This action might not be possible to undo. Are you sure you want to continue?
Andrei Moroianu
Current version March 16, 2004.
Contents
Introduction 4
Part 1. Complex geometry 5
1. Complex structures and holomorphic maps 6
2. Holomorphic forms and vector ﬁelds 12
3. Complex and holomorphic vector bundles 17
Part 2. Hermitian and K¨ahler structures 23
4. Hermitian bundles 24
5. Hermitian and K¨ahler metrics 27
6. The curvature tensor of K¨ahler manifolds 32
7. Examples of K¨ahler metrics 37
Part 3. The Laplace operator 43
8. Natural operators on Riemannian and K¨ahler manifolds 44
9. Hodge and Dolbeault theory 49
Part 4. Prescribing the Ricci tensor on K¨ahler manifolds 55
10. Chern classes 56
11. The Ricci form of K¨ahler manifolds 61
12. The Calabi conjecture 65
13. K¨ahler–Einstein metrics 69
Part 5. Vanishing results 73
14. Weitzenb¨ock techniques 74
15. The Hirzebruch–Riemann–Roch formula 79
16. Further vanishing results 83
Part 6. Calabi–Yau manifolds 87
17. Ricci–ﬂat K¨ahler metrics 88
18. Constructions of Calabi–Yau manifolds 91
Bibliography 97
3
4 CONTENTS
Introduction
These notes, based on a graduate course I gave at Hamburg University in 2003, are intended to
students having basic knowledges of diﬀerential geometry. I assume, in particular, that the reader
is familiar with following topics:
• diﬀerential manifolds, tensors, Lie groups;
• principal ﬁbre bundles, vector bundles, connexions, holonomy groups;
• Riemannian metrics, de Rham decomposition theorem, Levi–Civita connexion, Killing
vector ﬁelds.
This background material is well covered in the classical literature, and can be found for instance
in [8], Ch. 14.
The main purpose of these notes is to provide a quick and accessible introduction to diﬀerent
aspects of K¨ahler geometry. They should hopefully be useful for graduate students in mathematics
and theoretical physics. The text is self–contained with a few notable exceptions – the Newlander–
Nirenberg theorem, the Hodge theorem, the Calabi conjecture, the Hirzebruch–Riemann–Roch
formula, the Cheeger–Gromoll theorem and the Kodaira embedding theorem. I considered that
including the proofs of these results would have add too much technicality in the presentation
and would have doubled the volume of the text without bringing essentially new insights to our
objects of interest.
The text is organized as follows. In the ﬁrst part I quickly introduce complex manifolds, and in
Part 2 I deﬁne K¨ahler manifolds from the point of view of Riemannian geometry. Most of the
remaining material concerns compact manifolds. In Part 3 I review Hodge and Dolbeault theories,
and give a simple way of deriving the famous K¨ahler identities. Part 4 is devoted to the Calabi
conjecture and in Part 5 I obtain several vanishing results using Weitzenb¨ock techniques. Finally,
in Part 6, diﬀerent aspects of Calabi–Yau manifolds are studied using techniques from algebraic
geometry.
Most of the sections end up with a series of exercises whose levels of diﬃculty range from low to
medium.
Part 1
Complex geometry
1. Complex structures and holomorphic maps
1.1. Preliminaries. K¨ahler manifolds may be considered as special Riemannian manifolds.
Besides the Riemannian structure, they also have compatible symplectic and complex structures.
Here are a few examples of K¨ahler manifolds:
• (C
m
, ' , `), where ' , ` denotes the Hermitian metric ds
2
= Re(
¸
dz
i
d¯ z
i
);
• any oriented 2–dimensional Riemannian manifold;
• the complex projective space (CP
m
, FS) endowed with the FubiniStudy metric;
• every projective manifold, that is, submanifold of CP
m
deﬁned by homogeneous polyno
mials in C
m+1
.
We give here a short deﬁnition, which will be detailed later.
Definition 1.1. A K¨ahler structure on a Riemannian manifold (M
n
, g) is given by a 2–form Ω
and a ﬁeld of endomorphisms of the tangent bundle J satisfying the following
• algebraic conditions
a) J is an almost complex structure: J
2
= −Id.
b) g(X, Y ) = g(JX, JY ) ∀ X, Y ∈ TM.
c) Ω(X, Y ) = g(JX, Y ).
• analytic conditions
d) the 2–form Ω is symplectic: dΩ = 0.
e) J is integrable in the sense that its Nijenhuis tensor vanishes (see (4) below).
Condition a) requires the real dimension of M to be even. Obviously, given the metric and one of
the tensors J and Ω, we can immediately recover the other one by the formula c).
K¨ahler structures were introduced by Erich K¨ahler in his article [7] with the following motivation.
Given any Hermitian metric h on a complex manifold, we can express the fundamental two–form
Ω in local holomorphic coordinates as follows:
Ω = i
¸
h
α
¯
β
dz
α
∧ d¯ z
β
,
where
h
α
¯
β
:= h
∂
∂z
α
,
∂
∂¯ z
β
.
He then noticed that the condition dΩ = 0 is equivalent to the local existence of some function u
such that
h
α
¯
β
=
∂
2
u
∂z
α
∂¯ z
β
.
In other words, the whole metric tensor is deﬁned by a unique function! This remarkable (be
merkenswert) property of the metric allows one to obtain simple explicit expressions for the
Christoﬀel symbols and the Ricci and curvature tensors, and “a long list of miracles occur then”.
The function u is called K¨ahler potential.
1. COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 7
There is another remarkable property of K¨ahler metrics, which, curiously, K¨ahler himself did not
seem to have noticed. Recall that every point x in a Riemannian manifold has a local coordinate
system x
i
such that the metric osculates to the Euclidean metric to the order 2 at x. These
special coordinate systems are the normal coordinates around each point. Now, on a complex
manifold with Hermitian metric, the existence of normal holomorphic coordinates around each
point is equivalent to the metric being K¨ahler!
K¨ahler manifolds have found many applications in various domains like Diﬀerential Geometry,
Complex Analysis, Algebraic Geometry or Theoretical Physics. To illustrate their importance let
us make the following remark. With two exceptions (the so–called Joyce manifolds in dimensions
7 and 8), the only known compact examples of manifolds satisfying Einstein’s equations
R
αβ
= 0
(Ricci–ﬂat in modern language) are constructed on K¨ahler manifolds. Generic Ricci–ﬂat K¨ahler
manifolds, also called Calabi–Yau manifolds, will be studied later on in these notes.
1.2. Holomorphic functions. A function F = f +ig : U ⊂ C →C is called holomorphic if
it satisﬁes the Cauchy–Riemann equations:
∂f
∂x
=
∂g
∂y
and
∂f
∂y
+
∂g
∂x
= 0.
Let j denote the endomorphism of R
2
corresponding to the multiplication by i on C via the
identiﬁcation of R
2
with C given by z = x + iy → (x, y). The endomorphism j can be expressed
in the canonical base as
j =
0 −1
1 0
.
The diﬀerential of F (viewed as real function F : U ⊂ R
2
→R
2
) is of course the linear map
F
∗
=
¸
∂f
∂x
∂f
∂y
∂g
∂x
∂g
∂y
.
Then it is easy to check that the Cauchy–Riemann relations are equivalent to the commutation
relation jF
∗
= F
∗
j.
Similarly, we identify C
m
with R
2m
via
(z
1
, . . . , z
m
) = (x
1
+ iy
1
, . . . , x
m
+ iy
m
) →(x
1
, . . . , x
m
, y
1
, . . . , y
m
),
and denote by j
m
the endomorphism of R
2m
corresponding to the multiplication by i on C
m
:
j
m
=
0 −I
m
I
m
0
.
A function F : U ⊂ C
n
→ C
m
is then holomorphic if and only if the diﬀerential F
∗
of F as real
map F : R
2n
→R
2m
satisﬁes j
m
F
∗
= F
∗
j
n
.
8
1.3. Complex manifolds. A complex manifold of complex dimension m is a topological
space M with an open covering  such that for every point x ∈ M there exists U ∈  containing
x and a homeomorphism φ
U
: U →
˜
U ⊂ C
m
, such that for every intersecting U, V ∈ , the map
between open sets of C
m
φ
UV
:= φ
U
◦ φ
−1
V
is holomorphic. A pair (U, φ
U
) is called a chart and the collection of all charts is called a holo
morphic structure.
Important example. The complex projective space CP
m
can be deﬁned as the set of complex lines
of C
m+1
(a line is a vector subspace of dimension one). If we deﬁne the equivalence relation ∼ on
C
m+1
−¦0¦ by
(z
0
, . . . , z
m
) ∼ (αz
0
, . . . , αz
m
), ∀ α ∈ C
∗
,
then CP
m
= C
m+1
−¦0¦/ ∼ . The equivalence class of (z
0
, . . . z
m
) will be denoted by [z
0
: . . . : z
m
].
Consider the open cover U
i
, i = 0, . . . , m of CP
m
deﬁned by
U
i
:= ¦[z
0
: . . . : z
m
] [ z
i
= 0¦
and the maps φ
i
: U
i
→C
m
,
φ
i
([z
0
: . . . : z
m
]) =
z
0
z
i
, . . . ,
z
i−1
z
i
,
z
i+1
z
i
. . . ,
z
m
z
i
.
It is then an easy exercise to compute
φ
i
◦ φ
−1
j
(w
1
, . . . , w
m
) =
w
1
w
i
, . . . ,
w
i−1
w
i
,
w
i+1
w
i
, . . . ,
w
j
w
i
,
1
w
i
,
w
j+1
w
i
, . . . ,
w
m
w
i
,
which is obviously holomorphic on its domain of deﬁnition.
A function F : M → C is called holomorphic if F ◦ φ
−1
U
is holomorphic for every U ∈ . This
property is local. To check it in the neighborhood of a point x it is enough to check it for a single
U ∈  containing x.
The most important object on a complex manifold from the diﬀerential geometric point of view is
the almost complex structure J, which is a ﬁeld of endomorphisms of the tangent bundle deﬁned
as follows. For every X ∈ T
x
M, choose U ∈  containing x and deﬁne
J
U
(X) = (φ
U
)
−1
∗
◦ j
n
◦ (φ
U
)
∗
(X).
If we take some other V ∈  containing x, then φ
V U
= φ
V
◦φ
−1
U
is holomorphic, and φ
V
= φ
V U
◦φ
U
,
so
J
V
(X) = (φ
V
)
−1
∗
◦ j
n
◦ (φ
V
)
∗
(X) = (φ
V
)
−1
∗
◦ j
n
◦ (φ
V U
)
∗
◦ (φ
U
)
∗
(X)
= (φ
V
)
−1
∗
◦ (φ
V U
)
∗
◦ j
n
◦ (φ
U
)
∗
(X) = (φ
U
)
−1
∗
◦ j
n
◦ (φ
U
)
∗
(X)
= J
U
(X),
thus showing that J
U
does not depend on U and their collection is a well–deﬁned tensor J on M.
This tensor clearly satisﬁes J
2
= −Id.
1. COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 9
Definition 1.2. A (1,1)–tensor J on a diﬀerential manifold M satisfying J
2
= −Id is called an
almost complex structure. The pair (M, J) is then referred to as almost complex manifold.
A complex manifold is thus in a canonical way an almost complex manifold. The converse is only
true under some integrability condition (see Theorem 1.4 below).
1.4. The complexiﬁed tangent bundle. Let (M, J) be an almost complex manifold. We
would like to diagonalize the endomorphism J. In order to do so, we have to complexify the
tangent space. Deﬁne
TM
C
:= TM ⊗
R
C.
We extend all real endomorphisms and diﬀerential operators from TM to TM
C
by C–linearity.
Let T
1,0
M (resp. T
0,1
M) denote the eigenbundle of TM
C
corresponding to the eigenvalue i (resp.
−i) of J. The following algebraic lemma is an easy exercise.
Lemma 1.3. One has
T
1,0
M = ¦X −iJX [ X ∈ TM¦ , T
0,1
M = ¦X + iJX [ X ∈ TM¦.
and TM
C
= T
1,0
M ⊕T
0,1
M.
The famous Newlander–Nirenberg theorem can be stated as follows:
Theorem 1.4. Let (M, J) be an almost complex manifold. The almost complex structure J comes
from a holomorphic structure if and only if the distribution T
0,1
M is integrable.
Proof. We will only prove here the “only if” part. The interested reader can ﬁnd the proof
of the hard part for example in [5].
Suppose that J comes from a holomorphic structure on M. Consider a local chart (U, φ
U
) and let
z
α
= x
α
+ iy
α
be the α–th component of φ
U
. If ¦e
1
, . . . , e
2m
¦ denotes the standard basis of R
2m
,
we have by deﬁnition:
∂
∂x
α
= (φ
U
)
−1
∗
(e
α
) and
∂
∂y
α
= (φ
U
)
−1
∗
(e
m+α
).
Moreover, j
m
(e
α
) = e
m+α
, so we obtain directly from the deﬁnition
(1) J
∂
∂x
α
=
∂
∂y
α
.
We now make the following notations
∂
∂z
α
:=
1
2
∂
∂x
α
−i
∂
∂y
α
,
∂
∂¯ z
α
:=
1
2
∂
∂x
α
+ i
∂
∂y
α
.
From (1) we obtain immediately that
∂
∂zα
and
∂
∂¯ zα
are local sections of T
1,0
M and T
0,1
M respec
tively. They form moreover a local basis in each point of U. Let now Z and W be two local
10
sections of T
0,1
M and write Z =
¸
Z
α
∂
∂¯ zα
, W =
¸
W
α
∂
∂¯ zα
. A direct calculation then gives
[Z, W] =
n
¸
α,β=1
Z
α
∂W
β
∂¯ z
α
∂
∂¯ z
β
−
n
¸
α,β=1
W
α
∂Z
β
∂¯ z
α
∂
∂¯ z
β
,
which is clearly a local section of T
0,1
M.
An almost complex structure arising from a holomorphic structure is called a complex structure.
Remark. The existence of local coordinates satisfying (1) is actually the key point of the hard
part of the theorem. Once we have such coordinates, it is easy to show that the transition functions
are holomorphic: suppose that u
α
, v
α
is another local system of coordinates, satisfying
∂
∂v
α
= J
∂
∂u
α
.
We then have
(2)
∂
∂x
α
=
m
¸
β=1
∂u
β
∂x
α
∂
∂u
β
+
m
¸
β=1
∂v
β
∂x
α
∂
∂v
β
and
(3)
∂
∂y
α
=
m
¸
β=1
∂u
β
∂y
α
∂
∂u
β
+
m
¸
β=1
∂v
β
∂y
α
∂
∂v
β
.
Applying J to (2) and comparing to (3) yields
∂u
β
∂x
α
=
∂v
β
∂y
α
and
∂u
β
∂y
α
= −
∂v
β
∂x
α
,
thus showing that the transition functions are holomorphic.
1.5. Exercises.
(1) Prove Lemma 1.3.
(2) Let A+ iB ∈ Gl
m
(C). Compute the product
I
m
0
−iI
m
I
m
A B
−B A
I
m
0
iI
m
I
m
and use this computation to prove that for every invertible matrix A+iB ∈ Gl
m
(C), the
determinant of the real 2m2m matrix
A B
−B A
is strictly positive.
(3) Show that every almost complex manifold is orientable.
1. COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 11
(4) Let α > 1 be some real number. Let Γ be the subgroup of Gl
m
(C) generated by αI
m
.
Show that Γ acts freely and properly discontinuously on C
m
−¦0¦. Use this to prove that
S
1
S
2m−1
carries a complex structure.
12
2. Holomorphic forms and vector ﬁelds
2.1. Decomposition of the (complexiﬁed) exterior bundle. Let (M, J) be an almost
complex manifold. We now turn our attention to exterior forms and introduce the complexiﬁed
exterior bundle Λ
∗
C
M = Λ
∗
M⊗
R
C. The sections of Λ
∗
C
M can be viewed as complex–valued forms
or as formal sums ω + iτ, where ω and τ are usual real forms on M.
We deﬁne the following two sub–bundles of Λ
1
C
M:
Λ
1,0
M := ¦ξ ∈ Λ
1
C
M [ ξ(Z) = 0 ∀ Z ∈ T
0,1
M¦
and
Λ
0,1
M := ¦ξ ∈ Λ
1
C
M [ ξ(Z) = 0 ∀ Z ∈ T
1,0
M¦.
The sections of these sub–bundles are called forms of type (1,0) or forms of type (0,1) respectively.
Similarly to Lemma 1.3 we have
Lemma 2.1. One has
Λ
1,0
M = ¦ω −iω ◦ J [ ω ∈ Λ
1
M¦ , Λ
0,1
M = ¦ω + iω ◦ J [ ω ∈ Λ
1
M¦.
and Λ
1
C
M = Λ
1,0
M ⊕Λ
0,1
M.
Let us denote the k–th exterior power of Λ
1,0
(resp. Λ
0,1
) by Λ
k,0
(resp. Λ
0,k
) and let Λ
p,q
denote
the tensor product Λ
p,0
⊗Λ
0,q
. The exterior power of a direct sum of vector spaces can be described
as follows
Λ
k
(E ⊕F) · ⊕
k
i=0
Λ
i
E ⊗Λ
k−i
F.
Using Lemma 2.1 we then get
Λ
k
C
M · ⊕
p+q=k
Λ
p,q
M.
Sections of Λ
p,q
M are called forms of type (p, q). It is easy to check that a complex–valued k–form
ω is a section of Λ
k,0
M if and only if Z ω = 0 for all Z ∈ T
0,1
M. More generally, a k–form is
a section of Λ
p,q
M if and only if it vanishes whenever applied to p + 1 vectors from T
1,0
M or to
q + 1 vectors from T
0,1
M.
If J is a complex structure, we can describe these spaces in terms of a local coordinate system.
Let z
α
= x
α
+ iy
α
be the α–th coordinate of some φ
U
. Extending the exterior derivative on
functions by C–linearity we get complex–valued forms dz
α
= dx
α
+ idy
α
and d¯ z
α
= dx
α
− idy
α
.
Then ¦dz
1
, . . . , dz
m
¦ and ¦d¯ z
1
, . . . , d¯ z
m
¦ are local basis for Λ
1,0
M and Λ
0,1
M respectively, and a
local basis for Λ
p,q
M is given by
¦dz
i
1
∧ . . . ∧ dz
ip
∧ d¯ z
j
1
∧ . . . ∧ d¯ z
jq
, i
1
< . . . < i
p
, j
1
< . . . < j
q
¦.
To every almost complex structure J one can associate a (2,1)–tensor N
J
called the Nijenhuis
tensor, satisfying
(4) N
J
(X, Y ) = [X, Y ] + J[JX, Y ] + J[X, JY ] −[JX, JY ], ∀ X, Y ∈ (
∞
(TM).
Proposition 2.2. Let J be an almost complex structure on M
2m
. The following statements are
equivalent:
2. HOLOMORPHIC FORMS AND VECTOR FIELDS 13
(a) J is a complex structure.
(b) T
0,1
M is integrable.
(c) d(
∞
(Λ
1,0
M) ⊂ (
∞
(Λ
2,0
M ⊕Λ
1,1
M).
(d) d(
∞
(Λ
p,q
M) ⊂ (
∞
(Λ
p+1,q
M ⊕Λ
p,q+1
M) ∀ 0 ≤ p, q ≤ m.
(e) N
J
= 0.
Proof. (a) ⇐⇒(b) is given by Theorem 1.4.
(b) ⇐⇒ (c) Let ω be a section of Λ
1,0
M. The Λ
0,2
M–component of dω vanishes if and only if
dω(Z, W) = 0 ∀ Z, W ∈ T
0,1
M. Extend Z and W to local sections of T
0,1
M and write
dω(Z, W) = Z(ω(W)) −W(ω(Z)) −ω([Z, W]) = −ω([Z, W]).
Thus
dω(Z, W) = 0 ∀ Z, W ∈ T
0,1
M, ∀ ω ∈ Λ
1,0
M
⇐⇒ ω([Z, W]) = 0 ∀ Z, W ∈ T
0,1
M, ∀ ω ∈ Λ
1,0
M
⇐⇒ [Z, W] ∈ T
0,1
M ∀ Z, W ∈ T
0,1
M.
(c) ⇐⇒ (d) One implication is obvious. Suppose now that (c) holds. By conjugation we get
immediately d(
∞
(Λ
0,1
M) ⊂ (
∞
(Λ
0,2
M ⊕Λ
1,1
M). It is then enough to apply Leibniz’ rule to any
section of Λ
p,q
M, locally written as a sum of decomposable elements ω
1
∧ . . . ∧ ω
p
∧ ¯ τ
1
∧ . . . ∧ ¯ τ
q
,
where ω
i
∈ (
∞
(Λ
1,0
M) and ¯ τ
i
∈ (
∞
(Λ
0,1
M).
(b) ⇐⇒ (e) Let X, Y ∈ (
∞
(TM) be local vector ﬁelds and let Z denote the bracket Z :=
[X + iJX, Y + iJY ]. An easy calculation gives Z − iJZ = N
J
(X, Y ) − iJN
J
(X, Y ). Thus
Z ∈ T
0,1
M ⇐⇒ N
J
(X, Y ) = 0, which proves that T
0,1
M is integrable if and only if N
J
≡ 0
2.2. Holomorphic objects on complex manifolds. In this section (M, J) will denote
a complex manifold of complex dimension m. We start with the following characterization of
holomorphic functions.
Lemma 2.3. Let f : M →C be a smooth complex–valued function on M. The following assertions
are equivalent:
(1) f is holomorphic.
(2) Z(f) = 0 ∀ Z ∈ T
0,1
M.
(3) df is a form of type (1, 0).
Proof. (2) ⇐⇒ (3). df ∈ Λ
1,0
M ⇐⇒ df(Z) = 0 ∀ Z ∈ T
0,1
M ⇐⇒ Z(f) = 0 ∀ Z ∈
T
0,1
M.
(1) ⇐⇒ (3). The function f is holomorphic if and only if f ◦ φ
−1
U
is holomorphic for every
holomorphic chart (U, φ
U
), which is equivalent to f
∗
◦ (φ
U
)
−1
∗
◦ j
m
= if
∗
◦ (φ
U
)
−1
∗
, that is, f
∗
◦
J = if
∗
. This last equation just means that for every real vector X, df(JX) = idf(X), hence
idf(X + iJX) = 0 ∀ X ∈ TM, which is equivalent to df ∈ Λ
1,0
M.
14
Using Proposition 2.2, for every ﬁxed (p, q) we deﬁne the diﬀerential operators ∂ : (
∞
(Λ
p,q
M) →
(
∞
(Λ
p+1,q
M) and
¯
∂ : (
∞
(Λ
p,q
M) →(
∞
(Λ
p,q+1
M) by d = ∂ +
¯
∂.
Lemma 2.4. The following identities hold:
∂
2
= 0,
¯
∂
2
= 0, ∂
¯
∂ +
¯
∂∂ = 0.
Proof. We have 0 = d
2
= (∂ +
¯
∂)
2
= ∂
2
+
¯
∂
2
+(∂
¯
∂ +
¯
∂∂), and the three operators in the last
term take values in diﬀerent sub–bundles.
Definition 2.5. A vector ﬁeld Z in (
∞
(T
1,0
M) is called holomorphic if Z(f) is holomorphic for
every locally deﬁned holomorphic function f. A p–form ω of type (p, 0) is called holomorphic if
¯
∂ω = 0.
Definition 2.6. A real vector ﬁeld X is called real holomorphic if X − iJX is a holomorphic
vector ﬁeld.
Lemma 2.7. Let X be a real vector ﬁeld on a complex manifold (M, J). The following assertions
are equivalent:
• X is real holomorphic
• L
X
J = 0
• The ﬂow of X consists of holomorphic transformations of M.
Although not explicitly stated, the reader might have guessed that a map f : (M, J
1
) → (N, J
2
)
between two complex manifolds is called holomorphic if its diﬀerential commutes with the complex
structures at each point: f
∗
◦ J
1
= J
2
◦ f
∗
.
Proof. The equivalence of the last two assertions is tautological. In order to prove the
equivalence of the ﬁrst two assertions, we ﬁrst notice that a complex vector ﬁeld Z is of type
(0,1) if and only if Z(f) = 0 for every locally deﬁned holomorphic function f. Suppose that X is
real holomorphic and let Y be an arbitrary vector ﬁeld and f a local holomorphic function. As
(X + iJX)f = 0, we have (X − iJX)f = 2X(f). By deﬁnition X(f) is then holomorphic so by
Lemma 2.3 we get (Y +iJY )(X(f)) = 0 and (Y +iJY )(f) = 0. This implies [Y +iJY, X](f) = 0.
This holds for every holomorphic f so [Y +iJY, X] has to be of type (0,1), that is [JY, X] = J[Y, X].
Hence L
X
J(Y ) = L
X
(JY )−J(L
X
Y ) = [X, JY ]−J[X, Y ] = 0 for all vector ﬁelds Y , i.e. L
X
J = 0.
The converse is similar and left to the reader.
We close this section with the following important result:
Proposition 2.8. (The local i∂
¯
∂–Lemma). Let ω ∈ Λ
1,1
M ∩ Λ
2
M be a real 2–form of type
(1,1) on a complex manifold M. Then ω is closed if and only if every point x ∈ M has an open
neighborhood U such that ω[
U
= i∂
¯
∂u for some real function u on U.
Proof. One implication is clear from Lemma 2.4:
d(i∂
¯
∂) = i(∂ +
¯
∂)∂
¯
∂ = i(∂
2
¯
∂ −∂
¯
∂
2
) = 0.
2. HOLOMORPHIC FORMS AND VECTOR FIELDS 15
The other implication is more delicate and needs the following counterpart of the Poincar´e Lemma
(see [2] p. 25 for a proof):
Lemma 2.9.
¯
∂–Poincar´e Lemma. A
¯
∂–closed (0,1)–form is locally
¯
∂–exact.
Let ω be a closed real form of type (1,1). From the Poincar´e Lemma, there exists locally a 1–form
τ with dτ = ω. Let τ = τ
1,0
+ τ
0,1
be the decomposition of τ in forms of type (1,0) and (0,1).
Clearly, τ
1,0
= τ
0,1
. Comparing types in the equality
ω = dτ =
¯
∂τ
0,1
+ (∂τ
0,1
+
¯
∂τ
1,0
) + ∂τ
1,0
,
we get
¯
∂τ
0,1
= 0 and ω = (∂τ
0,1
+
¯
∂τ
1,0
). The
¯
∂–Poincar´e Lemma yields a local function f such that
τ
0,1
=
¯
∂f. By conjugation we get τ
1,0
= ∂
¯
f, hence ω = (∂τ
0,1
+
¯
∂τ
1,0
) = ∂
¯
∂f+
¯
∂∂
¯
f = i∂
¯
∂(2Im(f)),
and the Proposition follows, with u := 2Im(f).
2.3. Exercises.
(1) Prove Lemma 2.1.
(2) Prove that the object deﬁned by formula (4) is indeed a tensor.
(3) Show that a almost complex structure on a real 2–dimensional manifold is always inte
grable.
(4) Show that ¦dz
α
¦ and ¦
∂
∂zα
¦ are dual basis of Λ
1,0
M and T
1,0
M at each point of the local
coordinate system.
(5) Show that a 2–form ω is of type (1,1) if and only if ω(X, Y ) = ω(JX, JY ), ∀ X, Y ∈ TM.
(6) Let M be a complex manifold with local holomorphic coordinates ¦z
α
¦.
• Prove that a local vector ﬁeld of type (1,0) Z =
¸
Z
α
∂
∂zα
is holomorphic if and only
if Z
α
are holomorphic functions.
• Prove that a local form of type (1,0) ω =
¸
ω
α
dz
α
is holomorphic if and only if ω
α
are holomorphic functions.
(7) If X is a real holomorphic vector ﬁeld on a complex manifold, prove that JX has the
same property.
(8) Prove the converse in Lemma 2.7.
16
(9) Show that in every local coordinate system one has
∂f =
n
¸
α=1
∂f
∂z
α
dz
α
and
¯
∂f =
n
¸
α=1
∂f
∂¯ z
α
d¯ z
α
.
(10) Let N be a manifold, and let T be a complex sub–bundle of Λ
1
C
N such that T ⊕Λ
1
N =
Λ
1
C
N. Show that there exists a unique almost complex structure J on N such that
T = Λ
1,0
N with respect to J.
3. COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 17
3. Complex and holomorphic vector bundles
3.1. Holomorphic vector bundles. Let M be a complex manifold and let π : E → M be
a complex vector bundle over M (i.e. each ﬁber π
−1
(x) is a complex vector space). E is called
holomorphic vector bundle if there exists a trivialization with holomorphic transition functions.
More precisely, there exists an open cover  of M and for each U ∈  a diﬀeomorphism ψ
U
:
π
−1
(U) →U C
k
such that
• the following diagram commutes:
π
−1
U
π
ψ
U
GG
U C
k
pr
U
ÑÑÓ
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
Ó
U
• for every intersecting U and V one has ψ
U
◦ψ
−1
V
(x, v) = (x, g
UV
(x)v), where g
UV
: U∩V →
Gl
k
(C) ⊂ C
k
2
are holomorphic functions.
Examples. 1. The tangent bundle of a complex manifold M
2m
is holomorphic. To see this, take
a holomorphic atlas (U, φ
U
) on M and deﬁne ψ
U
: TM[
U
→U C
m
by ψ
U
(X
x
) = (x, (φ
U
)
∗
(X)).
The transition functions g
UV
= (φ
U
)
∗
◦ (φ
V
)
−1
∗
are then clearly holomorphic.
2. The cotangent bundle, and more generally the bundles Λ
p,0
M are holomorphic. Indeed, using
again a holomorphic atlas of the manifold one can trivialize locally Λ
p,0
M and the chain rule
dz
α
1
∧ . . . ∧ dz
αp
=
¸
β
1
,..,βp
∂z
α
1
∂w
β
1
∂z
αp
∂w
βp
dw
β
1
∧ . . . ∧ dw
βp
shows that the transition functions are holomorphic.
For every holomorphic bundle E one deﬁnes the bundles Λ
p,q
E := Λ
p,q
M ⊗E of E–valued forms
on M of type (p, q) and the
¯
∂–operator
¯
∂ : (
∞
(Λ
p,q
E) → (
∞
(Λ
p,q+1
E) in the following way. If a
section σ of Λ
p,q
(E) is given by σ = (ω
1
, . . . , ω
k
) in some local trivialization (where ω
i
are local
(p, q)–forms), we deﬁne
¯
∂σ := (
¯
∂ω
1
, . . . ,
¯
∂ω
k
). Suppose that σ is written σ = (τ
1
, . . . , τ
k
) in some
other trivialization of E. Then one has τ
j
=
¸
g
jk
ω
k
for some holomorphic functions g
jk
, thus
¯
∂τ
j
=
¸
g
jk
¯
∂ω
k
, showing that
¯
∂σ does not depend on the chosen trivialization. By construction
one has
¯
∂
2
= 0 and
¯
∂ satisﬁes the Leibniz rule:
¯
∂(ω ∧ σ) = (
¯
∂ω) ∧ σ + (−1)
p+q
ω ∧ (
¯
∂σ), ∀ ω ∈ (
∞
(Λ
p,q
M), σ ∈ (
∞
(Λ
r,s
E).
Notice that the bundles Λ
p,q
M are not in general holomorphic bundles for q = 0.
3.2. Holomorphic structures. An operator
¯
∂ : (
∞
(Λ
p,q
E) → (
∞
(Λ
p,q+1
E) on a complex
vector bundle E satisfying the Leibniz rule is called a pseudo–holomorphic structure. If, moreover,
¯
∂
2
= 0, then
¯
∂ is called a holomorphic structure.
18
A section σ in a pseudo–holomorphic vector bundle (E,
¯
∂) is called holomorphic if
¯
∂σ = 0.
Lemma 3.1. A pseudo–holomorphic vector bundle (E,
¯
∂) of rank k is holomorphic if and only if
each x ∈ M has an open neighborhood U and k holomorphic sections σ
i
of E over U such that
¦σ
i
(x)¦ form a basis of E
x
(and hence on some neighborhood of x).
Proof. If E is holomorphic, one can deﬁne for every local holomorphic trivialization (U, ψ
U
)
a local basis of holomorphic sections by σ
i
(x) := ψ
−1
U
(x, e
i
), ∀ x ∈ U. Conversely, every local
basis of holomorphic sections deﬁnes a local trivialization of E, and if ¦σ
i
¦ and ¦˜ σ
i
¦ are two such
holomorphic basis, we can write σ
i
=
¸
g
ij
˜ σ
j
, which immediately yields (applying
¯
∂ and using
Leibniz’ rule) that
¯
∂g
ij
= 0, hence the transition functions are holomorphic.
Theorem 3.2. A complex vector bundle E is holomorphic if and only if it has a holomorphic
structure
¯
∂.
