)-module, by subject to a
single relation
(9.3)

k
X
j=0

(1)j cj kj = k c1 k1 + + (1)k ck = 0

for some cj H 2j (M, R).

Theorem 9.6 The classes cj in equation (9.3) are the Chern classes of E.
Another functorial definition arises via classifying spaces. Let Gk (Cn ) = Gk,n denote
the Grassmannian manifold of k-dimensional linear subspaces of Cn . There are standard
inclusions Cn , Cn 0 Cn+1 , which induce holomorphic embeddings Gk,n Gk,n+1.
The union over n, with the direct limit topology (in which a set is closed if and only if its
intersection with each Gk,n is closed), is called the infinite Grassmann manifold Gk .
For each n N, there is a universal or tautological rank k vector bundle Uk,n Gk,n Cn ,
whose fibre at x Gk,n is the k-dimensional subspace of Cn represented by the point x.
There are bundle maps Uk,n Uk,n+1 compatible with the embeddings of base spaces, and
the direct limit is the universal bundle Uk Gk .
Consider the standard flag 0 = C0 C1 C2 Cn . Every k-plane W Cn gives
rise to a non-decreasing sequence dim(W Ci ) whose consecutive terms differ by at most
one. Let 1 m1 < m2 < < mk n denote the dimensions in which jumps occur, that
is
dim(W C mi 1 ) = i 1.
dim(W C mi ) = i,
Proposition 9.7 Each
P sequence 1 m1 < m2 < < mk n gives rise to an open cell of
(complex) dimension i (mi i) consisting of k-planes in Cn having the prescribed incidence
on the standard flag. The closure of such a cell is a subvariety of Gk,n called a Schubert
cycle. The number of Schubert cycles of dimension r is equal to the number of partitions of
r into at most k integers, each of which is at most n k.
In the infinite Grassmannian, there is a Schubert cycle of (complex) dimension j corresponding to each partition of j into at most k integers.
86

Theorem 9.8 If j k, then the jth Chern class of the universal bundle is the cohomology
class which takes the value 1 on the Schubert cycle corresponding to the partition (1, . . . , 1)
and vanishes on all other Schubert cycles.
The Chern classes of the universal bundle generate the integral cohomology ring H (Gk ),
and there are no relations among them. In particular, H (Gk , Z) has no torsion and no
non-zero elements of odd dimension.
The importance of the universal bundle Uk Gk is that the Grassmannian is the classifying space for the unitary group U(k). More concretely, there is a one-to-one correspondance
between equivalence classes of complex rank k vector bundles over a CW complex M and
homotopy classes of maps : M Gk . The correspondance associates the bundle E = Uk
to a classifying map . The Chern classes of E may now be defined as the pullbacks of
the universal Chern classes under .
There is a beautifully simple intuitive principle which describes the Poincare duals of the
Chern classes of E: The (k j + 1)st Chern class is Poincare dual to the cycle on which
j generic smooth sections of E are linearly dependent. In particular, the top Chern class
ck (E) is the Euler class, which is Poincare dual to the zero set of a generic section. This
principle has a precise obstruction-theoretic statement. To describe this, it is convenient to
associate to E the bundle of j-frames Vj (E), whose fibre over x M is the Stiefel manifold
of j-frames in Ex .
Theorem 9.9 Let E M be a complex vector bundle of rank k over a CW complex. Then
it is possible to construct a section of Vj (E) over the (2k 2j + 1)-skeleton of M, and
ckj+1(E) is the primary obstruction to extending over the (2k 2j + 2)-skeleton.
Example 9.10 The top Chern class is the Euler class; applying this fact to the tangent
bundle of an almost-complex manifold gives the generalized Gauss-Bonnet Theorem
 n Z
Z
i
(9.4)
(M) =
cn (M) =
det .
2
M
M
One way to see this geometrically is to embed M , Gn,N for N sufficiently large by a
classifying map for T M. The top Chern class cn (M) is Poincare dual to the Schubert cycle
consisting of all n-planes in CN which lie in the hyperplane C N 1 . Let v = eN be the Nth
standard basis vector, and define a smooth vector field on M by orthogonal projection to
Tx M CN . This vector field vanishes exactly when Tx M CN 1 , so the Euler number of
M is equal to the intersection of M and the Poincare dual of cn (M).
2
Example 9.11 By investigating the homotopy groups of the unitary groups, Bott showed
that if E S 2n is a complex vector bundle over the 2n-sphere, then (n1)! divides (S 2n , E).
87

Applying this result to the tangent bundle E = T S 2n , for which = 2, implies that if the
sphere S 2n admits an almost-complex structure, then n 3.
2
The Chern character is a ring homomorphism from the K-theory of M to the rational
cohomology H (M, Q). It is defined for vector bundles in terms
P of a formal
Qfactorization of
j
the total Chern class. Let x be an indeterminate,
and write j cj (E)x = j (1 + j x). The
P
Chern character of E is defined to be j exp j . In terms of a curvature matrix,


1
ch(E) = tr exp
.
2i
Theorem 9.12 If E and F are vector bundles over M, then
ch(E F ) = ch(E) + ch(F ),

ch(E F ) = ch(E) ch(F ).

If the total Chern class of E = T M is factored as above, then the Todd class of M is
defined by
k
Y
j
H (M, Q).
td(M) =
1 exp j
j=0

In terms of the Chern classes of E and M, the Chern character and Todd class are given by

1
ch(E) = k + c1 (E) +
c1 (E)2 2c2 (E) +
2
(9.5)

1
1
td(M) = 1 + c1 (M) +
c1 (M)2 + c2 (M) +
2
12
Let E M be a holomorphic vector bundle over a compact complex manifold. The Euler
characteristic of M with respect to E is
(M, E) =

n
X

(1)i hi (M, E).

i=0

The holomorphic Euler characteristic of M is defined to be (M, OM ). The Riemann-RochHirzebruch Theorem expresses the Euler characteristic (M, E) as the integral over M of a
universal polynomial in the curvature forms of connections on E and T M:
Z
Theorem 9.13 The Euler characteristic is given by (M, E) =
ch(E) td(M).
M

Example 9.14 Let M be a curve of genus g, and let denote the positive generator of
H 2 (M, Z). Let E M be a holomorphic vector bundle of rank k, and define the degree of
E to be
Z
d = deg E =
c1 (E).
M

88

Using equation (9.5), the Todd class of M and Chern character of E are given by
td(M) = 1 + (1 g)

ch(E) = k + d.

By Theorem 9.13, (M, E) = d + k(1 g). If E is a line bundle, then h0 (M, E) h1 (M, E) =
(M, E) = d + 1 g. Consequently, if the degree of E is at least g, then E has a non-trivial
holomorphic section.
2
Example 9.15 Let M = P2 , and let H 2 (M, Z) denote the positive generator. By
Example 9.2, the total Chern class of T P2 is equal to 1 + 3 + 3 2 , so the Todd class is
td(P2 ) = 1 + (3/2) + 3 2 by equation (9.5). For each integer k, consider the rank-two
vector bundles
T (k) := T P2 OP2 (k),

T (k) := T P2 OP2 (k).

By Theorem 9.12, the Chern character of T (k) is







2
2
2
1 + k + (1/2)
ch OP2 (k) = 2 + 3 + (3/2)
ch T (k) = ch T P


= 2 + (2k + 3) + k 2 + 3k + (3/2) 2 .

