NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Version of December 3, 2014

Introduction

These are notes on complex geometry intended to accompany Math 545 at SBU in Fall,
2014. As explained in the course syllabus, the overall goal of the course is to present some
of the basic material on complex manifolds that are needed to work in complex or algebraic
geometry. We will try to emphasize concrete examples and calculations wherever possible.

The present notes generally speaking closely follow the presentations in the texts of
Griffiths-Harris [2] and Huybrechts [3] – and in later parts the notes of Schnell [4] – which
the reader can consult for more details. In particular, we borrow freely from these sources
without explicit attribution.

1. Preliminaries – Local Theory

This section presents some preliminary facts concerning holomorphic functions of several
complex variables, following closely the presentations of Griffiths-Harris [2] and especially
Huybrechts [3].

Holomorphic Functions of Several Variables. We will be concerned with complex-

valued functions defined on some open subset of Cn . In the absence of any distinguished
directions in Cn , we generally write
z = (z1 , . . . , zn )
for the standard linear coordinate functions, with real and imaginary parts xi and yi :
√
zi = xi + −1 · yi .
However sometimes it is useful to think of Cn as a product Cn = Ck × C` , with coordinates
(z, w) = (z1 , . . . , zk , w1 , . . . , w` ).
As we will be concerned with functions that admit local expressions as Taylor series, it is
convenient to use multi-indices J = (j1 , . . . , jn ) in order to introduce the shorthand
z J =def z1j1 · . . . · znjn .
1
2 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

ji , so that for example the polynomial |J|=m z J consists of the sum of

P P
We put |J| =
all momomials of degree m in z1 , . . . , zn . Finally, given ε = (ε1 , . . . , εn ) with εi > 0, and
c = (c1 , . . . , cn ) ∈ Cn , we denote by
∆(ε; c) =def z ∈ Cn | |zi − ci | < εi


the polydisk of (mult-)radius ε centered at c. Write T(ε; c) for the torus {z ∈ Cn | |zi − c| =
εi }.

As in the case of functions of one variable, there are various equivalent ways to charac-
terize holomorphic functions of several variables. For an open subset U ⊆ Cn , write C ∞ (U )
for the collection of all C ∞ complex-valued functions on U .
Theorem 1.1. Let U ⊂ Cn be an open set, and let f ∈ C ∞ (U ) be a smooth C-valued
function on U . Then the following three conditions are equivalent:

(i). f satisfies the Cauchy-Riemann equations on U , i.e. writing

√
f (z) = f (x, y) = u(x, y) + −1 · v(x, y),
one has:
∂u ∂v ∂u ∂v
− =0 , + =0 for i = 1, . . . , n.
∂xi ∂yi ∂yi ∂xi

(ii). f is holomorphic separately in each of the variables zi , i.e. for each c = (c1 , . . . , cn ) ∈
U , the function
f (c1 , . . . , ci−1 , zi , ci+1 , . . . , cn )
is analytic in zi in a neighborhood of ci .
(iii). Given any point c = (c1 , . . . , cn ) in U , there is an open neighborhood V = V (c) of c
in U in which f (z) can be represented by a convergent power series:
∞
X
f (z) = aJ · (z − c)J .
|J|=0

Definition 1.2. (Holomorphic functions). One says that f is holomorphic on U if the

conditions of Theorem 1.1 are satisfied. We write O(U ) for the ring of all such functions.

Sketch of Proof of Theorem 1.1. The equivalences (i) ⇐⇒ (ii) and (iii) =⇒ (i) follow from
the one-variable theory. For the implication (ii) =⇒ (iii), one chooses 0 < ε  1 and uses
the Cauchy integral formula in each variable separately to write
Z Z
1 f (ξ1 , . . . , ξn )
f (z1 , . . . , zn ) = √ · ... · dξ1 · · · dξn
|ξn −cn |=ε (ξ1 − z1 ) · . . . · (ξn − zn )
(2π −1)n
|ξ1 −c1 |=ε
Z
1 f (ξ1 , . . . , ξn )
= √ · · dξ1 · · · dξn .
(2π −1)n T(ε;c) (ξ1 − z1 ) · . . . · (ξn − zn )
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 3

Now use the identity

∞
1 1 X (zi − ci )ji
= ·
(ξi − zi ) (ξi − ci ) j =0 (ξi − ci )ji
i

to replace the integrand by a power series, and then integrate term by term. 

By reduction to the one-variable case, it is elemenatry that the maximum modulus

principle holds in higher dimensions, as well as the fact that a holomorphic function is
determined by its values on any small open set (Exercise 1.1). However unlike the situation
in one dimension, when n ≥ 2 domains of holomorphicity are quite special. For example,
one has a famous result of Hartogs:
Theorem 1.3. (Hartogs’ Theorem) Let ∆(ε) = ∆(ε; 0) be a polydisk centered at the
origin of Cn , and let
∆(ε0 ) ( ∆(ε)
be a smaller polydisk, i.e. a polydisk centered at 0 with ε0i < εi for every i. Assume that f is
a holomorphic function on
U = ∆(ε) − ∆(ε0 ).
If n ≥ 2, then f extends to a holomorphic function on all of ∆(ε).

In particular, if f is holomorphic away from a point in a domain of dimension ≥ 2, then it

extends holomporphically to the whole domain. Of course when n = 1 this as well as the
statement of Hartogs’ Theorem fails.

Sketch of Proof of Hartogs’ Theorem. We assume for simplicity of exposition that

ε = (1, . . . , 1) , ε0 = ( 31 , . . . , 13 ).
Note that then U contains the union of the two “shells”
1 1
2
< |z1 | < 1 , 2
< |z2 | < 1 .
Now for fixed (w2 , . . . , wn ) with |wi | < 1, the function
φw (z1 ) = f (z1 , w)
1
is analytic in the annulus 2
< |z1 | < 1, and hence is given by a Laurent series
∞
X
f (z1 , w) = am (w) · z1m . (*)
m=−∞

Using the expression of the am (w) as Cauchy-type integrals, one sees that the am (w) are
holomprphic in w. On the other hand, if |w2 | > 12 then φw (z1 ) is analytic on the whole unit
disk |z1 | < 1, and therefore am (w) = 0 for m < 0 when 21 < |w2 | < 1. It follows from the
Identity principle (Exercise 1.1 (ii)) implies that am (w) = 0 when m < 0 for every w with
|wi | < 1. But then (*) expresses f as a holomorphic function on all of ∆(ε). We refer to [3,
Proposition 1.1.4] for details. 
4 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Concerning the Cauchy-Riemann equations, it is convenient to introduce the differential

operators
√ √
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= − −1 , = + −1 . (1.1)
∂zi 2 ∂xi ∂yi ∂ z̄i 2 ∂xi ∂yi
(The conceptual meaning of these operators will emerge later.) Then one has:
Proposition 1.4. A C ∞ function f on U satisfies the Cauchy-Riemann equations – and
hence is holomorphic on U – if and only if
∂f
= 0 for i = 1, . . . , n. 
∂ z̄i

We conclude this subsection with a Lemma that will be useful on several occasions.
Lemma 1.5. Working with variables (z, w1 , . . . , wn−1 ) on Cn , fix r > ε > 0 and εi > 0. Let
g(z, w1 , . . . , wn−1 ) be a function which is analytic on the cylindrical shell
S = {r − ε < |z| < r + ε} × {|wi | < εi }.
Then the function
Z
f (w1 , . . . , wn−1 ) =def g(ξ, w1 , . . . , wn−1 ) · dξ
{|z|=r}

is holomorphic on the polydisk {|wi | ≤ εi }.

The reader is asked to prove this as Exercise 1.5. (Or see [3, Lemma 1.1.3].)

Weierstrass Preparation Theorem. An elementary fact from the one-variable theory is

that the zeroes of an analytic function consist of isolated points. In particular, one has a
complete picture of the (quite trivial) local geometry of these zeroes.

In higher dimensions the situation is of course vastly more complicated and interest-
ing: the zeroes of an analytic function form (by definition) a hypersurface in its domain
of definition. The most one can say – which however is already something – is that such
a hypersurface is described locally by the vanishing of a monic polynomial in one variable
whose coeffients are holomorphic functions of the remaining ones. This is the content of the
Weierstrass preparation theorem.

Turning to details, let f be a holomorphic function defined in a polydisk centered at the

origin 0 ∈ Cn , with f (0) = 0. We single out the first variable z1 , and write
gw (z1 ) = f (z1 , w) = f (z1 , w1 , . . . , wn−1 ),
which we view as an analytic function of z1 depending holomorphically on w. We assume
that
g0 (z1 ) 6≡ 0 in a neighborhood of 0. (1.2)
−1 n−1
The idea is to study the zero-locus Zeroes(f ) = {f (0)} by projecting to C .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 5

z1

Figure 1. Zeroes of a function with Weierstrass polynomial p = (z1 − w)(z12 − w3 ).

Theorem 1.6. (Weierstrass Preparation Theorem) In a suitable polydisk ∆ centered

at 0 one can write f uniquely as a product
f = p · h, (1.3)
where h is a holomorphic function that does not vanish at any point of ∆, and p is a is
Weierstrass polynomial in z1 , i.e. a function of the form
p = z1d + a1 (w) · z d−1 + . . . + ad−1 (w) · z1 + ad (w),
where the ai (w) are holomorphic functions vanishing at the origin.

In other words, the zeroes of f in ∆ are given by those of its Weierstrass polynomial. See
Figure 1: the zeroes of the Weierstrass polynomial are shown in the indicated polydisk
around the origin.

Proof. We start with some preliminary remarks. Fix 0 < ε  r, and consider for the moment
a function g(z) in a single complex variable z = z1 that is analytic in the disk {|z| < r + ε}.
We assume that g has no zeroes on the circle {|z| = r}. Denote by α1 , . . . , αd the zeroes of
g in ∆ = {|z| < r} (listed with multiplicity), and put
g 0 (ξ)
Z
1
σk = √ ξk · · dξ.
2π −1 |z|=r g(ξ)
A standard residue calculation shows that
d
X
σk = αik .
j=1

On the other hand, consider the polynomial p(z) = z d + a1 z d−1 + . . . + ad−1 z + ad defined by
p(z) = (z − α1 ) · . . . · (z − αd ).
6 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

By the theory of symmetric functions, the coefficients ai of p are given by (universal) poly-
nomials in the σk . On the other hand, we can write g = p · h, where – thanks to the Riemann
singularity theorem – h is an analytic function that is nowhere vanishing on ∆. The idea of
the proof of Theorem 1.6 is to repeat this discussion with g(z) replaced by gw (z1 ), and to
show that the resulting constructions vary holomorphically with w.

Specifically, choose 0 < ε  r  1 and 0 < εi  1 so that the cylindrical shall

S = {r − ε ≤ |z1 | ≤ r + ε} × {|wi | ≤ εi }
does not meet the zero locus of f , and so that 0 is the only zero of g0 (z1 ). Put
g 0 (ξ)
Z
1
σk (w) = √ ξk · w · dξ.
2π −1 |z1 |=r gw (ξ)
These are holomorphic in w (Exercise 1.5), and since for fixed w the integral σ0 (w) counts
the number of zeroes (with multiplicities) of gw (z1 ) in the disk of radius r in the z1 -plane, it
follows in particular that the number of such zeroes – say d – is independent of w. Denote
these zeroes by
α1 (w), . . . , αd (w).
These are not single-valued functions of w, but as above the functions ai (w) defined by the
equation
pw (z1 ) = (z1 − α1 (w)) · . . . · (z1 − αd (w))
= z1d + a1 (w) · z d−1 + . . . + ad−1 (w) · z1 + ad (w)
are polynomials in the σk (w), and hence holomorphic in w. Moreover by construction
α1 (0) = . . . = αd (0) = 0,
and hence the ai (w) vanish at the origin. This constructs the required Weierstrass polynomial
for f . It remains to show that we can find a holomorphic function h(z1 , w) on ∆ = ∆(ε; 0)
so that (1.3) holds.

To this end, consider the function

gw (z1 )
h0 (z1 , w) =def .
pw (z1 )
For fixed w, h0 (z1 , w) is bounded on {|z1 | ≤ r + ε} for suitable 0 < ε  1, and hence extends
to an analytic function on this disk thanks to the one-variable Riemann extension theorem.
Moreover, for fixed w this extension is given inside {|z1 | < r} by
Z
1 h0 (ξ, w)
h(z1 , w) = √ · dξ.
2π −1 |z1 |=r (ξ − z1 )
Now the integrand on the right is analytic in a neighborhood of the cylinder S, and it
follows that h is holomorphic in the w variables as well (Exercise 1.5 again). This produces
the required function satisfying (1.3). 
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 7

The last step in the argument just completed also yields the multivariable analogue of
the Riemann extension theorem.

Theorem 1.7. (Riemann Extension Theorem) Let f be a holomorphic function on an

open subset U of Cn , and consider a holomorphic function

h0 : U − Zeroes(f ) −→ C.

Assume that h0 is locally bounded near Zeroes(f ). Then h0 extends to a holomorphic function
h on all of U . 

We conclude with a variant of the preparation theorem, concerning divisibility by Weier-

strass polynomials.

Theorem 1.8. (Weierstrass Division Theorem) In the setting of Theorem 1.6, suppose
that f is holomorphic in a neighborhood of the origin, and let p = p(z1 , w) be a Weierstrass
polynomial in z1 of degree d. Then in a suitable polydisk ∆ about the origin, one can uniquely
write
f = q · p + s,
where q is holomorphic on ∆ and s = s(z1 , w) is a polynomial in z1 of degree < d whose
coefficients are holomorphic functions of w.

Sketch of Proof. Keeping notation as in the proof of the Weierstrass preparation theorem,
define
Z
1 f (ξ, w)
q(z1 , w) = √ · · dξ.
2π −1 |z1 |=r p(ξ, w) · (ξ − z1 )
We assume that p(z1 , w) has no zeroes on the shell

S = {r − ε ≤ |z1 | ≤ r + ε} × {|wi | ≤ εi },

and then q is holomorphic in a neighborhood of 0. So the issue is to show that

s(z1 , w) =def f (z1 , w) − q(z1 , w) · p(z1 , w)

is a polynomial in z1 of degree < d. To this end, write

f (ξ, w) · p(z1 , w)
Z Z
1 f (ξ, w) 1
s(z1 , w) = √ · · dξ − √ · · dξ
2π −1 |z1 |=r ξ − z1 2π −1 |z1 |=r p(ξ, w) · (ξ − z1 )
 
p(ξ, w) − p(z1 , w)
Z
1 f (ξ, w)
= √ · · · dξ.
2π −1 |z1 |=r p(ξ, w) (ξ − z1 )

But since p is a Weierstrass polynomial of degree d in z1 , the second term in the integral is
a polynomial of degree ≤ d − 1 in z1 , as required. For the uniqueness statement, see [3, p.
15]. 
8 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Algebraic Properties of the Ring of Holomorphic Germs. We now apply the Weier-
strass theorems to study the algebraic properties of the ring of germs of holomorphic functions
in a neighborhood of the origin 0 ∈ Cn .

For local questions, it is useful to identify two functions if they agree in a neighborhood
of a fixed point, say 0 ∈ Cn : this leads to the notion of the germ of a holomorphic function.
More formally, consider pairs (U, f ) where U is a neighborhood of 0 ∈ Cn , and f is a
holomorphic function on U . Define an equivalence relation ∼ on such pairs by declaring that
(U1 , f1 ) ∼ (U2 , f2 ) if there is a neighborhood V ⊆ U1 ∩ U2 such that
f1 |V ≡ f2 |V.
The resulting equivalence classes are called germs of holomorphic functions about 0, and
form a ring
On = OCn ,0 . (1.4)
Via Taylor expansions, there is an isomorphism
On ∼= C{z1 , . . . , zn }
of On with the ring of convergent power series in z1 , . . . , zn . This is a local integral domain:
On has a unique maximal ideal, consisting of (germs of) functions vanishing at the origin.
Of course one can discuss in a similar manner germs of holomorphic functions at any fixed
point a ∈ Cn , or germs of smooth (or continuous, or ...) functions at a point.
Remark 1.9. As we shall see later, the notation OCn ,0 indicates the fact that On is the
stalk at 0 of the sheaf OCn of holomorphic functions on Cn .

The ring On is very well-behaved algebraically. To begin with:

Proposition 1.10. The ring On is a unique factorization domain.

In other words, every non-unit f ∈ On can be written as a finite product of irreducible

elements, and this expression is unique up to multiplication by units and the order of factors.

Sketch of Proof. Induction on n, the case n = 1 being clear.1 Given f ∈ On vanishing at the
origin, we can assume after a linear change of coordinates that the hypothesis (1.2) of the
Weierstrass preparation theorem are satisfied. Thus we can write
f = p · u,
where p ∈ On−1 [z1 ] is (the germ of) a Weierstrass polynomial, and u ∈ On is a unit. Since
a polynomial ring over a UFD is itself a UFD, it follows by the induction hypothesis that
On−1 [z1 ] is a UFD. Thus p can be written essentially uniquely as a finite product of irreducible
elements of On−1 [z1 ], and it is elementary that any factor of a Weierstrass polynomial is
again a Weierstrass polynomial. The essential point is then to show that if p is irreducible
in On−1 [z1 ], then p remains irreducible considered as an element of On . To this end, suppose
1In fact, any f ∈ C{z} can be written uniquely in the form f = z m · u for some unit u ∈ Cz.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 9

that p = f1 · f2 is a product of two germs vanishing at 0. Then f1 , f2 are represented by

functions satisfying (1.2), and hence we can write
f1 = u1 · q1 , f2 = u2 · q2 ,
where q1 , q2 ∈ On−1 [z1 ] are germs of Weierstrass polynomials, and u1 , u2 ∈ On are units.
Then
p = (q1 q2 ) · (u1 u2 ),
and by the uniqueness of the Weiestrass expression it follows that u1 u2 = 1. Thus p = q1 · q2
is itself reducible in On−1 [z1 ]. 

Even more is true:

Theorem 1.11. The ring On is Noetherian, i.e. every ideal in On is finitely generated.

Proof. By induction on n, we may suppose that On−1 is Noetherian, and then it follows
by the Hilbert basis theorem that the polynomial ring On−1 [z1 ] is also Noetherian. Taking
z1 , w2 , . . . , wn as the coordinates on Cn , we may view
On−1 [z1 ] ⊆ On
as a subring of On . Now let I ⊆ On be any non-trivial ideal. Note to begin with that –
possibly after a linear change of coodinates – I contains a Weierstrass polynomial p(z1 , w)
in z1 . In fact, if f ∈ I is any element satisfying (1.2), then f = p · h, for some Weierstrass
polynomial p and some unit h ∈ On , and hence p ∈ I. On the other hand, the ideal
I 0 = I ∩ On−1 [z1 ]
is finitely generated, and we choose generators p1 , . . . , pt ∈ I 0 . We assert that p, p1 , . . . , pt
generate I. In fact, consider any f ∈ I. By the Weierstrass division theorem, we can write
f = q · p + s,
where s ∈ On−1 [z1 ]. But evidently s ∈ I, so s is an On linear combination of p1 , . . . , pt , and
we are done. 

We conclude this subsection by discussing ideals in On . These are related to (germs of)
analytic subvarieties of Cn . We start with a definition:
Definition 1.12. (Analytic subset or variety) Let U ⊆ Cn be an open set. An analytic
subset or subvariety 2 of U is a subset X ⊆ U with the property that for each point x ∈ X,
there is a neighborhood V 3 x of x in U such that
V ∩ X = Zeroes{f1 , . . . , fk }
for some holomorphic functions f1 , . . . , fk ∈ O(V ).

2In algebraic geometry, the term “variety” is sometimes reserved for an irreducible set.
10 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

For example, the zero-locus Zeroes(f ) ⊆ U of any holomorphic function f ∈ O(U ) is an

analytic variety.

We will be concerned with germs at 0 ∈ Cn of analytic subvarieties. Such a germ is

given by a subvariety 0 ∈ X ⊆ Cn , where X1 , X2 determine the same germ if
X1 ∩ V = X2 ∩ V
for some open neighborhood of 0. Every (germ of a) subvariety X determines an ideal in
OCn ,0 , namely
I(X) = {f ∈ OCn ,0 | f ≡ 0 on X in a neighborhood of 0}.
Conversely, any ideal J ⊆ OCn ,0 determines the germ of a subvariety X = Zeroes(J). In
fact, since J is finitely generated (Theorem 1.11), we can choose finitely many generators
f1 , . . . , fk ∈ J, and we then take X to be the germ of their common zeroes.

It is elementary that given the germ of an analytic subvariety X of Cn ,

Zeroes(I(X)) = X (1.5)
(Exercise 1.10). Furthermore, starting with an ideal J ⊆ OCn ,0 , one has


J ⊆ I Zeroes(J) . (*)
However, in general equality cannot be expected to hold in
(*). The reason is that if f ∈
OCn ,0 is a function with the property that f m ∈ I Zeroes(J) – in other words, if f m vanishes
on the analytic set Zeroes(J) – then f itself already vanishes on Zeroes(J), ie


f ∈ I Zeroes(J) .
This shows that
√
J =def f | f m ∈ J for some m > 0


⊆ I Zeroes(J) . (**)
The celebrated Nullstellensatz asserts that equality does hold in (**).
Theorem 1.13. (Nullstellensatz) For any ideal J ⊆ OCn ,0 , one has

√
I Zeroes(J) = J.

It follows that there is a one-one correspondence between (germs of) analytic subvarieties
of Cn and radical ideals in OCn ,0 (Exercise 1.11). We will not discuss the proof of the
Nullstellensatz here. For a partial proof, see [3].

Finally, we discuss irreducibility of germs.

Definition 1.14. A analytic germ X ⊆ Cn is irreducible if it cannot be written as the union
X = X1 ∪ X2
of two non-trivial germs with X1 , X2 ( X.
Proposition 1.15. (i). A germ X is irreducible if and only if its ideal I(X) is a prime
ideal in OCn ,0 .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 11

(ii) Any germ X can be written as a union

X = X 1 ∪ . . . ∪ Xk
of irreducible germs with no inclusions among the Xα . This decomposition is unique
up to ordering of the factors. 

Holomorphic Mappings. In this subsection we consider holomorphic mappings between

domains in Cn and Cm . These will be the local models for holomorphic mappings between
complex manifolds. We follow here Huybrechts’ presentation [3, pages 10 - 13] fairly closely.

Consider open sets

U ⊆ Cn , V ⊆ Cm .
Definition 1.16. A holomorphic mapping
f : U −→ V (1.6)
is a mapping given by holomorphic coordinate functions:


f (z) = f1 (z), . . . , fm (z) , fj ∈ O(U ). (1.7)
One says that f is biholomorphic if f has a holomorphic inverse (in which case of course
n = m).

√
As a matter of notation, we take wj = uj + −1 · vj to be the coordinate functions on Cm ,
and we write √
w = f (z) = u(z) + −1 · v(z).

The first point is to discuss the derivatives of such a mapping. Suppose for the moment
that f is merely C ∞ . We have natural identifications
Cn = R2n , Cm = R2m ,
so we can view f as a mapping between domains in R2n and R2m . In particular, for each
a ∈ U we can form the derivative
DR f = (DR f )a : Ta R2n −→ Tf (a) R2m . (1.8)
This derivative is given by the usual real Jacobian matrix JR (f ) of f : with respect to the
standard real bases of the spaces in question, JR f is represented by the block matrix:
    
∂u ∂u
 ∂x ∂y
JR (f ) = 
   ,
 (1.9)
∂v ∂v


∂x ∂y
 
∂ui
where eg ∂u


∂x
denotes the matrix ∂xj
, and we have supressed evaluation of the derivatives
at a.
12 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Now we want to bring the complex structures back into the picture. Namely, there are
natural identifications
Ta Cn = Cn , Tf (a) Cm = Cm .
In particular,√the source and target of (1.8) are actually complex vector spaces, with multi-
plication by −1 given by
√ ∂ ∂ √ ∂ ∂
−1 · = , −1 · = . (*)
∂xj ∂yj ∂ui ∂vi
∂
We may provisionally take , ∂
∂xj ∂ui
as complex bases of the tangent spaces in question.3

The basic fact is that f is holomorphic if and only if its derivative is complex linear:
Proposition 1.17. The C ∞ -mapping f : U −→ V is holomorphic if and only if the derivative
(DR f )a : Ta Cn −→ Tf (a) Cm
is C-linear for every a ∈ U . In this case, the derivative is represented with respect to the
standard bases by the complex Jacobian matrix
 
∂fi
J(f )a =def (a)
∂zj
of f , where as above
∂fi √
 
∂fi 1 ∂fi
= − −1 · .
∂zj 2 ∂xj ∂yj

When f is holomprphic, we will write

Dfa : Ta Cn −→ Tf (a) Cm
for the corresponding C-linear transformation

Proof. This is a restatement of the Cauchy-Riemann equations using (1.9). Omitting again
evaluation of the derivatives, JR (f ) is C-linear if and only if one has the matrix identities
       
∂u ∂v ∂u ∂v
= , = − ,
∂x ∂y ∂y ∂x
which are the Cauchy–Riemann equations. Moreover if these hold, then the C-linear mapping
determined by DR f is given by the m × n complex matrix
√
     
∂u ∂v ∂f
+ −1 · = . 
∂x ∂x ∂z

As a consequence, we find:

3Later we will prefer to work with the complex tagent vectors ∂

∂zj .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 13

Corollary 1.18. Assuming that f is holomorphic, the R-linear derivative

DR fa : Ta R2n −→ Tf (a) R2m
is injective or surjective if and only if the complex linear mapping
J(f )a : Cn −→ Cm
defined by the complex Jacobian is so. 

For future reference, it will be useful to record the relation between the real and complex
derivatives in the case when f is merely C ∞ . To this end, it is convenient to complexify the
real tangent spaces in question, i.e. we consider DR f as defining a C-linear mapping
C2n = Ta R2n ⊗R C −→ Tf (a) R2m ⊗R C = C2m . (*)
We take the complex tangent vectors
∂ ∂ ∂ ∂
, , ,
∂zj ∂ z̄j ∂wi ∂ w̄i
as bases for the two sides in (*). Then the matrix representing (*) has the block form
 ∂f
∂f

∂z ∂ z̄
 , (1.10)
∂ f¯ ∂ f¯



∂z ∂ z̄

where eg ∂f ∂fi



∂z
denotes the matrix ∂zj
, and we have again supressed mention of evaluation
at a. This yields convenient way of dealing with the chain rule.

The next point is that the inverse and implicit function theorems hold in the complex
setting.
Theorem 1.19. (Inverse and Implicit Function Theorems) Let U ⊆ Cn and V ⊆ Cm
be open sets, and let
f : U −→ V
be a holomorphic mapping. Assume that
Dfa : Ta U −→ Tf (a) V
is surjective at some point a ∈ U (and hence also in a neighborhood of a).

(i). If n = m then there exist neighborhoods U 0 3 a, V 0 3 f (a) such that f restricts to a

biholomorphic isomorphsm
∼
=
f : U 0 −→ V 0 .
(ii). If n > m then one can locally realize the fibre of f through a as the graph of a
holomorphic mapping. More precisely, after perhaps reindexing the coordinates so
that  
∂fi
∂zj 1≤i,j≤m
14 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

is non-singular, one can find open subsets

U1 ⊆ Cn−m , U2 ⊆ Cm , with U1 × U2 ⊆ U ,
together with a holomorphic mapping g : U1 −→ U2 such that
f −1 (f (a)) ∩ (U1 × U2 ) = (w, g(w)) | w ∈ U1 .


Sketch of Proof. Consider first the case n = m as in (i). Thanks to Proposition 1.18, it
follows from the usual smooth inverse function theorem that a local inverse g = f −1 exists as
a C ∞ mapping, and the issue is to show that g is holomorphic.  Tothis end, we differentiate
the relation g ◦ f = Id. With notation as in (1.10) one has ∂(Id) ∂ z̄
= 0, and this yields the
matrix equation
       ¯  
∂g ∂f ∂g ∂f ∂g
0 = · + · = · J(f ).
∂w ∂ z̄ ∂ w̄ ∂ z̄ ∂ w̄
Since J(f ) is non-singular, it follows that ∂∂gw̄ = 0, i.e. g is holomorphic. The proof of (ii)


is similar, and we refer to [3, p. 12] for details. 

Exercises for Section 1.