Proof. The “only if” part follows directly from the discussion above. Suppose, conversely,
that E is a complex bundle over M of rank k with holomorphic structure
¯
∂ satisfying Leibniz’
rule and
¯
∂
2
= 0. In order to show that E is holomorphic, it is enough to show, using Lemma
3.1, that one can trivialize E around each x ∈ M by holomorphic sections. Let ¦σ
1
, . . . , σ
k
¦ be
local sections of E which form a basis of E over some open set U containing x. We deﬁne local
(0,1)–forms τ
ij
on U by the formula
¯
∂σ
i
=
k
¸
j=1
τ
ij
⊗σ
j
.
The condition
¯
∂
2
= 0, together with Leibniz’ rule, yields
0 =
¯
∂
2
σ
i
=
k
¸
j=1
¯
∂τ
ij
⊗σ
j
−
k
¸
j,l=1
τ
il
∧ τ
lj
⊗σ
j
,
hence
(5)
¯
∂τ
ij
=
k
¸
l=1
τ
il
∧ τ
lj
, ∀ 1 ≤ i, j ≤ k.
From now on we will use the summation convention on repeating indexes. Suppose one can ﬁnd
a map f : U
→Gl
k
(C), f = (f
ij
) such that
(6) 0 =
¯
∂f
ij
+ f
il
τ
lj
, ∀ 1 ≤ i, j ≤ k,
for some open subset U
of U containing x. It is then easy to check that the local sections s
j
of
E over U
deﬁned by s
j
:= f
jl
σ
l
are holomorphic:
¯
∂s
j
=
¯
∂f
jl
⊗σ
l
+ f
jr
τ
rl
⊗σ
l
= 0.
The theorem thus follows from the next lemma.
3. COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 19
Lemma 3.3. Suppose that τ := (τ
ij
) is a gl
k
(C)–valued (0,1)–form on U satisfying
¯
∂τ = τ ∧τ, or
equivalently (5). Then for every x ∈ U there exists some open subset U
of U containing x and a
map f : U
→Gl
k
(C), f = (f
ij
) such that
¯
∂f + fτ = 0, or equivalently such that (6) holds.
Proof. The main idea is to deﬁne an almost complex structure locally on U C
k
using τ, to
show that its integrability is equivalent to (5), and ﬁnally to obtain f as the matrix of some frame
deﬁned by τ in terms of holomorphic coordinates given by the theorem of Newlander–Nirenberg.
We denote by N the product U C
k
. We may suppose that U is an open subset of C
m
with
holomorphic coordinates z
α
and denote by w
i
the coordinates in C
k
.
It is an easy exercise to check that any complement T of Λ
1
N in the complexiﬁed bundle Λ
1
C
N of
some manifold N
2n
, with iT = T, deﬁnes an almost complex structure on N, such that T becomes
the space of (1,0)–forms on N.
Consider the sub–bundle T of Λ
1
N ⊗C generated by the 1–forms
¦dz
α
, dw
i
−τ
il
w
l
[ 1 ≤ α ≤ m, 1 ≤ i ≤ k¦.
We claim that the almost complex structure induced on N by T is integrable. By Proposition
2.2, we have to show that d(
∞
(T) ⊂ (
∞
(T ∧ Λ
1
C
N). It is enough to check this on the local basis
deﬁning T. Clearly d(dz
α
) = 0 and from (5) we get
d(dw
i
−τ
il
w
l
) = −∂τ
il
w
l
−
¯
∂τ
il
w
l
+ τ
il
∧ dw
l
= −∂τ
il
w
l
−τ
is
∧ τ
sl
w
l
+ τ
is
∧ dw
s
= −∂τ
il
w
l
+ τ
is
∧ (dw
s
−τ
sl
w
l
),
which clearly is a section of (
∞
(T ∧ Λ
1
C
N). We now use the Newlander–Nirenberg theorem and
complete the family ¦z
α
¦ to a local holomorphic coordinate system ¦z
α
, u
l
¦ on some smaller
neighborhood U
of x. Since du
l
are sections of T, we can ﬁnd functions F
li
and F
lα
, 1 ≤ i, l ≤ k,
1 ≤ α ≤ m such that
du
l
= F
li
(dw
i
−τ
ik
w
k
) + F
lα
dz
α
.
We apply the exterior derivative to this system and get
0 = dF
li
∧ (dw
i
−τ
ik
w
k
) + F
li
(−dτ
ik
w
k
+ τ
ik
∧ dw
k
) + dF
lα
∧ dz
α
.
We evaluate this last equality for w
i
= 0, and get
(7) 0 = dF
lk
(z, 0) ∧ dw
k
+ F
li
(z, 0)τ
ik
∧ dw
k
+ dF
lα
∧ dz
α
.
If we denote f
lk
(z) := F
lk
(z, 0), then the Λ
0,1
U
–part of dF
lk
(z, 0) is just
¯
∂f
lk
. Therefore, the
vanishing of the Λ
0,1
U
⊗Λ
1,0
C
k
–components of (7) just reads
0 =
¯
∂f
lk
+ f
li
τ
ik
.
20
3.3. The canonical bundle of CP
m
. For a complex manifold (M
2m
, J), the complex line
bundle K
M
:= Λ
m,0
M is called the canonical bundle of M. We already noticed that K
M
has a
holomorphic structure.
On the complex projective space there is some distinguished holomorphic line bundle called the
tautological line bundle. It is deﬁned as the complex line bundle π : L → CP
m
whose ﬁber L
[z]
over some point [z] ∈ CP
m
is the complex line < z > in C
m+1
.
We consider the canonical holomorphic charts (U
α
, φ
α
) on CP
m
and the local trivializations ψ
α
:
π
−1
U
α
→ U
α
C of L deﬁned by ψ
α
([z], w) = ([z], w
α
). It is an easy exercise to compute the
transition functions:
ψ
α
◦ ψ
−1
β
([z], λ) = ([z], g
αβ
([z])λ), where g
αβ
([z]) =
z
α
z
β
,
which are clearly holomorphic. The aim of this subsection is to prove the following
Proposition 3.4. The canonical bundle of CP
m
is isomorphic to the m + 1
st
power of the
tautological bundle.
Proof. A trivialization for p : Λ
m,0
CP
m
→CP
m
is given by (φ
∗
α
)
−1
: p
−1
U
α
→U
α
Λ
m,0
C
m
,
so the transition functions are h
αβ
:= (φ
∗
α
)
−1
◦ (φ
∗
β
). Let now ω := dw
1
∧ . . . ∧ dw
m
be the
canonical generator of Λ
m,0
C
m
. We introduce holomorphic coordinates on U
α
∩ U
β
: a
i
:=
z
i
zα
for
i ∈ ¦0, . . . m¦ −¦α¦ and b
i
:=
z
i
z
β
for i ∈ ¦0, . . . m¦ −¦β¦. Then
φ
∗
α
(ω) = da
0
∧ . . . ∧ da
α−1
∧ da
α+1
∧ . . . ∧ da
m
and
φ
∗
β
(ω) = db
0
∧ . . . ∧ db
β−1
∧ db
β+1
∧ . . . ∧ db
m
.
Therefore we can write
(8) db
0
∧ . . . ∧ db
β−1
∧ db
β+1
∧ . . . ∧ db
m
= h
αβ
da
0
∧ . . . ∧ da
α−1
∧ da
α+1
∧ . . . ∧ da
m
.
On the other hand, for every i = α, β we have a
i
= b
i
a
β
and a
β
b
α
= 1. This shows that
da
i
= a
β
db
i
+b
i
da
β
for i = α, β and da
β
= −
1
b
2
α
db
α
= −a
2
β
db
α
, and an easy algebraic computation
then yields
da
0
∧ . . . ∧ da
α−1
∧ da
α+1
∧ . . . ∧ da
m
= (−1)
α−β
a
m+1
β
db
0
∧ . . . ∧ db
β−1
∧ db
β+1
∧ . . . ∧ db
m
.
Using (8) we thus see that the transition functions are given by
h
αβ
= (−1)
α−β
a
−m−1
β
= (−1)
α−β
z
α
z
β
m+1
.
Finally, denoting c
α
:= (−1)
α
we have c
α
h
αβ
c
−1
β
= g
m+1
αβ
, which proves that
K
CP
m
· L
m+1
.
3. COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 21
3.4. Exercises.
(1) Prove that any holomorphic function on a compact manifold f : M → C is constant.
Hint: use the maximum principle.
(2) Let E → M be a rank k complex vector bundle whose transition functions with respect
to some open cover ¦U
α
¦ of M are g
αβ
. Show that a section σ : M → E of E can be
identiﬁed with a collection ¦σ
α
¦ of smooth maps σ
α
: U
α
→C
k
satisfying σ
α
= g
αβ
σ
β
on
U
α
∩ U
β
.
(3) Let πE →M be a complex vector bundle over a complex manifold M. Prove that E has
a holomorphic structure if and only if there exists a complex structure on E as manifold,
such that the projection π is a holomorphic map.
(4) The tautological line bundle. Let L be the complex line bundle π : L →CP
m
whose ﬁber
L
[z]
over some point [z] ∈ CP
m
is the complex line < z > in C
m+1
. Prove that E is a
holomorphic line bundle. Hint: Use the local trivializations ψ
α
: π
−1
U
α
→U
α
C deﬁned
by ψ
α
([z], w) = ([z], w
α
).
(5) Show that the tautological line bundle L has no non–trivial holomorphic sections.
(6) The hyperplane line bundle. Let H := L
∗
be the dual of L. Thus, the ﬁber of H over
some point [z] ∈ CP
m
is the set of linear maps < z >→C. Find local trivializations for
H with holomorphic transition functions. Find the dimension of the space of holomorphic
sections of H.
Part 2
Hermitian and K¨ahler structures
4. Hermitian bundles
4.1. Connections on complex vector bundles. Let M be a diﬀerentiable manifold (not
necessarily complex) and let E →M be a complex vector bundle over M.
Definition 4.1. A (C–linear) connection on a E is a C–linear diﬀerential operator ∇ : (
∞
(E) →
(
∞
(Λ
1
(E)) satisfying the Leibniz rule
∇(fσ) = df ⊗σ + f∇σ, ∀ f ∈ (
∞
(M).
One can extend any connection to the bundles of E–valued p–forms on M by
∇(ω ⊗σ) = dω ⊗σ + (−1)
p
ω ∧ ∇σ,
where the wedge product has to be understood as
ω ∧ ∇σ :=
n
¸
i=1
ω ∧ e
∗
i
⊗∇
e
i
σ
for any local basis ¦e
i
¦ of TM with dual basis ¦e
∗
i
¦.
The curvature of ∇ is the End(E)–valued 2–form R
∇
deﬁned by
R
∇
(σ) := ∇(∇σ)) ∀ σ ∈ (
∞
(E).
To check that this is indeed tensorial, we can write:
∇
2
(fσ) = ∇(df ⊗σ + f∇σ) = d
2
f ⊗σ −df ∧ ∇σ + df ∧ ∇σ + f∇
2
σ = f∇
2
σ.
More explicitly, if ¦σ
1
, . . . , σ
k
¦ are local sections of E which form a basis of each ﬁber over some
open set U, we deﬁne the connection forms ω
ij
∈ Λ
1
(U) (relative to the choice of the base) by
∇σ
i
= ω
ij
⊗σ
j
.
We deﬁne the two–forms R
∇
ij
by
R
∇
(σ
i
) = R
∇
ij
⊗σ
j
,
and compute
R
∇
ij
⊗σ
j
= R
∇
(σ
i
) = ∇(ω
ij
⊗σ
j
) = (dω
ij
) ⊗σ
j
−ω
ik
∧ ω
kj
⊗σ
j
,
showing that
(9) R
∇
ij
= dω
ij
−ω
ik
∧ ω
kj
.
4.2. Hermitian structures and connections. Let E → M be a complex rank k bundle
over some manifold M. We do not assume for the moment that M has an almost complex
structure.
Definition 4.2. A Hermitian structure H on E is a smooth ﬁeld of Hermitian products on the
ﬁbers of E, that is, for every x ∈ M, H : E
x
E
x
→C satisﬁes
• H(u, v) is C–linear in u for every v ∈ E
x
.
• H(u, v) = H(v, u) ∀ u, v ∈ E
x
.
4. HERMITIAN BUNDLES 25
• H(u, u) > 0 ∀ u = 0.
• H(u, v) is a smooth function on M for every smooth sections u, v of E.
It is clear from the above conditions that H is C–anti–linear in the second variable. The third
condition shows that H is non–degenerate. In fact, it is quite useful to think to H as to a
C–anti–linear isomorphism H : E →E
∗
.
Every rank k complex vector bundle E admits Hermitian structures. To see this, just take a
trivialization (U
i
, ψ
i
) of E and a partition of the unity f
i
subordinate to the open cover ¦U
i
¦
of M. For every x ∈ U
i
, let (H
i
)
x
denote the pull–back of the Hermitian metric on C
k
by the
C–linear map ψ
i
[
Ex
. Then H :=
¸
f
i
H
i
is a well–deﬁned Hermitian structure on E.
Suppose now that M is a complex manifold. Consider the projections π
1,0
: Λ
1
(E) → Λ
1,0
(E)
and π
0,1
: Λ
1
(E) → Λ
0,1
(E). For every connection ∇ on E, one can consider its (1,0) and
(0,1)–components ∇
1,0
:= π
1,0
◦ ∇ and ∇
0,1
:= π
0,1
◦ ∇. From Proposition 2.2, we can extend
these operators to ∇
1,0
: (
∞
(Λ
p,q
(E)) →(
∞
(Λ
p+1,q
(E)) and ∇
0,1
: (
∞
(Λ
p,q
(E)) →(
∞
(Λ
p,q+1
(E))
satisfying the Leibniz rule:
∇
1,0
(ω ⊗σ) = ∂ω ⊗σ + (−1)
p+q
ω ∧ ∇
1,0
σ, ∇
0,1
(ω ⊗σ) =
¯
∂ω ⊗σ + (−1)
p+q
ω ∧ ∇
0,1
σ,
for all ω ∈ (
∞
(Λ
p,q
M), σ ∈ (
∞
(E). Of course, ∇
0,1
is a pseudo–holomorphic structure on E for
every connection ∇.
For every section σ of E one can write
R
∇
(σ) = ∇
2
σ = (∇
1,0
+∇
0,1
)
2
(σ) = (∇
1,0
)
2
(σ) + (∇
0,1
)
2
(σ) + (∇
1,0
∇
0,1
+∇
0,1
∇
1,0
)(σ),
so the Λ
0,2
part of the curvature is given by
(R
∇
)
0,2
= (∇
0,1
)
2
.
Theorem 3.2 shows that if the Λ
0,2
–part of the curvature of some connection D on E vanishes,
then E is a holomorphic bundle with holomorphic structure
¯
∂ := ∇
0,1
. The converse is also true:
simply choose an arbitrary Hermitian metric on E and apply Theorem 4.3 below.
We say that ∇ is a H–connection if H, viewed as a ﬁeld of C–valued real bilinear forms on E, is
parallel with respect to ∇. We can now state the main result of this section:
Theorem 4.3. For every Hermitian structure H in a holomorphic bundle E with holomorphic
structure
¯
∂, there exists a unique H–connection ∇ (called the Chern connection) such that ∇
0,1
=
¯
∂.
Proof. Let us ﬁrst remark that the dual vector bundle E
∗
is also holomorphic, with holomor
phic structure denoted by
¯
∂, and that any connection ∇ on E induces canonically a connection,
also denoted by ∇, on E
∗
by the formula
(10) (∇
X
σ
∗
)(σ) := X(σ
∗
(σ)) −σ
∗
(∇
X
σ), ∀ X ∈ TM, σ ∈ (
∞
(E), σ
∗
∈ (
∞
(E
∗
).
Note also that ∇
0,1
=
¯
∂ on E just means that ∇σ ∈ (
∞
(Λ
1,0
(E)) for every holomorphic section σ
of E. From (10), if this property holds on E, then it holds on E
∗
, too.
26
After these preliminaries, suppose that ∇ is a H–connection with ∇
0,1
=
¯
∂. The C–anti–linear
isomorphism H : E →E
∗
is then parallel, so for every section σ of E and every real vector X on
M we get
∇
X
(H(σ)) = ∇
X
(H)(σ) + H(∇
X
σ) = H(∇
X
σ).
By the C–anti–linearity of H, for every complex vector Z ∈ TM
C
we have
∇
Z
(H(σ)) = H(∇¯
Z
σ).
For Z ∈ T
1,0
M, this shows that
(11) ∇
1,0
(σ) = H
−1
◦ ∇
0,1
(H(σ)) = H
−1
◦
¯
∂(H(σ)),
hence ∇ =
¯
∂ + H
−1
◦
¯
∂ ◦ H, which proves the existence and uniqueness of ∇.
Remark. The (0,2)–component of the curvature of the Chern connection vanishes. Indeed,
R
0,2
(σ) = ∇
0,1
(∇
0,1
(σ)) =
¯
∂
2
(σ) = 0.
Its (2,0)–component actually vanishes, too, since by (11),
∇
1,0
(∇
1,0
(σ)) = ∇
1,0
(H
−1
◦
¯
∂(H(σ)) = H
−1
◦
¯
∂
2
(H(σ)) = 0.
4.3. Exercises.
(1) Let E → M be a complex vector bundle and denote by E
∗
and
¯
E its dual and its
conjugate. (Recall that for every x ∈ M, the ﬁbre of E
∗
over x is just the dual of E
x
and
the ﬁbre
¯
E
x
of
¯
E is equal to E
x
as a set, but has the conjugate complex structure, in the
sense that the action of some complex number z on
¯
E
x
is the same as the action of ¯ z on
E
x
). If g
αβ
denote the transition functions of E with respect to some open cover ¦U
α
¦ of
M, ﬁnd the transition functions of E
∗
and
¯
E with respect to the same open cover.
(2) Show that a Hermitian structure on a complex vector bundle E deﬁnes an isomorphism
between E
∗
and
¯
E as complex vector bundles.
(3) Let E → M be a rank k complex vector bundle. Viewing local trivializations of E as
local basis of sections of E, show that if the transition functions of E with respect to
some local trivialization take values in the unitary group U
k
⊂ Gl
k
(C) then there exists
a canonically deﬁned Hermitian structure on E.
(4) Prove the naturality of the Chern connection with respect to direct sums and tensor
products of holomorphic vector bundles.
5. HERMITIAN AND K
¨
AHLER METRICS 27
5. Hermitian and K¨ahler metrics
5.1. Hermitian metrics. We start with the following
Definition 5.1. A Hermitian metric on an almost complex manifold (M, J) is a Riemannian
metric h such that h(X, Y ) = h(JX, JY ), ∀ X, Y ∈ TM. The fundamental form of a Hermitian
metric is deﬁned by Ω(X, Y ) := h(JX, Y ).
The extension (also denoted by h) of the Hermitian metric to TM
C
by C–linearity satisﬁes
(12)
h(
¯
Z,
¯
W) = h(Z, W), ∀ Z, W ∈ TM
C
.
h(Z,
¯
Z) > 0 ∀ Z ∈ TM
C
−¦0¦.
h(Z, W) = 0, ∀ Z, W ∈ T
1,0
M and ∀ Z, W ∈ T
0,1
M.
Conversely, each symmetric tensor on TM
C
with these properties deﬁnes a Hermitian metric by
restriction to TM (exercise).
Remark. The tangent bundle of an almost complex manifold is in particular a complex vector
bundle. If h is a Hermitian metric on M, then H(X, Y ) := h(X, Y ) −ih(JX, Y ) = (h−iΩ)(X, Y )
deﬁnes a Hermitian structure on the complex vector bundle (TM, J), as deﬁned in the previous
section. Conversely, any Hermitian structure H on TM as complex vector bundle deﬁnes a
Hermitian metric h on M by h := Re(H).
Remark. Every almost complex manifold admits Hermitian metrics. Simply choose an arbitrary
Riemannian metric g and deﬁne h(X, Y ) := g(X, Y ) + g(JX, JY ).
Let z
α
be holomorphic coordinates on a complex Hermitian manifold (M
2m
, J, h) and denote by
h
α
¯
β
the coeﬃcients of the metric tensor in these local coordinates:
h
α
¯
β
:= h
∂
∂z
α
,
∂
∂¯ z
β
.
Lemma 5.2. The fundamental form is given by
Ω = i
m
¸
α,β=1
h
α
¯
β
dz
α
∧ d¯ z
β
.
The proof is left as an exercise.
5.2. K¨ahler metrics. Suppose that the fundamental form Ω of a complex Hermitian mani
fold is closed. From the i∂
¯
∂–Lemma we get locally a real function u such that Ω = i∂
¯
∂u, which
in local coordinates reads
h
α
¯
β
=
∂
2
u
∂z
α
∂¯ z
β
.
This particularly simple expression for the metric tensor in terms of one single real function
deserves the following
28
Definition 5.3. A Hermitian metric h on an almost complex manifold (M, J) is called a K¨ahler
metric if J is a complex structure and the fundamental form Ω is closed:
h is K¨ahler ⇐⇒
N
J
= 0
dΩ = 0
A local real function u satisfying Ω = i∂
¯
∂u is called a local K¨ahler potential of the metric h.
Our aim (as Riemannian geometers) is to express the K¨ahler condition in terms of the covariant
derivative of the Levi–Civita connection of h. We start with doing so for the Nijenhuis tensor.
Lemma 5.4. Let h be a Hermitian metric on an almost complex manifold (M, J), with Levi–Civita
covariant derivative ∇. Then J is integrable if and only if
(13) (∇
JX
J)Y = J(∇
X
J)Y, ∀ X, Y ∈ TM.
Proof. Let us ﬁx a point x ∈ M and extend X and Y to vector ﬁelds on M parallel with
respect to ∇ at x. Then we can write
N
J
(X, Y ) = [X, Y ] + J[JX, Y ] + J[X, JY ] −[JX, JY ]
= J(∇
X
J)Y −J(∇
Y
J)X −(∇
JX
J)Y + (∇
JY
J)X
= (J(∇
X
J)Y −(∇
JX
J)Y ) −(J(∇
Y
J)X −(∇
JY
J)X),
thus proving that (13) implies N
J
= 0. Conversely, suppose that N
J
= 0 and denote by
A(X, Y, Z) = h(J(∇
X
J)Y − (∇
JX
J)Y ), Z). The previous equation just reads A(X, Y, Z) =
A(Y, X, Z). On the other hand, A is skew–symmetric in the last two variables, since J and ∇
X
J
are anti–commuting skew–symmetric operators. Thus A(X, Y, Z) = −A(X, Z, Y ), so by circular
permutations we get
A(X, Y, Z) = −A(Y, Z, X) = A(Z, X, Y ) = −A(X, Y, Z),
which implies (13).
Theorem 5.5. A Hermitian metric h on an almost complex manifold is K¨ahler if and only if J
is parallel with respect to the Levi–Civita connection of h.
Proof. One direction is obvious, since if J is parallel, then N
J
clearly vanishes, and as
Ω = h(J, ), we also have ∇Ω = 0, so in particular dΩ = 0. Conversely, suppose that h is K¨ahler
and denote by B the tensor B(X, Y, Z) := h((∇
X
J)Y, Z). As J and ∇
X
J anti–commute we have
B(X, Y, JZ) = B(X, JY, Z).
From (13) we get
B(X, Y, JZ) + B(JX, Y, Z) = 0.
Combining these two relations also yields
B(X, JY, Z) + B(JX, Y, Z) = 0.
5. HERMITIAN AND K
¨
AHLER METRICS 29
We now use dΩ = 0 twice, ﬁrst on X, Y, JZ, then on X, JY, Z and get:
B(X, Y, JZ) + B(Y, JZ, X) + B(JZ, X, Y ) = 0,
B(X, JY, Z) + B(JY, Z, X) + B(Z, X, JY ) = 0.
Adding these two relations and using the previous properties of B yields 2B(X, Y, JZ) = 0, that
is, J is parallel.
5.3. Characterization of K¨ahler metrics. We will now prove the analytic characterization
of K¨ahler metrics described in the ﬁrst section.
Theorem 5.6. Let h be a Hermitian metric on a complex manifold (M
2m
, J). Then h is K¨ahler
if and only if around each point of M there exist holomorphic coordinates in which h osculates to
the standard Hermitian metric up to the order 2.
Proof. Suppose that we can ﬁnd holomorphic local coordinates z
α
= x
α
+iy
α
around x ∈ M
such that h
α
¯
β
=
1
2
δ
αβ
+ r
αβ
, with r
αβ
(x) = 0
∂r
αβ
∂x
γ
(x) =
∂r
αβ
∂y
γ
(x) = 0
at x. Then
dΩ = i
m
¸
α,β,γ=1
∂h
α
¯
β
∂x
γ
dx
γ
+
∂h
α
¯
β
∂y
γ
dy
γ
∧ dz
α
∧ d¯ z
β
clearly vanishes at x. As x was arbitrary, this means dΩ = 0.
Conversely, if the metric is K¨ahler, for every x ∈ M we take an orthonormal basis of T
x
M of the
form ¦e
1
, . . . , e
m
, Je
1
, . . . , Je
m
¦ and choose a local holomorphic coordinate system (z
α
= x
α
+iy
α
)
around x such that
e
α
=
∂
∂x
α
(x) and Je
α
=
∂
∂y
α
(x).
The fundamental 2–form Ω can be written as
Ω = i
¸
α,β
1
2
δ
αβ
+
¸
γ
a
αβγ
z
γ
+
¸
γ
a
αβ¯ γ
¯ z
γ
+ o([z[)
dz
α
∧ d¯ z
β
,
where o([z[) denotes generically a function whose 1–jet vanishes at x. The condition h
α
¯
β
= h
β¯ α
together with Lemma 5.2 implies
(14) a
αβ¯ γ
= α
βαγ
,
and from dΩ = 0 we ﬁnd
(15) a
αβγ
= a
γβα
.
We now look for a local coordinate change in which the fundamental form has vanishing ﬁrst order
terms. We put
z
α
= w
α
+
1
2
¸
β,γ
b
αβγ
w
β
w
γ
,
30
where b
αβγ
are complex numbers satisfying b
αβγ
= b
αγβ
. This coordinate change is well–deﬁned
locally thanks to the holomorphic version of the local inversion theorem. We then have
dz
α
= dw
α
+
¸
β,γ
b
αβγ
w
β
dw
γ
,
whence (using Einstein’s summation convention)
Ω = i
1
2
δ
αβ
+ a
αβγ
z
γ
+ a
αβ¯ γ
¯ z
γ
+ o([z[)
dz
α
∧ d¯ z
β
= i
1
2
δ
αβ
+ a
αβγ
w
γ
+ a
αβ¯ γ
¯ w
γ
+ o([z[)
(dw
α
+ b
αετ
w
ε
dw
τ
) ∧ (d ¯ w
β
+ b
βετ
¯ w
ε
d ¯ w
τ
)
= i
1
2
δ
αβ
+ a
αβγ
w
γ
+ a
αβ¯ γ
¯ w
γ
+ b
βγα
w
γ
+ b
αγβ
¯ w
γ
+ o([z[)
dw
α
∧ d ¯ w
β
.
If we choose b
βγα
= −a
αβγ
, (which is possible because of (15) which ensures that a
αβγ
is symmetric
in α and γ), then from (14) we get
b
αγβ
= −a
βαγ
= −a
αβ¯ γ
,
showing that
Ω = i
1
2
δ
αβ
+ o([z[)
dw
α
∧ d ¯ w
β
.
5.4. Comparison of the Levi–Civita and Chern connections. Our next aim is to ex
press the
¯
∂–operator on a Hermitian manifold in terms of the Levi–Civita connection. In order to
do so, we have to remember that TM is identiﬁed with a complex vector bundle via the complex
structure J. In other words, a product iX for some X ∈ TM is identiﬁed with JX.
Lemma 5.7. For every section Y of the complex vector bundle TM, the
¯
∂–operator, as TM–valued
(0,1)–form is given by
¯
∂
∇
Y (X) =
1
2
(∇
X
Y + J∇
JX
Y −J(∇
Y
J)X),
where ∇ denotes the Levi–Civita connection of any Hermitian metric h on M.
Proof. We ﬁrst recall that
¯
∂f(X) =
1
2
(X + iJX)(f), so
¯
∂
∇
(fY )(X) = f
1
2
(∇
X
Y + J∇
JX
Y −J(∇
Y
J)X) +
1
2
(X(f)Y + JX(f)JY )
= f
¯
∂
∇
Y (X) +
¯
∂f(X)Y,
which shows that the above deﬁned operator
¯
∂
∇
satisﬁes the Leibniz rule. Moreover, a vector
ﬁeld Y is a holomorphic section of TM if and only if it is real holomorphic. By Lemma 2.7, this
5. HERMITIAN AND K
¨
AHLER METRICS 31
is equivalent to L
Y
J = 0, which means that for every vector ﬁeld X ∈ (
∞
(TM) one has
0 = (L
Y
J)X = L
Y
(JX) −JL
Y
X = [Y, JX] −J[Y, X]
= ∇
Y
JX −∇
JX
Y −J∇
Y
X + J∇
X
Y
= (∇
Y
J)X −∇
JX
Y + J∇
X
Y = J(∇
X
Y + J∇
JX
Y −J(∇
Y
J)X),
thus showing that
¯
∂
∇
Y vanishes for every holomorphic section Y . This proves that
¯
∂
∇
=
¯
∂.
A Hermitian manifold (M, h, J) two natural linear connections: the Levi–Civita connection ∇
and the Chern connection
¯
∇ on TM as Hermitian vector bundle.
Proposition 5.8. The Chern connection coincides with the Levi–Civita connection if and only if
h is K¨ahler.
Proof. Let H := h − iΩ denote the Hermitian structure of TM. By deﬁnition, J is parallel
with respect to the Chern connection, which is a complex connection. Thus, if ∇ =
¯
∇ then J
is ∇–parallel, so h is K¨ahler by Theorem 5.5. Conversely, suppose that h is K¨ahler. Then the
Levi–Civita connection is a well–deﬁned complex connection in TM since ∇J = 0, by Theorem
5.5 again. Moreover, it is a H–connection since ∇h = 0 and ∇Ω = 0. Finally, the condition
∇
0,1
=
¯
∂ follows from Lemma 5.7, as ∇
0,1
X
=
1
2
(∇
X
+ i∇
JX
) =
1
2
(∇
X
+ J∇
JX
).
5.5. Exercises.
(1) Prove that every Hermitian metric on a 2–dimensional almost complex manifold is K¨ahler.
(2) Prove that the fundamental form of a Hermitian metric is a (1,1)–form.
(3) If h
α
¯
β
denote the coeﬃcients of a Hermitian metric tensor in some local holomorphic
coordinate system, show that h
α
¯
β
= h
β¯ α
.
(4) Show that the extension of a Hermitian metric h by C–linearity is a symmetric bilinear
tensor satisfying
h(
¯
Z,
¯
W) = h(Z, W), ∀ Z, W ∈ TM
C
.
h(Z,
¯
Z) > 0 ∀ Z ∈ TM
C
−¦0¦.
h(Z, W) = 0, ∀ Z, W ∈ T
1,0
M and ∀ Z, W ∈ T
0,1
M.
and conversely, any symmetric complex bilinear tensor satisfying this system arises from
a Hermitian metric.
32
6. The curvature tensor of K¨ahler manifolds
6.1. The curvature tensor. Let (M
n
, g) be a Riemannian manifold with Levi–Civita con
nection ∇ and denote by R its curvature tensor:
R(X, Y )Z := ∇
X
∇
Y
Z −∇
Y
∇
X
Z −∇
[X,Y ]
Z, ∀ X, Y, Z ∈ (
∞
(TM).
It is sometimes convenient to identify the curvature tensor with the following tensor:
R(X, Y, Z, T) := h(R(X, Y )Z, T), ∀ X, Y, Z, T ∈ TM.
The symmetries of the curvature operator then read:
• R(X, Y, Z, T) = −R(X, Y, T, Z);
• R(X, Y, Z, T) = R(Z, T, X, Y );
• R(X, Y, Z, T) + R(Y, Z, X, T) + R(Z, X, Y, T) = 0 (ﬁrst Bianchi identity);
• (∇
X
R)(Y, Z, T, W) +(∇
Y
R)(Z, X, T, W) +(∇
Z
R)(X, Y, T, W) = 0 (second Bianchi iden
tity).
The Ricci tensor of M is deﬁned by
Ric(X, Y ) := Tr¦V →R(V, X)Y ¦,
or equivalently
Ric(X, Y ) =
2m
¸
i=1
R(e
i
, X, Y, e
i
),
where e
i
is a local orthonormal basis of TM. We recall that the Ricci tensor of every Riemannian
manifold is symmetric, as can be easily seen from the symmetries of the curvature. A Riemannian
manifold (M
n
, g) is called Einstein if the Ricci tensor Ric is proportional to the metric tensor g
at each point of M
Ric(X, Y ) = λg(X, Y ) ∀ X, Y ∈ T
x
M.
If n ≥ 3, it is easy to check that λ (which a priori depends on x) has to be constant on M (see
[8]).
Suppose now that (M
2m
, h, J) is a K¨ahler manifold. Since J is ∇–parallel, the curvature tensor
has more symmetries:
(16) R(X, Y )JZ = JR(X, Y )Z, ∀ X, Y, Z ∈ (
∞
(TM).