A similar calculation shows ch T (k) = 2 + (2k 3) + k 2 3k + (3/2) 2 . Theorem 9.13
applied to T (k) gives (T (k)) = k 2 1 = (k + 1)(k 1). When k = 0, this is easily verified,
since T P2 = 1 is the bundle of holomorphic one-forms on P2 . The Dolbeault isomorphism
gives hi (T P2 ) = h1,i (P2 ), so by direct calculation
(T P2 ) = h1,0 (P2 ) h1,1 (P2 ) + h1,2 (P2 ) = 0 1 + 0 = 1
as expected. If k > 0, then h0 (T (k)) = 0; indeed, if the bundle T (k) had a non-trivial
holomorphic section, then tensoring with a section of OP2 (k) would give a non-trivial section
of T P2 , that is, a non-trivial holomorphic one-form on P2 .
Applying Theorem 9.13 to T (k) gives (T (k)) = k 2 + 6k + 8 = (k + 3)2 1. For k = 0,
h0 (T P2 ) h1 (T P2 ) + h2 (T P2) = (T P2 ) = 8.
Since h0 (T P2 ) is the dimension of the space of holomorphic vector fields on P2 , and the
automorphism group of P2 isVthe eight-dimensional Lie group P GL(3, C), h0 (T P2 ) = 8.
The canonical bundle KP2 = 2 T P2 is isomorphic to OP2 (3), so Kodaira-Serre duality
gives h2 (T P2) = h0 (T (3)) which is zero as noted above. Consequently, h1 (T P2 ) = 0. The
vector space H 1 (T M) has a very interseting interpretation, due to Kodaira and Spencer:
it is exactly the space of infinitesimal deformations of the complex structure of M. The
previous calculation shows that P2 has no non-trivial infinitesimal deformations, that is, the
complex structure of P2 is infinitesimally rigid. Much more is known:
89

Theorem 9.16 Let M be a complex surface having the homotopy type of P2 . Then M is
biholomorphic to P2 .
In more detail, the theorem asserts the following. An almost-complex structure J on the
underlying four-manifold P2 may be regarded as a section of the quotient GL(4, R)/GL(2, C)
of the real frame bundle of T P2 by the bundle of almost-complex frames. The theorem
asserts that if J is an integrable almost-complex structure homotopic to the standard almostcomplex structure J0 , then J is biholomorphic to J0 .
Surprisingly, it is not known whether or not there exists a complex structure of general
type on the underlying real four-manifold P2 .
2

10

Vanishing Theorems and Applications

Line bundles are of fundamental importance in the study of complex manifolds for two
principal reasons. First, their sections play the role of global holomorphic functions, which
are lacking on compact manifolds. Second, line bundles have a cohomological interpretation,
which helps make their classification amenable to homological methods. The latter is of
course mostly of importance because of the former; results about vanishing of cohomology can
be used to extract information about global sections of line bundles. Interesting geometric
theorems may in turn be expressible in terms of sections of line bundles.

10.1

Ampleness and Positivity of Line Bundles and Divisors

Let p : L M be a holomorphic line bundle over a (connected) compact complex manifold of

dimension n. Assume further that at each point x M there is a section which is non-zero at
x; in this event, L is said to be generated by global sections. Associated functorially to L is a
holomorphic map from M to a projective space, defined as follows. Let V = VL = H 0 (M, L)
denote the space of global sections of L. The map iL : M P(V ) sends each point x M
to the set of sections of L which vanish at x. Because L is generated by global sections, the set
of sections vanishing at x M is a hyperplane in V ; indeed, with respect to a trivialization
of L, evaluation at x is a linear functional V C. A hyperplane in V determines a linear
functional on V up to a multiplied constant, which is exactly a point of the projective space
P(V ). The map iL : M P(V ) is holomorphic, but is not generally an embedding.
Holomorphicity is easily seen (equation (10.1) below) by choosing a basis {si }N
i=0 of sections
N +1
of L, which defines an isomorphism V ' C
. With respect to a trivialization, the map
iL is given by sending x M to evaluation of [s0 : : sN ] at x, which is an element of
P(CN +1 ). For simplicity, it is customary to write
(10.1)

x M 7 iL (x) = [s0 (x) : : sN (x)] P(CN +1 ),

90

from which holomorphicity of iL is clear.

A line bundle p : L M is very ample if the map iL is an embedding, i.e. is one-to-one
and has differential of rank n everywhere. In this event, an unravelling of definitions shows
that L itself is the pullback of the hyperplane bundle [H] = OPN (1) by the map iL . If some
tensor power Lk is very ample, then L is ample. A divisor D is (very) ample if the associated
line bundle [D] is (very) ample.
There is a precise differential-geometric analogue of ampleness, called positivity. A (1, 1) > 0 for every (1, 0) tangent vector Z. A positive (1, 1)-form is
form is positive if (Z, Z)
exactly a fundamental form of a Kahler metric. A (1, 1)-class H1,1
(M, R) is positive if
there is a positive (1, 1)-form representing ; a positive (1, 1)-class is also called a K
ahler
class, since such classes are exactly those represented by Kahler forms. A holomorphic line
bundle p : L M over a complex manifold is positive if there is an Hermitian structure in
L whose first Chern form 1 (L, h) is positive. Equivalently, p : L M is positive if c1 (L)
is a Kahler class. Finally, a divisor is positive if the Poincare dual (1, 1)-class is positive.
By an obvious sign change, it makes sense to speak of negative (1, 1)-forms and classes, line
bundles, and divisors.
A very ample line bundleand hence an ample line bundleis positive; indeed, the
hyperplane bundle [H] = OPn (1) is positive, so if L is very ample, then L = iL [H] is
positive. The converse assertion is the Kodaira Embedding Theorem, see Theorem 10.10
below.
Remark 10.1 There is another positivity condition a divisor may possess: A divisor D
is numerically effective (or nef) if the intersection product D C is non-negative for every
smooth curve C M. A positive divisor is nef, as is easily seen by Poincare duality. A nef
divisor is not generally positive (i.e. ample), however. There is a numerical condition, called
the Nakai-Moishezon Criterion, under which a divisor is ample. Specifically, if D k V > 0
for every smooth, k-dimensional subvariety of M (k = 1, . . . , n arbitrary), then D is ample.
It is not difficult to show this condition is necessary.
2
Let p : L M be a holomorphic line bundle, and let f : M M be a biholomorphism
(or an automorphism). If the action of f lifts to a bundle map fe : L L, then there
is an induced projective automorphism covering f . It can be shown that if p : L M is
very ample, then every automorphism of M lifts to L. Since iL : M PN is an embedding,
every automorphism of M is induced by an automorphism of the ambient projective space.

