Complex Manifolds

Prof. Dr. Martin SCHLICHENMAIER

lecture notes written by

Alain LEYTEM

Master course, 2nd semester

Contents

1 Holomorphic functions in 1 variable 4

1.1 Notations and definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The Cauchy-Riemann equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Cauchy integral formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Holomorphic functions in n variables 6

2.1 Notations and definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Power series in n variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Analyticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Identity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Maximum Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 Hartog’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Real differentiable manifolds 12

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Topological properties of manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Differentiable functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Orientability of differentiable manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Complex manifolds 20
4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Complex manifolds are orientable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Holomorphic functions on complex manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Complex submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 Examples of complex manifolds 26

5.1 Cn and open subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Submanifolds of Cn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3 The projective space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.4 Submanifolds of Pn (C) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.5 Complex tori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.6 Complex Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.7 Grassmannians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6 Sheaf of holomorphic functions 43

6.1 Global holomorphic functions on a compact complex manifold . . . . . . . . . . . . . . . . . . . . 43
6.2 Sheaves : definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Morphisms of sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.4 Restriction of sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.5 Ideal sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.6 Germs and stalks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.7 Remarks on sheafification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

1
Complex Manifolds Section 0.0 SCHLICHENMAIER, Leytem

7 Meromorphic functions 53
7.1 Construction of meromorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.2 Definition using UFDs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.3 Definition using exceptional sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.4 Analytic sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.5 Application to meromorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

8 Analytic sets and singularities 64

8.1 Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.2 Regular and singular points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.3 Irreducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.4 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

9 Holomorphic vector bundles 72

9.1 Biholomorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
9.2 Vector bundles : basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
9.3 Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.4 Cocycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.5 Isomorphic bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.6 Frame of a vector bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

10 Operations on vector bundles 85

10.1 Induced operations on vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
10.2 Sub-bundles and quotient bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.3 Pull-back bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.4 Associated frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.5 Line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

11 Tangent vectors and differentials 97

11.1 The real picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
11.2 The complex picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11.3 Differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.4 De Rham and Dolbeault cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

References 107

2
Complex Manifolds Section 0.0 SCHLICHENMAIER, Leytem

Introduction
Let M be a set with a collection of coordinate charts (Ui , ϕi )i∈I , i.e.
∼
∀ i ∈ I, ϕi : Ui ⊆ M −→ ϕi (Ui ) ⊆ Cn ∼
= R2n where ϕi (Ui ) is open in Cn

M is called a complex manifold if these charts satisfy

S
a) The chart domains Ui cover M : i∈I Ui = M
b) ϕi (Ui ∩ Uj ) is an open subset of Cn , ∀ i, j ∈ I
c) The locally defined transition maps ϕj ◦ ϕ−1 i : ϕi (Ui ∩ Uj ) → ϕj (Uj ∩ Ui ) are holomorphic (in n variables).

In this case, M is a manifold of complex dimension n and of real dimension 2n.

Interpretation :
Complex manifolds locally look like Cn ∼
= R2n .

Examples of complex manifolds :

1) Cn : non-compact manifold
2) Pn (C) = CPn−1 , the complex projective space : compact manifold
3) Complex tori (see figure 1); they can for example be constructed as a quotient of the type

T = C / L where L is a lattice over Z2

4) If K = C, algebraic varieties ”without singularities” are non-compact complex manifolds.

5) But not all complex manifolds are of algebraic type.
It turns out that compact and non-compact complex manifolds have a very different behaviour, e.g. if a globally
defined function on a compact manifold is globally holomorphic, it must be constant.

Figure 1: the torus T = C / L embedded into R3

Recalls :
− holomorphic functions in 1 variable
− holomorphic functions in n variables for n > 1
− differentiable manifolds
− holomorphic differentiable forms

3
Chapter 1

Holomorphic functions in 1 variable

1.1 Notations and definitions

We know that C = R ⊕ i R ≡ R2 as vector spaces.
Let z ∈ C ⇒ ∃! x, y ∈ R such that z = x + iy ≡ (x, y) where i2 = −1 ⇒ x = Re z, y = Im z , z̄ = x − iy
p
|z| = x2 + y 2 ⇒ |z|2 = x2 + y 2 = (x + iy) · (x − iy) = z · z̄
Let U ⊆ C be an open subset of C. A complex function is a map
f : U ⊆ C → C : z 7→ w = f (z) ⇔ f : U ⊆ R2 → R2 : (x, y) 7→ u(x, y) , v(x, y)

Since C ∼
= R2 , these are equivalent descriptions of complex functions : f (z) ≡ f (x, y) = u(x, y) + i · v(x, y).

Thus any complex function has the partial derivatives :

∂f ∂f ∂u ∂u ∂v ∂v
, ⇔ , , ,
∂z ∂ z̄ ∂x ∂y ∂x ∂y
z+z̄ z−z̄
The relation between the variables is given by x = 2 , y= and defines a change of variables, so
2i

∂f ∂f ∂x ∂f ∂y ∂f 1 ∂f 1 1  ∂f ∂f 
= · + · = · + · = · −i· (1.1)
∂z ∂x ∂z ∂y ∂z ∂x 2 ∂y 2i 2 ∂x ∂y
∂f ∂f ∂x ∂f ∂y ∂f 1 ∂f −1 1  ∂f ∂f 
= · + · = · + · = · +i· (1.2)
∂ z̄ ∂x ∂ z̄ ∂y ∂ z̄ ∂x 2 ∂y 2i 2 ∂x ∂y
because of the chain rule. Notice that these are equations on R2 since f : U → C ⇔ f : U → R2 .

1.1.1 Definition A
Let U ⊆ C be open, f : U → C be a complex function and w ∈ U .
∂f ∂f
Assume that f is differentiable in the real sense with continuous partial derivatives ∂x and ∂y .
∂f
f is called holomorphic at w ∈ U ⇔ ∂ z̄ (w)
= 0.
f is called holomorphic in U if f is holomorphic at all points w ∈ U .

1.1.2 Definition B
Let U ⊆ C be open. A complex function f : U ⊆ C → C is called analytic in U if f has a local series expansion
for all w ∈ U , i.e. ∀ w ∈ U , ∃ ak ∈ C such that
∞
X
f (z) = ak · (z − w)k (1.3)
k=0

and this series converges in a neighborhood of w, e.g. in an open disc around w : D(w, ε) = { z ∈ C , |z −w| < ε }.

4
Complex Manifolds Section 1.3 SCHLICHENMAIER, Leytem

1.1.3 Remark
As a power series, the convergence will always be absolutely and uniformly, i.e.
∞
X ∞
X
ak · (z − w)k < ∞, ∀ z ∈ D(w, ε) ⇒ ak · (z − w)k < ∞, ∀ z ∈ D(w, ε)
k=0 k=0

and ∀ w ∈ U , the convergence of the series is independent of the chosen point z in D(w, ε).
Analytic functions are obviously holomorphic since there is no z̄ in (1.3), thus Def. B ⇒ Def. A.

1.2 The Cauchy-Riemann equations

Let f = u + i · v. Using (1.2), the condition of f being holomorphic (Definition A) can be rewritten as
∂f 1  ∂f ∂f  ∂(u + i · v) ∂(u + i · v)
=0 ⇔ · +i· =0 ⇔ +i· =0
∂ z̄ 2 ∂x ∂y ∂x ∂y
∂u ∂v ∂u ∂v  ∂u ∂v   ∂v ∂u 
⇔ +i· +i· + i2 · =0 ⇔ − +i· + =0
∂x ∂x ∂y ∂y ∂x ∂y ∂x ∂y
∂u ∂v ∂u ∂v
⇔ = and =− (1.4)
∂x ∂y ∂y ∂x
The identities (1.4) are called the Cauchy-Riemann differential equations and are equivalent to Definition A.

1.2.1 Theorem
Let U ⊆ C be open and f : U → C be a complex function such that f is real differentiable, i.e. the partial
derivatives ∂f ∂f
∂x and ∂y exist and are continuous. Then the following conditions are equivalent :
1) f is holomorphic in U .
2) f is analytic in U .
3) f satisfies the Cauchy-Riemann differential equations in U .
None of these equivalences is true for functions in real variables !
Thus holomorphic functions are (as power series) differentiable up to every order, i.e. they are of class C ∞ .

1.2.2 Examples
− holomorphic : polynomial functions P (z), exponential functions exp(z), trigonometric functions sin z, cos z
P (z)
rational functions Q(z) (without poles, otherwise they are called meromorphic)
− not holomorphic : conjugates z̄, modules |z|2 = z z̄

1.3 Cauchy integral formula

1.3.1 Theorem
Let D be an open disc in C and assume that f ∈ C ∞ (D̄) with f holomorphic in D. Then
I
1 f (w)
f (z) = · dw , ∀ z ∈ D
2πi ∂D w − z
where ∂D is traveled in the trigonometric sense.

1.3.2 Generalization
Let U ⊆ C be a simply connected open set in C and f : U → C be holomorphic in U . Let γ be a simple closed
path in U and z ∈ U such that z ∈
/ im γ. If n(z, γ) denotes the winding number of z with respect to γ, then
Z
1 f (w)
n(z, γ) · f (z) = · dw
2πi γ w − z

5
Chapter 2

Holomorphic functions in n variables

2.1 Notations and definitions

∼ R2 n ∼


We know that Cn = = R2n is a real vector space of dimension 2n and an n-dimensional complex vector
n ∼ 2n
space. The isomorphism C = R is given by

(z1 , . . . , zn ) ∈ Cn 7−→ (x1 , y1 , x2 , y2 , . . . , xn , yn ) ∈ R2n with zk = xk + i · yk , ∀ k ∈ {1, . . . , n}

It is not possible to define a scalar product (symmetric, bilinear, positive definite) on Cn . It must be replaced
by the Hermitian product, defined by
n
X
h z, w i := z̄k · wk for z, w ∈ Cn
k=1

This product is linear in the second variable, conjugate-linear in the

p first argument, positive definite and satisfies
h z, w i = h w, z i. Finally the module of z ∈ Cn is equal to kzk = h z, z i, hence
s s s s
n n n
n n
x2k + yk2 = x2k + yk2
P P P P P
kzk = z̄k · zk = |zk |2 =
k=1 k=1 k=1 k=1 k=1

2.1.1 Definitions
Let U ⊆ Cn be open, f : U → C be a complex function and w ∈ U . f is called holomorphic at w ∈ U if f is
continuous at w and if for all k ∈ {1, . . . , n}, the function zk 7→ f (z1 , . . . , zk , . . . , zn ) is holomorphic at wk , i.e. f
is holomorphic in each variable separately.
f is called holomorphic in U if it is holomorphic at every point w ∈ U .
And a function F : U ⊆ Cn → Cm is holomorphic ⇔ every component function Fi : U → C is holomorphic.

The same argument as in the case n = 1 shows that f : U → C is holomorphic in U ⇔ the Cauchy-Riemann
equations are satisfied, i.e. if f = u + i · v for u, v : R2n → R, then
∂f
f holomorphic ⇔ = 0 , ∀ k ∈ {1, . . . , n}
∂ z̄k
∂u ∂v ∂u ∂v
⇔ = and =− , ∀ k ∈ {1, . . . , n}
∂xk ∂yk ∂yk ∂xk

2.2 Power series in n variables

The condition for a complex function in more variables to be analytic in some open set U is more difficult to
formulate since we first need to define multi-index power series.
n
Let z = (z1 , . . . , zn ) ∈ Cn , α = (α1 , . . . , αn ) ∈ Nn0 and aα = aα1 ,...,αn ∈ C. The length of α is |α| :=
P
αi .
k=1

6
Complex Manifolds Section 2.3 SCHLICHENMAIER, Leytem

If z α := z1α1 · z2α2 · . . . · znαn , then a formal power series in n variables can be written as
∞
X ∞ X
X
g(z) = g(z1 , . . . , zn ) = aα · z α = aα · z α (2.1)
α,|α|=0 k=0 |α|=k

is finite since α ∈ Nn0 , i.e. only finitely many α can satisfy |α| = k.
P
where the sum
|α|=k

2.2.1 Example
Let n = 2 and consider only 0 ≤ k ≤ 2, then
2 X
X
aα · z α = a00 z10 z20 + a10 z11 z20 + a01 z10 z21 + a20 z12 z20 + a11 z11 z21 + a02 z10 z22
k=0 |α|=k = a00 + a10 z1 + a01 z2 + a20 z12 + a11 z1 z2 + a02 z22
Saying that the power series is formal just means that we do not worry about convergence problems for the
moment. If all aα are zero except finitely many, we obtain a usual polynomial in n variables. And for polynomials
we can plug in values for z = (z1 , . . . , zn ) since the sum is finite, thus always converges.

2.2.2 Problem
∞
aα · (z − w)α .
P
Consider power series of the form
|α|=0
Recall that convergence of a power series in 1 variable is defined as the convergence of the sequence of partial
sums. This definition is however not possible if n > 1 since there is natural linear ordering defined on Nn0 ; the
”partial sums” in (2.1) are not indexed, thus there is no ”sequence” which could converge in the usual sense.

2.2.3 Definition : convergence for n > 1

Let z ∈ Cn . ∞
aα · z α converges to c ∈ C ⇔ ∀ ε > 0, there exist a finite index set I0 ⊂ Nn0 such that
P
The power series
|α|=0
X
for all finite set I satisfying I0 ⊆ I ⊂ Nn0 , we have : aα · z α − c < ε
α∈I
In this case we denote c :=Plim aα z α and this
P
limit c is always unique. Moreover this convergence is an
aα z α → c, then |aα | z α also converges, but not necessarily to |c|.
P
absolute convergence, i.e. if

2.3 Analyticity
2.3.1 Definition
Let U 6= ∅ be open in Cn and f : U → C be a complex function on U .
f is called complex analytic in U ⇔ ∀ w ∈ U , there exist a neighborhood Uw ⊆ U of w and a power series
∞
X
aα · (z − w)α where (z − w)α = (z1 − w1 )α1 · (z2 − w2 )α2 · . . . · (zn − wn )αn
|α|=0

which converges to f (z) for all z ∈ Uw . This means that f can locally be written at any point as a power series
which converges in a certain neighborhood of this point.

f is called real analytic in U ⇔ on U , it can locally be expanded as a power series in z and z̄.
Obviously : f complex analytic ⇒ f real analytic.

2.3.2 Examples
• f (z) = z 2 : f is a power series as polynomial ; it is complex analytic since it only depends on z
• g(z) = |z|2 = z z̄ : g is real analytic, but not complex analytic

7
Complex Manifolds Section 2.5 SCHLICHENMAIER, Leytem

2.3.3 Theorem
Let f be a complex function defined on some open set U in Cn .
Then f is holomorphic in U ⇔ f is complex analytic in U .

2.4 Identity Theorem

Let A ⊆ C and a ∈ C. We say that a is an accumulation point of A if any open set U ⊆ C containing a intersects
A in some point distinct than a. By taking U = D(a, n1 ) with n → +∞, this is equivalent to :

a accumulation point of A ⇔ ∃ sequence (an )n with an ∈ A, an 6= a, ∀ n, such that lim an = a

n→+∞

In particular, the condition an 6= a, ∀ n implies that A must necessarily be infinite.

If B ⊆ C, we say that A has an accumulation point in B if ∃ z0 ∈ B such that z0 in an accumulation point of A.

2.4.1 Theorem (n = 1)
Let U ⊆ C be open, connected and U 6= ∅ and let M ⊆ U be such that M has an accumulation point in U .
Let f, g : U → C be holomorphic in U such that f|M = g|M . Then f = g on U .

Example :

If 2 globally holomorphic functions f, g : C → C are equal on M = { n1 | n ∈ N }, then they have to coincide

everywhere on C as well.

2.4.2 Theorem (n > 1)

Let U ⊆ Cn be open, connected and U 6= ∅. Let W ⊆ U be such that W 6= ∅ and W is open.

If f, g : U → C are holomorphic in U such that f|W = g|W , then f = g on U .

2.4.3 Remark
The condition ”W ⊆ U open” is stronger than the condition ”M ⊆ U has an accumulation point in U ”.
Indeed : ∀ w ∈ W , ∃ εw > 0 such that w ∈ D(w, εw ) ⊂ W ⊆ U since W is open. Thus one can define a sequence
of distinct points in this open disc converging to the center w ⇒ w is an accumulation point of W in U .
In fact, the condition about having an accumulation point is not sufficient for n > 1 :

Consider C2 with M = (z1 , z2 ) ∈ C2 | z2 = 0 ⇒ every point in M is an accumulation point of M in C2 .
Let g, h : C2 → C be holomorphic functions on C2 such that g 6= h outside of M (this always exists) and define

f1 (z1 , z2 ) := z2 · g(z1 , z2 ) , f2 (z1 , z2 ) := z2 · h(z1 , z2 )

Then f1 , f2 are holomorphic on C2 with f1 = f2 = 0 on M , but f1 6= f2 on C2 since g 6= h.

2.4.4 Identity Theorem for power series

α
bα z α represent the same holomorphic function, then aα = bα , ∀ α ∈ Nn0 .
P P
If aα z and
Thus : if ∃ α0 ∈ Nn0 such that aα0 6= bα0 , then the corresponding power series define different holomorphic maps.

8
Complex Manifolds Section 2.6 SCHLICHENMAIER, Leytem

2.5 Maximum Principle

2.5.1 Recall
Let V ⊆ Rm be open. For a function h : V → C of real variables x1 , . . . , xm , the Laplacian is defined as
m
X ∂2h ∂2h ∂2h ∂2h
∆h := = + + ... +
i=1
∂x2i 2
∂x1 2
∂x2 ∂x2m

h is called harmonic on V if ∆h(x) = 0, ∀ x ∈ V .

Let now f : U → C be a holomorphic function on an open set U ⊆ Cn and consider the identification Cn ∼ = R2n ,
i.e. we identify f (z1 , . . . , zn ) ≡ f (x1 , y1 , . . . , xn , yn ). Then f , Re f and Im f are harmonic on U (as functions of
2n real variables).

2.5.2 Theorem
Let U 6= ∅ be open in Cn and f : U → C be holomorphic in U . Then the absolute value function
q
|f | : U → R : z 7→ |f (z)| = f (z) · f (z)

is in general not holomorphic or harmonic (unless f is constant), but |f | is continuous and real analytic in U .
Moreover |f | does not have a maximum in U .
If U is bounded, this means that its maximum ”lies on the boundary” of U , given by ∂U := U ∩ (Cn \ U ).

2.5.3 Example
Consider the case of an open disc U = D(z0 , r) with center z0 ∈ Cn and radius r > 0 as in figure 2.1.
If |f | is continuous on the compact set Ū = D̄(z0 , r), we already know that it must have a maximum in Ū . The
theorem says that this maximum has to be on the boundary ∂U of Ū .

Figure 2.1: the open disc U = D(z0 , r) with boundary ∂U

2.6 Hartog’s Lemma

2.6.1 Recalls
n = 1 : Let U ⊆ C be open and convex with p ∈ U .
If f : U → C is holomorphic in U \ {p} and continuous in U (in particular at p), then f is holomorphic in U .
If f is holomorphic in U \ {p} and bounded in a neighborhood of p (without being assumed to be defined or
continuous at p), then f extends uniquely to a holomorphic function on U , hence f does not have a pole at p.

2.6.2 Definition
Let w = (w1 , w2 , . . . , wn ) ∈ Cn and r ∈ R, r > 0.
A polydisc of radius r centered at w, denoted ∆(w, r), is equal to the Cartesian product of discs

∆(w, r) = D(w1 , r) × D(w2 , r) × . . . × D(wn , r)

= (z1 , z2 , . . . , zn ) ∈ Cn , |z1 − w1 | < r, |z2 − w2 | < r, . . . , |zn − wn | < r



It should not be confused with the usual disc D(w, r) = z ∈ Cn , kz − wk < r ⊆ Cn .

9
Complex Manifolds Section 2.6 SCHLICHENMAIER, Leytem

2.6.3 Hartog’s Lemma

Let n ≥ 2, U ⊆ Cn be open and p ∈ U . Assume that f is holomorphic in U \ {p} (see figure 2.2). Then f extends
uniquely to a holomorphic function on U (even if f may a priori not be continuous or bounded at p).

Figure 2.2: f is not holomorphic at p

a
Proof. The extension is necessarily unique because U \ {p} is open in U . If f has 2 holomorphic extensions, these
will coincide on U \ {p} (as extensions of f ), thus by the Identity Theorem 2.4.2 they are also equal on U .

It suffices to prove the statement for n = 2 and for p = (0, 0) ∈ C2 by applying a shift (which is a holo-
morphic transformation).
Moreover since we work locally around p, it suffices to extend f to a holomorphic function in a small open
neighborhood V of p and we may assume that V is a polydisc centered at p of small enough radius. Hence


p = (0, 0) ∈ V ⇒ ∃ r > 0 such that V = ∆ (0, 0), r ⊂ U

f is well-defined and holomorphic on U \ (0, 0) , in particular it is holomorphic on the boundary of V . Define
Z
1 f (z1 , w2 )
F (z1 , z2 ) := · dw2 , ∀ (z1 , z2 ) ∈ V
2πi |w2 |=r w2 − z2

where z1 is a parameter in the integral. F is well-defined since |w2 | = r > 0 ⇒ (z1 , w2 ) 6= (0, 0).
Moreover (z1 , z2 ) ∈ V ⇒ |z2 | < r, hence w2 − z2 6= 0, so numerator and denominator are well-defined.

Thus ∀ w2 ∈ C such that |w2 | = r, the map

f (z1 , w2 )
(z1 , z2 ) 7−→ g(z1 , z2 , w2 ) :=
w2 − z2
is holomorphic in V . It remains to check that F is holomorphic in V as well. Using the complex version of
differentiation of parameter-dependent integrals, we find

z1 7→ g(z1 , z2 , w2 ) is holomorphic in D(0, r), ∀ z2 , w2 such that |z2 | < r, |w2 | = r

z2 7→ g(z1 , z2 , w2 ) is holomorphic in D(0, r), ∀ z1 , w2 such that |z1 | < r, |w2 | = r
w2 7→ g(z1 , z2 , w2 ) is continuous on ∂D(0, r), ∀ z1 , z2 such that |z1 | < r, |z2 | < r
∂g ∂g ∂g ∂g
Moreover the partial derivatives ∂z1 , ∂z2 , ∂ z̄1 and ∂ z̄2 are continuous in all variables (whenever defined), hence

∂f
∂ z̄1 (z1 , w2 )
Z   Z
∂F 1 ∂ f (z1 , w2 ) 1
(z1 , z2 ) = · dw2 = · dw2 = 0
∂ z̄1 2πi |w2 |=r ∂ z̄1 w2 − z2 2πi |w2 |=r w2 − z2
Z   Z  
∂F 1 ∂ f (z1 , w2 ) 1 ∂ 1
(z1 , z2 ) = · dw2 = · f (z1 , w2 ) · dw2 = 0
∂ z̄2 2πi |w2 |=r ∂ z̄2 w2 − z2 2πi |w2 |=r ∂ z̄2 w2 − z2
1
since z2 7→ w2 −z2 is holomorphic ∀ w2 ∈ ∂D(0, r) and f is holomorphic in all variables with (z1 , w2 ) 6= (0, 0).

Finally F is holomorphic in both variables z1 and z2 in the whole polydisc, i.e. we constructed a function
which is holomorphic in V . In particular we can plug in
Z
1 f (0, w2 )
F (0, 0) = · dw2 ∈ C
2πi |w2 |=r w2

10
Complex Manifolds Section 2.6 SCHLICHENMAIER, Leytem

To show that F is an extension of f on V , we need that F|V \{p} = f|V \{p} . Fix some r1 ∈ ]0, r[ and define

V 0 := V ∩ (z1 , z2 ) ∈ C2 , |z1 | > r1



In order to use Cauchy’s integral formula, we have to exclude z1 = 0, otherwise the map w2 7→ f (0, w2 ) is not
holomorphic in D̄(0, r). Hence ∀ (z1 , z2 ) ∈ V 0 :
Z
1 f (z1 , w2 )
F (z1 , z2 ) = · dw2 = f (z1 , z2 ) with (z1 , z2 ) 6= (0, 0)
2πi |w2 |=r w2 − z2

So F = f on V 0 with F and f holomorphic in V \ {p}. But V 0 ⊂ V \ {p} is open, hence by the Identity Theorem
F = f on V \ {p}. So F is the required holomorphic extension of f on V , thus on U .

2.6.4 Remark
As a consequence of Hartog’s lemma we obtain that the singularities of holomorphic functions in several variables
cannot be isolated points (since it is possible to extend the function in this case).
It can even be shown that such an extension exists if f is holomorphic in some polydisc ∆1 without knowing its
behaviour in a smaller polydisc ∆2 and ∆1 \ ∆2 is still connected.

In this case f can also be uniquely extended to a holomorphic function in the whole big polydisc ∆1 .
This does not hold for arbitrary sets where a function is not defined. Let e.g. n = 2 and consider f (z1 , z2 ) = z11 .
f is defined and holomorphic on C \ N where N = { (0, z2 ) | z2 ∈ C }. Since N has however empty interior, the
set of singularities N does not contain any polydisc and the remark does not apply in this case.

2.6.5 Counter–examples
The condition n ≥ 2 is necessary.
Consider e.g. the function z 7→ z1 on U = C with p = 0. It cannot be extended to C since it is not bounded at 0.

For n ≥ 2, it is also important that the function is holomorphic, e.g. if n = 2, the map
1 1
z = (z1 , z2 ) 7−→ 2
=
kzk |z1 | + |z2 |2
2

is defined on Cn \ {0} but, since it is not holomorphic, cannot be extended to a holomorphic function on Cn .

11
Chapter 3

Real differentiable manifolds

3.1 Introduction
intuitive idea : Real manifolds locally look like Rn .
There are 2 possible approaches in order to define manifolds :

– start with an arbitrary set M , define charts domains and coordinate maps ϕi such that the transition
functions satisfy certain properties and construct a topology on M using the atlas of coordinate maps
– start with a given topological space (M, T ) and a collection of open sets Ui and homeomorphisms ϕi

We choose the second approach in the sequel.

Let (M, T ) be a topological space and {Ui }i∈J be a family of non-empty open subsets of M such that
[
M= Ui
i∈J

i.e. the Ui are an open covering of the space M . Let also be given a collection of maps (see figure 3.1)

∀ i ∈ J, ϕi : Ui → Wi ⊆ Rn

where every ϕi is a topological isomorphism : ϕi and ϕ−1

i are both bijective and continuous (one also says that
the ϕi are bicontinuous or homeomorphisms). Hence all Wi are open in Rn and the transition maps

ψji := ϕj ◦ ϕ−1
i : ϕi (Ui ∩ Uj ) ⊆ Rn −→ ϕj (Uj ∩ Ui ) ⊆ Rn

are also homeomorphisms (whenever defined).

Figure 3.1: two overlapping coordinate patches

A pair (Ui , ϕi ) is called a coordinate patch of M and the whole collection (Ui , ϕi )i∈J is called an atlas of M .
Since the transition maps ψji are all continuous, one says that M is a topological manifold.

12
Complex Manifolds Section 3.2 SCHLICHENMAIER, Leytem

3.1.1 Definition
A topological manifold M is called a (real) differentiable manifold if the transition maps ψji are differentiable,
i.e. infinitely often differentiable (of class C ∞ ), ∀ (i, j) ∈ J × J.
−1
Since ψji = ψij , this immediately implies that the transition functions are diffeomorphisms.
One can also impose weaker conditions, e.g. the ψji should only be C 1 , C 2 . . ., or stronger conditions, for instance
that the ψji have to be real analytic. In this case M would be a C k –manifold, resp. a real analytic manifold.
A topological manifold is of dimension n if ϕi : Ui → Wi ⊆ Rn , ∀ i ∈ J.

3.2 Topological properties of manifolds

In order to exclude ”exotic” manifold structures, one often requires that :
− M is Hausdorff
− M is connected
− M is paracompact
Unless explicitly mentioned, we always assume that these 3 conditions are satisfied in a differentiable manifold.

3.2.1 Definitions
Let (M, T ) be a topological space.
M is called Hausdorff if any 2 distinct points in M can be separated, i.e.

∀ x, y ∈ M, x 6= y : ∃ Ux , Uy open neighborhoods of x, y such that Ux ∩ Uy = ∅

M is called regular if any point and a closed set not containing that point can be separated, i.e.

∀ x ∈ M, ∀ F ⊆ M closed such that x ∈

/ F : ∃ Ux , VF open such that x ∈ Ux , F ⊆ VF , Ux ∩ VF = ∅

M is called normal if any 2 disjoint closed sets can be separated by open neighborhoods.
Let x ∈ M . A basis of neighborhoods Bx of x is a family of neighborhoods of x such that for any neighborhood
Ux of x, there is a neighborhood Vx ∈ Bx such that x ∈ Vx ⊆ Ux . A basis B of M is a collection of open sets
such that any open set in M can be written as a union of open sets from B.
M is called first-countable if every point in M has a countable basis of neighborhoods. M is called second-
countable if it has a countable basis.

M is called connected ⇔ (M = U ∪ V for some open sets U, V with U ∩ V = ∅ ⇒ U = ∅ or V = ∅),

i.e. M is connected if it cannot be written as a disjoint union of 2 non-empty open subsets. An open set U ⊆ M
is connected if it is connected with respect to the induced topology.
M is said to be path-connected if ∀ p, q ∈ M , there is a continuous map γ : [0, 1] → M such that γ(0) = p and
γ(1) = q (see figure 3.2). Such a map γ is called a path from p to q.
M is called locally connected if every point in M admits a basis of connected neighborhoods and it is locally
path-connected if every point in M admits a basis of path-connected neighborhoods.

Figure 3.2: γ is a path from p to q

Let now U ⊆ T be a family of open subsets of M .

U is called pointwise finite if ∀ x ∈ M , x only belongs to a finite number of sets in U. U is called locally finite if
∀ x ∈ M , there is a neighborhood Vx of x such that Vx intersects only a finite number of sets in U.
Assume now that U is an open covering of M . Another family of subsets V ⊆ T is called a refinement of U if it
is again an open covering of M and if ∀ V ∈ V, ∃ U ∈ U such that V ⊆ U .

13
Complex Manifolds Section 3.2 SCHLICHENMAIER, Leytem

M is called a Lindelöf space if every open covering of M has a countable subcover. M is called compact if every
open covering of M contains a finite subcover.
M is called locally compact if every point in M admits a compact neighborhood. And finally M is called
paracompact if it is Hausdorff and if every open covering of M has a locally finite refinement.
We also recall that continuous images of compact sets are again compact, i.e.

K ⊂ M compact, f : M → N continuous ⇒ f (K) ⊂ N compact

3.2.2 Results
Let (M, T ) be a topological space. One can show the following properties, which we are not going to prove :
− M compact ⇒ M paracompact (since any finite covering is a locally finite covering)
− M Hausdorff, second-countable and locally compact ⇒ M paracompact
− M second-countable ⇒ M first-countable and Lindelöf
− M regular and Lindelöf ⇒ M paracompact
− M paracompact and Hausdorff ⇒ M regular and normal
− M path-connected ⇒ M connected
− M connected and locally path-connected ⇒ M path-connected
− If M is locally path-connected, then connectedness and path-connectedness are equivalent.
− Any Hausdorff and second-countable manifold is locally compact, locally connected and locally path-connected.

Conclusion :
Any topological manifold which is Hausdorff and second-countable is also paracompact. Moreover connectedness
and path-connectedness coincide for differentiable manifolds of our purpose (see section 3.2).
The fact that we consider connected differentiable manifolds M ensures that the dimension of M is well-defined.
Otherwise we have to consider each connected component of M separately.

3.2.3 Example
 S
Let M be a manifold and U = Ui i∈I be an open covering : M = i∈I Ui .
 S
Again, a refinement of U is an open covering V = Vj j∈J such that M = j∈J Vj and

∀ j ∈ J, ∃ i ∈ I (not necessarily unique) such that Vj ⊆ Ui

Being a Hausdorff and second-countable manifold, we can say that R is paracompact. As an example, let
[
In = ] − n, n[ ⇒ R = In : open covering (see figure 3.3)
n∈N
T
But this covering is not locally finite because n∈N In = {0} : 0 ∈ In , ∀ n ∈ N, hence also every neighborhood of
0 intersects infinitely many In .
Figure 3.3: the intervals In cover R

As long as we have infinitely many In necessary to cover R, the cover is not locally finite at any point : every
point of R belongs to infinitely many In . But we need infinitely many to cover R. So consider

In− := ] − n, −n + 2[ ⊆ In , In+ := ]n − 2, n[ ⊆ In

⇒ { In | n ∈ N } ∪ { In+ , In− | n ≥ 3 } is again a covering and a refinement of the previous one (even if it contains
more sets than before, see definition). And now one can take the subcover (see figure 3.4)

I1 , I2 , In+ , In− | n ≥ 3


14
Complex Manifolds Section 3.3 SCHLICHENMAIER, Leytem

It still covers R and is locally finite at 0 since 0 ∈ I1 , I2 only. And it is also locally finite at any other point of R
since the intervals In+ , In− ”drift off” to ± ∞.

Figure 3.4: a locally finite covering of R

Note that this is not a proof of the fact that R is paracompact. In order to show this we need to show that every
open cover of R has a locally finite refinement.

3.3 Differentiable functions

3.3.1 Partition of unity
Let M be a topological space and U = {Ui }i∈J be an arbitrary open covering of M . A family of continuous
functions {τi }i∈J where τi : M → [0, 1], ∀ i ∈ J is called a partition of unity if
1) {τi }i∈J is locally finite, i.e. ∀ x ∈ M , there is an open neighborhood W of x such that τi|W = 0 for all but at
most finitely many i ∈ J
P
2) ∀ x ∈ M : τi (x) = 1, which makes sense by 1) since it is a finite sum whenever x is given
i∈J
A partition of unity is called subordinated to U if supp τi = { x ∈ M | τi (x) 6= 0 } ⊆ Ui , ∀ i ∈ J.

Partitions of unity can be used for ”gluing” local objects.

Let again M be a manifold and U = {Ui }i∈J be an open covering of M . Let ψi be local objects defined on Ui
(e.g. local functions) such that ψi = ψj on Ui ∩ Uj . If there exist a partition of unity τi : M → [0, 1] subordinated
to U, one can define a global object ψ by setting
X
ψ(x) := τj (x) · ψj (x)
j∈J
P P
Then ψ|Ui = ψi since x ∈ Ui ⇒ ψj (x) = ψi (x), ∀ j, thus ψ(x) = τj (x) · ψi (x) = ψi (x) · τj (x) = ψi (x) · 1.
j∈J j∈J

3.3.2 Theorem
Let M be a Haudorff topological space. Then
M is paracompact ⇔ for any open covering of M there exist a partition of unity subordinated to this covering.

3.3.3 Definition
Let M be a differentiable manifold with atlas (Ui , ϕi )i∈I . A map f : M → R is called a differentiable function if
and only if fi := f ◦ ϕ−1
i : Wi ⊆ Rn → R is a differentiable map, ∀ i ∈ I (see figure 3.5).

Strictly speaking, this condition is not really needed. It already suffices the local condition : ∀ x ∈ M , ∃ U ⊂ M
open neighborhood of x and ∃ (U, ϕ) chart of M at x such that f ◦ ϕ−1 : ϕ(U ) ⊆ Rn → R is differentiable.

3.3.4 Bump functions

Proposition : Every differentiable manifold has a differentiable partition of unity, called the bump functions.
(This does not mean that every partition of unity in a differentiable manifold is differentiable.)

15
Complex Manifolds Section 3.4 SCHLICHENMAIER, Leytem

Figure 3.5: composition of the functions f and ϕ−1

i to give a map fi : Rn → R

Let x ∈ M and U be an open neighborhood of x. A bump function γ is a smooth map γ : M → [0, 1] with
support contained in U and taking value 1 in a neighborhood V of x. It can be visualized as in figure 3.6.

Figure 3.6: a bump function with support in U

But there does not necessarily exist an analytic partition of unity since these bump functions are constructed by
using manipulations and transformations of the map
(
e−1/x if x > 0
g(x) = ⇒ g (n) (0) = 0, ∀ n ∈ N
0 if x ≤ 0
Consider small neighborhoods of the points a, b, c, d in figure 3.7 where the bump function γ begins to ”go
up” and ”go down”. All the derivatives vanish at these points, so the Taylor series of γ will be constant in
any neighborhood of them. Thus γ cannot be analytic since it does not coincide with its Taylor series in a
neighborhood of these points.
More generally, bump functions have compact support, but compactly supported functions are never analytic.
Figure 3.7: bump functions are not real analytic

3.4 Orientability of differentiable manifolds

Consider figure 3.8. Let M be a differentiable manifold with atlas (Ui , ϕi )i∈I . Since all ϕi are bijective, we have
∀ x ∈ Wi = ϕi (Ui ), ∃! a ∈ Ui such that ϕi (a) = x = (x1 , . . . , xn )
The n-tuple (x1 , . . . , xn ) represents the coordinates of a ∈ Ui in the considered chart or coordinate system.
These coordinates can however change by using another coordinate chart. If ∃ j 6= i such that a ∈ Ui ∩ Uj , then
ϕj (a) = y = (y1 , . . . , yn ) ∈ Wj = ϕj (Uj ) ⊆ Rn as well. The coordinates y and x are then related by the relation
y = ψji (x) = (ϕj ◦ ϕ−1
i )(x)

16
Complex Manifolds Section 3.4 SCHLICHENMAIER, Leytem

Figure 3.8: coordinate change in 2 overlapping charts

The ψji are thus also called transition functions or coordinate change maps.
y being a function of x, one can now consider the Jacobian matrix
 ∂y1 ∂y1 ∂y1   ∂y1 ∂y1 ∂y1 
  ∂x1 ∂x2 . . . ∂x n ∂x1 (x) ∂x2 (x) . . . ∂xn (x)
∂yk
J = J(ψji ) = =  ... .. .. ..  =  .. .. .. ..
 = J(x)
 
∂xl k,l . . .   . . . .
∂yn ∂yn ∂yn ∂yn ∂yn ∂yn
∂x1 ∂x2 . . . ∂xn ∂x1 (x) ∂x2 (x) . . . ∂xn (x)

J is a matrix depending on the point x ∈ ϕi (Ui ∩ Uj ) ⊆ Rn where
it is computed.

Since the transition maps ψji are bijective, we have that det J(x) 6= 0, ∀ x ∈ ϕi (Ui ∩ Uj ).

3.4.1 Definition


A manifold is called orientable ⇔ it admits an atlas (Ui , ϕi )i∈I such that det J(ψji ) > 0, ∀ i, j ∈ I.
Note that this is an inequality of functions, i.e. for given i, j the relation must hold for any point x ∈ ϕi (Ui ∩ Uj ).

Fact :
Complex manifolds are always orientable (proof, see section 4.2.3).
Examples of non-orientable manifolds are the Möbius strip and the real projective plane RP2 . Hence by the
above fact it is not possible to endow these manifolds with a complex manifold structure.

3.4.2 Formulation with differential forms

Consider the local coordinates x1 , x2 , . . . , xn with the standard orientation (order) and let y1 , y2 , . . . , yn be the
coordinates after the coordinate change. In order to determine the orientation of the coordinate change, we have
to compare the orientation of dx1 ∧ dx2 ∧ . . . ∧ dxn and dy1 ∧ dy2 ∧ . . . ∧ dyn . We have :
n n
∗
X ∂(ψji )k X ∂yk
∀ k ∈ {1, . . . , n} : ψji (dyk ) = · dxl = · dxl
∂xl ∂xl
l=1 l=1
∗


⇒ ψji (dy1 ∧ dy2 ∧ . . . ∧ dyn ) = det J(ψji ) · dx1 ∧ dx2 ∧ . . . ∧ dxn (3.1)
∗
where ψji denotes the pull-back of a differential form under the diffeomorphism ψji .


Hence by (3.1) the orientation of both differential forms coincide ⇔ det J(ψji ) > 0.

3.4.3 Example
Consider the n-dimensional sphere S n given by

S n = x = (x1 , x2 , . . . , xn+1 ) ∈ Rn+1 , kxk2 = 1 ⊂ Rn+1



S n is compact in Rn+1 as it is closed and bounded, hence it is paracompact. Moreover Rn+1 is Hausdorff and
second-countable, thus S n is also Hausdorff and second-countable since these properties are hereditary.