Exercise 1.1. (Maximum modulus and identity principles) Let U ⊆ Cn be a con-
nected open set. Reduce to the one-variable case to prove the following:

(i). If f ∈ O(U ) is a non-constant holomorphic function, then |f | cannot assume a

maximum on U . In particular, if U is bounded, and if f extends to a continuous
function on the closure U of U , then |f | assumes its maximum on the boundary of
U.
(ii). Suppose that V ⊆ U is a non-empty open set with the property that f is identically
zero on V . Then f is identically zero on U . Hence if g ∈ O(U ) is a second holomorphic
function which agrees with f on V , then f = g on all of U .
Exercise 1.2. Prove Proposition 1.4.
Exercise 1.3. (i). If f : U −→ C is a C ∞ function on an open set U ⊆ Cn , then
∂ f¯ ∂ f¯
   
∂f ∂f
= , =
∂ z̄ ∂z ∂z ∂ z̄
(ii). Let f : U −→ C, g : V −→ C be C ∞ functions on open subsets of C such that f ◦ g
is defined. . Then
∂ ∂f ∂g ∂f ∂ḡ
(f (g(z)) = · + ·
∂z ∂w ∂z ∂ w̄ ∂z
∂ ∂f ∂ḡ ∂f ∂g
(f (g(z)) = · + ·
∂ z̄ ∂ w̄ ∂ z̄ ∂w ∂ z̄
In particular, if f and g are analytic, then so is f ◦ g and the usual one-variable chain
∂
rule holds for ∂z .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 15

Exercise 1.4. (i). Prove that if f, g are holomorphic on U ⊆ Cn , then the product rule
holds for each of their z-derivatives, i.e.
∂(f g) ∂g ∂f
= f· +g· .
∂zi ∂zi ∂zi
(ii). Deduce that for a single variable z,
∂ m
(z ) = m · z m−1 .
∂z
(iii). In the setting of Theorem 1.1 (iii), show that the coefficient of z1j1 . . . znjn in the Taylor
series of f (z) is given by
1 ∂ j1 +···+jn f

aj1 ,...,jn = · j1 jn c .
j1 ! · · · jn ! ∂z1 · · · ∂zn
Exercise 1.5. Prove Lemma 1.5. (Cf. [3, Lemma 1.1.3].)
Exercise 1.6. Show that the Weierstrass polynomial associated to a given holomorphic
function f (z1 , w) is unique.
Exercise 1.7. Working in C2 with variables z, w, find the Weierstrass polynomial for
sin(w2 − z 3 ).
Exercise 1.8. Prove the Riemann Extension Theorem 1.7. Show that the conclusion holds
more generally if the hypersurface {f = 0} is replaced by any analytic subvariety X ⊆ U .
Exercise 1.9. Denote by An the ring of germs at 0 ∈ Cn of smooth functions. In contrast
to On , the ring An is not well-behaved algebraically. Specifically, prove that it is neither an
integral domain nor Noetherian.
Exercise 1.10. Prove the equality (1.5)
√
Exercise 1.11. Recall that an ideal J ⊆ OCn ,0 is said to be radical if J = J. Granting
the Nullstellensatz, prove that there is a one-to-one order-reversing correspondence between
germs of analytic subvarieties of Cn , and radical ideals in OCn ,0 .
Exercise 1.12. Prove Proposition 1.15 .
Exercise 1.13. In contrast to the case of holomorphic mappings between domains in C, it
is not the case in general that a holomorphic map between domains in higher dimensions is
an open mapping. For example, show that
f : C2 −→ C2 , (z, w) 7→ (z, zw)
is not an open mapping.
Exercise 1.14. Let f : U −→ V be a holomorphic mapping between domains U, V ⊆ Cn .
Then
det JR (f ) = | det J(f )|2 .
In particular, a biholomorphic mapping is always orientation-preserving.
16 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

2. Complex manifolds

In this section we introduce our principle objects of interest, namely complex manifolds.

Definition and First Examples. As one might expect, a complex manifold is simply a
differentiable manifold that carries holomorphic coordinate charts.
Definition 2.1. (Holomorphic atlas, complex manifold) Let X be a smooth manifold.

(i). A holomorphic atlas {(Ui , ϕi )} on X consists of a collection of open sets Ui ⊆ X

covering X, together with diffeomorphisms
∼
=
ϕi : Ui −→ ϕi (Ui ) ⊆ Cn
such that the transition functions
hi,j =def ϕi ◦ ϕ−1
j : ϕj (Ui ∩ Uj ) −→ ϕi (Ui ∩ Uj ) (2.1)
are biholomorphic.
(ii). Two such atlases {(Ui , ϕi )}, {(Ui0 , ϕ0i )} are equivalent if all the maps
ϕi ◦ ϕ0−1
j : ϕ0j (Ui ∩ Uj0 ) −→ ϕi (Ui ∩ Uj0 ) (2.2)
are biholomorphic.
(iii). A complex manifold of of (complex) dimension n is a smooth manifold of (real)
dimension 2n together with an equivalence class of holomorphic atlases.

Observe that since biholomorphic maps are orientation preserving (Exercise 1.14), a complex
manifold X is automatically orientable as a real manifold.

In the usual way, all of the notions studied locally in the previous section make sense on
such a manifold.
Definition 2.2. (Holomorphic function, mapping) Let X be a complex manifold.

(i). A holomorphic function on X is a function f : X −→ C with the property that there

is an atlas {(Ui , ϕi )} for X for which all the functions
f ◦ ϕ−1
i : ϕi (Ui ) −→ C

are holomorphic functions.

(ii). A smooth mapping f : X −→ Y is holomorphic if one can find atlases {(Ui , ϕi )} and
{(Vj , ψj )} for X and Y , with f (Ui ) ⊆ Vj (i) for some index j(i), with the property
that
ψj ◦ f ◦ ϕ−1
i : ϕi (Ui ) −→ ψ(Vi )
is a holomorphic mapping.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 17

These are independent of the choice of open set or the atlas thanks to the fact that the
functions appear in (2.1) and (2.2) are biholomorphic.

In the future, we will abbreviate this sort of definition by saying for example that
a function f on a complex manifold is holomorphic if it is given in local coordates by a
holomorphic function.
Definition 2.3. (Isomorphism) An isomorphism between complex manifolds is a holo-
morphic mapping
f : X −→ Y
that possesses a holomorphic inverse. In this case one says that X and Y are isomorphic or
biholomorphic.
Example 2.4. Any open subset U ⊆ X of a complex manifold is iteself a complex manifold.
In particular, any open subset U ⊆ Cn is a complex manifold of dimension n.
Example 2.5. (Riemann sphere) The standard Riemann sphere S = C ∪ {∞} is a
compact complex manifold (of dimension 1). Recall that in the finite part of the plane one
takes the usual coordinate function z as the local coordinate, while in a neighborhood of
the point at infinity one uses the coordinate w with w = z1 . In other words, the transition
function associated to this atlas is
1
h : C − {0} −→ C − {0} , z 7→ .
z
The Riemann sphere is also the one-dimensional instance of complex projective space Pn :
we will study this manifold in detail in the next subsection.
Example 2.6. (Analytic hypersurfaces) Let U ⊆ Cn+1 be an open set, let f : U −→ C
be a holomorphic function, and denote by
X = Zeroes(f ) ⊆ U
the zero-locus of f . Assume that for each point a ∈ X there is at least one index i such that
∂f
(a) 6= 0.
∂zi
Then X is a complex manifold of dimension n. (In fact, it follows from the Implicit Function
Theorem 1.19 (ii) that the n remaining coordinate functions on Cn+1 give local coordinates
in a neighborhood of a.) This construction leads to a vast number of important concrete
examples. Note that when n = 1 we get further examples of one-dimensional complex
manifolds.
Definition 2.7. (Riemann surface, complex curve) A Riemann surface is a complex
manifold of dimension one. These are also called smooth complex curves, especially when
they arise as in the previous two examples.
Example 2.8. Generalizing the construction of the previous example, suppose that V is a
complex manifold of dimension n + k, and that
X ⊆ V
18 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

is a subset defined as the common zeroes of a collection of k functions f = (f1 , . . . , fk ) such

that the Jacobian J(f )a has maximal rank at every point a ∈ X. Then again X is a complex
manifold of dimension n.

In the situation of the previous example, we say that X is a closed submanifold of V .

More generally:
Definition 2.9. (Closed submanifold). Let Y be a complex manifold of dimension n + k,
and let X ⊆ Y be a closed smooth submanifold of real dimension 2n. One says that X is
a closed submanifold of Y if there exists a holomorphic atlas {(Ui , ϕi )} on Y sich that if
Ui ∩ X 6= ∅, then
≡
ϕi : (Ui ∩ X) −→ ϕ(Ui ) ∩ Cn ,
where Cn ⊆ Cn+k is a linearly embedded copy of Cn in Cn+k .

It is equivalent to ask that X be locally cut out in Y by a k-tuple of holomorphic functions

having maximal rank (Exercise 2.4).

One can also talk about possibly non-closed submanifolds, whose definition we leave to
the reader. However since we are mainly interested in closed ones, we adopt the
Convention 2.10. By a “submanifold” X ⊆ Y , we always mean unless otherwise stated a
closed submanifold.
Example 2.11. (Complex Lie Groups) A complex Lie group is a conplex manifold G,
which is simultaneously a group, in such a way that multiplication and inversion are given
by holomorphic mappings
mult : G × G −→ G , inv : G −→ G.
For example, there is a natural identification
2
Matn×n (C) = Cn
2
of the set of all n × n complex matrices with Cn : the coordinate functions are the entries
of the matrix. Then
GLn (C) =def {A ∈ Matn×n | det(A) 6= 0}
is an open subset of Matn×n (C), hence a complex manifold, and thanks to Cramer’s rule
multiplication and inversion are holomorphic maps. Thus GLn (C) is a complex Lie group.
The other standard linear groups – notably the special linear, orthognal and symplectic
groups SLn (C), On (C), Sp2n (C) – are closed submanifolds of GLn (C), and hence themselves
complex Lie groups (Exercise 2.6). On the other hand, observe that the unitary and special
unitary groups Un and SUn are not complex submanifolds of GLn (C). In fact, it turns out
that the only compact connexted complex Lie groups are complex tori, which we discuss
next.
Example 2.12. (Complex tori, I). Let
Λ ⊆ Cn
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 19

be a lattice, i.e. a discrete subgroup of maximal rank. It is known that Λ ∼

= Z2n , and the
embedding Λ ⊂ Cn is determined by choosing 2n complex vectors
λ1 , . . . , λ2n ∈ Cn
that are linearly independent over R, and taking
Λ = Z · λ1 + . . . + Z · λ2n .
Now put
X = XΛ = Cn /Λ,
and denote by π : Cn −→ X the quotient mapping. Then X is diffeomorphic to a torus
(real) dimension 2n – i.e. a product ot 2n copies of the circle S 1 – and π is its universal
covering. A moment’s thought shows that there is a unique way to put a complex structure
on X in such a way that π is holomorphic: in fact, π is locally a diffeomorphism, and one
takes local inverses as coordinate charts. (See Exercise 2.7). Thus XΛ is a compact complex
manifold of dimension n, which is called a complex torus. Note however that whereas the
diffeomorphism type of XΛ doesn’t depend on the particular lattice Λ ⊂ Cn one starts with,
we will see in the next examples that its complex isomorphism type definitely does.
Example 2.13. (Morphisms of complex tori) Consider two lattices
Λ , Λ0 ⊆ Cn ,
defining complex tori XΛ , XΛ0 . If A : Cn −→ Cn is a linear transformation such that
A(Λ) ⊆ Λ0 , (*)
then evidently A induces a holomorphic mapping fA : XΛ −→ XΛ0 . Conversely, we will show
that any holomorphic mapping
f : XΛ −→ XΛ0 (**)
is essentially of this form. Granting this, it follows that for “most” pairs of lattices Λ, Λ0 , the
corresponding tori XΛ , XΛ0 cannot be biholomorphic.4 Suppose then given a holomorphic
mapping f as in (**). Then f lifts to a holomorphic mapping F : Cn −→ Cn on the universal
covering spaces with the property that
F (z + λ) − F (z) ∈ Λ0 (***)
for every z ∈ Cn , λ ∈ Λ, and after replacing F by a translate we may suppose that F (0) = 0.
As Λ0 is discrete, the left hand side of (***) is constant as a function of z, which implies
that all the holomorphic partials of F are Λ-periodic. These partials are therefore bounded,
and hence constant thanks to Liouville. Therefore F (z) = Az for some linear transformation
A, and of course A must then satisfy (*) since by construction F decends to a morphism
Cn /Λ −→ Cn /Λ0 .

4Roughly speaking, a lattice Λ ⊆ Cn depends on 2n2 complex parameters, whereas an n × n complex

matrix is determined on n2 complex parameters, so for “most” Λ, Λ0 there should be no non-zero matrices
satisfying (*). See Exercise 2.8 for some concrete examples.
20 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Remark 2.14. It is an interesting (but somewhat counter-intuitive) fact that if one starts
with a “random” lattice Λ ⊆ Cn , the resulting torus XΛ does not contain any proper analytic
submanifolds or subvarieties of positive dimension. Equivalently, there are no non-trivial Λ-
periodic sub varieties of Cn . On the other hand, if for instance
√
Λ = Zn + −1 · Zn ⊆ Cn ,
then XΛ is a product of one-dimensional complex tori, and so carries lots of submanifolds
(e.g. sub-products).
Example 2.15. (One-dimensional tori). We consider in more detail one-dimensional
complex tori. Denote by H the upper half plane in C. After a linear change of coordinates,
any lattice in C is spanned by 1 and some complex number τ ∈ H: write
Λτ = Z + Z · τ ⊆ C,
and let Xτ denote the corresponding one-dimensional torus. We will show that
aτ + b
Xτ ∼
= Xτ 0 ⇐⇒ τ 0 = (*)
cτ + d
for some  
a b
∈ SL2 (Z).
c d
Granting this for the moment, it follows that isomorphism classes of one-dimensional complex
tori are parametrized by H/SL2 (Z) under the action defined by (*). This quotient space has
been much studied: for example, a fundamental domain is given (modulo some identifications
on the boundary) by
|τ | ≥ 1 , |Re(τ )| ≤ 12 .
In fact, it turns out (somewhat non-trivially) that H/SL2 (Z) can itself be given the structure
of a one-dimensional complex manifold with
H/SL2 (Z) ∼
= C.
In summary, isomorphism classes of one-dimensional complex tori are parametrized by the
complex plane: this is the first example of a moduli space, ie a complex manifold (or analytic
variety) that parametrizes isomorphism classes of complex manifolds of fixed diffeomorphism
type. As for the proof of (*), it follows from Example 2.13 that Xτ ∼ = Xτ 0 if and only if
there is a complex number µ ∈ C∗ such that multiplication by µ carries Λτ 0 onto Λτ . This
is equivalent to asking that
µ · τ 0 = aτ + b , µ · 1 = cτ + d
 
0 aτ +b a b
for some integers a, b, c, d ∈ Z. Hence τ = ,
and the resulting matrix must have
cτ +d c d
determinant = 1 since it is invertible and maps the upper half plane to iteself.
Example 2.16. (Hopf manifolds). Given a real number 0 < λ < 1 and an integer k ∈ Z,
define
σk : Cn − {0} −→ Cn − {0} , σk (z1 , . . . , zn ) = (λk z1 , . . . , λk zn ).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 21

This defines a fixed point-free action of Z on Cn − {0}, and every point z ∈ Cn − {0} has
an open neighborhood disjoint from all of its images under the action. It follows that the
quotient
X = (Cn − {0}) /Z
carries in the natural way the structure of a complex manifold of dimension n, called the
Hopf manifold. It is diffeomorphic to the product S 2n−1 × S 1 . (See Exercise 2.9.) The
interest in this construction, as we will see later, is that it gives an example of a complex
manifold which cannot carry a Käbler metric.

Especially for algebraic geometers – who like to define spaces by means of defining equa-
tions – the construction in Example 2.6 and its generalizations are fundamental. But it
suffers from one serious drawback, namely that positive-dimensional closed submanifolds of
Cn are never compact. In fact, any submanifold of Cn carries lots of non-constant holo-
morphic functions (for instance the coordinate functions). On the other hand, one has the
elementary but basic
Proposition 2.17. Let X be a compact connected complex manifold. Then the only global
holomorphic functions on X are constants.

Proof. If f is a holomorphic function on X, then |f | assumes a maximum at some point of

X. By the maximum principle, f must then be constant. 

It is very important to have some ambient complex manifold in which one can “write
down” (using equations of some form) many compact submanifolds. We turn next to the
construction of projective space, which serves this function.

Projective Space. There are various equivalent definitions of projective space. As our base
construction we will take
Definition 2.18. Let C∗ act on Cn+1 − {0} by scalar multiplication, i.e.
λ · (a0 , . . . , an ) = (λ · a0 , . . . , λ · an ) for λ 6= 0. (2.3)
We define
Pn = Pn (C) =def Cn+1 − {0} /C∗

to be the quotient by this action, and we denote by

π : Cn+1 − {0} −→ Pn


(2.4)
the quotient map.

This specifies Pn as a set and a topological space. We will show momentarily how it carries
the structure of a complex manifold. Note that all the fibres of π are copies of C∗ sitting in
a one-dimensional complex subspace of Cn+1 .
22 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

A couple of preliminary remarks will be useful. First, note that any non-zero vector
0 6= a ∈ Cn+1 determines a unique one-dimensional complex subspace
C∼= C · a ⊆ Cn+1 .
Moreover, two non-zero vectors a, a0 ∈ Cn+1 determine the same subspace if and only if
a0 = λ · a for some λ ∈ C∗ .
Hence
Viewpoint 2.19. The projective space Pn parametrizes all one-dimensional vector sub-
spaces of Cn+1 .

This being the case, it is sometimes useful to take a coordinate-free approach. Specifically,
given any complex vector space V of dimension n + 1, we denote by P(V ) the n-dimensional
projective space of one-dimensional vector subspaces of V .5

Next, there is a very useful way to specify points in projective space by so-called homo-
geneous coordinates. Namely, we can view (2.3) as determining an equivalence relation on
all non-zero vectors 0 6= a ∈ Cn+1 , where
a0 ∼ a ⇐⇒ a0 = λ · a for λ ∈ C∗ ,
and then evidently n
o
Pn = a ∈ Cn+1 − {0} / ∼ .
Given a non-zero vector a = (a0 , . . . , an ), denote by
[a] = [a0 , . . . , an ]
its equivalence class. Thus [a] determines a point π(a) ∈ Pn with homogeneous coordi-
nates [a0 , . . . , an ]. In other words, points in Pn are specified by homogeneous coordinates
[a0 , . . . , an ] with at least one non-zeroi entry, where
[a00 , . . . , a0n ] = [a0 , . . . , an ] ⇐⇒ a0i = λ · ai for some λ ∈ C∗ .
Remark 2.20. Note that one can replace C in these definitions and constructions by any
field k, leading to
Pn (k) = k n+1 − {0} /k ∗ .

Proposition 2.21. Complex projective space Pn carries a unique structure of a complex

manifold in such a way that the map
π : Cn+1 − {0} −→ Pn
is a locally trivial holomorphic C∗ -bundle.
5Following
a convention introduced by Grothendieck – which is a more general context is essentially forced
on one – algebraic geometers sometimes use P(V ) to mean the projective space of one-dimensional quotients
of V . So if you are reading a book you have to be careful to figure out which of these two possibilities
“P(V )” refers to. We’ll try to stick to the subspace convention here.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 23

In other words, Pn carries the structure of a complex manifold in such a way that first of all
π is holomorphic. Moreover we can find an open covering {Ui } of Pn by open sets with the
property that
π −1 (Ui ) ∼
= Ui × C∗ (2.5)
as complex manifolds.

Sketch of Proof. For each index 0 ≤ i ≤ n, let

Pn ⊇ Ui = {[a0 , . . . , an ] | ai 6= 0.}
This is the image of the open set {ai 6= 0} ⊆ Cn+1 , hence is open in Pn . By dividing through
by ai , the homogeneous coordinates of any point in Ui can be expressed uniquely as a vector
whose ith coordinate is 1:
[a0 , . . . , an ] = [ aa0i , . . . , 1, . . . aani ].
Thus there is a natural identification
Ui −→ Cn , [a0 , . . . , an ] 7→ ( aa0i , . . . , ai−1
ai
, ai+1
ai
, . . . , aani ).
We take these maps as local coordinates on Ui . The reader should check that the resulting
transition functions are biholomorphic. We likewise leave it to the reader to construct the
isomorphism (2.5). 
Example 2.22. (P1 ). In the case n = 1 we recover the Riemann sphere S, i.e. P1 = S. In
fact, keeping the notation of the previous proof, consider a point [a0 , a1 ] ∈ P1 . On the open
set U0 = {a0 6= 0} take we use z = a1 /a0 as the local coordinate, while on U1 = {a1 6= 0} we
use w = a0 /a1 . Thus
1
w= on U0 ∩ U1 = C∗ ,
z
as required.

It is perhaps not yet apparent that Pn is compact. To see this, let k k denote the
standard Euclidean norm on Cn , and consider the unit sphere
S 2n+1 = {k z k = 1} ⊆ Cn+1 .
Consider the restriction
h : S 2n+1 −→ Pn
to S 2n+1 of the basic quotient mapping π in (2.4). The compactness of Pn is a consequence
of the fact that h is surjective, which in turn follows from the fact that
 
a
π(a) = π
kak
for any non-zero vector a ∈ Cn+1 . To say the same thing a little differently, consider the unit
circle S 1 = {|z| = 1} ⊆ C of complex numbers of length 1. Then S 1 acts on S 2n+1 by scalar
multiplication, and h is the quotient map. The mapping h is called the Hopf vibration: the
reader should check that it expresses S 2n+1 as the total space of a locally trivial S 1 bundle
over Pn .
24 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Example 2.23. The case n = 1 of this discussion is already very interesting. Recalling that
P1 = S 2 is the Riemann sphere, the Hopf fibration in this case is a smooth mapping
h : S 3 −→ S 2
expressing the three-sphere as a circle bundle over the two-sphere. In particular, for any
point p ∈ S 2 , the fibre h−1 (p) is a circle in S 3 . The interesting fact is that for distinct
p, q ∈ S 2 , these circles are linked. (One can verify this pictorially by making a stereographic
projection from S 3 − {point} to R3 and working out explicitly the equations of two fibres.)

We next discuss the paving of Pn by affine spaces. By construction there is a holomorphic

inclusion
Cn = U0 ⊆ Pn , (z1 , . . . , zn ) 7→ [1, z1 , . . . , zn ].
The complement Pn − U0 consists of the set of points [0, z1 , . . . , zn ] whose first homogeneous
coordinate vanishes. This is in turn isomorphic to a projective space of dimension n − 1,
with homogeneous coordinates [z1 , . . . , zn ]. In other words:
Pn = Cn t Pn−1 ,
and one thinks of Pn−1 as the “hyperplane at infinity” that one adds to compactify Cn :
there is one point at infinity added for each complex direction in Cn . Repeating the same
discussion starting with the hyperplane at infinity, one ends up by expressing Pn as a disjoint
union of affine spaces:
Pn = Cn t Cn−1 t . . . t C1 t C0 .
The analogous thing works for projective space over any field, and the resulting decomposi-
tion is the analogue of the fact familiar from topology that Pn has the structure of a CW
complex with one cell in each even (real) dimension 0, 2, . . . , 2n.

We now turn to describing loci in projective space Pn by means of polynomial equations.

Note that if f ∈ C[z0 , . . . , zn ] is a polynomial, the value of f at a point [a] = [a0 , . . . , an ] ∈ Pn
is not well-defined, since in general
f (a0 , . . . , an ) 6= f (λ · a0 , . . . , λ · an ).
On the other hand, suppose that f is homogeneous of degree d, i.e. every monomial in f has
degree d. Then
f (λ · a0 , . . . , λ · an ) = λd · f (a0 , . . . , an )
for every λ ∈ C∗ . Therefore it makes sense to talk about whether or not f vanishes at a
point [a] ∈ Pn .
Definition 2.24. A projective algebraic set X ⊆ Pn is the locus of common zeroes of a
collection
F1 , . . . , FN ∈ C[Z0 , . . . , Zn ]
6
of homogeneous polynomials. We write
X = Zeroes(Fα ).
6One often uses capital letters to denote homogeneous polynomials and variables.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 25

Example 2.25. (Linear subspaces). Suppose that

L1 (Z0 , . . . , Zn ), . . . , Le (Z0 , . . . , Zn )
are linear polynomials that are C-linearly independent. Then the Lα cut out a sumbanifold
Λ ⊆ Pn
that is isomorphic to Pn−e . Submanifolds arising in this manner are called linear subspaces
of Pn .
Example 2.26. (A smooth conic). Working on P2 with homogeneous coordinates X, Y, Z,
consider the conic curve
C = {XY − Z 2 = 0}.
This is illustrated schematically in Figure 2, which shows the intersections of C with the
three “axes” X = 0, Y = 0 and Z = 0.7 It is interesting to consider the intersections of C
with the three copies UZ , UY , UZ of C2 obtained as complement of these axes. We identify
UZ = {Z 6= 0} with CC 2 by taking affine coordinates
X Y
x = , y = ,
Z Z
2
and then C0 ⊆ C is defined by the equation
xy − 1 = 0.
Note that (the real points of) this curve is a hyperbola with the x- and y-axes as asymptotes:
this corresponds to the fact that the compact curve C ⊆ P2 meets the line Z = 0 at infinity at
two points. On the other hand, consider the intersection C ∩ UX . Here the affine coordinates
are
Y Z
y = , z = ,
X X
and the defining equation is y − z 2 = 0. Here we get a parabola, reflecting the fact that the
compact curve C is tangent to the line X = 0 at infinity.

Remark 2.27. (Dehomogenization). As the previous example illustrates, if X ⊆ Pn is

defined by homogeneous polynomials
Fα (Z0 , . . . , Zn ) ∈ C[Z0 , . . . , Zn ],
and if as above
Ui = {Zi 6= 0} = Cn
denotes the ith standard copy of Cn sitting inside Pn , then
X ∩ Ui ⊆ Ui = Cn
is cut out by the (generally inhomogeneous polynomials) obtained by setting Zi = 1 in Fα ,
and treating the remaining variables as affine coordinates on Cn .8
7One typically tried to visualize complex varieties by drawing their real parts. This what the Figure shows
is the real conic XY − Z 2 = 0 in P2 (R). But in happy situations, with a little practice these real drawings
can be quite suggestive.
8However, if one starts with an algebraic set

X0 ⊆ U0 = Cn
26 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

XY - Z2 = 0

Z=0

Y=0 X=0

Figure 2. The conic XY = Z 2 in P2 .

Definition 2.28. (Complex projective manifold). A complex projective manifold is a
complex manifold X that arises as a closed complex sumbanifold
X ⊆ Pr
of some projective space Pr .

Thanks to the compactness of Pr , any projective manifold is compact. It may happen (and
in fact always does) that a given complex manifold can be realized in different ways as
submanifolds of projective spaces of various dimensions.
Remark 2.29. (Chow’s Theorem). A remarkable theorem of Chow states that if X ⊆ Pr
is a complex submanifold, then in fact X is the algebraic set cut out by a collection of
homogeneous polynomials. The same statement remains true if one asks merely that X be
an analytic subvariety of Pr , ie locally cut out by holomprohic functions. Unfortunately we
won’t be able to prove this in Math 545, but it’s good to keep the fact in mind.

The following statement generalizes Example 2.6:

Proposition 2.30. Let F ∈ C[Z0 , . . . , Zn ] be a homogeneous polynomial of degree d, and
assume that the partial derivatives
∂F ∂F
,...,
∂Z0 ∂Zn
defined by a collection of polynomials {fα (z1 , . . . , zn )} and looks for the “projective closure” X ⊆ Pn of X0
– ie the smallest projective algebraic set restricting to X0 on U0 – then one typically needs to consider more
equations than the homogenizations of the fα .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 27

(which are homogeneous polynomials of degree d − 1) have no common zeroes in Pn . Then

X = Zeroes(F ) ⊆ Pn
is an (n − 1)-dimensional sub manifold of Pn .

Proof. This follows from Example 2.6 (ie the implicit function theorem) together with Euler’s
Theorem: if F is homogeneous of degree d, then
∂F ∂F
+ ... + = d · F.
∂Z0 ∂Zn
We leave detais to the reader. 

Given a projective manifold

X ⊆ Pr ,
we next discuss a convenient way to write down holomorphic mappings
f : X −→ Ps .
For this, consider
F0 , . . . , Fs ∈ C[Z0 , . . . , Zn ]
homogeneous polynomials of the same degree d. Assume that
Zeroes(F0 , . . . , Fs ) ∩ X = ∅.
Note that given a point x = [a] ∈ X, the homogeneous vector [F0 (a), . . . , Fs (a)] gives a
well-defined point in Ps : the values Fi (a) are determined up to a common scalar multiple,
and they don’t vanish simultaneously for x ∈ X.
Proposition 2.31. The mapping f : X −→ Ps defined by
f (x) = [F0 (x), . . . , Fs (x)]
is holomorphic. If Y ⊆ Ps is a complex submanifold, and if Im(f ) ⊆ Y , then f gives rise to
a holomorphic mapping
f : X −→ Y. 
Example 2.32. Define
ν : P1 −→ P2 via [s, t] 7→ [s2 , t2 , st].
The image of ν is the conic C = {XY − Z 2 = 0} discussed above, and in fact ν establishes
an isomorphism P1 ∼ = C. It is a nice exercise to write down the inverse of ν.
Example 2.33. (PGL action on Pn ) Using Example 2.31 (or directly) one sees that
SLn+1 (C)
PGLn+1 (C) =def
C∗ · Id
acts on Pn by linear automorphisms. In fact, it turns out that Aut(Pn ) = PGLn+1 (C).
28 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Finally, we discuss products. Given projective manifolds

X ⊆ Pr , Y ⊆ Ps ,
it is natural to ask how one can realize the product X ×Y as a projective manifold. The essen-
tial point is to realize Pr × Ps projectively, and this is accomplished by the Segre embedding
Pr × Ps ⊆ Prs+r+s ,
defined as follows. View Prs+r+s as the projective space of all (r + 1) × (s + 1) matrices, so
that a point in Prs+r+s is described by giving a non-zero (r + 1) × (s + 1) matrix [ci,j ] defined
up to scalars. Then put
σ = σr,s : Pr × Ps −→ Prs+r+s , [a0 , . . . , ar ] × [b0 , . . . , bs ] 7→ ai bj .
 
(2.6)
For example, when r = s = 1 the Segre mapping becomes
 
1 1 3 a0 b 0 a0 b 1
P × P −→ P , [a0 , a1 ] × [b0 , b1 ] 7→ , (2.7)
a1 b 0 a1 b 1
whose image is the (smooth) quadric in P3 arising as the set of all 2 × 2 matrices of rank 1.
Proposition 2.34. (Segre embedding). The Segre mapping σr,s defines a holomorphic
embedding
Pr × Ps ⊆ Prs+r+s
whose image is the (projectivized) set of matrices of rank ≤ 1.

Intrinsically, given vector spaces V and W , one can describe the Segre embedding as the
map
P(V ) × P(W ) ,→ P(V ⊗ W ) , [v] × [w] 7→ [v ⊗ w].
(Recall that V ⊗ W consists of linear combinations of pure tensors v ⊗ w; the Segre variety
consists of the projectivization of the set of all elements of V ⊗ W that happen to be of the
form v ⊗ w for some v ∈ V and w ∈ W .)