10.2

The Kodaira-Nakano Vanishing Theorem

Let : L M be an Hermitian holomorphic line bundle over a compact Kahler manifold

with Kahler form , let D denote the Chern connection of (L, h), and let D = D 1,0 + L be
the type decomposition. There is a map
L : Ap,q (L) Ap+1,q+1 (L),
91

whose adjoint with respect to the global inner product is denoted . In analogy with
equation (8.8) above, there is a commutator relation

(10.2)
[, L ] = 1(D 1,0 ) .
The main technical result of this section, the Kodaira-Nakano Vanishing Theorem, follows
from a short calculation using this equation. The result for p = 0 is due to Kodaira, while
Akizuki and Nakano proved the version stated here. (The content of Remark 10.3 below is
called Nakanos Inequality.)
Theorem 10.2 Let : L M be a positive line bundle over a compact complex manifold.
Then


H q M, p (L) = 0 for p + q > n = dimC M.

proof Choose an Hermitian structure h in L so that

the Chern form
1 (L, h) = 1

is a Kahler form. Then D 2 = R = = 1 = 1L; in words, the

curvature operator R has adual interpretation as the second covariant derivative D 2 and
as the algebraic operator 1L. Comparing these interpretations and using the various
commutation identities will prove the theorem.
Let Hp,q (L) be a harmonic (p, q)-form with values in L. It suffices to show that if
p + q > n, then = 0. By Proposition 8.15, [L, ] = n (p + q) acting on (p, q)-forms
with
values in L. The curvature operator is of type (1, 1), so D 2 = L D 1,0 + D 1,0 L = 1L.
In particular, since is harmonic, L = 0 and D 2 = L D 1,0 , so

(10.3)
L = 1 L D 1,0 + D 1,0 L ,
L = 1 L D 1,0 .
Simple algebra and integration by parts gives

(n p q) kk2 = ([, L] , ) = (L , ) (L , )

=
1( L D 1,0 , ) 1(D 1,0 L , )

by (10.3)

= kD 1,0 k2 + k(D 1,0 ) k2

since L = L 1(D 1,0 ) by equation (10.2). This is non-negative, so = 0 provided

p + q > n, as claimed.
Remark 10.3 Examination of the terms involved shows that, as operators on forms,


[, L] = 1 L , [, D 1,0] + D 1,0 (D 1,0 ) + (D 1,0 ) D 1,0


= 1 L , [, D 1,0] + 2 1,0 .
D

The bracket term does not contribute to ([, L] , ) since L = 0 and L = 0.

Corollary 10.4 If 1 q n 1 and k Z, then H q Pn , OPn (k) = 0.
92

10.3

Cohomology of Projective Manifolds

An extremely important and useful consequence of Theorem 10.2 is the Lefschetz Hyperplane
Theorem, which roughly asserts that the interesting cohomology of a projective manifold
is in the middle dimension. The proof is a typical application of a vanishing theorem: A
geometric problem is phrased in terms of exact sequences of sheaves, then isomorphisms are
extracted from the long exact cohomology sequence by cohomology vanishing.
Theorem 10.5 Let M PN be a smooth algebraic subvariety of complex dimension n + 1,
V M a smooth hypersurface, e.g. a hyperplane section of M. Then the inclusion map
p,q
i : V M induces an isomorphism i : Hp,q
(M, C) H (V, C) for p + q n 1, and an
p,q
injection i : Hp,q
(M, C) H (V, C) for p + q = n.
In particular, the map i : H k (M, Q) H k (V, Q) is an isomorphism for k n 1 and is
injective for k = n. Note that the real dimension of V is 2n. Even the special case M = Pn+1
gives interesting information, cf. Corollary 10.6 below.
proof By the Dolbeault theorem, it suffices to show that the map H q (M, pM ) H q (V, pV )
induced by restriction is an isomorphism for p + q n 1 and is injective for p + q = n.
Consider the following three sheaves on M:
pM

pM V
pV

=
=
=

germs of holomorphic p-forms on M

restriction of pM to V , extended by zero
germs of holomorphic p-forms on V , extended by zero

Intuitively, the sheaf pM V contains information in directions not tangent to V , even though

p
p 
the only
non-zero
stalks
are
over
points
of
V
.
There
are
sheaf
morphisms
r
:

M
M V and

p
p 
0
i : M V V . Let s H (M, [V ]) be given by local defining functions for V . The kernel of
r is the sheaf of germs of p-forms vanishing along V , which is the sheaf pM (V ) = pM [V ],
so there is an exact sequence

s
r
(10.4)
0 pM (V ) pM pM V 0
The restriction map i fits into the short exact sequence

(10.5)


ds
i
0 Vp1 (V ) pM V pV 0.

The 1-form ds defines a global section of the bundle V /M [V ], showing explicitly that the
map i loses exactly the non-tangential information about germs of p-forms along V .
By Kodaira-Serre duality and the Kodaira-Nakano vanishing theorem,


H q (M, pM (V )) = 0 and H q V, Vp1 (V ) = 0 for p + q < n.
93

Applying this vanishing to the long exact sequences associated to the sheaf sequences in

r
i
equations (10.4) and (10.5) gives maps H q (M, pM ) H q (M, pM V ) H q (V, pV ) which
are isomorphisms if p + q n 1 and are injective for p + q = n.
Corollary 10.6 Let V n PN be a complete intersection of complex dimension n. Then
the Dolbeault cohomology of V matches that of PN except possibly in the middle dimension.
Precisely, hp,p(V ) = 1 for 0 p n, 2p 6= n, and all other Hodge numbers hp,q (V ) with
p + q 6= n vanish. In particular, a complete intersection of positive dimension is connected,
and is simply-connected if of dimension at least two.
proof By induction on the codimension of V , Theorem 10.5 immediately implies hp,p (V ) =
1 for 0 2p n1, and all other Hodge numbers hp,q (V ) with p+q n1 vanish. KodairaSerre duality gives the information about Hodge numbers hp,q with p + q > n.
The next result gives topological information from which it is possible to calculate Hodge
numbers of some interesting varieties.
Proposition 10.7 If V = Vd,n is a smooth hypersurface of degree d in Pn+1 , then the
topological Euler characteristic is equal to

n 
X


1
n+2
(10.6)
(Vd,n ) = d
(d)j =
(1 d)n+2 1 + (n + 2)d .
d
j+2
j=0
proof The set of degree d polynomials defining a smooth hypersurface is the complement
of proper algebraic subvarieties and hence is connected. Two such polynomials therefore
have diffeomorphic zero loci, so (Vd,n ) depends only on d and n. Let H Pn+1 be a
hyperplane such that H V is smooth. Projection of V to H is a d-sheeted branched cover,
whose branch locus is a smooth, degree d variety of dimension n 1. Triangulate the branch
locus H V , and extend to a triangulation of H ' Pn . Pulling back gives a triangulation
of V , and calculating the Euler characteristic from this triangulation gives
(V ) = d (H) (d 1)(H V ),

or (Vd,n ) = d(n + 1) (d 1)(Vd,n1 ).

Since (Vd,0 ) = d, equation (10.6) follows by induction on n.

Example 10.8 (K3 surfaces) Let V be a smooth quartic hypersurface in P3 . By the second
adjunction formula, the canonical bundle of V is trivial, and by the Lefschetz hyperplane
theorem, h1,0 (V ) = 0. A complex surface satisfying these properties is called a K3 surface.
By Proposition 10.7, the Euler characteristic of V is 24. From Theorem 10.5, and Poincare
duality, b0 (V ) = b4 (V ) = 1 and b1 (V ) = b3 (V ) = 0, so b2 (V ) = 22. Most of the Hodge numbers may be found immediately from Corollary 10.6: h0,0 (V ) = h2,2 (V ) = 1, and hp,q (V ) = 0
if p + q = 1 or 3. To calculate the remaining Hodge numbers, note that since the canonical
94

bundle of V is trivial, h2,0 (V ) = h0,2 (V ) = 1, leaving h1,1 (V ) = 20. The significance of h1,1 is
seen by the Dolbeault theorem and Kodaira-Serre duality. Since the bundle of holomorphic
1-forms is dual to the holomorphic tangent bundle, and the canonical bundle is trivial,bigr
H 1 (V, 1V ) ' H 1 (M, T M).