This immediately implies
R(X, Y, JZ, JT) = R(X, Y, Z, T) = R(JX, JY, Z, T),
hence
Ric(JX, JY ) =
2m
¸
i=1
R(e
i
, JX, JY, e
i
) =
2m
¸
i=1
R(Je
i
, X, Y, Je
i
) = Ric(X, Y ),
since for every orthonormal basis ¦e
i
¦, the set ¦Je
i
¦ is also an orthonormal basis.
This last equation justiﬁes the following
6. THE CURVATURE TENSOR OF K
¨
AHLER MANIFOLDS 33
Definition 6.1. The Ricci form ρ of a K¨ahler manifold is deﬁned by
ρ(X, Y ) := Ric(JX, Y ), ∀ X, Y ∈ TM.
The Ricci form is one of the most important objects on a K¨ahler manifold. Among its properties
which will be proved in the next few sections we mention:
• the Ricci form ρ is closed;
• the cohomology class of ρ is (up to some real multiple) equal to the Chern class of the
canonical bundle of M;
• in local coordinates, ρ can be expressed as ρ = −i∂
¯
∂ log det(h
α
¯
β
), where det(h
α
¯
β
) denotes
the determinant of the matrix (h
α
¯
β
) expressing the Hermitian metric.
For the moment being we use the Bianchi identities for the curvature tensor to prove the following
Proposition 6.2. (i) The Ricci tensor of a K¨ahler manifold satisﬁes
Ric(X, Y ) =
1
2
Tr(R(X, JY ) ◦ J).
(ii) The Ricci form is closed.
Proof. Let (e
i
) be a local orthonormal basis of TM. (i) Using the ﬁrst Bianchi identity we
get
Ric(X, Y ) =
¸
i
R(e
i
, X, Y, e
i
) =
¸
i
R(e
i
, X, JY, Je
i
)
=
¸
i
(−R(X, JY, e
i
, Je
i
) −R(JY, e
i
, X, Je
i
))
=
¸
i
(R(X, JY, Je
i
, e
i
) + R(Y, Je
i
, X, Je
i
))
= Tr(R(X, JY ) ◦ J) −Ric(X, Y ).
(ii) From (i) we can write 2ρ(X, Y ) = Tr(R(X, Y ) ◦ J). Therefore,
2dρ(X, Y, Z) = 2((∇
X
ρ)(Y, Z) + (∇
Y
ρ)(Z, X) + (∇
Z
ρ)(X, Y ))
= Tr((∇
X
R)(Y, Z) ◦ J + (∇
Y
R)(Z, X) ◦ J + (∇
Z
R)(X, Y ) ◦ J) = 0
from the second Bianchi identity.
6.2. K¨ahler metrics in local coordinates. Let (M
2m
, h, J) be a K¨ahler manifold with
Levi–Civita covariant derivative ∇ and let (z
α
) be a system of local holomorphic coordinates. We
introduce the following local basis of the complexiﬁed tangent space:
Z
α
:=
∂
∂z
α
, Z
¯ α
:=
∂
∂¯ z
α
, 1 ≤ α ≤ m,
34
and we let subscripts A, B, C, . . . run over the set ¦1, . . . , m,
¯
1, . . . , ¯ m¦. We denote the components
of the K¨ahler metric in these coordinates by
h
AB
:= h(Z
A
, Z
B
).
Of course, since the metric is Hermitian we have
(17) h
αβ
= h
¯ α
¯
β
= 0, h¯
βα
= h
α
¯
β
= h
β¯ α
.
Let h
α
¯
β
denote the coeﬃcients of the inverse matrix of (h
α
¯
β
). The Christoﬀel symbols are deﬁned
by
∇
Z
A
Z
B
= Γ
C
AB
Z
C
.
Using the convention
¯
¯ α = α, we get by conjugation
Γ
C
AB
= Γ
¯
C
¯
A
¯
B
.
Since ∇ is torsion–free we have
Γ
C
AB
= Γ
C
BA
,
and since T
1,0
is ∇–parallel we must have
Γ
γ
A
¯
β
= 0.
These relations show that the only non–vanishing Christoﬀel symbols are
Γ
γ
αβ
and Γ
¯ γ
¯ α
¯
β
.
Now, in order to compute these coeﬃcients we notice that Γ
C
α
¯
δ
= 0 implies
(18) ∇
Zα
Z¯
δ
= 0,
hence
∂h
β
¯
δ
∂z
α
= h(∇
Zα
Z
β
, Z¯
δ
) = Γ
γ
αβ
h
γ
¯
δ
.
This proves the formulas
(19) Γ
γ
αβ
h
γ
¯
δ
=
∂h
β
¯
δ
∂z
α
and Γ
γ
αβ
= h
γ
¯
δ
∂h
β
¯
δ
∂z
α
.
The curvature tensor can be viewed either as (3,1)– or as (4,0)–tensor. The corresponding coeﬃ
cients are deﬁned by
R(Z
A
, Z
B
)Z
C
= R
D
ABC
Z
D
and R
ABCD
= R(Z
A
, Z
B
, Z
C
, Z
D
) = h
DE
R
E
ABC
.
From the fact that T
1,0
M is parallel we immediately get R
γ
AB
¯
δ
= R
¯ γ
ABδ
= 0, hence R
ABγδ
=
R
AB¯ γ
¯
δ
= 0. Using the curvature symmetries we ﬁnally see that the only non–vanishing components
of R are
R
α
¯
βγ
¯
δ
, R
α
¯
β¯ γδ
, R
¯ αβγ
¯
δ
, R
¯ αβ¯ γδ
and
R
δ
α
¯
βγ
, R
¯
δ
α
¯
β¯ γ
, R
δ
¯ αβγ
, R
¯
δ
¯ αβ¯ γ
.
6. THE CURVATURE TENSOR OF K
¨
AHLER MANIFOLDS 35
From (18) and (19) we obtain
R
δ
α
¯
βγ
Z
δ
= R(Z
α
, Z¯
β
)Z
γ
= −∇
Z¯
β
(∇
Zα
Z
γ
) = −∇
Z¯
β
(Γ
δ
αγ
Z
δ
) = −
∂Γ
δ
αγ
∂¯ z
β
Z
δ
,
therefore
(20) R
δ
α
¯
βγ
= −
∂Γ
δ
αγ
∂¯ z
β
.
Using this formula we can compute the components of the Ricci tensor:
Ric
γ
¯
β
= Ric¯
βγ
= R
A
A
¯
βγ
= R
α
α
¯
βγ
= −
∂Γ
α
αγ
∂¯ z
β
.
Let us denote by d the determinant of the matrix (h
α
¯
β
). Using the Lemma 6.3 below and (19) we
get
Γ
α
αγ
= Γ
α
γα
= h
α
¯
δ
∂h
α
¯
δ
∂z
γ
=
1
d
∂d
∂z
γ
=
∂ log d
∂z
γ
.
This proves the following simple expressions for the Ricci tensor
Ric
α
¯
β
= −
∂
2
log d
∂z
α
∂¯ z
β
,
and for the Ricci form
(21) ρ = −i∂
¯
∂ log d.
Lemma 6.3. Let (h
ij
) = (h
ij
(t)) be the coeﬃcients of a map h : R → Gl
m
(C) with h
ij
:= (h
ij
)
−1
and let d(t) denote the determinant of (h
ij
). Then the following formula holds
d
(t) = d
m
¸
i,j=1
h
ij
(t)h
ji
(t).
Proof. Recall the deﬁnition of the determinant
d =
¸
σ∈Sm
ε(σ)h
1σ
1
. . . h
mσm
.
If we denote
˜
h
ji
:=
1
d
¸
σ∈Sm, σ
i
=j
ε(σ)h
1σ
1
. . . h
i−1σ
i−1
h
i+1σ
i+1
. . . h
mσm
,
then we obtain easily
m
¸
j=1
h
ij
˜
h
ji
=
1
d
m
¸
j=1
¸
σ∈Sm, σ
i
=j
ε(σ)h
1σ
1
. . . h
mσm
=
1
d
¸
σ∈Sm
ε(σ)h
1σ
1
. . . h
mσm
=
1
d
d = 1,
and
m
¸
j=1
h
kj
˜
h
ji
=
1
d
m
¸
j=1
¸
σ∈Sm, σ
i
=j
ε(σ)h
1σ
1
. . . h
i−1σ
i−1
h
kσ
i
h
i+1σ
i+1
. . . h
mσm
= 0
36
for k = i since in the last sum each term corresponding to a permutation σ is the opposite of the
term corresponding to the permutation (ik) ◦ σ, where (ik) denotes the transposition of i and k.
This shows that
˜
h
ji
= h
ji
are the coeﬃcients of the inverse matrix of h. We now get
d
(t) =
¸
σ∈Sm
m
¸
i=1
ε(σ)h
iσ
i
(t)h
1σ
1
. . . h
i−1σ
i−1
h
i+1σ
i+1
. . . h
mσm
=
m
¸
i=1
m
¸
j=1
¸
σ∈Sm, σ
i
=j
ε(σ)h
ij
(t)h
1σ
1
. . . h
i−1σ
i−1
h
i+1σ
i+1
. . . h
mσm
= d
m
¸
i,j=1
h
ij
(t)
˜
h
ji
= d
m
¸
i,j=1
h
ij
(t)h
ji
(t).
6.3. Exercises.
(1) Let S := Tr(Ric) denote the scalar curvature of a K¨ahler manifold M with Ricci form ρ.
Using the second Bianchi identity, prove the formula:
δρ = −
1
2
JdS.
In particular, the Ricci form of M is harmonic if and only if the scalar curvature S is
constant.
(2) Prove that the curvature of a K¨ahler manifold, viewed as a symmetric endomorphism
of the space of complex 2–forms, maps Λ
0,2
and Λ
2,0
to 0. Compute the image of the
fundamental form through this endomorphism.
(3) Let h be a Hermitian metric on some complex manifold M
2m
and let z
α
= x
α
+ iy
α
be a local system of holomorphic local coordinates on M. Using (17) show that the
determinant of the complex mm matrix (h
α
¯
β
) is a positive real number whose square
is equal to the determinant of the real 2m2m matrix h
ij
representing the metric in the
local coordinate system (x
i
, y
i
).
(4) Let h and h
be two K¨ahler metrics on some complex manifold (M, J) having the same
(Riemannian) volume form. Prove that the Ricci tensors of h and h
are equal.
7. EXAMPLES OF K
¨
AHLER METRICS 37
7. Examples of K¨ahler metrics
7.1. The ﬂat metric on C
m
. Its coeﬃcients in holomorphic coordinates are
h
α
¯
β
= h
∂
∂z
α
,
∂
∂¯ z
β
=
1
4
h
∂
∂x
α
−i
∂
∂y
α
,
∂
∂x
β
+ i
∂
∂y
β
=
1
2
δ
αβ
,
so by Lemma 5.2 the fundamental form is
Ω = i
1
2
m
¸
α=1
dz
α
∧ d¯ z
α
=
i
2
∂
¯
∂[z[
2
.
Thus u(z) =
1
2
[z[
2
is a K¨ahler potential for the canonical Hermitian metric on C
m
.
7.2. The Fubini–Study metric on the complex projective space CP
m
. Consider the
holomorphic atlas (U
j
, φ
j
) on CP
m
described in the ﬁrst section. Let π : C
m+1
−¦0¦ →CP
m
be
the canonical projection
π(z
0
, . . . , z
m
) = [z
0
: . . . : z
m
].
This map is clearly surjective. It is moreover a principal C
∗
–ﬁbration, with local trivializations
ψ
j
: π
−1
U
j
→U
j
C
∗
given by
ψ
j
(z) = ([z], z
j
),
and satisfying ψ
j
◦ ψ
−1
k
([z], α) = ([z],
z
j
z
k
α).
Consider the functions u : C
m
→R and v : C
m+1
−¦0¦ →R deﬁned by u(w) = log(1 +[w[
2
) and
v(z) = log([z[
2
). For every j ∈ ¦0, . . . , m¦, we deﬁne f
j
= φ
j
◦ π.
C
m+1
−¦0¦
f
j
π
GG
U
j
⊂ CP
m
φ
j
y
y
y
y
y
y
y
y
y
y
y
y
y
y
y
y
y
y
C
m
The map f
j
is clearly holomorphic and a direct calculation yields u ◦ f
j
(z) = v(z) −log [z
j
[
2
. As
∂
¯
∂ log [z
j
[
2
= 0, this shows that (f
j
)
∗
(∂
¯
∂u) = ∂
¯
∂v for every j. We thus can deﬁne a 2–form Ω on
CP
m
by
Ω[
U
j
:= i(φ
j
)
∗
(∂
¯
∂u),
which satisﬁes
(22) π
∗
(Ω) = i∂
¯
∂v.
Clearly Ω is a closed real (1, 1)–form, so the tensor h deﬁned by
h(X, Y ) := Ω(X, JY ), ∀ X, Y ∈ TCP
m
is symmetric and Hermitian. The next lemma proves that h deﬁnes a K¨ahler metric on CP
m
.
Lemma 7.1. The tensor h is positive deﬁnite on CP
m
.
38
Proof. Let us ﬁx some local holomorphic chart φ
j
: U
j
→ C
m
. Clearly, h = (φ
j
)
∗
(
ˆ
h), where
ˆ
h is the symmetric tensor on C
m
deﬁned by
ˆ
h(X, Y ) := i∂
¯
∂u(X, JY ), ∀ X, Y ∈ TC
m
. We have
to prove that
ˆ
h is positive deﬁnite. Now, since U
m
is a group of holomorphic transformations of
C
m
preserving u, it also preserves
ˆ
h. Moreover, it is transitive on the unit sphere of C
m
, so it is
enough to prove that
ˆ
h is positive deﬁnite at a point p = (r, 0, . . . , 0) ∈ C
m
for some positive real
number r. We have
∂
¯
∂ log(1+[z[
2
) = ∂(
1
1 +[z[
2
(
m
¸
i=1
z
i
d¯ z
i
)) =
1
1 +[z[
2
m
¸
i=1
dz
i
∧d¯ z
i
−
1
(1 +[z[
2
)
2
(
m
¸
i=1
¯ z
i
dz
i
)∧(
m
¸
i=1
z
i
d¯ z
i
).
At p this 2–form simpliﬁes to
1
(1 +r
2
)
2
dz
1
∧ d¯ z
1
+ (1 +r
2
)
m
¸
i=2
dz
i
∧ d¯ z
i
,
which shows that
ˆ
h
p
(X, Y ) =
2
(1 +r
2
)
2
Re(X
1
¯
Y
1
+ (1 +r
2
)
m
¸
i=2
X
i
¯
Y
i
)
hence
ˆ
h is positive deﬁnite.
The K¨ahler metric on CP
m
constructed in this way is called the Fubini–Study metric and is usually
denoted by h
FS
.
7.3. Geometrical properties of the Fubini–Study metric. The Fubini–Study metric
was deﬁned via its fundamental 2–form, which was expressed by local K¨ahler potentials. We
provide here a more geometrical description of this metric, showing that it is the projection to
CP
m
of some symmetric tensor ﬁeld of C
m+1
−¦0¦.
Lemma 7.2. For every z ∈ C
m+1
− ¦0¦, the canonical projection π : C
m+1
− ¦0¦ → CP
m
is a
submersion, and the kernel of its diﬀerential π
∗z
: T
z
(C
m+1
−¦0¦) →T
π(z)
CP
m
is the complex line
generated by z.
Proof. Let z ∈ C
m+1
with z
j
= 0. The composition f
j
:= φ
j
◦ π is given by
f
j
(z
0
, . . . , z
m
) =
1
z
j
(z
0
, . . . , z
j−1
, z
j+1
, . . . , z
m
).
We put j = 0 for simplicity and denote f = f
0
. Its diﬀerential at z applied to some tangent vector
v is
f
∗
z
(v) =
1
z
0
(v
1
, . . . , v
m
) −
v
0
z
2
0
(z
1
, . . . , z
m
).
Thus v ∈ ker(π
∗
)
z
⇐⇒ v ∈ ker(f
∗
)
z
⇐⇒ v =
v
0
z
0
z. This shows that ker((π
∗
)
z
) is the complex
line generated by z, and for dimensional reasons (π
∗
)
z
has to be surjective.
7. EXAMPLES OF K
¨
AHLER METRICS 39
Consider the complex orthogonal z
⊥
of z in C
m+1
with respect to the canonical Hermitian metric,
i.e. the set
z
⊥
:= ¦y ∈ C
m+1
[
m
¸
j=0
z
j
¯ y
j
= 0¦.
This deﬁnes a codimension 1 complex distribution D in C
m+1
−¦0¦ with D
z
:= z
⊥
. Let X →X
⊥
denote the orthogonal projection onto z
⊥
in T
z
(C
m+1
−¦0¦) and deﬁne a bilinear symmetric tensor
˜
h on C
m+1
−¦0¦ by
˜
h(X, Y ) :=
2
[z[
2
'X
⊥
, Y
⊥
`,
where ', ` denotes the canonical Hermitian product.
Lemma 7.3. The (1,1)–form ϕ(X, Y ) :=
˜
h(JX, Y ) associated to the tensor
˜
h satisﬁes ϕ =
i∂
¯
∂ log([z[
2
) on C
m+1
−¦0¦.
Proof. It is enough to prove this relation at a point p = (r, 0, . . . , 0) ∈ C
m+1
− ¦0¦ for
some positive real number r because both members are U
m+1
–invariant and U
m+1
is transitive on
spheres. We have
∂
¯
∂ log([z[
2
) = ∂(
1
[z[
2
(
m
¸
i=0
z
i
d¯ z
i
)) =
1
[z[
2
m
¸
i=0
dz
i
∧ d¯ z
i
−
1
[z[
4
(
m
¸
i=0
¯ z
i
dz
i
) ∧ (
m
¸
i=0
z
i
d¯ z
i
).
At p this 2–form simpliﬁes to
1
r
2
m
¸
i=1
dz
i
∧ d¯ z
i
.
On the other hand, we have at p
−iϕ
∂
∂z
α
,
∂
∂¯ z
β
= −i
˜
h
i
∂
∂z
α
,
∂
∂¯ z
β
=
˜
h
∂
∂z
α
,
∂
∂¯ z
β
which vanishes if α = 0 or β = 0 and equals
1
r
2
δ
αβ
otherwise. Thus
−iϕ =
1
r
2
m
¸
i=1
dz
i
∧ d¯ z
i
.
Using this lemma and (22) we see that π
∗
h =
˜
h, showing that the Fubini–Study metric h
FS
on
CP
m
is given by the projection of the above deﬁned semi–positive symmetric tensor ﬁeld
˜
h.
Proposition 7.4. The group U
m+1
acts transitively by holomorphic isometries on (CP
m
, h
FS
)
Proof. For every A ∈ U
m+1
, z ∈ C
m+1
−¦0¦ and α ∈ C
∗
, we have A(αz) = αA(z), showing
that the canonical action of U
m+1
on C
m+1
− ¦0¦ descends to an action on CP
m
. For every
A ∈ U
m+1
, let
˜
A be the corresponding transformation of CP
m
. Looking at its expression in
the canonical holomorphic charts, it is easy to check that every
˜
A acts holomorphically on CP
m
.
40
In order to check that
˜
A preserves the Fubini–Study metric, we ﬁrst use (22) and the relation
v ◦ A(z) = log [Az[
2
= log [z[
2
= v(z) to get
π
∗
(
˜
A
∗
(Ω)) = A
∗
(i∂
¯
∂v) = i∂
¯
∂A
∗
v = i∂
¯
∂v = π
∗
Ω.
Lemma 7.2 shows that π
∗
is onto, so π
∗
is injective on exterior forms, hence
˜
A
∗
(Ω) = Ω. As
˜
A
also preserves the complex structure, too, this clearly implies that
˜
A is an isometry.
We will now use our computations in local coordinates from the previous section in order to show
that the Fubini–Study metric is Einstein. Since there exists a transitive isometric action on CP
m
,
it is enough to check this at some point, say p := [1 : 0 : . . . : 0] ∈ CP
m
. From Lemma 7.1 we see
that the fundamental form is given in the local chart φ
0
by
Ω =
i
1 +[z[
2
m
¸
i=1
dz
i
∧ d¯ z
i
−
i
(1 +[z[
2
)
2
(
m
¸
i=1
¯ z
i
dz
i
) ∧ (
m
¸
i=1
z
i
d¯ z
i
).
Lemma 7.5. Let dv denote the volume form on C
m
dv := dx
1
∧ dy
1
∧ . . . ∧ dx
m
∧ dy
m
=
i
2
dz
1
∧ d¯ z
1
∧ . . . ∧
i
2
dz
m
∧ d¯ z
m
.
Then the fundamental 2–form Ω satisﬁes
Ω ∧ . . . ∧ Ω =
2
m
m!
(1 +[z[
2
)
m+1
dv.
Proof. Both terms are clearly invariant by the action of U
m
on C
m
, which is transitive on
spheres, so it is enough to prove the equality at points of the form z = (r, 0, . . . , 0), where it is
actually obvious.
Now, for every Hermitian metric h on C
m
with fundamental form ϕ, the determinant d of the
matrix (h
α
¯
β
) satisﬁes
1
m!
ϕ
m
= d2
m
dv.
Applying this to our situation and using the lemma above yields
d = det(h
α
¯
β
) =
1
(1 +[z[
2
)
m+1
,
whence log d = −(m+ 1) log(1 +[z[
2
), so from the local formula (21) for the Ricci form we get
ρ = −i∂
¯
∂ log d = (m+ 1)i∂
¯
∂ log(1 +[z[
2
) = (m+ 1)Ω,
thus proving that the Fubini–Study metric on CP
m
is an Einstein metric, with Einstein constant
m+ 1.
7. EXAMPLES OF K
¨
AHLER METRICS 41
7.4. Exercises.
(1) A submersion f : (M, g) → (N, h) between Riemannian manifolds is called Riemannian
submersion if for every x ∈ M, the restriction of (f
∗
)
x
to the g–orthogonal of the tangent
space to the ﬁber f
−1
(f(x)) is an isometry onto T
f(x)
N. Prove that the restriction of the
canonical projection π to S
2m+1
deﬁnes a Riemannian submersion onto (CP
m
,
1
2
h
FS
).
(2) Show that (CP
1
, h
FS
) is isometric to the round sphere of radius 1/
√
2, S
2
(1/
√
2) ⊂ R
3
.
Hint: Use the fact that a simply–connected manifold with constant positive sectional
curvature K is isometric to the sphere of radius 1/
√
K.
(3) Show that for every Hermitian tensor h on C
m
with fundamental form ϕ, the determinant
d of the matrix (h
α
¯
β
) satisﬁes
ϕ
m
= d2
m
m!dv.
Part 3
The Laplace operator
8. Natural operators on Riemannian and K¨ahler manifolds
8.1. The formal adjoint of a linear diﬀerential operator. Let (M
n
, g) be an oriented
Riemannian manifold (not necessarily compact) with volume form dv and let E and F be Her
mitian vector bundles over M with Hermitian structures denoted by ', `
E
and ', `
F
.
Definition 8.1. Let P : (
∞
(E) → (
∞
(F) and Q : (
∞
(F) → (
∞
(E) be linear diﬀerential
operators. The operator Q is called a formal adjoint of P if
M
'Pα, β`
F
dv =
M
'α, Qβ`
E
dv,
for every compactly supported smooth sections α ∈ (
∞
0
(E) and β ∈ (
∞
0
(F).
Lemma 8.2. There exists at most one formal adjoint for every linear diﬀerential operator.
Proof. Suppose that P : (
∞
(E) →(
∞
(F) has two formal adjoints, denoted Q and Q
. Then
their diﬀerence R := Q−Q
satisﬁes
M
'α, Rβ`
E
dv = 0 ∀ α ∈ (
∞
0
(E), ∀ β ∈ (
∞
0
(F).
Suppose that there exists some σ ∈ (
∞
(F) and some x ∈ M such that R(σ)
x
= 0. Take a positive
function f on M such that f ≡ 1 on some open set U containing x and f = 0 outside a compact
set. Since R is a diﬀerential operator, the value of R(σ) at x only depends on the germ of σ at
x, so in particular R(fσ) has compact support and R(fσ)
x
= R(σ)
x
= 0. Applying the formula
above to the compactly supported sections α := R(fσ) and β := fσ of E and F we get
0 =
M
'α, Rβ`
E
dv =
M
[R(fσ)[
2
dv,
thus showing that the smooth positive function [R(fσ)[
2
has to vanish identically on M, contra
dicting the fact that its value at x is non–zero.
The formal adjoint of an operator P is usually denoted by P
∗
. From the above lemma it is
immediate to check that P is the formal adjoint of P
∗
and that Q
∗
◦ P
∗
is the formal adjoint of
P ◦ Q. The lemma below gives a useful method to compute the formal adjoint:
Lemma 8.3. Let P : (
∞
(E) → (
∞
(F) and Q : (
∞
(F) → (
∞
(E) be linear diﬀerential operators.
If there exists a section ω ∈ (
∞
(E
∗
⊗F
∗
⊗Λ
n−1
M) such that
(23) ('Pα, β`
F
−'α, Qβ`
E
)dv = d(ω(α, β)), ∀ α ∈ (
∞
(E), β ∈ (
∞
(F),
then Q is the formal adjoint of P.
Proof. The n−1–form ω(α, β) has compact support for every compactly supported sections
α and β. By Stokes’ theorem we see that the integral over M of its exterior derivative vanishes.
8. NATURAL OPERATORS ON RIEMANNIAN AND K
¨
AHLER MANIFOLDS 45
8.2. The Laplace operator on Riemannian manifolds. We start with an oriented Rie
mannian manifold (M
n
, g) with volume form dv. We denote generically by ¦e
1
, . . . , e
n
¦ a local
orthonormal frame on M parallel in a point and identify vectors and 1–forms via the metric g. In
this way we can write for instance dv = e
1
∧ . . . ∧ e
n
.
There is a natural embedding ϕ of Λ
k
M in TM
⊗k
given by
ϕ(ω)(X
1
, . . . , X
k
) := ω(X
1
, . . . , X
k
),
which in the above local basis reads
ϕ(e
1
∧ . . . ∧ e
k
) =
¸
σ∈S
k
ε(σ)e
σ
1
⊗. . . ⊗e
σ
k
.
The Riemannian product g induces a Riemannian product on all tensor bundles. We consider the
following weighted tensor product on Λ
k
M:
'ω, τ` :=
1
k!
g(ϕ(ω), ϕ(τ)),
which can also be characterized by the fact that the basis
¦e
i
1
∧ . . . ∧ e
i
k
[ 1 ≤ i
1
< . . . < i
k
≤ n¦
is orthonormal. With respect to this scalar product, the interior and exterior products are adjoint
operators:
(24) 'X ω, τ` = 'ω, X ∧ τ`, ∀ X ∈ TM, ω ∈ Λ
k
M, τ ∈ Λ
k−1
M.
We deﬁne the Hodge *–operator ∗ : Λ
k
M →Λ
n−k
M by
ω ∧ ∗τ := 'ω, τ`dv, ∀ω, τ ∈ Λ
k
M.
It is well–known and easy to check on the local basis above that the following relations are satisﬁed:
(25) ∗1 = dv, ∗dv = 1,
(26) '∗ω, ∗τ` = 'ω, τ`,
(27) ∗
2
= (−1)
k(n−k)
on Λ
k
M.
The exterior derivative d : (
∞
(Λ
k
M) →(
∞
(Λ
k+1
M)
d =
¸
i
e
i
∧ ∇
e
i
has a formal adjoint δ : (
∞
(Λ
k+1
M) →(
∞
(Λ
k
M) satisfying
δ = −(−1)
nk
∗ d∗ = −
¸
i
e
i
∇
e
i
.
To see this, let α ∈ Ω
p
and β ∈ Ω
p+1
be smooth forms. Then we have
'dα, β`dv = dα ∧ ∗β = d(α ∧ ∗β) −(−1)
p
α ∧ d ∗ β
= d(α ∧ ∗β) −(−1)
p+p(n−p)
α ∧ ∗ ∗ d ∗ β = d(α ∧ ∗β) −(−1)
np
'α, ∗d ∗ β`dv,
46
so Lemma 8.3 shows that d
∗
= (−1)
np+1
∗ d∗ on p + 1–forms.
Using the Hodge *–operator we get the following useful reformulation of Lemma 8.3: if there
exists a section τ ∈ (
∞
(E
∗
⊗F
∗
⊗Λ
1
M) such that
(28) 'Pα, β`
F
−'α, Qβ`
E
= δ(τ(α, β)), ∀ α ∈ (
∞
(E), β ∈ (
∞
(F),
then Q is the formal adjoint of P.
The Laplace operator ∆ : (
∞
(Λ
k
M) →(
∞
(Λ
k
M) is deﬁned by
∆ := dδ + δd,
and is clearly formally self–adjoint.
8.3. The Laplace operator on K¨ahler manifolds. After these preliminaries, let now
(M
2m
, h, J) be an almost Hermitian manifold with fundamental form Ω. We deﬁne the following
algebraic (real) operators acting on diﬀerential forms:
L : Λ
k
M →Λ
k+2
M, L(ω) := Ω ∧ ω =
1
2
¸
i
e
i
∧ Je
i
∧ ω,
with adjoint Λ satisfying
Λ : Λ
k+2
M →Λ
k
M, Λ(ω) :=
1
2
¸
i
Je
i
e
i
ω.
These natural operators can be extended to complex–valued forms by C–linearity.
Lemma 8.4. The following relations hold:
(1) The Hodge *–operator maps (p, q)–forms to (m−q, m−p)–forms.
(2) [X , Λ] = 0 and [X , L] = JX ∧ .
The proof is straightforward.
Let us now assume that M is K¨ahler. We deﬁne the twisted diﬀerential d
c
: (
∞
(Λ
k
M) →
(
∞
(Λ
k+1
M) by
d
c
(ω) :=
¸
i
Je
i
∧ ∇
e
i
ω
whose formal adjoint is δ
c
: (
∞
(Λ
k+1
M) →(
∞
(Λ
k
M)
δ
c
:= −∗ d
c
∗ = −
¸
i
Je
i
∇
e
i
.
Lemma 8.5. On a K¨ahler manifold, the following relations hold:
(29) [L, δ] = d
c
, [L, d] = 0
and
(30) [Λ, d] = −δ
c
, [Λ, δ] = 0.
8. NATURAL OPERATORS ON RIEMANNIAN AND K
¨
AHLER MANIFOLDS 47
Proof. Using Lemma 8.4 (2) and the fact that J and Ω are parallel we get
[L, δ] = −[L, e
i
∇
e
i
] = −[L, e
i
]∇
e
i
= Je
i
∧ ∇
e
i
= d
c
.
The second relation in (29) just expresses the fact that the K¨ahler form is closed. The two relations
in (30) follow by the uniqueness of the formal adjoint.
Corresponding to the decomposition d = ∂ +
¯
∂ we have the decomposition δ = ∂
∗
+
¯
∂
∗
, where
∂
∗
: (
∞
(Λ
p,q
M) →(
∞
(Λ
p−1,q
M), ∂
∗
:= −∗
¯
∂∗
and
¯
∂
∗
: (
∞
(Λ
p,q
M) →(
∞
(Λ
p,q−1
M),
¯
∂
∗
:= −∗ ∂ ∗ .
Notice that ∂
∗
and
¯
∂
∗
are formal adjoints of ∂ and
¯
∂ with respect to the Hermitian product H on
complex forms given by
(31) H(ω, τ) := 'ω, ¯ τ`.
We deﬁne the Laplace operators
∆
∂
:= ∂∂
∗
+ ∂
∗
∂ and ∆
¯
∂
:=
¯
∂
¯
∂
∗
+
¯
∂
∗
¯
∂.
One of the most important features of K¨ahler metrics is that these new Laplace operators are
essentially the same as the usual one:
Theorem 8.6. On any K¨ahler manifold one has ∆ = 2∆
∂
= 2∆
¯
∂
.
Proof. Our identiﬁcation of TM and T
∗
M via the metric maps (1, 0)–vectors to (0, 1)–forms
and vice–versa. From the fact that Λ
p,q
M are preserved by the covariant derivative follows easily
∂ =
¸
j
1
2
(e
j
+ iJe
j
) ∧ ∇
e
j
and
¯
∂ =
¸
j
1
2
(e
j
−iJe
j
) ∧ ∇
e
j
.
From the deﬁnition of d
c
we then get
(32) d
c
= i(
¯
∂ −∂),
and by adjunction
(33) δ
c
= i(∂
∗
−
¯
∂
∗
).
Applying (29) to a (p, q)–form and projecting onto Λ
p±1,q
M and Λ
p,q±1
M then yields
(34) [L, ∂
∗
] = i
¯
∂, [L,
¯
∂
∗
] = −i∂, [L, ∂] = 0, [L,
¯
∂] = 0,
and similarly from (30) we obtain
(35) [Λ, ∂] = i
¯
∂
∗
, [Λ,
¯
∂] = −i∂
∗
, [Λ, ∂
∗
] = 0, [Λ,
¯
∂
∗
] = 0.