Having defined an embedding Pr × Ps ⊆ Prs+r+s , one can define an algebraic subset of

P × Ps to be the intersection of Pr × Ps with an algebraic subset of Prs+r+s . However there
r

is an easier “internal” way to work with these. Specifically, one says that a polynomial
P (Z, W ) = p(Z0 , . . . , Zr , W0 , . . . , Ws )
is bihomogeneous of bidegree (d, e) if P is homogeneous of degree d in the Zi and homoge-
neous of degree e in the Wj . For example
Z03 W0 W1 + Z12 Z2 W22
is bihomogeneous of degree (3, 2), but Z1 − W1 is not bihomogeneous. If P is bihomogeneous
of bidegree (d, e), then
P (λZ, µW ) = λd µe P (Z, W ),
so the zero-locus of a bihomogeneous polynomial is a well-defined subseteq of Pr × Ps . We
can then define an algebraic subset of Pr × Ps to be the common zero-locus of a collection
of bihomogeneous polynomials; the reader is asked to check in Exercise 2.17 that this is the
same thing as intersecting the Segre variety with an algebraic subset of Prs+r+s .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 29

There is a similar “internal” description of algebraic subsets of Pr × Cs : these are

described by polynomials p(Z, w) that are homogeneous in the Z-variables, but arbitrary in
the w-variables. Equivalently, p(Z, w) is a homogeneous polynomial in Z whose coefficients
are polynomials in w. (One could allow the coefficients to be holomorphic functions in the
w.) The following example gives an important special case of this construction.
Example 2.35. (The blowing-up of Cn+1 at the origin.) Recalling that Pn parametrizes
one-dimensional subspaces of Cn+1 we can form the incidence correspondence in Pn × Cn+1
formed by pairs consisting of a point in [a] ∈ Pn and a vector v ∈ C · a lying on the line
C · a corresponding to [a]:
Z = ([a], v) | v ∈ C · a ⊆ Pn × Cn+1 .


This is an algebraic subset: in fact, v = (v0 , . . . , vn ) lies on the line spanned by [a] =
[a0 , . . . , an ] if and only if the vectors v and a are linearly dependent, so B is cut out in
Pn × Cn+1 by the 2 × 2 minors of the matrix
 
Z0 . . . Zn
.
w0 . . . wn
It is very interesting to consider the two projections
p : Z −→ Pn , b : Z −→ Cn+1 .
For any point [a] ∈ Pn , the fibre p−1 ([a]) is the one-dimensional subspace C · a spanned by
a, and in fact Z is locally a product of Pn and C. On the other hand, consider the fibres
of b. If 0 6= v ∈ Cn+1 , then v lies on a unique line C · v ⊆ Cn+1 , and b−1 (v) consists of the
corresponding point of Pn ; in other words, b is an isomorphism away from 0. On the other
hand, b−1 (0) is all of Pn . The mapping b : Z −→ Cn+1 is called the blowing-up of Cn+1 at
the origin. Homotopically, Z is obtained from Cn+1 by removing a small disk about 0, with
boundary S 2n+1 , and gluing in a copy of Pn via the Hopf map h : S 2n+1 −→ Pn . See Figure
3.

Holomorphic Line and Vector Bundles. Given a projective manifold, we have seen
(Proposition 2.31) that one can use homogeneous polynomials to define holomorphic map-
pings to Ps . But suppose one starts with an abstract compact complex manifold X. How
then can one attempt to construct maps – and ideally embeddings – φ : X −→ Pr ? The
answer is that such morphisms are defined via sections of holomorphic line bundles; φ nec-
essarily has the form
φ(x) = [s0 (x), . . . , sr (x)]
where si (x) ∈ Γ(X, L) are sections of a holomorphic line bundle without no common zeroes.
In this subsection we give the basic definitions and examples of line and vector bundles.

Let X be a connected complex manifold. There are various equivalent definition of

holomorphic line (or vector) bundles on X. We start with the most geometric, although it
is not always the most useful in practice.
30 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Figure 3. Blowing up the origin in Cn+1

First Viewpoint (Total Space). Intuitively, we can define a holomorphic vector bundle
on X to be a holomorphic family of vector spaces parametrized by X.
Definition 2.36. A holomorphic vector bundle on X of rank e is a complex manifold E,
together with a holomorphic mapping
p : E −→ X,
with the property that p realizes E a locally the product of X with an e-dimension vector
space in such a way that the transition functions are given by linear automorphisms.

The definition requires some elaboration. To begin with, we ask that there be given an
open covering {Ui } of X together with biholomorphic isomorphisms
∼
=
φi : p−1 (Ui ) −→ Ui × Ce
commuting with the projections to Ui :
∼
φi : p−1 (Ui )
= / Ui × Ce

p pr1
% {
Ui .
Now set
Uij = Ui ∩ Uj ,
and consider the “j-to-i” transition map
φij = φi ◦ φ−1 e e
j : Uij × C −→ Uij × C .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 31

Figure 4. Diagram of Transition Functions

The second requirement is that φij be given by

φij (x, v) = x , gij (x) · v ,
where
gij : Uij −→ GLe (C)
is a holomorphic mapping. In other words, the transition functions are given by a family of
linear automorphisms
gij (x) : Ce −→ Ce
varying holomophically with x. The situation is summarized in a big commutative diagram
shown in Figure 4. Note that it follows from the definitions that on the triple overlaps
Uijk = Ui ∩ Uj ∩ Uk , the transition matrices satisfy the cocycle condition:
gij · gjk = gik . (2.8)

Definition 2.37. A vector bundle of rank e = 1 is called a line bundle. Note that in this
case the gij (x) are simply nowhere-vanishing holomorphic functions on Uij , i.e.
gij ∈ O∗ (Ui ∩ Uj ).
Remark 2.38. These definitions make sense in any fixed geometric setting: one requires
that the local product isomorphisms φi and the transition matrices gij be morphisms in
the appropriate category. So for example one can discuss continuous real or complex vector
bundles on a topological space, or C ∞ vector bundles (real or complex) on a smooth manifold.
So for example, if M is a smooth manifold of dimension n, its tangent bundle T M is a real
vector bundle of rank n. (See Exercise 2.18.)
Example 2.39. The global product 1e = 1eX = X × Ce is called the trivial vector bundle
of rank e. As its name implies, this is not a very exciting example.
32 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Example 2.40. (OPn (−1)). As in Example 2.35, consider the incidence correspondence
Pn × Cn+1 ⊇ L = ([a], v) | v ∈ C · a .


Then the first projection p : L −→ Pn realizes L a line bundle on Pn , which is called (for
reasons that will become apparent later) OPn (−1).9 We work out the transition functions for
this bundle, assuming for ease of notation that n = 1. On P1 with homogeneous coordinates
[Z0 , Z1 ] we consider as usual the open sets
U0 = [1, ZZ10 ] , U1 = [ ZZ01 , 1]
 

. We can trivialize p over these sets. In fact

n
o
2 −1 Z1 Z1
U0 × C ⊇ p (U0 ) = [1, Z0
], t · (1, Z0
)
n
o
U1 × C2 ⊇ p−1 (U1 ) = [ ZZ01 , 1] , s · ( ZZ01 , 1) ,
and then
φ0 : p−1 (U0 ) −→ U0 × C , φ1 : p−1 (U1 ) −→ U1 × C
are defined by
 
φ0 [1, ZZ10 ] , t · (1, ZZ10 ) = [1, ZZ10 ], t


 
φ1 [ ZZ01 , 1] , s · ( ZZ01 , 1) = [ ZZ01 , 1], s

In order to determine the transition function g10 , we compute the composition φ1 ◦ φ0−1 :
[1, ZZ01 ], t 7→ [Z0 , Z1 ], t · (1, ZZ01 ) = [Z0 , Z1 ], ( ZZ10 t) · ( ZZ01 , 1) 7→ [ ZZ01 , 1], ( ZZ10 t) .

In other words,  
Z1

g10 [Z0 , Z1 ] = .
Z0
Similarly, with respect to the standard open covering Ui = {Zi 6= 0} of Pn , the transition
functions of OPn (−1) are given by
 
Zi
gij = . (2.9)
Zj
Example 2.41. (OPn (1)). In this example, we describe the hyperplane line bundle on Pn ,
which is also called OPn (1). Embed Pn into Pn+1 as the hyperplane Zn+1 = 0, and set
O = [0, . . . , 0, 1] ∈ Pn+1 . Linear projection from O defines a mapping
p : Pn+1 − {O} −→ Pn , [Z0 , . . . , Zn+1 ] 7→ [Z0 , . . . , Zn ].
Geometrically, p sends a point x ∈ Pn+1 − {O} to the unique point of intersection of the
line joining O and x with the hyperplane Pn ⊆ Pn+1 . (See Figure 5.) The reader should
check that p realizes Pn+1 − {O} as the total space of a line bundle on Pn , whose transition
functions with respect to the standard open covering {Ui } of Pn are given by
 
Zj
gij = . (2.10)
Zi
9This is sometimes also called the Hopf line bundle, or the tautological line bundle (since its fibre over a
point in Pn is “the line that that point is”).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 33

Figure 5. The hyperplane line bundle

Note that if p : E −→ X is a vector bundle, and x ∈ X is any point, then the fibre p−1 (x)
has the structrue of a vector space (defined up to isomophism): this comes from any of the
local trivializations of E and the fact that the comparison maps are linear isomorphisms.10
The general yoga is that any intrinsically defined notion or operation on vector spaces makes
sense for vector bundles. For example, if
p : E −→ X , q : F −→ X
are two vector bundles, then a homomorphism u from E to F is a holomorphic mapping
u : E −→ F,
commuting with p and q, such that u is linear on each fibre. (The reader should convince
him/herself that the linearity condition is well-defined.) Similarly, one can associate to
p : E −→ X the dual bundle E∗ −→ X whose fibre over a point x ∈ X is the vector space
of linear functionals on p−1 (x). (See Exercise 2.22.)

Second Viewpoint (Transition Data): Another viewpoint of the theory – which is

often more convenient for concrete calculations – is to take as basic objects the transition
data {Ui , gij } coming from local trivializations of E over an open covering {Ui } o9f X. Note
that if we’re given such an open covering, together with invertible matrices
gij ∈ GLe (O(Uij ))
satisfying the cocycle condition (2.8), then we can build a vector bundle p : E −→ X
described by the given data. In fact, one starts with the disjoint union of the products
Ui × Ce , and then uses the gij to glue these along the common open sets Uij × Ce . Working
with transition data offers a very concrete way of dealing with vector bundles, and we will
make it a point to explicate all the constructions in these terms. Of course we will then also
have to understand when two data {Ui , gij } and {Ui , gij0 } give rise to isomorphic bundles.
10Compare Exercise 2.20.
34 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Figure 6. Section of a vector bundle

For our purposes, the most important feature of a bundle are its global sections:
Definition 2.42. Let p : E −→ X be a holomorphic vector bundle. A global section of E is
a holomorphic mapping
s : X −→ E such that p ◦ s = idX .

The condition on p ◦ s = idX guarantees that s(x) ∈ p−1 (x) for every x ∈ X. Thus a section
picks out a vector in each fibre of E over X. We denote by
Γ(X, E) or Γ(X, O(E))
the set of all global sections of E. Thanks to the vector space structure on the fibres of p,
we can add sections and multiply them by functions. Thus Γ(X, O(E)) is a module over the
ring of O(X) of holomorphic functions on X; in particular, Γ(X, E) is naturally a complex
vector space. The zero section 0 = 0E ∈ Γ(X, O(E)) is the section that picks out the zero
vector in each fibre. See Figure 6.
Example 2.43. A section of the trivial line bundle 1 is just a holomorphic function on X.
Example 2.44. Working in the C ∞ setting, a section of the tangent bundle of a smooth
manifold M is a vector field on M .

It is important to work out the meaning of sections in terms of transition data. Suppose
that p : E −→ X is described by data {Ui , gij }: in other words, {Ui } is an open cover of
X on which p trivializes, and gij ∈ GLe (O(Uij )) are the corresponding transition matrices.
Then the composition
∼
=
φi ◦ s : Ui −→ p−1 (Ui ) −→ Ui × Ce
is the graph of a vector-valued holomorphic function
fi : Ui −→ Ce :
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 35

equivalently, fi is a vector of holomorphic functions. A moment’s thought shows that on the

overlaps Uij , these satisfy the basic compatibility relation
fi = gij · fj . (2.11)

It is very suggestive to think of the data {Ui , fi } as a sort of “twisted” vector-valued

holomorphic function on X. If E = 1e – so that one may take each gij to be the identity
matrix – then (2.11) simply means that the fj patch together to a global vector-valued
holomorphic function on X, and of course when X is compact the fi must therefore be
constant. On the other hand, as we shall see shortly, the presence of non-trivial transition
matrices gij raises the possibility of having many non-trivial holomorphic sections even in
the compact setting.

The most important case of this principle occurs when e = rank E = 1, in which case
sections of suitable line bundles serve as substitutes for global holomorphic functions. This
is illustrated by the basic
Proposition 2.45. Global sections of OPn (1) are identified with linear forms on Pn , ie
Γ Pn , OPn (1) ∼


= linear forms in Z0 , . . . , Zn .

Proof. For simplicity we will treat the case n = 1. Recall that in general the transtion
functions of OPn (1) with respect to the standard open covering {Ui } of Pn are given by
gij = Zj
Zi
. In the case n = 1 this gives the one transition function
Z0
g10 = .
Z1
Now say s ∈ Γ(P1 , OP1 (1)). Then s|U0 and s|U1 are given by entire functions s0 ( ZZ01 ) and
s1 ( ZZ10 ) satisfying
     
s1 ZZ10 = ZZ01 · s0 ZZ10 . (*)
Writing
  2

Z1
s 0 = a0 + a1 + a2 ZZ10 + . . .
Z0
   2
s1 = b0 + b1 ZZ10 + b2 ZZ01 + . . . ,
(*) immediately implies that
a2 = a3 = . . . = 0 , b2 = b3 = . . . = 0,
and
a0 = b 1 , a 1 = b 0 .
Thus s may be identified with the linear form
s = a0 Z0 + a1 Z1 ,
s s
with s0 = Z0
and s1 = Z1
. 
36 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Remark 2.46. (Finite dimensionality). Note that in particular Γ(Pn , OPn (1)) is a finite-
dimensional vector space. This is a special case of a non-trivial theorem to the effect that if
X is a compact complex manifold, then the space Γ(X, O(E)) of sections of any holomorphic
vector bundle is finite-dimensional.

Suppose that E −→ X and L −→ X are respectively a vector bundle and a line bundle
given by transition data {Ui , gij } and {Ui , hij }. Then we may form the vector bundle
E⊗L:
this is a bundle having the same rank as E, and transition matrices hij · gij .11 In particular
for k ≥ 1, the k-fold tensor product L⊗k is described by the transition functions (hij )k . For
projective space, the computation of Proposition 2.45 generalizes to yield:
Proposition 2.47. Set

⊗k
OPn (k) = OPn (1) .
Then

Γ Pn , OPn (k) ∼

= homogeneous polynomials of degree k in Z0 , . . . , Zn .

If L is a line bundle given by data {Ui , gij } , put

L−1 = L∗ , L−k = (L−1 )⊗k .
Then L−k has transition functions gij−k . Noting in particular that
L ⊗ L∗ ∼= 1,
we see that the set of all isomorphism classes of complex line bundles is an abelian group
under tensor product: this is the Picard group Pic(X) of X.

We next discuss homomorphisms and isomorphisms from the viewpoint of transition

data. Suppose that p : E −→ X and q : F −→ X are two bundles of ranks e and f ,
described by data {Ui , gij } and {Ui , hij }, and consider a homomorphism ν : E −→ F. Via
the identifications
p−1 (Ui ) = Ui × Ce , p−1 (Vi ) = Ui × Cf ,
we see that ν|Ui is described by a mapping
Ui × Ce −→ Ui × Cf , (x, v) 7→ x, νi (x) · v ,

where νi (x) is an e × f matrix of holomorphic functions on Ui , or equivalently a family

νi (x) : Ce −→ Cf
of linear transformations varying holomorphically with x ∈ Ui . On the overlaps Uij = Ui ∩Uj ,
these must satisfy the compatibility condition
νi ◦ gij = hij ◦ νj . (2.12)
11This follows from the fact that if E and L are vector spaces of ranks e and 1 respectively, and if
g : E −→ E , h : L −→ L
are linear transformations, then the induced linear transformation E ⊗ L −→ E ⊗ L is given by h · g.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 37

Conversely, a collection of matrices {νi (x)} satisfying (2.12) determines a homorphism ν :

E −→ F.

We can now explicate when data {Ui , gij } and {Ui , gij0 } determine isomorphic bundles.
By the discussion of the previous paragraph, this happens if and only if there exist invertible
matrices
νi ∈ GLe (O(Uij ))
of holomorphic functions such that
gij = νi−1 ◦ gij0 ◦ νj
on Uij . In particular, the bundle determined by data {Ui , gij } is isomorphic to the trivial
bundle if and only if
gij = νi−1 ◦ νj .

Finally suppose that f : Y −→ X is a holomorphic mapping. Then there is a natural

way to pull back a holomorphic vector bundle p : E −→ X to arrive at a holomorphic
bundle f ∗ E −→ Y on Y . For example, if E is described by transition data {Ui , gij }, we
may describe f ∗ E by taking the open cover {f −1 (Ui )}, and using as transition matrices the
pull-backs f ∗ gij ∈ GLe (OY (f −1 (Uij )). Geometrically, if Ex = p−1 (x) denotes the fibre of E
over x, then f ∗ (E)y = Ef (y) .

Mappings to projective space. We close this section by discussing mappings to projective

space.

Let X be a complex manifold, and L a holomorphic lind bundle on X. Suppose given

global sections


s0 , . . . , sr ∈ Γ X, O(L) .
We ask that the si do not simultaneously vanish at any point x ∈ X. In other words
Given any x ∈ X, assume that there is at least one index α such that sα (x) 6= 0. (2.13)
We claim that then for any x ∈ X the expression
[s0 (x), . . . , sr (x)]
determines a well-defined point in Pr . In brief, we locally represent sα by a function fα in a
neighborhood of x on which L trivializes, and set
[s0 (x), . . . , sr (x)] =def [f0 (x), . . . , fr (x)]. (*)
The fα are not canonically defined, but thanks to (2.11) the vector arising from another
choice of local representatives differs from the given one by multiplication by a non-zero
scalar, so as a homogeneous vector the right-hand side of (*) is well-defined. With this
understood, it is immediate that the mapping
φ : X −→ Pr , x 7→ [s0 (x), . . . , sr (x)]
38 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

is holomorphic. Conversely, any holomorphic mapping φ : X −→ Pr arises in this manner.

In fact, writing Z0 , . . . , Zr for the homogeneous coordinates on Pr , one simply takes
L = φ∗ OPr (1) , sα = φ∗ (Zα ).

Given a compact complex manifold X, we can now begin to approach the question
whether X admits a holomorphic embedding into some projective space. In order to produce
such an embedding one needs to produce a holomorphic

line bundle L on X that in the first
place carries enough sections sα ∈ Γ X, O(L) so that (2.13) holds, and so that moreover
the resulting holomorphic mapping
φ : X −→ Pr
actually be an embedding. It is a remarkable fact that in many ways the most appealing
characterization of when this happens is differentio-geometric in nature. Specifically, one
starts by choosing a Hermitian metric on the fibres of L, which gives rise to a curvature form
Θ, which is a closed two-form representing the so-called first Chern class of L:12
[Θ] = c1 (L) ∈ H 2 (X, Z).
The first basic theorem – the Lefschetz (1, 1) Theorem (Theorem 6.29) – allows one under
suitable hypotheses to identifiy which cohomology classes arise in this fashion. The second
fundamental result – the Kodaira Embedding Theorem (Theorem 7.24) – states that if
Θ = Θ(L) satisfies a certain positivity hypothesis, then some large tensor power of L carries
enough holomorphic sections to define an embedding. Putting these two theorems together,
one arrives at a complete characterization of when a compact complex manifold admits a
projective embedding: it should carry a Kähler metric with rational periods (Exercise 7.7).

Our main goal in the rest of the class is to prove these results. This requires that we
study calculus on complex manifolds, to which we turn in the following sections.

Exercises for Section 2.

Exercise 2.1. Given complex manifolds X and Y , the product X × Y is a complex manifold
in such a way that the two projections
X × Y −→ X , X × Y −→ Y
are holomorphic.
Exercise 2.2. Let X be a complex manifold of dimension n ≥ 2, and fix a point x ∈ X.
Then any holomorphic function on X − {x} extends to a holomorphic function on all of X.

Exercise 2.3. Let U ⊆ Cn be an open set, and f = f (z) ∈ O(U ) a non-zero holomorphic
function. Denote by X ⊆ U the complement of the zero-locus of f :
X = U − {f = 0} ⊆ U.
Show that X is isomorphic to a closed submanifold of U × C. (Hint: consider the function
f (z) · t − 1 ∈ O(U )[t].)
12We’re ignoring some normalizing constants here
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 39

Exercise 2.4. Let Y be a complex manifold of dimension n + k. A subset X ⊆ Y is a closed

submanifold if and only if one can find an open covering {Uα } of Y , together with a k-tuple
{fα } of holomorphic functions on each Uα , such that
Uα ∩ X = Zeroes(fα ),
and J(fα ) has maximal rank on Uα ∩ X.
Exercise 2.5. (Hyperelliptic Riemann surfaces) Working in C2 with coordinates z, w,
let
X = {w2 − p2g+2 (z) = 0}
be the curve defined as the zero-locus of a polynomial w2 − p2g+2 (z) where p2g+2 (z) is a
polynomial of degree 2g + 2 in z with distinct roots.

(i). Prove that the condition of Example 2.6 is met, and hence that X is a complex
manifold of dimension 1.
(ii). By studying the projection
π : X −→ C , (z, w) 7→ z,
show that one can compactify X by adding two points “at infinity” to obtain a
compact Riemann surface X in such a way that π extends to a holomorphic map
π : X −→ S = C ∪ {∞}.
(Hint: Because p has even degree, the inverse image under π of the complement D∗
of a disk of very large radius consists of the disjoint union of two copies of D∗ . But
D∗ is a punctured neighborhood of the point at infinity, and therefore we need to
add two points to compactify X.)

It is instrucitve to think through why X is diffeomorphic to a compact surface of genus g.

When g ≥ 2, such a curve or Riemann surface is called hyperelliptic.

Exercise 2.6. (Linear groups) Keeping the notation of Example 2.11, let
det : GLn (C) −→ C∗
be the holomorphic function that takes a matrix to its determinant. Prove that the complex
Jacobian of det is non-zero at any invertible matrix A. (Reduce to the case A = Id.)
Conclude that
SLn (C) =def {A | det A = 1}
GLn (C), and hence a complex Lie group. The complex orthogonal and symplectic groups
can be treated similarly.
Exercise 2.7. (Covering spaces). Let
f : Y −→ X
be a covering space in the sense of topology.
40 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

(i), If X is a complex manifold, then Y carries a unique structure of a complex manifold

such f is holomorphic and the deck transformations of f are holomorphic isomor-
phisms.
(ii). Conversely, if Y is a complex manifold, and the deck transformations of f are holo-
morphic, then there is a unique complex structure on X that renders f a holomorphic
mapping.
Exercise 2.8. (Non-isomorphic tori). Let
√
Λ0 = Zn + −1 · Zn ⊆ Cn ,
and let Λ ⊆ Cn be a lattice spanned by 2n vectors λj ∈ Cn . Assume that the 2n2 complex
numbers appearing as coordinates of the λj are algebraically independent over Q. Then Λ0
and Λ determine non-isomorphic complex tori.
Exercise 2.9. Verify the assertions made in Example 2.16 concerning the construction of
the Hopf manifold. I.e. prove that
X = Cn − {0} /Z

is indeed a complex manifold in such a way that the quotient mapping

Cn − {0} −→ X
is holomorphic, and that X ≈ S 2n−1 × S 1 .
Exercise 2.10. Fill in the details in the proof of Proposition 2.21.
Exercise 2.11. Let K = Fq be a finite field with q elements. How many points does the
finite set Pn (K) contain?
Exercise 2.12. Verify pictorially that the fibres of the Hopf fibration h : S 3 −→ S 2 are
simply linked circles in S 3 .
Exercise 2.13. Let C ⊆ P2 be the plane curve
Y 2 Z − X 3 − X 2 Z = 0.
As in Example 2.26, draw the restrictions of C to each of the affine coordinate charts
UZ , UY , UX , and explain how the pictures fit together.
Exercise 2.14. Fill in the details of the proof of Proposition 2.30, including a proof of
Euler’s theorem. (For the latter, note that it suffices to check it for a single monomial.)
Exercise 2.15. Prove Proposition 2.31.
Exercise 2.16. Prove Proposition 2.34, that the Segre embedding realizes Pr × Ps as the
sub manifold of Prs+r+s arising as matrices of rank 1. (Write out what σr,s looks like on the
product of the standard coordinate patches in Pr and Ps .)
Exercise 2.17. (Algebraic subsets of Pr × Ps ).

(i). Prove that a subset Z ⊆ Pr × Ps is the common zero-locus of a collection of biho-

mogeneous polynomials if and only if it is the intersection of the Segre variety with
an algebraic subset of Prs+r+s . (Note that a homogenous polynomial of degree d
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 41

on Prs+r+s pulls back under the Segre mapping to a bihomogeneous polynomial of

bidegree (d, d), so one direction is clear.)
(ii). Find bihomogeneous polynomials defining the diagonal ∆ ⊆ Pr × Pr .
Exercise 2.18. Let M be a smooth manifold of dimension n, and T M the tangent bundle
of M . What are the transition matrices gij of T M ?
Exercise 2.19. Verify that the transition functions for the hyperplane line bundle on Pn
constructed in Example 2.41 are given as stated in (2.10)
Exercise 2.20. (A locally trivial affine bundle). Working in P1 × P1 , denote by
B ⊆ P1 × P1 the complement of the diagonal:
B = P1 × P1 − ∆,

and consider (either of) the projections

p : B −→ P1 .
Taking the standard open subsets U0 , U1 ⊆ P1 , show that there are isomorphisms
∼
=
φi : p−1 (Ui ) −→ Ui × C
commuting with the projections to Ui , but that p cannot be the total space of a line bundle
over P1 , i.e. one cannot arrange for the comparison maps to be given by linear isomor-
phisms.13 (In fact, observe that B is isomorphic via the Segre to a sumanifold of C3 , and
deduce that p does not admit any sections.)
Exercise 2.21. Prove that there is a surjective map of vector bundles
1n+1 −→ OPn (1)
of vector bundles on Pn .
Exercise 2.22. If E −→ X is a vector bundle described by transition data {Ui , gij }, find
the transition matrices of the dual bundle E∗ −→ X.
Exercise 2.23. Prove Proposition 2.47, that sections of OPn (k) are given by homogeneous
polynomials of degree k.
Exercise 2.24. (Projective bundles). Let p : E −→ X be a vector bundle, say described
by transition data {Ui , gij }. Then one can form the projective bundle
π : P(E) −→ X,
whose fibre over x is the projective space of one-dimensional subspaces of Ex = p−1 (x). This
may be constructed by starting with the products Ui × Pe−1 , and using the gij – viewed as
defining automorphisms of Pe−1 – to glue them along Uij × Pe−1 .

(i). Show that if L is a line bundle, then

P(E ⊗ L) ∼
= P(E).

13Rather, they are given by affine isomorphisms C −→ C.

42 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

(ii). Show that there is a “tautological” line sub-bundle

OP(E) (−1) ⊆ π ∗ (E) :
if we view a point in P(E) as the choice of a point x ∈ X together with a one-
dimension subspace Λ of the fibre Ex = p−1 (x) then the fibre of OP(E) (−1) at (x, Λ)
is Λ. What are the transition functions of this line bundle?

3. Complex and Hermitian Linear Algebra

In this section we discuss the linear algebra underlying calculus on complex manifolds.
We follow closely the presentation of Huybrechts in [3, Chapter 1.2].

Almost complex structures on a vector space. Let V be a finite-dimensional real

vector space.
Definition 3.1. An almost complex structure on V is an R-linear endomorphism
J : V −→ V such that J 2 = Id.

Note that giving such a J is equivalent to equipping V with the structure of a complex
vector space by taking √
−1 · v = J(v).
In particular, the existence of an almost complex structure on V implies that dimR V is
even, say
dimR V = 2n.
Remark 3.2. Suppose that V carries an almost complex structure J. Then one can find
vectors xi ∈ V such that the vectors

xi , yi =def J(xi )
form a basis for V (Exercise 3.1). With respect to this basis J is given by the block matrix
0 −1
 
1 0 
0 −1
 
 
1 0
 
 
...

Given a real vector space V with dimR V = m, set

VC =def V ⊗R C.
This is naturally a complex vector space of complex dimension m, which contains V as a
(real) subspace. It is useful (if slightly abusive) to think of elements of VC as having the
form √
v + −1 · w for v, w ∈ V , (3.1)
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 43

with the obvious expression for scalar multiplication by a complex number. Note that there
is an operation of complex conjugation on VC , given by the rule
√ √
v + −1 · w = v − −1 · w.
The real subspace V ⊆ VC consists of those vectors invariant under complex conjugation.

Now suppose that J : V −→ V is an almost complex structure. Then J extends to a

C-linear map
√ √
JC : VC −→ VC , JC (v + −1 · w) = J(v) + −1 · J(w).
By abuse of notation we will sometimes write simply J in place of JC .
√
The eigenvalues of this transformation are ± −1, and the corresponding eigenspaces
play a crucial role in the story:
Definition 3.3. We set
√
V 1,0 =

µ ∈ VC | Jµ = −1µ
√
V 0,1 =

µ ∈ VC | Jµ = − −1µ
These are complex subspaces of VC .
Lemma 3.4. There is a canonical direct sum decomposition:
VC = V 1,0 ⊕ V 0,1
of complex vector spaces. The corresponding projections
π 1,0 : VC −→ V 1,0 , π 0,1 : VC −→ V 0,1
are given by
√
π 1,0 (µ) = 1


2
µ− −1 · J(µ)
√
π 0,1 (µ) = 1


2
µ + −1 · J(µ) .
Finally, complex conjugation defines an R-linear isomorphism V 1,0 ∼
=R V 0,1 , ie
V 1,0 = V 0,1 .