The space H 1 (M, T M) is the space of infinitesimal deformations of the complex structure,
and this calculation indicates that there is a 20-dimensional family of complex surfaces
deformation equivalent to a smooth quartic in P3 . An easy count shows that only a 19dimensional family can be accounted for by quartic hypersurfaces themselves, and indeed a
generic deformation of a quartic hypersurface is not algebraic.
A great deal about K3 surfaces is known. Among other things, every K3 surface is
Kahlerian, every smooth K3 surface is diffeomorphic to the Fermat quartic in P3 , and
there is a branched covering from the Teichm
uller spacewhich is diffeomorphic to a
20-dimensional ballto the moduli space of K3 surfaces. K3 surfaces enjoy a number of
striking differential-geometric properties, some of which are discussed in more detail later,
though one can be mentioned here: A smooth K3 surface admits three integrable almostcomplex structures and an Hermitian metric which is Kahler with respect to all three. These
almost-complex structures satisfy the same algebraic relations that the unit quaternions do,
so there is actually an S 2 of integrable almost-complex structures.
K3 surfaces are of interest to algebraic geometers because of their role in the classification
of complex surfaces, to four-manifold topologists because of their curious intersection form,
to number theorists because of their relationship to elliptic curves and higher-dimensional
Abelian varieties, and to theoretical physicists working on supersymmetric field theories
because of their differential-geometric properties.
2
Example 10.9 (Calabi-Yau 3-folds) Consider a smooth quintic hypersurface V P4 .
Again, the canonical bundle is trivial, and by Proposition 10.7 (V ) = 200. By Corollary 10.6, hp,p (V ) = 1 for 0 p 3, all other Hodge numbers hp,q (V ) with p + q 6= 3 vanish,
and h0,3 (V ) = h3,0 (V ) = 1 because the canonical bundle is trivial. Thus h1,2 (V ) = h2,1 (V ) =
101, and by Kodaira-Serre duality this may be interpreted as the dimension of the space of
infinitesimal deformations
of the complex structure. The space of quintic polynomials in five
variables has dimension 95 = 126, while the automorphism group of P4 is 24-dimensional.
Dividing out by non-zero scalars accounts for the last parameter: 101 = 126 24 1. Thus
every small deformation of a smooth quintic hypersurface is a smooth quintic hypersurface.
These manifolds are examples of what are usually called Calabi-Yau three-folds, for reasons that are explained later. Generally, a Calabi-Yau three-fold is a three-dimensional
Kahlerian manifold with no holomorphic one-forms (h0,1 = 0) and with trivial canonical
bundle (h0,3 = 0). Calabi-Yau three-folds are three-dimensional analogues of K3 surfaces,
but are much less well understood. Even Calabi-Yau three-folds which arise as complete intersections in weighted projective spaces are not classified (up to deformation type), though
it is known that the number of deformation types is less than 8,000.
2
95

10.4

The Kodaira Embedding Theorem

As mentioned earlier, every positive divisor is ample. This is the content of the Kodaira
Embedding Theorem. An alternate statement (Corollary 10.11) gives an intrinsic criterion
for a compact Kahlerian manifold to be projective algebraic.
Theorem 10.10 Let p : L M be a positive line bundle over a compact complex manifold.
Then there exists a positive integer k0 such that if k k0 , then Lk M is very ample.
proof (Brief sketch) Proving the Kodaira Embedding Theorem amounts to establishing
the following conditions for k  1, i.e. that there is a k0 such that if k k0 , then
The bundle Lk is generated by global sections, i.e. for every x M there is a section
s H 0 (M, Lk ) with s(x) 6= 0;
Sections of Lk separate points, that is, if x and y are distinct points of M, then there
is a section s with s(x) = 0 and s(y) 6= 0;
Sections of Lk separate tangent vectors, in the sense that if Z and W are vectors
tangent to M at x, then there is a section s such that ds(x)Z = 0 and ds(x)W 6= 0.
Each of these conditions has an interpretation in terms of morphisms of ideal sheaves and
sheaves of germs of sections of Lk . Various morphisms are shown to be surjective by using
the Kodaira-Nakano Vanishing Theorem.
Corollary 10.11 Let (M, J, g) be a compact Kahler manifold whose Kahler form is rational,
2
that is, lies is H1,1
(M, C) H (M, Q). Then there is an integer N > 0 and an embedding
N
i : M P such that the pullback of the Fubini-Study form on PN is an integral multiple
of the Kahler form of g.
proof The long exact sequence associated to the exponential sheaf sequence on M contains
the terms
c1

H 1 (M, OM
)
H 2 (M, Z) H 2 (M, OM ).
Because the Kahler form 0 of g is rational, there is a positive integer m such that the
(1, 1)-form = m0 is integral. For type reasons, the image of in H 2 (M, OM ) is zero.
By exactness, there is a holomorphic line bundle p : L M with c1 (L) = > 0. By
Theorem 10.10, L is ample, so there is an integer k > 0 such that Lk is very ample and the
restriction of the Fubini-Study metric to the image of M is k = km0 .
Example 10.12 A metric g whose Kahler form is rational is called a Hodge metric, and a
manifold admitting a Hodge metric is called a Hodge manifold. Not every Kahlerian manifold

96

is a Hodge manifold; an example is provided by the two-dimensional compact complex torus

whose lattice in C2 is generated by
 
 
 
 
1
0
2

5 .
,
,
,
0
1
3
7
In fact, this torus admits no non-constant meromorphic functions.
2
Theorem 10.10 has a number of geometrically interesting consequences. Some of these
are listed below; the proofs are for the most part immediately apparent.
Corollary 10.13 Let M1 and M2 be projective algebraic manifolds. Then M1 M2 is projective algebraic.
Corollary 10.14 Let M be a compact Kahlerian manifold with b2 (M) = 1. Then M is
projective algebraic. In particular, if M is a compact complex curve or a compact, irreducible
Hermitian symmetric space, then M is projective.
f M is a finite unbranched covering of compact complex maniCorollary 10.15 If : M
f is.
folds, then M is projective if and only if M

The next two results are immediate consequences of the proof of Theorem 10.10, but are
not immediate from the sketch given above.

f M is a blow-up of M at one
Corollary 10.16 If M is projective algebraic, and if : M
f is projective algebraic.
point, then M
Corollary 10.17 Let L M be an ample line bundle. For every line bundle L0 M,
there is an integer k0 such that if k k0 , then Lk L0 is very ample.