Now, the relation
¯
∂
2
= 0 together with (35) gives
−i(
¯
∂∂
∗
+ ∂
∗
¯
∂) =
¯
∂[Λ,
¯
∂] + [Λ,
¯
∂]
¯
∂ =
¯
∂Λ
¯
∂ −
¯
∂Λ
¯
∂ = 0
and similarly
∂
¯
∂
∗
+
¯
∂
∗
∂ = 0.
48
Thus,
∆ = (∂ +
¯
∂)(∂
∗
+
¯
∂
∗
) + (∂
∗
+
¯
∂
∗
)(∂ +
¯
∂)
= (∂∂
∗
+ ∂
∗
∂) + (
¯
∂
¯
∂
∗
+
¯
∂
∗
¯
∂) + (
¯
∂∂
∗
+ ∂
∗
¯
∂) + (∂
¯
∂
∗
+
¯
∂
∗
∂)
= ∆
∂
+ ∆
¯
∂
.
It remains to show the equality ∆
∂
= ∆
¯
∂
, which is an easy consequence of (35):
−i∆
∂
= −i(∂∂
∗
+ ∂
∗
∂) = ∂[Λ,
¯
∂] + [Λ,
¯
∂]∂ = ∂Λ
¯
∂ −∂
¯
∂Λ + Λ
¯
∂∂ −
¯
∂Λ∂
= ∂Λ
¯
∂ +
¯
∂∂Λ −Λ∂
¯
∂ −
¯
∂Λ∂ = [∂, Λ]
¯
∂ +
¯
∂[∂, Λ] = −i
¯
∂
∗
¯
∂ −i
¯
∂
¯
∂
∗
= −i∆
¯
∂
.
8.4. Exercises.
(1) Consider the extension of J as derivation
J : Λ
k
M →Λ
k
M, J(ω) :=
¸
i
Je
i
∧ e
i
ω.
Show that the following relations hold:
• J is skew–Hermitian.
• J(α ∧ β) = J(α) ∧ β + α ∧ J(β) for all forms α ∈ Ω
p
M and β ∈ Ω
k
M.
• The restriction of J to Λ
p,q
M equals to the scalar multiplication by i(q −p).
• [J, Λ] = 0 and [J, L] = 0.
(2) Let ω be a k–form on a n–dimensional Riemannian manifold M. Prove that e
i
∧(e
i
ω) =
kω and e
i
(e
i
∧ ω) = (n −k)ω.
(3) Show that 0 = dd
c
+ d
c
d = dδ
c
+ δ
c
d = δδ
c
+ δ
c
δ = δd
c
+ d
c
δ on every K¨ahler manifold.
(4) Prove that [J, d] = d
c
and [J, d
c
] = −d on K¨ahler manifolds.
(5) Show that the Laplace operator commutes with L, Λ and J on K¨ahler manifolds.
9. HODGE AND DOLBEAULT THEORY 49
9. Hodge and Dolbeault theory
9.1. Hodge theory. In this section we assume that (M
n
, g) is a compact oriented Rie
mannian manifold. From now on we denote the space of smooth complex–valued k–forms by
Ω
k
M := (
∞
(Λ
k
M ⊗ C) and by Z
k
(M) the space of closed complex k–forms on M. Since the
exterior derivative satisﬁes d
2
= 0, one clearly has dΩ
k−1
M ⊂ Z
k
M. We deﬁne the De Rham
cohomology groups by
H
k
DR
(M, C) :=
Z
k
M
dΩ
k−1
M
.
The De Rham theorem says that the k
th
cohomology group of M with complex coeﬃcients is
naturally isomorphic to the k
th
De Rham cohomology group:
H
k
(M, C) · H
k
DR
(M, C).
We now denote by H
k
(M, C) the space of complex harmonic k–forms on M, i.e. forms in the
kernel of the Laplace operator:
H
k
(M, C) := ¦ω ∈ Ω
k
M [ ∆ω = 0¦.
Lemma 9.1. A form is harmonic if and only if it is closed and δ–closed.
Proof. One direction is clear. Suppose conversely that ω is harmonic. Since M is compact
and d and δ are formally adjoint operators we get
0 =
M
H(∆ω, ω)dv =
M
H(dδω + δdω, ω)dv =
M
[δω[
2
+[dω[
2
dv,
showing that dω = 0 and δω = 0.
Theorem 9.2. (Hodge decomposition theorem). The space of k–forms decomposes as a direct
sum
Ω
k
M = H
k
(M, C) ⊕δΩ
k+1
M ⊕dΩ
k−1
M.
Proof. Using Lemma 9.1 it is immediate to check that the three spaces above are orthogonal
with respect to the global Hermitian product on Ω
k
M given by
(ω, τ) :=
M
H(ω, τ)dv.
The hard part of the theorem is to show that the direct sum of these three summands is the whole
space Ω
k
M. A proof can be found in [2], pp. 84–100.
The Hodge decomposition theorem shows that every k–form ω on M can be uniquely written as
ω = dω
+ δω
+ ω
H
,
where ω
∈ Ω
k−1
M, ω
∈ Ω
k+1
M and ω
H
∈ H
k
(M, C). If ω is closed, we can write
0 = (dω, ω
) = (dδω
, ω
) =
M
[δω
[
2
dv,
showing that the second term in the Hodge decomposition of ω vanishes.
50
Proposition 9.3. (Hodge isomorphism). The natural map f : H
k
(M, C) → H
k
DR
(M, C) given
by ω →[ω] is an isomorphism.
Proof. First, f is well–deﬁned because every harmonic form is closed (Lemma 9.1). The
kernel of f is zero since the spaces of harmonic forms and exact forms are orthogonal, so in
particular their intersection is ¦0¦. Finally, for every De Rham cohomology class c, there exists a
closed form ω such that [ω] = c. We have seen that the Hodge decomposition of ω is ω = dω
+ω
H
,
showing that
f(ω
H
) = [ω
H
] = [dω
+ ω
H
] = [ω] = c,
hence f is surjective.
The complex dimension b
k
(M) := dim
C
(H
k
DR
(M, C)) is called the k
th
Betti number of M and is
a topological invariant in view of De Rham’s theorem.
Proposition 9.4. (Poincar´e duality). The spaces H
k
(M, C) and H
n−k
(M, C) are isomorphic. In
particular b
k
(M) = b
n−k
(M) for every compact n–dimensional manifold M.
Proof. The isomorphism is simply given by the Hodge *–operator which sends harmonic
k–forms to harmonic n −k–forms.
We close this section with the following interesting application of Theorem 9.2.
Proposition 9.5. Every Killing vector ﬁeld on a compact K¨ahler manifold is real holomorphic.
Proof. Let X be a Killing vector ﬁeld, that is L
X
g = 0. We compute the Lie derivative of
the fundamental 2–form with respect to X using Cartan’s formula:
L
X
Ω = d(X Ω) + X dΩ = d(X Ω),
so L
X
Ω is exact. On the other hand, since the ﬂow of X is isometric, it commutes with the
Hodge *–operator, thus L
X
◦ ∗ = ∗ ◦ L
X
. As we clearly have d ◦ L
X
= L
X
◦ d, too, we see that
L
X
◦ δ = δ ◦ L
X
, whence
d(L
X
Ω) = L
X
(dΩ) = 0
and
δ(L
X
Ω) = L
X
(δΩ) = 0,
because Ω is coclosed, being parallel. Thus L
X
Ω is harmonic and exact, so it has to vanish by the
easy part of Theorem 9.2. This shows that the ﬂow of X preserves the metric and the fundamental
2–form, it thus preserves the complex structure J, hence X is real holomorphic.
9. HODGE AND DOLBEAULT THEORY 51
9.2. Dolbeault theory. Let (M
2m
, h, J) be a compact Hermitian manifold. We consider the
Dolbeault operator
¯
∂ acting on the spaces of (p, q)–forms Ω
p,q
M := (
∞
(Λ
p,q
M) ⊂ Ω
p+q
M. Let
Z
p,q
M denote the space of
¯
∂–closed (p, q)–forms. Since
¯
∂
2
= 0, we see that
¯
∂Ω
p,q−1
M ⊂ Z
p,q
M.
We deﬁne the Dolbeault cohomology groups
H
p,q
M :=
Z
p,q
M
¯
∂Ω
p,q−1
M
.
In contrast to De Rham cohomology, the Dolbeault cohomology is no longer a topological invariant
of the manifold, since it strongly depends on the complex structure J.
We deﬁne the space H
p,q
M of
¯
∂–harmonic (p, q)–forms on M by
H
p,q
M := ¦ω ∈ Ω
p,q
M [ ∆
¯
∂
ω = 0¦.
As before we have
Lemma 9.6. A form ω ∈ Ω
p,q
M is
¯
∂–harmonic if and only if
¯
∂ω = 0 and
¯
∂
∗
ω = 0.
The proof is very similar to that of Lemma 9.1 and is left as an exercise.
Theorem 9.7. (Dolbeault decomposition theorem). The space of (p, q)–forms decomposes as a
direct sum
Ω
p,q
M = H
p,q
M ⊕
¯
∂
∗
Ω
p,q+1
M ⊕
¯
∂Ω
p,q−1
M.
Proof. Lemma 9.6 shows that the three spaces above are orthogonal with respect to the
global Hermitian product
(, ) :=
M
H(, )dv
on Ω
p,q
M, and a proof for the hard part, which consists in showing that the direct sum of the
three summands is the whole space Ω
p,q
M, can be found in [2], pp. 84–100.
This shows that every (p, q)–form ω on M can be uniquely written as
ω =
¯
∂ω
+
¯
∂
∗
ω
+ ω
H
,
where ω
∈ Ω
p,q−1
M, ω
∈ Ω
p,q+1
M and ω
H
∈ H
p,q
M. This is called the Dolbeault decomposition
of ω. As before,the second summand in the Dolbeault decomposition of ω vanishes if and only if
¯
∂ω = 0. Specializing for q = 0 yields
Proposition 9.8. A (p, 0)–form on a compact Hermitian manifold is holomorphic if and only if
it is
¯
∂–harmonic.
Corollary 9.9. (Dolbeault isomorphism). The map f : H
p,q
M → H
p,q
M given by ω → [ω] is
an isomorphism.
The proof is completely similar to the proof of the Hodge isomorphism.
We denote by h
p,q
the complex dimension of H
p,q
M. These are the Hodge numbers associated to
the complex structure J of M.
52
Proposition 9.10. (Serre duality). The spaces H
p,q
M and H
m−p,m−q
M are isomorphic. In
particular h
p,q
= h
m−p,m−q
.
Proof. Consider the composition of the Hodge *–operator with the complex conjugation
¯ ∗ : Ω
p,q
M →Ω
m−p,m−q
M, ¯ ∗(ω) := ∗¯ ω.
We have
¯∗∆
¯
∂
(ω) = ∗(
¯
∂
¯
∂
∗
+
¯
∂
∗ ¯
∂)ω = ∗(∂∂
∗
+ ∂
∗
∂)¯ ω
= −∗ (∂ ∗
¯
∂ ∗ +∗
¯
∂ ∗ ∂)¯ ω =
¯
∂
∗
¯
∂(¯ ∗ω) −∗
2
¯
∂ ∗ ∂¯ ω
=
¯
∂
∗
¯
∂(¯ ∗ω) −
¯
∂ ∗ ∂ ∗
2
¯ ω =
¯
∂
∗
¯
∂(¯ ∗ω) +
¯
∂
¯
∂
∗
(¯ ∗ω) = ∆
¯
∂
(¯ ∗ω).
This shows that ¯∗ is a (C–anti–linear) isomorphism from H
p,q
M to H
m−p,m−q
M.
If M is K¨ahler much more can be said about Hodge and Betti numbers, due to Theorem 8.6.
Firstly, the fact that ∆ = 2∆
¯
∂
shows that H
p,q
M ⊂ H
p+q
M. Secondly, since ∆
¯
∂
leaves the spaces
Ω
p,q
M invariant, we deduce that ∆ has the same property, thus proving that the components of
a harmonic form in its type decomposition are all harmonic. This shows that
H
k
M = ⊕
p+q=k
H
p,q
M.
Moreover, as ∆
¯
∂
is a real operator, it commutes with the complex conjugation (in the general case
we only have that ∆
¯
∂
α = ∆
∂
¯ α) so the complex conjugation deﬁnes an isomorphism between the
spaces H
p,q
M and H
q,p
M. Consider now the fundamental form Ω ∈ Ω
1,1
M. Since Ω
m
is a non–
zero multiple of the volume form, we deduce that all exterior powers Ω
p
∈ Ω
p,p
M are non–zero.
Moreover, they are all harmonic since Ω is parallel so Ω
p
is parallel, too, and a parallel form is
automatically harmonic. We thus have proved the
Proposition 9.11. In addition to Poincar´e and Serre dualities, the following relations hold be
tween Betti and Hodge numbers on compact K¨ahler manifolds:
(36) b
k
=
¸
p+q=k
h
p,q
, h
p,q
= h
q,p
, h
p,p
≥ 1 ∀ 0 ≤ p ≤ m.
In particular (36) shows that all Betti numbers of odd order are even and all Betti numbers of
even order are non–zero.
9.3. Exercises.
(1) Prove that the complex manifold S
1
S
2k+1
carries no K¨ahler metric for k ≥ 1.
(2) Let V be an Euclidean vector space, identiﬁed with V
∗
via the metric. Prove that the
Lie algebra extension of an endomorphism A of V to Λ
k
V is given by the formula
A(ω) := A(e
i
) ∧ e
i
ω,
for every orthonormal basis ¦e
i
¦ of V .
9. HODGE AND DOLBEAULT THEORY 53
(3) The global i∂
¯
∂–Lemma. Let ϕ be an exact real (1, 1)–form on a compact K¨ahler manifold
M. Prove that ϕ is i∂
¯
∂–exact, in the sense that there exists a real function u such that
ϕ = i∂
¯
∂u.
(4) Show that there exists no global K¨ahler potential on a compact K¨ahler manifold.
(5) Let (M, h) be a compact K¨ahler manifold whose second Betti number is equal to 1. Show
that if the scalar curvature of M is constant, then the metric h is Einstein.
Part 4
Prescribing the Ricci tensor on K¨ahler manifolds
10. Chern classes
10.1. ChernWeil theory. The comprehensive theory of Chern classes can be found in [8],
Ch.12. We will outline here the deﬁnition and properties of the ﬁrst Chern class, which is the
only one needed in the sequel. The following proposition can be taken as a deﬁnition
Proposition 10.1. To every complex vector bundle E over a smooth manifold M one can asso
ciate a cohomology class c
1
(E) ∈ H
2
(M, Z) called the ﬁrst Chern class of E satisfying the following
axioms:
• (naturality) For every smooth map f : M → N and complex vector bundle E over N,
one has f
∗
(c
1
(E)) = c
1
(f
∗
E), where the left term denotes the pull–back in cohomology
and f
∗
E is the pull–back bundle deﬁned by f
∗
E
x
= E
f(x)
∀ x ∈ M.
• (Whitney sum formula) For every bundles E, F over M one has c
1
(E ⊕ F) = c
1
(E) +
c
1
(F), where E ⊕F is the Whitney sum deﬁned as the pull–back of the bundle E F →
M M by the diagonal inclusion of M in M M.
• (normalization) The ﬁrst Chern class of the tautological bundle of CP
1
is equal to −1 in
H
2
(CP
1
, Z) · Z, which means that the integral over CP
1
of any representative of this
class equals −1.
Let E → M be a complex vector bundle. We will now explain the Chern–Weil theory allows
one to express the images in real cohomology of the Chern classes of E using the curvature of an
arbitrary connection ∇ on E. Recall the formula (9) for the curvature of ∇ in terms of ∇:
(37) R
∇
(σ
i
) =: R
∇
ij
σ
j
= (dω
ij
−ω
ik
∧ ω
kj
)σ
j
,
where ¦σ
1
, . . . , σ
k
¦ are local sections of E which form a basis of each ﬁber over some open set U
and the connection forms ω
ij
∈ Λ
1
(U) (relative to the choice of this basis) are deﬁned by
∇σ
i
= ω
ij
⊗σ
j
.
Notice that although the coeﬃcients R
∇
ij
of R
∇
depend on the local basis of sections (σ
i
), its
trace is a well–deﬁned (complex–valued) 2–form on M independent of the chosen basis, and can
be computed as Tr(R
∇
) =
¸
R
∇
ii
in the local basis (σ
i
). To compute this explicitly we use the
following summation trick:
¸
i,k
ω
ik
∧ ω
ki
=
¸
k,i
ω
ki
∧ ω
ik
= −
¸
i,k
ω
ik
∧ ω
ki
,
where the ﬁrst equality is given by interchanging the summation indices and the second by the
fact that the wedge product is skew–symmetric on 1–forms. From (37) we thus get
(38) Tr(R
∇
) = d(
¸
ω
ii
),
where of course the trace of the connection form ω = (ω
ij
) does depend on the local basis (σ
i
).
This shows that Tr(R
∇
) is closed, being locally exact.
10. CHERN CLASSES 57
If ∇ and
˜
∇ are connections on E, the Leibniz rule shows that their diﬀerence A :=
˜
∇ − ∇ is
a zero–order operator, more precisely a smooth section in Λ
1
(M) ⊗ End(E). Thus Tr(A) is a
well–deﬁned 1–form on M and (38) readily implies
(39) Tr(R
˜
∇
) = Tr(R
∇
) + dTr(A),
We thus have proved the following
Lemma 10.2. The 2–form Tr(R
∇
) is closed and its cohomology class [Tr(R
∇
)] ∈ H
2
(M, C) does
not depend on ∇.
It is actually easy to see that [Tr(R
∇
)] is a purely imaginary class, in the sense that it has a
representative which is a purely imaginary 2–form. Indeed, let us choose an arbitrary Hermitian
structure h on E and take ∇ such that h is ∇–parallel. If we start with a local basis ¦σ
i
¦ adapted
to h, then we have
0 = ∇(δ
ij
) = ∇(h(σ
i
, σ
j
)) = h(∇σ
i
, σ
j
) + h(σ
i
, ∇σ
j
)
= ω
ij
+ ω
ji
.
From (37) we get
R
∇
ij
= dω
ij
−ω
ik
∧ ω
kj
= −ω
ji
−ω
ki
∧ ω
jk
= −ω
ji
+ ω
jk
∧ ω
ki
= −R
∇
ji
,
showing that the trace of R
∇
is a purely imaginary 2–form.
Theorem 10.3. Let ∇ be a connection on a complex bundle E over M. The real cohomology
class
c
1
(∇) :=
¸
i
2π
Tr(R
∇
)
is equal to the image of c
1
(E) in H
2
(M, R).
Proof. We have to check that c
1
(∇) satisﬁes the three conditions in Proposition 10.1. The
naturality is straightforward. Recall that if f : M → N is smooth and π : E → N is a rank k
vector bundle, then
f
∗
(E) := ¦(x, v) [ x ∈ M, v ∈ E, f(x) = π(v)¦.
If ¦σ
i
¦ is a local basis of sections of E, then
f
∗
σ
i
: M →f
∗
(E), x →(x, σ
i
(f(x)))
is a basis of local sections of f
∗
E. The formula
f
∗
∇(f
∗
σ) := f
∗
(∇σ)
deﬁnes a connection on f
∗
E (one has to check the classical formulas for basis changes in order to
prove that f
∗
∇ is well–deﬁned), and with respect to this basis we obviously have
R
f
∗
∇
ij
= f
∗
(R
∇
ij
),
whence c
1
(f
∗
∇) = f
∗
(c
1
(∇)).
58
The Whitney sum formula is also easy to check. If E and F are complex bundles over M with
connections ∇ and
˜
∇ then one can deﬁne a connection ∇⊕
˜
∇ on E ⊕F by
(∇⊕
˜
∇)(σ ⊕ ˜ σ)(X) := ∇σ(X) ⊕
˜
∇˜ σ(X).
If ¦σ
i
¦, ¦˜ σ
j
¦ are local basis of sections of E and F then ¦σ
i
⊕0, 0 ⊕ ˜ σ
j
¦ is a local basis for E ⊕F
and the curvature of ∇ ⊕
˜
∇ in this basis is a block matrix having R
∇
and R
˜
∇
on the principal
diagonal. Its trace is thus the sum of the traces of R
∇
and R
˜
∇
.
We ﬁnally check the normalization property. Let L → CP
1
be the tautological bundle. For any
section σ : CP
1
→L of L we denote by σ
0
: U
0
→C and σ
1
: U
1
→C the expressions of σ in the
standard local trivializations of L, given by ψ
i
: π
−1
U
i
→U
i
C, ψ
i
(w) = (π(w), w
i
).
The Hermitian product on C
2
induces a Hermitian structure h on L. Let ∇ be the Chern connec
tion on L associated to h. We choose a local holomorphic section σ and denote its square norm
by u. If ω is the connection form of ∇ with respect to the section σ, ∇σ = ω ⊗ σ, then we can
write:
X(u) = X(h(σ, σ)) = h(∇
X
σ, σ) + h(σ, ∇
X
σ) = ω(X)u + ¯ ω(X)u, ∀X ∈ TCP
1
.
This just means ω + ¯ ω = d log u. On the other hand, since σ is holomorphic and ∇
0,1
=
¯
∂, we see
that ω is a (1, 0)–form. Thus ω = ∂ log u. From (38) we get
(40) R
∇
= dω = d∂ log u =
¯
∂∂ log u.
We thus have to check the following condition:
i
2π
CP
1
¯
∂∂ log u = −1.
It is clearly enough to compute this integral over U
0
:= CP
1
−¦[0 : 1]¦. We denote by z := φ
0
=
z
1
z
0
the holomorphic coordinate on U
0
. We now take a particular local holomorphic section σ such
that σ
0
(z) = 1. From the deﬁnition of σ
0
(as the image of σ through the trivialization ψ
0
of L),
we deduce that σ(z) is the unique vector lying on the complex line generated by (z
0
, z
1
) in C
2
,
whose ﬁrst coordinate is 1, i.e. σ(z) = (1, z). This shows that u = [(1, z)[
2
= 1 + [z[
2
. In polar
coordinates z = r cos θ + ir sin θ one can readily compute
¯
∂∂f =
i
2
r
∂
2
f
∂r
2
+
1
r
∂
2
f
∂θ
2
+
∂f
∂r
dr ∧ dθ.
Applying this formula to f := log(1 +r
2
) we ﬁnally get
i
2π
CP
1
¯
∂∂ log u =
i
2π
[0,∞)×[0,2π]
i
2
r
∂
2
f
∂r
2
+
∂f
∂r
dr ∧ dθ
= −
1
2
∞
0
d
r
∂f
∂r
=
1
2
lim
r→∞
r
∂f
∂r
= − lim
r→∞
r
2
2r
1 +r
2
= −1.
10. CHERN CLASSES 59
If M is an almost complex manifold, we deﬁne the ﬁrst Chern class of M – denoted by c
1
(M) –
to be the ﬁrst Chern class of the tangent bundle TM, viewed as complex vector bundle:
c
1
(M) := c
1
(TM).
In the next sections we will see that a representative of the ﬁrst Chern class of a K¨ahler manifold
is
1
2π
ρ, where ρ denotes the Ricci form.
10.2. Properties of the ﬁrst Chern class. Let M be a complex manifold and let E, F be
two complex vector bundles over M.
Proposition 10.4. (i) c
1
(E) = c
1
(Λ
k
E), where k denotes the rank of E.
(ii) c
1
(E ⊗F) = rk(F)c
1
(E) + rk(E)c
1
(F).
(iii) c
1
(E
∗
) = −c
1
(E), where E
∗
denotes the dual of E.
Proof. (i) Consider any connection ∇ in E, inducing a connection
˜
∇ on Λ
k
E. If σ
1
, . . . , σ
k
denotes a local basis of sections of E, then σ := σ
1
∧ . . . ∧ σ
k
is a local non–vanishing section of
Λ
k
E. Let ω := (ω
ij
) and ˜ ω be the connection forms of ∇ and
˜
∇ relative to these local basis:
∇σ
i
= ω
ij
⊗σ
j
and
˜
∇σ = ˜ ω ⊗σ.
We then compute
˜
∇σ =
˜
∇(σ
1
∧ . . . ∧ σ
k
)
=
¸
i
σ
1
∧ . . . ∧ σ
i−1
∧ (
¸
j
ω
ij
⊗σ
j
) ∧ σ
i+1
∧ . . . ∧ σ
k
=
¸
i=j
ω
ij
⊗σ,
which proves that ˜ ω = Tr(ω). From (37) we then get
Tr(R
˜
∇
) = R
˜
∇
= d˜ ω − ˜ ω ∧ ˜ ω = d˜ ω
and
Tr(R
∇
) =
¸
i=j
(dω
ij
−ω
ik
∧ ω
kj
) =
¸
i=j
dω
ij
= dTr(ω) = d˜ ω,
thus proving that c
1
(E) = c
1
(Λ
k
E).
(ii) Let us denote by e and f the ranks of E and F. Because of the canonical isomorphism
Λ
ef
(E ⊗ F)
∼
= (Λ
e
E)
⊗f
⊗ (Λ
f
F)
⊗e
, it is enough to check this relation for line bundles E and F.
Any connections ∇
E
and ∇
F
on E and F respectively induce a connection ∇ on E ⊗ F deﬁned
by
∇(σ
E
⊗σ
F
) := (∇
E
σ
E
) ⊗σ
F
+ σ
E
⊗(∇
F
σ
F
).
The corresponding connection forms are then related by
ω = ω
E
+ ω
F
60
so clearly
R
∇
= dω = d(ω
E
+ ω
F
) = R
∇
E
+ R
∇
F
.
(iii) Again, since (Λ
k
E)
∗
is isomorphic to Λ
k
(E
∗
), we can suppose that E is a line bundle. But in
this case the canonical isomorphism E ⊗E
∗
· C (where C denotes the trivial line bundle) shows
that 0 = c
1
(C) = c
1
(E ⊗E
∗
) = c
1
(E) + c
1
(E
∗
).
10.3. Exercises.
(1) Consider the change of variables z = r cos θ + ir sin θ. Show that for every function
f : U ⊂ C →C the following formula holds:
¯
∂∂f =
i
2
r
∂
2
f
∂r
2
+
1
r
∂
2
f
∂θ
2
+
∂f
∂r
dr ∧ dθ.
(2) Show that the ﬁrst Chern class of a trivial bundle vanishes.
(3) Show that if E is a complex line bundle, there is a canonical isomorphism E ⊗E
∗
· C.
(4) Let ∇ be any connection on a complex bundle E and let ∇
∗
be the induced connection
on the dual E
∗
of E deﬁned by
(∇
∗
X
σ
∗
)(σ) := X(σ
∗
(σ)) −σ
∗
(∇
X
σ).
Show that
R
∇
∗
(X, Y ) = (R
∇
(X, Y ))
∗
,
where A
∗
∈ End(E
∗
) denotes the adjoint of A, deﬁned by A
∗
(σ
∗
)(σ) := −σ
∗
(A(σ)).
11. THE RICCI FORM OF K
¨
AHLER MANIFOLDS 61
11. The Ricci form of K¨ahler manifolds
11.1. K¨ahler metrics as geometric U(m)–structures. We start by a short review on G–
structures which will help us to characterize K¨ahler and Ricci–ﬂat K¨ahler metrics. Let M be an
n–dimensional manifold and let G be any closed subgroup of Gl
n
(R).
Definition 11.1. A topological G–structure on M is a reduction of the principal frame bundle
Gl(M) to G. A geometrical G–structure is given by a topological G–structure G(M) together with
a torsion–free connection on G(M).
Let us give some examples. An orientation on M is a Gl
+
n
(R)–structure. An almost complex
structure is a Gl
m
(C)–structure, for n = 2m. A Riemannian metric is a O
n
–structure. In general,
if the group G can be deﬁned as the group preserving an element of some Gl
n
representation
ρ : Gl
n
(R) → End(V ), then a G–structure is simply a section σ in the associated vector bundle
Gl(M)
ρ
V with the same algebraic properties as v
0
in the sense that for every x ∈ M there
exists u ∈ Gl(M) with σ(x) = [u, v
0
]. To see this, let G be given by
G := ¦g ∈ Gl
n
(R) [ ρ(g)(v
0
) = v
0
¦.
If G(M) is a G–structure, we deﬁne a section in Gl(M)
ρ
V by σ(x) := [u, v
0
] where u is an
arbitrary element of the ﬁber G(M)
x
. This deﬁnition clearly does not depend on u. Conversely,
the set ¦u ∈ Gl(M) [ σ = [u, v
0
]¦ deﬁnes a reduction of the structure group of Gl(M) to G. In
this setting, the G–structure is geometrical if and only if there exists a torsion–free connection on
M with respect to which σ is parallel.
Proposition 11.2. The U
m
–structure deﬁned by an almost complex structure J together with a
Hermitian metric h on a manifold M is geometrical if and only if the metric is K¨ahler.
Proof. The point here is that if G is a closed subgroup of O
n
then there exists at most one
torsion–free connection on any G–structure (by the uniqueness of the Levi–Civita connection). As
U
m
= O
2m
∩Gl
m
(C), the U
m
structure is geometrical if and only if the tensor deﬁning it (namely
J) is parallel with respect to the Levi–Civita connection, which by Theorem 5.5 just means that
h is K¨ahler.
11.2. The Ricci form as curvature form on the canonical bundle. We now turn back
to our main objects of interest. Let (M
2m
, h, J) be a K¨ahler manifold with Ricci form ρ and
canonical bundle K := Λ
m,0
M. As before, we will interpret the tangent bundle TM as a complex
(actually holomorphic) Hermitian vector bundle over M, where the multiplication by i corresponds
to the tensor J and the Hermitian structure is h − iΩ. From Proposition 5.8 we know that the
Levi–Civita connection ∇ on M coincides with the Chern connection on TM.
Lemma 11.3. The curvature R
∇
∈ (
∞
(Λ
2
M ⊗ End(TM)) of the Chern connection and the cur
vature tensor R of the Levi–Civita connection are related by
R
∇
(X, Y )ξ = R(X, Y )ξ,
where X, Y are vector ﬁelds on M and ξ is a section of TM.
62
Proof. The proof is tautological, provided we make explicit the deﬁnition of R
∇
. Let ¦e
i
¦
denote a local basis of vector ﬁelds on M and let ¦e
∗
i
¦ denote the dual local basis of Λ
1
M. Then
R
∇
ξ = ∇
2
ξ = ∇(e
∗
i
⊗∇
e
i
ξ) = de
∗
i
⊗∇
e
i
ξ −e
∗
i
∧ e
∗
j
⊗∇
e
j
∇
e
i
ξ.
Denoting X
i
:= e
∗
i
(X) and Y
i
:= e
∗
i
(Y ) we then obtain
R
∇
(X, Y )ξ = de
∗
i
(X, Y )∇
e
i
ξ −(e
∗
i
∧ e
∗
j
)(X, Y )∇
e
j
∇
e
i
ξ
= (X(Y
i
) −Y (X
i
) −e
∗
i
([X, Y ]))∇
e
i
ξ −(X
i
Y
j
−X
j
Y
i
)∇
e
j
∇
e
i
ξ
= −∇
[X,Y ]
ξ + (X(Y
i
) −Y (X
i
))∇
e
i
ξ −X
i
∇
Y
∇
e
i
ξ + Y
i
∇
X
∇
e
i
ξ
= −∇
[X,Y ]
ξ −∇
Y
∇
X
ξ +∇
X
∇
Y
ξ = R(X, Y )ξ.
We are now ready to prove the following characterization of the Ricci form ρ on K¨ahler manifolds:
Proposition 11.4. The curvature of the Chern connection of the canonical line bundle K is equal
to iρ acting by scalar multiplication on K.
Proof. We ﬁx some notations: let r and r
∗
be the curvatures of the Chern connections of
K := Λ
m,0
M and K
∗
:= Λ
0,m
M. They are related by r = −r
∗
(exercise). Moreover, the connection
induced on Λ
m
(TM) with the induced Hermitian structure by the Chern connection on TM is
clearly the Chern connection of Λ
m
(TM). It is easy to check that Λ
m
(TM) is isomorphic to K
∗
,
so from the proof of Proposition 10.4 and from Lemma 11.3 we get
r
∗
(X, Y ) = Tr(R
∇
(X, Y )) = Tr(R(X, Y )).
Since we will now use both complex and real traces, we will make this explicit by a superscript.
By Proposition 6.2 we then obtain
iρ(X, Y ) = iRic(JX, Y ) =
i
2
Tr
R
(R(X, Y ) ◦ J)
=
i
2
(2iTr
C
(R(X, Y )) = −Tr
C
(R(X, Y ))
= −r
∗
(X, Y ) = r(X, Y ),
where we used the fact that
Tr
R
(A
R
◦ J) = 2iTr
C
(A)
for every skew–hermitian endomorphism A.
11.3. Ricci–ﬂat K¨ahler manifolds. Let (M
2m
, h, J) be a K¨ahler manifold with canonical
bundle K (endowed with the Hermitian structure induced from the K¨ahler metric on TM) and
Ricci form ρ. We suppose, for simplicity, that M is simply connected. Then the previous results
can be summarized as follows:
Theorem 11.5. The ﬁve statements below are equivalent:
(1) M is Ricci–ﬂat.