Again we leave the proof for the reader (Exercise 3.2).

Note that JC : VC −→ VC satisfies JC2 = Id, and hence defines a complex structure on
JC . However this is not the complex structure defined √ via (3.1), which is the one we always
use. In fact, by definition JC acts
√ by multiplication by −1 on the subspace V 1,0 ⊆ VC , but
it acts by multiplication by − −1 on V 0,1 ⊆ VC . In fact:
Proposition 3.5. The composition
π 1,0
V ⊆ VC −→ V 1,0
gives rise to a C-linear isomorphism
√
(V, J) ∼
= (V 1,0 , · −1)
44 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

of complex vector spaces. On the other hand, the composition

π 0,1
V ⊆ VC −→ V 0,1
√
determines a complex antilinear isomorphism between (V, J) and (V 0,1 , · −1).

It is useful to have explicit descriptions of bases. As in Remark 3.2, consider a basis of

V of the form {xi , yi = J(xi )}. Then
√
zi =def 12 xi − −1 · yi ∈ V 1,0


√ (3.2)
z i =def 12 xi + −1 · yi ∈ V 0,1

are C-bases of the two spaces in question, and in particular {zi , z i } is a complex basis for
VC .14

We now turn to forms. Given a real vector space V , write as usual

V ∗ = HomR (V, R)
for the dual space of linear functionals on V .√Put (V ∗ )C = V ∗ ⊗R C. We may view elements
of this space as functionals of the form f + −1 · g for f, g ∈ V ∗ , and in particular one has
identifications
(V ∗ )C = HomR (V, C) = HomC (VC , C) = VC∗ .

Now suppose that J : V −→ V is an almost complex structure on V . then J induces an

almost complex structure
J ∗ : V ∗ −→ V ∗ , J ∗ (f )(v) = f J(v) .

Then as before we get an eigenspace decomposition (V ∗ )C = (V ∗ )1,0 ⊕ (V ∗ )0,1 , where

√
(V ∗ )1,0 = f ∈ HomR (V, C) | f (Jv) = −1 · f (v)

√ (3.3)
(V ∗ )0,1 = f ∈ HomR (V, C) | f (Jv) = − −1 · f (v)


Note that
(V ∗ )1,0 ∼
= HomC−lin (V, J), C) , (V ∗ )0,1 ∼
= HomC−antilin (V, J), C).
If we choose a basis {xi , yi = J(yi )} for V as above, and denot by xi , y i the corresponding
dual basis of V ∗ , then
√ √
z i = xi + −1y i , z i = xi − −1y i (3.4)
are bases for (V ∗ )1,0 and (V ∗ )0,1 respectively, dual to the bases {zi } and {z i } appearing in
(3.2).
14In
effect we have seen these expressions before. If we think of V as the (real) tangent space to a complex
manifold, then (3.2) corresponds to the definitions
∂ 1 ∂ √ ∂
∂ 1 ∂ √ ∂

= − −1 , = + −1 .
∂zi 2 ∂xi ∂yi ∂z i 2 ∂xi ∂yi
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 45

Next we turn to exterior products. Goven as before a real vector space V with dimR V =
2n, we consider the real and complex vector spaces
Λk V , Λk VC ,
of real and complex dimensions 2n


k
respectively. Now suppose that J : V −→ V is an
almost complex structure, giving rise to the decomposition
VC = V 1.0 ⊕ V 0,1 .
Then there is a canonical isomorphism:
Λk VC = Λk V 1,0 ⊕ V 0,1


M
Λp (V 1,0 ) ⊗ Λq (V 0,1 )


=
p+q=k

as complex vector spaces. We define

Λp,q V = V p,q = Λp (V 1,0 ) ⊗ Λq (V 0,1 ), (3.5)
so that M
Λk V = V p,q . (3.6)
p+q=k

Explicitly, if
α1 , . . . , αp ∈ V 1,0 and β1 , . . . , βq ∈ V 0,1 ,
then
α1 ∧ . . . ∧ αp ∧ β1 ∧ . . . ∧ βq ∈ V p,q ,
and an arbitrary element of V p,q is a C-linear combination of terms of this form. As for the
decomposition (3.6), starting with µ1 , . . . , µk ∈ VC , start by writing
µi = µ1,0
i + µ0,1
i

as a sum of vectors of types (1, 0) and (0, 1). Then expand out the wedge product
µ1 ∧ . . . ∧ µk = (µ11,0 + µ0,1 1,0 0,1
1 ) ∧ . . . ∧ (µk + µk )

via linearity, and for each (p, q) with p + q = k, collect together those terms that are wedges
of p vectors of type (1, 0) and q vectors of type (0, 1). Note that
V p,q = V q,p .
In terms of the basis {z i , z i } appearing in (3.4), (V ∗ )pq has as basis the elements
z i1 ∧ . . . ∧ z ip ∧ z j1 ∧ . . . ∧ z jq
with i1 < . . . < ip , j1 < . . . < jq . Finally, consider the endomorphism
Λk JC : Λk VC −→ Λk VC
determined by J. Then the reader should verify that:
√
Λk JC acts on Λp,q V by multiplication by ( −1)p−q . (3.7)
46 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Metrics and the fundamental form. We now introduce metrics. Focusing still on a
finite-dimensional real vector space V , assume now that we are given a positive definite
inner product < , > on V .
Definition 3.6. An almost complex structure J on V is compatible with < , > if
< Jv, Jw > = < v, w > for all v, w ∈ V.

In other words, we ask that J ∈ O(V, < , >).

Note that a compatible complex structure J on (V, < , >) determines an orientation on
V : choosing vectors x1 , . . . , xn ∈ V such that
{x1 , Jx1 , . . . , xn , Jxn } (*)
is a basis of V , we require that the ordered basis (*) determine the positive orientation of V .
Example 3.7. It is instructive to consider the case dimR (V ) = 2. If we fix an orientation
on V , then a Euclidean structure < , > determines a unique compatible almost complex
structure. In fact, choose any non-zero vector v ∈ V , and define J by requiring that
< v, Jv >= 0 , ||v|| = ||Jv|| = 1 , J 2 (v) = −v
and that {v, Jv} be a positively oriented basis of V . Conversely, if J is an almost complex
structure on V , then up to scalar multiples there is a unique Euclidean inner product < , >
on V with respect to which J is compatible and {v, Jv} gives a positive orientation. See
also Exercise 3.3.

Fix a Euclidean space (V, < , >, J) with a compatible almost complex structure. These
data are combined in the crucial
Definition 3.8. The fundamental form associated to < , > and J is the two-form ω ∈ Λ2 V ∗
defined by
ω(v, w) = < Jv, w > = − < v, Jw > .
Proposition 3.9. The form ω is in fact alternating, and under the inclusion
Λ2 V ∗ ⊆ Λ2 VC∗ ,
ω has type (1, 1), ie:
ω ∈ Λ2 V ∗ ∩ Λ1,1 V ∗ .

Proof. One has

ω(w, v) = < Jw, v > = < J 2 w, Jv > = < −w, Jv > = − < Jv, w > = ω(v, w),
so ω is alternating. For the second statement, observe that ω has type (1, 1) if and only if
ω(Jv, Jw) = ω(v, w) (Exercise 3.4). But this is clear, since:
ω(Jv, Jw) = < J 2 v, Jw > = < Jv, w > = ω(v, w)
thanks to the fact that J is compatible with < , >. 
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 47

We now discuss Hermitian forms arising from the data (V, J, < , >) of an almost complex
structure compatible with a Euclidean inner product. Define
√
h(v, w) = < v, w > − −1 · ω(v, w)
√ (3.8)
= < v, w > − −1· < Jv, w > .

This is a Hermitian form on V with respect to the complex structure given by J.15 For
another viewpoint, denote by < , >C−lin the C-linear extension of < , > to VC . Then we can
extend < , > to a Hermitian inner product < , >herm on VC by setting
< µ, ν >herm = < µ, ν >C−lin .
√
Then under the isomorphism (V, J) ∼
= (V 1,0 , · −1) given by Proposition 3.5, one has
1
2
· h( , ) = < , >herm |V 1,0 . (3.9)

It is convenient to write these definitions explicitly in terms of coordinates. Choose a

basis {xi , yi = J(xi )} for V as before, and set
hij = h(xi , xj ).
Then (hij ) is an Hermitian matrix, and
√ √
h(yi , yj ) = hij , h(xi , yj ) = − −1 · hij , h(yi , xj ) = −1 · hij . (3.10)
Moreover, by definition
ω = −Im(h) , < , > = Re(h),
and consequently
ω(xi , xj ) = ω(yi , yj ) = −Im(hij )
(3.11)
ω(xi , yj ) = Re(hij ),
and
< xi , xj > = < yi , yj > = Re(hij )
(3.12)
< xi , yj > = Im(hij ).
In particular, writing xi , y i for the basis dual to xi , yi , it follows that
X n
X
i j i j
Re(hij ) · xi ∧ y j .


ω = − Im(hij ) · x ∧ x + y ∧ y + (3.13)
i<j i,j=1
√ √
Finally, as in (3.2), put zi = 12 (xi − −1yi ) and z i = 12 (xi + −1yi ), with dual basis z i , z i .
Then (3.11) leads to the basic expression
√ n
−1 X
ω = · hij · z i ∧ z j . (3.14)
2 i,j=1

15We take Hermitian forms to be C-linear in the first argument, and conjugate linear in the second.
48 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

The Hodge ∗-operator. Before proceeding, we pause to recall some facts about the Hodge
∗-operator. Let W be a real vector space of dimension d equipped with a positive definite
inner product g =< , >. Giving < , > is the same thing as giving a symmetric linear mapping
T : W −→ W ∗ , T (v)(w) = < v, w >
with the property that T (v)(v) > 0 for all 0 6= v ∈ W . On the other hand, T determines by
fuctoriality a symmetric mapping
Λk T : Λk W −→ Λk W ∗
satisfying the analogous positivity property. In other words, an inner product < , > on W
induces one on each exterior product Λk W , which we continue to denote by < , >. Concretely,
if
e1 , . . . , ed ∈ W (*)
is an orthonormal basis for W , then < , > is defined on Λk W by taking as an orthonormal
basis the multi-vectors
eI = ei1 ∧ . . . ∧ eik for I = {i1 < . . . < ik }.
We also fix the volume form
vol = e1 ∧ . . . ∧ ed ∈ Λd W,
arising from the orthonormal basis (*).

There is a natural non-degenerate pairing

Λk W ⊗ Λd−k W −→ Λd W , α ⊗ β 7→ α ∧ β,
and we have fixed a basis vol for the one-dimensional space on the right. We can then define
the Hodge ∗-operator
∗ : Λk W −→ Λd−k W , β 7→ ∗β
by requiring that
α ∧ ∗β = < α, β > · vol (3.15)
for every α, β ∈ Λk W . This extends by linearity to a map
∗ : Λ∗ W −→ Λ∗ W.

The basic properties of this operator are summarized in the following Proposition, whose
proof we leave to the reader.
Proposition 3.10. Fix as above an orthonormal basis e1 , . . . , ed ∈ W .

(i). If I = {i1 < . . . < ik } and J = {j1 < . . . < jd−k } are disjoint multi-indices such that
I ∪ J = [1, d], then
∗ eI = σI,J · eJ ,
where eI = ei1 ∧ . . . ∧ eik , with eJ defined similarly, and σI,J is the sign of the
permutation (I, J).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 49

(ii). Given α ∈ Λk W, β ∈ Λd−k W , one has

< α, ∗β > = (−1)k(d−k) < ∗α, β > .

(iii). If α ∈ Λk W , then
∗ ∗ α = (−1)k(d−k) · α.

(iv). The operator ∗ is an isometry of Λ∗ W, < , > .

Now let (V, < , >, J) be as above a real Euclidean space of dimension 2n with a compat-
ible almost complex structure. We apply the previous discussion to W = V ∗ , which inherits
an inner product < , > and orientation from V and J. In particular, one then gets an inner
product < , > on Λ∗ V ∗ , and the Hodge ∗-operator
∗ : Λ∗ V ∗ −→ Λ∗ V ∗ (3.16)
is defined. Moreover, the inner product < , > on V ∗ extends to a positive definite Hermitian
form < , >herm on VC∗ , which then determines a positive definite Hermitian inner product
< , >herm on Λ∗ VC∗ . Thanks to the fact (Exercise 3.5) that the subspaces
V ∗1,0 , V ∗0,1 ⊆ VC∗
are orthogonal with respect to < , >herm , one finds that
Λk V ∗ = ⊕p+q=k Λp,q VC∗
is an orthogonal decomposition with respect to < , >herm . Finally, the Hodge ∗-operator
(3.16) extends C-linearly to
∗ : Λk VC∗ −→ Λ2n−k VC∗ ,
and satisfies
α ∧ ∗β = < α, β >herm ·vol. (3.17)
p0 ,q 0
Observe that if β ∈ Λp,q V ∗ and α ∈ Λ V ∗ , then
< α, β >herm = 0 unless p0 = q , q 0 = p.
It follows that
∗ Λp,q VC∗ ⊆ Λn−q,n−p VC∗ .

(Note the switch in indices!)

Example 3.11. Let {xi , yi = Jxi } be an orthonormal basis of V , and let {z i , z i } be the
corresponding bases of Λ1,0 VC∗ and Λ0,1 VC∗ . Given disjoint subsets I, J, K ⊆ [1, n], consider a
form of the type
α = zI ∧ zI ∧ zJ ∧ zK .
Then
∗α = (constant) · z L ∧ z L ∧ z J ∧ z K ,
where L is the complement of I ∪ J ∪ K in [1, n] (Exercise 3.9).
50 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Example 3.12. One has ∗1 = vol, ∗vol = 1, and

ωj ω n−j
∗ =
j! (n − j)!
1
(Exercise 3.9). In particular, this shows again that vol = n!
· ω n (Exercise ??).

Lefschetz operator and decomposition. Consider once again a triple (V, < , >, J) con-
sisting of a real Euclidean space of dimension 2n and a compatible almost complex structure.
Denote by
ω ∈ Λ2 V ∗
the corresponding fundamental form.
Definition 3.13. The Lefschetz operator L associated to ω is the operator given by wedge
product with ω:
L : Λk V ∗ −→ Λk+2 V ∗ , α 7→ ω ∧ α.
The dual or adjoint Lefschetz operator is the operator
Λ : Λk+2 V ∗ −→ Λk V ∗
determined by requiring that
< Λα, β > = < α, Lβ > for all β ∈ Λk V ∗ .

We sometimes view L and its adjoint Λ as operators on the full exterior algebra of V :
L : Λ∗ V ∗ −→ Λ∗ V ∗ , Λ : Λ∗ V ∗ −→ Λ∗ V ∗ .

Note that it follows from Proposition 3.10 that the Hodge ∗-operator is invertible. We
can then use it to express Λ in terms of L:
Proposition 3.14. One has
Λ = ∗−1 ◦ L ◦ ∗.

Proof. Since
< Λα, β > = < α, Lβ > = < Lβ, α >,
the issue is to show that if α ∈ Λk+2 V ∗ and β ∈ Λk V ∗ , then
< Lβ, α > ·vol = < β, ∗−1 L ∗ α > ·vol.
But
< β, ∗−1 L ∗ α > ·vol = β ∧ L(∗α)
= β ∧ ω ∧ ∗α
= ω ∧ β ∧ ∗α
= < Lβ, α > ·vol,
as required. 
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 51

Passing to VC , one extends L and Λ by C-linearity to

L : Λk VC∗ −→ Λk+2 VC∗ , Λ : Λk+2 VC∗ −→ Λk VC∗ .
These are adjoint with respect to < , >herm , ie
< Λα, β >herm = < α, Lβ >herm ,
and the equality in Proposition 3.14 remains valid. Moreover, as ω ∈ Λ1,1 V ∗ , one finds that
L (V ∗ )pq ⊆ (V ∗ )p+1,q+1 , Λ (V ∗ )p+1,q+1 ⊆ (V ∗ )pq .

We now come to the Lefschetz decomposition. The first point is to define primitive
forms:
Definition 3.15. (Primitive vectors). A k-vector α ∈ Λk V ∗ is primitive if
Λ(α) = 0.
We denote by P k ⊆ Λk V ∗ the subspace of primitive elements.

We will see shortly that α ∈ Λk V ∗ is primitive if and only if k ≤ n and Ln−k+1 (α) = 0. One
uses the same definition to define the subspace of complex primitive forms PCk ⊆ Λk VC∗ .

The basic result is then

Theorem 3.16. As before let (V, < , >, J) be a Euclidean real vector space of dimension 2n
together with a compatible almost complex structure.

(i). One can express Λk V ∗ as a < , >-orthoginal direct sum

M
Λk V ∗ = Li P k−2i .
i≥0
k ∗
In other words, any α ∈ Λ V can be expressed uniquely as a sum
k
α = αk + Lαk−2 + L2 αk−4 + . . . + L[ 2 ] α k ,
k−2[ ]
2

where the αj ∈ P j are primitive forms.

(ii). The mapping
Ln−k : Λk V ∗ −→ Λ2n−k V ∗
is an isomorphism.
(iii). If k > n then P k = 0, while if k ≤ n then
P k = {α ∈ Λk V ∗ | Ln−k+1 α = 0}.

The statement in (i) is the Lefschetz decomposition. The assertion in (ii) will later on lead
to the Hard Lefschetx theorem. The next few pages of this subsection will be devoted to the
proof of the Theorem.
52 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Example 3.17. When k = 2 the Lefschetz decomposition is

Λ2 V ∗ = P 2 ⊕ R · ω.
For k = 3, it takes the form Λ3 V ∗ = P 3 ⊕ ω · V ∗ .
Remark 3.18. In the global setting, when we are dealing with differential forms on a
projective manifold, the operator L corresponds to cup product with the cohomology class
of a hyperplane. The primitive forms then represent the cohomology classes that don’t “come
from” a linear space section of the given variety.

It is nowadays traditional to prove Theorem 3.16 by realizing Λk V ∗ as a representation

of the Lie algebra sl2 . To this end, we introduce a counting operator H on Λ∗ V ∗ , and work
out the commutation relations among L, Λ and H.
Definition 3.19. Define
H : Λ∗ V −→ Λ∗ V
to be the operator that acts on the subspace Λk V ∗ by multiplication by (k − n), ie
2n
X
H = (k − n) · π k ,
k=1
k ∗ ∗ k ∗
where π : Λ V −→ Λ V is the projection.

We extend this C-linearly to an operator on Λ∗ VC∗ .

We now have three operators

H , L , Λ : Λ∗ V ∗ −→ Λ∗ V ∗ .
The following very basic result gives the commutation relations among these.
Proposition 3.20. One has
[H, L] = 2L , [H, Λ] = −2Λ
[L, Λ] = H.

Proof. For the first statement, let α ∈ Λk V ∗ . Then Lα ∈ Λk+2 V ∗ , and so

HLα = (k + 2 − n) · Lα , LHα = (k − n) · Lα,
whence [H, L]α = 2Lα. Similarly, [L, Λ] = −2Λ. To show that [L, Λ] = H, one proceeds by
induction on dim V . If dimR V = 2, so that n = 1, then L1 = ω, Λω = 1, and the assertion
is clear. Now suppose that we write (V, < , >, J) as a direct sum
(V, < , >, J) = (V1 , < , >1 , J1 ) ⊕ (V2 , < , >2 , J2 ).
of proper subspaces of real dimensions 2n1 and 2n2 that are preserved by J and are orthogonal
for < , >. Write ωi , Li , Λi for the indicated data on Λ∗ Vi , and recall that
M
Λk V ∗ = Λk1 V1∗ ⊗ Λk2 V2∗ .
k1 +k2 =k
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 53

Then to begin with

ω = ω1 + ω2 ∈ Λ2 V1∗ ⊕ Λ2 V2∗ ⊆ Λ2 V1 ⊕ V2 ,

and so L acts on Λk1 V1∗ ⊗ Λk2 V2∗ via

L = L1 ⊗ 1 + 1 ⊗ L2 . (*)
We claim that similarly
Λ = Λ 1 ⊗ 1 + 1 ⊗ Λ2 . (**)
In fact, temporarily denote the right hand side of (**) by Λ0 . If α = α1 ⊗ α2 and β = β1 ⊗ β2 ,
then
< α, β > = < α1 , β1 >1 · < α2 , β2 >2 ,
and then a computation using (*) shows that
< α, Lβ > = < Λ0 α, β >,
which establishes (**). It follows that [L, Λ] acts on Λk1 V1∗ ⊗ Λk2 V2∗ via
[L1 , Λ1 ] ⊗ 1 + 1 ⊗ [L2 , Λ2 ].
But by induction [Li , Λi ] = Hi , and hence
[L, Λ](α1 ⊗ α2 ) = (k1 − n1 ) · α1 ⊗ α2 + (k2 − n2 ) · α1 ⊗ α2


= (k1 + k2 ) − (n2 + n2 ) · α1 ⊗ α2 ,
as required. 

We next review the structure of representations of sl2 : this will lead to the proof of
Theorem 3.16. Recall that sl2 is the Lie algebra of all 2 × 2 matrices of trace 0; we will work
with real matrices and representations, but nothing changes if one complexifies. Put
     
0 1 1 0 0 0
X = , B = , Y = .
0 0 0 −1 1 0
These satisfy the commutation relations
[B, X] = 2X , [B, Y ] = −2Y , [X, Y ] = B.
Lemma 3.21. Let W be an irreducible finite-dimensional representation of sl2 . Then there
exists a non-negative integer m such that W is the direct sum
W = W−m ⊕ W−m+2 ⊕ . . . ⊕ Wm−2 ⊕ Wm ,
where Wj ⊆ W is a one-dimensional eigenspace for B, with eigenvalue j. Moreover, Y maps
Wj onto Wj−2 and X maps Wj onto Wj+2 . In particular, any vector u ∈ W can be written
uniquely in the form
u = u0 + Xu1 + X 2 u2 + . . . + X m um
where Y uj = 0 for each j.
54 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Proof. Let v ∈ W be an eigenvector of B with eigenvalue λ.16 Then

B(Xv) = [B, X](v) + XB(v)
= (2 + λ)X(v),
so Xv is an eigenvector of B with eigenvaule λ + 2. Similarly, Y v is an eigenvector with
eigenvalue λ − 2. It follows that we can find a primitive vector w ∈ W , ie an eigenvector of
B such that Y w = 0. There also exists an integer n such that
X m (w) 6= 0 , X m+1 (w) = 0.
Now the elements X j w (j ≥ 0) span a sub-representation of W , and they are linearly
independent since they are eigenvectors of B with distinct eigenvalues. It follows that
M m
W = R · X j (w).
j=0

So it suffices to show that the B-eigenvalue of w is −m. To this end, suppose that Bw = λw.
Then
Y Xw = XY w − Bw = −λw,
and
Y X 2 w = XY Xw − BXw
= −λXw − (λ + 2)Xw.
In general


Y X kw = − λ − (λ + 2) − . . . − λ + (2(k − 1) (X k−1 w)
= (−kλ − k 2 + k)X k−1 w.
Since X m+1 w = 0 but X m w 6= 0, this implies that
−(m + 1)λ − (m + 1)2 + (m + 1) = 0,
ie λ = −m. 
2
Remark 3.22. Let C denote the standard two-dimensional representation of sl2 . Then the
representation W considered in the previous lemma arises as Symm C2 .
Corollary 3.23. Let U be a finite dimensional representation of sl2 , and let
P = {u ∈ U | Y u = 0} ⊆ U
be the subspace of primitive vectors. Assume that B acts on P by multiplication by −m.
Then m ≥ 0, and
U = P ⊕ X(P ) ⊕ X 2 (P ) . . . ⊕ X m (P ).
Moreover X m+1 (P ) = 0, while
X m (u) 6= 0 ∀ 0 6= u ∈ P.

Proof. In fact, U is a direct sum of irreducible representations, and then the corollary follows
from the previous lemma and its proof. 
16It will emerge shortly that B actually has real eigenvalues.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 55

We now give the

Proof of Theorem 3.16. Proposition 3.20 asserts that Λ∗ V ∗ is a representation of sl2 , with
L ↔ X , H ↔ B , Λ ↔ Y.
Furthermore, by definition H acts as multiplication by k−n on P k V ∗ ⊆ Λk V ∗ . The existence
of the Lefschetz decomposition then follows from the previous corollary as does statement
(iii) of the Theorem. Moreover the corollary shows that the mapping
Ln−k : P k V ∗ −→ Λn−k V ∗
is injective. It then follows from (i) that
Ln−k : Λk V ∗ −→ Λ2n−k V ∗
is injective, and since both sides have the same dimension Ln−k must be an isomorphism. 

Note that the operators Ln−k−j and ∗Lj both map Λk V ∗ to Λ2n−k−j V ∗ , so it is natural
to ask for a comparison of these. To this end, write
J = Λ∗ J ∗ : Λ∗ V ∗ −→ Λ∗ V ∗
for the endomorphism of Λ∗ V ∗ determined by the almost complex structure J.
Proposition 3.24. Let α ∈ P k be a primitive k-form. Then
k(k+1) j!
∗Lj (α) = (−1) 2 · · Ln−k−j J(α).
(n − k − j)!

The Proposition is proven using the Lefschetz decomposition by an induction on dim V , in

the spirit of the proof of Proposition 3.20. However the calculations are somewhat involved,
and we will not repeat them here. We refer the interested reader to [3, pages 37-38].
Example 3.25. Take α = 1 ∈ Λ0 V ∗ (so k = 0) and j = 0. Then the Proposition asserts
j ω n−j
that ∗ ωj! = (n−j)! , as observed in Example 3.12.

The operators L, Λ, H pass by linearity to the complexification Λ∗ VC∗ , and all the results
of this subsection remain valid in this setting. Moreover since these operators act bihomo-
geneously with respect to the (p, q) grading on Λ∗ VC∗ , the Lefschetz decomposition respects
this bigrading. Specifically, let
P p,q = PCk ∩ Λp,q VC∗ .
Then PCk = ⊕p+q=k P p,q , and
M
Λp,q VC∗ = Lj P p−i,q−i .
i≥0

Moreover, since L is real:

P p,q = P q,p .
√
Finally, recall that if α ∈ P p,q then J(α) = ( −1)p−q ·α, and hence Proposition 3.24 assumes
a particularly simple shape in this case.
56 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Example 3.26. Taking k = 2, note that Λ2,0 V ∗ and Λ2,0 V ∗ are primitive for reasons of
type. The Lefschetz decomposition in this case takes the form
Λ2 VC∗ = P 2,0 ⊕ P 1,1 ⊕ C · ω ⊕ P 0,2 .

We conclude by studying the sign of an important quadractic form defined with the help
of ω. Define
Q : Λk V ∗ × Λk V ∗ −→ R
by requiring that
k(k+1)
(−1) 2 α ∧ β ∧ ω n−k = Q(α, β) · vol.
We extend this by linearity to a symmetric bilinear form
Q : Λk VC∗ × Λk VC∗ −→ C,
called the Hodge-Riemann pairing. The basic fact is:
0 0
Theorem 3.27. (Hodge-Riemann bilinear relations). If α ∈ Λp,q V ∗ and β ∈ Λp ,q V ∗ ,
then
Q(α, β) = 0 unless p = q 0 , q = p0 .
If p + q ≤ n and α ∈ P p,q is a primitive (p, q)-form, then
√
( −1)(p−q) · Q(α, α) = n − (p + q) !· < α, α >herm .


(*)
√
In particular ( −1)(p−q) Q(α, α) > 0 if α 6= 0.

Proof. Only (*) needs proof, for which we apply Proposition 3.24 with j = 0 and k = p + q
to the primitive form (α) ∈ P q,p . One finds that
k(k+1) 1 √
∗α = (−1) 2 · · ( −1)q−p · α ∧ ω n−k ,
(n − k)!
so that
k(k+1) √
α ∧ ω n−k = (−1) 2 · (n − k)! · ( −1)p−q · ∗α.
Recalling that
< α, α >herm ·vol = α ∧ ∗α,
(*) follows. 
Example 3.28. Take k = 2, and consider the form
Q : Λ1,1 V ∗ × Λ1,1 V ∗ −→ R , (α, β) 7→ α ∧ β ∧ ω n−2
(where we identity Λn V ∗ with R via vol). Recall that
Λ1,1 V ∗ = P 1,1 ⊕ (R · ω).
Theorem 3.27 shows that Q is negative definite on P 1,1 , but positive-definite on the one-
dimensional subspace spanned by ω. So on Λ1,1 V ∗ Q has one positive eigenvalue, and the
rest negative. This will lead to the Hodge Index theorem in the global setting.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 57

Example 3.29. Suppose that dimC V = n = 2m is even, and let

P = P m,m = {α ∈ Λm,m V ∗ | α ∧ ω = 0}.
Then the quadratic form
P × P −→ R , (α, β) 7→ α ∧ β
is positive definite if m is even, and negative definite if m is odd.

Exercises for Section 3.

Exercise 3.1. Prove the existence of the vectors x1 , . . . , xn ∈ V asserted in Remark 3.2 .
Exercise 3.2. Prove Lemma 3.4.
Exercise 3.3. Let (V, < , >) be an oriented real Euclidean space with dimR = 4. Show
that the possible compatible complex structures that determine the given orientation are
parametrized by a copy of the two-dimensional sphere S 2 . (Hint: Think of R4 as the
quaternions.)
Exercise 3.4. Prove that a two-form η ∈ Λ2 V ∗ is of type (1, 1) in Λ2 VC∗ if and only if
η(Jv, Jw) = η(v, w) for all v, w ∈ V . (Hint: see (3.7).)
Exercise 3.5. Prove that with respect to the Hermitian form < , >herm , VC is the orthogonal
direct sum of V 1,0 and V 0,1 .
Exercise 3.6. Prove the expression for ω appearing in equation (3.14).
Exercise 3.7. Show that one can always choose an basis {xi , yi = Jxi } which is orthonormal
with respect to the form < , >. Then show that with respect to this basis,
√ n n
−1 X i i
X
ω = · z ∧z = xi ∧ y i .
2 i=1 i=1

Exercise 3.8. With the usual notation, show that

1
vol = · ωn.
n!
Exercise 3.9. Prove the assertions of Example 3.11, and Example ??
Exercise 3.10. Show that if α ∈ Λk V ∗ , then
[Li , Λ](α) = i(k − n + i − 1) · α.
(Hint: Argue by induction on i using Proposition 3.20 together with the identity
[Li , Λ] = L[Li−1 , Λ] + [L, Λ]Li−1 .