17
Complex Manifolds Section 3.4 SCHLICHENMAIER, Leytem

We want to endow S n with a differentiable manifold structure. For this, consider the hemispheres

U(k,j) := x ∈ S n | (−1)j · xk > 0 , j = 0, 1, k = 1, 2, . . . , n + 1



which are open in Rn+1 and in S n . They form an open covering of S n (the case n = 1 is given in figure 3.9),
called the standard covering. As charts, we take

h(k,j) : U(k,j) −→ B n (0, 1) ⊂ Rn : (x1 , . . . , xk , . . . , xn+1 ) 7−→ (x1 , . . . , xk−1 , xk+1 , . . . , xn+1 )

This just corresponds to dropping the k th position and hence projecting the point on the sphere down into the
ball B n (0, 1) = { y ∈ Rn , kyk2 < 1 }. Intuitively, this projection corresponds to a flattening of the sphere.

Moreover h(k,j) is well-defined because

n+1
X n+1
X
x2l < x2l = 1 ⇒ (x1 , . . . , xk−1 , xk+1 , . . . , xn+1 ) ∈ B n (0, 1)
l=1 l=1
l6=k

as xk > 0 for j = 0 or xk < 0 for j = 1. h(k,j) is in addition bijective and its inverse map is equal to
s
 n 
−1 n j
yi2 , yk , yk+1 , . . . , yn
P
h(k,j) : B (0, 1) −→ U(k,j) : (y1 , . . . , yk , . . . , yn ) 7−→ y1 , . . . , yk−1 , (−1) · 1 −
i=1

3.4.4 Exercise
Show that S n is a differentiable and orientable real manifold.

Figure 3.9: the standard covering of S 1 together with a positive orientation

The domains U(k,j) form a covering of S n and all images are open. It remains to check that ψ = h(k,j) ◦ h−1
(l,m) is
differentiable.
Observe that the case k = l is trivial since either j = m ⇒ ψ = id or j 6= m ⇒ dom ψ = ∅ :
 
dom ψ = h(l,m) U(l,m) ∩ U(k,j) = h(l,m) x ∈ S n | (−1)j · xk > 0 and (−1)m · xl > 0


(
x ∈ Rn | (−1)j · xk > 0 if k < l
=  n j
x ∈ R | (−1) · xk−1 > 0 if k > l
s
 n 
⇒ ψ(x1 , . . . , xn ) = h(k,j) x1 , . . . , xl−1 , (−1)m · 1 − x2i , xl , xl+1 , . . . , xn
P
i=1
 p 
 x1 , . . . , xk−1 , xk+1 , . . . , xl−1 , (−1)m · 1 − x2i , xl , xl+1 , . . . , xn
P
=   (3.2)
 x , . . . , x , (−1)m · p1 − P x2 , x , x , . . . , x
1 l−1 i l l+1 k−2 , xk , xk+1 , . . . xn

18
Complex Manifolds Section 3.4 SCHLICHENMAIER, Leytem

The explicit expressions in (3.2) allow to conclude that ψ, if it is defined, is differentiable and even real analytic
(since identity, square and square root are analytic). This atlas does however not (yet) satisfy the orientability
condition. But it can nevertheless be used to find an appropriate atlas later on. We set
s
(−1)m · (−2xu ) (−1)m+1 · xu
 n

∂ m
x2i
P
yu := (−1) · 1 − = p P 2 = p , ∀ u ∈ {1, . . . , n}
∂xu 1 − x2i
P
i=1 2 1 − xi

Consider the case k > l (k < l is similar). Then the Jacobian matrix J(ψ) is given by
 
1 1
..  .. 
. 
 . 

l−1   1 

l y1 . . . yl−1 yl . . . yk−2 yk−1 yk
 ... yn 

l+1   1 

..  .. 
. 
 . 

k−1   1 

k 
 0 1 

..  .. 
.  . 
n 1
y1 yl−1 yl yk−2 yk yk+1 yn yk−1
1 0
1 0
1 0
⇒ det J(ψ) = (−1)l−1 · (−1)n−(k−1) ·


1 0
1 0
1 0
1 0

and expanding with respect to the last column gives

xk−1
det J(ψ) = (−1)n−k+l · (−1)n+1 · yk−1 = (−1)2n−k+l+1 · (−1)m+1 · p


1 − x2i
P


  
Since (−1)j · xk−1 > 0 and 2n is even, we obtain that sign det J(ψ) = sign (−1)l−k+m+j .

In general this is not always positive. We admit that the factor (−1)l−k+m+j can be eliminated by adding
some powers of (−1) in the definition of h(k,j) such that the orientation is preserved (see figure 3.9). This
modification will finally define an orientation-preserving atlas for S n .

19
Chapter 4

Complex manifolds

4.1 Definition
Consider figure 4.1. Let (M, U) be a real differentiable manifold with atlas U = (Ui , ϕi )i∈J of (real) dimension
2n and assume that M is connected (as topological space) :
ϕi : Ui → Wi ⊆ R2n , ψji : ϕi (Ui ∩ Uj ) ⊆ R2n → ϕj (Uj ∩ Ui ) ⊆ R2n
where Wi ⊆ R2n is open and ψji is differentiable whenever defined.
We identify Cn with R2n using the standard identification (which is compatible with the orientation of M ) :
(z1 , z2 , . . . , zn ) ←→ (Re z1 , Im z1 , Re z2 , Im z2 , . . . , Re zn , Im zn ) = (x1 , y1 , x2 , y2 , . . . , xn , yn )
Denote Uji := ϕi (Ui ∩ Uj ) ; then ψji : Uji ⊆ Cn → Cn becomes a complex map on the open set Uji ⊆ Cn .
Figure 4.1: a differentiable manifold of real dimension 2n

Such a manifold (M, U) is called a complex manifold ⇔ the maps ψji are biholomorphic ∀ i, j ∈ J.
−1
Note that it is enough to require that the ψji are holomorphic only because ψji = ψij :
ψji , ψij holomorphic ⇒ ψji , ψij biholomorphic
As for differentiable manifolds, we always assume in the following that complex manifolds are connected.
A one-dimensional connected compact complex manifold is called a Riemann surface.

Remark :
Two different atlases can define the same manifold if they are compatible. Recall that a chart (U, ϕ) is said to
be compatible with an atlas U = (Ui , ϕi )i∈I if and only if
− the sets ϕi (U ∩ Ui ) and ϕ(U ∩ Ui ) are open in Cn , ∀ i ∈ I
− ∀ i ∈ I, the functions ϕi ◦ ϕ−1 and ϕ ◦ ϕ−1i are holomorphic whenever defined
Moreover 2 atlases U and V are compatible if every chart of U is compatible with V and vice-versa.
One can then show that adding charts to a manifold which are compatible with the existing atlas will not change
the structure of the manifold. This is why manifolds are in general equipped with an equivalence class of atlases.

20
Complex Manifolds Section 4.2 SCHLICHENMAIER, Leytem

4.2 Complex manifolds are orientable

In the sequel, we use the following short-hand notation. For a function

f : U ⊆ Cn → Cm : (z1 , . . . , zn ) 7−→ f1 (z1 , . . . , zn ) , f2 (z1 , . . . , zn ) , . . . , fm (z1 , . . . , zn )

 
∂f ∂fi
:= ∈ Mat(m × n, C)
∂z ∂zj i,j

4.2.1 Definitions
Fix the indices i, j ∈ J. The holomorphic map ψji : Uji ⊆ Cn → Cn : z 7→ w = ψji (z) has the n components

(ψji )k : Uji ⊆ Cn → C , k = 1, . . . , n

∂wk ∂(ψji )k
All components are holomorphic too, thus (z) = (z) = 0, ∀ k, l ∈ {1, . . . , n}, ∀ z ∈ Uji .
∂ z̄l ∂ z̄l
For complex functions, one can now define 3 different types of Jacobian matrices :
 
∂w ∂wk
holomorphic Jacobian matrix : Jhol (z) = (z) = (z) ∈ Mat(n × n, C)
∂z ∂zl k,l
∂w ∂w
! ∂w ∂w
!
∂z (z) ∂ z̄ (z) ∂z ∂ z̄
complex Jacobian matrix : J(z) = ∂ w̄ ∂ w̄
= ∂ w̄ ∂ w̄ (z) ∈ Mat(2n × 2n, C)
∂z (z) ∂ z̄ (z) ∂z ∂ z̄
 ∂(Re w ) ∂(Re w ) ∂(Re w ) 
∂x1
1
∂y1
1
∂x2
1
. . . ∂(Re
∂yn
w1 )
 
 ∂(Im w1 ) ∂(Im w1 ) ∂(Im w1 ) . . . ∂(Im w1 ) 
 ∂x1 ∂y1 ∂x2 ∂yn 
 ∂(Re w ) ∂(Re w ) ∂(Re w ) ∂(Re w2 ) 

real Jacobian matrix : Jreal (z) =   (z) ∈ Mat(2n × 2n, R)
2 2 2
 ∂x1 ∂y1 ∂x2 ... ∂yn
 .
.. .
.. .
.. .. .
..


 . 

∂(Im wn ) ∂(Im wn ) ∂(Im wn ) ∂(Im wn )
∂x1 ∂y1 ∂x2 . . . ∂yn

J can be seen as the complexified version of Jreal .

M is now called orientable ⇔ there is an atlas of M such that det Jreal (z) > 0, ∀ z ∈ Uji .
Since w = ψji is bijective, all 3 determinants are necessarily non-zero. Obviously J 6= Jreal (the first one is
complex, the second one real only), but there is a relation between their determinants.

4.2.2 Lemma
For any holomorphic function f : U ⊆ C → C, we have f (z) = f (z̄), ∀ z ∈ U .
Proof. Let z0 ∈ U . Since f is holomorphic, hence analytic at z0 , we can write f as a power series in z which
converges in some neighborhood of z0 :
∞
X
f (z) = an · (z − z0 )n
n=0

Hence the result follows from the fact that conjugation is a continuous and linear operation on C.

4.2.3 Theorem
Any complex manifold is orientable (as a real manifold) and has always even real dimension.
Proof. Even dimension is a consequence of the identification Cn ∼ = R2n .
Since the wk = (ψji )k are holomorphic in all variables zl , we get ∀ k, l ∈ {1, . . . , n} :

∂wk ∂ w̄k ∂wk ∂w ∂ w̄

=0 and = =0 ⇒ = =0
∂ z̄l ∂zl ∂ z̄l ∂ z̄ ∂z

21
Complex Manifolds Section 4.3 SCHLICHENMAIER, Leytem

∂w
∂w
! !
∂z 0 ∂z 0 
∂w
 
∂w
 
∂w
 
∂w

⇒ J= ∂ w̄
= ⇒ det(J) = det · det = det · det
∂w ∂z ∂z ∂z ∂z


0 ∂ z̄ 0 ∂z
because det is a linear expression. Hence
 2
∂w 2
det(J) = det = det(Jhol ) > 0 since det(Jhol ) 6= 0
∂z


It remains to compute det Jreal (z) . For this, we want to find a relation between det(J) and det(Jreal ), using
the relations (1.1) and (1.2) to calculate terms of the type
   
∂wk ∂(Re wk ) ∂(Im wk ) 1 ∂(Re wk ) ∂(Re wk ) i ∂(Im wk ) ∂(Im wk )
= +i· = · −i· + · −i·
∂zl ∂zl ∂zl 2 ∂xl ∂yl 2 ∂xl ∂yl
   
1 ∂(Re wk ) ∂(Im wk ) i ∂(Im wk ) ∂(Re wk )
= · + + · − (4.1)
2 ∂xl ∂yl 2 ∂xl ∂yl
∂(Re wk ) ∂(Re wk ) ∂(Im wk ) ∂(Im wk )
Note that the terms ∂xl , ∂yl , ∂xl and ∂yl form a 2 × 2–sub-matrix inside Jreal . Similarly
   
∂wk 1 ∂(Re wk ) ∂(Im wk ) i ∂(Im wk ) ∂(Re wk )
= · − + · + (4.2)
∂ z̄l 2 ∂xl ∂yl 2 ∂xl ∂yl
   
∂ w̄k 1 ∂(Re wk ) ∂(Im wk ) i ∂(Im wk ) ∂(Re wk )
= · − − · + (4.3)
∂zl 2 ∂xl ∂yl 2 ∂xl ∂yl
   
∂ w̄k 1 ∂(Re wk ) ∂(Im wk ) i ∂(Im wk ) ∂(Re wk )
= · + − · − (4.4)
∂ z̄l 2 ∂xl ∂yl 2 ∂xl ∂yl

Hence by a rearrangement of J (linear combinations, permutations of lines and columns), one obtains det(Jreal )
from det(J). Indeed consider the base case n = 1 :
∂w ∂w
!
1 1 ∂(Re w1 ) ∂(Re w1 ) !
∂z1 ∂ z̄1 ∂x1 ∂y1
J= ∂ w̄1 ∂ w̄1
, Jreal = ∂(Im w1 ) ∂(Im w1 )
∂z1 ∂ z̄1 ∂x1 ∂y1

Replacing the values in J by (4.1), (4.2), (4.3) and (4.4), we find that det(J) = det(Jreal ), hence J is obtained
by linear combinations of the values in Jreal . For n ≥ 2, we then have
 ∂w1 ∂w1
. . . ∂w ∂w1

∂z1 ∂z2 ∂ z̄1
1
∂ z̄2 ... 1
 ∂w2 ∂w2 ∂w ∂w 
 ∂z . . . ∂ z̄1 ∂ z̄2 . . . 2
2 2
 1 ∂z2
 .. .. .. .. ..

.. .. 
 . . . . . . .
J =  ∂ w̄1 ∂ w̄1
 ∂ w̄1 ∂ w̄1

n + 1
 ∂z1 ∂z2 . . . ∂ z̄1 ∂ z̄2 . . . 
 ∂ w̄2 ∂ w̄2 ∂ w̄2 ∂ w̄2

 . . . . . . n + 2
 ∂z 1 ∂z 2 ∂ z̄ 1 ∂ z̄ 2  .
.. .. .. .. .. .. ..
. . . . . .

In order to obtain the 2 × 2–sub-matrices as described above, we have to bring the row n + i to position i + 1 and
similarly for the columns. Since we get the same number of changes for rows and columns, the total number of
permutations is even, hence the sign of the determinant does not change. And in order to obtain Jreal we then
make the same linear combinations as in the case n = 1 in each one of these 2 × 2–sub-matrices, so finally the
whole determinant did not change. Thus det(Jreal ) = det(J) > 0, showing that the manifold is orientable.

4.2.4 Remark
The converse of this theorem is not true : not all orientable manifolds are complex manifolds.
For example S n is orientable, but S 3 cannot be a complex manifold since dim S 3 = 3 is odd. In fact only S 2 is
a complex manifold since it is equal to S 2 = P1 (C) = CP1 , the complex projective plane (see section 5.3.8).
More generally one can show that any 2-dimensional orientable compact differentiable manifold admits a complex
structure.

22
Complex Manifolds Section 4.3 SCHLICHENMAIER, Leytem

4.3 Holomorphic functions on complex manifolds

4.3.1 Definition
Let U be an open subset (not necessarily a chart domain) of a complex manifold (M, U) with atlas U = (Ui , ϕi )i∈J .
A complex-valued function f : U → C is called a holomorphic function on U ⇔ fi := f ◦ ϕ−1 i is holomorphic
on the open set ϕi (U ∩ Ui ) ⊆ Cn , ∀ i ∈ J (see figure 4.2). Note that U ∩ Ui can be empty.
Similarly a function F : U ⊆ M → Cm is holomorphic on an open subset U of M if all complex-valued coordinate
functions are holomorphic on U .

Figure 4.2: composition of the functions f and ϕ−1

i to give a map fi : Cn → C

More generally : Let (M, U) and (N, V) be 2 complex manifolds of dimension m and n respectively with atlases
U = (Ui , ϕi )i∈I and V = (Vj , ψj )j∈J a f (a) ⊆ Cm ⊆ Cn .
A continuous map f : M → N is called a holomorphic map if and only if (see figure 4.3) the maps

ψj ◦ f ◦ ϕ−1 : ϕi f −1 (Vj ) ∩ Ui ⊆ Cm → (ψj ◦ f ) f −1 (Vj ) ∩ Ui ⊆ Cn

i

are holomorphic, ∀ i ∈ I, ∀ j ∈ J. Note that ϕi f −1 (Vj ) ∩ Ui is open in Cm because

Vj open, f continuous ⇒ f −1 (Vj ) open , Ui open ⇒ f −1 (Vj ) ∩ Ui open

ϕi bicontinuous ⇒ ϕi f −1 (Vj ) ∩ Ui is open

We have to restrict ourselves to this smaller open set since otherwise some operations may not be well-defined.
Figure 4.3: a holomorphic map f : M → N

Remark :
It is enough to check this condition for a subcover, e.g. in the case where U or V contain ”superfluous” charts.
This is justified by the fact that manifolds are equipped with an equivalence class of atlases.

4.3.2 Proposition
Let (M, U) be a complex manifold and p ∈ M . For any U ∈ U with p ∈ U , we can add a coordinate chart (U, ϕ)
which is compatible with the atlas U such that ϕ : U → Cn and ϕ(p) = 0 ∈ Cn .
One says that (U, ϕ) is a centered coordinate chart at p.

23
Complex Manifolds Section 4.4 SCHLICHENMAIER, Leytem

Proof. p ∈ M ⇒ there is a chart (U, ψ) of M such that p ∈ U (because the chart domains cover M ). Define

ϕ(x) := ψ(x) − ψ(p) ∈ Cn , ∀ x ∈ U

We denote τw the translation by w ∈ Cn , i.e. τw (z) = z + w. Since τw−1 = τ−w , this map is clearly biholomorphic.
So ϕ = τ−ψ(p) ◦ ψ, showing that ϕ is also biholomorphic because ψ and τ−ψ(p) are. This already ensures that ϕ
is compatible with the existing atlas U. Moreover ϕ trivially satisfies ϕ(p) = 0.

4.3.3 Consequence
Given a holomorphic function f : M → C and a point p ∈ M , one can always suppose that there exists a
coordinate chart (U, ϕ) centered at p. Hence by identifying U with ϕ(U ) (which is natural because ϕ is bijective
and biholomorphic), one can identify p = 0 and write f|U = f ◦ ϕ−1 which is holomorphic, thus f|U can be written
as a (centered) power series in z = (z1 , . . . , zn ) ∈ Cn .

4.4 Complex submanifolds

4.4.1 Proposition
Every connected open set in a complex manifold is again a complex manifold of the same dimension.
Proof. Let (M, U) be a complex manifold with atlas U = (Ui , ϕi )i∈I . Let N ⊆ M , N 6= ∅ such that N is open
and connected. N can be covered by a family of open subsets in U (see figure 4.4). Define V = (Vi , ψi )i∈I , where

Vi := N ∩ Ui , ψi := ϕi |N ∩U
i

Then (N, V) is also a complex manifold since Vi is open in N and ψi is the restriction of the homeomorphism ϕi
to an open subset of M , hence ψi is still bijective onto ψi (Vi ). In particular we have that dim N = dim M .
Moreover the transition functions ψj ◦ ψi−1 are biholomorphic since they are just restrictions of the ϕj ◦ ϕ−1
i .

Figure 4.4: an open covering of N ⊂ M

4.4.2 Model of a submanifold

As an example consider the complex manifold M = C ∼ = R2 and the closed subset N = S 1 = { z ∈ C , |z| = 1 }.
Again N can be covered by open subsets in the atlas of M .

Let U ⊆ M be open such that U ∩ N 6= ∅ and define V := U ∩ N ⇒ V is open in N . V locally looks like R and
loc loc
U ∼
= R2 = R × R ∼
= V ×R

This is the model of a submanifold : locally N looks like R in R2 ∼

= C.

24
Complex Manifolds Section 4.4 SCHLICHENMAIER, Leytem

Remark :
Arbitrary closed subsets of complex manifolds can in general not define submanifolds. Consider figure 4.5.
At the top it is not admissible since it does not look like R in R × R.
Figure 4.5: N is not a complex submanifold of C

loc
But : this is only true if the manifold is complex or differentiable. For topological manifolds, we have that Λ ∼
= R.
Topologically, pointed lines are equivalent to straight lines since no differentiability conditions are required.

4.4.3 Definition
Let M be a complex manifold of dimension n and N ⊆ M be closed. N is called a (complex) submanifold of M
⇔ ∀ y ∈ N , there is a coordinate chart (U, ϕ) of M with ϕ : U → W ⊆ Cn , W open, y ∈ U such that
∼
W 0 = W ∩ Ck × {0}


ϕ|U ∩N : U ∩ N −→

for some 0 ≤ k ≤ n and where Ck ,→ Cn is embedded in the standard way, i.e. (z1 , . . . , zk ) 7→ (z1 , . . . , zk , 0, . . . , 0).
In addition k is then equal to the dimension of N .
Note that this chart must not necessarily belong to the chosen atlas U of M ; it only has to be compatible with
U (since then adding it to the atlas will not change the manifold structure).
In particular, a submanifold should be a manifold itself. Its atlas is given by restrictions of the charts in the atlas
of the manifold in which it is contained (see figure 4.6). This makes sense since, as restrictions, the transition
functions in the submanifold are then still bijective and biholomorphic. However, by definition, submanifolds do
not need to be connected : a submanifold of a complex manifold is thus again a complex manifold if and only if
it is connected.
Figure 4.6: an atlas of N is given by restrictions of the atlas of M

Concerning the example of S 1 in 4.4.2, it must therefore be said that S 1 is only a real submanifold of C since
dimR S 1 = 1 is odd. It is not possible to endow S 1 with a complex manifold structure, thus it cannot be a
complex submanifold of C neither (any connected complex submanifold is a complex manifold).

4.4.4 Equivalent characterization

Using the Implicit Function Theorem and the Constant Rank Theorem, one can show that :
A subset N in a manifold M of dimension n is a complex submanifold of dimension k if and only if it can locally
be written as the zero set of locally holomorphic functions for which the Jacobian matrix has maximal rank, i.e.
∀ y ∈ N , there is an open neighborhood U of y in M (we may choose U sufficiently small such that (U, ϕ) is a
chart at y) and there are holomorphic functions fi : U → C, i = 1, . . . , n − k (as defined in 4.3.1) such that
n−k
∂(fi ◦ ϕ−1 )
\  
fi−1


U ∩N = {0} , rk (z) = n − k , ∀ z ∈ U ∩ N (4.5)
i=1
∂zj i,j

This does not mean that submanifolds are affine algebraic varieties since the functions fi must be holomorphic,
which is not an algebraic characterization. Note that there must always exist the same number of functions, i.e.
∀ y ∈ N and for any set of functions fi describing U ∩ N , exactly n − k of them are necessary and sufficient.

25
Chapter 5

Examples of complex manifolds

5.1 Cn and open subsets

A global chart is given by (Cn , id), but this is not the only possibility.
− For any fixed a ∈ Cn , one can e.g. take (Cn , ϕa = id −a).
− One can also use the fact that Cn is a C-vector space and that any basis defines a global coordinate chart :
n
X
∀ z ∈ Cn , ∃! ai ∈ C such that z = ai ei ⇒ ϕ(z) := (a1 , . . . , an ) ∈ Cn
i=1
n
where {ei }i=1,...,n can be any basis of C . 
Moreover any open connected subset of Cn is a complex manifold, e.g. the unit ball B n := z ∈ Cn , kzk < 1 .

5.2 Submanifolds of Cn
5.2.1 Linear subspaces
H ⊆ Cn is called a linear subspace of Cn if it is the solution set of a system of homogeneous linear equations, i.e.
if there exist k linear forms li : Cn → C such that
k
T
H= ker li = V (l1 , l2 , . . . , lk )
i=1

In particular, linear subspaces of Cn define affine linear varieties in Cn (since linear maps on Cn are polynomials).

Equivalently a linear subspace of Cn can also be given as the solution set of the matrix equation A · z = 0 :
   
a11 . . . a1n z1
A =  ... .. ..  , a ∈ C , z =  ..  ⇒ H = z ∈ Cn | A · z = 0 ∈ Ck
 
. .  ij .
ak1 ... akn zn
In this case, the linear forms li : Cn → C are given by li (z) = ai1 z1 + ai2 z
2 + . . . + ain zn . Thus their Jacobian
matrix J(l1 , . . . , lk ) is constant and equal to A. Moreover rk J(l1 , . . . , lk ) = rk(A) = k ⇔ the lr are linearly
independent and in this case a linear subspace defines a submanifold of Cn of codimension k, i.e.


rk J(l1 , . . . , lk ) = k ⇔ dim H = n − k
Indeed this defines a splitting Cn = Ck × Cn−k by the characterization (4.5).

26
Complex Manifolds Section 5.2 SCHLICHENMAIER, Leytem

5.2.2 Singular points

Let K be a complex affine variety in Cn defined by p polynomials g1 , . . . , gp .
A point z0 ∈ K is called a singular point of K if the rank of the Jacobian matrix of the gi drops at z0 , i.e.


z0 ∈ K is singular ⇔ rk J(g1 , . . . , gp )(z0 ) is not maximal

A smooth variety is an affine variety which does not contain singular points.
Any linear subspace where the defining linear forms are linearly independent is for example a smooth algebraic
variety since J(f1 , . . . , fk )(z) = A is of maximal rank for all z ∈ Cn .

5.2.3 Exercise
Let f : C → C be a holomorphic function on Cn and consider its vanishing set (which is not a linear subspace !)
n

V (f ) = z ∈ Cn | f (z) = 0 = f −1 {0}


f being holomorphic, thus continuous, V (f ) is closed in Cn , but until now it is not yet a submanifold of Cn .
 
∂f ∂f ∂f
∀ z ∈ Cn : grad f (z) = (z) , (z) , . . . , (z)
∂z1 ∂z2 ∂zn

a) Show that V (f ) is a (not necessarily connected) submanifold of Cn ⇔ grad f (z) 6= 0, ∀ z ∈ V (f ).

Since f is globally defined and V (f ) is the zero set of the holomorphic function f , we can take U = Cn , thus

V (f ) ∩ Cn = f −1 {0}

So by (4.5) it is necessary and sufficient to show that the Jacobian matrix associated to f has maximal rank on
the set U ∩ V (f ) = V (f ). But there is only one (globally defined) function, so n − k = 1 and


rk J(f )(z) = 1 ⇔ J(f )(z) 6= 0 ⇔ grad f (z) 6= 0 , ∀ z ∈ V (f )

b) B := { z ∈ Cn | grad f (z) = 0 } = (grad f )−1 {0}

B is closed since f is holomorphic, i.e. in particular C ∞ . Show that N := V (f ) \ B is a submanifold of Cn \ B.

It is necessary to consider N as a subset of Cn \ B since it is the complement of a closed set in a closed set, hence
in general it is not closed in Cn any more. It is however closed in Cn \ B since

N = V (f ) ∩ Cn \ B and V (f ) is closed in Cn

Moreover Cn \ B is open in Cn , hence it is a complex manifold itself.

Proving that N is a submanifold of Cn \ B is now done exactly as in a) since grad f (z) 6= 0, ∀ z ∈ N .

c) Show that dim V (f ) = n − 1 if V (f ) is a submanifold of Cn .

This follows from the fact that V (f ) is the zero set of a single holomorphic function ⇒ n − k = 1 and

dim V (f ) = k = n − (n − k) = n − 1

By (4.5) : dim N = k ⇔ there are n − k (locally) holomorphic functions fi ; here n − k = 1 ⇒ k = n − 1.

Remark :
If V (f ) is a submanifold of Cn , it is called a hypersurface. And in the case where f is a polynomial function (in
particular it is holomorphic), it is also a hypersurface in the context of affine varieties.

27
Complex Manifolds Section 5.3 SCHLICHENMAIER, Leytem

5.2.4 Examples
For n = 2, let g1 (z1 , z2 ) = z1 − z22 , g2 (z1 , z2 ) = z12 − z22 , g3 (z1 , z2 ) = z12 + z22 , g4 (z1 , z2 ) = z12 + z22 − 1 on C2 .

J(g1 )(z1 , z2 ) = grad g1 (z1 , z2 ) = 1 −2z2 ⇒ rk J(g1 )(z1 , z2 ) = 1, ∀ (z1 , z2 ) ∈ C2

J(g2 )(z1 , z2 ) = grad g2 (z1 , z2 ) = 2z1 −2z2 ⇒ rk J(g2 )(z1 , z2 ) = 1 ⇔ (z1 , z2 ) 6= (0, 0)



J(g3 )(z1 , z2 ) = grad g3 (z1 , z2 ) = 2z1 2z2 ⇒ rk J(g3 )(z1 , z2 ) = 1 ⇔ (z1 , z2 ) 6= (0, 0)



J(g4 )(z1 , z2 ) = grad g4 (z1 , z2 ) = 2z1 2z2 ⇒ rk J(g4 )(z1 , z2 ) = 1, ∀ (z1 , z2 ) ∈ V (g4 )

V (g1 ) is a submanifold (hypersurface) of C2 .

(0, 0) ∈ V (g2 ), V (g3 ) is a singular point, thus V (g2 ) and V (g3 ) are not submanifolds of C2 , but of C2 \ {(0, 0)}.
Moreover g2 (z1 , z2 ) = (z1 − z2 ) (z1 + z2 ) and g3 (z1 , z2 ) = (z1 − i z2 ) (z1 + i z2 ), hence V (g2 ) and V (g3 ) consist of
2 complex lines intersecting in the point (0, 0).

/ V (g4 ), thus grad f 6= 0 on V (g4 ), which is hence a submanifold of Cn .

(0, 0) is also singular for g4 , but (0, 0) ∈

5.2.5 Generalization
Let f1 , . . . , fk be holomorphic functions on Cn . Then the common zero set of the fi is

V (f1 , . . . , fk ) = z ∈ Cn | f1 (z) = f2 (z) = . . . = fk (z) = 0



It is closed and defines a submanifold of Cn ⇔ rk J(f1 , . . . , fk )(z) is maximal, ∀ z ∈ V (f1 , . . . , fk ).

5.3 The projective space

Consider Cn+1 and let P be the space of all lines in Cn+1 passing through the origin (0, . . . , 0) (see figure 5.1).
Since every line is uniquely determined by 2 points passing through it, we hence obtain that every point in
Cn+1 \ {(0, . . . , 0)} defines such a line. Now we introduce a relation ∼ on Cn+1 \ {(0, . . . , 0)} by

z1 ∼ z2 ⇔ ∃ λ ∈ C∗ = C \ {0} such that z2 = λ · z1

i.e. all points lying on a same complex line are identified (figure 5.1). One shows that ∼ is an equivalence relation.

Figure 5.1: lines in Cn+1 passing through the origin and the relation ∼

The n-dimensional complex projective space is then given by

Pn (C) = CPn := Cn+1 \ {(0, . . . , 0)} / ∼ = [z] z ∈ Cn+1 \ {(0, . . . , 0)}



where each equivalence class [z] = { z 0 ∈ Cn+1 \ {0} | z 0 ∼ z } represents a line in Cn+1 passing through the
origin. Elements in Pn (C) are given by such equivalence classes.

28
Complex Manifolds Section 5.3 SCHLICHENMAIER, Leytem

Goal :
We want to show that Pn (C) is a compact complex manifold of dimension n. Compare e.g. the case RP1 ∼
= S1
by identifying antipodal points (see figure 5.2; when arriving at π, the circle closes again). Thus

S 1 compact ⇒ RP1 compact

Figure 5.2: RP1 is obtained from S 1 by identifying antipodal points

5.3.1 Topology
First we need to define a topology on Pn (C). Consider the canonical projection map

ν : Cn+1 \ {0}  Pn (C) : z 7→ [z]

which maps a point to the line it defines. Cn+1 \ {0} is open in Cn+1 , hence it is a topological space itself.
Then we endow Pn (C) with the usual quotient topology, i.e. the finest topology such that the projection map
ν is continuous. This means that U ⊆ Pn (C) is open ⇔ ν −1 (U ) is open in Cn+1 \ {0}. By the properties of
preimages, this defines indeed a topology on Pn (C).
− Anecdote to the quotient topology :
One can of course endow Pn (C) with the trivial topology since ν(∅) = ∅ and ν −1 Pn (C) = Cn+1 \ {0}, thus ν

would be continuous. But this topology is too small in order to provide interesting results.
The discrete topology is however not possible since if all points are open in Pn (C), then ∀ [z] ∈ Pn (C) :

ν −1 [z] = z 0 ∈ Cn+1 \ {0} ν(z 0 ) = [z] = { z 0 ∈ Cn+1 \ {0} z 0 ∼ z } = λ · z | λ ∈ C∗


 

This is equal to the line in Cn+1 defined by [z] without the origin, which is not open in Cn+1 \ {0}, thus ν is not
continuous with respect to the discrete topology.

Proposition :
ν is an open map and hence : U ⊆ Pn (C) is open ⇔ ∃ W ⊆ Cn+1 \ {0} open such that ν(W ) = U .
Proof. Let V ⊆ Cn+1 \ {0} be open.
We have to show that ν(V ) is open in Pn (C), i.e. that ν −1 ν(V ) is open in Cn+1 \ {0}. But

[
ν −1 ν(V ) =


λ·V (5.1)
λ∈C∗

⊂ : if ν(x) ∈ ν(V ), then ∃ v ∈ V s.t. ν(x) = ν(v) ⇔ [x] = [v] ⇔ x ∼ v ⇔ ∃ λ ∈ C∗ s.t. x = λ v ⇒ x ∈ λ · V

⊃ : if ∃ λ ∈ C∗ s.t. x ∈ λ · V , then ∃ v ∈ V s.t. x = λ v ⇒ [x] = [v] ⇔ ν(x) = ν(v) ∈ ν(V ) ⇒ x ∈ ν −1 ν(V )
Note that ν −1 ν(V ) represents a cone in Cn+1 (see figure 5.3).

And λ · V is open in Cn+1 \ {0} since the map ϕ : V → λ · V , ϕ(v) = λ v is a homeomorphism if λ 6= 0, so

[
λ · V = ν −1 ν(V ) is open in Cn+1 \ {0} as a union of open sets


λ · V = ϕ(V ) is open ⇒
λ∈C∗

It follows that : U ⊆ Pn (C) is open ⇔ ∃ W open in Cn+1 \ {0} such that W = ν −1 (U ), thus ν(W ) = U :

ν(W ) = ν ν −1 (U ) = id(U ) = U since ν is surjective and has thus a right inverse

And finally ν(W ) is open in Pn (C) since W is open and ν is an open map.

29
Complex Manifolds Section 5.3 SCHLICHENMAIER, Leytem

5.3.2 Hausdorff
Exercise : Show that Pn (C) with respect to the quotient topology is a Hausdorff space.

Let [x], [y] ∈ Pn (C) such that [x] 6= [y] ⇒ ∃ x, y ∈ Cn+1 \ {0} where x and y do not lie on the same line. Let

`x := λ x | λ ∈ C∗ , `y := λ y | λ ∈ C∗
 

`x ∩ `y = ∅. Now choose U open in Cn+1 \ {0} such that x ∈ U and U ∩ `y = ∅ (which is possible since x and y
are not on the same line) as in figure 5.3. Denote the cone through U by CU = { λ z | λ ∈ C∗ , z ∈ U }. Then
choose V open in Cn+1 \ {0} such that y ∈ V and V ∩ CU = ∅. We obtain that CU ∩ CV = ∅ too. Hence ν(U )
and ν(V ) are open neighborhoods (ν is open) containing
respectively. And ν(U ) ∩ ν(V ) = ∅ because
[x] and [y]
the corresponding cones are disjoint : ν −1 ν(U ) = CU , ν −1 ν(V ) = CV and CU ∩ CV = ∅.

Figure 5.3: preimages under the quotient map ν correspond to cones in Cn+1

Note : Not any quotient space of a Hausdorff space is again Hausdorff. Here we really use the fact that we are
dealing with Cn+1 (which is Hausdorff and regular) and an open projection map.

5.3.3 Compactness
We shall show that Pn (C) is compact. Let

Cn+1 \ {0}
ϕ1
/ S 2n+1 ϕ2
/ Pn (C)
ν
4

where Cn+1 \ {0} ∼ z

= R2n+2 \ {0} is non-compact, ϕ1 (z) = kzk and ϕ2 (y) = [y], ∀ y ∈ S 2n+1 .
This diagram commutes because  z 
(ϕ2 ◦ ϕ1 )(z) = kzk = [z] = ν(z)
ϕ1 is (obviously) continuous and ϕ2 is continuous since similarly as in (5.1), ∀ U ⊆ Pn (C) open :
[ [
ϕ−1
2 (U ) =
λ
|λ| · U = α·U
λ∈C∗ |α|=1

z


In addition ϕ2 is surjective since for [z] ∈ Pn (C), we have ϕ2 kzk = [z]. Hence since S 2n+1 is compact (closed
2n+2
and bounded in R \ {0}), we obtain that P (C) = im ϕ2 = ϕ2 (S 2n+1 ) is also compact (as recalled in 3.2.1).
n

5.3.4 Atlas
Now we are going to define charts and chart domains for Pn (C).
Let {Wi }i∈J be an open covering of M = Cn+1 \ {0}. Then {ν(Wi )}i∈J will be an open covering of Pn (C) since ν
is open and surjective. However it will not be a coordinate covering : since ν is not bijective, the coordinates on
Cn+1 \ {0} do not induce coordinates of Pn (C). ”ν −1 ” does not define a global chart of Pn (C), so this approach
is not helpful for defining a manifold structure.

30
Complex Manifolds Section 5.3 SCHLICHENMAIER, Leytem

Definition :
To any [z] ∈ Pn (C), we associate the homogeneous coordinates

[z] = (z0 : z1 : z2 : . . . : zn ) ∈ Pn (C)

where (z0 , z1 , . . . , zn ) are the coordinates of a representative z ∈ Cn+1 \{0} of [z], modulo the equivalence relation,
i.e. z 0 ∈ [z] ⇔ ∃ λ ∈ C∗ such that z 0 = λ z. Hence

(z0 : z1 : z2 : . . . : zn ) = (λ z0 : λ z1 : λ z2 : . . . : λ zn ) , ∀ λ ∈ C∗

Homogeneous coordinates are therefore not unique in general. In particular (0 : 0 : . . . : 0) ∈ / Pn (C) since the
n+1
origin (0, 0, . . . , 0) was excluded from C . If [z] = (z0 : z1 : . . . : zn ), then ∃ i ∈ {0, . . . , n} such that zi 6= 0.

Now we define for any k ∈ {0, . . . , n} the subset

Uk := [z] = (z0 : z1 : . . . : zn ) ∈ Pn (C) zk 6= 0



This is well-defined since if [z] = (z0 : . . . : zn ) ∈ Pn (C) such that zk 6= 0 in this homogeneous representation,
then zk 6= 0 in all homogeneous representations since we are only allowed to multiply by non-zero constants λ.
Thus Uk is a well-defined subset of Pn (C).
In particular if [z] ∈ Uk , then we can choose λ = z1k and take as homogeneous coordinates
z z1 zn 
0
[z] = (z0 : z1 : . . . : zk : . . . : zn ) = : : ... : 1 : ... :
zk zk zk

Indeed, in [z] (set of points in Cn+1 ) there is a unique representation of [z] with a 1. Hence if [z] ∈ Uk , we
can assume without loss of generality that [z] = (y0 : y1 : . . . : 1 : . . . : yn ) for some yi ∈ C. Moreover this
representation of [z] is now unique.

Proposition :
Uk is open in Pn (C), ∀ k ∈ {0, . . . , n}.

Proof.
ν −1 (Uk ) = z ∈ Cn+1 \ {0} z ∈ Cn+1 \ {0}
 
ν(z) = [z] ∈ Uk = zk 6= 0
n+1
and this is an open subset of C \ {0}, thus Uk is open.

Proposition :
Uk ∼
= Cn , ∀ k ∈ {0, . . . , n}. Moreover this equivalence is homeomorphic.
Proof. For [z] ∈ Uk , we define the maps
z z1 zn 
0
ϕk : Uk −→ Cn : [z] 7−→ ,,...,b k, . . . ,
zk zk zk
ψk : Cn −→ Uk : (w1 , . . . , wn ) 7−→ (w1 : w2 : . . . : wk−1 : 1 : wk : . . . : wn )
λzi
ϕk is well-defined since λz k
= zzki , ∀ i 6= k.
This already gives a 1-to-1 correspondence since ϕk ◦ ψk = idCn and ψk ◦ ϕk = idUk ⇒ ψk = ϕ−1
k .
It remains to check that ϕk is bicontinuous, i.e. a homeomorphism.

− ϕk is continuous : let W ⊆ Cn be open.