(2) The Chern connection of the canonical bundle K is ﬂat.
11. THE RICCI FORM OF K
¨
AHLER MANIFOLDS 63
(3) There exists a parallel complex volume form, that is, a parallel section of Λ
m,0
M.
(4) M has a geometrical SU
m
–structure.
(5) The Riemannian holonomy of M is a subgroup of SU
m
.
In the non–simply connected case, the last 3 statements are only local.
Proof. (1) ⇐⇒(2) is a direct consequence of Proposition 11.4.
(2) ⇐⇒(3) follows from the general principle that a connection on a line bundle is ﬂat if and only
if there exists a parallel section (globally deﬁned if π
1
(M) = 0, and locally deﬁned otherwise).
(3) ⇐⇒ (4) The special unitary group SU
m
can be deﬁned as the stabilizer of a vector in the
canonical representation of U
m
onto Λ
m,0
C. Thus, there exists a parallel section in Λ
m,0
M if and
only if the geometrical U
m
–structure deﬁned by the K¨ahler metric can be further reduced to a
geometrical SU
m
–structure.
(4) =⇒ (5) If G(M) is a G–structure, the holonomy of a connection in G(M) is contained in G.
Now, if M has a geometrical SU
m
–structure, the torsion–free connection deﬁning it is just the
Levi–Civita connection, therefore the Riemannian holonomy group is a subgroup of SU
m
.
(5) =⇒ (4) The reduction theorem ([8], Ch. 2, Thm. 7.1) shows that for every ﬁxed frame u, the
holonomy bundle (that is, the set of frames obtained from u by parallel transport) is a Hol
u
(M)–
principal bundle, and the Levi–Civita connection can be restricted to it. Thus, if the Riemannian
holonomy Hol(M) of M is a subgroup of SU
m
, we get a geometrical SU
m
–structure simply by
extending the holonomy bundle to SU
m
.
Notice that by Theorem 10.3 and Proposition 11.4, for a given K¨ahler manifold (M, h, J), the
vanishing of the ﬁrst Chern class of (M, J) is a necessary condition for the existence of a Ricci–
ﬂat K¨ahler metric on M compatible with J. The converse statement is also true if M is compact,
and will be treated in the next section.
11.4. Exercises.
(1) Let G be a closed subgroup of Gl
n
(R) containing SO
n
. Show that every G–structure is
geometrical.
(2) Let M
n
be a connected diﬀerentiable manifold. Prove that M is orientable if and only if
its frame bundle Gl
n
(M) is not connected.
(3) Show that a U
m
–structure on M deﬁnes an almost complex structure together with a
Hermitian metric.
(4) Show that a geometrical Gl
m
(C)–structure is the same as an integrable almost complex
structure. Hint: Start with a torsion free connection ∇ and consider the connection
˜
∇
64
deﬁned by
˜
∇
X
Y := ∇
X
Y −A
X
Y , where A
X
Y =
1
4
(2J(∇
X
J)Y +(∇
JY
J)X+J(∇
Y
J)X).
Use the proof of Lemma 5.4 to check that A is symmetric if and only if J is integrable.
(5) Let A be a skew–hermitian endomorphism of C
m
and let A
R
be the corresponding real
endomorphism of R
2m
. Show that
Tr
R
(A
R
◦ J) = 2iTr
C
(A).
(6) The special unitary group SU
m
is usually deﬁned as the subgroup of U
m
⊂ Gl
m
(C)
consisting of complex unitary matrices of determinant 1. Prove that SU
m
is equal to the
stabilizer in U
m
of the form dz
1
∧ . . . ∧ dz
m
.
(7) Let (L, h) be a complex line bundle with Hermitian structure over some smooth manifold
M. Prove that the space of Hermitian connections is an aﬃne space over the real vector
space (
∞
(Λ
1
M). Equivalently, there is a free transitive group action of (
∞
(Λ
1
M) on the
space of Hermitian connections on L.
(8) If L is a complex line bundle over M, show that every real closed 2–form in the cohomology
class c
1
(L) ∈ H
2
(M, R) is
i
2π
times the curvature of some connection on L.
12. THE CALABI CONJECTURE 65
12. The Calabi conjecture
12.1. An overview. We have seen that the ﬁrst Chern class of any compact K¨ahler manifold
is represented by
1
2π
ρ. Conversely, we have the following famous result due to Calabi and Yau
Theorem 12.1. Let M
m
be a compact K¨ahler manifold with fundamental form ϕ and Ricci form
ρ. Then for every closed real (1, 1)–form ρ
1
in the cohomology class 2πc
1
(M), there exists a unique
K¨ahler metric with fundamental form ϕ
1
in the same cohomology class as ϕ, whose Ricci form is
exactly ρ
1
.
Before giving an outline of the proof, we state some corollaries.
Corollary 12.2. If the ﬁrst Chern class of a compact K¨ahler manifold vanishes, then M carries
a Ricci–ﬂat K¨ahler metric.
Corollary 12.3. If the ﬁrst Chern class of a compact K¨ahler manifold is positive, then M is
simply connected.
Proof. By the Calabi theorem M has a K¨ahler metric with positive Ricci curvature, so the
result follows from Theorem 15.6 below.
The ﬁrst step in the proof of Theorem 12.1 is to reformulate the problem in order to reduce it
to a so–called Monge–Amp`ere equation. We denote by H the set of K¨ahler metrics in the same
cohomology class as ϕ. The global i∂
¯
∂–Lemma shows that
(41) H =
u ∈ (
∞
(M) [ ϕ + i∂
¯
∂u > 0,
M
uϕ
m
= 0
(this last condition is needed since u is only deﬁned up to a constant).
Now, if g and g
1
are K¨ahler metrics with K¨ahler forms ϕ and ϕ
1
in the same cohomology class,
we denote by dv :=
1
m!
ϕ
m
and dv
1
:=
1
m!
ϕ
m
1
their volume forms and consider the real function f
deﬁned by e
f
dv = dv
1
. Since [ϕ] = [ϕ
1
] we also have [ϕ
m
] = [ϕ
m
1
], that is
(42)
M
e
f
dv =
M
dv.
Let ρ and ρ
1
denote the corresponding Ricci forms. Since iρ is the curvature of the canonical
bundle K
M
, for every local holomorphic section ω of K
M
we have
(43) iρ =
¯
∂∂ log g(ω, ¯ ω) and iρ
1
=
¯
∂∂ log g
1
(ω, ¯ ω).
It is easy to check that the Hodge operator acts on Λ
m,0
simply by scalar multiplication with
ε := i
m
(−1)
m(m+1)
2
. We thus have
(44) εω ∧ ¯ ω = ω ∧ ∗¯ ω = g(ω, ¯ ω)dv
and similarly
(45) εω ∧ ¯ ω = g
1
(ω, ¯ ω)dv
1
= e
f
g
1
(ω, ¯ ω)dv.
66
From (43)–(45) we get
(46) iρ
1
−iρ = ∂
¯
∂f.
This shows that the Ricci form of the modiﬁed K¨ahler metric ϕ
1
= ϕ+i∂
¯
∂u can be computed by
the formula
(47) ρ
1
= ρ −i∂
¯
∂f, where f = log
(ϕ + i∂
¯
∂u)
m
ϕ
m
.
Now given closed real (1, 1)–form ρ
1
in the cohomology class 2πc
1
(M), the global i∂
¯
∂–Lemma
shows that there exists some real function f such that ρ
1
= ρ −i∂
¯
∂f. Moreover, f is unique if we
impose the normalization condition (42). We denote by H
the space of smooth functions on M
satisfying this condition. The Calabi conjecture is then equivalent to the following
Theorem 12.4. The mapping Cal : H →H
deﬁned by
Cal(u) = log
(ϕ + i∂
¯
∂u)
m
ϕ
m
is a diﬀeomorphism.
We ﬁrst show that Cal is injective. It is clearly enough to show that Cal(u) = 0 and u ∈ H
implies u = 0. If Cal(u) = 0 we have ϕ
m
1
= ϕ
m
, and since 2–forms commute we obtain
0 = ϕ
m
1
−ϕ
m
= i∂
¯
∂u ∧
m−1
¸
k=0
ϕ
k
1
∧ ϕ
m−k−1
.
Using the formula 2i∂
¯
∂ = dd
c
and the fact that ϕ and ϕ
1
are closed forms we get after multipli
cation by u
0 = 2iu∂
¯
∂u ∧
m−1
¸
k=0
ϕ
k
1
∧ ϕ
m−k−1
= udd
c
u ∧
m−1
¸
k=0
ϕ
k
1
∧ ϕ
m−k−1
= d
ud
c
u ∧
m−1
¸
k=0
ϕ
k
1
∧ ϕ
m−k−1
−du ∧ d
c
u ∧
m−1
¸
k=0
ϕ
k
1
∧ ϕ
m−k−1
.
Integrating over M and using Stokes’ theorem yields
(48) 0 =
m−1
¸
k=0
M
du ∧ Jdu ∧ ϕ
k
1
∧ ϕ
m−k−1
.
Now, since ϕ
1
deﬁnes a K¨ahler metric, there exists a local basis ¦e
1
, Je
1
, . . . , e
m
, Je
m
¦ orthonormal
with respect to g such that
ϕ =
m
¸
j=1
e
j
∧ Je
j
and ϕ
1
=
m
¸
j=1
a
j
e
j
∧ Je
j
,
12. THE CALABI CONJECTURE 67
where a
j
are strictly positive local functions. This shows easily that for every k
ϕ
k
1
∧ ϕ
m−k−1
= ∗(
m
¸
j=1
b
k
j
e
j
∧ Je
j
), b
k
j
> 0.
In fact one can compute explicitly
b
k
j
= k!(m−k −1)!
¸
j
1
=j,...,j
k
=j
j
1
<...<j
k
a
j
1
. . . a
j
k
.
This shows that the integrand in (48) is strictly positive unless du = 0. Thus u is a constant, so
u = 0 because the integral of u dv over M vanishes. Therefore Cal is injective.
To prove that it is a local diﬀeomorphism, we compute its diﬀerential at some u ∈ H. By
changing the reference metric if necessary, we may suppose without loss of generality that u = 0.
For v ∈ T
0
H we compute
Cal
∗
(v) =
d
dt
[
t=0
(Cal(tv)) =
d
dt
[
t=0
(ϕ + i∂
¯
∂tv)
m
ϕ
m
= m
i∂
¯
∂v ∧ ϕ
m−1
ϕ
m
= Λ(i∂
¯
∂v) = −
¯
∂
∗
¯
∂v = −∆
¯
∂
v.
From the general elliptic theory we know that the Laplace operator is a bijection of the space
of functions with zero integral over M. Thus Cal
∗
is bijective, so the Inverse Function Theorem
shows that Cal is a local diﬀeomorphism.
The surjectivity of Cal, which is the hard part of the theorem, follows from a priori estimates,
which show that Cal is proper. We refer the reader to [6] for details.
12.2. Exercises.
(1) Show that ∗ω = i
m(m+2)
ω for all ω ∈ Λ
m,0
M on a Hermitian manifold M of complex
dimension m.
(2) Prove that the mapping
u →ϕ + i∂
¯
∂u
is indeed a bijection from the set deﬁned in (41) to the set of K¨ahler metrics in the
cohomology class [ϕ].
(3) Prove that the total volume of a K¨ahler metric on a compact manifold only depends on
the cohomology class of its fundamental form.
(4) Show that ∗(ϕ
m−1
) = (m− 1)!ϕ on every Hermitian manifold M of complex dimension
m. Using this, prove that if (M, ϕ) is K¨ahler, then ρ∧ϕ
m−1
= (m−1)!S, where ρ denotes
the Ricci form and S is the scalar curvature of the K¨ahler metric deﬁned by ϕ.
68
(5) Let (M
m
, J) be a compact complex manifold. Show that the integral over M of the scalar
curvature of a K¨ahler metric only depends on the cohomology class of its fundamental
form ϕ. More precisely one has
M
Sdv = 2πmc
1
(M) ∪ [ϕ]
m−1
.
13. K
¨
AHLER–EINSTEIN METRICS 69
13. K¨ahler–Einstein metrics
13.1. The Aubin–Yau theorem. We turn our attention to compact K¨ahler manifolds
(M, g) satisfying the Einstein condition
Ric = λg, λ ∈ R.
We will exclude the case λ = 0 which was treated above. If we rescale the metric by a positive
constant, the curvature tensor does not change, so neither does the Ricci tensor, which was deﬁned
as a trace. This shows that we may suppose that λ = ε = ±1. The K¨ahler–Einstein condition
reads
ρ = εϕ, ε = ±1.
As the ﬁrst Chern class of M is represented by
ρ
2π
, we see that a necessary condition for the
existence of a K¨ahler–Einstein manifold on a given compact K¨ahler manifold is that its ﬁrst
Chern class is deﬁnite (positive or negative). In the negative case, this condition turns out to be
also suﬃcient:
Theorem 13.1. (Aubin, Yau) A compact K¨ahler manifold with negative ﬁrst Chern class admits
a unique K¨ahler–Einstein metric with Einstein constant ε = −1.
We will treat simultaneously the two cases ε = ±1, in order to emphasize the diﬃculties that
show up in the case ε = 1.
As before, we ﬁrst reformulate the problem. Let (M
2m
, g, J, ϕ, ρ) be a compact K¨ahler manifold
with deﬁnite ﬁrst Chern class c
1
(M). By deﬁnition, there exists a positive closed (1, 1)–form
representing the cohomology class 2πεc
1
(M). We can suppose without loss of generality that this
form is equal to ϕ (otherwise we just change the initial K¨ahler metric). Then [ϕ] = 2πεc
1
(M) =
[ερ], so the global i∂
¯
∂–Lemma shows that there exists some function f with
(49) ρ = εϕ + i∂
¯
∂f.
We are looking for a new K¨ahler metric g
1
with fundamental form ϕ
1
and Ricci form ρ
1
such that
ρ
1
= εϕ
1
. Suppose we have such a metric. From our choice for ϕ we have
[2πϕ] = εc
1
(M) = [2περ
1
] = [2πϕ
1
].
From this equation and the global i∂
¯
∂–Lemma it is clear that there exists a unique function u ∈ H
such that ϕ
1
= ϕ+i∂
¯
∂u. Now the previously obtained formula (47) for the Ricci form of the new
metric reads
(50) ρ
1
= ρ −i∂
¯
∂ log
(ϕ + i∂
¯
∂u)
m
ϕ
m
.
Using (49) and (50), the K¨ahler–Einstein condition for g
1
becomes
(51) εϕ + i∂
¯
∂f −i∂
¯
∂ log
(ϕ + i∂
¯
∂u)
m
ϕ
m
= εϕ
1
,
70
which is equivalent to
(52) log
(ϕ + i∂
¯
∂u)
m
ϕ
m
+ εu = f + const.
Conversely, if u ∈ (
∞
+
(M) satisﬁes this equation, then the K¨ahler metric ϕ
1
:= ϕ+i∂
¯
∂u is K¨ahler–
Einstein (we denote by (
∞
+
(M) the space of all smooth functions u on M such that ϕ+i∂
¯
∂u > 0).
The Aubin–Yau theorem is therefore equivalent to the fact that the mapping
Cal
ε
: (
∞
+
(M) →(
∞
(M) Cal
ε
(u) := Cal(u) + εu
is a diﬀeomorphism.
The injectivity of Cal
−
can be proved as follows. Suppose that Cal
−
(u
1
) = Cal
−
(u
2
) and denote
by ϕ
1
:= ϕ + i∂
¯
∂u
1
and ϕ
2
:= ϕ + i∂
¯
∂u
2
. Then
log
ϕ
m
1
ϕ
m
−u
1
= log
ϕ
m
2
ϕ
m
−u
2
,
hence, denoting the diﬀerence u
2
−u
1
by u:
(53) log
(ϕ
1
+ i∂
¯
∂u)
m
ϕ
m
1
= u.
At a point where u attains its maximum, the (1, 1)–form i∂
¯
∂u is negative semi–deﬁnite, since we
can write (for any vector X parallel at that point)
i∂
¯
∂u(X, JX) =
1
2
(dd
c
u)(X, JX) =
1
2
(X(d
c
u(JX)) −JX(d
c
u(X)))
=
1
2
(H
u
(X, X) + H
u
(JX, JX)) ≤ 0,
since the Hessian H
u
of u is of course negative semi–deﬁnite at a point where u reaches its
maximum. Taking into account (53) we see that u ≤ 0 at each of its maximum points, so u ≤ 0
on M. Similarly, u ≥ 0 at each minimum points, so ﬁnally u = 0 on M, thus proving the injectivity
of Cal
−
.
We have already computed the diﬀerential of Cal at u = 0 applied to some v ∈ T
0
(
∞
+
(M):
Cal
∗
(v) = −∆
¯
∂
v.
Consequently
Cal
−
∗
(v) = −v −∆
¯
∂
v
is a bijection of (
∞
(M) since the self–adjoint elliptic operator v →
1
2
∆v + v has obviously no
kernel and its index is zero.
As before, the surjectivity of Cal
−
is harder to prove and requires non–trivial analysis (see [6]).
13. K
¨
AHLER–EINSTEIN METRICS 71
13.2. Holomorphic vector ﬁelds on compact K¨ahler–Einstein manifolds. Let M
2m
be a compact K¨ahler manifold. We start by showing the following
Lemma 13.2. Let ξ be a holomorphic (real) vector ﬁeld with dual 1–form also denoted by ξ. Then
ξ can be decomposed in a unique manner as
ξ = df + d
c
h + ξ
H
,
where f and h are functions with vanishing integral and ξ
H
is the harmonic part of ξ in the usual
Hodge decomposition.
Proof. Since ξ is holomorphic we have L
ξ
J = 0, so [ξ, JX] = J[ξ, X] for every vector ﬁeld X.
Thus ∇
JX
ξ = J∇
X
ξ, so taking the scalar product with some JY and skew–symmetrising yields
dξ(JX, JY ) = dξ(X, Y ), i.e. dξ is of type (1, 1). The global dd
c
–Lemma shows that dξ = dd
c
h
for some function h. The form ξ −d
c
h is closed, so the Hodge decomposition theorem shows that
ξ −d
c
h = df + ξ
0
for some function f and some harmonic 1–form ξ
0
. Comparing this formula with the Hodge decom
position for ξ and using the fact that harmonic 1–forms are L
2
–orthogonal to d(
∞
(M), d
c
(
∞
(M)
and δΩ
2
(M) shows that ξ
0
equals ξ
H
, the harmonic part of ξ. Finally, the uniqueness of f and h
follows easily from the normalization condition, together with the fact that d(
∞
(M) and d
c
(
∞
(M)
are L
2
–orthogonal.
Next, we have the following characterization of real holomorphic and Killing vector ﬁelds on
compact K¨ahler–Einstein manifolds with positive scalar curvature.
Lemma 13.3. A vector ﬁeld ξ (resp. ζ) on a compact K¨ahler–Einstein manifold M
2m
with positive
scalar curvature S is Killing (resp holomorphic) if and only if ξ = Jdh (resp. ζ = df +d
c
h) where
h (resp. f and h) are eigenfunctions of the Laplace operator corresponding to the eigenvalue
S
m
.
Proof. The Ricci tensor of M satisﬁes Ric(X) =
S
2m
X for every vector X. Let ξ be a vector
ﬁeld on M. If we view as usual TM as a holomorphic vector bundle, then the Weitzenb¨ock formula
(see (65) below) yields
(54) 2
¯
∂
∗
¯
∂ξ = ∇
∗
∇ξ + iρξ = ∇
∗
∇ξ −Ric(ξ) = ∇
∗
∇ξ −
S
2m
ξ.
The Bochner formula (Exercise 3 in the next section) reads
(55) ∆ξ = ∇
∗
∇ξ + Ric(ξ) = ∇
∗
∇ξ +
S
2m
ξ.
Since S > 0, this shows that there are no harmonic 1–forms on M.
Suppose that ζ is holomorphic. From Lemma 13.2 we then can write ζ as a sum ζ = df + d
c
h,
where f and h have vanishing integrals over M. Now, subtracting (54) from (55) yields ∆ζ =
S
m
ζ,
so ∆(df +d
c
h) = d(
S
m
f) +d
c
(
S
m
h), and since ∆ commutes with d and d
c
, and the images of d and
d
c
are L
2
–orthogonal, this yields ∆f =
S
m
f +c
1
and ∆h =
S
m
h +c
2
. Finally the constants have to
vanish because of the normalization condition.
72
If ξ is Killing, then ξ is holomorphic by Proposition 9.5. The codiﬀerential of every Killing vector
ﬁeld vanishes, and moreover δ anti–commutes with d
c
. Thus 0 = δξ = δdf, showing that df = 0,
so ξ = d
c
h with ∆h =
S
m
h.
Conversely, suppose that ξ = df + d
c
h and f and h are eigenfunctions of the Laplace operator
corresponding to the eigenvalue
S
m
. Then ∆ξ =
S
m
ξ, so from (55) we get
S
m
ξ = ∇
∗
∇ξ +
S
2m
ξ.
Then (54) shows that ξ is holomorphic.
If moreover df = 0, we have
L
ξ
ϕ = d(ξ ϕ) + ξ dϕ = d(Jξ) = −ddh = 0,
where ϕ is the fundamental form of M. Together with L
ξ
J = 0, this shows that L
ξ
g = 0, so ξ is
Killing.
We are now ready to prove the following result of Matsushima:
Theorem 13.4. The Lie algebra g(M) of Killing vector ﬁelds on a compact K¨ahler–Einstein man
ifold M with positive scalar curvature is a real form of the Lie algebra h(M) of (real) holomorphic
vector ﬁelds on M. In particular h(M) is reductive, i.e. it is the direct sum of its center and a
semi–simple Lie algebra.
Proof. Let F : g(M) ⊗C →h(M) be the linear map given by F(ξ + iζ) := ξ + Jζ. Since J
maps holomorphic vector ﬁelds to holomorphic vector ﬁelds, F is well–deﬁned. The two lemmas
above clearly show that F is a vector space isomorphism. Moreover, F is a Lie algebra morphism
because Killing vector ﬁelds are holomorphic, so
F([ξ + iζ, ξ
1
+ iζ
1
]) = [ξ, ξ
1
] −[ζ, ζ
1
] + J([ξ, ζ
1
] + [ζ, ξ
1
]) = [ξ, ξ
1
] + J
2
[ζ, ζ
1
] + [Jξ, ζ
1
] + [Jζ, ξ
1
]
= [ξ, ξ
1
] + [Jζ, Jζ
1
] + [Jξ, ζ
1
] + [Jζ, ξ
1
] = [F(ξ + iζ), F(ξ
1
+ iζ
1
)].
The last statement follows from the fact that the isometry group of M is compact, and every Lie
algebra of compact type, as well as its complexiﬁcation, is reductive.
There exist compact K¨ahler manifolds with positive ﬁrst Chern class whose Lie algebra of holomor
phic vector ﬁelds is not reductive. Therefore such a manifold carries no K¨ahler–Einstein metric,
thus showing that Theorem 13.1 cannot hold in the positive case.
Part 5
Vanishing results
14. Weitzenb¨ock techniques
14.1. The Weitzenb¨ock formula. The aim of the next 3 sections is to derive vanishing
results under certain positivity assumptions on the curvature using Weitzenb¨ock techniques.
The general principle is the following: let (E, h) → M be some holomorphic Hermitian bundle
over a compact K¨ahler manifold (M
2m
, g, J), with holomorphic structure
¯
∂ : (
∞
(Λ
p,q
M ⊗ E) →
(
∞
(Λ
p,q+1
M⊗E) and Chern connection ∇ : (
∞
(Λ
p,q
M⊗E) →(
∞
(Λ
1
C
M⊗Λ
p,q
M⊗E). If
¯
∂
∗
and
∇
∗
are the formal adjoints of
¯
∂ and ∇, it turns out that the diﬀerence of the diﬀerential operators
of order two ∇
∗
∇ and 2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
) is a zero–order operator, depending only on the curvature of
the Chern connection:
(56) 2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
) = ∇
∗
∇+1,
where 1 is a section in End(Λ
p,q
M ⊗ E). If 1 is a positive operator on Λ
p,0
M ⊗ E, then every
holomorphic section of Λ
p,0
M ⊗E is ∇–parallel, and if 1 is strictly positive on Λ
p,0
M ⊗E, then
this holomorphic bundle has no holomorphic section. This follows by applying (56) to some
holomorphic section σ of Λ
p,0
M ⊗ E, taking the scalar product with σ and integrating over M,
using the fact that
¯
∂
∗
vanishes identically on Λ
p,0
M ⊗E.
We start with the following technical lemma:
Lemma 14.1. If ¦e
j
¦ is a local orthonormal basis in TM (identiﬁed via the metric g with an
orthonormal basis of Λ
1
M), and ∇ denotes the Chern connection of E, as well as its prolongation
to Λ
p,q
M ⊗E using the Levi–Civita connection on the left–hand side of this tensor product, then
¯
∂,
¯
∂
∗
, ∇
∗
and ∇
∗
∇ are given locally by
(57)
¯
∂ : (
∞
(Λ
p,q
M ⊗E) →(
∞
(Λ
p,q+1
M ⊗E)
¯
∂σ =
1
2
(e
j
−iJe
j
) ∧ ∇
e
j
(σ),
(58)
¯
∂
∗
: (
∞
(Λ
p,q
M ⊗E) →(
∞
(Λ
p,q−1
M ⊗E)
¯
∂
∗
σ = −
1
2
(e
j
+ iJe
j
) ∇
e
j
(σ),
(59) ∇
∗
: (
∞
(Λ
1
C
M ⊗Λ
p,q
M ⊗E) →(
∞
(Λ
p,q
M ⊗E) ∇
∗
(ω ⊗σ) = (δω)σ −∇
ω
σ,
(60) ∇
∗
∇ : (
∞
(Λ
p,q
M ⊗E) →(
∞
(Λ
p,q
M ⊗E) ∇
∗
∇σ = ∇
∇e
j
e
j
σ −∇
e
j
∇
e
j
σ.
Proof. If (E, h
E
) and (F, h
F
) are Hermitian bundles, their tensor product inherits a natural
Hermitian structure given by
h
E⊗F
(σ
E
⊗σ
F
, s
E
⊗s
F
) := h
E
(σ
E
, s
E
)h
F
(σ
F
, s
F
).
The Hermitian structure, on Λ
p,q
M ⊗ E with respect to which one deﬁnes the adjoint operators
above is obtained in this way from the Hermitian structure h of E and the Hermitian structure H
of Λ
p,q
M given by (31). We will use the same symbol H for this Hermitian structure, by a slight
abuse of notation.
The relation (57) is more or less tautological, using the deﬁnition of
¯
∂ and the fact that e
j
−iJe
j
is a (1, 0)–vector, identiﬁed via the metric g with a (0, 1)–form. Of course, the wedge product
there only concerns the Λ
p,q
M–part of σ.
14. WEITZENB
¨
OCK TECHNIQUES 75
For σ ∈ (
∞
(Λ
p,q
M ⊗E) and s ∈ (
∞
(Λ
p,q−1
M ⊗E) we deﬁne the 1–form α by
α(X) :=
1
2
H((X + iJX) σ, s).
By choosing the local basis ¦e
j
¦ parallel in a point for simplicity, then we get at that point:
−δα = e
j
(α(e
j
)) =
1
2
H((e
j
+ iJe
j
) ∇
e
j
σ, s) +
1
2
H((e
j
+ iJe
j
) σ, ∇
e
j
s)
=
1
2
H((e
j
+ iJe
j
) ∇
e
j
σ, s) +
1
2
H(σ, (e
j
−iJe
j
) ∧ ∇
e
j
s)
=
1
2
H((e
j
+ iJe
j
) ∇
e
j
σ, s) + H(σ,
¯
∂s),
thus showing that the operator −
1
2
(e
j
+ iJe
j
) ∇
e
j
is the formal adjoint of
¯
∂.
The proof of (59) is similar: for ω ⊗ σ ∈ (
∞
(Λ
1
C
M ⊗ Λ
p,q
M ⊗ E) and s ∈ (
∞
(Λ
p,q
M ⊗ E) we
deﬁne the 1–form α by
α(X) := H((ω(X))σ, s)
and compute
−δα = e
j
(α(e
j
)) = −H((δω)σ, s) + H((ω(e
j
))∇
e
j
σ, s) + H((ω(e
j
))σ, ∇
e
j
s)
= H(∇
ω
σ −(δω)σ, s) + H(ω ⊗σ, ∇s),
whence ∇
∗
(ω ⊗σ) = (δω)σ −∇
ω
σ.
Finally, we apply (59) to some section ∇σ = e
j
⊗∇
e
j
σ of Λ
1
C
M ⊗Λ
p,q
M ⊗E and get
∇
∗
∇σ = (δe
j
)∇
e
j
σ −∇
e
j
∇
e
j
σ = −g(e
k
, ∇
e
k
e
j
)∇
e
j
σ −∇
e
j
∇
e
j
σ
= g(∇
e
k
e
k
, e
j
)∇
e
j
σ −∇
e
j
∇
e
j
σ = ∇
∇e
j
e
j
σ −∇
e
j
∇
e
j
σ.
We are now ready for the main result of this section
Theorem 14.2. Let (E, h) → (M
2m
, g, J) be a holomorphic Hermitian bundle over a K¨ahler
manifold M. For vectors X, Y ∈ TM, let
˜
R(X, Y ) ∈ End(Λ
p,q
M ⊗ E) be the curvature operator
of the tensor product connection on Λ
p,q
M ⊗ E induced by the Levi–Civita connection on Λ
p,q
M
and the Chern connection on E. Then the following formula holds
(61) 2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
) = ∇
∗
∇+1,
where 1 is the section of End(Λ
p,q
M ⊗E) deﬁned by
(62) 1(σ) :=
i
2
˜
R(Je
j
, e
j
)σ −
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) (
˜
R(e
j
, e
k
)σ).
76
Proof. The proof is a simple computation in a local orthonormal frame parallel at a point,
using Lemma 14.1, (57),(58) and (60):
2
¯
∂
∗
¯
∂ = −
1
2
(e
k
+ iJe
k
) ∇
e
k
((e
j
−iJe
j
) ∧ ∇
e
j
)
= −
1
2
(e
k
+ iJe
k
) ((e
j
−iJe
j
) ∧ ∇
e
k
∇
e
j
)
= −(g(e
k
, e
j
) + ig(Je
k
, e
j
))∇
e
k
∇
e
j
+
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) ∇
e
k
∇
e
j
= ∇
∗
∇−ig(Je
k
, e
j
)∇
e
k
∇
e
j
+
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) ∇
e
j
∇
e
k
+
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
)
˜
R(e
k
, e
j
)
= ∇
∗
∇−
i
2
g(Je
k
, e
j
)
˜
R(e
k
, e
j
) +
1
2
(e
j
−iJe
j
) ∧ ∇
e
j
((e
k
+ iJe
k
) ∇
e
k
)
+
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
)
˜
R(e
k
, e
j
)
= ∇
∗
∇−2
¯
∂
¯
∂
∗
+1.
14.2. Vanishing results on K¨ahler manifolds. Most of the applications will concern the
case q = 0. The expression of the curvature term becomes then particularly simple, since the last
term in (62) automatically vanishes. Let ρ
(p)
denote the action of the Ricci–form of M on Λ
p,0
given by
ρ
(p)
ω := ρ(e
j
) ∧ e
j
ω.
It is easy to check that this action preserves the space Λ
p,0
.
Proposition 14.3. If q = 0, for every section ω ⊗ξ of Λ
p,0
M ⊗E we have
(63) 2
¯
∂
∗
¯
∂(ω ⊗ξ) = ∇
∗
∇(ω ⊗ξ) + i(ρ
(p)
ω) ⊗ξ +
i
2
ω ⊗R
E
(Je
j
, e
j
)ξ,
where R
E
is the curvature of E.
Proof. The curvature
˜
R of Λ
p,0
M ⊗E decomposes in a sum
(64)
˜
R(X, Y )(ω ⊗ξ) = (R(X, Y )ω) ⊗ξ + ω ⊗R
E
(X, Y )(ξ),
where R is the Riemannian curvature. It is an easy exercise to check that the Riemannian
curvature operator acts on forms by R(X, Y )(ω) = R(X, Y )e
k
∧e
k
ω. From Proposition (6.2) (i)
we have 2ρ = R(Je
j
, e
j
) as endomorphisms of the tangent space of M. Therefore Theorem 14.2
and (64) yield the desired result.
We are now ready to obtain the vanishing results mentioned above.
14. WEITZENB
¨
OCK TECHNIQUES 77
Theorem 14.4. Let M be a compact K¨ahler manifold. If the Ricci curvature of M is negative
deﬁnite (i.e. Ric(X, X) < 0 for all non–zero X ∈ TM) then M has no holomorphic vector ﬁeld.