4. Calculus on Complex Manifolds, I: Local Theory

We now apply the linear algebra of the previous section to the tangent and cotangent
spaces of a complex manifold. We start with the local picture, where we can make explicit
calculations.
58 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Differential forms on open subsets of Cn . Let U ⊆ Cn be an open set, and let

√
zi = xi + −1 · yi
be complex coordinates on U . Then for each point p ∈ U , the (real) tangent space Tp U has
a natural almost complex structure J : Tp U −→ Tp U . Concretely, we take as a basis of Tp U
the tangent vectors
∂ ∂ ∂ ∂
,..., , ,..., , (*)
∂x1 ∂xn ∂y1 ∂yn
and then J is given by
∂ ∂ ∂ ∂
7→ , 7→ − .
∂xi ∂yi ∂yi ∂xi
Thus on the dual basis dxi , dyi , the transpose of J – which we again write simply as J –
acts by
J(dxi ) = −dyi , J(dyi ) = dxi .
Write T U for the real tangent bundle of U . We view the vector fields (*) as giving a
trivialization of this bundle over U :
T U = U × R2n ,
and we can think of J as defining an endomorphism of T U .

We now apply the discussion of the previous section. Denote by

TC U = T U ⊗R C
the complexified tangent bundle of U , whose fibres are the complexified tangent spaces
(Tp U )C . Then we can decompose TC U as a sum of two complex sub-bundles
TC U = T 1,0 U ⊕ T 0,1 U,
√ √
on which (the complexification of) J acts as −1 · Id and − −1 · Id respectively. These
bundles have as a basis the complex tangent vector fields
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= − and = + .17
∂zi 2 ∂xi ∂yi ∂z i 2 ∂xi ∂yi
Furthermore, with the almost complex structure on T 1,0 U determined by multiplication by
√
−1,
√

T 1,0 U, −1 ∼ = (T U, J).

Similarly, we get a decomposition

TC∗ U = Λ1,0 T ∗ U ⊕ Λ0,1 T ∗ U
of the complexified cotangent bundle of U , with bases
√ √
dzi = dxi + −1 · dyi , dz i = dxi − −1 · dyi .
17This explains intrinsically the meaning of these complex vector fields, which were introduced in an ad
hoc fashion in the first section.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 59

As in the pointwise situation of the previous section, this leads to a C-linear direct sum
M
Λk TC∗ U = Λp,q T ∗ U,
p+q=k
p,q ∗
with a basis for Λ T being given by forms of the sort
dzI ∧ dz J , |I| = p , |J| = q.
This in turn leads to a decomposition of the spaces of smooth differential forms on U . Write
AkC (U ) for the space of smooth complex-valued differential k-forms on U , i.e. C ∞ global
sections of Λk TC∗ U . Then we have
M
AkC (U ) = Ap,q (U ), (4.1)
p+q=k

where Ap,q (U ) is the (infinite-dimensional) complex vector space of smooth sections of

Ap,q (U ), i.e. complex-valued forms of the type
X
fI,J · dzI ∧ dz J ,
|I|=p,|J|=q

the fI,J being smooth functions on U . We say that such a form is of type (p, q), and we
denote by
π p,q : AkC (U ) −→ Ap,q (U )
the projections.

So far all of these constructions are completely algebraic in nature, ie they depend only
on the point-wise situation studied in the previous section. Calculus enters the picture with
the exterior derivative. Given a smooth C-valued function f ∈ A0C (U ) on U , we define
∂f ∈ A1,0 (U ) , ∂f ∈ A0,1 (U )
to be respectively the (1, 0) and (0, 1) components of df ∈ A1 (U ), i.e. ∂f = π 1,0 (df ) and
∂f = π 0,1 (df ). Thus df = ∂f + ∂f , and in terms of local coordinates:
n
X ∂f
∂f = · dzi
i=1
∂z i
n
X ∂f
∂f = · dz i .
i=1
∂z i

So for instance f is holomorphic if and only if ∂f = 0.

Now consider a “monomial” (p, q)-form

α = f · dzI ∧ dz J ,
where I, J are multi-indices of sizes p and q respectively. Then dα = df · dzI ∧ dz J , and
therefore
dα = (∂f ∧ dzI ∧ dz J ) + (∂f ∧ dzI ∧ dz J ). (*)
60 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Therefore the derivative of an arbitrary (p, q)-form is the sum of a (p + 1, q)-form and a
(p, q + 1)-form. It follows that the full exterior derivative
d : AkC (U ) −→ Ak+1
C (U )

decomposes as a direct sum d = ∂ + ∂ of two operators ∂ and ∂

∂ : Ap,q (U ) −→ Ap+1,q (U )
∂ : Ap,q (U ) −→ Ap,q+1 (U ).
determined by the two expressions on the right in (*). Formally,
∂ = π p+1,q ◦ d , ∂ = π p,q+1 ◦ d,
and in terms of local coordinates:
n
!

X ∂f
∂ f · dzI ∧ dz J = · dzi ∧ dzI ∧ dz J
i=1
∂zi
n
!

X ∂f
∂ f · dzI ∧ dz J = · dz i ∧ dzI ∧ dz J .
i=1
∂z i

The following statement summarizes the properties of this operator.

Proposition 4.1. One has:

(i). d = ∂ + ∂.
2
(ii). ∂ 2 = ∂ = 0 and ∂∂ + ∂∂ = 0.
(iii). These operators satisfy the Leibnitz rule: if α ∈ Ap,q (U ), then
∂(α ∧ β) = ∂(α) ∧ β + (−1)p+q α ∧ ∂(β)
∂(α ∧ β) = ∂(α) ∧ β + (−1)p+q α ∧ ∂(β).

∂-Poincaré Lemma. In this subsection we will establish the analogue for ∂ of the classical
Poincaré lemma that a closed form is locally exact. Specifically, we aim to prove (most of)
the following:
Theorem 4.2. Let ∆ ⊆ Cn be a polydisk, and let
α ∈ Ap,q (∆)
be a ∂-closed smooth (p, q)-form on δ. If q ≥ 1, then there exists a smooth (p, q − 1)-form β
on ∆ such that
α = ∂β.

It turns out that most of the work occurs in the case n = 1. We start by recording a
C ∞ version of the Cauchy integral formula.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 61

Proposition 4.3. Let ∆ ⊂ C be a disk, let f ∈ C ∞ (∆) be a smooth function defined on an

open neighborhood of the closure of ∆, and fix a point z ∈ ∆. Then
∂f (w) dw ∧ dw
Z Z
1 f (w)dw 1
f (z) = √ + √ .
2π −1 ∂∆ w − z 2π −1 ∆ ∂w w − z

For the proof, see [2, p.2] or [3, p. 46].

The following proposition essentially proves the ∂-Poincaré lemma in one variable for
the form f (z)dz (which is automatically ∂-closed).
Proposition 4.4. Let ∆ ⊂ C be a disk, and let f ∈ C ∞ (∆) be a smooth function on an
open neighborhood of ∆. Then the function
Z
1 f (w)
h(z) = √ dw ∧ dw
2π −1 ∆ w − z
is defined and C ∞ in ∆, and satisfies
∂h
= f.
∂z

Proof. Fix a point z0 ∈ ∆, and choose ε > 0 such that

∆(z0 , ε) ⊂ ∆(z0 , 2ε) ⊂ ∆.
Starting with a C ∞ function that is ≡ 1 in ∆(z0 , ε) and ≡ 0 outside ∆(z0 , 2ε), we may write
f = f1 + f2
where f1 ≡ 0 outside ∆(z0 , 2ε) and f2 ≡ 0 in ∆(z0 , ε). Then for z ∈ ∆(z0 , ε) the function
Z
1 f2 (w)
h2 (z) =def √ dw ∧ dw
2π −1 ∆ w − z
is C ∞ , and we have ∂z
∂
h2 (z) = 0 since the integrand is holomorphic in z outside ∆(z0 , ε). We
next consider the corresponding integral of f1 . Since f1 has compact support, one can write
and write (for z ∈ ∆(z0 , ε)):
Z
1 f1 (w)
h1 (z) =def √ dw ∧ dw
2π −1 ∆ w − z
Z
1 f1 (w)
= √ dw ∧ dw
2π −1 C w
Z
1 f1 (u + z)
= √ du ∧ du,
2π −1 C u
√
where u = w − z. Now pass to polar coordinates u = re( −1)θ . One finds that
√ √
Z
1
h1 (z) = − f1 (z + re( −1)θ )e−( −1)θ dr ∧ dθ,
π C
62 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

which is C ∞ in z. Moreover we have

√ √
Z
∂h1 (z) 1 ∂f1
=− (z + re( −1)θ ) · e−( −1)θ dr ∧ dθ
∂z π C ∂z
∂f1 (w) dw ∧ dw
Z
1
= √
2π −1 ∆ ∂w w−z
= f1 (z),
where the last equality arises from Proposition 4.3 and the fact that f1 vanishes on ∂∆.
Therefore ∂h
∂z
= f in ∆(z0 , ε), as required. 

We now turn to the proof of (a slightly weakened form of) Theorem 4.2. To begin with,
we reduce to proving the theorem in the case p = 0. In fact, suppose that
X
α = fI,J dzI ∧ dz J
I,J
P
(#I = p, #J = q) is a form of type (p, q). Then we can write α = I dzI ∧ αI , where
X
αI = fI,J dz J
J

is a (0, q)-form. Then ∂α = 0 if and only if ∂αI = 0 for every I, and if each αI is ∂-exact
then so it α. So it suffices to prove the theorem for (0, q)-forms.

Now let ∆ = ∆(ε) ⊆ Cn be a bounded polydisk, and that for q > 0 we are given a
∂-closed (0, q)-form α ∈ A0,q (U ) defined and smooth on an open neighborhood U ⊇ ∆. We
will show that
α = ∂β for some β ∈ A0,q−1 (∆).18
P
Write then α = fI dz I , and let 1 ≤ k ≤ n be the least index such that no dz i appears
non-trivially in this expression for i > k. Then we can write
α = α1 ∧ dz k + α2 ,
where α2 is a (0, q)-form involving only the variables z1 , . . . , zk−1 . Setting
 
∂
∂i = · dz i ,
∂z i
the assumption ∂α = 0 implies that
∂i α1 = ∂i α2 = 0
for i > k, and hence the coefficients fI of α are holomorphic in the variables zk+1 , . . . , zn .

Now consider the function

Z
1 fI (z1 , . . . , zk−1 , w, zk+1 , . . . , zn )
hI = √ dw ∧ dw,
2π −1 B w − zk
18The general case – which allows α to be only defined on ∆ is treated by an exhaustion argument. See
[2, p. 26] or [3, p. 47] for the details.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 63

the integral being taken over a suitable disk B ⊂ C. By Proposition 4.4, one has
∂hI
= fI
∂z k
on B. Moreover hI is holomorphic in zk+1 , . . . , zn , and smooth in the other variables. This
being said, put X
γ = hI dz I−{k} .
I3k

Then ∂ i γ = 0 for i > k, and ∂ k γ = ±α1 . Thus

α ± ∂γ
is still ∂-closed, but does not involve any of the dz k , . . . , dz n , and the assertion follows by
induction on k.

Local geometry of Kähler forms. We now bring metrics into the picture.

As before, let U ⊆ Cn be an open set.

Consider a Riemannian metric g =< , > on U , i.e a positive definite bilinear form gx =<
, >x on real tangent space Tx U varying smoothly with x. We assume that g is compatible
with the (almost) complex structure on U , i.e. we assume that
< Jv, Jw > = < v, w >
for all local vector fields v, w on U . Then by our discussion of Hermitian linear algebra, g
determines a real (1, 1)-form ω in U given by
ω(v, w) = < Jv , w >,
and √
h = g − −1 · ω
is a positive-definite Hermitian metric on U . As in the pointwise setting, one calls ω the
fundamental form associated to the metric. If we define
∂ ∂

hij (z) = h , ,
∂xi ∂xj z
then hij is a positive-definite matrix of smooth functions, and
√
−1 X
ω = · hij (z) dzi ∧ dz j .
2 i,j

Since conversely ω determines g and h, one sometimes abusively refers to the fundamental
form ω as the compatible metric.

For example, if g is the constant standard metric on U , then h = Id is the constant

standard Hermitian metric, and
√
−1 X
ω = · dzi ∧ dz i .
2 i
64 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

We now come to a central definition

Definition 4.5. One says that the compatible metric g (or ω) is Kähler if
dω = 0.

Such metrics will play a central role in what follows. For now we want to prove an important
local characterization of Kähler metrics.

We say that the Hermitian metric h on U (or the underlying Riemannian metric g) os-
culates to the identity to order 2 at a point p ∈ U if in suitable local holomorphic coordinates
centered at p one write
h = Id + O(|z|2 ), (4.2)
where the term on the right indicates a matrix of functions vanishing to order ≥ 2 at 0. This
is a useful condition, because it allows one to check identities among first order differential
operators by working with the standard metric.

The basic result is that the Kähler condition is equivalent to the being able to choose
coordinates so that g osculates to the identity.
Theorem 4.6. Let g be a compatible metric on U 3 p, with fundamental form ω. Then
dω = 0 if and only if one can choose coordinates centered at p with respect to which g
osculates the identity to order 2.

Sketch of proof. Assuming that dω = 0 we will show that after a coordinate change we can
arrange for (4.2) to hold. To this end, write
√
−1 X
ω = · hij (z)dzi ∧ dz j ,
2 i,j

where
X X
hij (z) = δij + aijk zk + a0ijk z k + O(|z|2 ),
k k

and hij = hji . Then

X X
dhij = aijk dzk + a0ijk dz k + O(|z|).
k k

Now the condition dω = 0 implies that the form

X
dhij ∧ dzi ∧ dzj
i,j

vanishes at the origin, which is equivalent to the equalities

aijk = akji , a0ijk = a0ikj . (*)
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 65

Moreover, since h is hermitian, a0ijk = aj ik. Now set

n
1 X
wj = zj + · aijk zi zk .
2 i,k=1

Then a computation using the relations in (*) shows that in these new coordinates,
√
−1 X
ω = · dwj ∧ dwj + O(|w|2 ),
2
as required. (See [3, p. 49] for details.) 

Almost complex structures. Let X be a real manifold of even (real) dimension 2n. An
almost complex structure on X is a bundle endomorphism
J : T X −→ T X with J 2 = Id.
Thus J gives an almost complex structure Jx on each of the (real) tangent spaces Tx X. As
in §3, an almost complex structure gives rise to a decomposition into type
T ∗ X = T ∗1,0 ⊕ T ∗0,1 X,
and similarly for k-forms. Moreover we can define operators
∂ : Ap,q (X) −→ Ap+1,q (X) , ∂ : Ap,q (X) −→ Ap,q (X)
by composing the deRham d with projection onto the spaces of (p + 1, q) and (q, p + 1)-forms
respectively. Note that on functions d acts as ∂ + ∂, but there is no reason that this has to
be true on higher forms, since it could happen eg that there is a (1, 0) form α such that dα
has a non-trivial (2, 0) component.

Evidently every complex manifold has a canonical such almost complex structure, so
it is natural to for conditions under which an almost complex structure is integrable, i.e
arises from a complex structure. This is the content of a famous theorem of Newlander and
Nierenberg.
Theorem 4.7. (Newlander–Nierenberg). Let (X, J) be an almost complex manifold.
Then X carries a complex structure giving rise to J if and only if any of the following
equivalent conditions are satisfied:

(i). d = ∂ + ∂ on A1,0 (X).

(ii). d = ∂ + ∂ on A∗ (X).
2
(iii). ∂ = 0.

The proof that these statements are equivalent is elemantary (see [3, p. 107, 108]). The
serious assertion is that these imply the existence of a complex structure on X giving rise to
J.
66 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Exercises for Section 4.

Exercise 4.1. Let f : U −→ V be a holomorphic mapping. Show that the complexified real
derivative DC (f ) : TC U −→ TC V maps T 1,0 U to T 1,0 (V ), and that under the isomorphisms
of complex vector spaces
TU ∼= T 1,0 , T V ∼ = T 1,0 V,
the resulting homomorphism T 1,0 U −→ T 1,0 V corresponds to the C-linear mapping deter-
mined by DR f , as in Proposition 1.17.
Exercise 4.2. Use considerations of type to prove statements (ii) and (iii) of Proposition
4.1.

5. Dolbeaut Theorem and Sheaf Cohomology

In this section, we give a brief sketch of the theory of sheaves and their cohomology, and
prove Dolbeaut’s theorem.

Statement of Dolbeaut’s theorem. Let X be a complex manifold. Writing Ap,q (X) for
the space of global smooth (p, q)-forms on X, we have the ∂-operator
∂ : Ap,q (X) −→ Ap,q+1 (X)
2
satisfying ∂ = 0. Therefore we can make the
Definition 5.1. The Dolbeaut cohomology groups of X are defined as the vector space of
∂-closed (p, q)-forms modulo ∂-exact (p, q)-forms:


p,q ker ∂ : Ap,q (X) −→ Ap,q+1 (X)
H∂ (X) =
.
im ∂ : Ap,q−1 (X) −→ Ap,q (X)

One can see these groups as holomorphic analogues of the DeRham cohomology groups of
a smooth manifold. So it is natural to ask what invariants of X they compute. Dolbeaut’s
theorem asserts that they are isomorphic to certain basic sheaf-theoretic invariants of X:
Theorem 5.2. (Dolbeaut’s theorem, I). There are canonical isomorphisms:
H∂p,q (X) = H q (X, ΩpX ).
In particular,
H∂0,q (X) = H q (X, OX ).

The groups on the right are the cohomology of X with coefficients in the sheaf of holomorphic
p-forms and the structure sheaf of X, respectively. Our goal in the rest of the section is to
explain the meaning of these groups, and to sketch the proof of the theorem. (Along the
way we’ll give the sheaf-theoretic proof of De Rham’s theorem, which follows formally from
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 67

the d-Poincaré lemma in the same way that the Dolbeaut theorem will follow from the ∂-
Poincaré lemma.) However before turning to sheaf theory, we want to indicate a “twisted”
analogue of Dolbeaut cohomolgy.

Specifically, let E be a holomorphic vector bundle of rank e on X, given as above by

local data {Ui , gij }. Then one can form the complex vector bundle Λp,q T X ∗ ⊗C E of E-
valued (p, q)-forms, whicn in turn gives rise to the (infinite-dimensional) complex vector
space Ap,q (E) = Ap,q (X, E) of smooth E-valued (p, q)-forms: an element
α ∈ Ap,q (X, E)
is locally given by a vector of smooth (p, q)-forms which transform via the transition matrices
of E. Thus
Ap,q (X) = Ap,q (X, 1)
is recovered as the space of smooth (p, q)-forms with values in the trivial line bundle.

It is a wonderful fact that given any holomorphic vector bundle E, the ∂ operator on
forms extends canonically to an operator
∂ = ∂ E : Ap,q (X, E) −→ Ap,q+1 (X, E). (5.1)
The critical case to understand is that when p = 0:
∂ E : A0 (E) −→ A0,1 (E). (*)
In other words, we need to explain how to take the ∂-derivative of a smooth section of E.
On a trivializing open set Uj , we represent s ∈ A0 (E) by a vector sj of smooth sections,
and these data patch together via the rule si = gij sj on Uij . We locally define ∂s to be the
vector of (0, 1)-forms obtained by differentiating each of the components of sj , i.e.
∂ E (s) =locally ∂(sj ) ∈ A0,1 (Uj , E).
We need to check that the vectors of (0, 1)-forms so obtained transform in the required
manner. But ∂gij = 0 thanks the fact that the gij are matrices of holomorphic functions,
and hence differentiating the relation si = gij sj one finds that
∂si = ∂gij · sj + gij · ∂sj
= gij · ∂sj ,
as required. The reader can then check
Lemma 5.3. The operator ∂ E : A0 (E) −→ A0,1 (E) defined in (*) extends uniquely to an
operator
∂ = ∂ E : Ap,q (X, E) −→ Ap,q+1 (X, E)
satisfying the Leibnitz rule
∂ E (f · s) = (∂f ∧ s) + f · ∂ E (s),
2
and ∂ E = 0.
68 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

We naturally then define the Dolbeaut cohomology of X with coefficients in E to be

p,q p,q ker ∂ : Ap,q (E) −→ Ap,q+1 (E)
H (X, E) = H∂ (E) =
.
im ∂ : Ap,q−1 (E) −→ Ap,q (E)
The first statement of Dolbeaut’s theorem then generalizes to the assertion that these groups
involve the sheaf the cohomolgy of X with coefficients in the sheaf of holomorphic sections
of E:
Theorem 5.4. (Dolbeaut’s theorem, II). There are canonical isomorphisms:
H∂p,q (X, E) = H q (X, ΩpX ⊗ OX (E)).
In particular,
H∂0,q (X, E) = H q (X, OX (E)).

Sheaves. One can think of the theory of sheaves as a tool for managing the questions
involved in trying to pass from local to global constructions.

Let X be a topological space.

Definition 5.5. A pre-sheaf F of additive abelian groups on X is a rule that assigns to each
open set U ⊆ X an abelian group F (U ), and to each inclusion V ⊆ U a homomorphism
ρU,V : F (U ) −→ F (V ),
called a restriction map. These data are required to satisfy the following axioms:

(i). F (∅) = (0)

(ii). ρU,U = Id.
(iii). Given W ⊆ V ⊆ U , one has
ρU,W = ρV,W ◦ ρU,V .

More succinctly, one can say that a presheaf is a contravariant functor from the category
of open subsets of X to the category of abelian groups. One defines presheaves of rings, or
vector spaces, or ..., similarly. One typically calls the elements of F (U ) sections of F over
U.
Example 5.6. (Holomorphic functions and sections). Let X be a complex manifold.
The presheaf OX of holomorphic functions on X associates to an open set U the holomorphic
functions on U :
OX (U ) = {holomorphic functions on U },
with ρU,V being given by restriction. Similarly, given a holomorphic vector bundle E on X,
we define OX (E) by the rule
OX (E)(U ) = Γ(U, E) = {holomorphic sections of E over U },
again with the ρ being ordinary restrictions. Thus OX is a presheaf of rings (and OX (E) is
a sheaf of modules over OX in a sense that we will define shortly).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 69

Example 5.7. (Ideal of analytic subvariety). Let X be a complex manifold, and let
V ⊆ X be an analytic subvariety. Then one defines the ideal presheaf IV of V by taking
IV (U ) = {f ∈ OX (U ) | f |V ≡ 0}.
This is a sub-presheaf of OX in the sense that IV (U ) ⊆ OX (U ) for all U ⊆ X.
Example 5.8. (Smooth forms). Let M be a smooth real manifold. Then the presheaf
AkM of smooth k-forms on X is given by
AkM (U ) = Ak (U ),
with the evident restrictions. So for exampe A0M is the presheaf of smooth functions on M .
If X is a complex manifold, one defines presheaves Ap,q
X in the analogous fashion.

Example 5.9. (Constant presheaf ). Let X be a topological space. The constant presheaf
Zconst on X is the presheaf that assigns to U ⊆ X all constant functions U −→ Z, ie
Zconst (U ) = Z,
with all non-trivial restrictions being the identity.

In all the examples so far, the restriction homomorphisms ρ have been actual restrictions
of functions, sections or forms. Although this need not be true in general, it is customary to
lighten the notation by writing
ρU,V (s) = s|V
for s ∈ F (U ) and V ⊆ U .

A sheaf is a pre-sheaf whose sections are determined locally:

Definition 5.10. Let F be a presheaf on a topological space. One says that F is a sheaf if
the following condition is satisfied:

Let U be an open set, and let U = ∪Ui be an open covering of U . Suppose

given sections si ∈ F (Ui ) satisfying
si |(Ui ∩ Uj ) = sj |(Ui ∩ Uj )
for all i, j. Then there exists a unique section s ∈ F (U ) such that
si = s | Ui .

In other words, the condition asks that the si should patch together to give a section on all
of U .
Example 5.11. The presheaves in Examples 5.6, 5.7 and 5.8 are sheaves.
Example 5.12. The presheaf Zconst from Example 5.9 is not in general a sheaf. For example,
take X = R, and consider the disconnected subset U = R − {0} with the open cover
U = (−∞, 0) ∪ (0, ∞).
70 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Then evidently constant functions

s− ∈ Zconst (−∞, 0) , s+ ∈ Zconst (0, ∞)
do not necessarily patch together to give a constant function on U , even though they vacu-
ously agree on intersection of the two subsets of U in question. However for any topological
space X, the presheaf ZX of locally constant functions on X, given by
ZX (U ) = {locally constant functions U −→ Z},
is a sheaf. In particular, Zconst is not the natural object to look at, and one therefore typically
refers to ZX as the constant sheaf on X.

It will be particularly important to understand the sections of a sheaf F over all of X:

this group is written
Γ(X, F ) = Γ(F ) = F (X),
and is called the group of global sections of F .

There is a natural notion of a homomorphism of sheaves or presheaves, as well as sub-

sheaves.
Definition 5.13. A morphism φ : F −→ G of sheaves or pre-sheaves consists in giving for
each open U ⊆ X a homomorphism
φU : F (U ) −→ G(U )
such that the φU are compatible with the restriction morphisms for F and G.
Definition 5.14. Given sheaves or presheaves F , G, one says that F is a subsheaf of G if
F (U ) ⊆ G(U ) is a subgroup for every U , compatably with restrictions.
Example 5.15. Let X be a complex manifold. Then (locally) constant sheaf ZX sits as a
subsheaf
ZX ⊆ OX
∗
of the sheaf OX of holomorphic functions on X. On the other hand, denote by OX the sheaf
of nowhere-zero holomorphic functions, given by
∗
OX (U ) = OX (U )∗ :
we view this as a sheaf of multplicative abelian groups. Then the exponential defines a sheaf
homomorphism
∗
exp : OX −→ OX
via the maps
√
∗ −1·f )
OX (U ) −→ OX (U ) , f 7→ e(2π .
In fact, we will see shortly that ZX = ker exp, and that there is a short exact sequence
∗
0 −→ ZX −→ OX −→ OX −→ 0 (5.2)
which has very beautiful cohomological consequences.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 71

Example 5.16. Let X be a complex manifold, and fix an index p ≥ 0. Then the ∂-operator
gives rise to a sheaf homomorphism
∂ : Ap,q p,q+1
X −→ AX
of sheaves of vector spaces.19 In fact, these fit together into a complex
. . . −→ Ap,q−1 −→ Ap,q
X −→ A
p,q+1
−→ . . . (5.3)
of sheaves of vector spaces. The proof of the Dolbeaut theorem will revolve around the
analysis of this complex.

Recall that we defined the germ of a holomorphic function f at a given point x to be

the restriction of f to a small neighborhood of x, and identifying two such restrictions if
they agree in a possibly smaller neighborhood. Given any presheaf F , we can define in a
completely analogous way the germ of a section of F at a point: this leads to the notion of
the stalk of F .
Definition 5.17. Let F be presheaf on a topological space X, and let x ∈ X be a fixed
point. The stalk Fx of F at x is defined to be
n o.
Fx = (U, s) U 3 x , s ∈ F (U ) ∼
where (U1 , s1 ) ∼ (U2 , s2 ) if there is an open neighborhood x ∈ V ⊆ U1 ∩ U2 such that
s1 |V = s2 |V.
Given s ∈ F (U ), the germ sx ∈ Fx of s at x is determined by the equivalence class of (U, s).

Observe that a morphism φ : F −→ G of presheaves determines a homomorphism φx :

Fx −→ Gx for every x ∈ X.
Example 5.18. (Sections from stalks). If F is a sheaf, then for any open set U ⊆ X, a
section s ∈ F (U ) is determined by its germs sx ∈ Fx for x ∈ U . However this can fail if F
is merely a presheaf (Exercise 5.1).

There are many constructions that lead naturally to a presheaf F , and it is then impor-
tant to know that there is a canonical way to pass to a sheaf that is “closest” to F . This is
the sheafification of F , or the sheaf associated to F .
Proposition 5.19. Let F be a presheaf on a topological space X. Then there is a sheaf F + ,
together with a morphism
θ : F −→ F +
characterized by the property that any homomorphism φ : F −→ G from F to a sheaf G
factors uniquely through θ. Moreover, the canonical homomorphism
θx : Fx −→ Fx+
is an isomorphism for all x ∈ X.
19Note that since ∂ is not linear over the ring of smooth functions, 5.3 is not a homomorphism of A0X -
modules.
72 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

We refer eg to [] for the proof. However we can quicky explain the construction of F + .
Start by forming the disjoint union
a
U = Fx
x∈X

of all the stalks of F , and fix U ⊆ X. Then F + (U ) consists of all functions

t : U −→ T , with t(x) ∈ Fx for all x,
having the property that for any y ∈ U , there is a neighborhood U ⊇ V 3 y and a section
s ∈ F (V ) such that
t(z) = sz for all z ∈ V.

We can now define the kernel and cokernel of a homomorphism of sheaves.

Definition 5.20. Let
φ : F −→ G
be a homomorphism of sheaves. Then ker(φ) is the sheaf:


ker(φ)(U ) = ker φU : F (U ) −→ G(U ) .
Similarly, im(φ) is the sheaf associated to the presheaf


U 7→ im φU : F (U ) −→ G(U ) .

One says that φ is surjective if im(φ) = G.