10.5

The Hodge Conjecture

There is a generalization of the Kodaira-Nakano vanishing theorem which is reminiscent of

Corollary 10.17, and which has a useful GAGA-type consequence, see Corollary 10.19 below.
While Theorem 10.18 is immediate from Corollary 10.17, a complete proof of the former is
provided.
Theorem 10.18 Let p : L M be a positive line bundle over a compact complex manifold,
and let L0 M be an arbitrary line bundle. Then there exists a k0 N such that
H q (M, Lk L0 ) = 0

for k k0 , q > 0.
97

proof By hypothesis, there is an Hermitian structure in L whose first Chern form 0

1
is positive. Endow L0 KM
with an Hermitian structure h, and consider the first Chern
1
0
form = 1 (L KM , h). For each x M, there is an integer kx such that the (1, 1)-form
kx 0 + > 0 on some neighborhood of x. By compactness of M, there is a covering of M by
finitely many such neighborhoods; take k0 to be the maximum of the kx s. If k k0 , then
1
k0 + > 0 everywhere on M, that is, Lk L0 KM
is a positive line bundle. Interpreting
KM as the bundle associated to the sheaf nM and using Theorem

10.10 with p = n shows
1
q
k
0
q
n
k
0
that if q > 0, then H (M, L L ) = H M, M (L L KM ) = 0.
Corollary 10.19 Let M PN be a smooth submanifold. Then every holomorphic line
bundle L M is of the form [D] for some divisor. Equivalently, L has a meromorphic
section.
proof The strategy is to let D0 be a smooth hyperplane section of M; if s is a global
holomorphic section of L(kD0 ), and if t is a global section of [kD0 ], then t1 is a meromorphic
section of [kD0 ] = [kD0 ], so s t1 is a global meromorphic section of L.
The result is vacuous if M is a point. Assume the assertion is true for manifolds of
dimension n
1. Choose a smooth hyperplane section D0 = M H, and choose k so that
H 1 M, L(kD0 ) = 0 (by Theorem 10.18) and so that L(kD0 )|V has a holomorphic section
(induction hypothesis).
Let s H 0 (M, [D0 ]) be a global section vanishing along D0 ; tensoring with s and restriction from M to V fit into a short exact sheaf sequence



r

s
0 OM L((k 1)D0 ) OM L(kD0 ) OV L(kD0 )|V 0.




The long exact sequence contains the terms H 0 M, L(kD0 ) H 0 V, L(kD0 )|V 0, from
which it follows that L(kD0 ) has a non-trivial holomorphic section.

Remark 10.20 Corollary 10.19 follows from the Kodaira embedding theorem as well. Let
D0 = M H be a smooth hyperplane section of M. The line bundle [D0 ] is positive, hence
ample by Theorem 10.10, and by Corollary 10.17 there is an integer k such that L(kD0 ) is
very ample. Let s be a global holomorphic section of L(kD0 ), and let t be a global section
of [kD0 ]. Then t1 is a meromorphic section of [kD0 ] = [kD0 ], so s t1 is a global
meromorphic section of L.
2
Let (M, J, g) be a compact Kahler manifold. By Hodge theory, the de Rham cohomology
(or singular cohomology) has a filtration

Mh
M
i
p,q
q,p
r
r
H (M, R) =
H (M, C) H (M, C) H (M, R) =:
H p,q (M, R),
pq

pq
p+q=r

A natural question is to determine which cycles are Poincare dual to classes in H p,q (M, R).
A homology class is analytic if there is a representative which is a rational linear combination
98

of fundamental classes of analytic subvarieties of M. The Poincare dual of an analytic class

is of type (p, p).
Hodge Conjecture If M PN is smooth, then every class in H p,p(M, Q) is Poincare
dual to an analytic class.
The only case in which the Hodge Conjecture is known completely is for p = 1, by the
Lefschetz Theorem on (1, 1)-classes:
Theorem 10.21 Let M PN be a smooth projective variety. Then every (1, 1)-class in
H 1,1 (M, R) H 2 (M, Z) is Poincare dual to a divisor.
proof (Brief sketch) The map i : H 2 (M, Z) H 2 (M, O) induced by inclusion i : Z ,
OM factors as
0,2
H 2 (M, Z) Hd2 (M, C) H0,2
(M, C).
Consequently, if is an integral (1, 1)-class, then i = 0, which implies is in the image of
the Chern class map c1 : H 1(M, O ) H 2 (M, Z). By Corollary 10.17, is Poincare dual
to a divisor.

Exercises
Exercise 10.1 Prove that if E M is a vector bundle over a compact complex manifold,
then c1 (det E) = c1 (E) H 2 (M, Z).

Exercise 10.2 Let V Pn be a complete intersection of multi-degree (d1 , . . . , dk ). Calculate
the first Chern class of T V .


Exercise 10.3 Let (M, J, g) be a compact Kahler manifold, and let

be a d-exact real
= .
(p, p)-form. Prove that there exists a real (p 1, p 1)-form with 1
Suggestion: First write = d( + ),
then consider the type decomposition upon expanding

this out. Use the Hodge Theorem to write as the sum of a harmonic form and a -exact
form , then express in terms of .


Exercise 10.4 Let p : L M be a holomorphic line bundle over a compact Kahler

manifold, and let 2c1 (L) be a smooth (1, 1)-form. Prove that there is an Hermitian
structure in L whose first Chern form is .

then search for

Suggestion: Let 0 = 1 (L, h0 ) be a curvature form; write 0 = 1 ,
a smooth function u : M R such that the curvature of e2u h0 is .)


99

11

Curvature and Holomorphic Vector Fields

Let (M, J, g) be an Hermitian manifold, and let D be the Levi-Civita connection of the
Riemannian metric g. The curvature tensor R maps local real vector fields X, Y , and Z to
the local vector field
R(X, Y )Z = DX DY Z DY DX Z D[X,Y ] Z.
It is often convenient to use the four-tensor R0 defined by


0
(11.1)
R (X, Y, W, Z) = g R(X, Y )Z, W .
The following properties of R0 are well-known facts from differential geometry.
Proposition 11.1 The tensor R0 is skew-symmetric in X and Y , and in W and Z, and
is symmetric under exchange of (X, Y ) and (W, Z). Further, R0 satisfies the first Bianchi
identity
SY,W,Z R0 (X, , , ) = 0.
(cyclic sum)

If in addition g is Kahler with respect to J, then R0 (JX, JY, W, Z) = R0 (X, Y, W, Z).

The skew-symmetry properties of R0 encapsulate the fact that the curvature operator D 2 :
A0 (M) A2 (End T M) is a two-form with values in End T M, and that this endomorphism
is skew-symmetric. Alternately, R0 may be regarded as a symmetric endomorphism of the
bundle of two-forms on M.
The sectional curvature of a real two-plane P Tx M is the value R0 (e1 , e2 , e1 , e2 ) of the
curvature tensor on an orthonormal basis of P . Geometrically, the sectional curvature is the
Gaussian curvature at x of the surface in M obtained by exponentiating P . The sectional
curvature function K is defined on the Grassmannian bundle of real two-planes in T M. If
P is a complex line, i.e. a J-invariant real two-plane, then the sectional curvature is equal to
R0 (e, Je, e, Je). The restriction of the sectional curvature function to the bundle of complex
lines is called the holomorphic sectional curvature Khol .
If the sectional curvature function K : G2 (T M) R is constant, then the curvature
tensor has an explicit algebraic expression in terms of the metric g; in particular, for each
c R, there is a local model space with constant sectional curvature c. A similar fact
is true when g is a Kahler metric with constant holomorphic sectional curvature. If g is
(geodesically) completemeaning every geodesic extends to have domain Rand simplyconnected, then spaces of constant curvature are classified.
Theorem 11.2 Let (M, J, g) be a complete, simply-connected Kahler manifold of dimension n and constant holomorphic sectional curvature c. If c > 0, then g is isometric to a
100

multiple of the Fubini-Study metric on Pn ; if c = 0, then g is isometric to the flat metric

on Cn ; if c < 0, then
g is isometric to a multiple of the Bergmann metric, whose K
ahler
2
n

form is given by = 1 log(1 kzk ) on the unit ball in C .