Proof. Let us take p = 0 and E = T
1,0
M in Proposition 14.3. If ξ is a holomorphic vector
ﬁeld, we have
(65) 0 = 2
¯
∂
∗
¯
∂ξ = ∇
∗
∇ξ +
i
2
R(Je
j
, e
j
)ξ = ∇
∗
∇ξ + iρ(ξ).
Taking the (Hermitian) scalar product with ξ in this formula and integrating over M, using the
fact that ρξ = Ric(Jξ) = iRic(ξ) yields
0 =
M
H(∇
∗
∇ξ −Ric(ξ), ξ)dv =
M
[∇ξ[
2
−H(Ricξ, ξ)dv.
Thus, if Ric is negative deﬁnite, ξ has to vanish identically.
Theorem 14.5. Let M be a compact K¨ahler manifold. If the Ricci curvature of M vanishes, then
every holomorphic form is parallel. If the Ricci curvature of M is positive deﬁnite, then there
exist no holomorphic (p, 0)–forms on M for p > 0.
Proof. We take E to be trivial and apply (63) to some holomorphic (p, 0)–form ω. Since
ρ = 0 we get 0 = ∇
∗
∇ω. Taking the Hermitian product with ω and integrating over M yields the
result.
Suppose now that Ric is positive deﬁnite. From (63) applied to some holomorphic (p, 0)–form ω
we get
(66) 0 = ∇
∗
∇ω + iρ
(p)
(ω).
The interior product of a (0, 1)–vector and ω vanishes, showing that JX ω = iX ω. We thus
get
iρ
(p)
(ω) = iρ(e
j
) ∧ e
j
ω = iρ(Je
j
) ∧ Je
j
ω = −ρ(Je
j
) ∧ e
j
ω = Ric(ω).
Since Ric is positive, its extension to (p, 0)–forms is positive, too, hence taking the Hermitian
product with ω in (66) and integrating over M yields
M
[∇ω[
2
+ H(Ric(ω), ω)dv = 0,
showing that ω has to vanish
14.3. Exercises.
(1) Show that the extension to Λ
p
M ⊗ C of a positive deﬁnite symmetric endomorphism of
TM is positive deﬁnite.
(2) Prove the following real version of the Weitzenb¨ock formula:
∆ω = ∇
∗
∇ω +1ω, ∀ ω ∈ Ω
p
M,
78
where 1 is the endomorphism of Ω
p
M deﬁned by
1(ω) := −e
j
∧ e
k
(R(e
j
, e
k
)(ω)).
(3) Applying the above identity to 1–forms, prove the Bochner formula
∆ω = ∇
∗
∇ω + Ric(ω), ∀ ω ∈ Ω
1
M.
(4) Prove that there are no global holomorphic forms on the complex projective space.
15. THE HIRZEBRUCH–RIEMANN–ROCH FORMULA 79
15. The Hirzebruch–Riemann–Roch formula
15.1. Positive line bundles. In order to state another application of the Weitzenb¨ock for
mula we have to make the following
Definition 15.1. A real (1, 1)–form ϕ on a complex manifold (M, g, J) is called positive (resp.
negative) if the symmetric tensor A satisfying A(JX, Y ) := ϕ(X, Y ) is positive (resp. negative)
deﬁnite. A cohomology class in H
1,1
M ∩ H
2
(M, R) is called positive (resp. negative) if it can
be represented by a positive (resp. negative) (1, 1)–form. A holomorphic line bundle L over a
compact complex manifold is called positive (resp. negative) if there exists a Hermitian structure
on L with Chern connection ∇ and curvature form R
∇
such that iR
∇
is a positive (resp. negative)
(1, 1)–form.
The positivity of a holomorphic line bundle is a topological property on K¨ahler manifolds:
Lemma 15.2. A holomorphic line bundle L over a compact K¨ahler manifold M is positive if and
only if its ﬁrst Chern class is positive.
Proof. One direction is clear from the deﬁnition. Suppose, conversely, that c
1
(L) is positive.
That means that there exists a positive (1, 1)–form ω and a Hermitian structure h on L whose
Chern connection ∇ has curvature R
∇
such that [iR
∇
] = [ω] (the factor 2π can obviously be
skipped). From the global i∂
¯
∂–Lemma, there exists a real function u such that iR
∇
= ω + i∂
¯
∂u.
we now use the formula (40) which gives the curvature of the Chern connection in terms of the
square norm of an arbitrary local holomorphic section σ:
R
∇
= −∂
¯
∂ log h(σ, σ).
It is then clear that the curvature of the Chern connection
˜
∇ associated to
˜
h := he
u
satisﬁes for
every local holomorphic section σ:
iR
˜
∇
= −i∂
¯
∂ log
˜
h(σ, σ) = −i∂
¯
∂ log h(σ, σ) −i∂
¯
∂u = iR
∇
−i∂
¯
∂u = ω,
thus showing that L is positive.
In order to get a feeling for this notion, notice that the fundamental form of a K¨ahler manifold is
positive, as well as the Ricci form of a K¨ahler manifold with positive Ricci tensor. From Lemma
11.4 we know that the canonical bundle K of a K¨ahler manifold has curvature iρ. Thus K is
negative if and only if the Ricci tensor is positive deﬁnite.
Theorem 15.3. A negative holomorphic line bundle L over a compact K¨ahler manifold has no
non–vanishing holomorphic section.
Proof. Taking p = 0 and E = L in (63) shows that every holomorphic section ξ of E satisﬁes
(67) 0 = 2
¯
∂
∗
¯
∂ξ = ∇
∗
∇ξ +
i
2
R
∇
(Je
j
, e
j
)ξ.
80
By hypothesis we have iR
∇
(X, Y ) = A(JX, Y ), with A negative deﬁnite. Thus
i
2
R
∇
(Je
j
, e
j
) = −
1
2
A(e
j
, e
j
) = −
1
2
Tr(A)
is a strictly positive function on M. Consequently, taking the Hermitian product with ξ in (67)
and integrating over M shows that ξ has to vanish.
This result is consistent with our previous calculations on CP
m
. We have seen that the canonical
bundle K is negative, and that K is isomorphic to the m + 1
st
tensor power of the tautological
bundle L, which is thus negative, too. On the other hand, we have shown with a direct computation
that this last bundle has no holomorphic section.
15.2. The Hirzebruch–Riemann–Roch formula. Let E → M be a holomorphic vector
bundle over some compact Hermitian manifold M
2m
. We denote by Ω
k
(E) := (
∞
(Λ
0,k
M ⊗ E)
the space of E–valued (0, k)–forms on M. Consider the following elliptic complex
(68) Ω
0
(E)
¯
∂
→Ω
1
(E)
¯
∂
→. . .
¯
∂
→Ω
m
(E).
We deﬁne the cohomology groups
H
q
(M, E) :=
Ker(
¯
∂ : Ω
q
(E) →Ω
q+1
E)
¯
∂Ω
q−1
(E)
.
By analogy with the usual (untwisted) case, we denote
H
p,q
(M, E) := H
q
(M, Λ
p,0
M ⊗E).
For every Hermitian structure on E one can consider the formal adjoint
¯
∂
∗
of
¯
∂, and deﬁne the
space of harmonic E–valued (0, q)–forms on M by
H
q
(E) := ¦ω ∈ Ω
q
(E) [
¯
∂ω = 0,
¯
∂
∗
ω = 0¦.
The analog of the Dolbeault decomposition theorem holds true in this case and as a corollary we
have
Theorem 15.4. The cohomology groups H
q
(M, E) are isomorphic with the spaces of harmonic
E–valued (0, q)–forms:
H
q
(M, E) · H
q
(E).
We can view the elliptic complex (68) as an elliptic ﬁrst order diﬀerential operator simply by
considering
¯
∂ +
¯
∂
∗
: Ω
even
(E) →Ω
odd
(E).
The index of the elliptic complex (68) is deﬁned to be the index of this elliptic operator:
Ind(
¯
∂ +
¯
∂
∗
) := dim(Ker(
¯
∂ +
¯
∂
∗
)) −dim(Coker(
¯
∂ +
¯
∂
∗
)).
The holomorphic Euler characteristic Ξ(M, E) is deﬁned by
Ξ(M, E) :=
m
¸
k=0
(−1)
k
dimH
k
(M, E)
15. THE HIRZEBRUCH–RIEMANN–ROCH FORMULA 81
and is nothing else but the index of the elliptic complex (68).
Theorem 15.5. (Hirzebruch–Riemann–Roch formula) The holomorphic Euler characteristic of
E can be computed as follows
Ξ(M, E) =
M
Td(M)ch(E),
where Td(M) is the Todd class of the tangent bundle of M and ch(E) is the Chern character of
E.
The Todd class and the Chern character are characteristic classes of the corresponding vector
bundles that we will not deﬁne explicitly. The only thing that we will use in the sequel is that
they satisfy the naturality axiom with respect to pull–backs. If E is the trivial line bundle, the
holomorphic Euler characteristic Ξ(M, E) is simply denoted by Ξ(M) :=
¸
m
k=0
(−1)
k
h
0,k
(M).
For a proof of the Hirzebruch–Riemann–Roch formula see [3].
We will give two applications of the Riemann–Roch formula, both concerning the fundamental
group of K¨ahler manifolds under suitable positivity assumptions of the Ricci tensor. The ﬁrst one
is a theorem due to Kobayashi:
Theorem 15.6. A compact K¨ahler manifold with positive deﬁnite Ricci tensor is simply connected.
Proof. Theorem 14.5 shows that there is no holomorphic (p, 0)–form on M, so h
p,0
(M) = 0
for p > 0. Of course, the holomorphic functions are just the constants, so h
0,0
(M) = 1. Since M
is K¨ahler we have h
p,0
(M) = h
0,p
(M), thus Ξ(M) = 1.
By Myers’ Theorem, the fundamental group of M is ﬁnite. Let
˜
M be the universal cover of M,
which is therefore compact, too. Applying the previous argument to
˜
M we get Ξ(
˜
M) = 1. Now,
if π :
˜
M → M denotes the covering projection, we have, by naturality, Td(
˜
M) = π
∗
Td(M), and
an easy exercise shows that for every top degree form ω on M one has
˜
M
π
∗
ω = k
M
ω,
where k denotes the number of sheets of the covering. This shows that k = 1, so M is simply
connected.
Our second application concerns Ricci–ﬂat K¨ahler manifolds. By Theorem 11.5, a compact K¨ahler
manifold M
2m
is Ricci–ﬂat if and only if the restricted holonomy group Hol
0
(M) is a subgroup of
SU
m
. A compact K¨ahler manifold M with Hol
0
(M) = SU
m
is called Calabi–Yau manifold.
Theorem 15.7. Let M
2m
be a Calabi–Yau manifold. If m is odd, then Hol(M) = SU
m
, so there
exists a global holomorphic (m, 0)–form even if M is not simply connected. If m is even, then
either M is simply connected, or π
1
(M) = Z
2
and M carries no global holomorphic (m, 0)–form.
Proof. Let
˜
M be the universal covering of M. The Cheeger–Gromoll theorem (cf. [1], p. 168)
shows that
˜
M is compact (having irreducible holonomy). By Theorem 14.5, every holomorphic
82
form on M is parallel, and thus corresponds to a ﬁxed point of the holonomy representation. It is
easy to check that SU
m
has only two invariant one–dimensional complex subspaces on (p, 0)–forms,
one for p = 0 and one for p = m. Thus
Ξ(
˜
M) =
0 for m odd
2 for m even
Moreover, Ξ(
˜
M) = kΞ(M), where k is the order of the fundamental group of M. This shows that
Ξ(M) = 0 for m odd, hence h
m,0
M = 1, so M has a global holomorphic (m, 0)–form.
If m is even, then either M is simply connected, or k = 2 and Ξ(M) = 1. In this last case, we
necessarily have h
m,0
M = 0, so M carries no global holomorphic (m, 0)–form.
15.3. Exercises.
(1) Prove the Kodaira–Serre duality:
H
q
(M, E) · H
m−q
(M, E
∗
⊗K
M
)
for every holomorphic vector bundle E over a compact Hermitian manifold M.
(2) Prove that the operator
¯
∂ +
¯
∂
∗
: Ω
even
(E) →Ω
odd
(E)
is elliptic, in the sense that its principal symbol applied to any non–zero real 1–form is
an isomorphism.
(3) Prove that the index of the above deﬁned operator is equal to the holomorphic Euler
characteristic Ξ(M, E).
(4) Let π :
˜
M → M be a k–sheet covering projection between compact oriented manifolds.
Prove that for every top degree form ω on M one has
˜
M
π
∗
ω = k
M
ω.
Hint: Start by showing that to any open cover ¦U
i
¦ of M one can associate a closed cover
¦C
j
¦ such that for every j there exists some i with C
j
⊂ U
i
and such that the interiors
of C
j
and C
k
are disjoint for every j = k.
(5) Show that the representation of SU
m
on Λ
p
C
m
has no invariant one–dimensional subspace
for 1 ≤ p ≤ m−1.
16. FURTHER VANISHING RESULTS 83
16. Further vanishing results
16.1. The Schr¨odinger–Lichnerowicz formula for K¨ahler manifolds. Let L be a holo
morphic Hermitian line bundle over some K¨ahler manifold M
2m
with scalar curvature S. We
would like to compute the curvature term in the Weitzenb¨ock formula on sections of Λ
0,k
M ⊗L,
and to show that this term becomes very simple in the case where L is a square root of the canon
ical bundle. The reader familiar with spin geometry will notice that in this case Λ
0,∗
M ⊗ K
1
2
is
just the spin bundle of M and the operator
√
2(∂ +
¯
∂) is just the Dirac operator.
Let us denote by iα the curvature of the Chern connection of L. The ﬁrst term of the curvature
operator 1 applied to some section ω ⊗ξ ∈ Ω
0,k
M ⊗E can be computed as follows
1
1
(ω ⊗ξ) :=
i
2
˜
R(Je
j
, e
j
)(ω ⊗ξ) =
i
2
2(ρ
(k)
ω) ⊗ξ + iα(Je
j
, e
j
)ω ⊗ξ
= i(ρ
(k)
ω) ⊗ξ −
1
2
α(Je
j
, e
j
)ω ⊗ξ.
In order to compute the second curvature term we make use of the following algebraic result
Lemma 16.1. The Riemannian curvature operator satisﬁes
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω = 4iρ
(k)
ω
for every (0, k)–form ω.
Proof. Since the interior product of a (1, 0)–vector and a (0, k)–form vanishes we obtain
(69) X ω = iJX ω ∀ ω ∈ Ω
0,k
M.
The forms R(e
j
, e
k
)ω are still (0, k)–forms, since the connection preserves the type decomposition
of forms. By changing e
j
to Je
j
and then e
k
to Je
k
we get
e
j
∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω = Je
j
∧ (e
k
+ iJe
k
) R(Je
j
, e
k
)ω
= −Je
j
∧ (e
k
+ iJe
k
) R(e
j
, Je
k
)ω
= −iJe
j
∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω.
Thus
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω = 2e
j
∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω = 4e
j
∧ e
k
R(e
j
, e
k
)ω.
Now, using (69) twice we get
R(e
j
, e
k
, e
l
, e
s
)e
j
∧ e
k
∧ e
s
e
l
ω = −R(e
j
, e
k
, Je
l
, Je
s
)e
j
∧ e
k
∧ e
s
e
l
ω
= −R(e
j
, e
k
, e
l
, e
s
)e
j
∧ e
k
∧ e
s
e
l
ω,
so this expression vanishes. From the ﬁrst Bianchi identity we then obtain
R(e
j
, e
k
, e
l
, e
s
)e
j
∧ e
s
∧ e
k
e
l
ω = R(e
j
, e
l
, e
k
, e
s
)e
j
∧ e
s
∧ e
k
e
l
ω
+R(e
j
, e
s
, e
l
, e
k
)e
j
∧ e
s
∧ e
k
e
l
ω
= −R(e
j
, e
l
, e
k
, e
s
)e
j
∧ e
s
∧ e
l
e
k
ω
84
whence
R(e
j
, e
l
, e
k
, e
s
)e
j
∧ e
s
∧ e
k
e
l
ω = 0.
Finally we get
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) R(e
j
, e
k
)ω = 4e
j
∧ e
k
R(e
j
, e
k
)ω
= 4R(e
j
, e
k
, e
l
, e
s
)e
j
∧ e
k
(e
s
∧ e
l
ω)
= −4Ric(e
j
, e
l
)e
j
∧ e
l
ω = 4iRic(e
j
, Je
l
)e
j
∧ e
l
ω
= 4iρ
(k)
ω.
For every (1, 1)–form α and (0, k)–form ω we have as before
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) α(e
j
, e
k
)ω = 2e
j
∧ (e
k
+ iJe
k
) α(e
j
, e
k
)ω = 4e
j
∧ e
k
α(e
j
, e
k
)ω
= −4α
(k)
(ω).
The second term in (62) thus reads
1
2
(ω ⊗ξ) := −
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) (
˜
R(e
j
, e
k
)(ω ⊗ξ))
= −
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
) ((R(e
j
, e
k
)ω) ⊗ξ + iα(e
j
, e
k
)ω ⊗ξ)
= −2iρ
(k)
(ω) ⊗ξ + 2iα
(k)
(ω) ⊗ξ.
Suppose that the curvature of the line bundle L satisﬁes
R
L
:= iα =
1
2
iρ.
The formulas above show that the curvature term in the Weitzenb¨ock formula on Ω
0,k
M ⊗ L
satisﬁes
1(ω ⊗ξ) = (1
1
+1
2
)(ω ⊗ξ) = iρ
(k)
(ω) ⊗ξ −
1
2
α(Je
j
, e
j
)ω ⊗ξ
−2iρ
(k)
(ω) ⊗ξ + 2iα
(k)
(ω) ⊗ξ
= −
1
4
ρ(Je
j
, e
j
)ω ⊗ξ =
S
4
ω ⊗ξ.
This proves the
Theorem 16.2. (Schr¨odinger–Lichnerowicz formula). Let L = K
1
2
be a square root of the canon
ical bundle of a K¨ahler manifold M, in the sense that L has a Hermitian structure h such that
K
M
is isomorphic to L ⊗L with the induced tensor product Hermitian structure. Then, if Ψ is a
section of the complex vector bundle
ΣM := (Λ
0,0
M ⊕. . . ⊕Λ
0,m
M) ⊗K
1
2
16. FURTHER VANISHING RESULTS 85
and D :=
√
2(
¯
∂ +
¯
∂
∗
) is the Dirac operator on ΣM, the following formula holds
D
2
Ψ = ∇
∗
∇Ψ +
S
4
Ψ.
The Schr¨odinger–Lichnerowicz formula is valid in a more general setting (on all spin manifolds,
not necessarily K¨ahler), and it has important applications in geometry and topology (see [4], [9]).
16.2. The Kodaira vanishing theorem. Let M
2m
be a compact K¨ahler manifold and let
L be a positive line bundle over M. From the deﬁnition, we know that L carries a Hermitian
structure whose Chern connection ∇ has curvature R
∇
with iR
∇
> 0. We consider the K¨ahler
metric on M whose fundamental form is just iR
∇
. By a slight abuse of language, we denote
by ∂ : (
∞
(Λ
p,q
M ⊗ L) → (
∞
(Λ
p+1,q
M ⊗ L) the extension of ∇
1,0
to forms. Note that, whilst
¯
∂ is an intrinsic operator, ∂ depends of course on the Hermitian structure on L. We apply the
Weitzenb¨ock formula to some section ω ⊗ξ of Λ
p,q
M ⊗L:
(70) 2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
)(ω ⊗ξ) = ∇
∗
∇(ω ⊗ξ) +1(ω ⊗ξ).
The same computation actually yields the dual formula
(71) 2(∂
∗
∂ + ∂∂
∗
)(ω ⊗ξ) = ∇
∗
∇(ω ⊗ξ) +
˜
1(ω ⊗ξ).
Alternatively, one can apply (70) to a section ˜ ω ⊗ ξ
∗
of Λ
q,p
M ⊗ L
∗
and then take the complex
conjugate. Subtracting these two equations yields
(72) 2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
)(ω ⊗ξ) = 2(∂
∗
∂ + ∂∂
∗
)(ω ⊗ξ) + (1−
˜
1)(ω ⊗ξ).
We now compute this curvature term.
(1−
˜
1)(ω ⊗ξ) = i
˜
R(Je
j
, e
j
)(ω ⊗ξ) −
1
2
(e
j
−iJe
j
) ∧ (e
k
+ iJe
k
)
˜
R(e
j
, e
k
)(ω ⊗ξ)
+
1
2
(e
j
+ iJe
j
) ∧ (e
k
−iJe
k
)
˜
R(e
j
, e
k
)(ω ⊗ξ)
= i
˜
R(Je
j
, e
j
)(ω ⊗ξ) + iJe
j
∧ e
k
˜
R(e
j
, e
k
)(ω ⊗ξ)
−ie
j
∧ Je
k
˜
R(e
j
, e
k
)(ω ⊗ξ)
= i
˜
R(Je
j
, e
j
)(ω ⊗ξ) + 2iJe
j
∧ e
k
˜
R(e
j
, e
k
)(ω ⊗ξ)
= 2iρ(ω) ⊗ξ + iω ⊗R
∇
(Je
j
, e
j
)ξ + 2iJe
j
∧ e
k
R(e
j
, e
k
)ω ⊗ξ
+2iJe
j
∧ e
k
ω ⊗R
∇
(e
j
, e
k
)ξ
= 2iρ(ω) ⊗ξ −2mω ⊗ξ + 2iJe
j
∧ e
k
R(e
j
, e
k
)ω ⊗ξ + 2(p + q)ω ⊗ξ.
On the other hand, the expression Je
j
∧ e
k
R(e
j
, e
k
)ω can be simpliﬁed as follows:
Je
j
∧ e
k
R(e
j
, e
k
)ω = Je
j
∧ e
k
R(e
j
, e
k
)e
l
∧ e
l
ω
= −Ric(e
j
, e
l
)Je
j
∧ e
l
ω −Je
j
∧ R(e
j
, e
k
)e
l
∧ e
k
e
l
ω
= −ρ(ω) −R(e
j
, e
k
, e
l
, e
s
)Je
j
∧ e
s
∧ e
k
e
l
ω,
86
and from the Bianchi identity
2R(e
j
, e
k
, e
l
, e
s
)Je
j
∧ e
s
∧ e
k
e
l
ω = R(e
j
, e
k
, e
l
, e
s
)Je
j
∧ e
s
∧ e
k
e
l
ω
+R(e
j
, e
l
, e
k
, e
s
)Je
j
∧ e
s
∧ e
l
e
k
ω
= R(e
l
, e
k
, e
j
, e
s
)Je
j
∧ e
s
∧ e
k
e
l
ω
= −R(e
l
, e
k
, Je
j
, e
s
)e
j
∧ e
s
∧ e
k
e
l
ω = 0,
where the last expression vanishes because R(, , J, ) is symmetric in the last two arguments.
This shows that Je
j
∧ e
k
R(e
j
, e
k
)ω = −ρ(ω), so from the previous calculation we get
2(
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
)(ω ⊗ξ) = 2(∂
∗
∂ + ∂∂
∗
)(ω ⊗ξ) + 2(p + q −m)(ω ⊗ξ).
After taking the Hermitian product with ω⊗ξ (which we denote by σ for simplicity) and integrating
over M we get
(73)
M
[
¯
∂σ[
2
+[
¯
∂
∗
σ[
2
dv =
M
[∂σ[
2
+[∂
∗
σ[
2
+ (p + q −m)[σ[
2
dv.
If σ is a harmonic L–valued form, the left hand side term in (73) vanishes, thus proving the
Theorem 16.3. (Kodaira vanishing theorem). If L is a positive holomorphic line bundle on a
compact K¨ahler manifold M, one has H
p,q
(M, L) = 0 whenever p + q > m.
Part 6
Calabi–Yau manifolds
17. Ricci–ﬂat K¨ahler metrics
17.1. Hyperk¨ahler manifolds. In order to obtain the classiﬁcation (up to ﬁnite coverings)
of compact Ricci–ﬂat K¨ahler manifolds, we make the following
Definition 17.1. A Riemannian manifold (M
n
, g) is called hyperk¨ahler if there exist three com
plex structures I, J, K on M satisfying K = IJ such that g is a K¨ahler metric with respect to
each of these complex structures.
It is clear that a metric is hyperk¨ahler if and only if it is K¨ ahler with respect to two anti–commuting
complex structures. In the irreducible case, this can be weakened as follows:
Proposition 17.2. Let (M
n
, g) be a locally irreducible Riemannian manifold. If g is K¨ahler with
respect to two complex structures J and J
1
, and if J
1
is diﬀerent from J and −J, then (M, g) is
hyperk¨ahler.
Proof. The endomorphism JJ
1
+J
1
J is symmetric and parallel on M, so by local irreducibility
it has to be constant:
(74) JJ
1
+ J
1
J = αId, α ∈ R.
From the Cauchy–Schwartz inequality we get
α
2
= [αId[
2
= [JJ
1
+ J
1
J[
2
≤ 2([JJ
1
[
2
+[J
1
J[
2
) ≤ 4[J[
2
[J
1
[
2
= 4,
where the norm considered here is the operator norm. The equality case can only hold if JJ
1
=
βJ
1
J for some real number β. Together with (74) this shows that JJ
1
= γId for some real number
γ, so J
1
= ±J, which was excluded in the hypothesis. Therefore we have α
2
< 4. We then
compute using (74)
(J
1
+ JJ
1
J)
2
= (α
2
−4)Id,
so the parallel skew–symmetric endomorphism
I :=
1
√
4 −α
2
(J
1
+ JJ
1
J)
deﬁnes a complex structure anti–commuting with J, with respect to which g is K¨ahler.
Consider the identiﬁcation of C
2k
with H
k
given by (z
1
, z
2
) → z
1
+ jz
2
. We denote by I, J
and K the right product on H
k
with i, j and k respectively, which correspond to the following
endomorphisms of C
2k
:
I(z
1
, z
2
) = (iz
1
, iz
2
) J(z
1
, z
2
) = (−¯ z
2
, ¯ z
1
) K(z
1
, z
2
) = (−i¯ z
2
, i¯ z
1
).
Let us denote by Sp
k
the group of unitary transformations of C
2k
(that is, preserving the canonical
Hermitian product and commuting with I), which also commute with J (and thus also with K).
Clearly we have
Sp
k
=
M =
A B
−
¯
B
¯
A
∈ ´
2k
(C)
M
¯
M
t
= I
2k
.
It is tautological that a 4k–dimensional manifold is hyperk¨ahler if and only if the bundle of
orthonormal frames has a reduction to Sp
k
.
17. RICCI–FLAT K
¨
AHLER METRICS 89
Lemma 17.3. Sp
k
⊂ SU
2k
.
Proof. By deﬁnition we have Sp
k
⊂ U
2k
, so every matrix in Sp
k
is diagonalizable as complex
matrix and its eigenvalues are complex numbers of unit norm. If v is an eigenvector of some
M ∈ Sp
k
with eigenvalue λ ∈ S
1
then
MJv = JMv = Jλv =
¯
λJv = λ
−1
Jv,
showing that the determinant of M equals 1.
This shows that every hyperk¨ahler manifold is Ricci–ﬂat. A hyperk¨ahler manifold is called strict
if it is locally irreducible.
Let now M be an arbitrary compact Ricci–ﬂat K¨ahler manifold. The Cheeger–Gromoll theorem
([1], p.168) says that M is isomorphic to a quotient
M · (M
0
T
l
)/Γ,
where M
0
is a compact simply connected K¨ahler manifold, T
l
is a complex torus and Γ is a ﬁnite
group of holomorphic transformations. Let M
0
= M
1
. . . M
s
be the De Rham decomposition
of M
0
. Then M
j
are compact Ricci–ﬂat simply connected K¨ahler manifolds with irreducible
holonomy for all j. A symmetric space which is Ricci–ﬂat is automatically ﬂat, so the M
j
’s are
not symmetric. The Berger holonomy theorem then shows that M
j
is either Calabi–Yau or strict
hyperk¨ahler for every j. We thus have the following
Theorem 17.4. A compact Ricci–ﬂat K¨ahler manifold M is isomorphic to the quotient
M · (M
1
. . . M
s
M
s+1
. . . M
r
T
l
)/Γ,
where M
j
are simply connected compact Calabi–Yau manifolds for j ≤ s, simply connected com
pact strict hyperk¨ahler manifolds for s + 1 ≤ j ≤ r and Γ is a ﬁnite group of holomorphic
transformations.
17.2. Projective manifolds. A compact complex manifold (M
2m
, J) is called projective if it
can be holomorphically embedded in some complex projective space CP
N
. A well–known result of
Chow states that a projective manifold is algebraic, that is, deﬁned by a ﬁnite set of homogeneous
polynomials in the complex projective space.
Proposition 17.5. Every projective manifold has a positive holomorphic line bundle.
Proof. Let ϕ be the fundamental form of the Fubini–Study metric on CP
N
. It is easy to
check (e.g. using (22)) that the hyperplane bundle H on CP
N
has a connection with curvature
−iϕ. The restriction of this line bundle to any complex submanifold of CP
N
is thus positive.
Conversely, we have the celebrated
Theorem 17.6. (Kodaira embedding theorem). A compact complex manifold M with a positive
holomorphic line bundle L is projective.
90
A proof can be found in [2], p. 176. The main idea is to show that a suitable positive power L
k
of L has a basis of holomorphic sections ¦σ
0
, . . . , σ
N
¦ such that the holomorphic mapping
M →CP
N
x →[σ
0
(x) : . . . : σ
N
(x)]
is an embedding.
Corollary 17.7. Every Calabi–Yau manifold of complex dimension m ≥ 3 is projective.
Proof. For every compact manifold M, let / and /
∗
be the sheaves of smooth functions on
M with values in C and C
∗
respectively. The exact sequence of sheaves
0 →Z →/
exp
→ /
∗
→0
induces an exact sequence in
˘
Cech cohomology
→H
1
(M, /) →H
1
(M, /
∗
)
c
1
→H
2
(M, Z) →H
2
(M, /) →.
The sheaf / is ﬁne (that is, it admits a partition of unity), so H
1
(M, /) = 0 and H
2
(M, /) = 0,
thus proving that
H
1
(M, /
∗
) · H
2
(M, Z).
Notice that H
1
(M, /
∗
) is just the set of equivalence classes of complex line bundles over M, and
the isomorphism above is given by the ﬁrst Chern class.
The above argument shows that for every integer cohomology class γ ∈ H
2
(M, Z), there exists a
complex line bundle L with c
1
(L) = γ. Moreover, if ω is any complex 2–form representing γ in
real cohomology, there exists a connection ∇ on L such that
i
2π
R
∇
= ω. To see this, take any
connection
˜
∇ on L with curvature R
˜
∇
. Then since [ω] = c
1
(L) we get [2πω] = [iR
˜
∇
], so there
exists some 1–form θ such that 2πω = i(R
˜
∇
+ dθ). Clearly the curvature of ∇ :=
˜
∇+ iθ satisﬁes
the desired equation. If the form ω is real and of type (1, 1), then the complex bundle L has a
holomorphic structure, given by the (0, 1)–part of the connection whose curvature is ω.
Let now M
2m
be a Calabi–Yau manifold, m > 2. Since SU
m
has no ﬁxed point on Λ
2,0
C
m
, we
deduce that there are no parallel (2, 0)–forms on M, so by Theorem 14.5 we get h
2,0
(M) = 0. By
the Dolbeault decomposition theorem we obtain that any harmonic 2–form on M is of type (1, 1).
Consider the fundamental form ϕ of M. Since H
2
(M, Q) is dense in H
2
(M, R), and the space of
positive harmonic (1, 1)–forms is open in H
1,1
(M, R) = H
2
(M, R), we can ﬁnd a positive harmonic
(1, 1)–form ω such that [ω] ∈ H
2
(M, Q). By multiplying with the common denominator, we may
suppose that [ω] ∈ H
2
(M, Z). Then the argument above shows that there exists a holomorphic
line bundle whose ﬁrst Chern class is ω, thus a positive holomorphic line bundle on M. By the
Kodaira embedding theorem, M is then projective.
18. CONSTRUCTIONS OF CALABI–YAU MANIFOLDS 91
18. Constructions of Calabi–Yau manifolds
18.1. Divisors. Let M be a complex manifold. An analytic hypersurface of M is a subset
V ∈ M such that for every x ∈ V there exists an open set U
x
⊂ M containing x and a holomorphic
function f
x
deﬁned on U
x
such that V ∩U
x
is the zero–set of f
x
. Such an f
x
is called a local deﬁning
function for V near x. The quotient of any two local deﬁning functions around x is a non–vanishing
holomorphic function around x.
An analytic hypersurface V is called irreducible if it can not be written as the union of two smaller
analytic hypersurfaces. Every analytic hypersurface is a ﬁnite union of its irreducible components.
If V is an irreducible analytic hypersurface, with deﬁning function ϕ
x
around some x ∈ V , then
for every holomorphic function f around x, the order of f along V at x is deﬁned to be the largest
positive integer a such that
f
ϕ
a
x
is holomorphic around x. It can be shown that the order of f is a
well–deﬁned positive integer, which does not depend on x, and is denoted by o(f, V ).