We have to show that ϕ−1 −1 n n −1
k (W ) is open in Uk ⇔ ϕk (W ) is open in P (C) since Uk is open in P (C) and ϕk (W )
is by definition already contained in Uk . But
0 −1
ν −1 ϕ−1 where ν 0 = ν|ν −1 (Uk )



k (W ) = ϕk ◦ ν (W )

where one needs to take the restriction of ν, otherwise the composition with ϕk does not make sense.

31
Complex Manifolds Section 5.3 SCHLICHENMAIER, Leytem

The map z
0 z1 zn 
ϕk ◦ ν 0 : ν −1 (Uk ) ⊆ Cn+1 \ {0} −→ Cn : (z0 , z1 , . . . , zn ) 7−→ , ,...,b
k, . . . ,
zk zk zk
is continuous since zk 6= 0 on ν −1 (Uk ), thus ν −1 ϕ−1 \ {0} ⇒ ϕ−1

n+1
k (W ) is open in C k (W ) is open in Uk .

− ϕk is open : let U ⊆ Uk be open. We have to show that ϕk (U ) is open in Cn .

nz z zn  
0 1
ϕk (U ) = y ∈ Cn y = ϕk [z] for some [z] ∈ U = [z] ∈ U ⇔ z ∈ ν −1 (U )


, ,...,b k, . . . ,
zk zk zk

By denoting fk : Cn+1 → Cn , fk (z0 , . . . , zn ) = zzk0 , . . . , b

k, . . . , zznk , we obtain that

fk (z0 , . . . , zn ) z ∈ ν −1 (U ) = fk|ν −1 (Uk ) ν −1 (U )



ϕk (U ) =

where ν −1 (U ) is open since U is open and ν continuous and fk is an open map since (x, y) 7→ x
y is open on C2 ,
thus the restriction of fk to an open subset of Cn+1 is still open. Finally ϕk (U ) is open in Cn .

5.3.5 Connectedness


Pn (C) is connected because ν is surjective : Pn (C) = im ν = ν Cn+1 \ {0} , where Cn+1 \ {0} is connected and
ν is continuous, hence im ν is also connected.

5.3.6 Topological structure

P (C) is a topological manifold. The Uk are open in the projective space and cover Pn (C) since : if [z] ∈ Pn (C),
n

then [z] 6= (0 : . . . : 0), so ∃ l ∈ {0, . . . , n} such that zl 6= 0 ⇒ [z] ∈ Ul .

[n
⇒ {Uk }k=0,...,n open covering : Uk = Pn (C)
k=0

{Uk }k=0,...,n is called the standard affine covering of Pn (C). Hence Pn (C) already defines a topological manifold
of complex dimension n (real dimension 2n) since we can take the homeomorphism ϕk to be the associated chart
to the domain Uk . So U = (Uk , ϕk )k=0,...,n is a continuous atlas of Pn (C).

5.3.7 Complex structure

Figure 5.4: the transition maps ψlk are holomorphic

Consider figure 5.4. We have to show that the transition maps

ψlk := ϕl ◦ ϕ−1
k : ϕk (Uk ∩ Ul ) ⊆ Cn −→ ϕl (Ul ∩ Uk ) ⊆ Cn

are biholomorphic. We assume without restriction that l < k, so

Uk ∩ Ul = [z] ∈ Pn (C) zk 6= 0 and zl 6= 0 and ϕk (Uk ∩ Ul ) = (w1 , . . . , wn ) ∈ Cn

 
wl 6= 0

32
Complex Manifolds Section 5.4 SCHLICHENMAIER, Leytem

Let (w1 , . . . , wn ) ∈ ϕk (Uk ∩ Ul ), so wl 6= 0 and by applying ψlk = ϕl ◦ ϕ−1 k we obtain

ψlk (w1 , . . . , wn ) = ϕl ϕ−1

k (w1 , . . . , wn ) = ϕl (w1 : . . . : wl−1 : wl : wl+1 : . . . : wk−1 : 1 : wk : . . . : wn )
w wl−1 wl+1 wk−1 1 wk wn 
1
= ϕl : ... : :1: : ... : : : : ... :
wl wl wl wl wl wl wl
w wl−1 wl+1 wk−1 1 wk wn 
1 n
= ,..., , ,..., , , ,..., ∈C
wl wl wl wl wl wl wl
which is well-defined since wl 6= 0. Thus ψlk is holomorphic from Cn \ { w ∈ Cn | wl = 0 } to Cn as the map
w 7→ w n
wl is holomorphic in w, ∀ i 6= l. This now shows that P (C) is a complex manifold of complex dimension n.
i

5.3.8 Particular case

CP1 := P1 (C) is called the Riemann sphere or the projective line. Since there are only 2 chart domains in this
case, it can be decomposed as :

U0 = (z0 : z1 ) ∈ P1 (C) z0 6= 0 = (1 : w) w ∈ C ∼
 
=C
1 ∗
 
P (C) \ U0 = (0 : z1 ) z1 ∈ C = (0 : 1) = { pt }

/ P1 (C), P1 (C) \ U0 consists of exactly 1 point. This point (0 : 1) is often also denoted by ∞ and
Since (0 : 0) ∈
called the point at infinity. Hence we obtain the decomposition

P1 (C) ∼
= C ∪˙ {∞} ∼
= S2 (5.2)

The first isomorphy is nothing else than the Alexandroff compactification of C and the second one is obtained
by the stereographic projection since topologically R2 ∪ {∞} ∼
= S 2 . Of course (5.2) also holds true with U1 ∼
=C
and P1 (C) \ U1 = {∞} where (1 : 0) = ∞ in this case.
Using such a compactification argument in order to show that Pn (C) is compact however only works for n = 1.
In general one has the following decomposition (which is not a 1-point compactification) :

= Cn ∪˙ Pn−1 (C) ∼
Pn (C) ∼ = Cn ∪˙ Cn−1 ∪˙ Pn−2 (C) ∼
= ... ∼
= Cn ∪˙ Cn−1 ∪˙ . . . ∪˙ C2 ∪˙ C ∪˙ {∞}

where Cn ∼= Uk , Cn−1 ∼ = Pn (C) \ Uk , etc. This shows in particular that there is no unique way to set ∞ ; one
always has to specify if one chooses ∞ = (1 : 0 : . . . : 0), ∞ = (0 : 1 : . . . : 0), . . . , or ∞ = (0 : 0 : . . . : 1).

5.4 Submanifolds of Pn (C)

5.4.1 Proposition
Consider the compact complex manifold Pn (C) and let N be a connected submanifold of Pn (C).
Then N is also a compact complex manifold.
Proof. Submanifolds are closed by definition and a closed set in a compact set is again compact. Moreover N is
a complex manifold because it is connected (see section 4.4.3).
More generally : Let X be compact and F ⊆ X be closed.
We want to show that F is also compact, i.e. if {Ui }i∈J is an open covering of F , it admits a finite subcover.
Ui open in F means that Ui = Vi ∩ F for some Vi open in X, ∀ i ∈ J. Since {Ui }i∈J covers F , we get
[ [
S [
F = Ui = Vi ∩ F = Vi ∩ F ⇒ F ⊆ Vi
i∈J i∈J i∈J i∈J

Hence {Vi }i∈J ∪ {X \ F } is an open covering of X (F is closed). As X is compact, finitely many of them will
cover X : ∃ I ⊂ J finite such that {Vi }i∈I ∪ {X \ F } covers X, so the corresponding {Ui }i∈I will cover F .

33
Complex Manifolds Section 5.5 SCHLICHENMAIER, Leytem

5.4.2 Examples
1) zero sets of locally holomorphic functions on Pn (C)
2) zero sets of homogeneous polynomials
We know that Pn (C) is a compact manifold. In section 6.1, we will show the following properties :
− On a compact complex manifold there are no globally defined non-constant holomorphic functions.
− Let M ⊆ Cn be a (connected) compact submanifold of Cn . Then M is a point : ∃ c ∈ Cn such that M = {c}.
Thus it does not make much sense to consider zero sets of globally holomorphic functions on Pn (C). Note that
2) is not a particular case of 1) since polynomials are not functions on the projective space Pn (C) !

A polynomial on Cn+1 with coefficients in C is called homogeneous of degree m if it is of the form

X
f (X0 , X1 , . . . , Xn ) = cj0 ,...,jn · X0j0 · X1j1 · . . . · Xnjn , cj0 ,...,jn ∈ C
j0 +...+jn =m

Let g(X0 , X1 ) = X0 X1 , then is homogeneous of degree 2. But g is not a function on P1 (C) since evaluation is not
well-defined. Indeed let [z] ∈ P1 (C) be represented by z = (z0 , z1 ) ∈ [z] ⇒ g(z) = z0 z1 . But if z 0 = (z00 , z10 ) ∈ [z]
with zi0 = λ zi , then [z] = [z 0 ] = (z0 : z1 ) = (λz0 : λz1 ), but g(z 0 ) = (λz0 ) (λz1 ) = λ2 · z0 z1 6= g(z) in general.
The zero set of a homogeneous polynomial is however well-defined : if f is homogeneous of degree m, then

∀ λ ∈ C : f (λ z) = λm · f (z) , hence f (z) = 0 ⇔ f (λ z) = 0, ∀ λ ∈ C∗

5.4.3 Definition
A smooth projective variety of Pn (C) is a non-singular (see 5.2.2) zero set of finitely many homogeneous polyno-
mials. Hence any smooth projective variety is a (closed) submanifold of Pn (C). And we even have :

5.4.4 Chow’s Theorem

Any (closed) submanifold of Pn (C) is a smooth projective variety. We will not prove this result.

5.5 Complex tori

Complex tori are examples of compact complex manifolds which are not submanifolds of Pn (C).
For n = 1, a tori T can be visualized as in figure 1, p. 3 (only the surface ; it is ”hollow” inside). T is compact
as a subset of R3 and can be embedded into R4 ∼ = C2 . But because of 5.4.2, there are no non-trivial compact
complex submanifolds in C2 . So there is no chance to see the complex structure on this picture ; one only sees
the topology. As we will see in the following : a way of constructing T starting from C is to define T = C / L
where L = h 1 , τ iZ is the lattice generated by 1 and τ ∈ C, Im τ > 0 (see figure 5.5).

5.5.1 Definitions
L ⊂ Cn is called a lattice in Cn of real dimension 2n if it is of the form
 2n 
P
L = h ω1 , ω2 , . . . , ω2n iZ = mj ωj mj ∈ Z
j=1

where ωi ∈ Cn are linearly independent over R.

Remark :
The ωi cannot be linearly independent over C since m > n vectors in an n-dimensional vector space are always
linearly dependent. But it is possible over R since Cn has dimension 2n over R. Consider e.g.
• 1 and τ , Im τ > 0, are linearly dependent over C since 1 + −1
τ ·τ =0
• over R : let α, β ∈ R such that α · 1 + β · τ = 0 ⇒ β · τ ∈ C \ R since Im τ > 0, hence β = 0 and also α = 0

34
Complex Manifolds Section 5.5 SCHLICHENMAIER, Leytem

L is a discrete subspace of Cn , i.e. ∀ z ∈ L, there is a neighborhood Uz of z in Cn such that L ∩ Uz = {z} as it

can be seen in figure 5.5.
Figure 5.5: a (discrete) lattice generated by 1 and τ ∈ C, Im τ > 0

Let z, z 0 ∈ Cn . We define the relation z ∼L z 0 ⇔ z 0 − z ∈ L ⇔ ∃ ω ∈ L such that z 0 = z + ω.

Exercise :
∼L is an equivalence relation on Cn .
− reflexive : z ∼L z for ω = 0 ∈ L
− symmetric : z 0 = z + ω ⇒ z = z 0 + (−ω) since ω ∈ L ⇒ −ω ∈ L
− transitive : z 0 = z + ω, z 00 = z 0 + ω 0 ⇒ z 00 = z + (ω + ω 0 ) with ω, ω 0 ∈ L ⇒ ω + ω 0 ∈ L

Hence one can define equivalence classes of ∼L and the quotient of Cn by ∼L :

[z] = z 0 ∈ C | z 0 ∼L z = z + ω | ω ∈ L
 

As a set, Cn / ∼L can now be identified with F, set consisting of all points from Cn which are not equivalent to
any other element in F. F is called the fundamental parallelogram of L and given by (see figure 5.6)
 2n 
P
F= αj ωj 0 ≤ αj < 1
j=1
n
F is neither open nor closed in C , but it gives a 1-to-1 correspondence to the equivalence classes of ∼L . One
also considers the (topological) closure of F (see figure 5.6), which differs from F just by a set of measure zero :
 2n 
: closed in Cn
P
F= αj ωj 0 ≤ αj ≤ 1
j=1

Figure 5.6: the fundamental parallelogram F and its closure

Definition :
The n-dimensional complex torus T n is defined as the quotient of Cn by the equivalence relation ∼L , i.e.
T n := Cn / ∼L = Cn / L ⇒ Tn ∼
= F as sets
T n and F are in bijection only and don’t even have the same topological properties. F is for example simply
connected, but the torus is not.

5.5.2 Topology
We again endow T n with the quotient topology, i.e. with the finest topology such that the natural projection
map ν : Cn → T n : z 7→ [z] is continuous, so
W ⊆ T n is open ⇔ ν −1 (W ) is open in Cn

35
Complex Manifolds Section 5.5 SCHLICHENMAIER, Leytem

Proposition : ν is an open map.

Proof. Let U ⊆ Cn be open. ν is open ⇔ ν(U ) is open in T n ⇔ ν −1 ν(U ) is open in Cn . But

[
⇒ ν −1 ν(U ) =



U +ω
ω∈L

⊂ : if ν(x) ∈ ν(U ), then [x] = [u] for some u ∈ U ⇒ ∃ ω ∈ L such that x = u + ω

⊃ : if x = u + ω for some u ∈ U , ω ∈ L, then ν(x) = [x] = [u + ω] = [u] ∈ ν(U ) ⇒ x ∈ ν −1 ν(U )

Figure 5.7: preimages under the projection map ν correspond to translations

And U + ω is open in Cn since the map ϕ : U → U + ω, ϕ(u) = u + ω is a homeomorphism ∀ ω ∈ L, so

[
U + ω = ν −1 ν(U ) is open in Cn as a union of open sets



U + ω = ϕ(U ) is open ⇒
ω∈L

5.5.3 Compactness
T is compact (hence also paracompact). For this, let {Ui }i∈I be an open covering of T n .
n

By definition the ν −1 (Ui ) are open in Cn and they form an open covering of Cn since
[ S  [
ν −1 (Ui ) = ν −1 Ui = ν −1 (T n ) = Cn ν −1 (Ui ) ∩ F


⇒ F=
i∈I i∈I i∈I

F being compact, one can extract a finite subcovering {ν −1 (Uj ) ∩ F }j∈J (see figure 5.8), thus T n =
S
j∈J Uj :
[
[z] ∈ T n ⇒ ∃ z ∈ [z], ω ∈ L such that z + ω ∈ F ⊂ ν −1 (Uj ) ∩ F

j∈J

i.e. ∃ j ∈ J such that z + ω ∈ ν −1 (Uj ) ⇒ [z] = [z + ω] ∈ ν ν −1 (Uj ) = Uj since ν is surjective.

Figure 5.8: a finite open covering of F

5.5.4 Hausdorff
T n is Hausdorff. For this, let [x], [y] ∈ T n such that [x] 6= [y]. [x] and [y] are represented by x, y ∈ Cn and
∃ ω, ω 0 ∈ L such that x0 = x + ω, y 0 = y + ω 0 and x0 , y 0 ∈ F with x0 6= y 0 since [x] 6= [y].
Assume first that x, y ∈ F ◦ (interior of F). Then there are open neighborhoods Ux and Uy of x0 and y 0 such that
Ux ∩ Uy = ∅ since F ◦ ⊂ Cn is Hausdorff (see figure 5.9). ν is open, so ν(Ux ) and ν(Uy ) are open neighborhoods
of [x] = [x0 ] and [y] = [y 0 ] in T n . Moreover ν(Ux ) ∩ ν(Uy ) = ∅ since otherwise

[z] ∈ ν(Ux ) ∩ ν(Uy ) ⇒ ∃ w, w0 ∈ L such that z + w ∈ Ux , z + w0 ∈ Uy

which is a contradiction since there is a unique representative of [z] in F.

The proof is similar if at least one of x0 or y 0 lies in F \ F ◦ . The only difference is that the identification will
”disconnect” the neighborhoods in Cn (see figure 5.9). But ν(Ux ) and ν(Uy ) will not change in this case.

36
Complex Manifolds Section 5.5 SCHLICHENMAIER, Leytem

Figure 5.9: open neighborhoods of x0 and y 0 inside of F

5.5.5 Connectedness
T is connected since it is the image of the continuous surjective map ν : Cn → T n where Cn is connected, hence
n

im ν = ν(Cn ) = T n is connected as well.

5.5.6 Complex structure

We have to define charts and chart domains for T n . In particular we want the map ν to be holomorphic (as
defined in 4.3.1) with respect to this atlas.

Let [z] ∈ T n and let zF be the unique representing element of [z] in F. We consider an open
ball Bz in
Cn around zF , small enough such that Bz does not meet other points z 0 ∈ [z] : Bz ∩ ν −1 {ν(zF )} = {zF }.
Note that Bz is not completely contained in F if zF lies on the boundary of F.

Repeating this argument for any [z] ∈ T n , we obtain a covering of F by these Bz , hence by compactness it
suffices to consider only finitely many of them. We denote this finite number of Bz by Bj for j ∈ J finite. Let

∀ j ∈ J : Uj := ν(Bj ) = [x] ∈ T n x ∈ Bj


Uj is open in T n since ν is open and {Uj }j∈J is an open covering of T n since

[ [  S 

Uj = ν(Bj ) = ν Bj = ν F = T n
j∈J j∈J j∈J

In addition, Bj ( ν −1 (Uj ). More precisely, we have that ν −1 (Uj ) = ˙ α∈L Vαj (see figure 5.10), where the Vαj are
S

open in C, V0j = Bj and each Vαj gives a 1-to-1 correspondence with Uj .

Figure 5.10: ν −1 (Uj ) consists of several pieces

Now we can define the atlas

W = Wjα , ϕjα (j,α)∈J×L
, Wjα = Uj , ∀ α ∈ L
ϕjα : Uj → Vαj : [x] 7→ xα , ϕ−1 j
jα : Vα → Uj : y 7→ [y] (5.3)

where xα is the unique representative of [x] in Vαj , hence ϕjα is indeed bijective. Moreover ϕ−1 −1
jα = ν|Vαj , so ϕjα
is continuous and for U ⊆ Vαj open, we have ϕ−1jα (U ) = ν(U ), which is open, hence ϕjα is continuous as well.

37
Complex Manifolds Section 5.5 SCHLICHENMAIER, Leytem

It remains to check that the maps ϕjα ◦ ϕ−1 n n

kβ : ϕkβ (Uk ∩ Uj ) ⊆ C → ϕjα (Uj ∩ Uk ) ⊆ C are holomorphic.
Consider figure 5.11. If y ∈ ϕkβ (Uk ∩ Uj ) ⊆ Vβk , then

ϕjα ◦ ϕ−1 −1 j




kβ (y) = ϕjα ϕkβ (y) = ϕjα [y] = yα ∈ Vα

where y 7→ yα is just a translation by α − β ∈ L, hence it is holomorphic with respect to y.

Finally, W defines an n-dimensional atlas of T n , so the torus is a complex manifold of dimension n.
Figure 5.11: the transition maps are holomorphic

With respect to this atlas, one can now show that ν : Cn → T n is holomorphic (as defined in 4.3.1) because
∀ (j, α) ∈ J × L , ϕjα ◦ ν ◦ id : ν −1 (Uj ) ⊆ Cn −→ Vαj ⊆ Cn : y 7−→ ϕjα [y] = yα

where ν −1 (Uj ) = ˙ β∈L Vβj as above. Thus y 7→ yα is again a translation, but here the shift depends on y since
S

ν −1 (Uj ) is a union of Vβj (see figure 5.10). This dependence is however holomorphic since Vβj ∩ Vβj0 = ∅ for β 6= β 0 ,
so no discontinuities appear : the shift is constant on any Vβj . We thus conclude that ϕjα ◦ ν is holomorphic.

Remark :
By 4.3.2, we can always assume that the atlas W also contains a centered chart at [z] for any [z] ∈ T n .

5.5.7 Proposition
Let M1 and M2 be 2 complex manifolds. Then M1 × M2 carries a natural structure of complex manifolds such
that the natural projections pi : M1 × M2 → Mi are holomorphic maps and the injections ij,a : Mj → M1 × M2
for j ∈ {1, 2}, a ∈ Mj : i1,a (m1 ) = (m1 , a) , i2,a (m2 ) = (a, m2 )
define submanifolds of M1 × M2 which are isomorphic to Mj . Furthermore dim(M1 × M2 ) = dim M1 + dim M2 .
Proof. m ∈ M1 × M2 ⇔ m = (m1 , m2 ) for some mi ∈ Mi
Let U = (Ui , ϕi )i∈I be an atlas of M1 and V = (Vj , ψj )j∈J be an atlas of M2 . Then we can define an atlas of
M1 × M2 by setting U × V := (Ui × Vj , ϕi × ψj )(i,j)∈I×J where {Ui × Vj } is an open covering of M1 × M2 and

(ϕi × ψj )(m) = (ϕi × ψj )(m1 , m2 ) = ϕi (m1 ), ψj (m2 ) ∈ Cdim M1 × Cdim M2

With respect to this atlas, the projections are holomorphic since in any local coordinate chart, we have
  

ϕk ◦ p1 ◦ (ϕi × ψj )−1 (x) = ϕk p1 ϕ−1 −1
= ϕk ϕ−1 −1



i (x1 ), ψj (x2 ) i (x1 ) = ϕk ◦ ϕi (x1 )
   
ψk ◦ p2 ◦ (ϕi × ψj )−1 (x) = ψk p2 ϕ−1 −1
= ψk ψj−1 (x2 ) = ψk ◦ ψj−1 (x2 )




i (x1 ), ψj (x2 )

and this is holomorphic since the transition maps are holomorphic. Concerning the injections, we have
i1,a (M1 ) = M1 × {a} = p−1 i2,a (M2 ) = {a} × M2 = p−1



2 {a} and 1 {a}
hence i1,a (M1 ) and i2,a (M2 ) are closed with respect to the product topology. In order to show that they indeed
define submanifolds of M1 × M2 , we use the characterization in (4.5). Denote n1 = dim M1 , n2 = dim M2 and
(1) (2) (n1 ) (1) (2) (n2 )
m ∈ M1 × M2 ⇒ (ϕi × ψj )(m) = z = (z1 , z2 ) = (z1 , z1 , . . . , z1 , z2 , z2 , . . . , z2 )

38
Complex Manifolds Section 5.6 SCHLICHENMAIER, Leytem

m = (m1 , m2 ) ∈ M1 × M2 ⇒ ∃ Vj ∈ V such that (m1 , m2 ) ∈ M1 × Vj . For l ∈ {1, . . . , n2 }, let

(l)
fl : M1 × Vj −→ C : (m1 , m2 ) 7−→ ψj (m2 ) − ψj (a) : lth coordinate

(l)
(l)
So fl (m1 , m2 ) = 0 ⇔ ψj (m2 ) = ψj (a) and fl (m1 , m2 ) = 0, ∀ l ⇔ ψj (m2 ) = ψj (a) ⇔ m2 = a, hence
    n2
\
fl−1 {0}


M1 × Vj ∩ M1 × {a} = M1 × {a} =
l=1

and this is exactly (4.5) with U = M1 × Vj , n → n1 + n2 and k = n1 . And the fl are holomorphic since locally

 (l)
(l)
(l)
fl ◦ (ϕi × ψj )−1 (z1 , z2 ) = fl ϕ−1 −1 −1



i (z1 ), ψj (z2 ) = ψj ψ j (z2 ) − ψj (a) = z2 − ψj (a) (5.4)

It remains to check that the rank of the Jacobian matrix is maximal. But by (5.4), this is simply given by
 
0 ... 0 1 0 ... 0
0 . . . 0 0 1 . . . 0
J(f1 , . . . , fn2 ) =  . .
 
.. .. .. . . .
 .. .. . . . . .. 
0 ... 0 0 0 ... 1

since fl is independent of z1 and just a projection with respect to z2 . Finally M1 × {a} ∼

= M1 is a submanifold
of M1 × M2 of dimension n1 .

5.5.8 Additive structure

Next we want to define an additive structure on T n .
As a vector space (Cn , +) (addition of vectors) is an abelian group. The lattice L ⊂ Cn is an additive subgroup
of Cn , thus also called the free abelian group generated by 2n elements. L being abelian, hence normal, the
quotient T n = Cn / L carries a natural group structure given by [z] + [y] = [z + y] and [z]−1 = [−z].

Exercise :
−1
The group operations + : T n × T n → T n and : T n → T n are holomorphic.
1) To check this, we have to compose + with the charts of T n given in (5.3). If we work locally, then
     
ϕjα ◦ + ◦ (ϕkβ × ϕlγ )−1 (x1 , x2 ) = ϕjα ϕ−1 −1
kβ (x1 ) + ϕlγ (x2 ) = ϕjα [x1 ] + [x2 ]
 
= ϕjα [x1 + x2 ] = (x1 + x2 )α

i.e. locally we only add the components of x = (x1 , x2 ), followed by a constant shift. This is holomorphic.

2) Similarly we obtain for −1 :

  
−1     
ϕjα ◦ −1 ◦ ϕ−1
kβ (x) = ϕjα ϕ−1
kβ (x) = ϕjα [x]−1 = ϕjα [−x] = (−x)α

Locally we just take the inverse of the coordinate and shift it again by a constant, so this is holomorphic too.
Hence T n admits a holomorphic group structure.

5.6 Complex Lie groups

5.6.1 Definition
Let (G, ·) be a complex manifold with a group law · : G × G → G. (G, ·) is called a complex Lie group if
1) the multiplication · : G × G → G is a holomorphic map with respect to the product structure
2) the inverse operation −1 : G → G : g 7→ g −1 is also holomorphic.
(T n , +) is thus a complex Lie group by the previous exercise.

39
Complex Manifolds Section 5.6 SCHLICHENMAIER, Leytem

5.6.2 Example
Consider the space of matrices M = Mat(n × n, C) ∼ = Cn·n .
This is complex manifold with respect to the global coordinate chart ϕ : A = (aij )ij 7→ (a11 , a12 , . . . , ann ). But
M is not yet a group since there is no inverse : A−1 does not exist for any A ∈ M . So let GL(n, C) ( M :

B A ∈ GL(n, C) ⇔ A is invertible ⇔ det(A) 6= 0

Denote Y := { A ∈ M | det(A) = 0 } = det−1 {0} . det is continuous and holomorphic since it is a polynomial

function in the entries aij of A, so Y is closed in Cn·n and GL(n, C) = M \ Y 6= ∅. GL(n, C) is thus an open
subset of M . Moreover GL(n, C) is connected since it is path-connected. Indeed :
2
−k
−1

n\

Y = M ∩ Y = det {0} = fi {0}
i=1

which is (4.5) with U = M , n → n2 , k = n2 − 1 and f1 = det. Moreover if A = (aij )ij ∈ Y , then

∂(f1 ◦ ϕ−1 )
∂(f1 ◦ ϕ−1 ) ∂(f1 ◦ ϕ−1 )

 



J(det)(A) = (aij )ij (aij )ij ... (aij )ij 6= 0 (5.5)
∂a11 ∂a12 ∂ann

since the expression det(A) cannot be independent of all its coordinates aij . This shows that Y is a 1-complex
codimensional submanifold of M .

Remark :
det is a homogeneous polynomial of degree n since det(λ · A) = λn · det(A). Hence Euler’s relation holds :
n X
n
X ∂ det(A)
aij · = n · det(A)
i=1 j=1
∂aij

Now consider figure 5.12 : codimC Y = 1, which means that codimR Y = 2, so in the real picture one can always
connect any A, B ∈ GL(n, C) by a continuous path which does not intersect Y . Finally GL(n, C) is open and
connected in M , hence it is a complex manifold itself.
Note that GL(n, C) cannot be a submanifold of M since it is not closed. Moreover we recall that GL(n, C) carries
the structure of an affine variety of Cn·n .

Figure 5.12: GL(n, C) is path-connected, but GL(n, R) is not

The above argument is not valid in a real manifold X. If Y is of codimension 1 in X, then X \ Y is in general
disconnected. Consider again figure 5.12 : if for example A, B ∈ GL(n, R) with det(A) < 0 and det(B) > 0,
then any path relating A and B must pass through a matrix with determinant zero since there is no continuous
way to pass from negative to positive numbers in R without passing through 0. In C, this is however possible by
”going around” 0 using complex-valued determinants.
Figure 5.13: passing from negative to positive numbers without hitting 0 is not possible in R

40
Complex Manifolds Section 5.7 SCHLICHENMAIER, Leytem

5.6.3 Proposition


GL(n; C) , · with respect to matrix multiplication is a complex Lie group.
−1
Proof. First of all GL(n, C) is closed with respect to · and since for A, B ∈ GL(n; C), we have :
1
det A−1 =


det(A · B) = det(A) · det(B) 6= 0 , 6= 0
det(A)
1) For showing that · is holomorphic, we prove that ϕ ◦ · ◦ (ϕ−1 × ϕ−1 ) is holomorphic on Cn·n × Cn·n . Let
A, B ∈ GL(n, C) with A = (aij ), B = (bkl ) and C = A · B, so C = (crs ) where
n
X
ϕ ◦ · ◦ (ϕ−1 × ϕ−1 ) (a11 , . . . , ann , b11 , . . . , bnn ) = (c11 , . . . , cnn )


crs = ark bks ⇒
k=1

crs is just a polynomial expression in the entries aij and bkl , thus ϕ ◦ · ◦ (ϕ−1 × ϕ−1 ) and hence · are holomorphic.

2) For the inversion of A, recall that A−1 = 1

det(A) · t (Aad ) where Aad is the algebraic adjoint of A given by

(Aad )ij = (−1)i+j · det(Aij )

where Aij is the (n − 1) × (n − 1)–submatrix of A obtained by erasing line i and column j :
 
 
A= 
 ij 

For example in the case n = 2, this gives det(A) = ad − bc 6= 0 and

     
a b ad d −c −1 1 d −b
A= ⇒ A = ⇒ A = ·
c d −b a ad − bc −c a
Hence in a local coordinate, we obtain in general
1 1 1
A−1 t
(Aad ) · (Aad )ji = · (−1)i+j · det(Aji )



ij
= · ij
=
det(A) det(A) det(A)
det is a (non-vanishing) polynomial in aij and the entries in t (Aad ) are also polynomials, hence −1 is holomorphic.
So GL(n, C) is a complex Lie group of dimension n2 .
Remark :
In the real case, · and −1 are therefore real analytic operations ⇒ GL(n, R) is a real Lie group.
Similarly one can show that U(n, R) and SU(n, R) are also real Lie groups. However U(n, C) and SU(n, C) are
not complex Lie groups since the definitions of these involve the conjugate-transpose matrix A∗ , whose entries
are not holomorphic functions in the variables a11 , a12 , . . . , ann . A more general theorem even states that :
Any compact connected complex Lie group is holomorphically isomorphic to a torus T n . (no proof)

5.6.4 Exercise
Show that SL(n, C) = { A ∈ GL(n, C) | det(A) = 1 } is a closed submanifold of GL(n, C) of dimension n2 − 1.

SL(n, C) = det−1 {1} , hence SL(n, C) is closed in GL(n, C). If f : GL(n, C) → C : A 7→ det(A) − 1, then

SL(n, C) = f −1 {0}

which is (4.5) with U = GL(n, C), n → n2 and k = n2 − 1. Moreover rk J(f ) = 1 since f differs from det only
by a constant, hence J(f )(A) is the same expression as in (5.5).

Remark :
Since f is a polynomial expression, SL(n, C) is even an affine variety of Cn·n .

41
Complex Manifolds Section 5.7 SCHLICHENMAIER, Leytem

5.7 Grassmannians
This is in fact a generalization of the projective space Pn (C) :
Instead of considering the lines in Cn passing through the origin, one can also look at the set of linear subspaces
of Cn of dimension k ≤ n (for the projective space, k = 1 since lines are 1-dimensional).
By a similar construction, one then obtains the so-called Grassmannians, denoted by Gr(n, k). It consists of the
k-planes in Cn and will again define a compact complex manifold. In particular, Gr(n + 1, 1) = Pn (C).

42
Chapter 6

Sheaf of holomorphic functions

6.1 Global holomorphic functions on a compact complex manifold

6.1.1 Theorem
Let M be a (connected) compact complex manifold and f : M → C be a global holomorphic function on M .
Then f is constant, i.e. there are no non-constant globally holomorphic functions on a compact complex manifold.
Proof. f is holomorphic, hence continuous. So |f | is also continuous on M (compact). Therefore |f | takes its
maximum value on M : ∃ x0 ∈ M such that |f (x0 )| ≥ |f (x)|, ∀ x ∈ M . Let
S := x ∈ M f (x) = f (x0 ) = f −1 {f (x0 )}


Then S is a closed subset of M since f is continuous and the points are closed in Cn . S 6= ∅ because x0 ∈ S. We
want to show that S = M .
Consider figure 6.1. Let (Ux0 , ϕ = (z1 , . . . , zn )) be a local complex coordinate chart around and centered at x0
(which exists by 4.3.2), i.e. ϕ(x0 ) = 0 and we denote z := ϕ(x), ∀ x ∈ Ux0 . Let also F := f ◦ ϕ−1 and consider
gz : C −→ C : λ 7−→ gz (λ) = F (λ z)
where z ∈ ϕ(Ux0 ) ⊆ Cn is a parameter. By construction gz is holomorphic in the variable λ and it satisfies
gz (1) = F (z) = (f ◦ ϕ−1 ) ϕ(x) = f (x). Moreover |gz | takes its maximum value at λ = 0 because

|gz (0)| = |F (0)| = |(f ◦ ϕ−1 )(0)| = |f ϕ−1 (0) | = |f (x0 )|

Figure 6.1: centered coordinate chart at x0

Hence |gz (λ)| ≤ |gz (0)|, ∀ λ ∈ C. By the maximum principle of holomorphic functions in 1 variable (λ) :
The absolute value of a non-constant holomorphic functions cannot take its maximum value in the interior of its
domain of definition (see section 2.5.2). But |gz | takes its maximum at 0 ∈ C = C◦ , so gz must be constant. In
particular gz (0) = gz (1), ∀ z ∈ ϕ(Ux0 ). Now we let z vary, i.e. if ∀ z, z 0 ∈ ϕ(Ux0 ), we have
F (z 0 ) = gz0 (1) = gz0 (0) = f (x0 ) = gz (0) = gz (1) = F (z)

43
Complex Manifolds Section 6.1 SCHLICHENMAIER, Leytem

gz and gz0 are both constant and coincide at 0 (see figure 6.2), hence they are equal and

F (z 0 ) = F (z), ∀ z, z 0 ∈ ϕ(Ux0 ) ⇔ f (x0 ) = f (x), ∀ x, x0 ∈ Ux0

i.e. f (x) = f (x0 ), ∀ x ∈ Ux0 . f is constant in the coordinate neighborhood around x0 .

Figure 6.2: gz and gz0 coincide everywhere

Now let y ∈ M be arbitrary. Since the manifold M is connected, hence path-connected by 3.2.2, there is a
continuous path γ : [0, 1] → M such that γ(0) = x0 and γ(1) = y. Consider a covering of open sets in M of this
path (see figure 6.3), for example [
im γ ⊂ Ux
x∈im γ


where Ux denotes a small open neighborhood of x in M . But im γ = γ [0, 1] , so im γ is compact and we can
extract a finite subcovering open neighborhoods Uxi , i = 0, . . . , n.

Figure 6.3: im γ can be covered by finitely many Uxi

We hence know that f is constant on Ux0 . Ux0 ∩ Ux1 is non-empty and open in Ux0 ∪ Ux1 , hence by the Identity
Theorem (via using local coordinates), f is constant on Ux0 ∪ Ux1 since it is holomorphic. Iterating this process
(finitely many Uxi ) finally gives that f is constant on Ux0 ∪ . . . ∪ Uxn with y ∈ Uxn , i.e. f (y) = f (x0 ).
Since this can be done for any y ∈ M , we obtain that f is constant everywhere.

6.1.2 Theorem
1) There are no connected compact submanifolds in Cn but the points.
2) Any (not necessarily connected) compact submanifold of Cn only consists of finitely many points.
Hence there is no non-trivial compact complex substructure in Cn .

Proof. 1) Let M ⊂ Cn be a connected compact submanifold and i : M ,→ Cn the canonical injection. Then

M  / Cn /C
i pi

where i is holomorphic (since it the restriction of the identity) and pi : Cn → C are the coordinate functions,
which are also holomorphic. Thus the fi := pi ◦ i are holomorphic from M to C, hence all fi are constant by the
previous theorem since M is connected. This means that the coordinate evaluation on M always gives the same
value : ∀ z ∈ M , fi (z) = pi (z) = zi = ci for some ci ∈ C, i.e. all points in M have the same coordinate. This
implies that M can only consist of a single point : M = {c} = {(c1 , . . . , cn )}.
2) Now let M be an arbitrary compact submanifold of Cn . Since M is locallySconnected (see section 3.2.2), the
connected components Ci of M are open and closed and M writes as M = i Ci . By compactness, there is a
finite subfamily of Ci covering M . Since the components do not intersect, this implies that M can only have
finitely many connected components, which are in addition compact since they are closed. It follows from 1) that
every of these finitely many components is just given by a point, hence M consists of finitely many points.

44
Complex Manifolds Section 6.2 SCHLICHENMAIER, Leytem

6.1.3 Conclusion
As a consequence we conclude that Whitney’s Embedding Theorem does not apply to complex manifolds since
non-trivial compact complex manifolds of dimension k (which exist) cannot be embedded into C2k+1 . Moreover
we cannot deduce the structure of a compact complex (sub)manifold from its algebra of holomorphic functions
(since there are no interesting ones). For this we have to introduce the notion of a sheaf.

6.2 Sheaves : definitions

Let M be a complex manifold and U ⊆ M be open, so U is again a (not necessarily connected) complex manifold.
We denote 
O(U ) := f : U → C is a holomorphic function on U
This definition makes sense since holomorphic functions on U do not need to be restrictions of globally holomor-
phic functions. By convention O(∅) = {0}, the zero function.

6.2.1 Proposition
O(U ) is a C–algebra for all U ⊆ M open.
Proof. 0 ∈ O(U )
f, g ∈ O(U ) ⇒ f + g ∈ O(U ) and f ∈ O(U ), α ∈ C ⇒ α · f ∈ O(U ), so (O(U ), +) is a vector space over C
moreover f, g ∈ O(U ) ⇒ f · g ∈ O(U ) and · is compatible with +, thus (O(U ), ·) is a (commutative) ring
Hence is an algebra over C.

O(M ) denotes the C–algebra of globally holomorphic functions.

Let U, V ⊆ M be open such that V ⊆ U . If f is holomorphic on U , then f|V is holomorphic on V because
holomorphy is a local condition, i.e. f ∈ O(U ) ⇒ f|V ∈ O(V ).
Restriction of holomorphic functions can be seen as a map ρU
V : O(U ) → O(V ) : f 7→ f|V where O(U ) and O(V )
are both C–algebras. This restriction is compatible with the C–algebra structures, e.g.

(f + g)|V = f|V + g|V , (α · f )|V = α · f|V , (f · g)|V = f|V · g|V (6.1)

6.2.2 Definition
A sheaf F of abelian groups / vector spaces / rings / algebras is an assignment U 7→ F(U ) where U is an open set
in M and F(U ) is an abelian group / vector space / ring / algebra, such that ∀ V ⊆ U open, there are maps

ρU U
V : F(U ) → F(V ) : f 7→ ρV (f ) = f|V

called the restrictions morphisms, which are a family of homomorphisms of abelian groups / linear maps between
vector spaces / ring homomorphisms / algebra homomorphisms such that
1) ρU
U = idF (U ) , ∀ U ⊆ M open

2) for any open sets V ⊆ U ⊆ W , we have ρW U W

V = ρV ◦ ρU

For any open set U ⊆ M and any open covering {Ui }i∈J of U , the following conditions hold :
3) if f, g ∈ F(U ) such that ρU U
Ui (f ) = ρUi (g) ⇔ f|Ui = g|Ui , ∀ i ∈ J, then f = g

4) if fi ∈ F(Ui ) such that fi|Ui ∩Uj = fj|Ui ∩Uj , ∀ i, j ∈ J, then ∃ f ∈ F(U ) such that f|Ui = fi , ∀ i ∈ J

Remarks :
− By 3), the f in 4) is always unique. Consider figure 6.4 for an interpretation of 4).
− Note that f and g are not necessarily functions and that ρU V is not necessarily the restriction of functions.
− However, when considering functions with the usual restriction, then 1), 2) and 3) are always satisfied.
− If F only satisfies the conditions 1) and 2), then F is called a presheaf.