At each point x of a standard model space, the holomorphic sectional curvature achieves
the maximum of the (absolute value of) the sectional curvature, and the sectional curvatures
are 1/4-pinched in the sense that
(1/4)|Khol (x)| |K(x)| |Khol (x)|.
In fact, there is a simple formula for the sectional curvature of a two-plane P in terms of the
angle between P and JP . The integral Fubini-Study metric on Pn has holomorphic sectional
curvature 4.

11.1

Ricci Curvature

The Ricci tensor of g is the symmetric, real two-tensor r whose value on vector fields X and
Y is defined to be the trace of the endomorphism V 7 R(V, X)Y ; symbolically,


r(X, Y ) = tr V 7 R(V, X)Y .
The Ricci tensor may be expressed more concretely in terms of the tensor R0 and a unitary
frame {ei }ni=1 for Tx M as
n
X
r(X, Y ) =
R0 (ei , X, ei , Y ).
i=1

Because r is symmetric, it is determined by the real, J-invariant quadratic form Q(X) =

r(X, X); in fact, the Ricci curvature is determined by restricting further to the unit tangent
bundle, that is, to the sphere bundle of unit vectors in T M.
As a real, symmetric, J-invariant two-tensor, r is associated to a real (1, 1)-form , called
the Ricci form of g, and defined by
(11.2)

(X, Y ) = r(X, JY ).

The profound differences between Riemannian and Kahlerian geometry are in large part due
to existence of the Ricci form, and to the following properties it enjoys.
Proposition 11.3 The Ricci form is closed,
R is
V1,1 and represents 2c1 (M). If the curvature
viewed as a symmetric endomorphism of
T M, then = R(). The Ricci form is 1
times the curvature of the canonical bundle of M with the Hermitian structure
P induced by g;
concretely, if the Kahler form of g is given in local coordinates by = 1 gdz d
z ,
then the Ricci form is equal to

(11.3)
= 1 log det(g).
101

The geometric significance of Proposition 11.3 is that on a compact Kahlerian manifold

the Ricci form has a cohomological interpretation, that the Ricci form varies within a fixed
de Rham class as the Kahler metric varies arbitrarily, and that the Ricci form depends only
on the complex structure J and the volume form of the metric. The following useful fact
may be proven in a fashion very similar to the proof of part v. of Proposition 8.11.
Proposition 11.4 Let be a Kahler class, and let 0 be a smooth representative. For every
representative
, there is a smooth, real-valued function, unique up to added constants, such
.
that = 0 + 1 f
Intuitively, smooth forms in a fixed Kahler class are parametrized by smooth functions, and
the set of Kahler forms in a fixed Kahler class is (identified with) a convex open set in
an infinite-dimensional vector space. The importance is that various geometric existence
questionssuch as existence of an Einstein metric, which reduces to an overdetermined
nonlinear system of PDEs in the Riemannian situationreduce to a single PDE in the
Kahlerian situation.

11.2

Holomorphic Vector Fields

It is assumed throughout this section that (M, J, g) is a fixed compact (connected) Kahler
manifold. Every holomorphic vector field Z H 0 (M, T 1,0 M) corresponds to an infinitesimal
automorphism of J, that is, to a real vector field X = Re Z such that LX J = 0.
Theorem 11.5 The group Aut(M) of automorphisms of M is a complex Lie group whose
Lie algebra H is exactly the space of holomorphic vector fields on M.
Since the space of holomorphic vector fields on a compact manifold is finite-dimensional,
the automorphism group of a compact complex manifold is finite-dimensional. A generic
compact manifold has no automorphisms at all, and under various hypotheses strong restrictions can be put on the number of automorphisms. The following theorem for Riemann
surfaces is due to Hurwitz.
Theorem 11.6 Let M be a compact curve of genus g 2. Then | Aut(M)| 84(g 1).
In words, a compact Riemann surface of genus at least two has only finitely many automorphisms, and the number of automorphisms is bounded by the genus. A generic Riemann
surface of genus at least two has no automorphisms, and while the bound in Theorem 11.6
is sharp for infinitely many genera g 3, it also fails to be sharp for infinitely many genera.
Indeed, a result of Accola asserts that for infinitely many genera g 3, the order of the
automorphism group of a compact Riemann surface of genus g is bounded above by 8(g + 1).
102

Example 11.7 Let M P2 be the (singular) curve with equation x2 z 4 (x z) y 7 = 0.

Then M has genus three and 168 automorphisms. There is a smooth planar model of this
curve, called the Klein quartic curve. The automorphism group of the Klein quartic curve
is the simple group of order 168.
2
To study the space of holomorphic vector fields, it is convenient to dualize to the space
of (0, 1)-forms and use Hodge theory. If X is a vector field, then the dual one-form X [ is
defined by X [ (Y ) = g(X, Y ). If X is a real vector field, then X [ is a real one-form, while
if X is of type (1, 0) (e.g. if X is holomorphic) then X [ is of type (0, 1). The inverse map
from one-forms to vector fields is denoted ], and these maps are read flat and sharp
respectively. The terminology comes from tensor calculus, since flat lowers indices and sharp
raises them. Specifically, if g are the components of g in a local holomorphic coordinate

system, and if g is the inverse matrix, then

n
X
[
=
gX d
z ,
X
X =

z
=1
,=1
[

n
X

n
X
=1

z
d

]

n
X

,=1

.
z

If : M C is a smooth function, then the (complex) gradient of with respect to g is the

smooth (1, 0) vector field
] =: ] .
gradg = ()
The Lichnerowicz ideal is defined to be h1 = {Z h : (Z) = 0 for all Z1,0
},
namely, the space of holomorphic vector fields which annihilate the space of holomorphic
one-forms under the natural pairing. Of course, the pairing of a holomorphic one-form with
a holomorphic vector field is a global holomorphic function, hence is constant. The ideal
h1 h actually contains the derived subalgebra [h, h]. This is trivial from the equation
2d(Z, W ) = Z (W ) W (Z) ([Z, W ]),
since all terms except the last are automatically zero for all holomorphic one-forms and all
holomorphic vector fields Z and W .
Theorem 11.8 Let Z be a holomorphic vector field, and let = Z [ be the associated oneform. Then
with unique up to an added constant.
The Hodge decomposition of is = H + ,
if and only if Z h1 , i.e. (Z) = 0 for every holomorphic one-form .
=
X = Re Z is Killing if and only if Re is constant.

103

proof Because g is Kahler, there exist local holomorphic normal coordinates; in such a
= 0 at x.
coordinate system, it is obvious that Z is holomorphic at x M if and only if
The Hodge decomposition of is
G) + (G);