Definition 18.1. A divisor D in a compact complex manifold M is a ﬁnite formal sum with
integer coeﬃcients of irreducible analytic hypersurfaces of M.
D :=
¸
i
a
i
V
i
, a
i
∈ Z.
A divisor D is called eﬀective if all a
i
≥ 0 for all i.
The set of divisors is clearly a commutative group under formal sums.
A meromorphic function on a complex manifold M is an equivalence class of collections (U
α
, f
α
, g
α
)
where ¦U
α
¦ is an open covering of M, and f
α
, g
α
are holomorphic functions deﬁned on U
α
such
that f
α
g
β
= f
β
g
α
on U
α
∩ U
β
for all α, β. Two such collections (U
α
, f
α
, g
α
) and (U
β
, f
β
, g
β
) are
equivalent if f
α
g
β
= f
β
g
α
on U
α
∩U
β
for all α, β. A meromorphic function can be always expressed
locally as
f
g
, where f and g are locally deﬁned holomorphic functions.
We deﬁne similarly a meromorphic section of a holomorphic line bundle L as an equivalence
class of collections (U
α
, σ
α
, g
α
) where σ
α
is a local holomorphic section of L over U
α
and g
α
is a
holomorphic function on U
α
, such that σ
α
g
β
= σ
β
g
α
on U
α
∩ U
β
for all α, β.
A meromorphic function h deﬁnes a divisor (h) in a canonical way by
(h) := (h)
0
−(h)
∞
,
where (h)
0
and (h)
∞
denote the zero–locus (resp. the pole–locus) of h taken with multiplicities.
More precisely, for every x in M, one can write the function h as h =
fx
gx
near x. If V is
an irreducible analytic hypersurface containing x, we deﬁne the order of h along V at x to be
o(f
x
, V ) −o(g
x
, V ), and this is a well–deﬁned integer independent on x, denoted by o(h, V ). Then
(h) =
¸
V
o(h, V )V,
92
where the above sum is ﬁnite since for every open set U
x
where h =
fx
gx
, there are only ﬁnitely
many irreducible analytic hypersurfaces along which f
x
of g
x
have non–vanishing order.
Similarly, if σ is a global meromorphic section of a line bundle L, one can deﬁne the order o(σ, V )
of σ along any irreducible analytic hypersurface V using local trivializations of L. This clearly
does not depend on the chosen trivialization, since the transition maps do not vanish, so they do
not contribute to the order. As before, one deﬁnes a divisor (σ) on M by
(σ) =
¸
V
o(σ, V )V.
If D =
¸
a
i
V
i
and f
i
are local deﬁning functions for V
i
near some x ∈ M (of course we can take
f
i
= 1 if V
i
does not contain x), then the meromorphic function
¸
f
a
i
i
is called a local deﬁning function for D around x.
Definition 18.2. Two divisors D and D
are called linearly equivalent if there exists some mero
morphic function h such that
D = D
+ (h).
In this case we write D ≡ D
.
Clearly two meromorphic sections σ and σ
of L deﬁne linearly equivalent divisors (σ) = (σ
)+(h),
where h is the meromorphic function deﬁned by σ = σ
h.
18.2. Line bundles and divisors. To any divisor D we will associate a holomorphic line
bundle [D] on M in the following way. Take an open covering U
α
of M and local deﬁning
meromorphic functions h
α
for D deﬁned on U
α
. We deﬁne [D] to be the holomorphic line bundle on
M with transition functions g
αβ
:=
hα
h
β
. It is easy to check that g
αβ
are non–vanishing holomorphic
functions on U
α
∩ U
β
satisfying the cocycle conditions, and that the equivalence class of [D] does
not depend on the local deﬁning functions h
α
.
Example. Let H denote the hyperplane ¦z
0
= 0¦ in CP
m
and consider the usual open covering
U
α
= ¦z
α
= 0¦ of CP
m
. Then 1 is a local deﬁning function for H on U
0
and
z
0
zα
are local deﬁning
functions on U
α
. The line bundle [H] has thus transition functions g
αβ
=
z
β
zα
, which are exactly
the transition function of the hyperplane line bundle introduced in Section 3, which justiﬁes its
denomination.
If D and D
are divisors, then clearly [−D] = [D]
−1
and [D + D
] = [D] ⊗ [D
]. We call Div(M)
the group of divisors on M, and Pic(M) := H
1
(M, O) the Picard group of equivalence classes
of holomorphic line bundles (where O denotes the sheaf of holomorphic functions). Then the
arguments above show that there exists a group homomorphism
[ ] : Div(M) →Pic(M) D →[D].
Notice that the line bundle associated to a divisor (h) is trivial for every meromorphic function
h. This follows directly from the deﬁnition: for any open cover U
α
on M, h[
Uα
is a local deﬁning
18. CONSTRUCTIONS OF CALABI–YAU MANIFOLDS 93
function for the divisor (h) on U
α
, so the transition functions for the line bundle [(h)] are equal
to 1 on any intersection U
α
∩ U
β
. Thus [ ] descends to a group homomorphism
[ ] : Div(M)/
≡
−→Pic(M).
Suppose now that [D] = 0 for some divisor D on M. That means that the line bundle [D] is trivial,
so there exists an open cover ¦U
α
¦ of M and holomorphic non–vanishing functions f
α
: U
α
→C
∗
such that
f
α
f
b
= g
αβ
=
h
α
h
b
on U
α
∩ U
β
,
where h
α
is a local deﬁning meromorphic function for D on U
α
. This shows the existence of a
global meromorphic function H on M such that H[
Uα
=
hα
fα
. Moreover, as f
α
does not vanish on
U
α
, the divisor associated to H is just D. This proves the injectivity of [ ] on isomorphism classes
of divisors.
Every holomorphic line bundle of a projective manifold has a global meromorphic section (see [2]
p.161). If L ∈ Pic(M) is a holomorphic line bundle, we have seen that a global meromorphic
section σ of L deﬁnes a divisor (σ) on M. We claim that [(σ)] = L. If g
αβ
denote the transition
functions of L with respect to some trivialization (U
α
, ψ
α
), the meromorphic section σ deﬁnes
meromorphic functions σ
α
on U
α
such that g
αβ
=
σα
σ
β
. From the deﬁnition, σ
α
is a deﬁning
meromorphic section for (σ) on U
α
, thus L is just the line bundle associated to (σ). We have
proved the
Theorem 18.3. If the manifold M is projective, the homomorphism [ ] descends to an isomorphism
Div(M)/
≡
∼
=
−→Pic(M).
18.3. Adjunction formulas. Let V ⊂ M be a smooth complex hypersurface of a compact
complex manifold M. We will show that the normal and co–normal bundles of V in M can be
computed in terms of the divisor V .
Proposition 18.4. (First adjunction formula) The restriction to V of the line bundle [V ] asso
ciated to the divisor V is isomorphic to the holomorphic normal bundle of V in M:
N
V
= [V ][
V
.
Proof. Let i : V → M be the inclusion of V into M. By deﬁnition, the normal bundle N
V
is the co–kernel of the inclusion i
∗
: T
1,0
V →T
1,0
M[
V
and its dual, the co–normal bundle N
∗
V
, is
deﬁned as the kernel of the projection i
∗
: Λ
1,0
M[
V
→Λ
1,0
V . Thus N
∗
V
is spanned by holomorphic
(1, 0)–forms on M vanishing on V .
Let f
α
be local deﬁning functions for V on some open covering U
α
. By deﬁnition, the quotients
g
αβ
:=
fα
f
β
are the transition functions of [V ] on U
α
∩ U
β
. Moreover, since f
α
vanishes along V
which is smooth, we see that df
α
[
V
is a non–vanishing local section of N
∗
V
. Now, since f
α
= g
αβ
f
β
,
we get
df
α
[
V
= (f
β
dg
αβ
+ g
αβ
df
β
)[
V
= g
αβ
[
V
df
β
[
V
.
94
Thus the collection (U
α
, df
α
[
V
) deﬁnes a global holomorphic section of N
∗
V
⊗[V ][
V
, showing that this
tensor product bundle is trivial. This proves that N
∗
V
= [−V ][
V
and consequently N
V
= [V ][
V
.
Consider now the exact ﬁbre bundle sequence
0 →N
∗
V
→Λ
1,0
M[
V
→Λ
1,0
V →0.
Taking the maximal exterior power in this exact sequence yields
K
M
[
V
· K
V
⊗N
∗
V
= K
V
−[V ][
V
,
so
K
V
· (K
M
⊗[V ])[
V
.
This is the second adjunction formula.
We will use the following theorem whose proof, based on the Kodaira vanishing theorem, can be
found in [2], p. 156.
Theorem 18.5. (Lefschetz Hyperplane Theorem). Let V be a smooth analytic hypersurface in
a compact complex manifold M
2m
such that [V ] is positive. Then the linear maps H
i
(M, C) →
H
i
(V, C) induced by the inclusion V → M are isomorphisms for i ≤ m − 2 and injective for
i = m−1. If m ≥ 3 then π
1
(M) = π
1
(V ).
Our main application will be the following result on complete intersections in the complex pro
jective space.
Theorem 18.6. Let P
1
, P
k
be homogeneous irreducible relatively prime polynomials in m + 1
variables of degrees d
1
, . . . d
k
. Let N denote the subset in CP
m
deﬁned by these polynomials:
N := ¦[z
0
: . . . : z
m
] ∈ CP
m
[ P
i
(z
0
, . . . , z
m
) = 0, ∀1 ≤ i ≤ k¦.
Then, if N is smooth, we have K
N
· [qH][
N
, where q = (d
1
+ . . . + d
k
) − (m + 1) and H is the
hyperplane divisor in CP
m
.
Proof. Notice ﬁrst that N is smooth for a generic choice of the polynomials P
i
. We denote
by V
i
the analytic hypersurface in CP
m
deﬁned by P
i
and claim that
(75) V
i
∼
= d
i
H.
This can be seen as follows. While the homogeneous polynomial P
i
is not a well–deﬁned function
on CP
m
, the quotient h
i
:=
P
i
z
d
i
0
is a meromorphic function. More precisely, h
i
is deﬁned by
the collection (U
α
,
P
i
z
d
i
α
,
z
d
i
0
z
d
i
α
). Clearly the zero–locus of h
i
is (h
i
)
0
= V
i
and the pole–locus is
(h
i
)
∞
= d
i
H
0
, where H
0
is just the hyperplane ¦z
0
= 0¦. This shows that (h
i
) = V
i
−d
i
H
0
, thus
proving our claim.
Let now, for i = 1, . . . , k, N
i
denote the intersection of V
1
, . . . , V
i
. Since N
i+1
= N
i
∩V
i+1
, we have
(76) [N
i+1
][
N
i
= [d
i+1
H][
N
i
.
18. CONSTRUCTIONS OF CALABI–YAU MANIFOLDS 95
This follows from the fact that if V is an irreducible hypersurface in a projective manifold M and
N is any analytic submanifold in M then
[V ][
N
· [V ∩ N].
We claim that
(77) K
N
i
· [n
i
H][
N
i
,
where n
i
:= (d
1
+ . . . + d
i
) − (m + 1). For i = 1 this follows directly from the second adjunction
formula together with (75), using the fact that K
CP
m
= [−(m + 1)H]. Suppose that the formula
holds for some i ≥ 1. The second adjunction formula applied to the hypersurface N
i+1
of N
i
,
together with (76) yields
K
N
i+1
= ([N
i+1
] ⊗K
N
i
)
N
i+1
= ([d
i+1
H] ⊗[n
i
H])[
N
i+1
= [n
i+1
H][
N
i+1
.
Thus (77) is true for every i, and in particular for i = k. This ﬁnishes the proof.
Corollary 18.7. Let d
1
, . . . , d
k
be positive integers and denote their sum by m+1 := d
1
+. . .+d
k
.
Suppose that m ≥ k + 3. If P
1
, P
k
are generic homogeneous irreducible polynomials in m + 1
variables of degrees d
1
, . . . d
k
, then the manifold
N := ¦[z
0
: . . . : z
m
] ∈ CP
m
[ P
i
(z
0
, . . . , z
m
) = 0, ∀1 ≤ i ≤ k¦
carries a unique (up to rescaling) Ricci–ﬂat K¨ahler metric compatible with the complex structure
induced from CP
m
. Endowed with this metric, N is Calabi–Yau.
Proof. Theorem 18.6 shows that the ﬁrst Chern class of N vanishes. The condition m ≥ k+3
together with Lefschetz Hyperplane Theorem applied inductively to the analytic hypersurfaces
N
i
⊂ N
i+1
show that N is simply connected and b
2
(N) = b
2
(CP
m
) = 1, and moreover the
restriction of the K¨ahler form of CP
m
to N is a generator of the second cohomology group of N.
The Calabi conjecture shows that there exists a unique Ricci–ﬂat metric on N up to rescaling.
If this metric were reducible, we would have at least two independent elements in the second
cohomology of N, deﬁned by the K¨ahler forms of the two factors. Since b
1
(N) = 1 this is
impossible. Thus N is either Calabi–Yau or hyperk¨ahler. The latter case is however impossible,
since every compact hyperk¨ahler manifold has a parallel (2, 0)–form, thus its second Betti number
cannot be equal to 1.
Bibliography
[1] A. Besse, Einstein Manifolds, Ergeb. Math. Grenzgeb., 10, Springer, Berlin, 1981.
[2] P. Griﬃth, J. Harris, Principles of Algebraic Geometry, Wiley, New York, 1978.
[3] F. Hirzebruch, Topological methods in algebraic geometry, Springer, New York, 1966.
[4] M. Gromov, B. Lawson, Jr., The classiﬁcation of simply connected manifolds of positive scalar curvature,
Ann. of Math. 111 (1980), 423–434.
[5] L. H¨ormander, An introduction to complex analysis in several variables, third edition, NorthHolland Pub
lishing Co., Amsterdam, 1990.
[6] D. Joyce, Compact Manifolds with Special Holonomy, Oxford University Press, 2000.
[7] E. K¨ahler,
¨
Uber eine bemerkenswerte Hermitesche Metrik, Abh. Math. Sem. Hamburg Univ. 9 (1933), 173–186.
[8] S. Kobayashi, K. Nomizu, Foundations of Diﬀerential Geometry I, II, Interscience Publishers, 1963, 1969.
[9] A. Lichnerowicz, Spineurs harmoniques, C. R. Acad. Sci. Paris 257 (1963), 7–9.
97
Current version March 16, 2004.
Contents
Introduction Part 1. 2. 3. Part 4. 5. 6. 7. 1. Complex geometry Complex structures and holomorphic maps Holomorphic forms and vector ﬁelds Complex and holomorphic vector bundles 2. Hermitian and K¨hler structures a Hermitian bundles Hermitian and K¨hler metrics a The curvature tensor of K¨hler manifolds a Examples of K¨hler metrics a 4 5 6 12 17 23 24 27 32 37 43 44 49 55 56 61 65 69 73 74 79 83 87 88 91 97
Part 3. The Laplace operator 8. Natural operators on Riemannian and K¨hler manifolds a 9. Hodge and Dolbeault theory Part 4. Prescribing the Ricci tensor on K¨hler manifolds a 10. Chern classes 11. The Ricci form of K¨hler manifolds a 12. The Calabi conjecture 13. K¨hler–Einstein metrics a Part 5. Vanishing results 14. Weitzenb¨ck techniques o 15. The Hirzebruch–Riemann–Roch formula 16. Further vanishing results Part 6. Calabi–Yau manifolds 17. Ricci–ﬂat K¨hler metrics a 18. Constructions of Calabi–Yau manifolds Bibliography
3
4 CONTENTS Introduction These notes. This background material is well covered in the classical literature. In the ﬁrst part I quickly introduce complex manifolds. in particular. . 14. Part 4 is devoted to the Calabi a conjecture and in Part 5 I obtain several vanishing results using Weitzenb¨ck techniques. the Hirzebruch–Riemann–Roch formula. the Calabi conjecture. based on a graduate course I gave at Hamburg University in 2003. • Riemannian metrics. Levi–Civita connexion. diﬀerent aspects of Calabi–Yau manifolds are studied using techniques from algebraic geometry. the Cheeger–Gromoll theorem and the Kodaira embedding theorem. I assume. are intended to students having basic knowledges of diﬀerential geometry. The text is self–contained with a few notable exceptions – the Newlander– Nirenberg theorem. They should hopefully be useful for graduate students in mathematics a and theoretical physics. • principal ﬁbre bundles. the Hodge theorem. that the reader is familiar with following topics: • diﬀerential manifolds. Ch. de Rham decomposition theorem. tensors. holonomy groups. I considered that including the proofs of these results would have add too much technicality in the presentation and would have doubled the volume of the text without bringing essentially new insights to our objects of interest. and can be found for instance in [8]. The text is organized as follows. Most of the a remaining material concerns compact manifolds. The main purpose of these notes is to provide a quick and accessible introduction to diﬀerent aspects of K¨hler geometry. o in Part 6. Finally. and in Part 2 I deﬁne K¨hler manifolds from the point of view of Riemannian geometry. vector bundles. Most of the sections end up with a series of exercises whose levels of diﬃculty range from low to medium. and give a simple way of deriving the famous K¨hler identities. In Part 3 I review Hodge and Dolbeault theories. Killing vector ﬁelds. Lie groups. connexions.
Part 1 Complex geometry .
every projective manifold. denotes the Hermitian metric ds2 = Re( dzi d¯i ). Complex structures and holomorphic maps 1. a hαβ dzα ∧ d¯β . Condition a) requires the real dimension of M to be even. • analytic conditions d) the 2–form Ω is symplectic: dΩ = 0. where .1. which will be detailed later. b) g(X. they also have compatible symplectic and complex structures. submanifold of CPm deﬁned by homogeneous polynomials in Cm+1 . z ¯ . a a Given any Hermitian metric h on a complex manifold. Y ∈ T M. a Besides the Riemannian structure. Here are a few examples of K¨hler manifolds: a • • • • (Cm . e) J is integrable in the sense that its Nijenhuis tensor vanishes (see (4) below). K¨hler manifolds may be considered as special Riemannian manifolds. Preliminaries. Y ) = g(JX. F S) endowed with the FubiniStudy metric. Y ). A K¨hler structure on a Riemannian manifold (M n . Obviously. the complex projective space (CPm . . c) Ω(X. The function u is called K¨hler potential. given the metric and one of the tensors J and Ω. we can immediately recover the other one by the formula c). z any oriented 2–dimensional Riemannian manifold. g) is given by a 2–form Ω a and a ﬁeld of endomorphisms of the tangent bundle J satisfying the following • algebraic conditions a) J is an almost complex structure: J 2 = −Id. ). and “a long list of miracles occur then”. Y ) = g(JX. Definition 1. hαβ = ¯ ∂zα ∂ zβ ¯ hαβ := h ¯ In other words. ∂zα ∂ zβ ¯ He then noticed that the condition dΩ = 0 is equivalent to the local existence of some function u such that ∂2u . K¨hler structures were introduced by Erich K¨hler in his article [7] with the following motivation. we can express the fundamental two–form Ω in local holomorphic coordinates as follows: Ω=i where ∂ ∂ .1. .1. We give here a short deﬁnition. JY ) ∀ X. that is. the whole metric tensor is deﬁned by a unique function! This remarkable (bemerkenswert) property of the metric allows one to obtain simple explicit expressions for the Christoﬀel symbols and the Ricci and curvature tensors.
. The endomorphism j can be expressed in the canonical base as 0 −1 j= . xm . Algebraic Geometry or Theoretical Physics.2.1. . With two exceptions (the so–called Joyce manifolds in dimensions 7 and 8). . y). ym ). . also called Calabi–Yau manifolds. on a complex manifold with Hermitian metric. a Complex Analysis. curiously. . the existence of normal holomorphic coordinates around each point is equivalent to the metric being K¨hler! a K¨hler manifolds have found many applications in various domains like Diﬀerential Geometry. . Recall that every point x in a Riemannian manifold has a local coordinate system xi such that the metric osculates to the Euclidean metric to the order 2 at x. zm ) = (x1 + iy1 . A function F : U ⊂ Cn → Cm is then holomorphic if and only if the diﬀerential F∗ of F as real map F : R2n → R2m satisﬁes jm F∗ = F∗ jn . will be studied later on in these notes. These special coordinate systems are the normal coordinates around each point. K¨hler himself did not a a seem to have noticed. Generic Ricci–ﬂat K¨hler a manifolds. xm + iym ) → (x1 . y1. we identify Cm with R2m via and denote by jm the endomorphism of R2m corresponding to the multiplication by i on Cm : jm = 0 −Im . Similarly. . . . . which. . COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 7 There is another remarkable property of K¨hler metrics. F∗ = . the only known compact examples of manifolds satisfying Einstein’s equations Rαβ = 0 a (Ricci–ﬂat in modern language) are constructed on K¨hler manifolds. To illustrate their importance let us make the following remark. A function F = f + ig : U ⊂ C → C is called holomorphic if it satisﬁes the Cauchy–Riemann equations: ∂f ∂g = ∂x ∂y and ∂f ∂g + = 0. 1. . 1 0 The diﬀerential of F (viewed as real function F : U ⊂ R2 → R2 ) is of course the linear map ∂f ∂f ∂x ∂g ∂x ∂y ∂g ∂y Then it is easy to check that the Cauchy–Riemann relations are equivalent to the commutation relation jF∗ = F∗ j. . . Im 0 (z1 . . . Now. Holomorphic functions. . ∂y ∂x Let j denote the endomorphism of R2 corresponding to the multiplication by i on C via the identiﬁcation of R2 with C given by z = x + iy → (x.
V ∈ U.... . m of CPm deﬁned by and the maps φi : Ui → Cm . Consider the open cover Ui . wi wi wi wi wi wi wi which is obviously holomorphic on its domain of deﬁnition. A complex manifold of complex dimension m is a topological space M with an open covering U such that for every point x ∈ M there exists U ∈ U containing ˜ x and a homeomorphism φU : U → U ⊂ Cm . . . . . then CPm = Cm+1 − {0}/ ∼ . thus showing that JU does not depend on U and their collection is a well–deﬁned tensor J on M.. If we deﬁne the equivalence relation ∼ on Cm+1 − {0} by (z0 . A function F : M → C is called holomorphic if F ◦ φ−1 is holomorphic for every U ∈ U. : zm ]) = It is then an easy exercise to compute φi ◦ φ−1 (w1 ..8 1. such that for every intersecting U. φU ) is called a chart and the collection of all charts is called a holomorphic structure. . . . This tensor clearly satisﬁes J 2 = −Id. . Important example.. .. U so JV (X) = (φV )−1 ◦ jn ◦ (φV )∗ (X) = (φV )−1 ◦ jn ◦ (φV U )∗ ◦ (φU )∗ (X) ∗ ∗ −1 = (φV )∗ ◦ (φV U )∗ ◦ jn ◦ (φU )∗ (X) = (φU )−1 ◦ jn ◦ (φU )∗ (X) ∗ = JU (X). ∗ If we take some other V ∈ U containing x. . This U property is local. To check it in the neighborhood of a point x it is enough to check it for a single U ∈ U containing x. . zi zi zi zi φi ([z0 : . . : zm ]. .. . i = 0. A pair (U. . . . .. and φV = φV U ◦φU . . zm ) will be denoted by [z0 : . wm ) = j wi−1 wi+1 wj 1 wj+1 wm w1 ... .. Complex manifolds. . . . then φV U = φV ◦φ−1 is holomorphic... . . .. αzm ).3. The equivalence class of (z0 . The most important object on a complex manifold from the diﬀerential geometric point of view is the almost complex structure J. choose U ∈ U containing x and deﬁne JU (X) = (φU )−1 ◦ jn ◦ (φU )∗ (X).. ∀ α ∈ C∗ . : zm ]  zi = 0} zi−1 zi+1 zm z0 . . For every X ∈ Tx M. . . . . The complex projective space CPm can be deﬁned as the set of complex lines of Cm+1 (a line is a vector subspace of dimension one). . the map m between open sets of C φU V := φU ◦ φ−1 V is holomorphic. zm ) ∼ (αz0 .. . Ui := {[z0 : . which is a ﬁeld of endomorphisms of the tangent bundle deﬁned as follows. ...
1 M = {X + iJX  X ∈ T M}. We will only prove here the “only if” part.2.1 M) denote the eigenbundle of T M C corresponding to the eigenvalue i (resp. ∂zα 2 ∂xα ∂yα ∂ 1 ∂ ∂ := +i . We extend all real endomorphisms and diﬀerential operators from T M to T M C by C–linearity.1 M.0 M (resp. Let now Z and W be two local .1)–tensor J on a diﬀerential manifold M satisfying J 2 = −Id is called an almost complex structure.4.4.1. If {e1 . . A (1.0 M and T 0. Consider a local chart (U.3. T 0. The pair (M. COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 9 Definition 1. T 1. J) be an almost complex manifold. The complexiﬁed tangent bundle. Deﬁne T M C := T M ⊗R C. ∗ ∂yα ∂ . φU ) and let zα = xα + iyα be the α–th component of φU . Proof. we have to complexify the tangent space. −i) of J. e2m } denotes the standard basis of R2m . They form moreover a local basis in each point of U. A complex manifold is thus in a canonical way an almost complex manifold.1 M respecz ¯ tively. Let (M. ∂ zα ¯ 2 ∂xα ∂yα J ∂ ∂xα = ∂ From (1) we obtain immediately that ∂zα and ∂∂α are local sections of T 1. In order to do so. Suppose that J comes from a holomorphic structure on M. The following algebraic lemma is an easy exercise. The converse is only true under some integrability condition (see Theorem 1. Let (M. jm (eα ) = em+α . . The famous Newlander–Nirenberg theorem can be stated as follows: Theorem 1. ∂yα Moreover. J) is then referred to as almost complex manifold. . J) be an almost complex manifold. We would like to diagonalize the endomorphism J. Lemma 1. Let T 1. One has and T M C = T 1. T 0. The almost complex structure J comes from a holomorphic structure if and only if the distribution T 0. The interested reader can ﬁnd the proof of the hard part for example in [5]. 1. so we obtain directly from the deﬁnition (1) We now make the following notations 1 ∂ ∂ ∂ := −i .0 M ⊕ T 0.0 M = {X − iJX  X ∈ T M} . we have by deﬁnition: ∂ = (φU )−1 (eα ) ∗ ∂xα and ∂ = (φU )−1 (em+α ). .4 below).1 M is integrable.
1 M. ∂vα ∂uα We then have m m ∂uβ ∂ ∂vβ ∂ ∂ (2) = + ∂xα ∂xα ∂uβ β=1 ∂xα ∂vβ β=1 and (3) ∂ = ∂yα m β=1 ∂uβ ∂ ∂vβ ∂ + . An almost complex structure arising from a holomorphic structure is called a complex structure. 1. (1) Prove Lemma 1. satisfying ∂ ∂ =J .3.β=1 ¯ ¯ ∂ zα ∂ zβ ¯ ¯ α. Compute the product Im 0 −iIm Im A B −B A Im 0 iIm Im and use this computation to prove that for every invertible matrix A + iB ∈ Glm (C). vα is another local system of coordinates. W ] = Zα − Wα .1 M and write Z = Zα ∂∂α . The existence of local coordinates satisfying (1) is actually the key point of the hard part of the theorem. ∂ zα ∂ zβ α.10 sections of T 0. Remark. (2) Let A + iB ∈ Glm (C).β=1 which is clearly a local section of T 0. A direct calculation then gives z ¯ n ∂Wβ ∂ ∂Zβ ∂ [Z. (3) Show that every almost complex manifold is orientable. the determinant of the real 2m × 2m matrix A B −B A is strictly positive. Exercises. W = z ¯ n Wα ∂∂α . ∂yα ∂uβ β=1 ∂yα ∂vβ m Applying J to (2) and comparing to (3) yields ∂vβ ∂uβ ∂vβ ∂uβ = and =− .5. Once we have such coordinates. it is easy to show that the transition functions are holomorphic: suppose that uα . ∂xα ∂yα ∂yα ∂xα thus showing that the transition functions are holomorphic. .
. Let Γ be the subgroup of Glm (C) generated by αIm . COMPLEX STRUCTURES AND HOLOMORPHIC MAPS 11 (4) Let α > 1 be some real number. Use this to prove that S 1 × S 2m−1 carries a complex structure.1. Show that Γ acts freely and properly discontinuously on Cm − {0}.
. . To every almost complex structure J one can associate a (2. a k–form is a section of Λp. . JY ] − [JX. Λ0.0 M}. Decomposition of the (complexiﬁed) exterior bundle.0M if and only if Z ω = 0 for all Z ∈ T 0.1 M. The following statements are equivalent: N J (X. . .1 M. i=0 Using Lemma 2.1 M} C Λ0.0 ⊗Λ0. i1 < . dzm } and {d¯1 .0 M or to q + 1 vectors from T 0. j1 < . .1.1 ) by Λk.1)–tensor N J called the Nijenhuis tensor. . The sections of Λ∗ M can be viewed as complex–valued forms C C or as formal sums ω + iτ .0) or forms of type (0. The exterior power of a direct sum of vector spaces can be described as follows Λk (E ⊕ F ) ⊕k Λi E ⊗ Λk−i F.0 M = {ω − iω ◦ J  ω ∈ Λ1 M} . Similarly to Lemma 1. C Λ1.3 we have Lemma 2. Y ∈ C ∞ (T M). C p.q M.0 0. . < ip .1 Then {dz1 .q denote the tensor product Λp.q Sections of Λ M are called forms of type (p. It is easy to check that a complex–valued k–form ω is a section of Λk.1 M. JY ]. q). Y ] + J[X.1. Let us denote the k–th exterior power of Λ1. and a z z local basis for Λp. We now turn our attention to exterior forms and introduce the complexiﬁed exterior bundle Λ∗ M = Λ∗ M ⊗R C. . Λ0. We deﬁne the following two sub–bundles of Λ1 M: C and Λ1.1 M = {ω + iω ◦ J  ω ∈ Λ1 M}.0 M ⊕ Λ0.q M if and only if it vanishes whenever applied to p + 1 vectors from T 1. ∧ d¯jq .q . Extending the exterior derivative on functions by C–linearity we get complex–valued forms dzα = dxα + idyα and d¯α = dxα − idyα .0 (resp. . Holomorphic forms and vector ﬁelds 2.1) respectively. . Let J be an almost complex structure on M 2m . . z 1.1 M := {ξ ∈ Λ1 M  ξ(Z) = 0 ∀ Z ∈ T 1. where ω and τ are usual real forms on M. Y ) = [X.1 we then get Λk M ⊕p+q=k Λp.k ) and let Λp.2. Let zα = xα + iyα be the α–th coordinate of some φU . Y ] + J[JX.q M is given by z z {dzi1 ∧ . ∀ X.0 M := {ξ ∈ Λ1 M  ξ(Z) = 0 ∀ Z ∈ T 0. C The sections of these sub–bundles are called forms of type (1. . < jq }. Λ0. .0 (resp. J) be an almost complex manifold. . . d¯m } are local basis for Λ M and Λ M respectively. . we can describe these spaces in terms of a local coordinate system. Let (M. satisfying (4) Proposition 2. If J is a complex structure. One has and Λ1 M = Λ1. More generally. ∧ dzip ∧ d¯j1 ∧ .12 2.
q M. W ) = 0 ∀ Z. (2) ⇐⇒ (3).1 M). which is equivalent to f∗ ◦ (φU )−1 ◦ jm = if∗ ◦ (φU )∗ . φU ). .1 M). Proof. Y ∈ C ∞ (T M) be local vector ﬁelds and let Z denote the bracket Z := [X + iJX.1 M. df ∈ Λ1.1 M.1 M.q M ⊕ Λp. ¯ (b) ⇐⇒ (e) Let X. .q+1M) ∀ 0 ≤ p. W ) = 0 ∀ Z. N J = 0. which proves that T 0. Y + iJY ]. W ) = Z(ω(W )) − W (ω(Z)) − ω([Z. q ≤ m. (2) Z(f ) = 0 ∀ Z ∈ T 0. This last equation just means that for every real vector X.2 M ⊕ Λ1. W ]). We start with the following characterization of holomorphic functions. It is then enough to apply Leibniz’ rule to any section of Λp. Y ). hence idf (X + iJX) = 0 ∀ X ∈ T M.1 M.1 M. W ∈ T 0. (a) ⇐⇒ (b) is given by Theorem 1. . Holomorphic objects on complex manifolds. Extend Z and W to local sections of T 0. ¯ ¯ where ωi ∈ C ∞ (Λ1. df (JX) = idf (X). W ∈ T 0. (b) ⇐⇒ (c) Let ω be a section of Λ1.1 M) ⊂ C ∞ (Λ0. W ]) = −ω([Z. W ] ∈ T 0. HOLOMORPHIC FORMS AND VECTOR FIELDS 13 (a) (b) (c) (d) (e) J is a complex structure.0 M [Z.3. ∧ ωp ∧ τ1 ∧ .0 M) ⊂ C ∞ (Λ2. Thus Z ∈ T 0.0 M. The Λ0. W ∈ T 0.2 M–component of dω vanishes if and only if dω(Z. The function f is holomorphic if and only if f ◦ φU is holomorphic for every −1 holomorphic chart (U.0 M ⊕ Λ1.1 M is integrable if and only if N J ≡ 0 2. dC ∞ (Λ1. . Y ) − iJN J (X. By conjugation we get immediately dC ∞ (Λ0.0 M.1 M ⇐⇒ N J (X.1 M ∀ Z.1 M is integrable. J) will denote a complex manifold of complex dimension m.4. W ]) = 0 ∀ Z. Let f : M → C be a smooth complex–valued function on M. (c) ⇐⇒ (d) One implication is obvious. (3) df is a form of type (1. The following assertions are equivalent: (1) f is holomorphic.0 M) and τi ∈ C ∞ (Λ0.0 M ω([Z. which is equivalent to df ∈ Λ1.q M) ⊂ C ∞ (Λp+1. An easy calculation gives Z − iJZ = N J (X.1 M). 0). . locally written as a sum of decomposable elements ω1 ∧ .2.2. ∀ ω ∈ Λ1. ∀ ω ∈ Λ1.1 −1 (1) ⇐⇒ (3). In this section (M. Suppose now that (c) holds.0 M ⇐⇒ df (Z) = 0 ∀ Z ∈ T 0. f∗ ◦ ∗ J = if∗ . 0.1 M and write dω(Z. W ∈ T 0. Thus ⇐⇒ ⇐⇒ dω(Z. dC ∞ (Λp. Lemma 2. that is. T 0. Y ) = 0.1 M ⇐⇒ Z(f ) = 0 ∀ Z ∈ T M. Proof. ∧ τq .