45
Complex Manifolds Section 6.2 SCHLICHENMAIER, Leytem

Figure 6.4: fi and fj coincide on Ui ∩ Uj

6.2.3 Theorem
The holomorphic functions on a complex manifold M define a sheaf of C–algebras.
Proof. because holomorphy is a local condition and the restriction is compatible with the C–algebra structure :
ρU U U
V (f + g) = ρV (f ) + ρV (g) , ρU U U
V (f · g) = ρV (f ) · ρV (g)

The sheaf of C-valued holomorphic functions on a complex manifold M is denoted by OM . Similarly one can
∞ ω
show that the differentiable and the analytic functions on M also define sheaves, denoted by CM and CM .

If M is a compact complex manifold, we know by 6.1 that OM (M ) = C. This is not true for C ∞ (M ).
Moreover OM (U ) can be non-trivial for some (non-compact) open subset U ⊂ M because if U is small enough,
it can be identified with some open subset of Cn , which has a lot of holomorphic functions.

6.2.4 Counter–example
Consider the complex plane M = C and B(U ), the C–algebra of bounded holomorphic functions on U ⊆ M open
together with the restriction of functions ρU V . Thus 1), 2), 3) are satisfied. But 4) is violated because boundedness
is not a local property. Indeed consider the case U = C and U is covered by {Ui }i∈N where

Ui = z ∈ C , |z| < i
S
are open, i.e. C = i∈N Ui . Let fi ∈ B(Ui ) be given by fi (z) = z. So any fi is bounded on Ui and the fi glue at
the intersections : fi|Ui ∩Uj = fj|Ui ∩Uj , ∀ i, j ∈ N. But there is no global holomorphic function which extends the
fi since B(C) = C : by Liouville, every bounded globally holomorphic (entire) function on C must be constant.
Hence no one of the fi can be extended to C since it is non-constant in any neighborhood of every point in Ui .
other argument :
If there is an extension f of the fi , it is necessarily of the form f (z) = z (this is the only candidate). But this f
is not bounded on C : @ K ∈ R such that |f (z)| = |z| ≤ K, ∀ z ∈ C. Finally
@ f ∈ B(C) such that f|Ui = fi , ∀ i ∈ N

6.2.5 Exercise
Let M = C and consider F : U 7→ F(U ) = J{0} (U ) where J{0} (U ) = O(U ) if 0 ∈
/ U and for 0 ∈ U , we set

J{0} (U ) = f ∈ OC (U ) | f (0) = 0 ( OC (U )
(strict inclusion because of the non-zero constant functions) together with the usual restriction of functions ρUV.
Show that F is a sheaf on M . It is called the vanishing sheaf of the point 0 ∈ C. More generally, one can also
consider the vanishing sheaf of arbitrary subsets of a more general manifold M (see section 6.5.2). S
The conditions 1), 2) and 3) are satisfied. So let U ⊆ C be open with an open covering U = i∈J Ui and let
fi ∈ J{0} (Ui ) such that fi|Ui ∩Uj = fj|Ui ∩Uj , ∀ i, j ∈ J.
If 0 ∈
/ U , then ∃ f ∈ J{0} (U ) = OC (U ) such that f|Ui = fi , ∀ i ∈ J since OC is a sheaf (see figure 6.5).
Hence we can assume that 0 ∈ U . ∀ i ∈ J, fi ∈ J{0} (Ui ) ⊂ OC (Ui ) ⇒ ∃ f ∈ OC (U ) such that f|Ui = fi , ∀ i ∈ J.
We have to check that this f satisfies f (0) = 0. 0 ∈ U ⇒ ∃ i0 ∈ J such that 0 ∈ Ui0 , hence
f (0) = f|Ui0 (0) = fi0 (0) = 0 since fi0 ∈ J{0} (Ui0 ) ⇒ f ∈ J{0} (U )

46
Complex Manifolds Section 6.4 SCHLICHENMAIER, Leytem

Figure 6.5: if 0 ∈
/ U , then J{0} coincide with the sheaf of holomorphic functions on U

6.2.6 Definition
Since OM is a sheaf of C–algebras, OM (U ) is in particular a commutative ring for any U ⊆ M open. Hence one
can consider modules over these rings.
A sheaf F with restrictions ρU
V is called a sheaf of OM –modules if F(U ) is a module over OM (U ) for any U ⊆ M
open and the module structure is compatible with the restrictions, i.e.

∀ U ⊆ M open , ∀ f ∈ F(U ) , ∀ h ∈ OM (U ) : ρU U U U
V (h ∗U f ) = ρV (h) ∗V ρV (f ) = h|V ∗V ρV (f ) (6.2)

where h ∗U f ∈ F(U ) and h|V ∗V ρU

V (f ) ∈ F(V ).

6.2.7 Examples
1) If F is a sheaf of OM –modules and M is compact, then F(M ) is a module over OM (M ) = C, i.e. F(M ) is a
vector space over C.
2) J{0} is a sheaf of OC –modules with the definition h ∗ f := h · f for h ∈ OC (U ), f ∈ J{0} (U ), U ⊆ C open. This
is well-defined because (h · f )(0) = 0 too, hence h · f ∈ J{0} (U ). Moreover this module structure is compatible
with the restrictions as in (6.1).
3) OM is a sheaf of OM –modules since any commutative ring OM (U ) can be considered as a module over itself.

6.3 Morphisms of sheaves

Let F and G be sheaves of the same type, i.e. both are sheaves of abelian groups / vector spaces / rings or algebras.
G
Denote the restrictions of F by ρFV U and those of G by ρV U . A morphism of sheaves ψ : F → G is a family
{ψU }U ⊆M of maps ψU : F(U ) → G(U ) indexed by the open sets in M such that if V ⊆ U are open :

F(U )
ψU
/ G(U )

ρF
VU ρG
VU
ρGV U ◦ ψU = ψV ◦ ρF
VU
 
F(V )
ψV
/ G(V )

and any ψU is a homomorphism of abelian groups / linear map between vector spaces / ring homomorphism or
algebra homomorphism.
If F and G are sheaves of OM –modules, it is in addition required that ∀ V ⊆ U open :

∗0U
OM (U ) × F(U )
∗U
/ F(U ) ψU
/ G(U ) o O(U )M × G(U ) OM (U ) (6.3)
ρF
VU ρG
VU ρO
VU
  ∗0V 
OM (V ) × F(V )
∗V
/ F(V ) ψV
/ G(V ) o OM (V ) × G(V ) OM (V )

where any ψU is a module homomorphism, i.e. ψU (h ∗U f ) = h ∗0U ψU (f ), ∀ h ∈ OM (U ), ∀ f ∈ F(U ).

47
Complex Manifolds Section 6.4 SCHLICHENMAIER, Leytem

6.4 Restriction of sheaves

6.4.1 Definition
Let F be a sheaf over a complex manifold M and U ⊆ M be open. The restriction of F to U , denoted by F|U ,
is defined by F|U (U 0 ) := F(U 0 ), ∀ U 0 ⊆ U open (hence U 0 is also open in M ).
The restriction morphisms of F|U are coming from the restrictions of F. Hence F|U is a sheaf on U since the
conditions 1), 2), 3) and 4) hold for any open set in M , thus also for U and open subsets of U .

Example :
If M is a complex manifold, U ⊆ M open, non-empty and connected, then (OM )|U = OU .

6.4.2 Definitions
∼ Ok = OM ⊕ . . . ⊕ OM as sheaves of
A sheaf F of OM –modules is called a free sheaf of rank k ∈ N0 if F = M
OM –modules. F ∼ k k
= OM means that F and OM are isomorphic in the category of sheaves of OM –modules. As a
consequence, we obtain that for all U ⊆ M open :

F(U ) ∼
= OM (U )k = OM (U ) × . . . × OM (U ) (6.4)

In general (6.4) is however not sufficient to say that F ∼ = OM ⊕ . . . ⊕ OM since the compatibility conditions with
the restrictions still need to be satisfied. In fact : F ∼
= OMk
⇔ (6.4) and (6.3).
It is however sufficient if the OM (U )–module F(U ) consists of functions and ρU V is the restriction of functions
since (6.3) is always satisfied for functions with usual restriction.

A sheaf F of OM –modules is called locally free of finite rank k

k
⇔ ∀ x ∈ M , ∃ k ∈ N0 and there is an open neighborhood U of x in M such that F|U ∼ k ∼
= OU = (OM )|U .
This k is necessarily the same for any x ∈ M :
Let x, y ∈ M with x ∈ U , y ∈ V , U, V ⊆ M open such that F(U ) ∼ = Ok (U ) and F(V ) ∼ = Ol (V ). M being
(path-)connected, let γ : [0, 1] → M be a continuous path from x to y. For any z ∈ im γ, consider an open
neighborhood Uz of z such that F(Uz ) ∼ = Okz (Uz ). im γ is compact, hence there are finitely many zi such that
[ n
[
im γ ⊂ Uz ⇒ im γ ⊂ Ui
z∈im γ i=1

with U1 = U , Un = V and F(Ui ) ∼

= Oki (Ui ), ∀ i ∈ {1, . . . , n}, k = k1 , l = kn . In particular, we have for example
F(U ) ∼
= Ok (U ) , F(U ∩ U2 ) ∼
= Ok (U ∩ U2 ) , F(U2 ) ∼
= Ok2 (U2 ) , F(U2 ∩ U ) ∼
= Ok2 (U2 ∩ U )
∼ Ok2 (U2 ∩ U ), which implies that k = k2 since the rank of a module is uniquely given. By
hence Ok (U ∩ U2 ) =
induction, we obtain that k = k2 = k3 = . . . = kn−1 = kn = l since there are only finitely many indices.

6.4.3 Examples
1) Let M = C. Then J{0} is a sheaf of OC –modules as shown in 6.2.7. Moreover J{0} is a free sheaf of rank 1 :
J{0} (C) = OC (C) · p where p : C → C, p(z) = z ⇒ p(0) = 0. This means that ∀ ϕ ∈ J{0} (C), ∃ ψ ∈ OC (C) such
that ϕ = ψ · p : ϕ has no constant term. ϕ is holomorphic on C, hence saying that ϕ ∈ J{0} (C) means that the
power series expansion of ϕ around 0 satisfies a0 = 0 :
∞
X ∞
X
k
ϕ(z) = ak · z = 0 + z · ak · z k ⇒ ϕ(0) = 0
k=0 k=1

In order to satisfy the compatibility conditions of the restrictions, we thus need that J{0} (U ) = OC (U ) · p|U for
all U ⊆ M open. But this is true, hence (6.4) is satisfied with k = 1 and we get J{0} ∼= OC .

2) Let M = P1 (C) with a = (0 : 1) 6= ∞, U ⊆ P1 (C) open and define


Ja (U ) := f ∈ OM (U ) if a ∈ U, then f (a) = 0

48
Complex Manifolds Section 6.5 SCHLICHENMAIER, Leytem

Ja is a sheaf on P1 (C) (same proof as for J{0} ), called the vanishing sheaf of a. It is locally free of rank 1 because
C∼= { (α : 1) | α ∈ C } and a = (0 : 1), so Ja (U ) ∼
= J{0} (U ) for all U ( P1 (C) open.
Ja is however not a free sheaf because there are no non-zero holomorphic functions on P1 (C) that vanish at a
given point :
OM P1 (C) = C , Ja P1 (C) = {0} ⇒ Ja (M ) ∼



6 OM (M )
=
Thus (6.4) is not satisfied, meaning that Ja cannot be free of rank 1.

other argument :
Assume that Ja (M ) ∼
= OM (M ). Since OM (M ) has rank 1 (basis given by f = 1), Ja (M ) must therefore also be
generated by 1 element over OM (M ), i.e. ∃ g ∈ OM (M ) such that Ja (M ) = OM (M ) · g. Since Ja (M ) = {0}, we
have g = 0. In order to satisfy the compatibility conditions of the restrictions, we thus need that
Ja (U ) = OM (M ) · g|U = OM (M ) · 0|U = {0}
which is not true since Ja (U ) 6= {0} for a small open set U ⊂ P1 (C) (because U locally looks like C).

6.5 Ideal sheaves

6.5.1 Definition
Let M be a complex manifold and OM be the sheaf of C-valued holomorphic functions on M .
A sheaf J on M is called an ideal sheaf if J is a subsheaf of OM , i.e. J (U ) ⊆ OM (U ), ∀ U ⊆ M open, denoted
by J ⊆ OM , such that J (U ) is an ideal in the commutative ring OM (U ) for any U .
This implies in particular that J is also a sheaf of OM –modules since every ideal is a module over the considered
ring by setting :
OM (U ) × J (U ) −→ J (U ) : (h, f ) 7−→ h ∗ f := h · f ∈ J (U )

6.5.2 Example
Let A ⊆ M be an arbitrary subset of the complex manifold M . For U ⊆ M open, we define

JA (U ) := f ∈ OM (U ) f (x) = 0, ∀ x ∈ A ∩ U
JA (U ) is an ideal in OM (U ) : for all h ∈ OM (U ), h · f still vanishes at any x ∈ A ∩ U if f ∈ JA (U ).
JA defines an ideal sheaf and is called the vanishing sheaf of A over OM .
Proof. The conditions 1), 2) and 3) are satisfied since we consider functions and usual restrictions.
S
Let U ⊆ M be open and U = i∈I Ui be an open covering. Denote Uij := Ui ∩ Uj .

For all i ∈ I, let fi ∈ JA (Ui ) such that fi|Uij = fj|Uij , ∀ i, j ∈ I. We have to show that ∃ f ∈ JA (U ) such that
f|Ui = fi , ∀ i ∈ I. By 3), the only candidate for functions with the canonical restriction is
f (x) := fi (x) if x ∈ Ui for some i ∈ I
a) f is well-defined since if x ∈ Ui ∩ Uj , then fi (x) = fi|Uij (x) = fj|Uij (x) = fj (x) = f (x).
b) f is holomorphic on U since holomorphy is a local condition, i.e. it is sufficient to check it on a small open
neighborhood of any point x ∈ U . But this is true since ∀ x ∈ U , ∃ i0 ∈ I such that x ∈ Ui0 and f|Ui0 = fi0 is
holomorphic on the open set Ui0 since fi0 ∈ OM (Ui0 ) by assumption.
c) check that f (x) = 0, ∀ x ∈ A ∩ U : S S
A∩U =A∩ i Ui = i (A ∩ Ui )
Hence if x ∈ A ∩ U , then ∃ i ∈ I (not necessarily unique) such that x ∈ A ∩ Ui , implying that
f (x) = fi (x) = 0 since fi ∈ JA (Ui )

49
Complex Manifolds Section 6.6 SCHLICHENMAIER, Leytem

6.5.3 Proposition
If A ⊆ M is open, A 6= ∅ and U ⊆ M is open and connected such that A ∩ U 6= ∅, then JA (U ) = {0}.
Proof. follows from the Identity Theorem : If f ∈ OM (U ) where U is connected and contains the non-empty
open set A ∩ U on which f vanishes, then f vanishes on the whole open set U .

Remark :
This result does not have much applications in practise since one often requires that A has to be closed, e.g.

M = C2 , A = {(0, 0)} ⇒ JA (U ) = f ∈ OM (U ) f (0, 0) = 0 if (0, 0) ∈ U = J{(0,0)} (U )



6.6 Germs and stalks

Let M be a complex manifold and x ∈ M be fixed.
The idea of germs and stalks is to consider holomorphic functions in a (non-determined) neighborhood of x.

6.6.1 Definition
Let U and U 0 be open in M such that x ∈ U ∩ U 0 and let g ∈ OM (U ), f ∈ OM (U 0 ). We define

(g, U ) ∼x (f, U 0 ) ⇔ ∃ W ⊆ U ∩ U 0 open, x ∈ W such that g|W = f|W

i.e. two functions with their corresponding domains are equivalent with respect to x if there exists a smaller open
neighborhood of x on which both functions concide.

∼x is an equivalence relation (reflexivity and symmetry are clear).

− transitivity : if g|W = f|W and f|W 0 = h|W 0 , then g|W ∩W 0 = h|W ∩W 0 where W ∩ W 0 is open and non-empty
We denote the equivalence class simply by [f ] = fx := (g, U ) (f, U 0 ) ∼x (g, U ) ; the domain of f does not


need to be specified. Such an equivalence class is called a germ of holomorphic functions. And the set of all
germs of holomorphic functions (the set of all equivalence classes) is called the stalk at x and denoted by

OM,x = fx | f holomorphic in some open neighborhood of x

We point out that germs do not have a domain of definition, but the representatives of a given germ have one.

This construction can also be done for arbitrary sheaves :

Let F be a sheaf on M and U, V ⊆ M be open. Fix x ∈ U ∩ V and let f ∈ F(U ), g ∈ G(V ). Then

f ∼x g ⇔ ∃ W ⊆ U ∩ V open, x ∈ W such that ρU V

W (f ) = ρW (g)

The germs are again given by the fx and the stalk at x (the set of all germs) is denoted by Fx .

6.6.2 Proposition
Fx has the same algebraic properties as the ”objects of F”, i.e. as the F(U ) where U ⊆ M is open.
Proof. Consider for example the case where F(U ) is an abelian group. We have to show that Fx can also be
endowed with an abelian group structure. Let U, V ⊆ M open, x ∈ U ∩ V , f ∈ F(U ), g ∈ F(V ) and set
 
[f ] + [g] := [f + g] := f|U ∩V + g|U ∩V (6.5)

where f|U ∩V , g|U ∩V ∈ F(U ∩ V ) since the sum of f and g can only be defined in a smaller neighborhood.

50
Complex Manifolds Section 6.7 SCHLICHENMAIER, Leytem

This definition is also independent of the representing elements of f and g. Let f 0 ∼x f and g 0 ∼x g with
0 0 0
f|W = f|W and g|W 0 = g|W 0 . Since x belongs to all of these open set, we have that A := U ∩ V ∩ W ∩ W 6= ∅ :

0 0
f|A = f|A and g|A = g|A
0 0
⇒ f|A + g|A = f|A + g|A ⇒ f 0 + g 0 ∼x f + g

Finally definition (6.5) implies that [f ]+[g] has exactly the same properties as f +g, so Fx is an abelian group.

Remark :
A more sophisticated way to prove this result is to use the concept of a filtrant inductive limit. Indeed :

Fx = lim
−→ F(U )
U 3x

which intuitively means that we take F(U ) and let U → {x}. Inductive limits preserve the properties of F(U ).

6.6.3 Corollary
OM,x is a C–algebra for all x ∈ M . In particular it is a commutative ring.
Proof. follows from proposition 6.6.2 since OM (U ) is a C–algebra for all U ⊆ M open (as showed in 6.2.1)

6.6.4 Proposition
∀ x ∈ M , OM,x is a local ring, i.e. a ring which contains only one maximal ideal.
Proof. Recall that a ring R is a local ring ⇔ R \ R× is an ideal. We define

Mx := [f ] ∈ OM,x f (x) = 0

Mx is well-defined since if g ∼x f and f (x) = 0, then g(x) = 0 too since x belongs to any open neighborhood
around x : all representatives of [f ] have the same value at x.
Moreover Mx is an ideal in OM,x with the definitions [f1 ] + [f2 ] = [f1 + f2 ], ∀ [f1 ], [f2 ] ∈ Mx as above and
[g] · [f ] = [g · f ], where [g] ∈ OM,x and the product g · f is done in the neighborhood U ∩ V if the representing
elements are f ∈ OM (U ), g ∈ OM (V ) because f (x) = 0 ⇒ (g · f )(x) = 0 too.
It remains to show that the elements in OM,x \ Mx are invertible. Let [g] ∈ / Mx ⇒ g(x) 6= 0, ∀ g ∈ [g].
Let (g, U ) be a representing element of [g]. Since g is holomorphic, hence continuous on U , ∃ U 0 ⊆ U open such
that x ∈ U 0 and g(y) 6= 0, ∀ y ∈ U 0 . Thus g1 is well-defined and holomorphic on U 0 . Moreover

[g]−1 =
1 1
g since [g] · g = [1]
1 0 ×


Thus g,U satisfies all the conditions and [g] is invertible in OM,x . This shows that OM,x \ Mx ⊆ OM,x .
×
And OM,x ⊆ OM,x \ Mx because Mx 6= OM,x . If an element in Mx would be invertible, then [1] ∈ Mx , hence
Mx = OM,x because it is an ideal. This contradiction finishes the proof.
Remark :
This does not hold for arbitrary Fx (since not any Fx must be a ring).

6.7 Remarks on sheafification

F
Assume that G is only a presheaf on M . Then Gx also exists for all x ∈ M y Gy . Set |G| := Gx and consider
x∈M
∀ x ∈ M , px : Gx → {x} : [g] 7→ x
|G| is defined as the disjoint union of all the Gx and called the total space of presheaves. Hence all the maps px
induce a map P : |G| → M : [g] 7→ x where x is the unique x such that [g] ∈ Gx (see figure 6.6).
We endow |G| with the initial topology with respect to P , i.e. the smallest topology that makes P continuous.
This topology hence consists of all the preimages under P of open sets in M .

51
Complex Manifolds Section 6.7 SCHLICHENMAIER, Leytem

Figure 6.6: for any [g] ∈ |G| there is a unique x ∈ M such that [g] ∈ Gx

For U ⊆ M open, we then define

Ĝ(U ) := s : U → P −1 (U ) ⊂ |G| such that s is continuous and P ◦ s = idU }



Such an element s ∈ Ĝ(U ) is called a continuous section on U .

One can show that Ĝ is now a sheaf ; it is the associated sheaf of G. And if G was already a sheaf, then Ĝ = G.

52
Chapter 7

Meromorphic functions

7.1 Construction of meromorphic functions

7.1.1 Proposition
Let M be a complex manifold and U ⊆ M be open and connected.
Then OM (U ) is an integral domain, i.e. if f, g ∈ OM (U ) are such that f · g = 0 on U , then f = 0 or g = 0.
Proof. Let f · g = 0 and assume that f 6= 0. We have to show that g = 0 on U , i.e. g(z) = 0, ∀ z ∈ U .
We first consider the case where U is completely contained in some chart domain V . Let (ϕ, V ) be the associated
coordinate chart. f 6= 0 ⇒ ∃ z0 ∈ U such that f (z0 ) 6= 0. f continuous ⇒ ∃ W ⊆ U open such that z0 ∈ W
and f (z) 6= 0, ∀ z ∈ W . By bijectivity of ϕ, W is in 1-to-1 correspondence with ϕ(W ), i.e. any z ∈ W ⊆ U ⊆ V
can uniquely be written as z = ϕ−1 (x) for some x ∈ ϕ(W ) ⊆ ϕ(U ) ⊆ ϕ(V ).

∀ z ∈ W : 0 = (f · g)(z) = f (z) · g(z) = f (z) · (g ◦ ϕ−1 )(x) with f (z) 6= 0

−1
⇒ (g ◦ ϕ )(x) = 0 , ∀ x ∈ ϕ(W ) ⊆ ϕ(U )

g ◦ ϕ−1 is holomorphic on ϕ(U ) ⊆ Cn , which is open and connected (as image of U , which is connected), and
vanishes on ϕ(W ) ⊆ ϕ(U ), which is open and non-empty since ϕ(z0 ) ∈ ϕ(W ). Thus g ◦ ϕ−1 = 0 on ϕ(U ) by the
Identity Theorem and since ϕ is bijective this exactly means that g = 0 on U .
S
Now let U ⊆ M be covered by chart domains : U = i∈I Vi . f 6= 0 ⇒ ∃ z0 ∈ U and ∃ i0 ∈ I such that z0 ∈ Vi0
with f (z0 ) 6= 0. Moreover ∃ W ⊆ Vi0 ⊆ U open such that z0 ∈ W and f (z) 6= 0, ∀ z ∈ W . By the same argument
as above, we conclude that g = 0 on Vi0 . Let y ∈ U be arbitrary. We have to show that g(y) = 0 too.
Since U is connected (hence path-connected), there is a continuous path γ : [0, 1] → M such that γ(0) = z0
and γ(1) = y. im γ is compact and covered by the Vi , hence it can be covered by finitely many of them, say
Vi0 , V1 , . . . , Vn with y ∈ Vn . We know that g = 0 on Vi0 , hence it is also zero on Vi0 ∩ V1 , which is non-empty
and open in V1 , hence (again via local coordinates), g = 0 on V1 . We repeat this argument until Vn , so finally
g = 0 on Vn and g(y) = 0. Hence g = 0 on U since y ∈ U was arbitrary.

53
Complex Manifolds Section 7.1 SCHLICHENMAIER, Leytem

Remark :
∞ ω
This is not true for CM or CM if M is a real differentiable manifold. Consider e.g. the following example :
∞
Figure 7.1: CM is not an integral domain

f (x) = 0, ∀ x > 0 and g(x) = 0, ∀ x < 0 ⇒ f · g = 0 on R, but f 6= 0 and g 6= 0

7.1.2 Definition 1
Let U ⊆ M be open and connected. Since OM (U ) is an integral domain, it does not contain zero divisors, hence

M(U ) := Quot OM (U ) = S −1 OM (U )

exists and is a field, where S = OM (U ) \ {0} is a multiplicative set. Its elements are called meromorphic ”func-
tions” on U . This definition however implies a certain number of problems.
f
1) An element h̄ ∈ M(U ) is given by an equivalence class h̄ = g where f, g ∈ OM (U ), but it is not a function.

f f0
Recall : ∼ 0 ⇔ f · g0 − f 0 · g = 0
g g

If h̄ is represented by f and g, we may define

f (z)
h : U → C , h(z) := (division)
g(z)

First of all, this is only a function on U \ V (g) where V (g) = { z ∈ U | g(z) = 0 } is the zero set of g.
0
Problem : This is not independent of the representatives f, g since maybe V (g) ( V (g 0 ) if fg ∼ fg0 .
At the moment we are not yet able to solve this problem. For this we need the notion of analytic sets (this will
be explained in section 7.4).

2) For n > 1, putting h(z) := ∞ if z ∈ V (g) does not work, see example 7.1.3.

3) M is in general not a sheaf. Consider an open subset U which is not connected, i.e. U = U1 ∪ U2 , U1 ∩ U2 = ∅
for U1 , U2 ⊆ M open. Then OM (U ) contains zero divisors, e.g.
( (
1 if z ∈ U1 0 if z ∈ U1
f1 (z) = , f2 (z) =
0 if z ∈ U2 1 if z ∈ U2

f1 ∈ OM (U ), f2 ∈ OM (U ), f1 6= 0, f2 6= 0, but f1 · f2 = 0.
But Quot(R) is not defined if R is not an integral domain (since R \ {0} is not multiplicative), hence the mapping

U 7−→ M(U )

is not even well-defined and M cannot be a sheaf.

In order to solve this problem, one can do the following (which we do not develop in detail) :
Since the manifold M is locally connected, it admits a basis consisting of connected open sets, on which the
mapping U 7→ M(U ) makes sense. This can be extended to all open sets and M will define a presheaf on M .
And finally we may take as M the sheaf associated to M.

54
Complex Manifolds Section 7.2 SCHLICHENMAIER, Leytem

7.1.3 Example
n = 1 : in local coordinates, 2 holomorphic functions f and g on M can locally be written as
f (z) = (z − z0 )k · fˆ(z) , g(z) = (z − z0 )l · ĝ(z)
fˆ(z)
where z0 ∈ C, fˆ(z0 ) 6= 0, ĝ(z0 ) 6= 0 and k, l ∈ Z can be zero. Then one defines h(z) := (z − z0 )k−l · .
ĝ(z)
− if k > l, h is a holomorphic function with a zero at z0
− if k = l, h is a holomorphic function with with a non-zero value at z0
− if k < l, then h is not holomorphic at z0 , but we can put h(z0 ) := ∞
In this last case, we say that z0 is a pole of order k − l of h. Moreover one can show that the definition of h is
independent of the chosen coordinates.

n > 1 : poles are in general not well-defined for n > 1

Consider M = C2 and f (z) = f (z1 , z2 ) = zz12 ⇒ V (z2 ) = { (z1 , 0) | z1 ∈ C }. Let x = (z1 , 0) ∈ V (f ) ; we
approach x from different directions (see figure 7.2). Let α and β be zero-sequences and set

z1 + α(n) z1 α(n)
xn := z1 + α(n), 0 + β(n) ⇒ f (xn ) = = +
β(n) β(n) β(n)
z1
If z1 6= 0, then β(n) → ∞, hence |f (xn )| → ∞, so the points (z1 , 0), z1 6= 0 could be considered to be poles.
α(n)
However for z1 = 0 and x = (0, 0), we get f (xn ) = β(n) . By definition lim f (z) exists if we obtain the same limit
z→0
for any zero-sequence. But this is not satisfied here, e.g. for c ∈ C, let
c 1
α(n) =
, β(n) = ⇒ f (xn ) = c −→ c as n → +∞
n n
1 1 1
α(n) = 2 , β(n) = ⇒ f (xn ) = −→ 0 as n → +∞
n n n
1 1
α(n) = , β(n) = 2 ⇒ f (xn ) = n −→ ∞ as n → +∞
n n
We conclude that by approaching (0, 0) from different directions, one can obtain any complex value, including 0
and ∞. Hence putting f (0, 0) := ∞ is not a good choice.
Figure 7.2: approaching the point x from different directions

7.1.4 Conclusion
If h̄ ∈ M(U ), then h̄ = f
g and one can define h(x) := fg(x)
(x)
as a function as long as

x∈ / V (g) = y ∈ U g(y) = 0
V (g) is closed since g is holomorphic, hence continuous. We assume that g has no essential singularities.
0
A meromorphic function is only a (holomorphic) functions outside of this zero set. Moreover if fg ∼ fg0 where g 0
has other zeros than g, we have to look for the smallest set of points we have to take out.

7.2 Definition using UFDs

Recall :
Let m, n ∈ N and d = gcd(n, m).
n and m are relatively prime ⇔ d = 1 ⇔ n and m do not have common prime factors.
Z is a unique factorization domain (”unique” means up to order of the factors and multiplication by units).

55
Complex Manifolds Section 7.2 SCHLICHENMAIER, Leytem

7.2.1 Theorem
If M is a complex manifold and x ∈ M , then the local rings OM,x are UFDs. No proof will be given.

Hence fx , gx ∈ OM,x are relatively prime ⇔ ∃ U, V, W ⊆ M open, f ∈ OM (U ), g ∈ OM (V ) such that

W ⊆ U ∩ V and f|W , g|W are relatively prime (i.e. they do not have common factors).
Given g ∈ OM (U ), we can decompose g as g = p · g1 · . . . · gl where all factors are holomorphic, p is a unit and
all gi are indecomposable (up to units). The units are the holomorphic functions f without zeros in U (since the
inverse f1 is then still holomorphic in U ).
Example : for c ∈ C \ {0} , z1 z23 · c = c · z1 · z2 · z2 · z2 and c is a unit.

7.2.2 Definition
A Weierstrass polynomial of degree m is a function W : Cn → C of the form
m−1
X
W (z1 , . . . , zn ) = znm + aj (z1 , . . . , zn−1 ) · znj
j=0

where the aj are holomorphic functions in a neighborhood of (0, . . . , 0) ∈ Cn−1 and aj (0, . . . , 0) = 0. Hence a
Weierstrass polynomial is a monic polynomial in the variable zn whose coefficients are holomorphic functions in
the remaining variables and vanish at the origin.

7.2.3 Weierstrass preparation theorem

Let p ∈ Cn and f : U → C be a holomorphic function on an open neighborhood U ⊆ Cn of p = (p1 , . . . , pn ). Let
k be the order of pn as a zero of f (p1 , . . . , pn−1 , · ) (where k = 0 is possible ; k ≥ 1 means that f (p) = 0).
Then locally around p (e.g. in some small polydisc) f writes uniquely as f (z) = h(z) · W (z − p) where h is a
holomorphic function in a neighborhood of p with h(p) 6= 0 and W is a Weierstrass polynomial of degree k.
Example : if k = 0, then one can choose h = f and W ≡ 1.

7.2.4 Definition 2
Let U ⊆ Cn be an arbitrary open subset. We define

M(U ) := objects that are locally given as quotients of 2 holomorphic functions
gi
i.e. ∀ f ∈ M(U ), there exists an open covering {Ui }i∈J of U such that f|Ui = hi where gi , hi ∈ OM (Ui ) are
relatively prime and gi · hj = gj · hi on Ui ∩ Uj , i.e.
g  g 
i j
= , ∀ i, j ∈ J (7.1)
hi |Ui ∩Uj hj |Ui ∩Uj

This is an alternative definition of meromorphic functions on U . Note that it suffices to define M(U ) for open
sets in Cn instead of open sets in a general complex manifold M since the definition of M only depends of the
local structure of U and M locally looks like Cn .
M with restriction of functions defines a sheaf since all conditions are local, so 3) and 4) are immediately satisfied.

In general, M(U ) 6= M(U ) for U ⊆ M open and connected. Let e.g. M be a compact complex manifold. Then


OM (M ) = C ⇒ M(M ) = Quot OM (M ) = Quot(C) = C

but M(M ) ( M(M ) since M(M ) is given by quotients of locally holomorphic functions (no global condition).

Proposition : (no proof)

If U ⊆ M is contractible (in particular connected), then M(U ) = M(U ).

56
Complex Manifolds Section 7.4 SCHLICHENMAIER, Leytem

7.2.5 Remark
Elements in M(U
S ) can also be interpreted as functions outside of a certain set A. Let f ∈ M(U ) with an open
covering U = i Ui ⇒ f|Ui = hgii and f is a well-defined (holomorphic) function on Ui \ V (hi ). ∀ x ∈ Ui ∩ Uj :
g  g 

i j

f|Ui |Uj (x) = (x) = (x) = f|Uj |Ui (x) ⇔ gi · hj = gj · hi on Ui ∩ Uj
hi |Uij hj |Uij
S
because of (7.1). Now let A := i V (hi ). Using local compactness of U and the fact that gi and hi are relatively
prime (all common factors are already taken out, which is possible in a UFD, so hi may only change by a unit),
one can show that A 6= U , hence f is a holomorphic function on U \ A 6= ∅.

7.3 Definition using exceptional sets

In this approach, we are looking for the smallest possible set A which must be removed in order to define
meromorphic functions on a complex manifold M that are holomorphic on M \ A.

7.3.1 Definition 3
A meromorphic function on a complex manifold M is a pair (A, f ) where A ⊆ M , f is a holomorphic function
on M \ A and A is minimal, i.e. ∀ x0 ∈ A, there exist an open neighborhood U of x0 and holomorphic functions
g, h on U such that
a) A ∩ U = V (h) = { x ∈ U | h(x) = 0 }
b) the germs gx0 and hx0 are relatively prime (as defined in 7.2.1), i.e. have no common factors except the units
g(x)
c) f (x) = h(x) , ∀x ∈ U \ A
A is called the exceptional set of f . In particular, f is holomorphic on M ⇔ A = ∅.

7.3.2 Remarks
These conditions already imply that A cannot be open (in particular A 6= M ), otherwise V (h) = A ∩ U is open,
non-empty since it contains x0 and h = 0 on A ∩ U ⊆ M , hence h = 0 on M by the Identity Theorem (via local
coordinates) since h is holomorphic on the non-empty open set A ∩ U ⊆ U .
A is minimal means that A ∩ U is exactly equal to the vanishing set of h, i.e. exactly all the zeros of h have to
be removed so that f is well-defined. This is the case because gx0 and hx0 are relatively prime, i.e. g cannot
compensate a vanishing factor of h in the denominator.
The functions g and h may differ from neighborhood to neighborhood, but they always need to define the same
g
function on non-empty intersections : if f = hgii on Ui \ A and if f = hjj on Uj \ A, then we need
gi gj
gi (x) · hj (x) = gj (x) · hi (x) , ∀ x ∈ Ui ∩ Uj ⇒ f= = on (Ui ∩ Uj ) \ A
hi hj

7.4 Analytic sets

7.4.1 Lemma
For n = 1, the zero set V (f ) of a non-constant holomorphic function f on an open set U is a discrete set, i.e.
∀ z ∈ V (f ), there is an open neighborhood V of z such that V ∩ V (f ) = {z}.
Proof. Assume that V (f ) ⊆ U is not discrete. Then ∃ z0 ∈ V (f ) such that any open neighborhood of z0 intersects
V (f ) in some point which is distinct from z0 , i.e. z0 is an accumulation point of V (f ) (see section 2.4). But
since f = 0 on V (f ) and f is holomorphic on U , the Identity Theorem for n = 1 implies that f = 0 on U , which
contradicts the fact that f is non-constant. Hence V (f ) must be a discrete set.

Remark :
This does not hold for n ≥ 2. Consider e.g. f (z1 , z2 ) = z1 ⇒ V (f ) = { (0, z2 ) | z2 ∈ C } is not discrete.

57
Complex Manifolds Section 7.4 SCHLICHENMAIER, Leytem

7.4.2 Definition
Let M be a complex manifold. A ⊆ M is called an analytic set if ∀ z ∈ M , there exist an open neighborhood U
of z and ∃ f1 , . . . , fk ∈ OM (U ) such that (see figure 7.3)
k
\
fj−1 {0}


A∩U = x∈U f1 (x) = f2 (x) = . . . = fk (x) = 0 = (7.2)
j=1

Hence analytic sets locally look like the zero set of finitely many holomorphic functions. These functions are not
uniquely determined by A. Moreover k may depend on z since the functions fi are in general not the same for
any z ∈ M . Note that (7.2) is trivially satisfied if A ∩ U = ∅ by taking f1 ≡ 1 ∈ OM (U ). If V ⊆ M is open and
connected, we can also talk about analytic sets in V .
Figure 7.3: A ⊆ M is an analytic set

Equivalently, A is an analytic set if and only if there is an open covering {Ui }i∈J of M such that ∀ i ∈ J,
∃ f1i , . . . , fki i ∈ OM (Ui ) satisfying A ∩ Ui = { x ∈ Ui | f1i (x) = . . . = fki i (x) = 0 }.
Note that the ki are not uniquely given since there may be many possibilities to describe A ∩ Ui as a zero set.
But one can always extract a minimal number of functions which are necessary and sufficient to describe A ∩ Ui .
In particular if A is connected, then it is possible to choose ki = kj , ∀ i, j ∈ J.

7.4.3 Examples
1) Zero sets of holomorphic functions are by definition analytic sets (take U = M ).
2) Algebraic varieties in Cn are analytic sets since they are given by the zero set of finitely many polynomials,
which are globally holomorphic functions on Cn .
3) The exceptional set A of a meromorphic function (A, f ) is always an analytic set by condition a). In particular
if A 6= ∅, then any exceptional set is a 1-codimensional analytic subset of M (since there is just 1 function).
4) For an analytic set, A = M is possible by taking the zero function everywhere : A ∩ M = { x ∈ M | 0(x) = 0 }.

5) There are analytic sets which are not of codimension 1. Consider e.g. M = C2 and A = {(0, 0)}. In or-
der to show that A is an analytic set, we have to find an open covering {Ui }i∈I of M and holomorphic functions
f1i , . . . , fki i ∈ OM (Ui ) such that A ∩ Ui = V (f1i , . . . , fki i ) ∩ Ui .
Take i = 1, U1 = C2 , f1 (z) = z1 and f2 (z) = z2 for z = (z1 , z2 ) ∈ C2 . Hence f1 , f2 ∈ OM (C2 ) and

V (f1 , f2 ) ∩ U1 = z ∈ C2 f1 (z) = f2 (z) = 0 = (z1 , z2 ) ∈ C2 z1 = z2 = 0 = {(0, 0)} = A = A ∩ U1

 

Thus A is an analytic set, but it is not of codimension 1 in C2 .

6) Points, a finite number of points and, more generally, any discrete set is an analytic set.
Let M be a complex manifold and A ⊆ M be discrete, i.e. ∀ a ∈ A, there is an open neighborhood Va of a such
that Va ∩ A = {a}. By choosing the Va small enough, we may assume that (Va , ϕa ) is a coordinate chart around
a. Hence one canS take as open covering of M the collection V = {U, Va | a ∈ A} where U ⊆ M is open and chosen
such that U ∪ a∈A Va = M and A ∩ U = ∅. This is possible since M is Hausdorff, e.g.
[ [
U= Ux where F = M \ Va
x∈F a∈A

and Ux is a small open neighborhood of x ∈ F not containing any points from A (see figure 7.4).