= H + (
The last term vanishes because Z is holomorphic. Set = G; the complex-valued function
is unique up to an added harmonic function, i.e. an added constant since M is compact.
To prove the second assertion, let be a holomorphic one-form. By Proposition 8.14,
is harmonic, and since (Z) = h, i (pointwise inner product) is constant, the global inner
product (, ) is equal to (Z) times the volume of M. Thus, if H = 0, then (, ) = 0.
Conversely, suppose (Z) = 0 for every holomorphic one-form . The (1, 0)-form H is
harmonic, hence holomorphic for type reasons. But by hypothesis, kHk2 = (, ) = 0.
The last assertion is a consequence of the Weitzenbock formula for the Laplace operator
acting on one-forms and is not given here.
The term H is dual to a nowhere-vanishing, autoparallel holomorphic vector field, called
is dual to the gradient part of Z. The (Abelian)
the autoparallel part of Z, and the term
Lie algebra of autoparallel vector fields is denoted a. If h0,1 (M, C) = 0, e.g. if M is simplyconnected, then H = 0 and every holomorphic vector field is a gradient field. The (finitedimensional) space of smooth, complex-valued functions with gradg holomorphic is called
the space of g-holomorphy potentials. While the space of holomorphy potentials depends on
the choice of metric, the dimension does not.
Corollary 11.9 If Z is a holomorphic vector field which vanishes somewhere, then Z is a
gradient field. If X = Re Z is Killing, and if is the real one-form dual to X, then the
following are equivalent:
i. The zero set of Z (or of X) is non-empty.
with pure imaginary.
ii. =
iii. = J du for some real-valued function u.
A smooth function is a holomorphy potential if and only if ] = 0. It is convenient
to introduce the fourth-order scalar Lichnerowicz operator L = ( ] ) ( ] ), whose kernel
is the space of g-holomorphy potentials. The Weitzenbock formula for L is found by standard differential-geometric techniques, namely integration by parts and application of Ricci
identities:
Proposition 11.10 Let denote the Levi-Civita connection of g, r the Ricci tensor, and
s = tr r the scalar curvature function. Then for every smooth function ,
(11.4)

2L = 2d + hr, i + s(gradg ).
104

In particular, the Lichnerowicz operator is a real operator if and only if the scalar curvature
function s is constant, a fact which will be of great importance later.

Part I

Metrics With Special Curvature

12

Einstein-K
ahler Metrics

Without further mention, all manifolds are assumed to be connected. A Riemannian metric
g is Einstein if the Ricci tensor is proportional to the metric tensor, that is, if there is
a real constant called the Einstein constantsuch that r = g. If g is Kahler with
respect to a complex structure J, then the triple (M, J, g) is an Einstein-Kahler metric (to a
complex geometer) or a Kahler-Einstein metric (to a differential geometer). A Kahler metric
is Einstein if and only if the Ricci form is proportional to the Kahler form:
(12.1)

= .

The importance of this equation is that while the Ricci form is 1 times the curvature of
the canonical bundle of M and hence represents the de Rham class 2c1 (M), the form is
positive. Consequently, the sign of an Einstein-Kahler metric (if any) is determined by the
complex structure. It is not known at present whether or not a compact, smooth manifold
can admit (Riemannian) Einstein metrics with different signs. The following summarizes
the obersvation for Kahlerian manifolds, and is an immediate consequence of the Kodaira
embedding theorem.
Proposition 12.1 Let (M, J, g) be a compact Einstein-K
ahler manifold with Einstein constant .
If < 0, then the canonical bundle KM is ample.
If = 0, then the canonical bundle is trivial.
If > 0, then the anticanonical bundle is ample.
A manifold (M, J) with KM ample is called a Fano manifold. The fact that the canonical bundle of a compact Einstein-Kahler manifold is either trivial or else has a sign strongly
restricts the complex structure of M. By way of illustration, the blow-up of P2 at three
collinear points cannot admit an Einstein-Kahler metric, while the blow-up at three noncollinear points admits a positive Einstein-Kahler metric. In addition, the first Chern class
105

of a complete intersection in PN always satisfies the conclusion of Proposition 12.1, while

the blow-up of P2 at nine or more points never satisfies the conclusion.
Under scaling, the Ricci tensor and form do not change. Thus a Kahlerian manifold
(M, J) admits an Einstein-Kahler metric if and only if it admits such a metric with = ,
= 1, 0, or 1. This restriction is assumed from now on.
The converses of the first two conditions in Proposition 12.1 are true (Theorems 12.2
and 12.3 below), while the converse of the third is known to be false. There are additional
necessary conditions in order that a Fano manifold admit an Einstein-Kahler metric, and
additional hypotheses under which these conditions are sufficient for existence. It has recently
been conjectured by Tian that a certain stability condition on (M, J) is sufficient to
guarantee existence. This is discussed in detail below.

12.1

The Calabi Conjectures

The so-called First Calabi Conjecture (see Theorem 12.2) is concerned with specification of
the Ricci curvature of a compact Kahlerian manifold, while the Second Calabi Conjecture
(see Theorem 12.3) is the converse of the first assertion in Proposition 12.1. The First Calabi
Conjecture was established by Yau in 1977, while the second was proven independently by
Yau and Aubin; the uniqueness assertion was known to Calabi in 1954.
Theorem 12.2 Let (M, J) be a compact Kahlerian manifold, let 0 be a positive (1, 1)-form
on (M, J), and let 2c1 (M) be an arbitrary smooth form. Then there is a unique K
ahler
metric g on (M, J) with [] = [0 ] and having Ricci form .
In words, given a representative of 2c1 (M), each Kahler class on (M, J) contains a
unique representative with the specified Ricci form. In particular, if c1 (M) = 0, then every
Kahler class contains a unique Einstein-Kahler metric with = 0.
Theorem 12.3 Let (M, J) be a compact complex manifold with ample canonical bundle.
Then there is a unique Einstein-K
ahler metric g with = .
Example 12.4 By the adjunction formula, if M n PN is a complete intersection of
multidegree d = d1 + + dN n , and if N + 1 = d, then M admits a Ricci-flat Kahler metric
in each Kahler class, while if N + 1 < d, then M admits a unique (up to scaling) negative
Einstein-Kahler metric. In particular, each smooth quartic surface in P3 and each smooth
quintic three-fold in P4 admits a Ricci-flat Kahler metric; for this reason, quintic three-folds
are called Calabi-Yau manifolds. The space of Kahler classes on a quartic surface is known to
be three-dimensional by the Hirzebruch Signature Theorem, so there is a three-dimensional
family of Ricci-flat Kahler metrics on each smooth quartic. As shown in Example 10.8, the
space of Kahler classes on a quintic threefold is one-dimensional, so each Calabi-Yau threefold
hasup to scalinga unique Ricci-flat Kahler metric.
2
106

Very little is known regarding existence if 2 < d < N +1. However, contrary to the natural
guess, the restriction of the Fubini-Study metric to M is never Einstein if 2 < d. In fact,
only the obvious complete intersections inherit an Einstein metric from the Fubini-Study
metric; this is a result of Kobayashi and Ochiai:
Theorem 12.5 Let M n PN be a smooth projective variety, and assume the restriction of
the Fubini-Study metric to M is Einstein. Then M is contained in a linear subspace Pn+1
and is either a linear subspace or a quadric hypersurface in Pn+1 .
The proofs of Theorems 12.2 and 12.3 are entirely hard analysis. However, it is easy to
reduce each theorem to a question of existence for a complex Monge-Amp`ere equation.
Recall

that if 0 is given, then every Kahler form cohomologous to 0 is given by = 0 + 1

for a smooth, real-valued function unique up to added constants. Let 0 be the Ricci form
of 0 , and let be the prospective Ricci form of the unknown metric. By Proposition 11.3,
 n

;
0 = 1 log
0n
this makes sense since the quotient of (non-vanishing) (n, n)-forms is a globally defined
smooth, real-valued function. Yau showed that (with respect to suitable topologies coming
from Holder norms) the map
 n

7 log
0n

> 0 to the space of C

is a diffeomorphism from the set of C 2+ functions with 0 + 1
functions on M. This involved proving the map is opena straightforward application of
the Inverse Function Theorem for Banach spacestogether with an argument that the map
is proper, which involved difficult estimates. As remarked above, that fact that this map is
injective was proven by Calabi in 1954.
To express Theorem 12.3 in terms of a Monge-Amp`ere equation, fix 0 2c1 (M)
(which by hypothesis is a Kahler class), and define the smooth, real-valued function f0 by