Example 5.21. Let φ : F −→ G be a homomorphism of sheaves. Then φ is surjective if
and only if given any point x ∈ X,a neighborhood U 3 x and a section s ∈ G(U ), one can
find a smaller neighborhood x ∈ V ∈ U such that
s|V = φV (t) for some t ∈ F (V ).
Example 5.22. Let X be a complex manifold. Then the exponential mapping
∗
exp : OX −→ OX
defined above is surjective, and
ker(exp) = ZX .
(Exercise 5.4.) In other words, the exponential sequence (5.2) is exact.

Note in particular that the surjectivity of φ does not imply the surjectivity of φU :
F (U ) −→ G(U ) for every open set U ⊆ X. For example, let X = C, and consider the
exponential map
∗
exp : OX −→ OX .
As noted in the previous example, this is surjective as a morphism of sheaves. However if
∗
U = C − {0}, then the function z ∈ OX (U ) is not in the image of exp (one cannot define a
single-valued branch of log(z) on C − {0}).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 73

The quotient of a sheaf by a subsheaf is defined similarly. Specifically, given F ⊆ G, one

defines F/G to be the sheaf associated to the subsheaf
 
U 7→ G(U )/F (U ) .
Thus a section s ∈ (G/F )(U ) is determined by the condition that one can find an open
covering U = ∪Ui , together with sections ti ∈ G(Ui ), such that ti maps to s|UI .

Having discussed quotients and kernels, one can then define exact sequences of sheaves:
0 −→ G −→ F −→ Q −→ 0,
which implies that G is a sub sheaf of F , and that Q = F/G. In general, a sequence of
sheaves on X is exact if and only if the corresponding sequences of stalks are exact at every
point.

Consider again now a complex manifold X.

Definition 5.23. A sheaf F on X is called an OX -module if F (U ) is an OX (U )-module for
each open set U ⊆ X, compatibly with the restriction maps.
Example 5.24. The sheaf OX (L) of sections of a holomorphic line bundle is an OX -module:
in fact, it is locally isomorphic to OX . Similarly for vector bundles. In this case Γ(X, OX (L))
is the space of global sections of L as discussed above.
Example 5.25. Fix a point P ∈ X, and consider the sheaf OX (L) ⊗ mP of sections of L
vanishing at P :

OX (L) ⊗ mP (U ) = {s ∈ Γ(U, L) | s(P ) = 0}.
This is an OX -module sitting as a sub-module of OX (L), and the quotient
G = OX (L)/(OX (L) ⊗ mP )
is a sky-scraper sheaf supported at P . The stalks of G are
(
C if x = P
Gx =
0 otherwise.
The natural map
OX (L)(U ) −→ G(U )
is given by evaluation at P . We usually denote such a sky-scraper sheaf as CP .

In our discussion of cohomology, it will be convenient to make a simplifying assumption.

Although we will not prove this here, we note the assumption holds in practice for all
“reasonable” sheaves of OX module on a complex manifold.
Simplifying Assumption 5.26. Let X be a complex manifold, and let
0 −→ F −→ E −→ G −→ 0
be an exact sequence of OX -modules on X. It will sometimes be convenient to assume that
there exist arbitrarily fine open coverings A = {Ui } of X such that the sequences
0 −→ F (UI ) −→ E( UI ) −→ G(UI ) −→ 0 (5.4)
74 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

are exact, where for each multi-index I = {i0 < . . . < ip }, UI denotes the intersection of the
corresponding elements of the covering.

Assume that for every x ∈ X, the stalks

Fx , Ex , Gx are finitely generated OX,x -modules.
Then in fact this condition is satisfied.

Cohomology of Sheaves. We now introduce (somewhat informally) the cohomology of X

with coefficients in a sheaf.

By way of motivation, consider an exact sequence

π
0 −→ F −→ E −→ G −→ 0
of sheaves on X, and suppose that we are in the situation of Fact 5.26. Writing as above
Γ(X, F ) = F (X) for the space of global sections of F , we want to investigate when the map
Γ(X, E) −→ Γ(X, G)
induced by π is surjective. For this, use 5.26 to choose an open covering A = {Ui } of X such
that we have a short exact sequence
π
0 −→ F (Ui ) −→ E(Ui ) −→ G(Ui ) −→ 0
for each i. Given a section s ∈ Γ(X, G), put
si = s | Ui ∈ G(Ui ).
Then the si lift to ti ∈ E(Ui ), ie π(ti ) = si . One cannot hope that the ti patch together to
give a global section t ∈ Γ(X, E) – after all, there is some ambiguity in choosing them – but
we can study systematically the extent to which this fails. In fact, on Uij = Ui ∩ Uj consider
the difference
rij = ti |Uij − tj |Uij .
Then


π(rij ) = π(ti ) − π(tj ) | Uij = si − sj | Uij = 0.
So we may view rij ∈ ker(π), i.e we may view as a section of F over Uij :
rij ∈ F (Uij ).
Note that on Uijk the (restrictions of the) rij satisfy


rij − rik + rkj | Uijk = 0.

The liftings ti are not unique: if we replace ti by

t0i = ti + ri , ri ∈ F (Ui ),
then the t0i are again liftings of si . Moreover the t0i give rise to rij
0
∈ F (Uij ) satisfying
0


(rij − rij ) = ri − rj | Uij .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 75

We claim now that s lifts to a section t ∈ Γ(X, E) if and only if we can find ri ∈ F (Ui ) such
that

rij = ri − rj | Uij .. (*)
In fact, supposing that (*) holds, define
t0i = ti − ri .
Then (t0i − t0j ) = 0 on Uij , so the t0i patch to give a global section t ∈ Γ(X, E) that maps to s.

We can rephrase this discussion more systematically by introducing the group

{rij ∈ F (Uij ) | rij − rjk + rki = 0}
Ȟ 1 (A, F ) = .
{ri − rj | ri ∈ F (Ui )}
Then in effect we have constructed a homomorphism
δ : Γ(X, G) −→ Ȟ 1 (A, F )
that fits into an exact sequence
δ
0 −→ Γ(X, F ) −→ Γ(X, E) −→ Γ(X, G) −→ Ȟ 1 (A, F ). (5.5)
In other words, we have cooked up a (cohomology) group H 1 which gives the obstruction to
the surjectivity of the map Γ(X, E) −→ Γ(X, G). As we shall see, in practice it is the H 0
and H 1 that are of most immediate importance, but to get our hands on the H 1 we often
need to consider an H 2 , etc. So we are led to introduce the whole panaply of cohomology
groups.

We now turn to the formal definitions. Let X be a topological space, let A = {Ui } be
an open covering of X, and let F be a sheaf on X.
Definition 5.27. The group of Cech p-cochains is
Y
C p (A, F ) = F (Ui0 i1 ...ip ).
i0 ,i1 ,...,ip

Thus an element γ ∈ C p (A, F ) consists in giving for each p + 1-fold intersection of the Ui a
section
γ(i0 , . . . , ip ) ∈ F (Ui0 i1 ...ip
of F over the corresponding open set.

Note that we do not require that the Ui appearing in the definition be distinct.

We next define the Cech boundary operator

δ : C p (A, F ) −→ C p+1 (A, F )
by the rule
p+1
X
(−1)j res γ(i0 , . . . , ibj , . . . ip+1 ) ,


δ(γ)(i0 , . . . , ip+1 ) =
j=0
where each of the sections on the right is restricted to the open set Ui0 ...ip+1 .
76 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

In the usual way, one checks that δ 2 = 0, and then we can take cohomology:
Definition 5.28. The Cech cohomology of F with respect to the covering A the cohomology
of the Cech complex just constructed:
Ȟ p (A, F ) = H p C ∗ (A, F ) .

Example 5.29. For any sheaf F and any open covering A,

Ȟ 0 (A, F ) = Γ(X, F ).
(Exercise 5.6.)
Example 5.30. The reader should check that the case p = 1 of Definition 5.28 agrees with
the group constructed by hand above.

Of course one would like to remove the dependence on the covering. Although we will
gloss over this point, for this one passes to refinements. In brief, suppose that A0 ≺ A is a
refinement of A. One can then construct chain maps
C p (A, F ) −→ C p (A0 , F )
which are uniquely defined up to homotopy. This gives rise to canonically defined homomor-
phisms
Ȟ p (A, F ) −→ Ȟ p (A0 , F ),
and one then passes to a direct limit:
Definition 5.31. The Cech cohomology of X with coefficients in a sheaf F is detined to the
the direct limit
Ȟ p (X, F ) = lim Ȟ p (A, F ),
−→
where the limit is taken over all open covers.

It turns out that for most of the sheaves F that arise in day to day complex or algebraic
geometry – specifically “coherent analytic” or “coherent algebraic” sheaves – there is a single
cover A with the property that all the cohomology groups Ȟ p (X, F ) can be computed as
H p (A, F ). So in practice one usually doesn’t need to pass to the limit.
∗
Example 5.32. Given a complex manifold X, denote by OX the sheaf of nowhere-vanishing
holomorphic functions on X: this is a sheaf of abelian groups under multiplication. Then
n o
∗
Ȟ 1 (X, OX ) = isomorphism classes of holomorphic line bundles on X .
∗
In fact, Ȟ 1 (A, OX ) classifies isomorphism classes of line bundles that trivialize on the open
∗
cover A: the corresponding transition functions gij ∈ OX (Uij ) give the cocycle determined
by a line bundle.

The most important property of the theory is that a short exact sequence
0 −→ F −→ E −→ G −→ 0 (*)
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 77

of sheaves on X should give rise to a long exact sequence on cohomology extending (5.5).
This is not universally true for Cech cohomology, but it does work on paracompact Hausdorff
spaces.
Theorem 5.33. Let X be a paracompact Hausdorff space, and suppose given a short exact
sequence (*) of sheaves on X. Then (*) induces a long exact sequence
0 −→ Ȟ 0 (X, F ) −→ Ȟ 0 (X, E) −→ Ȟ 0 (X, G) −→ Ȟ 1 (X, F ) −→ Ȟ 1 (X, E) −→ Ȟ 1 (X, E)Ȟ 2 (X, F ) −→ . . .

of cohomology groups.

We will not try to prove this in general, but the result is easy to derive assuming that
we are in the situation of the simplfying ssumption 5.26. In fact, it follows from (5.4) and
the definitions that the short exact exact sequence (*) gives rise to a short exact sequence
of complexes
0 −→ C • (A, F ) −→ C • (A, E) −→ C • (A, G) −→ 0.
This gives rise to a long exact sequence of the cohomology groups of A, and since we assume
that 5.26 holds for arbitrarily fine covers, we get the conclusion of 5.33 by passing to a limit.
Example 5.34. Consider the exponential sequence
∗
0 −→ ZX −→ OX −→ OX −→ 0
on a compact connected complex manifold X. Since
∗
H 0 (X, OX ) = C , H 0 (X, OX ) = C∗
(the only holomorphic functions on X are constant), the corresponding long exact sequence
on cohomology effectivelly starts with H 1 and takes the form:
∗
0 −→ H 1 (X, Z) −→ H 1 (X, OX ) −→ H 1 (X, OX ) −→ H 2 (X, Z) −→ H 2 (X, OX ) −→ . . . .
∗
Recalling that H 1 (X, OX ) classifies isomorphism classes of holomorphic line bundles on X,
this will later give us a rather precise description of all such. In particular this sequence will
lead to the proof of the Lefschetz (1, 1)-theorem.

Proof of the Dolbeaut theorem. We’re now ready to indicate the proof of the Dolbeaut
theorem.

Let X be a complex manifold of (complex) dimension n.

Definition 5.35. A holomorphic p-form on X is a (p, 0)-form η which can locally be ex-
pressed as
X
η = fI dzI ,
where the fI are holomorphic functions. Equivalently, η is a holomorphic section of the
holomorphic vector bundle Λp,0 TX∗ . We denote by ΩpX the holomorphic locally free sheaf of
holomorphic p-forms. Given a holomorphic vector bundle E on X, we speak similarly of
E-valued holomorphic p-forms, corresponding to the sheaf ΩpX ⊗ E.
78 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

We denote by Ap,q = Ap,qX the sheaf of smooth (p, q) forms on X. Thus the space A (X)
p,q

of global (p, q)-forms on X is just the group of global sections of this sheaf. Fixing an integer
p ≥ 0, we get as in Example 5.16 a complex of sheaves Ap,• :
0 −→ Ap,0 −→ Ap,1 −→ . . . −→ Ap,q −→ . . . (5.6)

Remainder of section not yet written...

Exercises for Section 5.

Exercise 5.1. Let F be a sheaf, and let U ⊆ X be an open set. Suppose that s, t ∈ F (U )
are two sections such that
sx = tx ∈ Fx
for all x ∈ U . Show that then s = t. However give an example to show that this can fail if
F is merely a presheaf.
Exercise 5.2. Let F = Zconst be the constant pre-sheaf on a space X. Show that the
sheafification F + of F is the sheaf ZX of locally constant integer valued functions on X.
Exercise 5.3. Let φ : F −→ G be a morphism of sheaves. Show that
ker(φ)x = ker(φx : Fx −→ Gx ) , im(φ)x = im(φx : Fx −→ Gx ).
In particular, φ is injective or surjective if and only if all the maps φx on stalks are so.
Exercise 5.4. Prove the assertions of Example 5.22
Exercise 5.5. (Locally free sheaves) Prove that a sheaf F on X is of the form OX (L)
for a holomorphic line bundle on X if and only if there is an open covering {Ui } of X such
that F |Ui ∼
= OUi for each i. State and prove the analogous assertion for vector bundles of
higher rank.
Exercise 5.6. Prove the assertion of Example 5.29
Exercise 5.7. Work out in detail the assertion of Example 5.32. Show similarly that iso-
morphism classes of smooth complex line bundles on a complex manifold X are classified
by
Ȟ 1 (X, A∗ ),
where A∗ denotes the sheaf of nowhere vanishing smooth C-valued functions on X.

6. Kähler Manifolds

In this section, much of the preceeding material comes together in the Kähler package,
which shows that compact complex manifolds that carry a Kähler metric have a number of
truly miraculous prorperties.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 79

The Kähler condition. Let X be a complex manifold of (complex) dimension n, and let
J be the corresponding almost complex structure. We suppose that X carries a Riemannian
metric g =< , > that is compatible with J, so that the fundamental two-form
ω ∈ A1,1 (X) , ω(v, w) = < Jv , w >
is defined. As in Definition 4.5, we say that (X, J, ω) is a Kähler manifold – and that g (or
ω) is a Kähler metric if dω = 0.20
Example 6.1. Any Riemann surface (complex manifold of dimension = 1) is Kähler.
Example 6.2. The standard constant metric g on Cn , with fundamental form
√
−1 X
ω = dzi ∧ dz i ,
2
is Kähler. Since the metric is invariant under translations, it follows that any complex torus
X = Cn /Λ is Kähler.
Example 6.3. (Submanifolds) If (X, g) is a Kähler manifold, then any complex subman-
ifold Y ⊆ X inherits a natural Kähler structure.
Example 6.4. Let (X, g, ω) be a compact Kahler manifold of dimension n. Since ω is closed,
it defines a class
[ω] ∈ H 2 (X, C).
Moreover since ω n is (a multiple of) the volume form determined by g, one has
Z
ω n > 0,
X

and hence [ω] 6= 0. In particular, if X is a compact Kähler manifold, then

H 2 (X, C) 6= 0,
and in fact all the even Betti numbers of X are non-zero. (In due time we will see that
the Hodge decomposition gives some much more subtle topological restrictions to a compact
complex manifold being Kähler, eg that the odd Betti numbers must be even.)
Example 6.5. (Hopf manifolds). We constructed in Example 2.16 a compact complex
manifold X of dimension n that is diffeomorphic to S 2n−1 × S 1 . It follows from the previous
example that if n > 1 then X cannot carry a Kähler metric.

Fubini-Study metric. What gives the Kähler condition its real importance is that projective
space carries a natural SU(n+1)-invariant Kähler metric, the so-called Fubini–Study metric.
Therefore any complex submanifold of projective space also carries a Kahler metric. We
construct the Fubini-Study metric by building an SU(n + 1)-invariant Hermitian metric
HFS on Pn . The Fubini–Study form ωFS will then arise as the negative imaginary part
ωFS = −Im HFS of this Fubini–Study metric.21
20One sometimes says that a manifold in Kähler if it admits a Kähler metric, but for now we’ll generally
suppose given a particular Kähler metric.
21The following paragraphs are lifted from Section 1.2.C of my positivity book.
80 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Consider the standard Hermitian inner product h(v, w) = t v · w on V = Cn+1 . Set

V 0 = V − {0} and denote by
ρ : V 0 −→ Pn = P(V )
the canonical map. Define to begin with a Hermitian metric H 0 on V 0 by associating to
x ∈ V 0 the Hermitian inner product
 
0 v w
Hx (v, w) = h , for v , w ∈ Tx V 0 = V
|x| |x|
p
and |x| = h(x, x). The metric H 0 is constructed so as to be invariant under the natural
C∗ -action on V 0 . Now ρ∗ T Pn is canonically a quotient of T V 0 , and so H 0 induces in the
usual manner a C∗ -invariant metric on ρ∗ T Pn , which then descends to a Hermitian metric
HFS on T Pn .

More explicitly, write Wx ⊆ V for the Hx0 -orthogonal complement to C · x ⊆ V , and let
πx : V −→ Wx be orthogonal projection:
h(v, x)
πx (v) = v − · x.
h(x, x)
Then Wx is identified with Tρ(x) Pn and πx with dρx , and
Hρ(x) dρx v, dρx w FS = Hx0 (πx v, πx w)

h(v, w)h(x, x) − h(v, x)h(x, w)

= .
h(x, x)2
If we take the usual affine local coordinates z1 , . . . , zn on Pn — corresponding to x =
(1, z1 , . . . , zn ) ∈ V 0 — then one finds22 that
ωFS =def −Im HFS
√ P
P
!
z̄α dzα ∧
P
−1 dzα ∧ dz̄α zα dz̄α
=locally · P −
2 .
2 1 + |zα |2
P
1 + |zα |2
By construction HFS is invariant under the natural SU(n + 1)-action on Pn , and hence so
too is ωFS .

We next verify that ωFS is indeed a Kähler form, i.e. that ωFS is closed. Following
Mumford, one can use the SU(n + 1) invariance to give a quick proof of this. In fact, given
p ∈ Pn choose an element γ ∈ SU(n + 1) such that γ(p) = p while dγp = −Id. Then for any
three tangent vectors u, v, w ∈ Tp Pn one has
dωFS (u, v, w) = γ ∗ (dωFS )(u, v, w) = dωFS (−u, −v − w),
and hence dωFS = 0. 
22It
may be useful to consider here an n-dimensional vector space W = Cn , with its standard Hermitian
product h(u, v) = t u · v. Then as w varies over W the expressions
0



ηw (u, v) = −Im h(u, v) , ηw (u, v) = −Im h(u, w)h(w, v)
define (1,
√
1)-forms η and η 0 on W , which
√
in terms of standard linear coordinates w1 , . . . , wn on W are given
by η = 2 · dwα ∧ dwα and η 0 = 2−1 ·
−1 P P
P

wα dwα ∧ wα dwα .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 81

Another approach to the Fubini–Study metric involves the Hopf map. Keeping the
notation of the previous example, consider the unit sphere
Cn+1 ⊇ S 2n+1 = S
with respect to the standard inner product h , i, with
p : S −→ Pn
the Hopf mapping. Denote by ωstd the standard Kähler form on Cn+1 , i.e.
X
ωstd = dxα ∧ dyα ,
where zα = xα + iyα are the usual complex coordinates on Cn+1 . Then ωFS is characterized
as the unique symplectic form on Pn having the property that
p∗ ωFS = ωstd | S.
(This follows from the construction in the previous example.)
Remark 6.6. (Normalization) There are various different normalizing conventions in the
literature. As shown in [3, p. 119] (where Huybrechts makes a different choice), one has
Z
ωF S = π.
P1
Noting that the Fubini-Study metric on Pn restricts to the Fubini–Study metric on any
linear subspace, it follows that π1 · ωFS represents a generator of
H 2 (Pn , Z) = Z.

The Kähler Identitites. We next study various pointwise and differential operators on a
complex manifold, and the relations that hold among them in the Kähler setting.

As in the beginnng of this section, consider a complex manifold X of (complex) dimension

n with a compatible metric g =<, > and fundamental form ω. Then all the constructions
and results from Section 3 extend to give pointwise operators on the tangent and cotangent
bundles of X and their exterior powers, as well as on the spaces of forms on X. Thus one
has the Hodge ∗-operator
∗ : Λk T ∗ X −→ Λ2n−k T ∗ X , ∗ : Ak (X) −→ A2n−k (X),
characterized by the property that
α ∧ ∗β = < α, β > · vol.
Similarly the Lefschetz operator α 7→ α ∧ ω gives
L : Λk T ∗ X −→ Λk+2 T ∗ X , L : Ak (X) −→ Ak+2 (X),
and its dual operator Λ =−1 ◦L ◦ ∗ gives
Λ : Λk T ∗ X −→ Λk−2 T ∗ X , Λ : Ak (X) −→ Ak−2 (X).
Note that while one can view L as a morphism of homolorphic vector bundles, this is not
the case for Λ. As before, we can extend these – under the same names – to the complexified
cotangent spaces.
82 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

We now bring differentiation into the picture. Specifically, we introduce three operators:
d∗ : Ak (X) −→ Ak−1 (X)
∂ ∗ : Ap,q (X) −→ Ap−1,q (X) (6.1)
∗
∂ : Ap,q (X) −→ Ap,q−1 (X)
defined as the compositions
d∗ = − ∗ d ∗
∂∗ = − ∗ ∂ ∗ (6.2)
∗
∂ = −∗ ∂∗
∗
Thus d∗ = ∂ ∗ + ∂ .

One can interpret these operators as (formal) adjoints to d, ∂, ∂ with respect to natural
inner products on the spaces of forms in question. Specifically, given a compact complex
manifold (X, g) as above define a Hermitian inner product on the spaces Ak (X), Ap,q (X) by
the formula Z
(α, β)X = α ∧ ∗β. (6.3)
X
If as in §3 we denote by < , >herm the Hermitian extension of g, then it follows from equation
(3.17) that Z
(α, β)X = < α, β >herm dvol. (6.4)
X

∗
The adjoint property of ∂ , ∂ ∗ , d∗ are given by
Proposition 6.7. For every α ∈ Ap,q−1 (X) and β ∈ Ap,q (X) one has
∗
(∂α, β)X = (α, ∂ β)X .
and analogously for ∂ ∗ and d∗ .

Proof. Note to begin with that

= ∂α ∧ ∗β + (−1)p+q−1 α ∧ ∂ ∗ β.


∂ α ∧ ∗β
Now d = ∂ on forms of type (n, n − 1) and hence by Stoke’s theorem
Z
p+q
(∂α, β)X = (−1) α ∧ ∂ ∗ β.
X

On the other hand, since ∂η = ∂η, and since ∗2 = (−1)k on k-forms, one has
− ∗ ∗∂ ∗ β = (−1)p+q ∂ ∗ β.
Therefore Z Z
∗ p+q
(α, ∂ β)X = α ∧ (− ∗ ∗∂ ∗ β) = (−1) α ∧ ∂ ∗ β,
X X
as required. 
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 83

We next study the commutation relations among these operators when X is Kähler.
The first remark is
Proposition 6.8. Assume that X is Kähler. Then
[∂, L] = [∂, L] = 0 (6.5)
∗
[∂ , Λ] = [∂ ∗ , Λ] = 0. (6.6)

Proof. Note that ∂ω and ∂ω are respectively the (2, 1) and (1, 2) components of dω, and
hence dω = 0 if and only if ∂ω = ∂ω = 0. Therefore
∂(α ∧ ω) = ∂α ∧ ω , ∂(α ∧ ω) = ∂α ∧ ω,
which gives (6.5). For (6.6) one first uses that Λ = ∗−1 L∗ (Proposition 3.14) and the fact
that ∗2 = (−1)k on k-forms to prove that
∗
[∂ , Λ] = − ∗ [∂, L]∗,
and the assertion follows. 

The deeper identity involves the commutation relations between Λ and ∂ and ∂.
Theorem 6.9. Assume that (X, g) is Kähler. Then
√ √ ∗
[Λ, ∂] = −( −1) ∂ ∗ , [Λ, ∂] = ( −1) ∂ (6.7)
∗ √ √
[L, ∂ ] = = −( −1) ∂ , [L, ∂ ∗ ] = ( −1) ∂ (6.8)

Turning to a sketch of the proof of the theorem, note that the assertion is an identity
between first-order differential operators. Therefore thanks to Theorem 4.6 it suffices to prove
it in a neighborhood of the origin in Cn with the standard (constant) metric. While one could
work on Cn with compactly supported forms, it is convient to keep dealing with compact
manifolds. Therefore, as do Griffiths–Harris, we will work in the following paragraphs with
a complex torus T = Cn /Λ with the flat Kähler metric coming from Cn . I will closely follow
Schnell’s presentation in [4].

In the sequel, we write dzI , dz J etc for the forms on T induced by the indicated trans-
lation invariant forms on Cn . We assume that the flat metric g is normalized so that
volg (T ) = 1, and we denote by h the corresponding Hermetian metric on T . Note that
h(dzI ∧ dz J , dzI ∧ dz J ) = 2|I|+|J| ,
and hence
(dzI ∧ dz J , dzI ∧ dz J )T = 2|I|+|J| .

Given an index 1 ≤ i ≤ n, we define operators:

ei : Ap,q (X) −→ Ap+1,q (T ) , α 7→ dzi ∧ α
(6.9)
ei : Ap,q (X) −→ Ap,q+1 (T ) , α 7→ dz i ∧ α.
84 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Thus √ √
−1 X  −1  X 
L(α) = dzi ∧ dz i ∧ α = ei ei (α).
2 2
One next defines the adjoint
e∗i : Ap,q (T ) −→ Ap−1,q (T )
by the condition that e∗i be adjoint with respect to <, >herm of the point-wise operator given
by the same formula as ei , so that
(ei α, β)T = (α, e∗i β)T .
A combinatorial argument now proves the:
Lemma 6.10. Fix an index 1 ≤ i ≤ n.

(i). If i 6∈ J then
e∗i (dzJ ∧ dz K ) = 0 , e∗i dzi ∧ dzJ ∧ dz K


= 2dzJ ∧ dz K .
(ii).
(
2 · Id if i = j
ej e∗i + e∗i ej =
0 otherwise

See [4, Lemma 22.1] for the proof.

Next, define differential operators

X X ∂φJ,K
∂i : Ap,q (T ) −→ Ap,q (T ) , φJ,K dzI ∧ dz K 7→ dzJ ∧ dz K
∂zi
X X ∂φJ,K (6.10)
∂ i : Ap,q (T ) −→ Ap,q (T ) , φJ,K dzJ ∧ dz K 7→ dzJ ∧ dz K
∂z i
Thus X X
∂ = ∂i ei , ∂ = ∂ i ei . (6.11)
Evidently these operators commute with each other, and with ei and e∗i . Furthermore,
integrating by parts shows:
Lemma 6.11. The adjoints of ∂i and ∂ i are given by
∗
∂i∗ = −∂ i , ∂ i = −∂i
in the sense that
(∂i α, β)T = −(α, ∂ i β)T , (∂ i α, β)T = −(α, ∂i β)T . 

We now prove the identity

√ ∗
[Λ, ∂] = −1 ∂ ,
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 85
√
assertion in (6.7) being similar. Since L = 2−1 ei ei , taking adjoints yields
P
the other
√
Λ = − 2−1 e∗i e∗i . Hence:
P
√ X
Λ∂ − ∂Λ = − 2−1 e∗i e∗i ∂j ej − ∂j ej e∗i e∗i


i,j
√ X
−1
∂j ei ∗ e∗i ej − ej e∗i e∗i


= − 2
i,j
√ X
−1
∂j e∗i e∗i ej + ej e∗i .


= − 2
i,j

Using Lemma 6.10 (ii), it follows that

√ X
[Λ, ∂] = −( −1) ∂j e∗j . (*)
P
On the other hand, taking adjoints in the relation ∂ = ∂ j ej and using Lemma 6.11 we see
that X ∗
X ∗
− ∂j e∗j = ∂ j e∗j = ∂ .
√ ∗
Thus (*) shows that [Λ, ∂] = ( −1) ∂ , as required. Finally (6.8) follows from (6.7) by
taking adjoints.

Laplacians and harmonic forms. Let (X, g) be a compact complex manifold with a
compatible metric. Associated to each of the operators d, ∂, ∂ ∗ there is a Laplace operator
whose kernel consists (by definition) of harmonic forms. The Hodge theorem will assert
that any cohomolgy class has a unique harmonic representative. In general there need be
no particular connection between d-, ∂ and ∂ ∗ -harmonic forms, but in the case of Kähler
manifolds the three Laplacians essentially coincide. As we shall see, fact has many remarkable
consequences.

We start by defining the three Laplace operators in question. As in the previous para-
graph, let (X, g) be a compact complex manifold with a compatable metric.
Definition 6.12. We define the Laplace operators associated to d, ∂, ∂ to be the operators
∆d : Ak (X) −→ Ak (X)
∆∂ : Ap,q (X) −→ Ap,q (X) (6.12)
∆∂ : Ap,q (X) −→ Ap,q (X)
given by:
∆d = dd∗ + d∗ d
∆∂ = ∂ ∂ ∗ + ∂ ∗ ∂ (6.13)
∗ ∗
∆∂ = ∂ ∂ + ∂ ∂.

We will be particularly interested in those forms killed by the Laplacian:

86 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Definition 6.13. A form α ∈ Ak (X) is said to be d-harmonic if

∆d α = 0,
and similarly α ∈ Ap,q (X) is ∂- or ∂-harmonic if
∆∂ α = 0 or ∆∂ α = 0.
We denote by
Hdk (X) = Hdk (X, g) ⊆ Ak (X)
H∂p,q (X) = H∂k (X, g) ⊆ Ap,q (X) (6.14)
H∂p,q (X)
= H∂p,q (X, g)
⊆ Ap,q (X)
the spaces of harmonic forms for the indicated Laplacians.