58
Complex Manifolds Section 7.4 SCHLICHENMAIER, Leytem

Figure 7.4: U and the Vai form an open covering of M

Since A ∩ U = ∅, we can take the constant function 1 on U . And on Va we may consider the holomorphic function

f a (x) = ϕa (x) − ϕa (a) ⇒ f a (x) = 0 ⇔ ϕa (x) = ϕa (a) ⇔ x = a

since ϕa is bijective. Hence f a ∈ OM (Va ) and A ∩ Va = {a} = { x ∈ Va | f a (x) = 0 } = V (f a ) ∩ Va , ∀ a ∈ A.

7) But not all analytic sets are discrete, as e.g. zero sets of holomorphic functions in several variables. Now we
show that such a zero set cannot ”end” somewhere.
Figure 7.5: A = V (f ) cannot ”end” at a certain point

Let U ⊆ M open, f ∈ OM (U ) and assume that A = V (f ) is ”bounded in U ”. Consider an ”endpoint” a ∈ A

and a chart (V, ϕ) around it (see figure 7.6). In the local situation ϕ(V ) ⊆ Cn , we obtain that ϕ(A ∩ V ) is closed
in ϕ(V ) since ϕ is a homeomorphism and zero sets of holomorphic functions are closed. Moreover

ϕ(A ∩ V ) = ϕ(A) ∩ ϕ(V ) ⇒ closed in ϕ(V )

  
= ϕ(x) ∈ ϕ(V ) f (x) = 0 = z ∈ Cn (f ◦ ϕ−1 )(z) = 0 = V (f ◦ ϕ−1 )

ϕ(A ∩ V ) = ϕ x∈V f (x) = 0

Hence we can choose a polydisc ∆ in Cn such that ϕ(A ∩ V ) ( ∆ ( ϕ(V ) and denote V 0 := ϕ(V ) \ ∆. Let
1
g(z) := , ∀z ∈ V 0
(f ◦ ϕ−1 )(z)

g is well-defined and holomorphic in V 0 since V (f ◦ ϕ−1 ) ⊂ ∆. A consequence of Hartog’s Lemma (section 2.6.4)
then states that g can be holomorphically extended on the polydisc ∆, i.e. g extends to a holomorphic function
on ϕ(V ). By uniqueness of this extension we obtain that g must exist everywhere on ∆, which means that f ◦ϕ−1
cannot have zeros on ∆ and hence that f has no zeros in A ∩ V : contradiction since a ∈ A ∩ V and f (a) = 0.

Figure 7.6: the local situation at the ”endpoint”

Remark :
Such an argument does not apply if A is ”not bounded” in U because if such a polydisc ∆ exists, then ϕ(V ) \ ∆
is not connected and Hartog’s Lemma does not apply (see figure 7.7).

59
Complex Manifolds Section 7.4 SCHLICHENMAIER, Leytem

Figure 7.7: ϕ(V ) \ ∆ is no longer connected

7.4.4 Exercise
Show that any submanifold of a complex manifold is an analytic set.
This directly follows from characterization (4.5). In fact, being a submanifold is a stronger condition than being
an analytic set since the maximal rank condition also needs to be satisfied.
The converse is false : not any analytic set A defines a complex submanifold since A may have singularities, i.e.
points at which the rank of the Jacobian matrix drops (hence the maximal rank condition is not satisfied).

7.4.5 Proposition
Let M be a complex manifold, U ⊆ M be open and A be an analytic set in U . Then A is closed in U .
Proof. Recall that M , hence U , is by definition a topological space. However since we require that the coordinate
charts are homeomorphisms, this topology must be locally homeomorphic to the standard topology of Cn .
If A = U , then A = A ∩ U = { x ∈ U | 0(x) = 0 } : closed in U . Hence we may assume that A 6= U . We have to
show that U \ A is open in U .
Let x0 ∈ U \ A. Since A is an analytic set, we know that there is a neighborhood V of x0 in U such that

A ∩ V = x ∈ V f1 (x) = . . . = fk (x) = 0

for some suitable fi ∈ OM (V ), i ∈ {1, . . . , k}. Not all i satisfy fi (x0 ) = 0 since otherwise x0 ∈ V ∩ A ⊆ A.
Assume for example that f1 (x0 ) 6= 0. Since f1 is holomorphic on V , we get by continuity that ∃ W ⊆ V open
such that x0 ∈ W and f1|W 6= 0, i.e. f1 (y) 6= 0, ∀ y ∈ W . Thus y ∈ / V ∩ A, ∀ y ∈ W ⊆ V ⇒ y ∈ / A, ∀ y ∈ W .
Hence W ∩ A = ∅ (see figure 7.8), which precisely means that W ⊆ U \ A with x0 ∈ W .
Figure 7.8: W is an open neighborhood of x0 not intersecting A

7.4.6 Recall
Let X be a topological space and A ⊆ X. The topological boundary of A is given by

∂A := Ā ∩ X \ A = x ∈ X ∀ U ⊆ X open set containing x, ∃ y, z ∈ U such that y ∈ A and z ∈
/A

Moreover we have the relations

A◦ ⊂ A ⊂ Ā , Ā = A ∪ ∂A , Ā = A◦ ∪˙ ∂A , A◦ = A \ ∂A , X \ A = X \ A◦ , ∂A◦ ⊆ ∂A

where A◦ = { x ∈ X | ∃ U open neighborhood of x such that U ⊆ A } contains all interior points of A. A◦ is

always open and ∂A is always closed, but may be different from ∂A◦ = ∂(A◦ ).
In particular : ∂∅ = ∅ and ∂X = ∅, but ∂A 6= ∅, ∀ A ∈
/ {∅, X}. Any non-trivial subset of X has boundary points.

7.4.7 Proposition
Let A be an analytic set in M such that A 6= M . Then A has empty interior, i.e. A◦ = ∅.

60
Complex Manifolds Section 7.5 SCHLICHENMAIER, Leytem

Proof. We already know that A is closed, hence Ā = A. Assume that A◦ 6= ∅. Then ∂A◦ 6= ∅ too because

A 6= M ⇒ A◦ ( M ⇒ A◦ ⊂ A ( M ⇒ A◦ ⊂ Ā = A ( M with A◦ = (A◦ )◦ ∪˙ ∂(A◦ ) = A◦ ∪˙ ∂A◦

If ∂A◦ = ∅, then A◦ = A◦ , which means that A◦ is closed and open. Since A◦ 6= ∅ and M is connected, this
implies that A◦ = M ⇒ A = M , which is a contradiction to our hypothesis. So let x0 ∈ ∂A◦ ⊂ A◦ ⇒ any
open neighborhood Ux0 of x0 intersects A◦ : Ux0 ∩ A◦ 6= ∅ and this intersection is still open.
Now take such a neighborhood U = Ux0 which is connected and small enough (chart domain) such that

A ∩ U = x ∈ U f1 (x) = . . . = fk (x) = 0

for some fi ∈ OM (U ) (which is possible since A is an analytic set). Hence ∀ i ∈ {1, . . . , k}, we have

fi|A∩U = 0 ⇒ fi|U ∩A◦ = 0 with U ∩ A◦ open, non-empty

Since U is connected and contains U ∩ A◦ 6= ∅, we obtain by the Identity Theorem (via local coordinates) that
fi|U = 0, ∀ i, i.e. fi = 0 as functions in OM (U ).

This implies that A ∩ U = U ⇒ U ⊆ A, which contradicts the fact that U contains elements which are not
in A. Indeed, U is a neighborhood of x0 ∈ ∂A◦ ⊆ ∂A, hence U ∩ (M \ A) 6= ∅ and this is not compatible with
U ⊆ A. Therefore A◦ = ∅.

Consequence :
If A is an analytic set in a complex manifold M , then A is either equal to the whole space or has empty interior.
In particular, not all closed sets in M are analytic sets, as e.g. in the following example :
Figure 7.9: A has a non-empty interior and is hence not an analytic set

7.4.8 Corollary
If A 6= M is analytic, then M \ A is open and dense in M , i.e. M \ A = M .

Proof. M \ A is open since A is closed. Moreover M \ A = M \ A◦ = M \ ∅ = M .

Example :
If M = C2 , A1 is a point and A2 is a line in C2 (both are analytic sets), then C2 \ {pt} = C2 \ line = C2 .

Figure 7.10: the complement of a point or a line in C2 is dense

61
Complex Manifolds Section 7.5 SCHLICHENMAIER, Leytem

7.5 Application to meromorphic functions

Let M be a complex manifold and (A, f ) be a meromorphic function on M , i.e. f ∈ OM (M \ A), where the
exceptional set A is an analytic set as shown in 7.4.3. Then it is possible to extend f to a bigger set (maybe f
cannot be extended on the whole A, but at least as far as possible).

7.5.1 Lemma
Let U ⊆ M be open, x0 ∈ U and g, h ∈ OM (U ) be relatively prime such that h(x0 ) = g(x0 ) = 0. Then in every
g(x)
neighborhood V of x0 and for every c ∈ C, ∃ x ∈ V such that h(x) = c.
The proof of this lemma uses the Weierstrass preparation theorem. As an example, consider the function
f (z1 , z2 ) = zz12 at (0, 0). In 7.1.3 we saw that f takes all possible complex values, including ∞, around (0, 0).

7.5.2 Proposition
Let Y ⊂ M be an open and dense subset of M and f ∈ OM (Y ) be holomorphic on Y . Assume that ∀ x0 ∈ M \ Y ,
∃ U open set containing x0 and ∃ g, h ∈ OM (U ) such that h(x) · f (x) = g(x), ∀ x ∈ U and the germs gx0 and hx0
are relatively prime (as defined in 7.2.1). If we define

A := x0 ∈ M \ Y | ∀ r ∈ R, ∀ V open neighborhood of x0 , ∃ x ∈ V ∩ Y such that |f (x)| > r (7.3)

then there exists a unique holomorphic extension fˆ of f to M \A ⊇ Y such that (A, fˆ) is a meromorphic function.
Remark : The condition in (7.3) is not uniform since we only require that ∃ x ∈ V ∩ Y instead of ∀ x ∈ V ∩ Y .
Proof. Let x0 ∈ M \ Y ⇒ ∃ U small open neighborhood of x0 and ∃ gU , hU ∈ OM (U ) such that

gU (x) = hU (x) · f (x) , ∀ x ∈ U (7.4)

If hU (x0 ) 6= 0, then by continuity ∃ W ⊆ U open such that x0 ∈ W and hU |W 6= 0. Hence hgUU will be bounded
in this neighborhood W around x0 , which means that x0 ∈ / A.
If hU (x0 ) = 0 and gU (x0 ) 6= 0, i.e. gU |W 6= 0 for some open neighborhood W of x0 , then (7.4) implies that
0 · f (x0 ) 6= 0, i.e. ”f (x0 ) = ∞”. More precisely, this means that |f | will be uniformly unbounded in any small
neighborhood W around x0 . It follows that x0 ∈ A.
If hU (x0 ) = gU (x0 ) = 0 where gU and hU are relatively prime, lemma 7.5.1 shows that x0 ∈ A. Hence
 [ [
A ∩ U = x ∈ U hU (x) = 0 ⇒ A= (A ∩ U ) since A ⊆ M \ Y ⊆ U
U 3x0 U 3x0

This shows in particular that A is an analytic set.

Moreover this whole argument is independent of the choice of gU and hU since they are relatively prime, i.e.
different representatives only differ by units on U (holomorphic functions which have no zeros on U ), hence A ∩ U
is always the same. Now one can extend f as follows : first we set

gU (x)
fˆU : (M \ A) ∩ U −→ C , fˆU (x) := = f|U (x)
hU (x)

This exists since A (the zero set of all hU ) has been removed. Moreover fˆU is well-defined on the intersections :
g g
if U1 , U2 have a non-empty intersection U12 := U1 ∩ U2 6= ∅ with fˆU1 = hUU1 and fˆU2 = hUU2 , then
1 2

g  g 
U1 U2
fˆU1 = fˆU2





|U12
= = f|U1 |U12
= f|U2 |U12
= |U12
hU1 |U12 hU2 |U12

Since A ⊆ M \ Y ⇒ Y ⊆ M \ A, we then can define

(
f (x) if x ∈ Y
fˆ : M \ A −→ C , fˆ(x) :=
fˆU (x) = gU (x) hU (x) if x ∈ M \ Y , x ∈ U

62
Complex Manifolds Section 7.5 SCHLICHENMAIER, Leytem

gU
This gives indeed a holomorphic extension of f since hU · f|U = gU by assumption ⇒ f|U = hU after A has
been removed. It remains to show that the extension is unique.
Let fˆ0 be any extension of f to M \ A. By definition, this means that fˆ|Y
0
= f = fˆ|Y , hence fˆ and fˆ0 coincide on
Y . Differences can thus only happen on M \ Y . But this is not the case since Y is dense in M . Every point in
M \ Y is a limit of sequences from Y , i.e.

∀ x0 ∈
/ Y : ∃ (xn )n ⊂ Y such that xn → x0 as n → +∞

In particular, if x0 ∈ M \ A, then fˆ(x0 ) and fˆ0 (x0 ) exist and by continuity, we have fˆ0 (xn ) → fˆ0 (x0 ). xn ∈ Y
implies that fˆ0 (xn ) = f (xn ) = fˆ(xn ) → fˆ(x0 ). By uniqueness of the limit we finally obtain that fˆ(x0 ) = fˆ0 (x0 )
too, i.e. fˆ and fˆ0 coincide on M \ A ⊇ Y .

7.5.3 Example
Consider C2 and denote the coordinate axes by l1 = { z ∈ C2 | z2 = 0 } and l2 = { z ∈ C2 | z1 = 0 }.
Let Y := C2 \ (l1 ∪ l2 ) ⇒ Y is open and dense in C2 and consider the function f (z) = zz21 , which is well-defined
on Y . Define A as in (7.3) ⇒ A ⊆ l1 ∪ l2 . By 7.1.3, we have that l1 ⊆ A since |f | → ∞ when approaching z ∈ l1
where z 6= (0, 0) ; on l1 \ {(0, 0)}, |f | even goes uniformly to ∞. However no point from l2 \ {(0, 0)} belongs to A
since f is not unbounded in any neighborhood of these points.

Moreover z1 and z2 are relatively prime, hence A = l1 and f can be extended to a holomorphic function fˆ on
C2 \ l1 . This is the maximal extension of f .

7.5.4 Theorem
Let M be a (connected) complex manifold.
Then the set of all meromorphic functions M(M ) is an algebra and a field extension of C.
Proof. C ⊂ M(M ) since any constant ∈ C can be written as c = 1c with V (1) = ∅. Now we have to define
additive and multiplicative structures on M(M ). Let (A, f ) and (A0 , f 0 ) be 2 meromorphic functions on M .
Since A and A0 are analytic sets (see 7.4.3), we know that A, A0 are closed and that M \ A and M \ A0 are dense
in M . Thus Y := M \ (A ∪ A0 ) = (M \ A) ∩ (M \ A0 ) is open and dense in M (as intersection of 2 dense subsets).
Thus f + f 0 and f · f 0 are defined on Y and can be extended by proposition 7.5.2 to some bigger set M \ A1
where A1 is again an analytic set. Hence M(M ) is already a C–algebra.
In order to construct the inverse of e.g. (A, f ), we write f locally as f = hg with A = V (h) and B = V (g). This
g
is independent of the choice of g and h because if f = hgii = hjj on Ui ∩ Uj (where numerator and denominator
are relatively prime), then gi · hj = gj · hi on Ui ∩ Uj . This implies that there cannot exist zeros on Ui ∩ Uj for
one representative which are not zeros for the other one. Now, as zero sets of holomorphic functions, A and B
are analytic sets and so is A ∪ B. If Y = M \ (A ∪ B), then f1 = hg is defined on Y and we can extend it to a
meromorphic function (A2 , f1 ) ∈ M(M ) such that M \ A2 is the maximal domain of definition of f1 .

7.5.5 Remarks
It is important to always extend the objects to the maximal domain. Consider e.g. the meromorphic functions
(A, f ) and (A, −f ). Then f + (−f ) = 0, but 0 is not only defined on M \ A. Extending it will define 0 as a
meromorphic function on M .
Hence by extending, one can (sometimes) reduce the exceptional set of a meromorphic function.

One can show that the extension M(M ) ⊃ C is a transcendental field extension. Moreover the dimension of
the manifold M is closely related to the transcendence degree of this extension. In fact they are often, but not
always, equal. We have for example equality in the case where M = Cn or if M is a projective variety.

63
Chapter 8

Analytic sets and singularities

8.1 Hypersurfaces
8.1.1 Recalls
Let M be a complex manifold. A ⊆ M is called an analytic set if there is an open covering {Ui }i∈J of M such
that ∀ i ∈ J, ∃ f1i , . . . , fki i ∈ OM (Ui ) satisfying
ki
\
A ∩ Ui = x ∈ Ui f1i (x) = . . . = fki i (x) = 0 = (fji )−1 {0}


j=1
Let A be an analytic set. We have showed that :
a) A is closed in M .
b) A = M or A◦ = ∅ (see figure 8.1).
c) if A 6= M , then M \ A is open and dense in M (see figure 8.1).
Figure 8.1: analytic sets have empty interior and their complement is dense

Submanifolds of M are analytic, but the converse is not true ; complex affine varieties can e.g. have singularities
(points where the rank of the Jacobian drops). In order to make them analytic, one first has to remove all
singularities . This will be the aim of this chapter.

8.1.2 Definition
A hypersurface is a non-empty analytic set that can locally be given by 1 non-constant function, i.e. A ⊆ M is
a hypersurface if there is an open covering {Ui }i∈J of M such that ∀ i ∈ J, ∃ fi ∈ OM (Ui ) satisfying
A ∩ Ui = x ∈ Ui fi (x) = 0 = fi−1 {0}


Not every analytic set can be written under such a form. Note that this does not mean that ki = 1, ∀ i ∈ J since
ki is not uniquely given (see section 7.4.2). Moreover it does not imply that A is of codimension 1 since A is not
a manifold (we first need to define the codimension of an analytic set, see section 8.3.2).

By definition, M cannot be a hypersurface. Indeed if A = M , then A ∩ Ui = M ∩ Ui = Ui = fi−1 {0} , so

f = 0 on Ui , which is non-empty and open, i.e. f = 0 on M by the Identity Theorem and this was excluded.
In addition one has to take care that A is well-defined on the intersections, i.e. if Ui and Uj are open such that
(A ∩ Ui ) ∩ Uj 6= ∅ with A ∩ Ui = V (fi ) for fi ∈ OM (Ui ) and A ∩ Uj = V (fj ) for fj ∈ OM (Uj ), we need that


 
V fi|Ui ∩Uj = V fj|Ui ∩Uj ⇔ x ∈ Ui ∩ Uj fi (x) = 0 = x ∈ Ui ∩ Uj fj (x) = 0 (8.1)
If (A ∩ Ui ) ∩ Uj = ∅, then (8.1) is not a restriction (see example 8.1.3).

64
Complex Manifolds Section 8.2 SCHLICHENMAIER, Leytem

8.1.3 Example
Let M = P1 (C) and {U0 , U1 } be the open covering of M given by the affine sets
   
U0 = (z0 : z1 ) z0 6= 0 = (1 : z1 ) z1 ∈ C , U1 = (z0 : z1 ) z1 6= 0 = (z0 : 1) z0 ∈ C

where e.g. U0 ∼
= { ω | ω ∈ C } by (z0 : z1 ) = (1 : z1
z0 ) = (1 : ω). Let f ∈ OM (U1 ) be given by
z0
f : U1 ( P1 (C) −→ C , f (z0 : z1 ) =
z1

f is well-defined on U1 and independent of the representative since λz z0 −1

λz1 = z1 . Moreover f
0
{0} = {(0 : 1)}.
Let A = {(0 : 1)} with f ∈ OM (U1 ) and g ∈ OM (U0 ) given by g = 1. With this we obtain

A ∩ U0 = ∅ = g −1 {0} A ∩ U1 = A = f −1 {0}



,

with (A ∩ U0 ) ∩ U1 = ∅, hence A is a well-defined analytic subset of P1 (C) and it is even a hypersurface since any
A ∩ Ui is given by 1 holomorphic function.
Note that f cannot be extended to U0 ∪ U1 by definition, but also since f is non-constant and there are no
non-constant global holomorphic functions on P1 (C) since it is compact.

8.2 Regular and singular points

8.2.1 Definitions
Let M be a complex manifold with dimC M = n and A ⊆ M be an analytic set.
a) x0 ∈ A is called a regular point (or a smooth point) of A if there is a chart (U, ϕ) around x0 and there exist
holomorphic functions f1 , . . . , fk ∈ OM (U ) such that (see figure 8.2)

∂(fi ◦ ϕ−1 )
 


U ∩ A = x ∈ U f1 (x) = . . . = fk (x) = 0 , J(x0 ) = ϕ(x0 ) (8.2)
∂zj i=1,...,k
j=1,...,n

where the k × n–Jacobian matrix J(x0 ) evaluated at ϕ(x0 ) ∈ Cn has maximal rank (i.e. rank k), if the local
coordinates are ϕ = (z1 , . . . , zn ). We will see in section 8.3.2 that k > n is not possible if this is the case.

Figure 8.2: x0 is a regular point of the Jacobian matrix has maximal rank

Note that the condition of U ∩ A being a zero set of a finite number of holomorphic functions is always satisfied
since A is analytic. We require that there are holomorphic functions which satisfy both conditions.
∂fi
In the following, we may omit the composition with the chart map ϕ−1 and simply write ∂z j
(x0 ) instead.
b) If x0 ∈ A is not a regular point, then it is called a singular point (or a singularity) of A.
c) We denote S(A) := { x ∈ A | x is singular }.

8.2.2 Examples
Let M = C2 be the complex plane and consider A = Vi = V (fi ) for fi = fi (z1 , z2 ) given by

f1 = z22 − 4z1 (z1 + 1)(z1 − 1) , f2 = z22 − z13 , f3 = z1 , f4 = z1 z2 , f5 = z12

65
Complex Manifolds Section 8.2 SCHLICHENMAIER, Leytem

Consider the trivial covering U = C2 ; all functions are holomorphic on C2 . Hence any Vi is an analytic set as
the zero set of a holomorphic function and using the notation in (8.2), we always have k = 1. This implies that
the 1 × 2–Jacobian matrix has maximal rank at a point x0 ∈ Vi if and only if
 
∂fi ∂fi


∂z1 (x0 ) ∂z2 (x0 ) 6= 0 , 0

a) ∂f1 ∂f1
(z1 , z2 ) = −4(z1 + 1)(z1 − 1) − 4z1 (z1 − 1) − 4z1 (z1 + 1) , (z1 , z2 ) = 2z2
∂z1 ∂z2
Together with the condition that (z1 , z2 ) ∈ V1 , the singular points of V1 are given by the solutions of the system
 
2
 z2 − 4z1 (z1 + 1)(z1 − 1) = 0
  z1 (z1 + 1)(z1 − 1) = 0 (1)

(z1 + 1)(z1 − 1) + z1 (z1 − 1) + z1 (z1 + 1) = 0 ⇔ 3z12 − 1 = 0 (2)
 
2z2 = 0 z2 = 0
 

where (1) and (2) are not compatible since (1) ⇔ z1 ∈ {−1, 0, 1} and none of these is a solution of (2). It
follows that S(V1 ) = ∅ since the system has no common solution : V1 is a curve without singular points. It is
actually given by a plane elliptic curve, which is a non-singular cubic. Figure 8.3 shows how V1 looks like in the
real case. In the complex case the curve will become connected.

b)  
2 3
 f2 (z1 , z2 ) = 0
  z2 − z1 = 0

∂f2
∂z (z1 , z2 ) = 0
⇔ −3z12 = 0 ⇔ z1 = z2 = 0 with (0, 0) ∈ V2
 ∂f12
 
∂z2 (z1 , z2 ) = 0 2z2 = 0


Hence (0, 0) is a singular point of V2 (the only one) and S(V2 ) = {(0, 0)}. V2 is called a cuspidal cubic (see figure
8.3). Another type of such curves is the nodal curve.

Figure 8.3: an elliptic, a cuspidal and a nodal curve

c) ∂f3 ∂f3


(z1 , z2 ) = 1 , (z1 , z2 ) = 0 ⇒ 1 , 0 6= 0 , 0
∂z1 ∂z2
Hence V3 has no singular points (which is intuitively clear since it is just a line, see figure 8.4) : S(V3 ) = ∅.

d) ∂f4 ∂f4
f4 (z1 , z2 ) = z1 z2 , (z1 , z2 ) = z2 , (z1 , z2 ) = z1
∂z1 ∂z2
For V4 , (0, 0) is the only candidate for a singular point and it is also one (see figure 8.4) : S(V4 ) = {(0, 0)}.

e) ∂f5 ∂f5
f5 (z1 , z2 ) = z12 , (z1 , z2 ) = 2z1 , (z1 , z2 ) = 0
∂z1 ∂z2
Hence the candidates for singularities are all points of the type (0, z2 ), ∀ z2 ∈ C. We have (0, z2 ) ∈ V5 , ∀ z2 ∈ C.
But V5 = { (0, z2 ) | z2 ∈ C }. Does this imply that all points of the curve are singularities ?

66
Complex Manifolds Section 8.2 SCHLICHENMAIER, Leytem

Figure 8.4: V3 and V4 only consist of the coordinate axes

No, because the definition of a singular point says that there does not exist any finite number of holomorphic
functions which satisfy both of the conditions in (8.2). Indeed, f5 is not the good choice, but f3 works because
V5 = V3 = C2 ∩ V3 = f3−1 {0}

and its Jacobian has rank 1. We took the wrong function to describe the analytic set V5 . In fact, the vanishing
ideal of V5 is given by h z1 i, and not by h z12 i since z12 is not irreducible. Finally, S(V5 ) = S(V3 ) = ∅.

Remark :
In practise, when dealing with a system of (polynomial) equations like

 f (z1 , z2 ) = 0
 (1)
∂f
∂z (z ,
1 2z ) = 0 (2)
 ∂f1

∂z2 (z1 , z2 ) = 0 (3)

it is easier to solve (2) and (3) first and then check if (1) is also satisfied because (1) is of higher degree than (2)
and (3). Solving (1) is in general much more complicated.

8.2.3 Proposition
The Jacobian matrix J(x0 ) in (8.2) obviously depends on the local coordinates ϕ = (z1 , . . . , zn ). However :
The rank of J(x0 ) does not depend on the chosen local coordinates.
Proof. This follows from the chain rule. Let (U, ϕ) and (V, ψ) be 2 coordinate charts around x0 and set
ϕ = (z1 , . . . , zn ) , ψ = (w1 , . . . , wn ) , ρ = ψ ◦ ϕ−1 , gi = fi ◦ ϕ−1 , hi = fi ◦ ψ −1
We have to show that the associated Jacobian matrices as given in (8.2) have the same rank. Note that
∀ i ∈ {1, . . . , k} : gi = fi ◦ ϕ−1 = fi ◦ ψ −1 ◦ ψ ◦ ϕ−1 = hi ◦ ρ








⇒ J(gi ) ϕ(x0 ) = J(hi ◦ ρ) ϕ(x0 ) = J(hi ) ρ ϕ(x0 ) ◦ J(ρ) ϕ(x0 ) = J(hi ) ψ(x0 ) ◦ J(ρ) ϕ(x0 ) (8.3)

Figure 8.5: a coordinate change from (U, ϕ) to (V, ψ)

Equation (8.3) thus shows that for any fixed indices i ∈ {1, . . . , k} and j ∈ {1, . . . , n} :
n
  ∂(fi ◦ ϕ−1 )
∂(hi ◦ ρ)
X ∂hi
∂wk

Jϕ (x0 ) = ϕ(x0 ) = ϕ(x0 ) = ψ(x0 ) · ϕ(x0 )
ij ∂zj ∂zj ∂wk ∂zj
k=1
n
X ∂(fi ◦ ψ −1 )
∂ρk


= ψ(x0 ) · ϕ(x0 ) = Jψ (x0 ) · J(ρ) ϕ(x0 )
∂wk ∂zj ij
k=1


where J(ρ) ϕ(x0 ) denotes the Jacobian matrix of the coordinate change ρ evaluated at the point ϕ(x0 ) ∈ Cn .

67
Complex Manifolds Section 8.2 SCHLICHENMAIER, Leytem

Using matrix notation, this means that

h
i−1
Jϕ (x0 ) = Jψ (x0 ) · J(ρ) ϕ(x0 ) ⇔ Jψ (x0 ) = Jϕ (x0 ) · J(ρ) ϕ(x0 )


The n × n–matrix J(ρ) ϕ(x0 ) is invertible at any point since ρ is a change of variables ; in particular it has
maximal rank. Hence Jϕ (x0 ) and Jψ (x0 ) have the same rank since they are related by an invertible matrix.

8.2.4 Theorem
Let M be a complex manifold and U ⊆ M be non-empty and open. Let A ⊆ M be an analytic set and S(A) be
the set of singular points in A. Assume that S(A) ∩ U = ∅, i.e. U does not contain singular points of A. Then
A ∩ U is a submanifold of U .
In particular, if S(A) = ∅ (A has no singular points), then the analytic set A is a submanifold of M .
Proof. Being a submanifold is a local statement, so let x0 ∈ U ∩ A. Since A is analytic, we know that A is locally
at x0 given by holomorphic functions f1 , . . . , fk . x0 ∈/ S(A) because S(A) ∩ U = ∅, so the matrix
 ∂f1 ∂f1 ∂f1 
∂z1 (x0 ) ∂z2 (x0 ) . . . ∂zn (x0 )
 .. .. .. .. 
 . . . . 
∂fk ∂fk ∂fk
∂z1 (x0 ) ∂z2 (x0 ) . . . ∂zn (x0 )

is of rank k. By permuting and renumbering the coordinates, we may assume that the first k columns are linearly
independent, hence  
∂fi 
det (x0 ) i=1,...,k 6= 0
∂zj j=1,...,k

Let (U 0 , ϕ) be a chart around x0 and denote z (0) := ϕ(x0 ). Then we define the map F : ϕ(U 0 ) ⊆ Cn → Cn by
(0)
F (z1 , . . . , zn ) := (f1 ◦ ϕ−1 )(z1 , . . . , zn ), . . . , (fk ◦ ϕ−1 )(z1 , . . . , zn ) , zk+1 − zk+1 , . . . , zn − zn(0)

 ∂f  
i
(x0 ) i=1,...,k
 ∂zj j=1,...,k
@
@ 
 
 
⇒ J(F ) z (0)


=
 0 0 . . . 0 1 0 . . . 0 

 0 0 ... 0 0 1 ... 0 
 . . . . . . . .
. . . . . . . .

 . . . . . . . .
0 0 ... 0 0 0 ... 1

In addition : F z (0) = 0 since ϕ−1 z (0) = x0 ∈ U 0 ∩ A = { x ∈ U 0 | f1 (x) = . . . = fk (x) = 0 } and

 

 ∂fi 
det J(F ) z (0) = det (x0 ) i=1,...,k · 1 · . . . · 1 6= 0
∂zj j=1,...,k

By the Inverse Function Theorem, we can thus invert the function F locally, i.e. there is an open neighborhood
V 0 of z (0) in ϕ(U 0 ) ⊆ Cn and an open neighborhood W of 0 in Cn such that F (V 0 ) ⊆ W is open and
F|V 0 : V 0 ⊆ ϕ(U 0 ) −→ W ⊆ Cn
is biholomorphic on V 0 . Since ϕ : U 0 → ϕ(U 0 ) is a homeomorphism, we may assume that V 0 = ϕ(V ) for some
open neighborhood V of x0 in U 0 ⊆ U . It follows that x0 ∈ V ⊆ U and
(F ◦ ϕ)|V : V ⊆ U −→ W ⊆ Cn
is also biholomorphic (by definition). Moreover since V ∩ A ⊆ U 0 ∩ A, i.e. fi (V ∩ A) = 0, ∀ i, we obtain that
k
 \
−1

(F ◦ ϕ)(V ∩ A) = w∈W w1 = . . . = wk = 0 = pi|W {0}
i=1

where pi : Cn → C is the projection onto the ith coordinate, which is holomorphic.

68
Complex Manifolds Section 8.3 SCHLICHENMAIER, Leytem

We conclude that (F ◦ ϕ)(U ∩ A) is a complex submanifold since it is locally a zero set of holomorphic functions
and the rank of the Jacobian of the projections is obviously maximal. Finally U ∩ A is a complex submanifold
as well since F ◦ ϕ is biholomorphic on V , i.e. the complex structure of U ∩ A is preserved under this map.

8.3 Irreducibility
8.3.1 Definition
An analytic set A is called reducible ⇔ ∃ A1 , A2 analytic sets such that A 6= A1 , A 6= A2 and A = A1 ∪ A2 .
A is called irreducible if it is not reducible.

Example :
Consider the examples in 8.2.2. We see that V4 = V (z1 z2 ) = V (z1 ) ∪ V (z2 ) is reducible.
V1 , V2 and V3 are irreducible (where we recall that V1 is not disconnected in the complex picture).

8.3.2 Results
In the following, we state a number of facts without (detailed) proof :
1) If A ⊆ M is an analytic subset, there exists a countable set of irreducible analytic sets {Aj }j∈J such that
S
a) A = j∈J Ai

b) The system is locally finite, i.e. ∀ x ∈ M , ∃ Ux open neighborhood of x such that Ux ∩ Aj 6= ∅ only for
finitely many j ∈ J (see figure 8.6)
c) If Aj1 6= Aj2 , then Aj1 * Aj2 . This is the maximality condition : the irreducible parts are maximal.
This ”decomposition” of A is called the decomposition into irreducible components.
Figure 8.6: a locally finite decomposition of A

In particular : if M is a compact complex manifold, then J can be chosen to be finite because

S S
M= Ux ⇒ M = Ui : I finite
x∈M i∈I

and every Ui satisfies Ui ∩ Aj 6= ∅ only for finitely many j ∈ J, hence there can only be finitely many Aj since
{Ui } is a covering of M .

c) is for example not satisfied in the following case :

The line A1 is irreducible and contains other irreducible subsets as e.g. the point A2 , hence A1 = A1 ∪ A2 , but
A2 6= A1 and A2 ⊂ A1 .

2) Let A ⊆ M be an analytic set and S(A) be the subset of singular points of A.

Then S(A) is an analytic set which is nowhere dense in A (”nowhere” means that it is not dense in any component
of A ; because ”not dense” does not exclude that it may be dense in some component of A).

69
Complex Manifolds Section 8.4 SCHLICHENMAIER, Leytem

Intuitive idea : let J be a k × n–Jacobian matrix. Then rk J(x) ≤ k at any point x ∈ A. How to check that rk J
drops at a point x0 ∈ A ?  
∂fi
J(x) = (x)
∂zj i=1,...,k
j=1,...,n

Recall that a minor of J is a submatrix of J obtained by removing one or more lines and columns from J. The
rank of J is then given by the dimension of the biggest minor inside J which is a square-matrix and has non-zero
determinant. Hence the rank at a point drops if all k × k–subdeterminants in J are zero. Consider for example
 
a11 a12 a13
J= ⇒ rk(J) ≤ 2
a21 a22 a23
     
a a12 a a13 a a13
rk(J) ≤ 1 ⇔ det 11 = 0 , det 12 = 0 and det 11 =0
a21 a22 a22 a23 a21 a23

Let now x0 ∈ / S(A) be a non-singular point ⇒ rk J(x0 ) = k. Since determinants are polynomial functions,
hence continuous, this implies that there is a neighborhood U of x0 on which the rank is still equal to k, i.e. all
point in U are non-singular points. It follows that S(A) cannot be dense in A since we cannot approach x0 ∈ A
by singular points.

3) If an analytic set A ⊆ M is irreducible, then A \ S(A) is connected.

Here we need the assumption that is A irreducible (see figure 8.7).

Figure 8.7: A \ S(A) is no longer connected since A is not irreducible

4) If an analytic set A ⊆ M is irreducible, then A \ S(A) is a (connected) complex manifold.

This follows from 2), 3) and 8.2.4 since S(A) is analytic, hence closed, so by choosing U = M \ S(A), we obtain
that A ∩ U = A \ S(A) is a complex submanifold of M , which is in addition connected by 3).
In particular, A \ S(A) has now a well-defined dimension given by the definition of a complex manifold. We set



dim A := dim A \ S(A) , codim A := dim M − dim A \ S(A)

Conclusion :
A has well-defined dimension and codimension (after removing singular points) ⇔ A is irreducible.

5) Let n = dim M . If A = V (f1 , . . . , fl ) is the vanishing set of a finite number of holomorphic functions
such that A is irreducible, then dim A ≥ n − l. Hence by adding an additional function to the zero set, the
dimension of A goes down by at most 1 (it may remain the same).

8.4 Divisors
8.4.1 Definitions
A prime divisor is an irreducible analytic subset of codimension 1, i.e. an irreducible hypersurface.
A divisor D is a formal sum X
D= nY · Y , nY ∈ Z (8.4)
Y

where we ”sum” over all prime divisors Y with the condition that the sum is locally finite, i.e. ∀ x ∈ M , ∃ Ux
open neighborhood of x such that only finitely many of the Y with Y ∩ Ux 6= ∅ have a non-zero coefficient nY .

70
Complex Manifolds Section 8.4 SCHLICHENMAIER, Leytem

In other words this means that for all x ∈ M , the set


Y prime divisor Ux ∩ Y 6= ∅ and nY 6= 0

must be finite. All values for nY such that this condition is satisfied are allowed.

8.4.2 Examples
P
If M is compact, we can again replace ”locally finite” by ”finite”, i.e. D = Y nY · Y with nY 6= 0 for at most
finitely many Y . In general, this condition is too strong, as shows the example in section 8.4.3.

Let M = C (or in general : a complex manifold of dimension 1). In this case the prime divisors are the points
because codimM Y = 1 ⇒ dimM Y = 0 and irreducibility implies that Y must be a single point (not many
points). The set of all prime divisors is thus the set of all points in M .

8.4.3 Divisor associated to a meromorphic function

Let h ∈ M(M ) be a meromorphic function and p ∈ M . The order or multiplicity of h at p is defined as


 0 if h is holomorphic at p with h(p) 6= 0

k if p is a zero of multiplicity k of h
ordp (h) :=


 −k if h has a singularity of order k at p
∞ if h is identically zero


Now consider M = C and let f : C → C be a meromorphic function. The divisor associated to f is given by
X
(f ) := ordp (f ) · {p}
p∈C

The previous example shows that (f ) is indeed a divisor (points are the only prime divisors). This sum is in
addition locally finite since the zero set of a (non-zero) holomorphic function on C is discrete. For example

(z) = 1 · {0} , (z 2 ) = 2 · {0} , (z − a) = 1 · {a}

We also see that one can obtain all the prime divisors (all points) by choosing all a ∈ C. Moreover we have :
1


∀ f, g ∈ M(C) : (f · g) = (f ) + (g) and f = −(f )

since ordp (f · g) = ordp (f ) + ordp (g) and ordp f1 = −ordp (f ) if f 6≡ 0.

Example : P
Let R : C → C , f (z) = sin(πz). Then V (f ) = Z and (f ) = 1 · {k}, which is locally finite (see figure 8.8).
k∈Z
Hence we see that requiring a finite sum in (8.4) is too restrictive (C is not compact).

Figure 8.8: the divisor associated to z 7→ sin(πz) is locally finite, but not finite

In general one can define (f ) for any meromorphic function f ∈ M(M ) where M is a 1-dimensional complex
manifold (i.e. a Riemann surface) since points are the only prime divisors in this case. still locally finite?
A divisor D is called a principal divisor if ∃ f ∈ M(M ) such that D = (f ).

For dim M ≥ 2, (f ) is in general not a divisor. However, one can then associate to a meromorphic function
f : M → C and a prime divisor Y the number (f, Y ) 7→ nY ∈ N0 such that f is holomorphic in Y and vanishes
along Y with multiplicity nY .

71
Chapter 9

Holomorphic vector bundles

9.1 Biholomorphic functions

Theorem
Any bijective holomorphic function f : U ⊆ Cn → f (U ) ⊆ Cn is biholomorphic.
We are going to discuss this result for n = 1. It also holds for n > 1, but this is much more difficult.

Open mapping theorem : (no proof)

If U is an open connected subset of C and f : U → C is a non-constant holomorphic function, then f is open.