0.
0 + 0 = 1 f
The function f0 is called the Ricci potential of 0 . With notation as above, it is easy to show
that is Einstein-Kahler with Einstein constant if and only if
 n

+ = f0 .
log
0n
Aubins proof of Theorem 12.3 used the so-called continuity method; Aubin introduced a
family of equations
 n

()t
log
+ t = f0 .
0n
107

By Theorem 12.2, equation ()0 has a solution. By the Inverse Function Theorem, if ()t0
has a solution, then there is an > 0 such that ()t has a solution for t < t0 + . It remains
to show that the set of t [0, 1] for which ()t has a solution is closed. It is only at this
point that the assumption = 1 must be introduced. Proving closedness involves a priori
estimates on and is the analogue of properness in Yaus proof. More explicitly, the idea is to
show that if ()t has a solution t for t < t0 , then in a suitable Holder norm, the functions t
remain bounded as t % t0 . The same general technique is used to prove existence of positive
Einstein-Kahler metrics, though the details are quite different for geometric reasons. The
problem is to connect geometric hypotheses about the complex manifold (M, J) to analytic
properties, namely, existence of a priori estimates for ()t with = 1.
At present, there are no explicit examples of non-positive Einstein-Kahler metrics on
(compact) simply-connected manifolds. However, even existence of Kahler metrics with
specified Ricci curvature gives useful information. A nice application is the following result
of Kobayashi.
Proposition 12.6 Let (M, J) be a Fano manifold. Then M is simply-connected.
proof (Sketch) By Theorem 12.2, there is a Kahler metric g with positive Ricci curvature. (Kobayashis original article contained this as a hypothesis, though added the present
weaker hypothesis modulo the First Calabi Conjecture.) There is a vanishing theorem, due
to Bochner and analogous to the Kodaira-Nakano Theorem, that a Kahler manifold with
positive Ricci tensor admits no non-trivial holomorphic p-forms if p > 0. (The proof is
a simple application of the Weitzenbock formula for the Laplacian acting on (p, 0)-forms.)
Consequently, the holomorphic Euler characteristic
(M, OM ) =

n
X

p 0

(1) h

(M, pM )

n
X

(1)p hp,0 (M)

p=0

p=0

is equal to 1.
f M be the universal covering space. By Myers Theorem (which holds for
Let : M
f is compact, so is a finite-sheeted
Riemannian manifolds with positive Ricci tensor), M
cover with d sheets. The pullback of g is a Kahler metric with positive Ricci curvature,
f, O f) = 1. By the Riemann-Roch-Hirzebruch Theorem (Theorem 9.13 above), the
so (M
M
Euler characteristic (M, OM ) is equal to the integral over M of the Todd class of M (since
the Chern character of O is equal to 1), which is a universal polynomial in the curvature
f so 1 = (M
f, O f) = d, the
of . The same formula holds locally for the Todd class of M,
M
number of sheets of the cover .
Another useful result, due to Yau, characterizes metrics of constant negative holomorphic
sectional curvature among compact Einstein-Kahler metrics. The idea of the proof is to
calculate Chern forms with respect to an Einstein-Kahler metric, then use the fact that
108

there is a unique simply-connected model space with constant negative holomorphic sectional
curvature.
Proposition 12.7 Let (M n , J) be a compact Kahlerian manifold with c1 (M) < 0. Then


(12.2)
(1)n 2(n + 1)c2 (M) c1 (M)n2 nc1 (M)n [M] 0,
with equality if and only if M is holomorphically covered by the unit ball in Cn .

12.2

Positive Einstein-K
ahler Metrics

In contrast to the existence problem for non-positive Einstein-Kahler metrics, the basic
restriction (i.e. that M be Fano) is not sufficient to guarantee existence of a positive Einstein
metric. The first non-trivial obstruction was found by Matsushima in 1954.
Theorem 12.8 Let (M, J, g) be a compact Einstein-K
ahler manifold with positive curvature.
Then the Lie algebra h is the complexification of the compact Lie algebra k of Killing vector
fields.
A Lie group whose Lie algebra is the complexification of a compact subalgebra (i.e. the
Lie algebra of a compact subgroup) is said to be reductive. Not every Fano manifold has
reductive Lie algebra of holomorphic vector fields; such manifolds cannot admit an EinsteinKahler metric. If M is obtained from Pn by blowing up a linear subspace of dimension `,
0 ` n 2, then M is Fano but admits no Einstein-Kahler metric.
Theorem 12.8 was generalized by Lichnerowicz in 1957 to Kahler metrics with constant
scalar curvature.
Theorem 12.9 Let (M, J, g) be a Kahler manifold with constant positive scalar curvature.
Then the Lie algebra h of holomorphic vector fields decomposes as the sum

h = a k 1k
of the space of autoparallel vector fields, Killing fields, and J(Killing fields).
proof By equation (11.4), which expresses the Lichnerowicz operator in terms of the
Laplace operator, Ricci tensor, and scalar curvature, the scalar curvature s is constant if
and only if L = (] ) (] ) is a real operator. If is a holomorphy potential, then Re and
Im are also holomorphy potentials since L is real. By Theorem 11.8, if ker L is pure
imaginary, then gradg is a Killing field. The theorem follows immediately.
Thus, the blow-up of Pn along a linear subspace admits no Kahler metric of constant
scalar curvature, regardless of Kahler class.
109

A second obstruction to existence of positive Einstein-Kahler metrics was discovered

in 1983 by Futaki, who generalized an integral invariant found in 1974 by Kazdan and
Warner (who had been attempting to determine which smooth functions on S 2 are Gaussian
curvature functions of some Riemannian metric). The Futaki invariant was generalized in
various ways between 19845 by Calabi, Futaki, and Bando.
Let (M, J) be a compact Kahlerian manifold, and let h be the space of holomorphic
vector fields on M. For each Kahler form , let f denote the normalized Ricci potential,
that is, the unique smooth, real-valued function with
Z

H = 1 f ,
f
= 0.
n!
M
Let = [] denote the Kahler class of g, and consider the linear functional
Z

n
(Xf ) .
(12.3)
F : X h 7 1
n!
M
The following summarizes the properties of the so-called Futaki character.
Theorem 12.10 The functional F defined by equation (12.3) is a Lie algebra character,
and depends only on the de Rham class = []. In particular, if there is a Kahler metric of
constant scalar curvature whose Kahler form represents , then the character F vanishes
identically.
Example 12.11 Let M n be obtained by blowing up Pn along skew linear subspaces of
complementary dimension (neither of which is a hyperplane). Then M is Fano, and the
set of Kahler classes for which F vanishes identically is a real-algebraic hypersurface in
the Kahler cone H+1,1 (M, R). The Futaki invariant Fc1 (M ) vanishes identically if and only
if n = 2k + 1 and the subspaces blown up are both of dimension k; in fact, M admits an
Einstein-Kahler metric exactly under this condition. In particular, there exist Fano manifolds
satisfying Matsushimas obstruction but having no Einstein-Kahler metric.
It is not at present known whether or not there exists a compact manifold M with nonreductive automorphism group but having vanishing Futaki character for every Kahler class,
nor whether or not there exists a compact manifold having reductive automorphism group
but non-vanishing Futaki character for every Kahler class.
2

110