We emphasize that these spaces depend on the chosen metric on X. It will turn out that
they are finite dimensional. Note that in the case of d, one only needs (X, g) to be a compact
oriented Riemannian manifold.
Example 6.14. To explain the terminolgy, consider the Laplace operator ∆d acting on
functions f ∈ A0c (Rn ) where Rn is given the usual flat Euclidean metric.23 Then d∗ f = 0
since f is a zero-form, hence with Euclidean coordinates x1 , . . . , xn one has
X 
∗ ∂f
∆d (f ) = d d f = − ∗ d ∗ dxi
∂xi
 X 
∂f
= −∗d ± dx1 ∧ . . . ∧ dxi ∧ . . . ∧ dxn
c
∂xi
X 2 
∂ f
= −∗ dx1 ∧ . . . ∧ dxn
∂x2i
X 2 
∂ f
=− .
∂x2i
So up to a sign, ∆d (f ) is just the classical Laplacian of f . One can compute ∆∂ (f ) and
∆∂ (f ) similarly (see [2, p. 83]).
Proposition 6.15. Let X be a compact complex manifold. Then a form α ∈ Ak (X) is
d-harmonic if and only if
dα = d∗ α = 0,
with the analogous condition for a form α ∈ Ap,q (X) to be ∂- or ∂-harmonic.

Proof. We use Proposition 6.7 (or more precisely its analogue for d∗ ). In fact, that result
shows that
(∆d α , α)X = (d d∗ α, α)X + (d∗ dα, α)X
= (d∗ α, d∗ α)X + (dα, dα)X .
Since both terms on the right are in any event non-negative, the assertion follows. 
23Assume for simplicity that n is even to avoid sign worries involving the definition of d∗ given above.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 87

The crucial point for us is that on a Kähler manifold, the three different Laplacians that
we have defined essentially coincide.
Theorem 6.16. Let (X, g) be a compact Kähler manifold. Then
1
∆∂ = ∆∂ = ∆ ,
2 d
∗
and ∆d commutes with the operators ∗, ∂, ∂, ∂ ∗ , ∂ , L and Λ, and it preserves the decomposi-
tion into type.

Proof. We first show that ∆∂ = ∆∂ . Using equation (6.7) in Theorem 6.9 and recalling that
∂ and ∂ anticommute we find:
∆∂ = ∂ ∗ ∂ + ∂ ∂ ∗
√

= ( −1) [Λ, ∂] ∂ + ∂ [Λ, ∂]
√

= ( −1) Λ∂∂ − ∂Λ∂ + ∂Λ∂ − ∂ ∂Λ
√




= ( −1) Λ∂ ∂ − ∂[Λ, ∂] + ∂ ∂Λ + [∂, Λ]∂ + Λ∂ ∂ − ∂ ∂Λ
√  √ ∗
  √ ∗

= ( −1) Λ∂ ∂ − ( −1)∂ ∂ − ∂ ∂Λ + −( −1)∂ ∂ + Λ∂ ∂ − ∂ ∂Λ
√  √ ∗ √ ∗

= ( −1) −( −1)∂ ∂ − ( −1)∂ ∂
= ∆∂ .

Note next that

∗ ∗
∂ ∂ + ∂ ∂ = 0.
In fact, by Theorem 6.9
√  ∗ ∗

( −1) ∂ ∂ + ∂ ∂ = ∂[Λ, ∂] + [Λ, ∂]∂
= ∂Λ∂ − ∂Λ∂
= 0.
Conjugating, it follows that likewise
∂ ∂ ∗ + ∂ ∗ ∂ = 0.
We now compute:
∗ ∗
∆d = (∂ + ∂)(∂ ∗ + ∂ ) + (∂ ∗ + ∂ )(∂ + ∂)
∗ ∗

= ∆∂ + ∆∂ + ∂ ∂ + ∂ ∂ + ∂ ∂ ∗ + ∂ ∗ ∂

= ∆∂ + ∆ ∂ .
The proofs that ∆d commutes with all the holomorphic and anti-holomorphic operators
are similar, and are left to the reader. That ∆d commutes with the projection operators
π p,q : Ak (X) −→ Ap,q (X) follows from the facts that ∆d = 2∆∂ and that ∆∂ preserves the
type decomposition since it is the compostion of operators that do. 
88 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

These identities already have an interesting corollary. Recall that a holomorphic p-form
on a complex manifold X is a (p, 0)-form η ∈ Ap,0 (X) such that ∂η = 0. Equivalently, η has
a local expression of the form X
η =locally fI dzI
where the fI are holomorphic. In complex dimension 1 any holomorphic (1, 0)-form is closed
simply because d = ∂ in (1, 0)-forms. However in higher dimensions there is no local reason
that a holomorphic form should be closed: eg η = z dw is not closed on C2 . However in the
global Kähler setting a holomorphic form is automatically closed:
Corollary 6.17. If X is a compact Kähler manifold, then any holomorphic form η on X is
closed.

We will see later (Exercise 6.4) that if η 6= 0, then moreover η cannot be exact.

Proof. We have ∂ ∗ η = 0 for reasons of type, hence η is ∂-harmonic thanks to Proposition

6.15. Therefore η is d-harmonic, and then the same Proposition shows that it is d- and
d∗ -closed. 

We close this subsection with a proposition that provides yet another characterization
of harmonic forms. It will be used in the next subsection to give an intuitive heuristic for
the Hodge decomposition theorem.
Proposition 6.18. With X as above, consider the Hermitian inner product (6.3), and let
α ∈ Ak (X)
be a d-closed form on X. Then α is d-harmonic if and only if α has minimial length in its
deRham cohomology class, i.e.
α ∈ Hdk (X) ⇐⇒ ||α + dη||2 > ||α||2
for every η ∈ Ak−1 (X) with dη 6= 0. The analogous statements hold for ∂ and ∂ cohology
and ∆∂ and ∆∂ -harmonic forms.

Proof of Proposition 6.18. First assume that α is harmonic, so that dα = d∗ α = 0. Then for
any η ∈ Ak−1 (X) one has
|| α + dη ||2 = ||α||2 + (α, dη)X + +(dη, α) + X ||dη||2
= ||α||2 + (d∗ α, η)X + (η, d∗ α)X + ||dη||2
= ||α||2 + ||dη||2 ,
so α has minimal norm. Conversely, if α has minimal norm, then taking η = t d∗ α in the
above we find that
||α + t dd∗ α|| = ||α||2 + 2t||d∗ α||2 + t2 ||dd∗ α||2 .
This has a minimum at t = 0 if and only if ||d∗ α|| = 0, ie iff α is harmonic. 
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 89

The Hodge Theorem. We now turn to the Hodge theorem and its consequences.

Let (X, g) be a compact connected complex manifold with a compatible complex struc-
ture. The Hodge theorem – which we will momentarily state precisely – asserts that there
is a unique harmonic representative in each d−, ∂− or ∂− cohoomology class. The proof
of the theorem is analytic in nature – it’s ultimately a consequence of regularity theorems
for elliptic PDE – and we do not attempt to give anything like a real explanation. However
there is an enjoyable heuristic argument (which most of us learned from Griffiths-Harris)
suggesting why the statement is not unreasonable.

The starting point is Proposition 6.18, which asserts that a form is harmonic (with
respect to any of the three Laplacians) if and only it has minimal length in its (d-, ∂- or ∂-)
cohomology class. Focusing for concreteness on the d-Laplacian ∆d and DeRham cohomolgy,
imagine – which unfortunately is not the case – that Ak (X) were a Hilbert space with respect
to the inner product (, )X and that
Im d : Ak−1 (X) −→ Ak (X) ⊆ Ak (X)

were a closed subspace. Then we could find a unique element of minimal lengh in the affine
subspace
[α] = α + Im(d),
which according to the Proposition would be the unique harmonic form in this cohomology
class. This provides at least some intuitive reason to imagine that something like the Hodge
theorem could be true.

Here is the formal statement of the Hodge decomposition theorem. To clarify where this
is and isn’t used, for the moment we do not impose the Kähler condition.
Theorem 6.19. (Hodge Theorem). Let (X, g) be a compact complex manifold with a
compatible metric. Then the three spaces
Hdk (X, g) ⊆ Ak (X) and H∂p,q (X, g) , H∂p,q (X, g) ⊆ Ap,q (X)
of harmonic forms are finite dimensional. Morover:

(i). There is an orthogonal direct sum decomposition

Ak (X) = d Ak−1 (X) ⊕ Hdk (X, g) ⊕ d∗ Ak+1 (X).
(ii). Similarly, there are orthogonal direct sum decompositions
Ap,q (X) = ∂ Ap−1,q (X) ⊕ H∂p,q (X, g) ⊕ ∂ ∗ Ap+1,q (X)
∗
Ap,q (X) = ∂ Ap,q−1 (X) ⊕ H∂p,q (X, g) ⊕ ∂ Ap,q+1 (X).

The orthogonality in the theorem is of course with respect to the Hermitian inner produce
(6.3). We will say a few words in Remark ?? about the general PDE facts underlying the
theorem.
90 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Corollary 6.20. In the situation of the Theorem, there are canonical isomorphisms
∼
=
Hdk (X, g) −→ HdR
k
(X)
∼
=
H∂p,q (X, g) −→ H∂p,q (X)
∼
=
H∂p,q (X, g) −→ H∂p,q (X).
Lemma 6.21. One has
ker d : Ak (X) −→ Ak+1 (X) = Hdk (X, g) ⊕ d Ak−1 (X),


(6.15)
with analogous statements for ker ∂ and ker ∂.

Proof. Evidently the RHS of (6.15) contains the LHS. For the reverse inclusion, suppose
that dα = 0. By the Hodge decomposition one can write α = dα1 + h + d∗ α2 , where h is
harmonic. It suffices to show that α2 = 0. Now since dα = 0, we have dd∗ α2 = 0, but then
||α22 || = (d∗ α2 , d∗ α2 ) = (dd∗ α2 , α2 ) = 0,
as required. 

Proof of Corollary 6.20. We prove the first statement. Since a harmonic form is d-closed,
there is a natural map
Hdk (X, g) −→ HdR
k
(X) , α 7→ [α]. (*)
It follows from the orthogonality of the Hodge decomposition that a non-zero harmonic form
cannot be exact, hence (*) is injective. Moreover, equation (6.15) implies that any closed
form is cohomologous to a (unique) harmonic form, hence (*) is surjective. 

Assume now that (X, g) is compact Kähler. Then the three Laplacians in play all
coincide, so we expect the Hodge decomposition to have particularly striking consequences.
This is very much the case. In fact, by the deRham theorem we have
H k (X, C) = Hdk (X)
and we write H p,q (X) = H∂p,q (X) for the ∂- cohomogy of X, so that
H p,q (X) = H q (X, ΩpX )
thanks to the Dolbeaut theorem.

Then we have the famous

Theorem 6.22. (Hodge Decomposition) If X is a compact Kähler manifold, then there
is a natural decomposition
H k (X, C) ∼
M
= H p,q (X), (6.16)
p+q
which is independent of the Kähler metric on X. Moreover there are isomorphisms
H q,p (X) = H p,q (X)
(6.17)
H p,q (X) = H n−q,n−p (X).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 91

Proof. Fix a Kähler metric g on X, and consider the resulting spaces of harmonic forms. In
the first place, by Corollary 6.20 we have isomorphisms
∼
=
H∂p,q (X, g) −→ H p,q (X),
∼
=
Hdk (X, g) −→ H k (X, C).
Furthermore, since ∆d commutes with taking (p, q)-components (Theorem 6.16), we have a
decomposition
Hdk (X, g) = ⊕p+q=k Hdp,q (X, g).
On the other hand, thanks to Theorem 6.16 it is the same to be harmonic for d or for ∂, and
therefore Hdp,q (X, g) = H∂p,q (X, g). Putting this together, we get isomorphisms
H k (X, C) ∼
= Hdk (X, g) = ⊕ H∂ (X, g) ∼
p,q
= ⊕ H p,q (X).
It remains to show that the resulting isomorphism between the outer terms is independent
of the choice of Kahler metric. Fixing a second Kähler metric g1 , this amount to showing
that if we have harmonic forms
α ∈ Hp,q (X, g) , α1 ∈ Hp,q (X, g1 )
representing the same cohomology class in H p,q (X) then α and α1 represent the same class
in H p+q (X, C). The hypothesis means that
α − α1 = ∂β,
and by the directness of the decomposition in Theorem 6.19 (ii) for ∂, it follows that α − α1
is orthogonal to H∂p,q (X, g) = Hdp,q (X, g). Since d(α − α1 ) = 0, it follows from Proposition
6.21 that α − α1 = dγ, as required. Finally if α ∈ Hp,q (X), then
α ∈ Hq,p (X) , ∗α ∈ Hn−q,n−p (X)
(Exercise 6.5), which gives the isomorphisms (6.17). 

We conclude this section with a useful result, called the ∂∂- Lemma.
Proposition 6.23. Let X be a compact Kähler manifold, and let
α ∈ Ap,q (X)
be a d-closed (p, q)-form. Then the following are equivalent:

(i). The form α is d-exact, i.e. α = dβ for some β ∈ Ap+q−1 (X);

(ii). The form α is ∂-exact, i.e. α = ∂β for some β ∈ Ap−1,q (X);
(iii). The form α is ∂-exact, i.e. α = ∂β for some β ∈ Ap,q−1 (X);
(iv). The form α is ∂∂-exact, i.e. α = ∂∂β for some β ∈ Ap−1,q−1 (X);
(v). The form α is perpendicular to the space Hp,q (X, g) of harmonic forms for an arbi-
trary Käher metric g on X.24
24Since we are in the Kähler setting, the same (p, q)-forms are harmonic for ∆d , ∆∂ and ∆∂ , so we don’t
need to specify which we refer to here.
92 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Proof. Since α is a form of pure type (p, q), it is d closed if and only if it is both ∂- and
∂-closed. Therefore it follows from Theorem 6.19 that each of the other conditions implies
(v), and clearly (iv) implies (i) – (iii). So it suffices to show that (v) ⇒ (iv). Since ∂α = 0
and α is perpendicular to H∂p,q (X), Lemma 6.21 for ∂ implies that α = ∂β for some β. Now
use the Hodge decomposition for ∂ to write
∗
β = ∂γ1 + γ2 + ∂ γ3 ,
where γ2 is harmonic. Therefore
∗
α = ∂β = ∂∂γ1 + ∂∂ γ3
∗
= −∂∂γ1 − ∂ ∂γ3 .
∗
On the other hand, ∂α = 0, so ∂∂ ∂γ3 = 0, which implies that
∗ ∗
(∂ ∂ ∂γ3 , ∂γ3 )X = ||∂ ∂γ3 ||2 = 0,
and hence
∗ ∗
−∂∂ γ3 = ∂ ∂γ3 = 0.
Thus α = ∂∂γ1 , as required. 

The Lefschetz theorems. We start by discussing some topological consequences of the

Hodge theorem.

Let (X, g) be a compact Kähler manifold, with Kähler form ω. Denote as usual by L and
Λ the Lefschetz operator and it’s dual. Since these commute with the Laplacian ∆ = ∆d ,
they give rise to homomorphisms
L : Hp,q (X) −→ Hp+1,q+1 (X) , Hp,q (X) −→ Hp−1,q−1 (X),
and hence to
L : H p,q (X) −→ H p+1,q+1 (X) , Λ : H p,q (X) −→ H p−1,q−1 (X).
Note that L is given by cup product with [omega] ∈ H 1,1 (X), and hence depends only on
this class. One can show that the same is true of Λ.

The first main result is the

Theorem 6.24. (Hard Lefschetz Theorem) Let X be a compact Kähler manifold of
complex dimension n. Then the (n − k)-fold iteration of the Lefschetz operator gives an
isomorhism
∼
=
Ln−k : H k (X, C) −→ H 2n−k (X, C).

Note that the two groups appearing here are Poincaré dual, so the deep fact is not that there
exists an isomorphism between the groups, but rather that (n − k)-fold iteration of L does
the job.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 93

Proof. Thanks to the Hodge decomposition it suffices to show that

Ln−k : Hk (X) −→ H2n−k (X) (*)
is an isomorphism, and since both source and target have the same dimension it is enough
to prove that it is injective. But this follows from our pointwise analysis in §3. In fact,
Theorem 3.16 implies that
Ln−k : Λk TC∗ X −→ Λ2n−k TC∗ X
is an isomorphism of smooth vector bundles, and hence
Ln−k : Ak (X) −→ A2n−k (X)
is bijective. Therefore (*) is indeed injective. 

We turn next to the Lefschetz decomposition. The primitive cohomology of X is defined

to be
P k (X, C) = ker Λ : H k (X, C) −→ H k−2 (X, C)

P p,q (X, C) = ker Λ : H p,q (X, C) −→ H p−1,q−1 (X, C)

Theorem 6.25. (Lefschetz decomposition). Let X be a compact Kähler manifold of

dimension n. Then
M
H k (X, C) = Li P k−2i (X, C)
i≥0
M
H p,q (X) = Li P p−i,q−i (X).
i≥0

Moreover
P k (X) = ker Ln−k+1 : H k (X, C) −→ H 2n−k+2 (X, C) ,

and similarly for P p,q .

Proof. By the Hodge decomposition theorem, primitive classes are (uniquely) represented by
primitive harmonic forms:
P k (X) =def ker Λ : Hk (X) −→ Hk−2 (X) ,


(*)
and similarly for P p,q . So it is equivalent to prove the Lefschetz decomposition for harmonic
forms. For this one can either reduce to Theorem 3.16, or one can note that L and Λ a
generate an SL2 action on primitive forms, and argue as in §3. We leave details to the
reader. 
Remark 6.26. Note that the condition (*) defining primitive harmonic forms operates
pointwise. Therefore if α ∈ P k is a primitive harmonic form, then α(x) ∈ Λk T ∗ X is a
primitive vector for every x ∈ X (although possibly α(x) = 0 at some x ∈ X.

Finally, we have the Hodge-Riemann bilinear relations.

94 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Theorem 6.27. Let X be a compact Kähler manifold of dimension n, with Kähler class ω.
Let 0 6= α ∈ P p,q (X). Then
√
Z
(p+q)(p+q−1)
p−q
( −1) (−1) 2 α ∧ α ∧ ω n−(p+1) > 0.
X

Proof. It is enough to prove the statement when α is a primitive harmonic form. Fix a point
x ∈ X at which α(x) 6= 0. Then thanks to Remark 6.26 the assertion follows from the
pointwise statement Theorem 3.27 established in §3. 

We now turn to the remarkable Lefschetz (1, 1)-theorem, which describes the holomor-
phic line bundles on a compact Kähler manifold. Recall first that isomorphism classes of
smooth complex line bundles on X are classified by H 2 (X, Z): this arises from the isomor-
phism
∼
=
H 1 (X, A∗ ) −→ H 2 (X, Z) , L 7→ c1 (L) (*)
∗
deduced from the exponential sequence 0 −→ Z −→ A −→ A −→ 0 where A is the sheaf
of smooth C-valued functions on X. Here c1 (L) ∈ H 2 (X, Z) is the first Chern class of a
line bundle L: for now one can take (*) as the definition of c1 (L), but we will discuss other
manifestations shortly.

It is then natural to ask the important

Question 6.28. Which elements in H 2 (X, Z) are the first Chern classes of holomorphic line
bundles on X?

For the answer, one considers the subgroup

H 1,1 (X, Z) ⊆ H 2 (X, Z)
defined as the set of all integral classes which, when mapped to H 2 (X, C), have type (1, 1)
under the Hodge decomposition of H 2 (X, C). In other words,
H 1,1 (X, Z) = γ ∈ H 2 (X, Z) | ι(γ) ∈ H 1,1 (X)


where ι : H 2 (X, Z) −→ H 2 (X, C) is the natural map.25 In other words, if we ignore torsion
and pretend that H 2 (X, Z) embeds in H 2 (X, C), then one has the more picuresque relation
H 1,1 (X, Z) = H 2 (X, Z) ∩ H 1,1 (X).
Theorem 6.29. (Lefschetz (1,1) theorem) A class γ ∈ H 2 (X, Z) first Chern class of a
holomorphic line bundle if and only γ ∈ H 1,1 (X, Z).
25Thus ι embeds H 2 (X, Z) modulo its torsion subgroup into
H 2 (X, C) = H 2 (X, Z) ⊗Z C.
We note that some authors define H 1,1 (X, Z) to be the intersection
 
im H 2 (X, Z) −→ H 2 (X, C) ∩ H 1,1 (X).
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 95

∗
Proof. We consider the exponential seqence 0 −→ Z −→ OX −→ OX −→ 0, which gives rise
to an exact sequence
∗ c
H 1 (X, OX 1
) −→ H 2 (X, Z) −→ H 2 (X, O). (6.18)
Our question is to understand the image of the map
∗
Pic(X) = H 1 (X, OX ) −→ H 2 (X, Z),
so the Theorem is equivalent to the assertion that
 
H 1,1 (X, Z) = ker H 2 (X, Z) −→ H 2 (X, O) . (*)
We assert:
Claim 6.30. Under the Dolbeaut isomorphism
H 2 (X, OX ) = H 0,2 (X),
the map H 2 (X, Z) −→ H 2 (X, OX ) coming from the exponential sequence associates to a
class γ ∈ H 2 (X, Z) the (0, 2)-component of its image in H 2 (X, C). (Exercise 6.9.)

Granting this, one immediately gets the inclusion ⊆ in (*), and the other inclusion follows
from the observation that since the image of H 2 (X, Z) −→ H 2 (X, C) consists of real classes,

2,0
ι(γ)0,2 = 0 ⇐⇒ ι(γ)0,2 = ι(γ) = ι(γ)2,0 = 0.


To complete the picture, we say a word about the Picard torus of a compact Kähler
manifold. Define  
Pic0 (X) = ker Pic(X) −→ H 2 (X, Z)
to be the group of topologically trivial holomorphic line bundles on X. This fits into the
exact sequence
0 −→ H 1 (X, Z) −→ H 1 (X, O) −→ Pic0 (X) −→ 0.
Now it follows from the Dolbeaut isomorphism that H 1 (X, OX ) = H 0,1 (X), and as above
the map H 1 (X, Z) −→ H 0,1 (X) occuring here is the composition
H 1 (X, Z) −→ H 1 (X, C) −→ H 0,1 (X)
coming from the Hodge decomposition of H 1 . Now note that
rank H 1 (X, Z) = b1 (X) = 2 · h01 (X) = 2 dimC H 1 (X, OX ).
This suggests that one might hope that H 1 (X, Z) sits as a lattice inside H 1 (X, OX ). In fact,
this is the case (Exercise 6.10). Therefore
Pic0 (X) = H 1 (X, OX )/H 1 (X, Z)
has in the natural way the structure of a complex torus, called the Picard torus of X.
Thus we have the beautiful picture that the set of isomorphism classes of holomorphic line
bundles having a given topological type is parametrized by a torus of complex dimension
h0,1 (X) = 21 b1 (X).
96 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Exercises for Section 6.

Exercise 6.1. Prove the assertion of Example 6.3.
Exercise 6.2. Let ω be a real closed (1, 1)-form on a complex manifold X. Assume that ω
is positive in the sense that it can locally be written in the form
√
−1 X
ω =locally hij (z) · dzi ∧ dz j ,
2
where hij (z) is a smooth family of positive definite Hermitian matrices on X. Prove that
then X carries a Kähler metric h of which ω is the corresponding fundamental 2-form.
Exercise 6.3. Prove that on a compac Kähler manifold, the Kähler form ω is harmonic.
Exercise 6.4. (a). Prove that a non-zero holomorphic form on a compact ähler manifold
is never exact.
(b). Give an example of a non-zero exact holomorphic form on C2 .
Exercise 6.5. Let X be a compact Kähler manifold, and let α ∈ Hp,q (X). Prove that then
α ∈ Hq,p (X) , ∗α ∈ Hn−q,n−p (X).
Exercise 6.6. (Functoriality of Hodge decomposition). . Prove that the Hodge de-
composition (6.16) is functorial in the sense that if f : Y −→ X is a holomorphic map
of compact Kähler manifolds, then the pull-back mapping f ∗ : H k (X, C) −→ H k (Y, C)
coincides under the Hodge decomposition with the direct sum of the pullbacks
f ∗ : H p,q (X) −→ H p,q (Y ).
(Hint: First use harmonic representatives to prove that this is true for an inclusion Y ⊆ X
of Y as a submanifold of X, and then deduce the general case from this by considering the
graph of f .)
Exercise 6.7. (The Kähler cone). Let X be a compact Kähler manifold. The Kähler
cone of X is the open convex cone
K ⊆ H 1,1 (X, R)
consisting of the classes of Kähler metrics on X.

(a). Prove that this is indeed an open convex cone. (Cf. [3, p. 130].)
(b). Describe the Kähler cone explitly for X = P1 × P1 .

We remark that in all but the simplest cases it can be difficult to compute this cone explicitly,
especially if one wants to know the subspace spanned by the integral classes. For example,
if X is a compact Riemann surface of genus g > 2, then K ∩ H 2 (X, Z) can depend on the
intrinsic geometry of X, and in fact it is not known what K ∩ H 2 (X, Z) is for “general” X.
See [PAG, §1.5.B] for a survey of the algebro-geometric analogue of this question. 
Exercise 6.8. Let X be a compact Kähler maifold.

(a). Prove that the odd betti numbers of X are even.

 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 97

(b). Prove that π1 (X) cannot be isomorphic to the free group on three generators.
(c). Prove that π1 (X) cannot be isomorphic to the free group on two generators. [Hint:
Otherwise there would be a two-sheeted covering space of X whose π1 is free on three
generators, and such a covering is again a compact Kähler manifold.]
Exercise 6.9. Prove Claim 6.30. (See [3, Lemma 3.3.1].)
Exercise 6.10. Using the facts that
H 1 (X, C) = H 1,0 (X) ⊕ H 0,1 (X) , H 0,1 = H 1,0 ,
show that the image of H 1 (X, Z) −→ H 0,1 (X) is a lattice. (See [3, Cor. 3.3.6].)
Exercise 6.11. (The case of complex tori) Let Λ ⊆ V be a lattice in an n-dimensional
complex vector space V . We consider the complex torus X = V /Λ with the flat (constant)
metric on V . As usual, we may consider forms on V that are invariant under translation by
Λ to be forms on X.

(a). Prove that the harmonic forms on X are just the constant ones of the sort
X
aI,J dzI ∧ dz J , (*)

where aI,J ∈ C. Deduce that

Hp,q (X) = Λp,q V ∗ ,
and deduce that the Hodge decompostion takes the form
Λk VC = ⊕Λp,q V ∗ .
(b). Choose a basis
λ1 , . . . , λ2n ∈ V
for Λ, and denote by λi ∧ λj the two (real) dimensional “subtorus” of X spanned by
λi , λj . Given a (1, 1)-form
X
η = aij dzi ∧ dz j ∈ Λ1,1 V ∗ ,

show that Z
1,1
η ∈ H (X, Z) ⇐⇒ η ∈ Z ∀i, j. (*)
γi ∧γj

(c). Show by explicit computation that there exist Λ ⊂ V that determine tori X for which
H 1,1 (X, Z) = 0. [Hint: By a suitable choice of coordinates one can suppose that
λn+1 = e1 , . . . , λ2n = en ,
where ei are the standard basis vectors. Then compute the integrals appearing in
(*) in terms of the coordinates of λ1 , . . . , λn , and check that for sufficiently general
choices of these λi there are no solutions to (*).]
98 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

7. Positivity for Line Bundles and the Kodaira Theorems

In this final section we study the notion of a positive line bundle on a Kähler manifold,
and we prove the Kodiara vanishing and embedding theorems.

Positive Line Bundles. We start with the notion of positivity for a (1, 1)-form on a com-
plex vector space.
Definition 7.1. Let (V, J) be a real vector space with an almost complex structure. A
(1, 1)-form η ∈ Λ1,1 V ∗ is positive if
η(v, Jv) > 0 for all 0 6= v ∈ V.
Equivalently, with notation as in §3, η is positive if one can write
√
−1 X
η = hij z i ∧ z j
2
where hij is a positive definite Hermitian metric. 

Yet another way to phrase the definition is to require that if W ⊆ V is any J-stable subspace
of real dimension 2 – so that (W, J|W ) is a two-dimensional almost complex vector space –
then η|W = c volW for some c > 0.

Similarly:
Definition 7.2. Let X be a complex manifold, with J the corresponding almost complex
structure, and let η be a (1, 1)-form on X. We say that η is positive if
ηx ∈ Λ1,1 Tx∗ X
is a positive form for every x ∈ X.

Equivalently, one asks that

√
−1 X
η =locally hij (z) dzi ∧ dz j
2
where hij (z) is a Hermitian matrix of smooth functions that is positive definite at every
point.
Example 7.3. Let (X, g) be a complex manifold with a compatible Riemannian metric,
and let ω ∈ A1,1 (X) be the associated (1, 1)-form. Then ω is positive, and conversely any
positive (1, 1)-form arises in this manner (cf Exercise 6.2).

The plan now is to introduce now a positivity notion for a holomorphic line bundle L
by asking in effect that c1 (L) be representable by a positive (1, 1)-form.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 99

Let L be a holomorphic line bundle on a X. A hermitian metric h on L is a smoothly

varying (positive definite) hermitian metric on all the fibres of L, and a hermitian holomor-
phic line bundle (L, h) is a holomorphic line bundle together with the choice of a hermitian
metric on L. (Hermitian metrics on vector bundles of higher rank are defined similarly.)

Suppose that (L, h) is a hermitian line bundle, and consider a holomorphic section
s ∈ Γ(U, L) of L on an open subset U ⊂ X. Then we take the fibre-wise norm square of s
to get a non-negative function
|s|2 = |s|2h = h(s, s)
on U . If L|U = U × C is trivial, then conversely a hermitian metric on L|U is specified by
giving a positive smooth function
h : U −→ R>0 (7.1)
which we declare to be the (square) length of the section x −→ 1 ∈ C. In general, if L is
described by transition data (Ui , gij ) then a metric is given by smooth functions
hi : Ui −→ R>0
such that
hi = |gij |2 hj on Uij . (7.2)
A simple argument with partitions of unity shows that any holomorphic (or, for that matter,
smooth) line bundle admits (many different) hermitian metrics. (Exercise 7.1.)