This already implies that the inverse of a bijective holomorphic map is continuous.
Hence by the Inverse Function Theorem it remains to show that ∂f ∂z (z0 ) 6= 0, ∀ z0 ∈ U (since f
−1
will then be
holomorphic in a neighborhood of f (z0 ), i.e. in a neighborhood of any point in f (U )).

approach 1 :
Since f is holomorphic, it is also analytic and ∀ z0 ∈ U , it can be written in a neighborhood of z0 as
∞
X 1 ∂kf
f (z) = ak · (z − z0 )k = a0 + a1 (z − z0 ) + a2 (z − z0 )2 + . . . where ak = · (z0 )
k! ∂z k
k=0
∂f
Assume that = 0, then a1 = 0 and f locally writes as f (z) = a0 + a2 (z − z0 )2 + . . ., which implies that f
∂z (z0 )
cannot be injective near z0 . Hence bijective holomorphic functions never have vanishing first derivatives.

approach 2 :
We know that f −1 is holomorphic around any point f (z0 ) such that ∂f ∂f
∂z (z0 ) 6= 0. Since ∂z is also holomorphic,
we know that its zero set is discrete ( ∂f
∂z is not identically zero because f is bijective). Let p be a zero of ∂z .
∂f
−1
Then f is continuous on f (Up ) and holomorphic on f (Up ) \ {f (p)}, where Up is a small open neighborhood of
p not containing other zeros of ∂f
∂z . It follows that f
−1
is holomorphic on f (Up ) (see recall in section 2.6.1). This
holds for any point of the discrete zero set, hence f −1 is holomorphic on f (U ).

For n > 1, the task is more complicated since matrices are involved, i.e. non-vanishing conditions must be
replaced by maximal rank and non-zero determinant conditions. Moreover the analytic expansion of a holomor-
phic function in several variables is more complicated (see section 2.3).

9.2 Vector bundles : basic notions

9.2.1 Definitions
Let M and E be complex manifold and π : E → M be a surjective holomorphic map. π is called a foot-map.
For any m ∈ M , we denote the fiber over m of π by Em := π −1 {m} . By abuse of notation, we will denote

Em = π −1 (m) in the following (π is not necessarily bijective, see figure 9.1).

72
Complex Manifolds Section 9.2 SCHLICHENMAIER, Leytem

Figure 9.1: fibers correspond to preimages under the foot-map

If U ⊆ M is open, then π −1 (U ) is open in E since π is holomorphic (hence continuous) and the assignment
denoted by π|U : π −1 (U ) → U gives a ”local situation”. E is called a fibration over the base manifold M .

π : E → M is called a family of vector spaces (over C) if Em = π −1 (m) is a vector space over C for all m ∈ M .
We only consider finite-dimensional vector spaces because otherwise E will be a infinite-dimensional manifold.

Let π : E → M and ρ : F → M be two families of vector spaces (over the same base manifold M ).
A morphism of families of vector spaces is a holomorphic map f : E → F such that the diagram

f
E +F (9.1)
π

~~
ρ

commutes, i.e. ρ ◦ f = π, and such that for all m ∈ M , the function fm given by

fm := f|π−1 (m) : Em −→ Fm

is a linear map over C between the finite-dimensional vector spaces Em
and Fm .

Note that fm is well-defined because of (9.1) : ∀ x ∈ π −1 (m), ρ f (x) = π(x) = m ⇒ f (x) ∈ ρ−1 (m) = Fm .
Hence f preserves the fibers of E and F : it maps a fiber over m of π to a fiber over m of ρ.

9.2.2 Example : the standard trivial family

Let M be a complex manifold of dimension n, k ∈ N0 , E := M × Ck and π = p1 (first projection).
p1 is holomorphic since M and M × Ck (product) are manifolds. Moreover ∀ m ∈ M :

π −1 (m) = p−1 {m} = {m} × Ck ∼ = Ck

1

where {m} × Ck is a vector space with respect to the operations

(m, v1 ) + (m, v2 ) = (m, v1 + v2 ) , λ · (m, v) = (m, λ v) , 0m := (m, 0)

for v1 , v2 , v, 0 ∈ Ck , λ ∈ C. Hence p1 : M × Ck → M is a family of vector spaces (see figure 9.2) and it is called
the standard trivial family of rank k.
Figure 9.2: the standard trivial family

The idea of a vector bundle is now that a bundle should be locally look like this standard trivial family.

73
Complex Manifolds Section 9.2 SCHLICHENMAIER, Leytem

9.2.3 Definition
A family of vector spaces π : E → M is called a (holomorphic) vector bundle ⇔ there is an open covering
{Ui }i∈J of M such that for all i ∈ J, the fibration π|π−1 (Ui ) : π −1 (Ui ) → Ui is isomorphic (as family of vector
spaces) to the standard trivial family p1 : Ui × Ck → Ui . In other words, for any Ui there is a morphism ψUi
such that
ψUi
,
π −1 (Ui ) U i × Ck (9.2)
π
p1
## {{
Ui
where ψUi is bijective and holomorphic (⇔ biholomorphic) and the map ψUi |π−1 (m) : Em → {m} × Ck is a linear
isomorphism, ∀ m ∈ Ui .
Note that π|π−1 (Ui ) is still surjective since π π −1 (Ui ) = Ui (π is surjective). One calls {Ui }i∈J a trivializing

open covering for the vector bundle E.

In the case where M is a compact complex manifold one can always achieve this with finitely many Ui .

9.2.4 Rank of a vector bundle

For a trivializing open covering {Ui }i∈J , we denote the vector bundle restricted to Ui by (see figure 9.3)
   
E|Ui := π|π−1 (Ui ) : π −1 (Ui ) → Ui ∼
= p1 : Ui × Ck → Ck

The same construction can be done for some arbitrary open set U ⊆ M , i.e. a vector bundle π : E → M can
always be restricted to π : E|U = π −1 (U ) → U .

Figure 9.3: restriction of a vector bundle to a trivializing open set

Due to the isomorphism ψUi in (9.2), every fiber in a vector bundle E has in particular dimension k and we
define the rank of E as rk E := k. By connectedness of M , this number k is the same for all m ∈ M :
Assume for example that for m ∈ Ui and m0 ∈ Uj open with Ui ∩ Uj 6= ∅, we have

E|Ui ∼
= U i × Ck i ,E|Uj ∼= Uj × Ckj ⇒ dim Em = ki , dim Em0 = kj
but : (Ui ∩ Uj ) × Cki ∼
= E|Ui |Uj = E|Uj |Ui ∼
= (Ui ∩ Uj ) × Ckj



(9.3)

The linear isomorphism implies that the dimensions are the same : ki = kj .
Let now x, y ∈ M be arbitrary. By path-connectedness, there is continuous path from x to y which can be
covered by finitely many chart domains. Repeating argument (9.3) for any of these finitely many open sets, we
obtain that the fibers over x and y are isomorphic. In particular : dim Ex = dim Ey = k.

We only consider vector bundles of finite rank (rank equal to 0 is possible). In particular, a vector bundle of
rank 1 is also called a line bundle (see section 10.5).

74
Complex Manifolds Section 9.3 SCHLICHENMAIER, Leytem

9.2.5 Example
F
The tangent bundle TM := Tm M with foot-map π : Tm M → M : vm 7→ m is a vector bundle of rank n.
m∈M
It is however not globally trivial, i.e. TM ∼
6 M × Cn . In fact, globally trivial bundles are very rare (an example
=
n
is M = C since there is just 1 chart in this case). Note that for any vector bundle E, we have :
 S 
Em ∼ Ck ∼
G G G
E = π −1 (M ) = π −1 π −1 {m} = = M × Ck


{m} = =
m∈M m∈M m∈M m∈M
k
since all k-dimensional vector spaces over C are isomorphic to C . BUT : this ”isomorphism” is only set-
theoretical, i.e. it is just a bijection between sets ; it is neither a (bi)holomorphic, nor a continuous identification.

Remark :
There are complex vector bundles over real differentiable manifolds which are not holomorphic.

9.3 Sections
For short, we say that E is a vector bundle over M if we mean that π : E → M is a vector bundle.

9.3.1 Definition
Let E be a vector bundle over M and U ⊆ M be open. Then E|U is a vector bundle over U .
A holomorphic map s : U → E|U is called a (global) section of E|U or a local section
of E if π|U ◦ s = idU .
Consider figure 9.4 : in particular, s(m) ∈ Em = π −1 (m), ∀ m ∈ U since π s(m) = m. Moreover s is injective
since it has a left inverse, thus s(U ) ∼
= U and s is biholomorphic onto its image. We denote

V(U ) := (global) sections of the vector bundle E|U
We will see that V defines indeed a sheaf of local sections of E.
Figure 9.4: a local section of a vector bundle

9.3.2 Sheaf of sections

First of all, V(U ) 6= ∅, ∀ U ⊆ M open : it always contains the zero section θ : U → E|U : m 7→ 0 ∈ Em where 0
is the unique zero element in the vector space Em (see figure 9.5). Indeed ∀ m ∈ U :
π θ(m) = π(0) = m because 0 ∈ Em = π −1 (m) and π π −1 (m) = {m}

θ is in addition holomorphic on U since its local form on a small neighborhood of m ∈ U is given by

θ : U −→ U × Ck : m 7−→ (m, 0)
where the first coordinate indicates the fiber and the second one the element in the fiber to which m is mapped.
This is obviously holomorphic, so finally θ ∈ V(U ).
Next we want to define a structure on V(U ) for any open set U ⊆ M .

1) V(U ) is a complex vector space with respect to the pointwise additional structure
s1 , s2 ∈ V(U ) ⇒ s1 + s2 ∈ V(U ) by (s1 + s2 )(m) := s1 (m) + s2 (m) ∈ Em
s ∈ V(U ), λ ∈ C ⇒ λ · s ∈ V(U ) by (λ · s)(m) := λ · s(m) ∈ Em
However we cannot say anything about its dimension (in fact dim V(U ) = ∞ in most cases).

75
Complex Manifolds Section 9.3 SCHLICHENMAIER, Leytem

Figure 9.5: the zero section

2) V(U ) is a module over OM (U ). Let f ∈ OM (U ) be holomorphic on U , s ∈ V(U ) and define f ∗ s ∈ V(U ) by

∀ m ∈ U : (f ∗ s)(m) := f (m) · s(m) ∈ Em (vector space)

| {z } | {z }
∈C ∈ Em

This is still a section : ∀ m ∈ U , π (f ∗ s)(m) = m since f (m) · s(m) ∈ Em = π −1 (m). Moreover the module

axioms as e.g. 1 ∗ s = s, (f + g) ∗ s = f ∗ s + g ∗ s, etc. are trivially satisfied.

It remains to check that f ∗ s is again holomorphic on U . Its local form is given by (see figure 9.6)

s : U −→ U × Ck : m 7−→ m, s(m) , f ∗ s : U −→ U × Ck : m 7−→ m , f (m) · s(m)

since (f ∗ s)(m) is still in the fiber over m of π. f and s being holomorphic, we get that f ∗ s is holomorphic too.

Figure 9.6: multiplication with complex numbers preserves the fiber Em

3) V : U 7→ V(U ) is a sheaf of complex vector spaces with the canonical restriction

ρU
V : V(U ) −→ V(V ) : s 7−→ s|V

if V ⊆ U . The gluing property 4) of a sheaf is satisfied since holomorphy is a local condition.

4) V : U 7→ V(U ) is a sheaf of OM –modules by the operation

∗ : OM (U ) × V(U ) −→ V(U ) : (f, s) 7−→ f ∗ s

This structure is compatible with the restrictions : ∀ m0 ∈ V ⊆ U , (f ∗ s)(m0 ) = f (m0 ) · s(m0 ), thus as in (6.2) :

(f ∗ s)|V = f|V ∗ s|V ⇒ ρU U

V (f ∗U s) = f|V ∗V ρV (s)

5) V is a locally free sheaf of OM –modules of rank k = rk E, i.e. ∃ open covering {Ui }i∈J of M such that

k
∀ i ∈ J : V|Ui ∼
= OUi (9.4)

This follows from the fact that if {Ui } is a trivialization cover of the bundle E, then E|Ui ∼ = Ui × Ck and this
∼
implies that V(Ui ) = OUi (Ui ) × . . . × OUi (Ui ) as OUi (Ui )–modules because the trivialization allows us to see a
holomorphic section locally as a usual holomorphic function whose image has k components. Since we are dealing
with functions and compatible restrictions of functions, (9.4) now follows by the remark in section 6.4.2.

76
Complex Manifolds Section 9.3 SCHLICHENMAIER, Leytem

9.3.3 Example
For any vector bundle E with a trivialization {Ui }i∈J we have certain standard sections êl given by

êl : Ui −→ π −1 (Ui ) ∼
= p−1
1 (Ui ) = Ui × C
k
: m 7−→ (m, el )

where el = (0, . . .
, 1, . . . , 0) is the lth standard basis vector of Ck . êl is a holomorphic section with respect to
p1 since p1 êl (m) = p1 (m, el ) = m by definition. We want to show that {êl }l=1,...,k is an OM (Ui )–basis of the
module V(Ui ), i.e. V(Ui ) is a free OM (Ui )–module of rank k.
k
P
− linear independence : assume that fl · êl = 0 for some fl ∈ OM (Ui ), which means that
l=1
∈C
Xk k z }| {
X
∀m ∈ M : fl (m) · êl (m) = 0 ⇔ fl (m) · (m, el ) = 0
l=1 l=1

{el }l=1,...,k is a basis of Ck ⇒ {(m, el )}l=1,...,k is a basis of {m} × Ck , so fl (m) = 0 for all l ∈ {1, . . . , k}.
This holds for all m ∈ M , which means that fl = 0 (zero function), ∀ l ∈ {1, . . . , k}.

− generating set : can all sections in V(Ui ) be written as an OM (Ui )–linear combination of the êl ? Let

s : Ui −→ Ui × Ck : m 7−→ m, s̃(m) where s̃ : Ui → Ck

k
Since {el }l=1,...,k is a basis of Ck , we get that ∀ m ∈ Ui , ∃ gl (m) ∈ C such that s̃(m) =
P
gl (m) · el .
k
P l=1
Hence as a candidate we consider the functions gl : m 7→ gl (m) ⇒ s = gl · êl is clear.
l=1
It remains to check that these functions are holomorphic on Ui , i.e. gl ∈ OM (Ui ). Consider the sequence

Ui
s / Ui × Ck p2
/ Ck pl
/C

where pl : (α1 , . . . , αk ) 7→ αl . The map pl ◦ p2 ◦ s : Ui → C is holomorphic since s, p2 and pl are.

But p2 ◦ s = s̃ and pl ◦ p2 ◦ s = gl by definition of gl (m) :



k
P 
pl ◦ p2 ◦ s (m) = pl p2 s(m) = pl p2 m, s̃(m) = pl s̃(m) = pl gl (m) · el = gl (m)
l=1

Hence every section on a trivializing open set can (uniquely) be written in such a form.

9.3.4 Examples
Sections of the tangent bundle TM of a complex manifold M are nothing but holomorphic vector fields on M :

X : M −→ TM : m 7−→ X(m) = Xm ∈ Tm M

Similarly the sections of the cotangent bundle T∗ M are the differential 1-forms.

9.3.5 Definitions
A vector bundle E over M is called a trivial vector bundle if it is globally isomorphic to the standard trivial
family, i.e. E ∼
= M × Ck as isomorphism of family of vector spaces. In this case, we even get that V ∼
= OMk
, i.e.
V will be a free sheaf of OM –modules of rank k. The proof in the same as in 5).

Two vector bundles E and F over the same base manifold M are said to be isomorphic if there exist two mor-
phisms of families of vector spaces f : E → F and g : F → E such that f ◦ g = idF and g ◦ f = idE .
An isoclass is just the set of equivalence classes of the equivalence relation E ∼ F ⇔ E ∼
= F.

Our next goal is to prove the following theorem :

77
Complex Manifolds Section 9.4 SCHLICHENMAIER, Leytem

9.3.6 Theorem
The category of isoclasses of holomorphic vector bundles of rank k over a manifold M is equivalent to the category
of locally free sheaves of OM –modules of rank k.
(Saying that the categories are equivalent means that there is a 1-to-1 correspondence between the objects.)
As a corollary, the isoclass of a trivial vector bundle is uniquely given by V ∼ k
= OM .

By the above, we already showed that to any vector bundle E of rank k one can associated a locally free sheaf
of rank k given by V, the sheaf of local sections of E.
It remains the question : How to assign a vector bundle of rank k to a locally free sheaf F of rank k ?
For this, we have to introduce the notion of cocycles.

9.4 Cocycles
9.4.1 Construction
Let π : E → M be a vector bundle of rank k with a trivializing covering U = {Ui }i∈J and let U, V ∈ U be
trivializing open subsets of M such that U ∩ V =
6 ∅. By definition, we have
ϕU ϕV
π −1 (U ) = E|U ∼
= U × Ck , π −1 (U ) = E|V ∼
= V × Ck
∼ ∼
where ϕU : E|U −→ U × Ck and ϕV : E|V −→ V × Ck are morphisms of families of vector spaces as in (9.2).
ϕV |U ∩V ϕU |U ∩V
(U ∩ V ) × Ck ∼
= E|U ∩V ∼
= (U ∩ V ) × Ck

−1
⇒ ϕU |U ∩V ◦ ϕV |U ∩V : (U ∩ V ) × Ck −→
∼
(U ∩ V ) × Ck (9.5)

Let x ∈ U ∩ V be a base point. ϕU and ϕV being morphisms, they have to respect the base point, i.e.

ϕU (Ex ) = ϕU |U ∩V Ex ⊆ {x} × Ck ϕV (Ex ) = ϕV |U ∩V Ex ⊆ {x} × Ck

and

Let v ∈ Ck . We can express the map in (9.5) as

ϕU |U ∩V ◦ (ϕV |U ∩V )−1 (x, v) = x , gU V (x)(v) = (x , gU V (x) · v

where (ϕV |U ∩V )−1 (x, v) ∈ Ex (fiber over x) and gU V is a function depending on x and v induced by the restrictions

ϕU |π−1 (x) : Ex → {x} × Ck and ϕV |π−1 (x) : Ex → {x} × Ck

which are linear isomorphisms by definition (9.2) of a trivialization. Indeed gU V (x) : Ck → Ck is given by
  

−1
gU V (x)(v) = p2 ϕU |π−1 (x) ϕV |π−1 (x) (x, v) (9.6)

Hence gU V (x) : Ck → Ck is a linear isomorphism for any fixed x ∈ U ∩ V (p2 : {x} × Ck → Ck is linear
and bijective) and since all vector spaces are finite-dimensional, it can thus be represented by an invertible
k × k–matrix :
∀ x ∈ U ∩ V : gU V (x) ∈ GL(k, C)
Moreover gU V is holomorphic with respect to x since the morphisms ϕU , ϕV are holomorphic, hence so is the
composition ϕU |U ∩V ◦ (ϕV |U ∩V )−1 . gU V is the second projection of this expression, thus it is holomorphic by
definition of the product manifold. Finally gU V is a holomorphic

matrix-valued function on (U ∩ V ) × Ck .
Equivalently, one can consider gU V ∈ GL k , OM (U ∩ V ) : matrix whose entries are holomorphic functions.
   
gU V11 gU V12 ... gU V1k gU V11 (x) gU V12 (x) ... gU V1k (x)
gU V21 gU V22 ... gU V2k  gU V21 (x) gU V22 (x) ... gU V2k (x)
gU V = . ..  ⇒ gU V (x) = 
   
.. .. .. .. .. ..
 ..

. . .   . . . . 
gU Vk1 gU Vk2 ... gU Vkk gU Vk1 (x) gU Vk2 (x) . . . gU Vkk (x)

78
Complex Manifolds Section 9.4 SCHLICHENMAIER, Leytem

This matrix is invertible for all x, i.e. det gU V (x) 6= 0, ∀ x ∈ U ∩ V , hence one can also consider

−1 1


gU V = · transpose of the adjoint of gU V ∈ GL k , OM (U ∩ V )
det(gU V )

Since the determinant is non-zero everywhere on U ∩ V and taking transpose and adjoint are polynomial opera-
−1 k
tions, gU V is also a holomorphic function on (U ∩ V ) × C .
−1
Such a construction of gU V and gU V can be done for any trivializing sets U, V ∈ U.
Note in addition that gU V is uniquely determined by the vector bundle E and the covering U since is made up
of the trivialization morphisms ϕU and ϕV .

9.4.2 Properties
Let U = {Ui }i∈J be a trivializing covering of a vector bundle E over M . For all U, V, W ∈ U, we have
1) gU U = id, i.e. gU U (x) = Id ∈ GL(k, C), ∀ x ∈ U
2) gU V = (gV U )−1 if U ∩ V 6= ∅
3) If U ∩ V ∩ W 6= ∅, then gU V · gV W = gU W on U ∩ V ∩ W (matrix multiplication), i.e.

gU V (x) · gV W (x) = gU W (x) , ∀ x ∈ U ∩ V ∩ W

Proof. 2) and 3) are true because

−1 
−1 −1
ϕU |π−1 (x) ◦ ϕV |π−1 (x) = ϕV |π−1 (x) ◦ ϕU |π−1 (x)

−1 
−1  
−1 
ϕU |π−1 (x) ◦ ϕW |π−1 (x) = ϕU |π−1 (x) ◦ ϕV |π−1 (x) ◦ ϕV |π−1 (x) ◦ ϕW |π−1 (x)

By using 1) and 2), condition 3) is equivalent to :

30 ) if U ∩ V ∩ W 6= ∅, then gU V · gV W · gW U = gU U = idU ∩V ∩W

1), 2), 3) are called the cocycle conditions for the family (gU V )U,V ∈U .

9.4.3 Definition
Let M be a complex manifold and U be a set of open subset
of M which is a covering of M . An object g is called
a cocycle if ∀ U, V ∈ U, we have gU V ∈ GL k , OM (U ∩ V ) with the convention gU V = id if U ∩ V = ∅ such that
the cocycle conditions 1), 2), 3) are satisfied.

Examples :
1) M : complex manifold with U = {M } and gM M = id
2) U is given by all open sets in M and gU V = id, ∀ U, V ∈ U
Setting everything equal to identity is of course always true. The important fact is to see that cocycles always
come together with an open covering of M .

9.4.4 Inverse construction

constructed, after choice of a trivialization covering U, a cocycle
In 9.4.1 we started with a vector bundle E and
(gU V )U,V ∈U with gU V ∈ GL k , OM (U ∩ V ) , ∀ U, V ∈ U. This cocycle depends on the chosen trivialization.
Now we take the opposite direction : Given a cocycle, we want to construct a (unique) vector bundle such that
its associated cocycles exactly correspond to the starting cocycle. More precisely :

Theorem :
Let M be a complex manifold. Suppose that we are given an open coordinate covering {Uα }α∈J of M together
with a cocycle (gαβ )α,β∈J . Then there exist a unique holomorphic vector bundle E over M such that {Uα }α∈J
is a trivializing open covering for E and the canonical cocycles of E are exactly given by the (gαβ )α,β∈J .

79
Complex Manifolds Section 9.4 SCHLICHENMAIER, Leytem

Proof. Let (gαβ )α,β∈J be a cocycle such that every Uα is a chart domain in M . We define
G
E 0 := Uα × Ck

α∈J

Being defined as a disjoint union, E 0 is not connected and therefore not a vector bundle (since it’s not a manifold).

In order to make E 0 a vector bundle, we first have to connected the different components Uα × Ck (which
are all open). There is nothing we can do if Uα ∩ Uβ = ∅ (see figure 9.7). So assume that Uα ∩ Uβ 6= ∅ ; we have
to glue the overlapping parts on the intersection (figure 9.7). If gαβ = id, this is not a problem. Otherwise we
do the following :
In order to identify the fibers, we introduce the equivalence relation ∼ on E 0 given by

(x, v) ∈ Uα × Ck , (y, w) ∈ Uβ × Ck : (x, v) ∼ (y, w) ⇔ x = y and v = gαβ (y) · w

Note that x = y already implies that Uα ∩ Uβ 6= ∅ since x ∈ Uα and y ∈ Uβ .

Figure 9.7: gluing the individuals parts of E is only possible on non-empty intersections

Exercise : Show that ∼ is an equivalence relation on E 0 .

This follows from the conditions 1), 2), 3) of being a cocycle. It suffices to check it on the second argument.
− reflexivity : (x, v) ∼ (x, v) since gαα = id, thus v = gαα (x) · v = Id · v
−1
− symmetry : if (x, v) ∼ (y, w) with v = gαβ (y) · w ⇒ w = gαβ (y) · v = gβα (y) · v
− transitivity : if v = gαβ (y) · w and w = gβγ (z) · u, then v = gαβ (y) · gβγ (z) · u = gαγ (z) · u since x = y = z.

Hence we can glue the disjoint union E 0 via the cocycle g. In general, this gluing is done in a non-trivial
way since gαβ is not necessarily given by the identity. Since {Uα }α∈J is a coordinate covering of M , i.e. every
chart domain intersects at least one other chart domain (M being connected), we obtain that

E := E 0 / ∼

is connected and induces a projection map p : E → M : [x, v] 7→ x (well-defined since (x, v) ∼ (y, w) ⇒ x = y).

In order to show that E is a complex manifold, we introduce the following notation :

Let [x, v] ∈ E with a (unique) first representative x ∈ M . Consider all the open chart domains Uα of M such
that x ∈ Uα . We denote vα := the unique second representative of [x, v] such that (x, vα ) ∈ Uα × Ck . vα is unique
since E 0 is a disjoint union of open sets of the type Uα × Ck . Hence we have a well-defined bijective map

ϕUα : E −→ Uα × Ck : [x, v] 7−→ (x, vα ) for α ∈ J fixed

We denote the quotient map by ν : E 0 → E and endow E with the quotient topology with respect to ν. As chart
domains of E, we then take Vα := ν(Uα × Ck ). Vα is open in E for any α because
G
ν −1 (Vα ) = ν −1 ν(Uα × Ck ) = Uβ × Ck : open in E 0

Uβ ∩Uα 6=∅

Hence {Vα }α∈J is an open covering of E and ν is an open continuous map. As chart maps, we take

φα : Vα −→ Cn+k : [x, v] 7−→ ϕα (x), vα ∈ ϕα (Uα ) × Ck

where n = dim M and (Uα , ϕα ) is the corresponding chart of M at x.

80
Complex Manifolds Section 9.5 SCHLICHENMAIER, Leytem

φα is well-defined since we are restricted to Vα and Uα × Ck , i.e. they only cocycles which may appear are the
gαα = id. Moreover φα is a topological homeomorphism since the chart (ϕα × id) of the product manifold is one.
The transition maps are holomorphic because
 
φα ◦ φ−1 : ϕβ (Uβ ) × Ck −→ ϕα (Uα ) × Ck : (x, v) 7−→ ϕα ϕ−1


β β (x) , gαβ (x) · v

where ϕα ◦ ϕ−1
β and gαβ are holomorphic. It follows that (Vα , φα )α∈J is an atlas of E and dim E = dim M + k.

With respect to this atlas, p is surjective (as projection) and holomorphic because
ϕα ◦ p ◦ φ−1 : ϕα (Uα ) × Ck −→ ϕα (Uα ) : (x, v) 7−→ ϕα ϕ−1


α α (x) = x (9.7)
is holomorphic (the local form of the projection is again a projection).
Moreover the restricted bijective maps ϕUα |Vα : Vα → Uα × Ck are biholomorphic (as defined in 4.3.1) because

(ϕα × id) ◦ ϕUα = φα and (ϕUα )−1 = ν|Uα ×Ck

Now let x ∈ Uα ⊆ M be fixed. We want to find the fiber over x of p. For this, note that
(p ◦ ϕ−1
Uα )(x, v) = x, ∀ v ∈ C
k
⇒ ϕ−1 k
Uα (x, C ) ⊆ p
−1
(x)

And if u ∈ p−1 (x), then p(u) = x ⇒ u ∈ Vα , hence ∃ v ∈ Ck such that u = ϕ−1

Uα (x, v) since ϕUα is bijective on
Vα . It follows that
p−1 (x) = ϕ−1 k
Uα (x, C )
This space carries a complex vector space structure with respect to the definition
∀ v, w ∈ Ck , ∀ λ ∈ C : ϕ−1 −1 −1
Uα (x, v) + ϕUα (x, w) := ϕUα (x, v + w) , λ · ϕ−1 −1
Uα (x, v) := ϕUα (x, λ v)

This definition moreover implies that the restriction to the fiber ϕUα |p−1 (x) is trivially linear and hence

ϕUα |p−1 (x) : p−1 (x) −→

∼
{x} × Ck (9.8)
is a linear isomorphism between k-dimensional vector spaces. We also have to add that this definition does not
k ∼ −1
depend on the choice of α since if x ∈ Uα ∩ Uβ , then ϕ−1 k
Uα (x, C ) = ϕUβ (x, C ). This follows from the fact that

ϕ−1 k
Uα (x, C ) ϕ−1 k
Uβ (x, C )

ϕUα ϕUβ
 ϕUα ◦ϕ−1 
/ {x} × Ck
Uβ
{x} × Ck

where the considered restrictions of ϕUα and ϕUβ are linear isomorphisms as shown in (9.8) and

(ϕUα ◦ ϕ−1 k


Uβ )(x, v) = x, gαβ (x) · v , ∀ v ∈ C

with gαβ (x) ∈ GL(k, C). Thus ϕ−1 −1 −1 k

Uβ ◦ (ϕUα ◦ ϕUβ ) ◦ ϕUα , restricted to ϕUα (x, C ), is an isomorphism too.

Similarly as in (9.8), one also shows that for all α ∈ J,

α ϕU
E|Uα = p−1 (Uα ) ∼
= Uα × Ck (9.9)
Thus {Uα }α∈J is a trivializing open covering for E and it finally follows that ϕUα is a morphism of families of
vector spaces and that p : E → M is indeed a vector bundle. Now let Uα ∩ Uβ 6= ∅ and consider its cocyles g̃Uα Uβ
as given in (9.6) :
   

−1

g̃Uα Uβ (x)(v) = p2 ϕUα |p−1 (x) ϕUβ |p−1 (x) (x, v) = p2 ϕUα ϕ−1
Uβ (x, v) = gαβ (x) · v = gαβ (x)(v)

We hence recover the initial cocycles : g̃Uα Uβ = gαβ .

And uniqueness of E follows now from the fact that the cocycles of a vector bundle are uniquely given once a
trivialization has been fixed, which as shown in (9.9) is the case here.

81
Complex Manifolds Section 9.5 SCHLICHENMAIER, Leytem

9.5 Isomorphic bundles

Let E and F be vector bundles over the same base manifold M such that E ∼ = F (as defined in 9.3.5). We choose
a common trivialization covering {Uα }α∈J of E and F (take intersections of the 2 individual trivializations).
φ ψα ϕα
E∼
=F = Uα × Ck ∼
⇒ E|Uα ∼ = F|Uα , ∀ α ∈ J
∼ ∼
where ϕα : F|Uα −→ Uα × Ck and ψα : E|Uα −→ Uα × Ck . ϕα and ψα are morphisms of families of vector spaces
and their restrictions to the fibers ϕα|Fx and ψα|Ex are linear isomorphisms onto {x} × Ck , ∀ x ∈ Uα . Let also

φ : E −→ F ⇒ φα := φ|Uα : E|Uα −→ F|Uα

ψα−1 : Uα × Ck −→ Uα × Ck : (x, v) 7−→ x , cα (x) · v


⇒ ϕα ◦ φα ◦

The last map being a biholomorphic morphism of families of vector spaces (hence the base point x is preserved),
we obtain that cα (x) is an invertible k × k–matrix for all x ∈ Uα . Moreover cα is holomorphic with respect to x
on Uα , thus

cα ∈ GL k , OM (Uα ) , ∀ α ∈ J (9.10)
One can show that the opposite direction is also true, i.e. if we start with a collection (cα )α∈J satisfying (9.10), one
can associate isomorphic vector bundles E and F to this collection whose cα defined as above exactly correspond
to the initial ones.

9.5.1 Definition
Given a trivialization {Uα }α∈J , we introduce a relation on cocycles gαβ given by
0 0
= cα ◦ gαβ ◦ c−1


gαβ ∼ gαβ ⇔ ∃ (cα )α∈J with cα ∈ GL k , OM (Uα ) such that gαβ β , ∀ α, β ∈ J

We call 2 cocycles cohomologous if they are equivalent with respect to the relation ∼.

Exercise : Show that ∼ is an equivalence relation.

− reflexivity : choose a collection with cα = id, ∀ α ∈ J ⇒ gαα = id ◦ gαα ◦ id−1
− symmetry : since the cα are invertible, we get gαβ0
= cα ◦ gαβ ◦ c−1
β ⇒ gαβ = c−1 0 −1 −1
α ◦ gαβ ◦ (cβ )
0
− transitivity : if gαβ = cα ◦ gαβ ◦ c−1 00 0 0 0−1
β and gβγ = cβ ◦ gβγ ◦ cγ , then

00

−1
= c0β ◦ cβ ◦ gβγ ◦ c−1

0−1
gβγ γ ◦ cγ = (c0β ◦ cβ ) ◦ gβγ ◦ (c0γ ◦ cγ , ∀ β, γ ∈ J

9.5.2 Theorem
We state without proof :
If we start with 2 cohomologous cocycles, the associated vector bundles constructed as in 9.4.4 are isomorphic.
This implies that the isoclasses of vector bundles are in 1-to-1 correspondence with the cohomology classes of
cocycles (after the choice of a common trivialization of the bundles).

9.5.3 Proof of theorem 9.3.6

Let F be a locally free sheaf of OM –modules of rank k. We have to find an associated vector bundle such that
the sheaf of sections V of this vector bundle is again given by F (up to isomorphism). a
ψα
k
Proof. Since F is locally free, there is an open covering {Uα }α∈J of M such that F(Uα ) ∼
= OM (Uα ) , ∀ α ∈ J,
where ψα is a module isomorphism. If Uαβ := Uα ∩ Uβ 6= ∅, we hence obtain

ψα|Uαβ ψβ|Uαβ
o / OM (Uα ∩ Uβ ) k

k

OM (Uα ∩ Uβ ) F(Uα ∩ Uβ )

−1
k
k
⇒ ψα|Uαβ ◦ ψβ|Uαβ : OM (Uα ∩ Uβ ) −→ OM (Uα ∩ Uβ )

82
Complex Manifolds Section 9.6 SCHLICHENMAIER, Leytem

k
−1
Since OM (Uα ∩ Uβ ) corresponds to a vector of k holomorphic functions, the isomorphism ψα|Uαβ ◦ ψβ|Uαβ


can uniquely be given by a cocycle gαβ ∈ GL k , OM (Uα ∩ Uβ ) (same construction in 9.4.1).
Let F be the vector bundle over M which is constructed by these cocycles as in 9.4.4 and V be its sheaf of local
sections, i.e. ∀ U ⊆ M open : s ∈ V(U ) ⇒ p ◦ s = idU where p : F → M : [x, v] 7→ x. Since both sheaves are
locally free, it suffices to consider the case where U is a trivializing open set. The local form (9.7) of p is then

p : U × Ck −→ U : (x, v) 7−→ x


and a local section s ∈ V(U ) is necessarily of the form s(x) = x, s̃(x) , where s̃ : U → Ck is holomorphic. This
form is also sufficient, hence s can be identified with s̃, which can again be seen as a k-tuple of holomorphic
function U → C. It follows that
k
V(U ) ∼
= OM (U ) ∼ = F(U )
for any trivializing open subset U ⊆ M , so V ∼
= F as sheaves of OM –modules. Hence the vector bundle F satisfies
all the assumptions of the theorem.

9.5.4 Bundle maps

Let π : E → M and p : F → N be 2 arbitrary vector bundles (not necessarily over the same base manifold). A
bundle map is a pair (f, g) of holomorphic maps f : E → F and g : M → N such that p ◦ f = g ◦ π and the
restriction f|π−1 (x) : Ex → Fg(x) is a linear map for all x ∈ M .

E
f
/F
π p
 
M
g
/N

A bundle map is in fact the generalization of a morphism of families of vector spaces and it allows us to form
the category of vector bundles since bundle maps are precisely the morphisms in this category.

9.6 Frame of a vector bundle

9.6.1 Definition
Let E be a holomorphic vector bundle of rank k over a complex manifold M . A frame of E over an open set
U ⊆ M is a set of sections Σ ⊂ V(U ) such that ∀ x ∈ U , the set of sections evaluated at x is a basis of the fiber
Ex . This immediately implies that Σ must have exactly k elements, i.e. Σ = {σ1 , σ2 , . . . , σk } for σi ∈ V(U ), ∀ i.
Thus the set {σ1 (x), σ2 (x) . . . , σk (x)} is a basis of the vector space Ex . In particular σi (x) ∈ Ex , ∀ i.

Remarks :
1) This does not imply that {σ1 , σ2 , . . . , σk } is a C–basis of V(U ) (since dim V(U ) = ∞ in general).
2) Frames do not exist over all open sets U ⊆ M .
3) But : given a trivialization covering {Uα }α∈J and a local trivialization E|Uα ∼ = Uα × Ck , there always exists a
frame over Uα given by {ê1 , . . . , êk } where êi (x) = (x, ei ), ∀ i ∈ {1, . . . , k}. Since {(x, ei )}i=1,...,k is a basis of the
vector space {x} × Ck ∼ = Ex , {ê1 (x), . . . , êk (x)} is thus a basis of Ex .

We conclude that, since any vector bundle is locally of this form (locally trivial) :
For any vector bundle E over M , there is a covering (the trivializing covering) such that there exists a frame
over any open set in this covering. In particular, frames of vector bundles depend on the chosen covering of M .
Furthermore there is even a stronger statement about this fact :

9.6.2 Theorem
There exists a frame of over an open set U ⊆ M ⇔ E|U is a trivial vector bundle (i.e. E|U ∼ = U × Ck ).
In particular, if there is a global frame of E, then E must be the trivial bundle. In other words, only the trivial
vector bundle has a global frame.

83
Complex Manifolds Section 9.6 SCHLICHENMAIER, Leytem

Proof. ⇐ : This was shown above : the frame is given by {ê1 , . . . , êk }.
⇒ : Let U ⊆ M be open and assume that a frame {σ1 , . . . , σk } of E over U is given. We want to construct a
∼
morphism of families of vector spaces ϕ : E|U −→ U × Ck . Let u ∈ E|U ; since E|U is equal to the disjoint union
of its fibers Ex for x ∈ U , we know that there is a unique element x ∈ U such that u ∈ Ex , hence
k
X
u= λi · σi (x)
i=1


and the coefficients λi are uniquely given since we have a basis. Then we set ϕ(u) := x, (λ1 , . . . , λk ) . ϕ is
obviously surjective and injective since the coefficients are unique and it is (bi)holomorphic since it’s a projection.
Moreover ϕ|Ex is a linear isomorphism for all x ∈ U (once x is fixed, Ex and {x} × Ck are two k-dimensional
vector spaces over C, hence isomorphic). So all conditions in (9.2) are satisfied and ϕ defines a trivialization.

9.6.3 Frames, sections and cocycles

Another construction that one can do by using frames is the following :
Let F be a vector bundle of rank k and assume that a frame {σ1 , . . . , σk } of F over an open subset U ⊆ M is
given. Let s ∈ V(U ) be a holomorphic section. Since s(x) ∈ Ex , ∀ x ∈ U , we can decompose
k
X k
X
∀ x ∈ U : s(x) = λxi · σi (x) ⇒ ∃! f1 , . . . , fk ∈ OM (U ) such that s = fi · σi (9.11)
i=1 i=1

where the functions fi (x) = λxi are holomorphic by a similar argument as in example 9.3.3. Thus one can assign


k
s 7−→ σ := f1 , f2 , . . . , fk ∈ OM (U )
Hence there is a 1-to-1 correspondence between holomorphic sections and k-tuples of holomorphic functions :
1:1
k
V(U ) ←→ OM (U )
because the fi are uniquely determined by the section. Note that this identification requires the fixing of a frame.

Let U and V now be 2 trivializing open sets with U ∩ V 6= ∅, so there is a frame {σiU }i over U and a frame {σjV }j
over V . We want to know what happens on U ∩ V .
Let gU V be the cocycle given as in (9.6) which defines the vector bundle, i.e. we have gU V = ”ϕU ◦ ϕ−1V ” where
∼
ϕU : E|U −→ U × Ck , ∼
ϕU |π−1 (x) : Ex −→ {x} × Ck
∼
ϕV : E|V −→ V × Ck , ∼
ϕV |π−1 (x) : Ex −→ {x} × Ck
We want to show that the 2 frames are related via gU V .
Let s ∈ V(U ∩ V ) ; using (9.11), s can now be decomposed in both frames as
7 → f 0 := f10 , f20 , . . . , fk0 , fi0 ∈ OM (U ∩ V )



s 7−→ f := f1 , f2 , . . . , fk , fi ∈ OM (U ∩ V ) and s −

9.6.4 Theorem
f = gU V · f 0 , where · denotes the matrix multiplication.
Proof. From linear algebra, we know that the transformation law for the coefficients is f = M · f 0 where f are
the old coordinates, f 0 the new ones and M is the basis change matrix from {σiU }i to {σjV }j . Hence it remains
to find the matrix M . Indeed the basis transformation is given by the cocycles because ∀ x ∈ U ∩ V :
ϕU |π−1 (x) ϕV |π−1 (x)
Ck ∼
= {x} × Ck ←− Ex = h σ1U (x), . . . , σkU i = h σ1V (x), . . . , σkV i = Ex −→ {x} × Ck ∼
= Ck
where the basis change {σiU }i → {σjV }j induces a basis change in Ck . Hence we know that the coordinate change

−1
is given by ϕU |π−1 (x) ◦ ϕV |π−1 (x) . But this is exactly the definition of the cocycles, so M = gU V .

9.6.5 Conclusion
Sections of E correspond to local vectors of functions with respect to certain frames which transform in a certain
manner, essentially given by the defining cocycles of the vector bundles.

84
Chapter 10

Operations on vector bundles

10.1 Induced operations on vector bundles

Let E and M be complex manifold and π : E → M be a surjective holomorphic map. We know that a vector
bundle of rank k is defined by
1) an open trivialization covering {Uα }α∈J of M
∼
2) local trivialization morphisms ϕUα : E|Uα −→ Uα × Ck that are biholomorphic and linear isomorphisms
when restricted to the fibers
3) the cocycles (also called patching functions) gαβ : Uα ∩ Uβ → GL(k, C) that are holomorphic such that
gαβ = ”ϕUα ◦ ϕ−1
Uβ ” and satisfying the cocycle relations

Problem :
Given a vector bundle, we want to construct a new one by using these data.
In the following, let π : E → M and p : F → M be 2 complex vector bundles of rank k and l respectively over
the same complex base manifold M . We choose a common trivialization {Uα }α∈J and denote the cocycles of E
by gαβ and those of F by hαβ .
Let also V and W be complex vector spaces of finite dimension k = dim V , l = dim W with bases B = {ei }i=1,...,k
and B 0 = {e0i }i=1,...,l respectively. The goal is to show that operations on vector spaces induce the same operations
on vector bundles. The idea is to apply the operations in each fiber (which is a vector space) of the bundle.

10.1.1 Dual bundle

Recall that the dual space of V is given by

V ∗ = HomC (V, C) = ϕ : V → C | ϕ linear and continuous ⇒ dim V ∗ = k



It is in fact not necessary to require continuity since any linear map in finite dimension is also continuous. We
also recall that for a linear map T : V → W , one defines its dual map t T : W ∗ → V ∗ (which is also linear) by

∀ α ∈ W ∗ , t T (α) := α ◦ T ⇔ t T (α)(v) = α T (v) ∈ C, ∀ v ∈ V

Whenever well-defined we have the relations (t T )−1 = t (T −1 ) and t (T1 ◦ T2 ) = t T2 ◦ t T1 .

Moreover if T is given by the matrix A (after choice of a basis), then the matrix associated to t T is t A (transpose).
Proof. The matrix A = (aij ) is defined by the relation T (ei ) = j aji e0j , ∀ i ∈ {1, . . . , k}. Let {εi }i=1,...,k and
P

{ε0i }i=1,...,k be the standard bases of V ∗ and W ∗ associated to B and B 0 , i.e. εi (ej ) = ε0i (e0j ) = δji , ∀ i, j. Then
t
T (ε0i ) = ε0i ◦ T = bji εj ∈ V ∗ ⇒ t T (ε0i )(ek ) = bji εj (ek ) = bki
P P
j j
P  P
t
T (ε0i )(ek ) = (ε0i ◦ T )(ek ) = ε0i T (ek ) = ε0i ajk e0j = ajk ε0i (e0j ) = aik


j j

Hence the coefficients of the matrix associated to t T satisfy bij = aji = t aij : we get the matrix t A.

85
Complex Manifolds Section 10.1 SCHLICHENMAIER, Leytem

Now we want to define the dual bundle E ∗ → M . We know that E =

F
Ex is locally given by
ϕUα x∈M
E|Uα ∼
= U α × Ck
Each fiber Ex being a vector space, we can take its dual space (Ex )∗ and we define the dual bundle E ∗ as
G
E ∗ := (Ex )∗
x∈M

i.e. we take the same foot-map π : E ∗ → M with π −1 (x) = (E ∗ )x = (Ex )∗ (take the dual in each fiber). In order
to describe the structure of E ∗ we have to find its trivialization and its cocycles. For this, we consider the map
∼
ϕx := ϕUα |Ex : Ex −→ {x} × Ck

which is a linear isomorphism and a basis of Ex is given by {ϕ−1 t k ∗ ∼

x (x, ei )}i=1,...,k . Thus ϕx : {x}×(C ) −→ (Ex )
∗
∗ t i
is an isomorphism too and a basis of (Ex ) is given by { ϕx (x, ε )}i=1,...,k . Finally we obtain that

−1
t
ϕx : (Ex )∗ −→
∼
{x} × (Ck )∗

This holds for any x ∈ M , hence a trivialization of E ∗ is given by ψUα : (E ∗ )|Uα −→

∼
Uα × (Ck )∗ where {Uα }α∈J

−1
is the same trivializing open covering as for E and ψUα = t ϕUα . For the cocycles this now implies that

−1  −1 
−1 −1 
−1
−1 t
◦ t ϕUβ = t
◦ t ϕUα = t ϕ−1 = t ϕUα ◦ ϕ−1

t
ψUα ◦ ψUβ
= ϕUα ϕUβ Uβ ◦ ϕUα Uβ

−1 t 
−1  t
: Uα ∩ Uβ −→ GL(k, C) : jαβ (x) = t gαβ (x)


⇒ jαβ = gαβ (x) = gβα (x)

These jαβ are in addition holomorphic and satisfy the cocycle relations : jαα = id, jβα = (jαβ )−1 and

jαβ (x) ◦ jβγ (x) = t gβα (x) ◦ t gγβ (x) = t gγβ (x) ◦ gβα (x) = t gγα (x) = jαγ (x)

In particular E ∗ is again a vector bundle of rank k.

10.1.2 Direct sum bundle

The ”direct sum” of V and W is defined as

V ⊕W ∼

=V ×W = (v, w) | v ∈ V, w ∈ W

Hence dim(V ⊕ W ) = k + l and a basis is given by B ∪ B 0 (here we do not mean the direct sum of 2 vector
subspaces of some bigger vector space). If V 0 and W 0 are 2 other vector spaces and we are given the linear maps
S : V → V 0 and T : W → W 0 , we can define the direct sum S ⊕ T by

S ⊕ T : V ⊕ W −→ V 0 ⊕ W 0 , (S ⊕ T )(v, w) := S(v), T (w)

If S and T are represented by matrices A and B (after choice of a basis), the matrix associated to S ⊕ T in the
induced basis given as above is  
A 0
S⊕T = (10.1)
0 B
since S only acts on V and T only acts on W .
We are now able to define the direct sum of 2 vector bundles E and F . As in 10.1.1 we have the local situation
α ϕU ψUα
E|Uα ∼
= Uα × Ck , F|Uα ∼
= Uα × Cl
Ex and Fx being vector spaces for any x ∈ M , we can take their direct sum Ex ⊕ Fx and hence define
G
E ⊕ F := (Ex ⊕ Fx )
x∈M

i.e. the fibers of the direct sum E ⊕ F are given by (E ⊕ F )x = Ex ⊕ Fx . We again have to find the trivialization
and the cocycles of this new vector bundle.

86
Complex Manifolds Section 10.1 SCHLICHENMAIER, Leytem

Consider the linear isomorphisms

∼
ϕx := ϕUα |Ex : Ex −→ {x} × Ck , ∼
ψx := ψUα |Fx : Fx −→ {x} × Cl

They induce the map ϕx ⊕ ψx : Ex ⊕ Fx → {x} × (Ck ⊕ Ck ) which is again an isomorphism because of (10.1).
It follows that the local trivialization of E ⊕ F is given by
∼
ρUα : (E ⊕ F )|Uα −→ Uα × (Ck ⊕ Cl )

with ρUα = ϕUα ⊕ ψUα and E ⊕ F is a vector bundle of rank k + l. For the cocycles note that
−1
A−1
  
A 0 0
= ⇒ ρ−1
Uα = (ϕUα ⊕ ψUα )
−1
= ϕ−1 −1
Uα ⊕ ψUα
0 B 0 B −1

and hence ρUα ◦ ρ−1 −1 −1 −1 −1

Uβ = (ϕUα ⊕ ψUα ) ◦ (ϕUβ ⊕ ψUβ ) = (ϕUα ◦ ϕUβ ) ⊕ (ψUα ◦ ψUβ ) so that finally
 
gαβ (x) 0
jαβ : Uα ∩ Uβ −→ GL(k + l, C) : jαβ (x) = gαβ (x) ⊕ hαβ (x) =
0 hαβ (x)

This is obviously holomorphic and jαβ satisfies the cocycle conditions since gαβ and hαβ do.

10.1.3 Tensor product bundle

The tensor product of V and W is given by

V ⊗ W = L2 (V ∗ × W ∗ , C) = ϕ : V ∗ × W ∗ → C | ϕ bilinear


i.e. elements of the tensor product are bilinear forms defined on the corresponding dual spaces. They write as a
finite linear combination of terms of the form v ⊗ w for some v ∈ V , w ∈ W , defined by the condition

(v ⊗ w)(f, g) := f (v) · g(w) ∈ C

and extended by linearity. Moreover a basis of V ⊗ W is given by {ei ⊗ e0j }i,j , thus dim(V ⊗ W ) = k · l.
If V 0 and W 0 are again 2 other vector spaces and S : V → V 0 and T : W → W 0 are linear maps, they induce
the tensor map
S ⊗ T : V ⊗ W −→ V 0 ⊗ W 0 , (S ⊗ T )(v ⊗ w) := S(v) ⊗ T (w) (10.2)
again extended by linearity. We will not discuss the matrix representation of this map. Moreover it must be said
that the representation of an element of the tensor product as a finite linear combination is not uniquely given,
hence (10.2) may a priori not be well-defined since it must be linearly extended. One can however show that the
map is independent of this decomposition and that (10.2) is always well-defined.

Concerning the tensor product of 2 vector bundles E and F , we again have the local situation
ϕU
α ψU
α
E|Uα ∼
= U α × Ck , F|Uα ∼
= U α × Cl
and define the tensor bundle E ⊗ F again as the disjoint union of the tensor products of all individual fibers, i.e.
G
E ⊗ F := (Ex ⊗ Fx )
x∈M

The constructions being exactly the same as in 10.1.2, we will omit the details and end up with the trivialization
∼
ρUα : (E ⊗ F )|Uα −→ Uα × (Ck ⊗ Cl )

where ρUα = ϕUα ⊗ ψUα , thus E ⊗ F is a vector bundle of rank k · l. And the cocycles are given by

jαβ : Uα ∩ Uβ −→ GL(k · l, C) : jαβ (x) = gαβ (x) ⊗ hαβ (x)

87
Complex Manifolds Section 10.1 SCHLICHENMAIER, Leytem

10.1.4 Exterior power bundles

For any r ∈ {0, 1, 2, . . . , k} we can define the rth exterior powers of V , given by
⊗r V = Lr (V ∗ × . . . × V ∗ , C) = ϕ : V ∗ × . . . × V ∗ → C ϕ linear in each argument


Λr V = Ar (V ∗ × . . . × V ∗ , C) = ϕ ∈ Lr (V ∗ × . . . × V ∗ , C) ϕ alternating


i.e. ϕ ∈ Λr V ⇔ ϕ ∈ ⊗r V and ϕ(α1 , . . . , αr ) = sign(σ) · ϕ(ασ(1) , . . . , ασ(r) ), ∀ α1 , . . . , αr ∈ V ∗ , ∀ σ ∈ Sr .

This implies that Λr V = {0} if r > k. By convention, we set ⊗0 V = Λ0 V = C.

A basis for ⊗r V is given by {ej1 ⊗ . . . ⊗ ejr }j1 ,...,jr where {ei }i=1,...,k is the standard basis of V and
(ej1 ⊗ . . . ⊗ ejr )(α1 , . . . , αr ) := α1 (e1 ) · . . . · αr (ejr ) ∈ C , ∀ α1 , . . . , αr ∈ V ∗
It follows that dim(⊗r V ) = k r and that any element from ⊗r V writes as a linear combination of such functions.
For T ∈ Λr V and S ∈ Λs V , we define the wedge product T ∧ S ∈ Λr+s V pointwise by
1 X
(T ∧ S)(α1 , . . . , αr+s ) := sign(σ) · T ασ(1) , . . . , ασ(r) · S ασ(r+1) , . . . , ασ(r+s)



·
r! s!
σ∈Sr+s
One can show that this product is associative, distributive
and anti-commutative. Moreover a basis of Λr V is
given by {ei1 ∧ . . . ∧ eir }1≤i1 <...<ir ≤k , hence dim(Λr V ) = kr .

If T : V → W is a linear map, it induces the 2 following exterior power maps :

⊗r T : ⊗r V −→ ⊗r W , ⊗r T (ej1 ⊗ . . . ⊗ ejr ) := T (ej1 ) ⊗ . . . ⊗ T (ejr )
Λr T : Λr V −→ Λr W , Λr T (ei1 ∧ . . . ∧ eir ) := T (ei1 ) ∧ . . . ∧ T (eir )
extended by linearity. One can again show that this is indeed well-defined. If S : Z → V is another linear map
and whenever T is a linear isomorphism, then ⊗r T and Λr T are again isomorphisms and they satisfy
(⊗r T )−1 = ⊗r (T −1 ) , ⊗r (T ◦ S) = ⊗r T ◦ ⊗r S , (Λr T )−1 = Λr (T −1 ) , Λr (T ◦ S) = Λr T ◦ Λr S (10.3)
∼ k
Let E now be a vector bundle with local trivialization ϕUα : E|Uα −→ Uα × C . As before we define the exterior
power bundles ⊗r E and Λr E by taking the exterior powers in each fiber :
G G
⊗r E := (⊗r Ex ) , Λr E := (Λr Ex )
x∈M x∈M

Similarly as in the previous examples, we then obtain the local trivializations

ψUα : (⊗r E)|Uα −→
∼
Uα × (⊗r Ck ) ρUα : (Λr E)|Uα −→
, ∼
Uα × (Λr Ck )

showing that ⊗r E and Λr E are holomorphic vector bundles of rank k r and kr respectively. The cocycles

jαβ : Uα ∩ Uβ −→ GL(k r , C) : jαβ (x) = ⊗r gαβ (x)

lαβ : Uα ∩ Uβ −→ GL kr , C : lαβ (x) = Λr gαβ (x)

are holomorphic and satisfy the cocycle conditions because of (10.3).

10.1.5 Particular case

For k = dim V , Λk V is a vector space of dimension kk = 1, hence Λk E will be a vector bundle of rank 1. It is a

line bundle, called the top exterior power of E. It is also called the determinant bundle and denoted Λk E = det E.
The reason is the following :
Let Vectf (C) be the category of finite-dimensional vector spaces over C and consider the functor
Λr : Vectf (C) −→ Vectf (C) : V 7−→ Λr V
HomC (V, W ) −→ HomC (Λr V, Λr W ) : T 7−→ Λr T
Take W = V . Since Λk V is 1-dimensional, we obtain that Λr T : Λr V → Λr V , i.e. Λr T is a linear map from a
1-dimensional space to another 1-dimensional space. Hence Λr T ∈ Mat(1, C) ∼= C and Λr T is just given by the
multiplication with a complex number. It actually turns out that Λ T (λ) = det T · λ, ∀ λ ∈ Λr V .
r

88
Complex Manifolds Section 10.2 SCHLICHENMAIER, Leytem

10.1.6 Generalization
We have seen that the construction of these new vector bundles is essentially always the same ; all you need is
an operation on vector spaces that induces an associated operation on linear maps between vector spaces. Thus
we have the following generalization :
If Vectf (C) denotes again the category of finite-dimensional vector spaces over C, the morphisms of this category
are given by linear maps between these vector spaces. Let F : Vectf (C) → Vectf (C) be a vector space functor,
for example :
F = ∗ ⇒ F(V ) = V ∗ and F(T ) = t T
F = Λr ⇒ F(V ) = Λr V and F(T ) = Λr T
where ∗ is a contravariant and Λr is a covariant functor. Then F induces a functor F 0 on the category of vector
bundles given by the following data :
To a vector bundle π : E → M with trivialization E|Uα ∼ = Uα × Ck and cocycles gαβ : Uα ∩ Uβ → GL(k, C), F 0
associates the vector bundle π : F(E) → M defined by
G
F(E) := F(Ex )
x∈M

with the trivialization F(E)|Uα ∼

= Uα ×F(C ) and cocycles F(gαβ ) : Uα ∩Uβ → GL(k 0 , C) where k 0 = dim F(Ck ).
k

Remark :
The same of course also applies for constructions involving bifunctors of vector spaces, as e.g. ⊕ and ⊗.

10.2 Sub-bundles and quotient bundles

10.2.1 Definition
Let π : E → M be a holomorphic vector bundle of rank k.
A vector sub-bundle of E is a family of vector subspaces {Fx ⊆ Ex }x∈M , indexed by M , such that
G
F := Fx
x∈M

is a complex submanifold of the manifold E and π|F : F → M is itself a vector bundle of rank l ≤ k. This
∼
implies that ∀ x ∈ M there is an open neighborhood U of x in M and a trivialization ϕU : E|U −→ U × Ck such
∼
that the restriction to F|U satisfies ϕU |F|U : F|U −→ U × Cl ⊆ U × Ck for l ≤ k.
The intuitive idea of a vector sub-bundle is that F ⊆ E such that for any point in M there is an open neighborhood
on which both bundles E and F are trivial and the trivialization maps for F are nothing but the restrictions to
F of the trivialization maps of E.

10.2.2 Examples
1) For an open set U ( M , E|U is not a sub-bundle of E since it is not indexed over M . In a sub-bundle F the
fiber over any point x ∈ M must be non-empty and of dimension l (see figure 10.1).
Figure 10.1: F is a sub-bundle of E

2) Any vector bundle E is a sub-bundle over itself.

3) The zero-bundle given by Fx = {0}, ∀ x ∈ M is a sub-bundle of any vector bundle E.
Moreover it is in 1-to-1 correspondence with M since x∈M {0} ∼
F
= M.

89
Complex Manifolds Section 10.2 SCHLICHENMAIER, Leytem

10.2.3 Vector space quotients

Let V and W be vector spaces of dimension k and l and consider the vector subspaces V 0 ⊆ V and W 0 ⊆ W of
dimension k 0 ≤ k and l0 ≤ l. Assume that bases for V and W are given by {ei }i=1,...,k and {e0i }i=1,...,l such that
{ei }i=1,...,k0 and {e0i }i=1,...,l0 are bases of V 0 and W 0 .
The quotient space V /V 0 is defined by the equivalence relation x ∼ y ⇔ ∃ v ∈ V 0 such that x = y + v 0 . We
denote the equivalence class of x ∈ V by x̄. It follows that dim(V /V 0 ) = k − k 0 and a basis of V /V 0 is given by
{ēi }i=k0 +1,...,k . Any element v ∈ V 0 satisfies v̄ = 0.
Let now f : V → W be a linear map such that f (V 0 ) ⊆ W 0 , i.e. f preserves the vector subspaces. In the bases
of V and W , f is then given by a matrix of the type
 
A B
for A : l0 × k 0 , B : l0 × (k − k 0 ) , D : (l − l0 ) × (k − k 0 )


f = aij i,j =
0 D

This happens because for any ei with i ∈ {1, . . . , k 0 }, we have ei ∈ V 0 ⇒ f (ei ) ∈ W 0 , hence
0
l
X l
X
0
∀ i ∈ {1, . . . , k } : f (ei ) = aji e0j + 0 · e0j ⇒ aji = 0, ∀ j ∈ {l0 + 1, . . . , l}
j=1 j=l0 +1

or in other words : aij = 0 if i is ”big” and j is ”small”. We moreover see that f|V 0 : V 0 → W 0 is represented by
the matrix A. f now induces the map f¯ : V /V 0 → W/W 0 defined by

f¯(x̄) := f (x)

This is well-defined since f is linear and satisfies f (V 0 ) ⊆ W 0 : f (x + v) = f (x) + f (v) = f (x) for any v ∈ V 0
because f (v) ∈ W 0 . Moreover f¯ is then given by the matrix D since all first basis vectors vanish in the quotient.

10.2.4 Definition
Let E be a vector bundle of rank k and F be a sub-bundle of E of rank l. The quotient bundle E/F is defined by
G
E/F := (Ex /Fx )
x∈M

i.e. the fibers of E/F are given by the quotients of the individual fibers : (E/F )x = Ex /Fx .

10.2.5 Cocycles for sub-bundles and quotients

Let E be a vector bundle of rank k, F a sub-bundle of E of rank l and E/F be the quotient bundle. If x ∈ M
∼ ∼
and U, V are 2 trivializing open sets for E with U ∩ V 6= ∅, then ϕU : E|U −→ U × Ck , ϕV : E|V −→ V × Ck and
 
hU V (x) kU V (x)
gU V (x) = ”ϕU ◦ ϕ−1
V ” = ∈ GL(k, C)
iU V (x) jU V (x)

with hU V (x) : l × l, kU V (x) : l × (k − l), iU V (x) : (k − l) × l and jU V (x) : (k − l) × (k − l). First of all we need
that iU V = 0 since F is a sub-bundle of E, hence the subspace Cl ⊂ Ck must be preserved. This is true because
 
hU V (x) kU V (x)
ψU = ϕU |F|U : F|U −→ ∼
U × Cl ⇒ gU V (x) = : Ck −→ Ck
0 jU V (x)

The cocycles of F are induced by ψU ◦ ψV−1 , i.e. by a simple restriction of ϕU ◦ ϕ−1

V and hence given by the first
l components of gU V in order to preserve the subspace Cl ⊂ Ck . Finally the cocycles of F are the

hU V : U ∩ V −→ GL(l, C)

In order to find a trivialization and cocycles of E/F , let {Uα }α∈J be a trivializing open covering for E with
∼
ϕUα : E|Uα −→ Uα × Ck , ∼
ϕUα |Ex = ϕx : Ex −→ {x} × Ck , ∼
ϕUα |Fx = ϕx|Fx : Fx −→ {x} × Cl

90
Complex Manifolds Section 10.3 SCHLICHENMAIER, Leytem

∼
This induces the map ϕ̄x : Ex /Fx −→ {x} × Ck /Cl ∼
= Ck−l and we construct similarly
∼
ψUα : (E/F )|Uα −→ Uα × Ck−l

ϕ̄x arising from the lower-right (k − l) × (k − l) matrix describing ϕx , ψUα is given by a similar expression and
−1
the cocycles are induced by the ψUα ◦ ψU β
, i.e. the cocycles of E/F are the lower-right part of gU V :

jU V : U ∩ V −→ GL(k − l, C)

10.3 Pull-back bundles

10.3.1 Definition
Let π : E → M be a holomorphic vector bundle of rank k and f : N → M be a holomorphic map of complex
manifolds. The pull-back of E along f is the vector bundle over N defined by

f ∗ E := (x, u) ∈ N × E | f (x) = π(u)



with the first projection p1 : f ∗ E → N as foot-map. Hence the following diagram commutes by definition :

f ∗E
p2
/E
p1 π
 
N
f
/M

The fiber of p1 over x ∈ N is therefore given by

(f ∗ E)x = p−1 ∗
 
1 (x) = (y, u) ∈ f E | y = x = (x, u) | u ∈ E, f (x) = π(u)
= (x, u) | u ∈ π −1 f (x) = {x} × π −1 f (x) = {x} × Ef (x) ∼



= Ef (x)

It follows that the rank of f ∗ E is also equal to k. Using the Implicit Function Theorem one can moreover show
that f ∗ E is a complex submanifold of N × E. In order to show that f ∗ E defines indeed a vector bundle, we have
to find a trivialization and the associated cocycles.
∼
Let ϕUα : E|Uα −→ Uα × Ck be a trivialization for E with cocycles gαβ : Uα ∩ Uβ → GL(k, C) and Uα , Uβ open
sets in M such that Uα ∩ Uβ 6= ∅. f being continuous, the sets Vα = f −1 (Uα ) are an open covering of N (note
however that Vα can be empty even if Uα 6= ∅). Now consider
∼
∀ y ∈ M, ϕy : Ey −→ {y} × Ck
∼
⇒ ∀ x ∈ N, ϕf (x) : Ef (x) −→ {f (x)} × Ck ∼
= {x} × Ck (10.4)

where f (x) ∈ Uα if x ∈ Vα . This is not really a composition (ϕf (x) 6= ϕx ◦ f ), but we can nevertheless take

ψVα : (f ∗ E)|Vα −→
∼
V α × Ck

where ψVα = ”ϕUα ◦f ” as a trivialization for f ∗ E. But (10.4) suggests that the cocycles are given by an expression
where any x has been replaced by f (x). And this is indeed true because we can take


jαβ : Vα ∩ Vβ −→ GL(k, C) : jαβ (x) = gαβ f (x)

i.e. jαβ = gαβ ◦ f , and this composition is well-defined. The relations can be summarized as follows :
gαβ
Uα ∩ Uβ / GL(k, C)
O 9
f
jαβ

Vα ∩ Vβ
Remark :
In the case where i : N ,→ M is the inclusion, i∗ E will be the restriction of the vector bundle E to E|N .

91
Complex Manifolds Section 10.4 SCHLICHENMAIER, Leytem

10.3.2 Problem
Let π1 : E → M and π2 : F → M be 2 vector bundles over the same base manifold M and ψ : E → F be a
holomorphic morphism of families of vector spaces, i.e. π2 ◦ ψ = π1 such that ψ preserves the fibers.

E
ψ
/F
π1
π2
 ~
M

Moreover the map ψx : Ex → Fx is linear (but not an isomorphism in general), hence we can consider the
finite-dimensional vector spaces ker ψx and im ψx . The idea is to define
G G
ker ψ := ker ψx ⊆ E and im ψ := im ψx ⊆ F
x∈M x∈M

There are however a certain number of problems with these definitions ; in particular ker ψ and im ψ do not
necessarily define sub-bundles of E and F . First of all im ψx is in general not always closed, hence ker ψ is
not closed and cannot be a complex submanifold of F . But the bigger problem is that, for both of them, the
dimensions of the fibers can ”jump” : dim(ker ψx ) and dim(im ψx ) are not uniquely determined by E and F , but
strongly depend on the linear maps ψx . Hence ker ψ is only a sub-bundle of E if the maps ψx all have the same
rank. And for im ψ it is in addition required that all images im ψx are closed.

Conclusion :
There are no kernels or cokernels in the category of vector bundles. In particular, since the category of (isoclasses
of) vector bundles of rank k is equivalent to the category of locally free sheaves of rank k, the locally free sheaves
do not admit kernels or cokernels neither. If one wants to consider kernels and cokernels, one has to pass to the
so-called coherent sheaves (sheaves where the rank can ”jump”).

10.4 Associated frames

Let M be a complex manifold and π : E → M be a holomorphic vector bundle of rank k with local trivialization
∼
ϕU : E|U −→ U × Ck

We recall that a frame for E over U (which always exists since U is a trivializing open set, see section 9.6.1) is a
collection {σ1 , . . . , σk } of local sections of E over U such that for all x ∈ U , {σ1 (x), . . . , σk (x)} is a basis for the
fiber Ex (k-dimensional vector space).
Let F be another holomorphic vector bundle over M with the same trivialization as E, i.e. locally

E|U ∼
= U × Ck , F|U ∼
= U × Cl

and assume that {σ1 , . . . , σk } is a frame for E over U and {τ1 , . . . , τl } is a frame for F over U . We want to find
expressions for the frames over U of the associated vector bundles constructed in 10.1.

1) The dual bundle E ∗ locally looks like (E ∗ )|U ∼ = U × (Ck )∗ . We have to find a basis of (Ex )∗ , ∀ x ∈ U .
Since a basis of Ex is already given, we simply take the canonical basis of the dual space. Thus a frame of E ∗
over U is given by {σ1∗ , . . . , σk∗ }, where σi∗ : U → (Ex )∗ ⇒ σi∗ (x) : Ex → C and σi∗ (x) is linear, ∀ x ∈ U . And
to define the σi∗ (x) on Ex , it is by linearity sufficient to give its value on each basis vector σj (x) of Ex . We set

σi∗ (x) : Ex → C , σi∗ (x) σj (x) := δij

2) The direct sum E ⊕ F locally looks like (E ⊕ F )|U ∼

= U × (Ck ⊕ Cl ) and a frame over U is given by

{σ1 , . . . , σk , τ1 , . . . , τl }

= Ck and {τ1 (x), . . . , τl (x)} is a basis for Fx ∼

Indeed, ∀ x ∈ U , {σ1 (x), . . . , σk (x)} is a basis for Ex ∼ = Cl , hence
∼
we know from linear algebra that {σ1 (x), . . . , σk (x), τ1 (x), . . . , τl (x)} is a basis for Ex ⊕ Fx = C ⊕ Cl .
k

92
Complex Manifolds Section 10.5 SCHLICHENMAIER, Leytem

3) The tensor product E ⊗ F locally looks like (E ⊗ F )|U ∼

= U × (Ck ⊗ Cl ) and a frame over U is given by

{σi ⊗ τj }i=1,...,k;j=1,...,l

where σi ⊗ τj (x) := σi (x) ⊗ τj (x) ∈ Ex ⊗ Fx ∼ = Ck ⊗ Cl , ∀ x ∈ U . Again this is the case because {σ1 (x), . . . , σk (x)}
and {τ1 (x), . . . , τl (x)} are bases for Ex and Fx , hence {σi (x) ⊗ τj (x)}i,j is a basis for Ex ⊗ Fx = (E ⊗ F )x .

4) The exterior powers ⊗k E and Λr E locally look like (⊗r E)|U ∼ = U × (⊗r Ck ) and (Λr E)|U ∼ = U × (Λr Ck ).
Hence as in 10.1.4, frames over U are given by {σj1 ⊗ . . . ⊗ σjr }j1 ,...,jr and {σi1 ∧ . . . ∧ σir }i1 <...<ir , where

(σj1 ⊗ . . . ⊗ σjr )(x) := σj1 (x) ⊗ . . . ⊗ σjr (x) and (σi1 ∧ . . . ∧ σir )(x) := σi1 (x) ∧ . . . ∧ σir (x)

In the case of the determinant bundle det E = Λk E of rank 1, this frame becomes

(σ1 ∧ . . . ∧ σk )(x) := σ1 (x) ∧ . . . ∧ σk (x) ∈ Λk (Ex ) ∼

= Λk (Ck ) ∼
=C

and actually corresponds to taking the volume form in each fiber.

10.5 Line bundles

10.5.1 Definition
A line bundle over a complex manifold M is a holomorphic vector bundle of rank 1, i.e. locally on a trivializing
open covering {Uα }α∈J it looks like Uα × C and all fibers are 1-dimensional.
An advantage of line bundles is that a lot of the vector bundle properties simplify in this case.

a) The cocycles are given by gU V : U ∩ V → GL(1, C) ∼ −1

= C \ {0} , · with invertibility condition gU V = gV U ,
so we need that gU V (x) 6= 0, ∀ x ∈ U ∩ V . More precisely, gU V is a map that associates to any x ∈ U ∩ V
the linear map given by multiplication with the non-zero complex number gU V (x), i.e. gU V is holomorphic with
respect to x and can be identified with a map U ∩ V → C. Hence gU V ∈ O∗ (U ∩ V ), where O∗ is the sheaf of
nowhere-vanishing holomorphic functions to ensure that gU1V is well-defined and holomorphic again.

b) A frame for a line bundle E over an open set U ⊆ M is just a holomorphic function σ ∈ O∗ (U ). Since
the fibers are 1-dimensional (hence isomorphic to C), we just need 1 non-zero vector (which can thus be identi-
fied with a non-zero complex number) to define a basis. This must be satisfied for any x ∈ U , hence the function
σ : U → C should not vanish on U (since if σ(x) = 0 for some x ∈ U , then {σ(x)} is not a basis of Ex ∼ = C).
−1
c) The dual bundle E ∗ is again a line bundle and its cocycles are jU V = (t gU V )−1 = gU V = gV U =
1
gU V , i.e.

1
jU V (x) = , ∀x ∈ U ∩ V
gU V (x)

The reason why the dual map has no affect on gU V is that a linear map T can be identified with its dual map
t
T in dimension 1 (because if T is represented by the matrix A, then the matrix of t T is t A = A too) ; T and t T
are not the same maps, but they have the same associated matrix and are hence equivalent.

d) The tensor product of 2 line bundle E and F over M is again a line bundle since rk(E ⊗ F ) = rk E · rk F = 1.
By choosing a common trivialization for E and F , we have locally E|U ∼
= U × C and F|U ∼ = U × C, hence

(E ⊗ F )|U ∼
= U × (C ⊗ C) ∼
=U ×C

If the cocycles of E and F are gU V and hU V , then the cocycles of the tensor bundle E ⊗ F are given by


jU V (x) = gU V (x) ⊗ hU V (x) = v 7→ gU V (x) · hU V (x) · v (10.5)

where gU V (x) ∈ C, hU V (x) ∈ C and the tensor product ⊗ in dimension 1 is nothing but the usual multiplication.
After identification we thus can write that jU V ∈ O∗ (U ∩ V ) is given by jU V (x) = gU V (x) · hU V (x), ∀ x ∈ U ∩ V .

93
Complex Manifolds Section 10.5 SCHLICHENMAIER, Leytem

10.5.2 Theorem
The set of isoclasses of line bundles over a complex manifold M is an abelian group under the ⊗-operation. This
group is called the Picard group.

Proof. Recall that E ∼ F ⇔ E ∼ = F as vector bundles as defined in 9.3.5

For associativity and commutativity we have to show that E ⊗ (F ⊗ G) ∼ = (E ⊗ F ) ⊗ G and E ⊗ F ∼ = F ⊗ E.
By theorem 9.5.2, we know that cohomologous cocycles define isomorphic vector bundles. Hence in our case it
already suffices to show that both vector bundles have the same cocycles (hence cohomologous cocycles). If the
cocycles of E, F, G are gU V , hV U , jU V , then (10.5) shows that the cocycles of the tensor products are

E ⊗ (F ⊗ G) −→ gU V · (hV U · jU V ) , (E ⊗ F ) ⊗ G −→ (gU V · hV U ) · jU V
E ⊗ F −→ gU V · hV U , F ⊗ E −→ hV U · gU V

Hence these 2 properties follow from associativity and commutativity of the complex numbers.
For the neutral element we have to find a line bundle E0 such that E0 ⊗ E ∼
= E ⊗ E0 ∼= E for any line bundle E.
We take E0 := M × C, the trivial bundle, whose cocycles are given by gU V (x) = 1, ∀ x ∈ U ∩ V since we have
∼
ϕU : E0|U −→ U × C ⇒ ϕU = id, ∀ U ⊆ M open ⇒ gU V = ”ϕU ◦ ϕ−1
V ” = id as well

Concerning the inverse, given a line bundle E, we want to find E −1 such that E ⊗ E −1 ∼ = E −1 ⊗ E ∼= E0 . We
take the dual bundle E ∗ = E −1 because the cocycles of E ∗ are gU1V , so gU V (x) · gU V1(x) = 1, ∀ x ∈ U ∩ V .

10.5.3 Proposition
A line bundle is trivial ⇔ it admits a nowhere-vanishing global section.
Proof. By theorem 9.6.2, a trivialization on an open set U ⊆ M is equivalent to the existence of a frame over U .
⇒ : If a line bundle E is trivial, it admits a frame over M . But since the fibers are 1-dimensional, there must
exists a global section σ : M → E such that {σ(x)} is a basis for Ex , ∀ x ∈ M . Moreover σ must be nowhere-
vanishing since σ(x) = 0 would not define a basis.
⇐ : If there exists a nowhere-vanishing global section σ, then {σ} defines a frame over M since any non-zero
vector defines a basis of a 1-dimensional vector space : σ(x) 6= 0 ⇒ {σ(x)} is a basis for Ex , ∀ x ∈ M . Hence
we can see σ ∈ O∗ (M ) by considering σ(x) ∈ Ex ∼ = C. Theorem 9.6.2 then implies that the line bundle is trivial.
∼
We recall that the morphism of families of vector spaces is given by ϕ : M × C −→ L, ϕ(x, λ) := λ · σ(x).

Lo
∼
ϕ
M ×C

{{
π p1

Remark :
Note that this is only true for line bundles and does not hold for vector bundles of higher rank.

10.5.4 Definition
Recall that vector bundles over a complex manifold M corresponds to locally free sheaves over M , given by the
sheaf of holomorphic sections. Hence the line bundles over M correspond to locally free sheaves of rank 1, also
called invertible sheaves. Finally we have seen in section 9.3.6 that the trivial vector bundle M × Ck is uniquely
k
given by the free sheaf OM . Hence the trivial line bundle M × C corresponds to the sheaf OM of holomorphic
functions, in particular because sections of the trivial bundle satisfy p1 ◦ s = id ⇒ s(x) = (x, s̃(x)) ∈ M × C,
where s̃ : M → C can be any holomorphic function. Therefore sections of the trivial line bundle can be identified
with holomorphic functions. f ∈ OM ⇔ f : M → C is a holomorphic section of the trivial bundle M × C.
In the following we denote by L the sheaf of holomorphic sections of a line bundle (i.e. L = V, but we want to
point out that we are dealing with a line bundle).

94
Complex Manifolds Section 10.5 SCHLICHENMAIER, Leytem

10.5.5 Proposition
Let π : L → M be a line bundle over M and s ∈ L(M ) be a global holomorphic section such that s 6≡ 0, i.e. s is
a globally holomorphic map G
s : M −→ L = Lx : x 7−→ vx ∈ Lx
x∈M

Then the set (s) := { x ∈ M | s(x) = 0 } (where 0 ∈

/ C, but 0 ∈ Lx ) is an analytic subset of M .
Proof. Recall that analytic subsets of M are locally given by zero sets of finitely many holomorphic functions (i.e.
sections of the trivial line bundle). Note however that (s) is not an analytic set by definition since s ∈
/ OM (M ) ;
s is not a map M → C and (s) 6= V (s). But s locally corresponds to a holomorphic C-valued function. If U ⊆ M
is a trivializing open set for L, then L|U ∼
= U × C and
s|U ≡ s̃ where s|U : U → L|U ⇒ s(x) = (x, s̃(x)), s̃ ∈ OM (U ), ∀ x ∈ U

Now if x ∈ U is such that s(x) = 0, the local form is given by x 7→ (x, 0) ∈ U × Lx ∼

= U × C. U being a trivializing
open set, we know by theorem 9.6.2 that there is a frame {σ} over U . Since s(x) ∈ Lx for all x and {σ(x)} is a
basis for the 1-dimensional fiber, we obtain that

∀ x ∈ U, ∃ λx ∈ C such that s|U (x) = λx · σ(x)

Define f : U → C, f (x) = λx ⇒ s|U (x) = f (x) · σ(x), ∀ x ∈ U where f is holomorphic in U by a similar

argument as in example 9.3.3. Since σ is nowhere-vanishing on U , we have

s|U (x) = 0 ∈ Lx ⇔ f (x) = 0 ∈ C

  
⇒ (s) ∩ U = x ∈ U | s(x) = 0 = x ∈ U | s|U (x) = 0 = x ∈ U | f (x) = 0 = V (f )

with f ∈ OM (U ), hence the trivializing covering for L defines a covering of M such that (s) is given by the zero
set of a holomorphic function on each open set. It follows that (s) is an analytic subset of M .
We also have to show that this description of (s) is independent of the chosen frame, i.e. independent of the
trivialization of L. If {τ } is another frame over U , we can write

s|U (x) = f (x) · σ(x) = g(x) · τ (x), f, g ∈ OM (U ), ∀ x ∈ U

and x ∈ U is a zero of s ⇔ f (x) = 0 = g(x). Since {σ(x)} and {τ (x)} are bases of Lx , there is a basis change
matrix A = A(x) ∈ GL(1, C) such that τ (x) = A(x) · σ(x), ∀ x ∈ U . A(x) being invertible for all x ∈ U , meaning
that A(x) 6= 0, ∀ x ∈ U , we get A ∈ O∗ (U ) and it follows that

∀ x ∈ U : s|U (x) = f (x) · σ(x) = g(x) · τ (x) = g(x) · A(x) · σ(x) with A(x) 6= 0

Hence f (x) = 0 ⇔ g(x) = 0 and (s) ∩ U = V (f ) = V (g) : we get the same local zero set independently of the
choice of the frame.