Let (L, h) be a hermitian line bundle described by local data (Ui , hi ) as in (7.2). We
claim that the (1, 1)-forms
ti =def −∂∂ log hi
patch together to give a globally defined form
Θ(L, h) ∈ A1,1 (X),
which is called the curvature form of (L, h). In fact, on Ui ∩ Uj one has
log hi = log |gij |2 + log hj ,
and
∂∂ log |gij |2 = ∂∂ gij g ij

= ∂∂ log gij − ∂∂ log g ij

= 0
since gij is holomorphic. Evidently Θ(L, h) is closed.
Remark 7.4. Let sU ∈ Γ(U, L) be a local section which is nowhere-vanishing on U . Then
one can alternatively define
Θ(L, h)U = −∂∂ |sU |2 ,
it following from a computation as above that these local expressions are independent of the
choice of sU .
Remark 7.5. We are avoiding here a systematic discussion of curvature of vector bundles.
For this, see eg [3, Chapter 4] or [2].
100 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

A basic fact is that Θ(L, h) essentially represents the first Chern class of L:
Proposition 7.6. One has
√
−1
c1 (L) = 2π
Θ(L, h) ∈ H 2 (X, C).

For a proof, see [].

Now we come to the basic definition

Definition 7.7. A holomorphic line bundle on a compact complex√manifold is positive (in
−1
the sense of Kodaira) if it carries a hermitian metric h such that 2π Θ(L, h) is a positive
(1, 1)-form.
Remark 7.8. In algebraic geometry, the analogous notion is for an algebraic line bundle to
be ample. In fact, at the end of the day (thanks to the Kodaira embedding theorem, Serre’s
GAGA theorem, and the Nakai-Moisheson criterion for amplitude) on a compact Kähler
manifold these turn out to be essentially equivalent notions.

The prototypical example is the hyperplane line bundle OPn (1) on projective space:
Proposition 7.9. The hyperplane line bundle on Pn carries a positive hermitian metric.

Proof. We will show that OPn (1) carries a Hermitian metric h whose curvature form is a
multiple of the Fubini–Study form ωFS . In fact, the standard Hermitian product hv, wi = t v·w
on V = Cn+1 gives rise to a Hermitian metric on the trivial bundle VPn on Pn = P(V ).
Then OPn (−1) inherits a metric as a sub-bundle of the trivial bundle Pn × V , which in turn
determines a metric h on OPn (1). Very explicitly, write [x] ∈ Pn for the point corresponding
to a vector x ∈ V − {0} and consider a section

s ∈ V ∗ = H 0 Pn , OPn (1) .

Then h is determined by the rule

2 |s(x)|2
s [x] h
= ,
hx, xi
where the numerator on the right is the squared modulus of the result of evaluating the
linear functional s on the vector x.

If we work with the usual affine coordinates z = (z1 , . . . , zn ) on Pn corresponding to the

point x = (1, z1 , . . . , zn ) ∈ V and take s ∈ V ∗ to be the functional given by projection onto
the zeroth coordinate, then

2 1
s [x] h = P .
1 + |zα |2
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 101

An explicit calculation [2, p. 30] shows that

√ √  
−1 −1 ¯ 1
Θ(OPn (1), h) =locally − · ∂ ∂ log P
2π 2π 1 + |zα |2
√ P
P
!
z̄α dzα ∧
P
−1 dzα ∧ dz̄α zα dz̄α
= · P −
2
2π 1 + |zα |2
P
1 + |zα |2
1
= · ωFS .
π
In particular, OPn (1) is positive in the sense of Kodaira. 
Remark 7.10. If L is a positive line bundle on a complex manifold X, and if Y ⊆ X is a
submanifold, then the restriction L|Y of L to Y is positive. (Exercise!) It follows that any
complex submanifold of Pn carries a postive line bundle.
Remark 7.11. Be aware that it typically happens that neither L nor L∗ is positive. For
example, the line bundle O(a, b) on P1 × P1 is positive if and only if a, b > 0.

An important fact is that on a Kähler manifold, the positivity of a line bundle L depends
only on the cohomology class of c1 (L).
Proposition 7.12. Assume that X is a compact Kähler manifold. Consider a hermitian
line bundle such that
√
−1
[ 2π Θ(L, h0 )] = [ω] ∈ H 2 (X, C).
where ω is a positive (1, 1) form. Then L carries a hermitian metric h such that
√
−1
2π
Θ(L, h) = ω.
In particular, L is positive.

Proof. Given h0 , we look for a real-valued function φ : X −→ R such that

h =def eφ h0
has curvature form ω. Now
Θ(L, eφ h0 ) = −∂∂φ + Θ(L, h0 ),
so we need to find φ satisfying
√ √
−1 −1
2π
Θ(L, h0 ) −ω = 2π
∂∂φ. (*)

But by assumption the left hand side of (*) is d-exact, hence by the ∂∂ lemma (Proposition
6.23) it is ∂∂-exact, as required. 

The very basic theorem for which we are aiming is this:

102 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Theorem 7.13. (Kodaira-Nakano Vanishing Theorem) Let L be a positive holomorphic

line bundle on a compact Kähler manifold X of dimension n. Then
H n,q (X, L) = 0 for q > 0,
i.e. H q (X, ΩnX ⊗ L) = 0 when q > 0. More generally, one has
H p,q (X, L) = 0
for p + q > n.

The first statement is the Kodaira vanishing theorem, while the second is usually called
Nakano vanishing.

Harmonic theory for hermitian line bundles. The first ingredient going into Theorem
7.13 is an extension of the Hodge decomposition theorem for hermitian line bundles. The
analysis underlying this result (which in any event we ignore) is nothing beyond that which
goes into the classical statement, once things have been set up properly. This set-up is what
we now discusss. Most of this material generalizes with no change to the case of hermitian
vector bundles of higher rank, but for simplicity we stick with line bundles.

Consider then a hermitian holomorphic line bundle (L, h) on a compact complex man-
ifold (X, g) of dimension n with a compatible Riemannian metric. The metric h allows us
first of all to define a Hodge ∗-operator on L-valued forms. Namely, one can view h as
giving a C-anti-linear isomorphism h : L −→ L∗ , and then one can define a C-anti-linear
isomorphism
∗L : Λp,q T ∗ X ⊗ L −→ Λn−p,n−q T ∗ X ⊗ L∗
by the rule


∗L α ⊗ s = ∗α ⊗ h(s) = ∗α ⊗ h(s) .
The metrics h and g also give rise to a global hermitian product on the spaces Ap,q (L) of L
valued forms via Z


α ⊗ s, β ⊗ t L = h(s, t) · (α ∧ ∗β), (7.3)
X
extending the definition in (6.3). In particular, if φ ∈ Ap,q (X, L), then
Z
φ ∧ ∗L φ = (φ, φ)L .
X

We now bring ∂ into the picture. Recall from (5.1) that because L is holomorphic there
is a canonically defined operator
∂ L : Ap,q (L) −→ Ap,q+1 (L).
We then define the adjoint
∗
∂ L : Ap,q (L) −→ Ap,q−1 (L)
by the formula
∗
∂ L = −∗L∗ ◦ ∂ L∗ ◦ ∗L .
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 103

(See Exercise 7.4). As in Proposition 6.7, this is adjoint to ∂ L with respect to the inner
product (7.3):
Lemma 7.14. For any α ∈ Ap,q (L) and β ∈ Ap,q+1 (L) one has
∗


α , ∂ Lβ L = ∂ Lα , β L.

(See [3, Lemma 4.1.12].)

Next one defines the Laplace operator for (L, h):

Definition 7.15. The Laplacian associated to (L, h) is the self-adjoint elliptic operator
∗ ∗
∆L : Ap,q (L) −→ Ap,q (L) , ∆L = ∂ L ∂ L + ∂ L ∂ L .
An L-valued form α ∈ Ap,q (L) is harmonic if
∆L (α) = 0.
One denotes by
Hp,q (L) ⊆ Ap,q (L)
the space of all such. As before an L-valued α is harmonic if and only if
∗
∂ L (α) = ∂ L (α) = 0.

Then one has:

Theorem 7.16. Let (L, h) be a Hermitian line bundle on a compact complex manifold X.
Then Hp,q (L) is finite dimensional, and one has an orthogonal direct sum decomposition
∗
Ap,q (X, L) = ∂ L Ap,q−1 (X, L) ⊕ Hp,q (L) ⊕ ∂ L Ap,q+1 (L).

And just as in §6, one concludes that every class in H∂p,q (X, L) = H q (X, ΩPX ⊗ L) has a
unique harmonic representative:
Corollary 7.17. The natural map
Hp,q (X, L) −→ H∂p,q (X, L)
is an isomorphism.

Before going on, we pause to note an important consequence of this picture, namely the
famous Serre duality theorem. Let L be a holomorphic line bundle on a compact complex
manifold X of dimension n. Specifically, there is a natural pairing
Z
p,q n−p,n−q ∗ n,n
H (X, L) × H (X, L ) −→ H (X) = C , (α, β) 7→ α ∧ β. (7.4)
X
∗
Here we view α ∧ β as an (n, n)-form with values in L ⊗ L = 1X .
Theorem 7.18. (Serre duality). The map (7.4) is a perfect pairing. In particular,
H p,q (X, L) ∼
= H n−p,n−q (X, L∗ )∗ .
104 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Proof. Fixing metrics on X and L, we can view (7.4) as the map

Z
p,q n−p,n−q ∗
H (X, L) × H (X, L ) −→ C (α, β) 7→ α∧β
X
on harmonic forms. Given 0 6= α ∈ Hp,q (X, L), we need to show that α pairs non-trivially
with some β ∈ Hn−p,n−q (X, L∗ ). But β =def ∗L α is harmonic (Exercise 7.5), and
Z Z
α∧β = α ∧ ∗L α = (α, α)L > 0,
X X
as required. 

Generalized Kähler identities and proof of Kodaira-Nakano vanishing. Let X be

a compact Kähler manifold of dimension n, and let (L, h) be a holomorphic line bundle on
X with a hermitian metric. The Kodaira vanishing theorem will follow from some analogues
of the Kähler identities, involving ∂ L in place of ∂. For this we need an analogue for L of ∂,
and this involves the notion of a connection. This presentation closely follows [4].

Recall then that a connection on L is a C-linear map

∇ = ∇L : A0 (L) −→ A1 (L)
that satisfies the Leibnitz rule:
∇(f · s) = df ⊗ s + f · ∇(s).
One says that ∇ is hermitian if it satisfies the identity
dh(s1 , s2 ) = h(∇s1 , s2 ) + h(s1 , ∇s2 ); (7.5)
here the the terms on the right are the 1-forms determined by the rules
h(∇s1 , s2 )(ξ) = h(∇ξ s1 , s2 ) , h(s1 , ∇s2 )(ξ) = h(s1 , ∇ξ s2 )
where ∇ξ : A0 (L) −→ A0 (L) is the “derivation” associated to a tangent vector ξ via ∇.
Using the decomposition A1 = A1,0 ⊕ A0,1 such a connection can be written as a sum
∇ = ∇1,0 ⊕ ∇0,1 , where
∇1,0 : A0 (L) −→ A1,0 (L) , ∇0,1 : A0 (L) −→ A0,1 (L).
One says that ∇L is compatible with the complex structure if ∇L0,1 = ∂ L .
Proposition 7.19. Given a holomorphic line bundle (L, h) with a hermitian metric on a
complex manifold X, there is a unique hermitian connection ∇L on L that is compatible with
the complex structure.

Sketch of Proof. We will give the local construction of ∇L . Note first that the difference
of any two connections on L is a 1-form on X. That said, choose a local trivialization
LU ∼= U × C of L over some open set U ⊆ X. Then on U sections of L are identified with
smooth functions, and ∇0,1 1,0
L = ∂ L acts simply as ∂. On the other hand, ∇L differs from ∂
by wedge product with a (1, 0)-form. In other words, one can write locally on U :


∇L = ∂ + θ + ∂
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 105

where θ ∈ A1,0 (U ). It remains to see what is the condition on θ imposed by the requirement
that ∇ be hermitian. For this, let e ∈ Γ(U, L) be the section determined by the constant
vector 1 ∈ C under the local trivialization of L. Then
h(e, e) = h
is the smooth function on U determining the metric as in equation (7.1), and ∂e = ∂e = 0.
Hence (7.5) yields
dh = h · θ + h · θ.
Since dh = ∂h + ∂h and since θ has type (1, 0), we find:
1
θ = · ∂h = ∂ log(h). (7.6)
h
This proves the existence and uniqueness of ∇L . 

One views ∇1,0

L as the L-valued analogue of ∂ L , and to emphasize this we shall subse-
quently write ∂L = ∇1,0
L . One extends these to operators

∂L : Ap,q (L) −→ Ap+1,q (L) , ∂ L : Ap,q (L) −→ Ap,q+1 (L)

via the Leibnitz rule. As before
2
∂L2 = ∂ L = 0, (7.7)
but now these operators no longer anti-commute. In fact:
Lemma 7.20. One has
∂L ∂ L + ∂ L ∂L = ΘL , (7.8)
in the sense that the composition of operators on the left acts by wedge product with the
curvature form ΘL .
Remark 7.21. In view of (7.7), one has
∂L ∂ L + ∂ L ∂L = ∇2L ,
and this is by definition the curvature of a connection. Here we are taking a low-tech
approach that avoids a general discussion of curvature.

Proof of Lemma 7.20. For simplicity, we check the stated identity when the two sides act
on smooth sections of L. Working locally, we may identify such a section with a smooth
function s. Then, using the local description of ∂L derived in the proof of Proposition 7.19,
we have
∂L ∂ L (s) = ∂(∂s) + θ ∧ ∂(s)
where θ = ∂(log h) is the 1-form appearing in (7.6). On the other hand,


∂ L ∂L (s) = ∂ L ∂s + θ ∧ s
= ∂∂(s) + ∂θ ∧ s − θ ∧ ∂s.
Therefore

∂L ∂ L + ∂ L ∂L (s) = ∂∂ log(h)) · s,
as required. 
106 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

There are three more operators that come into the picture. First, wedging with the (1, 1)
form ω determined by the metric g gives in the natural way an extenstion of the Lefschetz
operator
L : Ap,q (X, L) −→ Ap+1,q+1 (X, L) , α ⊗ s 7→ ω ∧ α ⊗ s.26
As before, we have the adjoint operator Λ:

Λ : Ap,q (X, L) −→ Ap−1,q−1 (X, L) , α ⊗ s 7→ Λα ⊗ s.

Note that these are point-wise operators, which coincide with the previously studied versions
when we take a local trivialization of L. One can check moreover that Λ is the global adjoint
of L with respect to the hermitian inner product ( , )L introduced above. Finally we denote
by ∂L∗ : Ap,q (L) −→ Ap−1,q (L) the adjoint of ∂L with respect to ( , )L . Then the Kähler
identity (6.7) generalizes to

Proposition 7.22. Let X be a compact Kähler manifold. Then

√
[Λ, ∂ L ] = − −1 ∂L∗ . (7.9)

This is sometimes called the Nakano identity. It fairly easily reduces to (6.7): see for instance
[3, Lemma 5.2.3].

We can now give the proof of Kodaira-Nakano vanishing.

Proof of Theorem 7.13. Since L is positive, it carries a hermitian metric h such that
√
−1
ω =def Θ(L, h)
2π
is a closed positive (1, 1)-form. We may then choose a Kähler metric on X with ω as the
corresponding Kähler form. With these choices, (7.8) becomes
√
∂L ∂ L + ∂ L ∂L = −2π −1 L,

where L is the Lefschetz operator. Taking adjoints, we find that

∗ ∗ √
∂L∗ ∂ L + ∂ L ∂L∗ = 2π −1 Λ. (7.10)

Now consider a harmonic form

α ∈ Hp,q (X, L).

26We apologize for the two conflicting meanings of the symbol L, but there doesn’t seem to be any easy
way to avoid this, and hopefully no confusion will result.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 107

Thanks to Corollary 7.17, the issue is to show that α = 0 if p + q > n. For this, we compute:
√ ∗ ∗
−1
Λα , (∂L∗ ∂ L + ∂ L ∂L∗ )α L



Λα , Λα L = 2π (by (7.10))
√ ∗ ∗
−1
Λα , ∂ L ∂L∗ α L


= 2π (since ∂ L α = 0)
√
−1
= 2π
∂ L Λα , ∂L∗ α)L
√
−1
− [Λ, ∂ L ]α, ∂L∗ α L


= 2π
(since ∂ L α = 0)
√
−1
√ ∗ ∗


= 2π
−1∂ L α, ∂ L α L
(by Proposition 7.22)
1
∂L∗ α, ∂L∗ α L .


= − 2π
But by comparing the signs of the first and last expressions we find that (Λα, Λα)L = 0, and
hence Λα = 0. But this means that α is locally represented by a primitive form. On the
other hand, by Theorem 3.16 (iii), there are no non-vanishing primitive forms of degree > n.
Hence α = 0, and we are done. 
Corollary 7.23. Let L be a positive line bundle on a compact Kähler manifold X. Then
given any line bundle P on X, there exists a constant m0 = m0 (L, P ) such that if m ≥ m0
then
H q (X, L⊗m ⊗ P ) = 0
for q > 0. 

This is a special case of the Serre vanishing theorem. The reader is asked to prove the
statement in Exercise 7.6.

The Kodaira embedding theorem. We now come to our final theorem, which ties to-
gether much of what we’ve studied up to now.
Theorem 7.24. (Kodaira Embedding Theorem). Let X be a compact Kähler manifold,
and let L be a positive holomorphic line bundle on X. Then for m  0 there exists an
embedding
φ = φm : X −→ Prm
(for some large r = rm ) such that
L⊗m = φ∗ OPr (1).

In other words, a compact Kähler manifold admits a projective embedding if and only if it
carries a positive line bundle.

The plan is to show first of all that L⊗m is globally generated for m  0, so that L⊗m
defines a holomorphic mapping φ : X −→ Pr . Then we will show that – possibly after
increasing m – the resulting mapping is actually an embedding. We will concentrate on
proving the first point, the second being similar.
108 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

Fix a point x ∈ X. We want to show that if m  0 then L⊗m has a holomorphic section
that doesn’t vanish at x. To this end, denote by mx ⊆ OX the ideal sheaf of holomorphic
functions vanishing at x, and consider the exact sequence
0 −→ mx −→ OX −→ O{x} −→ 0,
so that O{x} is a one-dimensional sky-scraper sheaf supported at x. Tensoring by L⊗m we
arrive at
0 −→ L⊗m ⊗ mx −→ L⊗m −→ L⊗m ⊗ O{x} −→ 0,
and the long exact cohomology sequence gives
ev
H 0 (X, L⊗m ) −→
x
C −→ H 1 (X, L⊗m ⊗ mx ). (7.11)
Here C = H 0 (L⊗m ⊗ O{x} ) and the map evx is evaluation at x. Now suppose one knew that
H 1 (X, L⊗m ⊗ mx ) = 0 for m  0. (7.12)
Then evaluation the evaluation map evx would be surjective, and we could conclude that
L⊗m has a global section that is non-vanishing at x.27 The question being reduced to a
vanishing statement involving a positive line bundle, one can hope to apply Theorem 7.13.
Unfortunately it doesn’t apply directly because – when n = dim X ≥ 2 – the ideal sheaf
mx ⊆ OX is not a line bundle. However we can circumvent this problem by blowing up the
point x ∈ X, which reduces one to a question about locally free sheaves.

Blowing up a point. Recall that in Example 2.35 we discussed the construction of the blow-
up of a point in Cn . Since this construction is local about a neighborhood of 0 ∈ Cn , we
will be able carry it over to any complex manifold X to produce a new manifold X 0 in which
the ideal sheaf of a point is in effect replaced by a line bundle.

We start by reviewing and analyzing the construction in Cn . Viewing Pp−1 as the space
of lines through the origin in Cn , consider
Z = ([a], v) | v ∈ C · a ⊆ Pn−1 × Cn .


Write
b : Z −→ Cn , f : Z −→ Pn−1
for the two projections. Thus f realizes Z as the total space of the line bundle OPn−1 (−1)
over Pn−1 . In particular, the fibre of b over 0 ∈ Cn is a copy of Pn−1 which we is called the
exceptional divisor of Z:
E = Pn−1 = b−1 (0) ⊆ Z.
Note that E is naturally identified as the projective space PT0 Cn of lines in the (holomorphic)
tangent space of Cn at 0. Furthermore, note that if ∆ is any neighborhood of 0 ∈ Cn , and
if ∆∗ = ∆ − {0}, then b restricts to an isomorphism
∼
=
b−1 (∆∗ ) −→ (∆∗ ). (7.13)
We set
∆0 = Bl0 (∆) =def b−1 (∆) :
27A similar argument with mx replaced by the ideal mx,y of two points x, y ∈ X would show that the
morphism φ is one-to-one, and replacing mx by m2x yields that dφ is injective.
 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 109

Figure 7. Blowing up a point x ∈ X.

this is the blowing-up of ∆ at 0. Thus ∆0 − E = b−1 (∆∗ ).

It is useful for computations to observe that if Ui ⊆ Pn−1 are the standard open subsets
of Pn−1 , then there are natural isomorphisms f −1 (Ui ) ∼ = Cn under which b is locally identified
n n
with the maps b : C −→ C given by


(w0 , w1 , . . . , wn−1 ) 7→ wi w0 , wi w1 , . . . , wi , . . . , wi wn−1 (7.14)
(Exercise 7.8). In these local coordinates on Z, the exceptional divisor E is (locally) defined
in Z by the equation wi = 0, and f is given by (w0 , . . . , wi , . . . wn−1 ) 7→ [w0 , . . . , 1, . . . , wn−1 ].

Denote by OZ (−E) ⊆ OZ the ideal sheaf of E, i.e. the sheaf of holomorphic functions
vanishing on E. Then, using the local description just given, one can check (Exercise 7.8):
Lemma 7.25. The ideal sheaf OZ (−E) is a line bundle on Z, and
OZ (−E) ∼ = f ∗ OPn−1 (1). (7.15)

(See Exercise 7.8.) This will allow us shortly to put a metric on OZ (−E). The reader should
also check by taking the Jacobian determinant of (7.14) that
b∗ ΩnCn = ΩnZ ⊗ OZ (−(n − 1)E);
very concretely, the pull-back via b of the form dz1 ∧ . . . ∧ dzn vanishes to order (n − 1) along
E.

We now globalize this discussion. Given a complex manifold X, fix a point x ∈ X.

Choose a neighborhood U of x ∈ X isomorphic to a neighborhood ∆ 3 0 of the origin in Cn ,
and write U ∗ = U − {x}. Using the local discussion above, and the identification of U with
∆, we can construct the blowing-up U 0 = Blx (U ) of U at x, and then we construct a new
manifold X 0 = Blx (X) by replacing U by U 0 via the isomorphism (7.13). The blowing-up
X 0 = Blx (X) comes with a mapping
µ : Blx (X) −→ X
110 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

that is an isomorphism over X − {x}. Moreover X 0 contains an exceptional divisor

E = µ−1 (x)
which is canonically idenfitied with the projective space PTx X of one-dimensional subspaces
in the holomorphic tangent space Tx X. This process is illustrated schematically in Figure
7. As in the local picture one has
µ∗ ΩnX = ΩnX 0 ⊗ OX 0 (−(n − 1)E). (7.16)
The fastidious reader can check that up to isomorphism Blx (X) does not depend on the
particular neighborhood U 3 x we took in the construction.
Remark 7.26. In an essentially identical manner, one can form the blowing up of X at a
finite number of points. With more effort, one can also construct the blow-up of X along
any closed submanifold Y ⊆ X: in this case the exceptional divisor is isomorphic to the
projectivization PNY /X of the complex normal bundle to Y in X.

The next point is to study the positivity of the relevant line bundles.
Lemma 7.27. Let L be a positive line bundle on X, and fix an integer a > 0 and an arbitrary
line bundle P on X. Then for m  0 the line bundle
µ∗ (L⊗m ⊗ P ) ⊗ OX 0 (−aE))
is positive on X 0 .

Proof. Fix a positive metric hL on L and an arbitrary metric hP on P . Then h⊗m L ⊗ hP pulls
back to a metric hm on µ∗ (L⊗m ⊗ P ). By taking m  0 we can arrange that it is positve
away from E, but it won’t be positive along E itself since Θ(hm ) vanishes on vectors tangent
to E. On the other hand, via the local description (7.15) of OX 0 (−E), we can find (using a
partition of unity) a metric ha on OX 0 (−aE) that in a neighborhood of E agrees with a power
of the pullback f ∗ h⊗a
FS of the Fubini-study metric on OPn−1 (1) (but we don’t know anything
about ha away from E). We claim that if m  0 then hm ⊗ ha is everywhere positive. This
will follow from Exercise 7.3 (a) away from E, so it remains to check the positivity along
E. For this, note that by construction Θ(ha ) is positive on tangent directions to E. On the
other hand, Θ(hm ) is positive on directions normal to E since (as we may assume) it’s the
pull-back of a positive (1, 1)-form on X. Therefore hm ⊗ ha is positive everywhere on X 0 for
m  0. 
Corollary 7.28. Let L be a positive line bundle on X. If m  0, then
H 1 X 0 , µ∗ (L⊗m ) ⊗ OX 0 (−E) = 0.

Proof. We apply the previous Lemma with P = (ΩnX )∗ and a = n = dim X. Then by (7.16),
one has:  
µ∗ L⊗m ⊗ OX 0 (−E) = ΩnX 0 ⊗ µ∗ L⊗m ⊗ (ΩnX )∗ ⊗ OX 0 (−nE) ,

so the assertion follows from Kodaira vanishing. 

 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014 111

We now indicate the proof of the Kodaira embedding theorem. Given a point x ∈ X,
we focus on showing that L⊗m has a section that doesn’t vanish at x for m  0: this implies
that the mapping φm appearing in Theorem 7.24 is well-defined, and as indicated above the
argument that it is one-to-one with injective derivative is similar. For simplicity we assume
n ≥ 2 (the case of Riemann surfaces being elementary given what we already know).

To this end, consider the blowing-up

µ : X 0 = Blx (X) −→ X.
We claim first that pull-back of sections yields an isomorphism
∼
=
H 0 (X, L⊗m ) −→ H 0 X 0 , µ∗ (L⊗m ) .

In fact, the map is obviously injective, so consider a section

s0 ∈ H 0 X 0 , µ∗ (L⊗m ) .

Then the restriction of s0 to X 0 − E can be viewed as a section s0 ∈ Γ(X − {x}, L⊗m ) of L⊗m
on the complement of x. But by Hartog’s theorem, s0 extends to a section s ∈ Γ(X, L⊗m )
which pulls back to s0 .

Now consider the commutative diagram

H 0 X 0 , µ∗ (L⊗m ) / H 0 X 0 , µ∗ L⊗m ⊗ OE )

 
evx
H 0 X, L⊗m ) / H 0 X, L⊗m ⊗ O{x}

The vertical map on the left is an isomorphism, and since µ∗ L is trivial along E, so is the
vertical map on the right. Thus we can identify the top map with evaluation of sections at
x ∈ X. On the other hand, it follows from Corollary 7.28 – and this is the serious point –
that the top vertical map is surjective if m  0. Therefore evx is surjective when m  0,
and so we have produced a section
sx ∈ Γ X, L⊗m with sx (x) 6= 0.

The starting value of m here might depend on x. However sx is non-vanishing in a neighbor-

hood of x, and then by compactness we can find one value of m that works at every point
of x. QED.

Exercises for Section 7.

Exercise 7.1. Prove that any smooth complex line bundle on a manifold admits a smooth
hermitian metric. [Hint: If h1 , h2 are hermitian metrics on X, and ρ : X −→ R 0 is a
smooth function, then h1 + h2 and ρ h1 are also Hermitian metrics. This allows one to glue
local metrics together via a partition of unity.)]
Exercise 7.2. If L1 , L2 are positive line bundles on a compact complex manifold X, then
so is L1 ⊗ L2 .
Exercise 7.3. Let X be a compact complex manifold.
112 NOTES ON COMPLEX GEOMETRY FOR MATH 545, FALL 2014

(a). Let ω be a positive (1, 1)-form on X, and let η be an arbitrary (1, 1)-form. Show that
if m  0, then mω + η is positive.
(b). Let L be a positive holomorphic line bundle on X, and let P be an arbitrary holo-
morphic line bundle on X. Show that if m  0, then L⊗m ⊗ P is again a positive
line bundle.
Exercise 7.4. Show that if L = X × C is the trivial line bundle with the constant metric,
∗ ∗
then ∂ L = ∂ .
Exercise 7.5. Prove that if α ∈ Hp,q (X, L), then
∗L α ∈ Hn−p,n−q (X, L∗ ).
Exercise 7.6. Prove Corollary 7.23. (Hint: Use Exercise 7.3 to write L⊗m0 ⊗ P = ΩnX ⊗ N
where N is a positive line bundle.)
Exercise 7.7. Prove that a compact Kähler manifold X admits a projective embedding if
and only if it carries a Kähler form ω with rational periods, ie with the property that
Z
ω ∈ Q
γ

for all γ ∈ H2 (X, Z). (Hint: The condition is equivalent to the condition that ω ∈ H 2 (X, Q),
and then after replacing ω by multiple one can assume that ω ∈ H 2 (X, Z). Now apply the
Lefschetz (1, 1)-theorem.) Such a manifold is sometimes called a Hodge manifold.
Exercise 7.8. Prove the local description of blowing up given in equation (7.14). Use this
to verify Lemma 7.25.

References

[1] Vladimir Arnol´d, Mathematical Methods of Classical Mechanics, Springer GTM.

[2] Phillip Griffiths and Joe Harris, Principles of Algebraic Geometry, .
[3] Daniel Huybrechts, Complex Geometry: An Introduction, Springer Universitext, 2005.
[4] Christian Schnell, Notes on complex manifolds, available from Christian’s Stony Brook web page:
http://www.math.sunysb.edu/~cschnell/pdf/notes/complex-manifolds.pdf