Remark :
Note that there exist line bundles which do not admit non-zero global holomorphic sections.

10.5.6 Proposition
Let π : L → M be a line bundle over M and s ∈ L(M ) be a global holomorphic section such that s 6≡ 0. Then

either : 1) (s) = ∅ ⇒ L = M × C and L ∼

= OM
or : 2) (s) 6= ∅ ⇒ the irreducible components of (s) have codimension 1

Hence if (s) is non-empty and irreducible, then it is a connected complex submanifold of codimension 1 after
removing the singular points in (s) (see section 8.3.2).
Proof. 1) is just a consequence of proposition 10.5.3 and theorem 9.3.6
2) If (s) 6= ∅, we already showed that it is analytic, hence closed. If we supposed that it is irreducible, we know
by 8.3.2 and theorem 8.2.4 that (s) \ S((s)) is in addition a connected submanifold of M . Moreover (s) is locally
defined by the zero set of 1 holomorphic function f , hence its codimension is 1.

95
Complex Manifolds Section 10.5 SCHLICHENMAIER, Leytem

Intuitively this can also be interpreted as follows : in local coordinates z1 , . . . , zn where n = dim M , (s) is defined
by the condition f (z1 , . . . , zn ) = 0. So we have n coordinates with 1 constraint, which yields n − 1 degrees of
freedom.
In general (s) is neither irreducible nor connected ; it may have different irreducible components, but all compo-
nents have codimension 1 (after removing singularities) by the previous argument. And then one can define


dim (s) := maximum of the dimensions of the components = 1

10.5.7 Sections and divisors

If π : L → M is a line bundle over M and s ∈ L(M ) a global non-zero holomorphic section, we want to assign a
divisor D to the analytic set (s) : X
D= nY · Y
Y

where nY ∈ Z, Y are irreducible analytic subsets of M of codimension 1 and the sum is locally finite. In order
to consider (s) as a divisor, we set nY to be the vanishing order of s along Y . This is possible since L is a line
bundle, i.e. the section s can be represented by a frame {σ} and a holomorphic function f ∈ OM (U ) such that
s|U (x) = f (x) · σ(x), ∀ x ∈ U where U ⊆ M is a trivializing open set. Then we set

ordx (s) := ordx (f ) , ∀ x ∈ U

This is independent of the chosen frame since sections defining frames do not have zeros in the trivializing set :

s|U (x) = f (x) · σ(x) = g(x) · τ (x) ⇒ ordx (s) = ordx (f ) = ordx (g)

where {τ } is another frame over U and g ∈ OM (U ). This is why we consider line bundles for this construction.
In the end we have defined a map L(M ) → Div(M ) : s 7→ (s)d := Y ordY (s) · Y . Note that Y is in general not
P
given by a point ; this is only the case if dim M = 1.

96
Chapter 11

Tangent vectors and differentials

11.1 The real picture

Let M be a complex manifold of complex dimension n. For the moment we forget about the complex structure,
i.e. we consider M as a real differentiable manifold of real dimension 2n. For this, we have the coordinates

∀ k ∈ {1, . . . , n} : zk = xk + i yk ⇒ (x1 , y1 , . . . , xn , yn ) : real coordinates

11.1.1 Definitions
∞
We denote by CM the sheaf of R-valued real differentiable functions on the manifold M . It is a sheaf of rings.
A point derivation of C ∞ (M ) at m ∈ M is a linear map Dm : C ∞ (M ) → R which satisfies the Leibniz rule :

∀ f, g ∈ C ∞ (M ), ∀ λ ∈ R : Dm (λ · f + g) = λ · Dm (f ) + Dm (g)
Dm (f · g) = Dm (f ) · g(m) + f (m) · Dm (g)

A tangent vector at m is a point derivation at m, i.e. given m ∈ M , a tangent vector is a derivation rule ∂m for
∞
functions which are locally defined at m (in the germ of m) such that ∂m : CM,m → R is linear and
∞
∀ f, g ∈ CM,m : ∂m (f · g) = (∂m f ) · g(m) + f (m) · (∂m g)

The set of all tangent vectors at m is called the (real) tangent space at m and is denoted by Tm M . We know
that Tm M is a 2n-real dimensional vector space and a basis is given by
n ∂ ∂ ∂ ∂ o
, , ... , ,
∂x1 |m ∂y1 |m ∂xn |m ∂yn |m
(partial derivatives evaluated at m). For short, we denote them in the following by ∂x1 |m , ∂y1 |m , etc.

11.1.2 The tangent bundle

F
The (real) tangent bundle of the manifold M , denoted by TR M , is defined as TR M := Tm M .
m∈M
Since Tm M ∼ = R2n , the foot-map π : TR M → M : u = ∂m 7→ m defines a real vector bundle of rank 2n and its
−1
fibers are π (m) = Tm M , ∀ m ∈ M . Moreover TR M is a 4n-real dimensional manifold with chart maps

ψα : π −1 (Uα ) ⊃ Tm M −→
∼
ϕα (Uα ) × R2n ⊂ R4n
P P

u = ∂m = i vi ∂xi |m + j wj ∂yj |m 7−→ ϕα (m) , (v1 , w1 , . . . , vn , wn ) (11.1)

where (Uα , ϕα )α∈J is a 2n-real dimensional atlas of M . A local frame is


Σ = ∂x1 , ∂y1 , . . . , ∂xn , ∂yn

where e.g. ∂xi : Uα → TR M|Uα : m 7→ ∂xi |m . Note that partial derivatives only exist in the charts since,
strictly-speaking, we do not calculate ∂xi f , but ∂xi (f ◦ ϕ−1 ∞
α ) if f ∈ CM (Uα ).

97
Complex Manifolds Section 11.2 SCHLICHENMAIER, Leytem

p ◦ ∂xi = idUα , so ∂xi and ∂yi are sections of TR M . And it is indeed a frame since
{∂x1 (m), . . . , ∂yn (m)} = {∂x1 |m , . . . , ∂yn |m }
is a basis of Tm M for all m ∈ Uα . By theorem 9.6.2 this shows in addition that a trivializing covering for TR M
is indeed given by a coodinate covering of M : TR M|Uα ∼= Uα × R2n .

Finally we want to compute the cocycles defining the tangent bundle. For this, consider the trivialization
TR M|Uα ∼= Uα × R2n via φα : u 7→ m, (v1 , w1 , . . . , vn , wn )

where m is the unique m ∈ Uα such that u ∈ Tm M and (v1 , w1 , . . . , vn , wn ) ∈ R2n are the coefficients of the
decomposition of u in the basis of Tm M as in (11.1).
Let now Uα and Uβ be 2 chart domains with non-empty intersection and associated charts
ϕα , ϕβ and the
respective coordinates (x01 , y10 , . . . , x0n , yn0 ) and (x1 , y1 , . . . , xn , yn ). If m, (v1 , w1 , . . . , vn , wn ) ∈ Uβ × R2n , then
 n n

 X X 
φα φ−1
β m, (v 1 , w1 , . . . , v n , wn ) = φ α v i · ∂ xi |m + wj · ∂y j |m
i=1 j=1
n
X n
X 
vi0 · ∂x0i |m + wj0 · ∂yj0 |m = m, (v10 , w10 , . . . , vn0 , wn0 )


= φα
i=1 j=1

Thus gUα Uβ (m) : (v1 , w1 , . . . , vn , wn ) 7→ (v10 , w10 , . . . , vn0 , wn0 )

and the relation between these coefficients is given
by the basis change {∂xi |m , ∂yj |m } → {∂x0i |m , ∂yj0 |m } in the tangent space Tm M . But we have the formula :
n
X
νj = Jji · µi where Jji = ∂ui (v j )|m : Jacobian matrix
i=1

when changing the local coordinates {ui } → {v j } and the coefficients vary {µi } → {ν j }, i.e. as vectors ν = J · µ.
The same happens here by changing the coordinates {x1 , y1 , . . . , xn , yn } → {x01 , y10 , . . . , x0n , yn0 }, hence
 
t t 0 0 0 0 ∂ψ
gUα Uβ (m) : (v1 , w1 , . . . , vn , wn ) 7−→ (v1 , w1 , . . . , vn , wn ) = · t (v1 , w1 , . . . , vn , wn )
∂(x, y)

where ψ = ϕα ◦ ϕ−1 0 0
β is the transition function (x , y ) = ψ(x, y). The cocycle gUα Uβ is therefore given by the
2n × 2n–Jacobian matrix of the transition map ψ. The coordinate change ψ being bijective, the determinant of
the Jacobian matrix is in addition non-zero.

11.1.3 Vector fields

A local vector field on M is a local section of the tangent bundle TR M . We denote the sheaf of local sections of
TR M (i.e. the sheaf of local vector fields on M ) by TR . Hence X is a vector field on U ⇔ X ∈ TR (U ). As sheaf
of local sections, TR is thus a locally free sheaf of OM –modules of rank 2n.
In a coordinate neighborhood Uα , a local vector field X ∈ TR (Uα ) (a local section) hence writes uniquely as
n
X n
X
∞
X= fi · ∂xi + gj · ∂yj for some fi , gj ∈ CM (Uα ) as in (9.11)
i=1 j=1

A global vector field on M is a differentiable map X : M → TR M such that ∀ m ∈ M , X(m) ∈ Tm M and there
is an open neighborhood U ⊆ M around m such that X|U is a local vector field as defined above.

11.2 The complex picture

Now we introduce the complex structure into the tangent space and the tangent bundle. Recall that
∂ 1  ∂ ∂  ∂ 1  ∂ ∂ 
zk = xk + i yk ⇒ = · −i· , = · +i· , ∀ k ∈ {1, . . . , n} (11.2)
∂zk 2 ∂xk ∂yk ∂ z̄k 2 ∂xk ∂yk
This transformation is biholomorphic and hence defines a holomorphic change of coordinates on M .

98
Complex Manifolds Section 11.2 SCHLICHENMAIER, Leytem

11.2.1 The complex tangent bundle

Let n = dimC M . We define the complex tangent space at m ∈ M as TC,m M := Tm M ⊗ C. Intuitively this just
means that we now also allow complex-valued functions. It is indeed the complexified version of the real tangent
space. The dimension of this space now strongly depends on whether we consider it as a real or a complex vector
space. First of all : dimR (Tm M ) = dimC (Tm M ) = 2n.
Proof. We know that dimR (Tm M ) = 2n and a basis over R is given by {∂x1 |m , ∂y1 |m , . . . , ∂xn |m , ∂yn |m }. Hence
Tm M only contains ”real values” since any u ∈ Tm M writes as a real linear combination of these basis vectors.
Now we want to find a basis of Tm M over C. And it is not given by {∂z1 |m , . . . , ∂zn |m } since even complex linear
combinations of the ∂zi |m do not generate all the real partial derivatives ∂xi |m and ∂yj |m . In order to satisfy
(11.2), we need to add at least all the ∂z̄j |m . This is also sufficient, hence a basis of Tm M over C is given by
{∂z1 |m , . . . , ∂zn |m , ∂z̄1 |m , . . . , ∂z̄n |m } ⇒ dimC (Tm M ) = 2n as well.
Note that this is not the only possibility : {∂x1 |m , ∂y1 |m , . . . , ∂xn |m , ∂yn |m } of course also defines a C-basis of
Tm M (since it is already a basis over R). The converse however is not true ! Inverting relation (11.2) yields

∂ ∂ ∂ ∂  ∂ ∂ 
∀ k ∈ {1, . . . , n} : = + , =i· − (11.3)
∂xk ∂zk ∂ z̄k ∂yk ∂zk ∂ z̄k
Since only real linear combinations are allowed it is not possible to generate all the ∂xi |m and ∂yj |m only with
the set {∂z1 |m , . . . , ∂zn |m , ∂z̄1 |m , . . . , ∂z̄n |m } which is therefore not an R-basis of Tm M .

The complex tangent space is obtained by tensorizing the real tangent space with C, hence



dimC TC,m M = dimC Tm M ⊗ C = dimC (Tm M ) · dimC (C) = 2n · 1 = 2n



dimR TC,m M = dimR Tm M ⊗ C = dimR (Tm M ) · dimR (C) = 2n · 2 = 4n

The occurrence of 4 can be interpreted as ”dividing” a complex C-valued function into real and imaginary part.
By the tensor property we moreover conclude that bases of TC,m M are

over C : ∂z1 |m , . . . , ∂zn |m , ∂z̄1 |m , . . . , ∂z̄n |m

over R : ∂x1 |m , ∂y1 |m , . . . , ∂xn |m , ∂yn |m , i · ∂x1 |m , i · ∂y1 |m , . . . , i · ∂xn |m , i · ∂yn |m

F
Now the complex tangent bundle is defined as TC M := TC,m M .
m∈M
As in the real case, we obtain that π : TC M → M is a complex vector bundle over M of complex rank 2n and
real rank 4n. But : it is NOT a holomorphic vector bundle ! This happens because the second part of the basis

∂z1 |m , . . . , ∂zn |m , ∂z̄1 |m , . . . , ∂z̄n |m

of TC,m M is not holomorphic (z̄k is involved).

More precisely, if U is a chart domain with local coordinates (z1 , . . . , zn ), then a frame for TC M|U is

∂z1 , . . . , ∂zn , ∂z̄1 , . . . , ∂z̄n ∼

 
= ∂x1 , ∂y1 , . . . , ∂xn , ∂yn

One can however define the holomorphic tangent bundle Thol M , which is of rank n and where a local frame is
given by {∂z1 , . . . , ∂zn }.

11.2.2 Complex vector fields

A local complex vector field X is a local section of the bundle TC M and is hence locally of the form
n
X n
X
X= γi · ∂zi + εj · ∂z̄j (11.4)
i=1 j=1

for some C-valued real differentiable functions γi and εj . But they are not necessarily holomorphic since TC M
is not a holomorphic vector bundle, hence local sections do not need to be holomorphic neither.

99
Complex Manifolds Section 11.3 SCHLICHENMAIER, Leytem

We denote the sheaf of local complex vector fields by TC . Such a vector field on an open set U ⊆ M is of
• the type (1, 0) if εj = 0, ∀ j ∈ {1, . . . , n} (only derivatives in the holomorphic directions).
• the type (0, 1) if γi = 0, ∀ i ∈ {1, . . . , n} (only derivatives in the anti-holomorphic directions).
• holomorphic type if it is of the type (1, 0) and the γi are holomorphic in U , i.e. ∂∂γz̄ki = 0, ∀ k, i ∈ {1, . . . , n}.

A global complex vector field on M is a differentiable map X : M → TC M such that ∀ m ∈ M , X(m) ∈ TC,m M
and there is an open neighborhood U ⊆ M around m such that X|U is a local complex vector field as in (11.4).
And a holomorphic vector field is a global complex vector field that it locally of holomorphic type.

As usual, we want that vector fields do not depend on the chosen coordinates. But this is satisfied here be-
cause M is a complex manifold, i.e. all coordinate changes w = ψ(z) are holomorphic :
∂w ∂w
! ∂w
!
0
 
∂w ∂z ∂ z̄ ∂z
= 0 ⇒ J(ψ) = ∂ w̄ ∂ w̄ = ∂ w̄
: 2n × 2n–matrix
∂ z̄ 0
∂z ∂ z̄ ∂ z̄

Hence when changing coordinates, permutations can only happen within the individual parts of X in (11.4). It
follows that the type of a vector field is independent under holomorphic coordinate transformations.
Note that this does not hold for real coordinate changes.

11.3 Differential forms

11.3.1 The cotangent bundle
The cotangent bundle is the dual bundle of the tangent bundle. Here again we can define the real, the complex
and the holomorphic cotangent bundle. We quickly go through some constructions :

*
Consider the dual space Tm M , which is again of real and complex dimension 2n. A basis (over R and C)
is given by {dx1|m , dy1|m , . . . , dxn|m , dyn|m }, whereas {dz1|m , . . . , dzn|m , dz̄1|m , . . . , , dz̄n|m } is a C-basis only. We
recall that dxi|m is the dual element associated to ∂xi |m , i.e.





dxi|m ∂xj |m = dyi|m ∂yj |m = dzi|m ∂zj |m = dz̄i|m ∂z̄j |m = δij
*
Similarly the dual space TC,m M is of complex dimension 2n and admits the same C-basis. Then we set

∗ G
TC* M := TC M = *
TC,m M
m∈M

and it follows that a local frame over a chart domain Uα ⊆ M is given by {dz1 , . . . , dzn , dz̄1 , . . . , , dz̄n }, where
dzi : Uα −→ TC* M|Uα : m 7−→ dzi|m
*
This is again not a holomorphic vector bundle, but one can define the holomorphic cotangent bundle Thol M,
which is of rank n and where a local frame is given by {dz1 , . . . , dzn }.


Finally one can also consider the exterior powers Λr TC* M for r ∈ {0, 1, . . . , 2n} which have the local frame

dzi1 ∧ . . . ∧ dzis ∧ dz̄j1 ∧ . . . ∧ dz̄jt s+t=r

11.3.2 Definitions
We denote by E the sheaf of C-valued functions on M which are differentiable with respect to the real structure.
Let V ⊆ M be open. A 1-differential on V is a differentiable map ω : V → TC* M such that ∀ m ∈ V ,
*
ω(m) ∈ TC,m M and there is an open neighborhood U ⊆ V of m such that ω|U defines a local section of the
cotangent bundle over U . Hence a 1-differential is locally of the form
n
X n
X n
X n
X
ω|U = ai · dxi + bj · dyj = εi · dzi + δj · dz̄j
i=1 j=1 i=1 j=1

where ai , bj , εi , δj ∈ E(U ) and by using the decomposition zk = xk + i yk ⇒ dzk = dxk + i dyk , ∀ k ∈ {1, . . . , n}.

100
Complex Manifolds Section 11.3 SCHLICHENMAIER, Leytem

The sheaf of 1-differentials is denoted by E 1 . Moreover we set :

• ω ∈ E (1,0) (V ) ⇔ δj = 0, ∀ j ∈ {1, . . . , n}, i.e. ω|U = i εi · dzi for εi ∈ E(U )
P

• ω ∈ E (0,1) (V ) ⇔ εi = 0, ∀ i ∈ {1, . . . , n}, i.e. ω|U = j δj · dz̄j for δj ∈ E(U )

P

The type of these 1-differentials is respected by holomorphic coordinate changes, but not by real ones. We have

E 1 = E (1,0) ⊕ E (0,1)

i.e. any element ω ∈ E 1 (V ) can uniquely be decomposed as ω = ω1 + ω2 with ω1 ∈ E (1,0) (V ), ω2 ∈ E (0,1) (V ) and
this decomposition is independent of the chosen coordinates.
Note that E 1 , E (1,0) and E (0,1) are locally free sheaves of E–modules of rank 2n, n, n respectively and hence
correspond to the sheaf of sections of a certain real vector bundle (for complex manifolds, we need OM –modules).

The sheaf of holomorphic 1-differentials is denoted by Ω1M and it consists of 1-differentials which are locally
over some open set U ⊆ M of the form n
X
ω|U = γi · dzi with γi ∈ OM (U )
i=1

i.e. the coefficient functions γi have to be holomorphic in U (not only differentiable).

Ω1M is not a sheaf of E–modules since f · γ is not necessarily holomorphic if f ∈ E(U ) and γ ∈ OM (U ). But it is a
locally free sheaf of OM –modules of rank n and hence corresponds to the sheaf of sections of some holomorphic
vector bundle of rank n. And this is not the cotangent bundle TC* M , but the holomorphic bundle Thol *
M.

11.3.3 Concept of a differential : real picture

Let M be a complex manifold of complex dimension n with complex coordinates (z1 , . . . , zn ) and real coordinates
{x1 , y1 , . . . , xn , yn }. Let also V ⊆ M be open and f ∈ E(V ) be a real differentiable function on V . To f we want
to associate a 1-differential, denoted by df and called the differential of f .
Let {Uα }α∈J be a coordinate covering of M . Then df is defined by the local condition
n n
X ∂f X ∂f
∀ α ∈ J : df|Uα ∩V = dxi + dyj
i=1
∂xi j=1
∂yj

where {∂x1 , ∂y1 , . . . , ∂xn , ∂yn } and {dx1 , dy1 , . . . , dxn , dyn } are frames over Uα of the tangent and cotangent bun-
dle respectively. Thus df is not a function any more : df ∈ / E(V ), but df ∈ E 1 (V ).

Properties :
1) df is well-defined, i.e. it is independent of the chosen local coordinates.
2) The differential satisfies the Leibniz rule : ∀ f, g ∈ E(V ), d(f · g) = df · g + f · dg.
Proof. 2) is clear since partial derivatives are linear and satisfy the Leibniz rule
1) For simplicity, assume that we are given a real differentiable manifold of real dimension m with a coordinate
covering {Uα }α∈J . Choose Uα and Uβ such that Uα ∩ Uβ 6= ∅ with local coordinates x = (x1 , . . . , xm ) and
y = (y1 , . . . , ym ) respectively, related by the differentiable coordinate change y = ψ(x) with invertible Jacobian
matrix J(ψ) = (Jij ). Let also f ∈ E(V ) for some open set V . We have to show that locally
m m
X ∂f X ∂f
dxi = dyj
i=1
∂xi j=1
∂yj

Given the respective covariant and contravariant transformation laws, we obtain :

∂ X ∂ X ∂yj ∂ X X ∂xi
= Jji · = · , dxi = (J −1 )ik dyk = dyk
∂xi j
∂y j j
∂xi ∂y j ∂yk
k k
X ∂f X ∂yj ∂f ∂xi X  ∂y j ∂xi  ∂f
⇒ dxi = · · dyk = · · dyk
i
∂xi ∂xi ∂yj ∂yk ∂xi ∂yk ∂yj
i,j,k k,j,i
XP  ∂f X ∂f X ∂f
= Jji · (J −1 )ik · dyk = δjk · dyk = dyj
i ∂yj ∂yj j
∂yj
k,j k,j

101
Complex Manifolds Section 11.3 SCHLICHENMAIER, Leytem

11.3.4 The complex picture

Under the same assumptions as in 11.3.3, let f ∈ E(V ) be a C-valued differentiable function on V . We set locally
n n
X ∂f X ∂f ¯
df := dzi + dz̄i =: ∂f + ∂f
i=1
∂zi i=1
∂ z̄i

Hence d = ∂ + ∂¯ where ∂ consists of the derivatives in the holomorphic direction and ∂¯ contains all derivatives in
the anti-holomorphic direction. Note that ∂ and ∂¯ are well-defined (i.e. independent of the chosen coordinates)
because we already showed in 11.2.2 that holomorphic coordinate changes do not mix the holomorphic and
anti-holomorphic parts. For V ⊆ M open we thus constructed the maps

d : E(V ) −→ E 1 (V ) , ∂ : E(V ) −→ E (1,0) (V ) , ∂¯ : E(V ) −→ E (0,1) (V )

¯ = 0.
We see that f ∈ E(V ) is holomorphic in V ⇔ ∂f
¯ ¯
This is the case because ∂f = 0 on V ⇒ ∂f|U = 0 for any open set U ⊆ V , hence if U is small enough :
X ∂f X ∂f X ∂f
¯ |U =
0 = ∂f dz̄i ⇒ (m) · dz̄i (m) = (m) · dz̄i|m = 0, ∀ m ∈ U
i
∂ z̄i i
∂ z̄i i
∂ z̄i
∂f
⇒ (m) = 0, ∀ m ∈ U, ∀ i ∈ {1, . . . , n} since dz̄1|m , . . . , dz̄n|m are linearly independent
∂ z̄i
∂f
⇒ = 0 in U, ∀ i ∈ {1, . . . , n} ⇒ f is holomorphic in U
∂ z̄i
which means that f is holomorphic in a small neighborhood of any point in V , i.e. f is holomorphic in V .
Hence OM = ker ∂¯ and this equality even holds in the sheaf-theoretical sense : OM (V ) = ker ∂(V
¯ ) for any open
set V ⊆ M . Moreover ker d is given by the sheaf of locally constant functions because (without details) :
X ∂f X ∂f basis ∂f ∂f
0 = df = dzi + dz̄i ⇒ = = 0, ∀ i ∈ {1, . . . , n}
i
∂zi i
∂ z̄i ∂zi ∂ z̄i

In particular : ker d(V ) = C if V is a connected open set.

Finally one can also consider d, ∂ and ∂¯ as morphisms of sheaves, i.e. we have the sequences of sheaves
d ∂ ∂¯
E −→ E 1 , E −→ E (1,0) , E −→ E (0,1)

The idea is now to continue these sequences to higher orders. In particular we are interested in the questions :
− Does any 1-differential come from a function, i.e. is d surjective ?
− How to integrate a differential form ?
d d ∂ ∂
E −→ E 1 −→ . . . , E −→ E (1,0) −→ . . .

11.3.5 Differentials of higher order

We recall that if W is an n-dimensional vector space, we can define its dual space W ∗ , consisting of all linear
forms on W , and then the exterior power Λp (W ∗ ), consisting of all p-linear alternating maps on W × . . . × W .
In the case of a complex manifold M we hence obtain the tangent bundle TC M , the cotangent bundle TC* M and
the exterior power bundle Λp (TC* M ), all of them not being a holomorphic vector bundle. If U is a coordinate
neighborhood with local coordinates (z1 , . . . , zn ) = (x1 , y1 , . . . , xn , yn ), a frame over U of these bundles is

TC M : ∂x1 , ∂y1 , . . . , ∂xn , ∂yn ∼

 
= ∂z1 , . . . , ∂zn , ∂z̄1 , . . . , ∂z̄n
TC* M : dx1 , dy1 , . . . , dxn , dyn ∼
 
= dz1 , . . . , dzn , dz̄1 , . . . , dz̄n
Λp (TC* M ) : dxi1 ∧ . . . ∧ dxis ∧ dyj1 ∧ . . . ∧ dyjt ∼
 
= dzi1 ∧ . . . ∧ dzis ∧ dz̄j1 ∧ . . . ∧ dz̄jt

for s + t = p, 1 ≤ i1 < . . . < is ≤ n, 1 ≤ j1 < . . . < jt ≤ n.

We already know that local vector fields and 1-differentials are local sections of the tangent bundle and the
cotangent bundle respectively, the associated sheaves being denoted by TC and E 1 .

102
Complex Manifolds Section 11.3 SCHLICHENMAIER, Leytem

Let V ⊆ M be open. A p-differential (or differential p-form) on V is a differentiable map ω : M → Λp (TC* M )

such that ∀ m ∈ V , ω(m) ∈ Λp (TC,m*
M ) and there is an open neighborhood U ⊆ V of m such that ω|U is a local
p *
section of Λ (TC M ) over U . As a local section it is thus locally given of the form
X X
ω|U = αI dxi1 ∧ . . . ∧ dxis ∧ dyj1 ∧ . . . ∧ dyjt = βI dzi1 ∧ . . . ∧ dzis ∧ dz̄j1 ∧ . . . ∧ dz̄jt (11.5)
I I

where αI , βI ∈ E(U ) and we denoted for short


I = (i1 , . . . , is , j1 , . . . , jt ) | s + t = p, 1 ≤ i1 < . . . < is ≤ n, 1 ≤ j1 < . . . < jt ≤ n

Since the individual parts dxi , dyj , dzi and dz̄j are 1-differentials and the wedge product is anti-commutative,
we obtain that
dxi ∧ dxi = dyj ∧ dyj = dzi ∧ dzi = dz̄j ∧ dz̄j = 0, ∀ i, j ∈ {1, . . . , n}
dxi ∧ dxj = −dxj ∧ dxi , dyi ∧ dyj = −dyj ∧ dyi , dzi ∧ dzj = −dzj ∧ dzi , dz̄i ∧ dz̄j = −dz̄j ∧ dz̄i

Moreover the descriptions with real coordinates xi , yj and complex coordinates zi , z̄j are equivalent (but only
over C !), so both notations can be used without risk of confusion. The relation between both of them is :

dz ∧ dz̄ = (dx + i dy) ∧ (dx − i dy) = dx ∧ dx − i dx ∧ dy + i dy ∧ dx + dy ∧ dy = −2i dx ∧ dy

We can even see the change (x1 , y1 , . . . , xn , yn ) ↔ (z1 , . . . , zn , z̄1 , . . . , z̄n ) as a usual coordinate change on M since
they are related by (11.2) and (11.3), which define a differentiable (but not a holomorphic) change of coordinates.

The sheaf of p-differentials (local sections) is denoted by E p . Moreover if such a form has s holomorphic differ-
entials dzi and t anti-holomorphic differentials dz̄j as in the second part of (11.5), then we write E s,t . Hence
M
Ep = E s,t
s+t=p

E p is given by all the possibilities to decompose p ∈ {0, . . . , 2n} into s holomorphic and t anti-holomorphic parts.
As an example consider E 2 = E 2,0 ⊕ E 1,1 ⊕ E 0,2 , so all possible complex types of 2-differentials locally write as
X   X n 
n X  X  
αi1 i2 dzi1 ∧ dzi2 + βij dzi ∧ dz̄j + γi1 i2 dz̄i1 ∧ dz̄i2
1≤i1 <i2 ≤n i=1 j=1 1≤i1 <i2 ≤n

for some locally differentiable functions αi1 i2 , βij , γi1 i2 .

11.3.6 The exterior derivative

The goal is now to extend the definition
P of thePdifferential of a function to all differential forms. Let ω ∈ E p (V ).
Then locally over U , ω looks like I αI dxI = I βI dzI for αI , βI ∈ E(U ). Here we used the short-hand notation

dxI := dxi1 ∧ . . . ∧ dxis ∧ dyj1 ∧ . . . ∧ dyjt , dzI := dzi1 ∧ . . . ∧ dzis ∧ dz̄j1 ∧ . . . ∧ dz̄jt

Then we define dω ∈ E p+1 (V ) by the local condition (over some small open set U ⊆ V )
n X n X n X n X
X ∂αI X ∂αI X ∂βI X ∂βI
dω|U := dxi ∧ dxI + dyi ∧ dxI = dzi ∧ dzI + dz̄i ∧ dzI (11.6)
i=1
∂xi i=1
∂yi i=1
∂zi i=1
∂ z̄i
I I I I

One can show that this is always well-defined (not only for holomorphic changes) and independent of the chosen
local coordinates, in particular the 2 above definitions coincide. Hence a global definition for dω is also given.
The map d : E p (V ) → E p+1 (V ) : ω 7→ dω is called the exterior derivative and is a generalization of df .

Properties :
1) The map d is linear and satisfies d ◦ d = 0.
2) ∀ ω ∈ E p (V ), η ∈ E q (V ) : d(ω ∧ η) = dω ∧ η + (−1)p · ω ∧ dη.

103
Complex Manifolds Section 11.4 SCHLICHENMAIER, Leytem

Proof. 2) is admitted
1) We only show how the computation is done in the sum involving dxi ; all other terms are similar. Locally :
 X X ∂α  X X ∂2α
I I
d(dω) = d dxi ∧ dxI = dxj ∧ dxi ∧ dxI
i
∂x i i,j
∂xi ∂x j
I I
X  X ∂ 2 αI 
= dxj ∧ dxi ∧ dxI since dxi ∧ dxi = 0
∂xi ∂xj
I i6=j
X  X ∂ 2 αI X ∂ 2 αI 
= dxj ∧ dxi ∧ dxI + dxj ∧ dxi ∧ dxI
i<j
∂xi ∂xj i>j
∂xi ∂xj
I
X X  ∂ 2 αI ∂ 2 αI 
= dxj ∧ dxi ∧ dxI + dxi ∧ dxj ∧ dxI
i<j
∂xi ∂xj ∂xj ∂xi
I
X X  ∂ 2 αI ∂ 2 αI 
= dxj ∧ dxi ∧ dxI − dxj ∧ dxi ∧ dxI = 0
i<j
∂xi ∂xj ∂xi ∂xj
I

where the last equality follows from Schwartz’ Theorem (partial derivatives commute) and anti-commutativity
of the wedge product.

11.4 De Rham and Dolbeault cohomology

11.4.1 Definitions
Let M be a complex manifold of complex dimension n and V ⊆ M be open. The vector space of all differential
forms on V is given by
M2n
E(V ) := E p (V )
p=0

hence any ω ∈ E(V ) writes uniquely as ω = ω0 + ω1 + . . . + ω2n with ωp ∈ E p (V ), ∀ p ∈ {0, . . . , 2n}. Note that
E p (V ) = {0} for p > 2n because a linear alternating form with more than 2n arguments in always zero.
By linearity

we can thus see the exterior derivative as a map d : E(V ) → E(V ) with the additional condition
d E p (V ) ⊂ E p+1 (V ), ∀ p ∈ {0, . . . , 2n}. So we can consider the sequence of sheaves
d d d d d
0 −→ E −→ E 1 −→ E 2 −→ . . . −→ E 2n −→ 0 (11.7)

which means that for any open set V ⊆ M (not necessarily a coordinate neighborhood), we have the sequence of
finite-dimensional vector spaces
d d d d d
{0} −→ E(V ) −→ E 1 (V ) −→ E 2 (V ) −→ . . . −→ E 2n (V ) −→ {0} (11.8)

Since d ◦ d = 0, we hence know that (11.7) is a complex of sheaves.

Let ω ∈ E(V ) be a differential form. We say that ω is closed if ω ∈ ker d, i.e. dω = 0. ω is called exact if
ω ∈ im d, i.e. there is a form η ∈ E(V ) such that ω = dη. We also define the two vector spaces :
p p
(V ) := ker d ∩ E p (V ) (V ) := im d ∩ E p (V ) = d E p−1 (V )


ZdR , BdR
p p
Hence ZdR (V ) contains all closed p-forms, called cocycles or cochains, and BdR (V ) contains all exact p-forms,
th
called coboundaries. The p de Rham cohomology of V is then given by
p p p
HdR (V ) := ZdR (V ) / BdR (V )

This now allows us to compute the cohomology of the sequence (11.8), which encodes the topology of M on V .
p
(11.8) is called an exact sequence if HdR p
(V ) = {0} for all p, i.e. if ZdR (V ) ∼ p
= BdR (V ), ∀ p ∈ {0, . . . , 2n}. Hence
the cohomology ”measures” the exactness of the sequence.

104
Complex Manifolds Section 11.4 SCHLICHENMAIER, Leytem

11.4.2 Results
Since d ◦ d = 0, we obtain that any exact form is closed. The converse however is not always true and depends
on the considered open set V . Actually :
p
the sequence (11.8) is exact ⇔ ZdR (V ) ∼ p
= BdR (V ), ∀ p ⇔ ∀ p : closed p-forms are exact on V

In other words, the sequence is exact if and only if we can integrate the differential forms on V . And we have
the following important result :

Lemma of Poincaré :
p
If V ⊆ M is a contractible open set, then HdR (V ) = {0}, ∀ p ≥ 1.
An open set V ⊆ M is called contractible if there is an x0 ∈ V and a differentiable maps φ : [0, 1] × V → V such
that φ(0, x) = x0 and φ(1, x) = x, ∀ x ∈ V .
Examples of contractible sets are e.g. Rn , Cn open balls and star-shaped sets.

Remark :
Since the kernel of d : E(V ) → E 1 (V ) is given by the locally constant functions on V , one usually extends the
sequence (11.7) to the augmented complex
i d d d d d
0 −→ CM −→ E −→ E 1 −→ E 2 −→ . . . −→ E 2n −→ 0

where i is the inclusion map and CM denotes the sheaf of locally constant functions on M .

11.4.3 Holomorphic p-forms

Recall that Ω1M is the sheaf of holomorphic 1-differentials which are locally of the form ω = i fi · dzi where the
P
coefficient-functions fi are holomorphic whenever defined.
In a similar way one can also define holomorphic differentials of higher order. Let V ⊆ M be open. A holomorphic
p-form on V is a differential p-form that is locally over U of the form
X
ω|U = fI dzi1 ∧ dzi2 ∧ . . . ∧ dzip
I

for fI ∈ OM (U ) and 1 ≤ i1 < . . . < ip ≤ n. The vector space of holomorphic p-forms on V is denoted by Ωp (V ),
with Ω0 (V ) = OM (V ). Then we set
Mn
Ω(V ) := Ωp (V )
p=0
p p
Since Ω (V ) ⊂ E (V ) for all p, the

exterior derivative d also applies to such forms, but the result may no longer
be a holomorphic form : d Ω(V ) * Ω(V ). ThusLwe have to consider the decomposition d = ∂ + ∂¯ and extend
the operators ∂ and ∂¯ as well. Let ω ∈ E p (V ) = E s,t (V ). We denote again

dzI := dzi1 ∧ dzi2 ∧ . . . ∧ dzis ∧ dz̄j1 ∧ dz̄j2 ∧ . . . ∧ dz̄jt

P
Hence if ω locally writes as ω = I αI dzI , we define ∂ω and ∂ω ¯ locally as

n X n X
X ∂αI ¯ =
X ∂αI
∂ω = dzi ∧ dzI , ∂ω dz̄j ∧ dzI
i=1
∂zi j=1
∂ z̄j
I I

i.e. it is just the decomposition of definition (11.6). Similarly as for d, one also shows that ∂ ◦ ∂ = ∂¯ ◦ ∂¯ = 0.
This implies a certain number of properties :

∂ : Ωp (V ) −→ Ωp+1 (V ) ∂ : E s,t (V ) −→ E s+1,t (V )

,
∂¯ Ω(V ) = {0} ∂¯ : E s,t (V ) −→ E s,t+1 (V )


,


In particular we can now see the map as ∂ : Ω(V ) → Ω(V ) with ∂ Ωp (V ) ⊂ Ωp+1 (V ).

105
Complex Manifolds Section 11.4 SCHLICHENMAIER, Leytem

We thus again have the (augmented) complex of sheaves

i ∂ ∂ ∂ ∂ ∂
0 −→ CM −→ OM −→ Ω1 −→ Ω2 −→ . . . −→ Ωn −→ 0

If for V ⊆ M open we denote dp := ∂|Ωp (V ) , we can also define the holomorphic de Rham cohomology by

H p (V, ΩpM ) := ker dp / im dp−1

11.4.4 Dolbeault cohomology

¯ is a (s, t + 1)-form.
We already know that if ω ∈ E s,t (V ) is a (s, t)-form, then ∂ω is a (s + 1, t)-form and ∂ω
Now fix p ∈ {0, . . . , n} and consider the (augmented) complex of sheaves

i ∂¯ ∂¯ ∂¯ ∂¯ ∂¯
0 −→ Ωp −→ E p,0 −→ E p,1 −→ E p,2 −→ . . . −→ E p,n −→ 0

where the kernel of ∂¯ : E p,0 (V ) → E p,1 (V ) is equal to Ωp (V ) since (p, 0)-forms locally write as
n X
X
¯ =
X ∂αI
ω= αI dzi1 ∧ dzi2 ∧ . . . ∧ dzip ⇒ ∂ω dz̄j ∧ dzi1 ∧ dzi2 ∧ . . . ∧ dzip
j=1
∂ z̄j
I I

¯ = 0 means that the derivatives of the coefficient-functions αI vanish with respect to all z̄j , i.e. the αI
and ∂ω
are holomorphic and ω ∈ Ωp (V ). The q th Dolbeault cohomology is then defined by

ker ∂¯ : E p,q (V ) → E p,q+1 (V )

p,q
H∂¯ (V ) :=
im ∂¯ : E p,q−1 (V ) → E p,q (V )

Result :
If ∆ ⊆ Cn is a polydisc, then H∂p,q
¯ (∆) = {0}, ∀ q ≥ 1.

106
Bibliography

[1] Martin Schlichenmaier, An Introduction to Riemann Surfaces, Algebraic Curves and Module Spaces.
Springer, 2007.
[2] Martin Schlichenmaier, lecture course Complex manifolds, academic year 2011.
[3] Aleksey Zinger, Notes on Vector Bundles, 2010.
webpage : http://www.math.sunysb.edu/˜azinger/mat531-spr10/vectorbundles.pdf

[4] Wikipedia

107