JY ]−J[X. We have 0 = d2 = (∂ + ∂)2 = ∂ 2 + ∂ 2 + (∂ ∂ + ∂∂). X] = J[Y. One implication is clear from Lemma 2.q M) and ∂ : C ∞ (Λp. the reader might have guessed that a map f : (M. This implies [Y + iJY. we ﬁrst notice that a complex vector ﬁeld Z is of type (0. A real vector ﬁeld X is called real holomorphic if X − iJX is a holomorphic vector ﬁeld.1) if and only if Z(f ) = 0 for every locally deﬁned holomorphic function f . Lemma 2. J2 ) between two complex manifolds is called holomorphic if its diﬀerential commutes with the complex structures at each point: f∗ ◦ J1 = J2 ◦ f∗ . The following identities hold: ¯ ∂ 2 = 0. Y ] = 0 for all vector ﬁelds Y .1 M ∩ Λ2 M be a real 2–form of type (1. X](f ) = 0.q+1M) by d = ∂ + ∂. Proof. As (X + iJX)f = 0. we have (X − iJX)f = 2X(f ).14 Using Proposition 2. LX J = 0. Let ω ∈ Λ1.3 we get (Y + iJY )(X(f )) = 0 and (Y + iJY )(f ) = 0. . ∂ 2 = 0. The converse is similar and left to the reader.4. ¯ ¯ ¯ ¯ Proof. We close this section with the following important result: ¯ Proposition 2. and the three operators in the last term take values in diﬀerent sub–bundles. Proof. Definition 2.7.8.0 M) is called holomorphic if Z(f ) is holomorphic for every locally deﬁned holomorphic function f . Suppose that X is real holomorphic and let Y be an arbitrary vector ﬁeld and f a local holomorphic function. This holds for every holomorphic f so [Y +iJY. 0) is called holomorphic if ¯ ∂ω = 0. that is [JY.1). Then ω is closed if and only if every point x ∈ M has an open ¯ neighborhood U such that ωU = i∂ ∂u for some real function u on U.4: ¯ ¯ ¯ ¯ ¯ d(i∂ ∂) = i(∂ + ∂)∂ ∂ = i(∂ 2 ∂ − ∂ ∂ 2 ) = 0.1) on a complex manifold M. By deﬁnition X(f ) is then holomorphic so by Lemma 2. Let X be a real vector ﬁeld on a complex manifold (M. Lemma 2.q M) → C ∞ (Λp. J1 ) → (N.e. i. ¯ ¯ ∂ ∂ + ∂∂ = 0.5. (The local i∂ ∂–Lemma). A p–form ω of type (p. X] has to be of type (0. q) we deﬁne the diﬀerential operators ∂ : C ∞ (Λp. A vector ﬁeld Z in C ∞ (T 1.q M) → ¯ ¯ C ∞ (Λp+1. for every ﬁxed (p. J). In order to prove the equivalence of the ﬁrst two assertions. Definition 2.6. Hence LX J(Y ) = LX (JY )−J(LX Y ) = [X. Although not explicitly stated. The equivalence of the last two assertions is tautological.2. The following assertions are equivalent: • X is real holomorphic • LX J = 0 • The ﬂow of X consists of holomorphic transformations of M. X].
0 M at each point of the local coordinate system.1 ¯ .0) ω = ωα dzα is holomorphic if and only if ωα are holomorphic functions. Y ∈ T M.1 +∂τ 1. (2) Prove that the object deﬁned by formula (4) is indeed a tensor. hence ω = (∂τ 0. Clearly.7. 2. (3) Show that a almost complex structure on a real 2–dimensional manifold is always integrable.0 M and T 1. 25 for a proof): ¯ ¯ ¯ Lemma 2. ∂–Poincar´ Lemma. .3. Let τ = τ + τ be the decomposition of τ in forms of type (1. ∂ • Prove that a local vector ﬁeld of type (1. ∀ X. From the Poincar´ Lemma.1). there exists locally a 1–form e 1.1 . (1) Prove Lemma 2. ¯ ¯ ¯ we get ∂τ 0.1 + ∂τ 1.9.1) if and only if ω(X.0) and (0. (5) Show that a 2–form ω is of type (1. prove that JX has the same property.0) Z = Zα ∂zα is holomorphic if and only if Zα are holomorphic functions.1). Exercises. τ 1. ¯ ¯ ¯ ¯ ¯ ¯ τ = ∂f and the Proposition follows.1 = 0 and ω = (∂τ 0. e Let ω be a closed real form of type (1. A ∂–closed (0. • Prove that a local form of type (1.1)–form is locally ∂–exact. (7) If X is a real holomorphic vector ﬁeld on a complex manifold.1 + ∂τ 1.1.0 ) = ∂ ∂f + ∂∂ f = i∂ ∂(2Im(f )). By conjugation we get τ 1.0 ) + ∂τ 1.0 ).0 0. Comparing types in the equality ¯ ¯ ω = dτ = ∂τ 0.2. JY ).1 τ with dτ = ω. (8) Prove the converse in Lemma 2. Y ) = ω(JX. (6) Let M be a complex manifold with local holomorphic coordinates {zα }. ∂ (4) Show that {dzα } and { ∂zα } are dual basis of Λ1.0 .0 = ∂ f . with u := 2Im(f ).0 = τ 0. The ∂–Poincar´ Lemma yields a local function f such that e 0. HOLOMORPHIC FORMS AND VECTOR FIELDS 15 The other implication is more delicate and needs the following counterpart of the Poincar´ Lemma e (see [2] p.1 + (∂τ 0.
0 N with respect to J. . Show that there exists a unique almost complex structure J on N such that C T = Λ1. z ∂ zα ¯ (10) Let N be a manifold. and let T be a complex sub–bundle of Λ1 N such that T ⊕ Λ1 N = C Λ1 N.16 (9) Show that in every local coordinate system one has n ∂f = α=1 ∂f dzα ∂zα n and ¯ ∂f = α=1 ∂f d¯α .
Holomorphic structures.2.q (E) is given by σ = (ω1 . ¯ ¯ ∂ 2 = 0.q+1E) on a complex vector bundle E satisfying the Leibniz rule is called a pseudo–holomorphic structure. . . . For every holomorphic bundle E one deﬁnes the bundles Λp. take a holomorphic atlas (U. If. where gU V : U ∩V → 2 Glk (C) ⊂ Ck are holomorphic functions.. . ¯ 3. we deﬁne ∂σ := (∂ω1 . The transition functions gU V = (φU )∗ ◦ (φV )−1 are then clearly holomorphic. . . .3. Complex and holomorphic vector bundles 3. COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 17 3. Let M be a complex manifold and let π : E → M be a complex vector bundle over M (i. Suppose that σ is written σ = (τ1 . .0M and the chain rule ∂zαp ∂zα1 ··· dwβ1 ∧ . φU ) on M and deﬁne ψU : T MU → U × Cm by ψU (Xx ) = (x. Notice that the bundles Λp.βp shows that the transition functions are holomorphic. Then one has τj = gjk ωk for some holomorphic functions gjk . More precisely. gU V (x)v).0M are holomorphic.q E) → C ∞ (Λp. If a section σ of Λp. By construction ¯ ¯ one has ∂ 2 = 0 and ∂ satisﬁes the Leibniz rule: ¯ ¯ ¯ ∂(ω ∧ σ) = (∂ω) ∧ σ + (−1)p+q ω ∧ (∂σ). Examples. each ﬁber π −1 (x) is a complex vector space). q)–forms). (φU )∗ (X)). moreover. v) = (x. An operator ∂ : C ∞ (Λp.q M). . thus ¯ ¯ ¯ ∂τj = gjk ∂ωk . ∗ 2. . The cotangent bundle.q+1E) in the following way. . Holomorphic vector bundles. E is called holomorphic vector bundle if there exists a trivialization with holomorphic transition functions. The tangent bundle of a complex manifold M 2m is holomorphic. To see this.1.. . and more generally the bundles Λp.. .q M are not in general holomorphic bundles for q = 0.q E := Λp. ∧ dwβp dzα1 ∧ . Indeed. ∂ωk ). . ∀ ω ∈ C ∞ (Λp. ∧ dzαp = ∂wβ1 ∂wβp β1 .s E). q) and the ∂–operator ∂ : C ∞ (Λp. . showing that ∂σ does not depend on the chosen trivialization. 1. then ∂ is called a holomorphic structure. . there exists an open cover U of M and for each U ∈ U a diﬀeomorphism ψU : π −1 (U) → U × Ck such that • the following diagram commutes: π −1 U π ψU G ÓÓ ÓÓ Ó ÓÓ ÓÓ prU ÓÓ ÓÓ ÑÓÓ U × Ck U −1 • for every intersecting U and V one has ψU ◦ψV (x.q E) → C ∞ (Λp.e. using again a holomorphic atlas of the manifold one can trivialize locally Λp. ωk ) in some local trivialization (where ωi are local ¯ ¯ ¯ (p.q M ⊗ E of E–valued forms ¯ ¯ on M of type (p. . σ ∈ C ∞ (Λr. τk ) in some other trivialization of E.
hence k (5) ¯ ∂τij = l=1 τil ∧ τlj . . Proof.1. and if {σi } and {˜i } are two such ¯ holomorphic basis. . j ≤ k. j ≤ k. Theorem 3. Suppose.18 ¯ ¯ A section σ in a pseudo–holomorphic vector bundle (E. A complex vector bundle E is holomorphic if and only if it has a holomorphic ¯ structure ∂. ψU ) −1 a local basis of holomorphic sections by σi (x) := ψU (x. . ∀ x ∈ U. If E is holomorphic. we can write σi = gij σj . Suppose one can ﬁnd a map f : U → Glk (C). A pseudo–holomorphic vector bundle (E. In order to show that E is holomorphic. f = (fij ) such that (6) ¯ 0 = ∂fij + fil τlj . The “only if” part follows directly from the discussion above.l=1 τil ∧ τlj ⊗ σj . that one can trivialize E around each x ∈ M by holomorphic sections. one can deﬁne for every local holomorphic trivialization (U. σk } be local sections of E which form a basis of E over some open set U containing x. ¯ that E is a complex bundle over M of rank k with holomorphic structure ∂ satisfying Leibniz’ ¯2 = 0. It is then easy to check that the local sections sj of E over U deﬁned by sj := fjl σl are holomorphic: ¯ ¯ ∂sj = ∂fjl ⊗ σl + fjr τrl ⊗ σl = 0.1. Proof. ¯ Lemma 3. . hence the transition functions are holomorphic. We deﬁne local (0.1)–forms τij on U by the formula k ¯ ∂σi = j=1 τij ⊗ σj . every local σ basis of holomorphic sections deﬁnes a local trivialization of E. ∀ 1 ≤ i. ∀ 1 ≤ i. using Lemma rule and ∂ 3. conversely.2. yields k k ¯ 0 = ∂ 2 σi = j=1 ¯ ∂τij ⊗ σj − j. ∂) of rank k is holomorphic if and only if each x ∈ M has an open neighborhood U and k holomorphic sections σi of E over U such that {σi (x)} form a basis of Ex (and hence on some neighborhood of x). The theorem thus follows from the next lemma. Conversely. together with Leibniz’ rule. for some open subset U of U containing x. it is enough to show. . ¯ The condition ∂ 2 = 0. ei ). From now on we will use the summation convention on repeating indexes. ∂) is called holomorphic if ∂σ = 0. Let {σ1 . which immediately yields (applying ∂ and using ˜ ¯ Leibniz’ rule) that ∂gij = 0.
1 ≤ i ≤ k}. Clearly d(dzα ) = 0 and from (5) we get ¯ d(dwi − τil wl ) = −∂τil wl − ∂τil wl + τil ∧ dwl = −∂τil wl − τis ∧ τsl wl + τis ∧ dws = −∂τil wl + τis ∧ (dws − τsl wl ). 0) is just ∂flk . to show that its integrability is equivalent to (5).1 1. We may suppose that U is an open subset of Cm with holomorphic coordinates zα and denote by wi the coordinates in Ck . It is enough to check this on the local basis C deﬁning T . 1 ≤ i. 0)τik ∧ dwk + dFlα ∧ dzα . By Proposition 2. and get (7) 0 = dFlk (z. We denote by N the product U × Ck . ¯ If we denote flk (z) := Flk (z. It is an easy exercise to check that any complement T of Λ1 N in the complexiﬁed bundle Λ1 N of C some manifold N 2n . The main idea is to deﬁne an almost complex structure locally on U × Ck using τ .1 U –part of dFlk (z. 1 ≤ α ≤ m such that dul = Fli (dwi − τik wk ) + Flα dzα . dwi − τil wl  1 ≤ α ≤ m. We evaluate this last equality for wi = 0. Therefore.3. ul } on some smaller neighborhood U of x. then the Λ0. We claim that the almost complex structure induced on N by T is integrable. 0). and ﬁnally to obtain f as the matrix of some frame deﬁned by τ in terms of holomorphic coordinates given by the theorem of Newlander–Nirenberg. which clearly is a section of C ∞ (T ∧ Λ1 N). 0) ∧ dwk + Fli (z. deﬁnes an almost complex structure on N. or equivalently such that (6) holds. .3. We apply the exterior derivative to this system and get 0 = dFli ∧ (dwi − τik wk ) + Fli (−dτik wk + τik ∧ dwk ) + dFlα ∧ dzα . or equivalently (5). with iT = T .2. f = (fij ) such that ∂f + f τ = 0. We now use the Newlander–Nirenberg theorem and C complete the family {zα } to a local holomorphic coordinate system {zα . l ≤ k. Proof.1)–form on U satisfying ∂τ = τ ∧ τ .0 k vanishing of the Λ U ⊗ Λ C –components of (7) just reads ¯ 0 = ∂flk + fli τik . Consider the sub–bundle T of Λ1 N ⊗ C generated by the 1–forms {dzα . such that T becomes the space of (1. Then for every x ∈ U there exists some open subset U of U containing x and a ¯ map f : U → Glk (C).0)–forms on N. Suppose that τ := (τij ) is a glk (C)–valued (0. we have to show that dC ∞ (T ) ⊂ C ∞ (T ∧ Λ1 N). COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 19 ¯ Lemma 3. the 0. Since dul are sections of T . we can ﬁnd functions Fli and Flα .
∧ dam = (−1)α−β am+1 db0 ∧ . φα ) on CPm and the local trivializations ψα : π −1 Uα → Uα × C of L deﬁned by ψα ([z]. λ) = ([z]. where gαβ ([z]) = . . ∧ dbβ−1 ∧ dbβ+1 ∧ . ∧ dam . Proof.0 Cm . We already noticed that KM has a holomorphic structure. . which proves that β KCPm Lm+1 . . The canonical bundle of CPm . ∧ daα−1 ∧ daα+1 ∧ . β we have ai = bi aβ and aβ bα = 1. . α so the transition functions are hαβ := (φ∗ )−1 ◦ (φ∗ ). . . . ∧ daα−1 ∧ daα+1 ∧ . ∧ dwm be the α β zi canonical generator of Λm. . for every i = α. . . . β Using (8) we thus see that the transition functions are given by hαβ = (−1)α−β a−m−1 = (−1)α−β β zα zβ m+1 . Let now ω := dw1 ∧ . m} − {β}. . . β and daβ = − b1 dbα = −a2 dbα . . On the other hand. β db0 ∧ . ∧ dbβ−1 ∧ dbβ+1 ∧ . . m+1 Finally. .3. It is an easy exercise to compute the transition functions: zα −1 ψα ◦ ψβ ([z]. For a complex manifold (M 2m . . J). and an easy algebraic computation 2 β α then yields da0 ∧ . . ∧ dbm = hαβ da0 ∧ . . This shows that dai = aβ dbi + bi daβ for i = α. . ∧ dam α and Therefore we can write (8) φ∗ (ω) = db0 ∧ . .4. gαβ ([z])λ).0 Cm . A trivialization for p : Λm. We consider the canonical holomorphic charts (Uα . . w) = ([z].20 3. It is deﬁned as the complex line bundle π : L → CPm whose ﬁber L[z] over some point [z] ∈ CPm is the complex line < z > in Cm+1 . The aim of this subsection is to prove the following Proposition 3. . . . . ∧ dbm . . On the complex projective space there is some distinguished holomorphic line bundle called the tautological line bundle. wα ). . We introduce holomorphic coordinates on Uα ∩ Uβ : ai := zα for zi i ∈ {0. . . the complex line bundle KM := Λm. . zβ which are clearly holomorphic. Then φ∗ (ω) = da0 ∧ . The canonical bundle of CPm is isomorphic to the m + 1st power of the tautological bundle. ∧ dbm . ∧ daα−1 ∧ daα+1 ∧ . . m} − {α} and bi := zβ for i ∈ {0.0 M is called the canonical bundle of M. . ∧ dbβ−1 ∧ dbβ+1 ∧ . denoting cα := (−1)α we have cα hαβ c−1 = gαβ .0 CPm → CPm is given by (φ∗ )−1 : p−1 Uα → Uα × Λm.
Find local trivializations for H with holomorphic transition functions. such that the projection π is a holomorphic map. Let L be the complex line bundle π : L → CPm whose ﬁber L[z] over some point [z] ∈ CPm is the complex line < z > in Cm+1 . Thus. Hint: use the maximum principle. wα ). Prove that E is a holomorphic line bundle. Hint: Use the local trivializations ψα : π −1 Uα → Uα × C deﬁned by ψα ([z]. (1) Prove that any holomorphic function on a compact manifold f : M → C is constant. the ﬁber of H over some point [z] ∈ CPm is the set of linear maps < z >→ C. Find the dimension of the space of holomorphic sections of H. (2) Let E → M be a rank k complex vector bundle whose transition functions with respect to some open cover {Uα } of M are gαβ . (3) Let πE → M be a complex vector bundle over a complex manifold M. COMPLEX AND HOLOMORPHIC VECTOR BUNDLES 21 3. (4) The tautological line bundle. (5) Show that the tautological line bundle L has no non–trivial holomorphic sections. Prove that E has a holomorphic structure if and only if there exists a complex structure on E as manifold.4. w) = ([z]. Let H := L∗ be the dual of L. Exercises.3. . Show that a section σ : M → E of E can be identiﬁed with a collection {σα } of smooth maps σα : Uα → Ck satisfying σα = gαβ σβ on Uα ∩ Uβ . (6) The hyperplane line bundle.
.
Part 2 Hermitian and K¨hler structures a .
• H(u. Definition 4. More explicitly. u) ∀ u. . that is.2. A (C–linear) connection on a E is a C–linear diﬀerential operator C ∞ (Λ1 (E)) satisfying the Leibniz rule (f σ) = df ⊗ σ + f σ. Connections on complex vector bundles. . .1. ∀ f ∈ C ∞ (M). for every x ∈ M. R (σi ) = Rij ⊗ σj . . Let M be a diﬀerentiable manifold (not necessarily complex) and let E → M be a complex vector bundle over M. if {σ1 . Hermitian bundles 4. i The curvature of R (σ) := 2 is the End(E)–valued 2–form R deﬁned by ( σ)) ∀ σ ∈ C ∞ (E). v) = H(v. Definition 4. . we can write: (f σ) = (df ⊗ σ + f σ) = d2 f ⊗ σ − df ∧ σ=f 2 σ. Hermitian structures and connections. (ωij ⊗ σj ) = (dωij ) ⊗ σj − ωik ∧ ωkj ⊗ σj . H : Ex × Ex → C satisﬁes • H(u. 4. we deﬁne the connection forms ωij ∈ Λ1 (U) (relative to the choice of the base) by We deﬁne the two–forms Rij by and compute showing that (9) Rij ⊗ σj = R (σi ) = σi = ωij ⊗ σj . Rij = dωij − ωik ∧ ωkj . v ∈ Ex . v) is C–linear in u for every v ∈ Ex . Let E → M be a complex rank k bundle over some manifold M.2. σ. We do not assume for the moment that M has an almost complex structure. σk } are local sections of E which form a basis of each ﬁber over some open set U.1. σ + df ∧ σ+f 2 To check that this is indeed tensorial.4. A Hermitian structure H on E is a smooth ﬁeld of Hermitian products on the ﬁbers of E. One can extend any connection to the bundles of E–valued p–forms on M by where the wedge product has to be understood as n : C ∞ (E) → (ω ⊗ σ) = dω ⊗ σ + (−1)p ω ∧ ω∧ σ := i=1 ω ∧ e∗ ⊗ i ei σ for any local basis {ei } of T M with dual basis {e∗ }.
(ω ⊗ σ) = ∂ω ⊗ σ + (−1)p+q ω ∧ 0.2. Consider the projections π 1.1 2 ) (σ) = ( 1. v of E.1 : C ∞ (Λp.1 : Λ1 (E) → Λ0. is parallel with respect to . For every section σ of E one can write R (σ) = so the Λ 0.0 : Λ1 (E) → Λ1. For every Hermitian structure H in a holomorphic bundle E with holomorphic ¯ structure ∂. v) is a smooth function on M for every smooth sections u. then it holds on E ∗ .q (E)) → C ∞ (Λp. too. From (10). part of the curvature is given by (R )0. • H(u. on E by the formula (10) )(σ) := X(σ ∗ (σ)) − σ ∗ ( X σ). HERMITIAN BUNDLES 25 • H(u.1 ¯ (ω ⊗ σ) = ∂ω ⊗ σ + (−1)p+q ω ∧ 1. σ ∈ C ∞ (E). and that any connection on E induces canonically a connection. To see this.0 σ.1 is a pseudo–holomorphic structure on E for σ=( 1. u) > 0 ∀ u = 0. one can consider its (1. Let us ﬁrst remark that the dual vector bundle E ∗ is also holomorphic. It is clear from the above conditions that H is C–anti–linear in the second variable.1 .q+1(E)) satisfying the Leibniz rule: 1.4.0 0. We say that is a H–connection if H. ψi ) of E and a partition of the unity fi subordinate to the open cover {Ui } of M. Proof.0 := π 1. ¯ then E is a holomorphic bundle with holomorphic structure ∂ := 0. it is quite useful to think to H as to a C–anti–linear isomorphism H : E → E ∗ . Suppose now that M is a complex manifold. for all ω ∈ C ∞ (Λp. In fact. there exists a unique H–connection (called the Chern connection) such that 0. Then H := fi Hi is a well–deﬁned Hermitian structure on E.0 (E) and π 0. We can now state the main result of this section: Theorem 4.q (E)) and 0.1 1. ∗ also denoted by . The converse is also true: simply choose an arbitrary Hermitian metric on E and apply Theorem 4.1 := π 0.0 (E)) for every holomorphic section σ of E. every connection .2 2 0.1 (E).0 ◦ and 0.0 : C ∞ (Λp.1 2 ) (σ) + ( 1. viewed as a ﬁeld of C–valued real bilinear forms on E. Of course.1 = ¯ ∂. we can extend these operators to 1.2 = ( ). For every connection on E.0 0.1 = ∂ on E just means that σ ∈ C ∞ (Λ1.3 below.0 + 0.0 )(σ).1 2 0.2 shows that if the Λ0. σ ∗ ∈ C ∞ (E ∗ ). if this property holds on E. let (Hi)x denote the pull–back of the Hermitian metric on Ck by the C–linear map ψi Ex .2 –part of the curvature of some connection D on E vanishes.1)–components 1. Every rank k complex vector bundle E admits Hermitian structures.0 2 ) (σ) + ( 0. From Proposition 2.1 σ.1 + 0. just take a trivialization (Ui . Theorem 3. ∀ X ∈ T M.3.q (E)) → C ∞ (Λp+1.0) and (0. For every x ∈ Ui . σ ∈ C ∞ (E).1 ◦ . The third condition shows that H is non–degenerate. ¯ Note also that 0. with holomor¯ phic structure denoted by ∂.q M). ( Xσ ∗ .
1 (σ)) = ∂ 2 (σ) = 0. suppose that is a H–connection with 0.3. (3) Let E → M be a rank k complex vector bundle. too. By the C–anti–linearity of H. The C–anti–linear ∗ isomorphism H : E → E is then parallel. this shows that (11) 1. . ¯ (1) Let E → M be a complex vector bundle and denote by E ∗ and E its dual and its ∗ conjugate. . (4) Prove the naturality of the Chern connection with respect to direct sums and tensor products of holomorphic vector bundles. Exercises. Its (2. for every complex vector Z ∈ T M C we have Z (H(σ)) = H( ¯ Z σ). 4. Remark.2 (σ) = 0. (Recall that for every x ∈ M.0 hence ¯ ¯ = ∂ + H −1 ◦ ∂ ◦ H. the ﬁbre of E over x is just the dual of Ex and ¯ ¯ the ﬁbre Ex of E is equal to Ex as a set.0)–component actually vanishes. since by (11). For Z ∈ T 1. Viewing local trivializations of E as local basis of sections of E. show that if the transition functions of E with respect to some local trivialization take values in the unitary group Uk ⊂ Glk (C) then there exists a canonically deﬁned Hermitian structure on E. so for every section σ of E and every real vector X on M we get X (H(σ)) = X (H)(σ) + H( X σ) = H( X σ).0 (H −1 ◦ ∂(H(σ)) = H −1 ◦ ∂ 2 (H(σ)) = 0.0 M. ¯ R0. but has the conjugate complex structure. which proves the existence and uniqueness of (σ) = H −1 ◦ 0.26 ¯ After these preliminaries. (2) Show that a Hermitian structure on a complex vector bundle E deﬁnes an isomorphism ¯ between E ∗ and E as complex vector bundles.0 ¯ ¯ ( 1.1 ( 0.2)–component of the curvature of the Chern connection vanishes. If gαβ denote the transition functions of E with respect to some open cover {Uα } of ¯ M. The (0. 1. in the ¯ sense that the action of some complex number z on Ex is the same as the action of z on ¯ Ex ).1 ¯ (H(σ)) = H −1 ◦ ∂(H(σ)).1 = ∂. ﬁnd the transition functions of E ∗ and E with respect to the same open cover.0 (σ)) = 1. Indeed.
each symmetric tensor on T M C with these properties deﬁnes a Hermitian metric by restriction to T M (exercise).1. W ) = h(Z. The fundamental form of a Hermitian metric is deﬁned by Ω(X. ∀ Z. Simply choose an arbitrary Riemannian metric g and deﬁne h(X. From the i∂ ∂–Lemma we get locally a real function u such that Ω = i∂ ∂u. Remark. J). Y ) := h(X. Y ) deﬁnes a Hermitian structure on the complex vector bundle (T M. J) is a Riemannian metric h such that h(X. Conversely. Hermitian and K¨hler metrics a 5. Hermitian metrics. ∀ X. h) and denote by hαβ the coeﬃcients of the metric tensor in these local coordinates: ¯ hαβ := h ¯ Lemma 5. Y ). then H(X. Y ∈ T M. We start with the following Definition 5. as deﬁned in the previous section. W ∈ T M C . If h is a Hermitian metric on M. W ) = 0. The extension (also denoted by h) of the Hermitian metric to T M C by C–linearity satisﬁes h(Z. K¨hler metrics. h(Z. The fundamental form is given by m ∂ ∂ . Let zα be holomorphic coordinates on a complex Hermitian manifold (M 2m . A Hermitian metric on an almost complex manifold (M. which in local coordinates reads ∂2u hαβ = . Conversely.1 M. ∂zα ∂ zβ ¯ Ω=i α. J.2. 5. HERMITIAN AND KAHLER METRICS 27 5. W ∈ T 0. Y ) := h(JX.0 M and ∀ Z. ∀ Z.β=1 hαβ dzα ∧ d¯β .1. Y ) := g(X. The tangent bundle of an almost complex manifold is in particular a complex vector bundle. . Suppose that the fundamental form Ω of a complex Hermitian mania ¯ ¯ fold is closed. any Hermitian structure H on T M as complex vector bundle deﬁnes a Hermitian metric h on M by h := Re(H). Y ) = (h − iΩ)(X. Y ) − ih(JX. ¯ ∂zα ∂ zβ ¯ This particularly simple expression for the metric tensor in terms of one single real function deserves the following . ¯ ¯ ¯ (12) h(Z.2. W ). Y ) = h(JX.¨ 5. Z) > 0 ∀ Z ∈ T M C − {0}. Y ) + g(JX. Remark. z ¯ The proof is left as an exercise. JY ). W ∈ T 1. JY ). Every almost complex manifold admits Hermitian metrics.
Y. Theorem 5. Z) + B(JX. We start with doing so for the Nijenhuis tensor. Z). JY ] − [JX. J). Lemma 5. Z) = 0. A Hermitian metric h on an almost complex manifold is K¨hler if and only if J a is parallel with respect to the Levi–Civita connection of h. X) = A(Z. Z).5. A Hermitian metric h on an almost complex manifold (M. suppose that h is K¨hler a and denote by B the tensor B(X. Y ). Y. Let h be a Hermitian metric on an almost complex manifold (M. JY. JY ] = J( X J)Y − J( Y J)X − ( JX J)Y + ( JY J)X = (J( X J)Y − ( JX J)Y ) − (J( Y J)X − ( JY J)X). so in particular dΩ = 0. Proof. . Y ) = −A(X. One direction is obvious. Z) = 0. Z).3. Then J is integrable if and only if (13) ( JX J)Y = J( X J)Y. Z) = −A(X. Y. Conversely. Y. X. Z) = h(J( X J)Y − ( JX J)Y ). Then we can write N J (X. X. Y. JZ) + B(JX. and as Ω = h(J·. Conversely. ∀ X. Y ∈ T M. Thus A(X. J) is called a K¨hler a metric if J is a complex structure and the fundamental form Ω is closed: h is K¨hler ⇐⇒ a NJ = 0 dΩ = 0 ¯ A local real function u satisfying Ω = i∂ ∂u is called a local K¨hler potential of the metric h. Y ] + J[X. Combining these two relations also yields B(X. Y. since J and X J are anti–commuting skew–symmetric operators. Z). then N J clearly vanishes. Let us ﬁx a point x ∈ M and extend X and Y to vector ﬁelds on M parallel with respect to at x. Y. suppose that N J = 0 and denote by A(X. As J and X J anti–commute we have B(X. Z. Y. JY. Z. From (13) we get B(X.4. we also have Ω = 0. The previous equation just reads A(X. since if J is parallel. Z) = A(Y. thus proving that (13) implies N J = 0. A is skew–symmetric in the last two variables. Proof. Z) = −A(Y. A(X.28 Definition 5. Y ] + J[JX. Y ) = [X. Z). Y. a Our aim (as Riemannian geometers) is to express the K¨hler condition in terms of the covariant a derivative of the Levi–Civita connection of h. so by circular permutations we get which implies (13). Y. Z) := h(( X J)Y. ·). JZ) = B(X. On the other hand. with Levi–Civita covariant derivative .
JY. . Then h is K¨hler a if and only if around each point of M there exist holomorphic coordinates in which h osculates to the standard Hermitian metric up to the order 2. J is parallel. Z. 2 β. As x was arbitrary. Je1 . Jem } and choose a local holomorphic coordinate system (zα = xα + iyα ) around x such that ∂ ∂ eα = (x) and Jeα = (x). then on X. . B(X. Proof. Conversely. We will now prove the analytic characterization a of K¨hler metrics described in the ﬁrst section. Adding these two relations and using the previous properties of B yields 2B(X. JY. Y. JZ) = 0.2 implies (14) and from dΩ = 0 we ﬁnd (15) aαβγ = aγβα . X. . if the metric is K¨hler. z Ω=i δαβ + aαβγ zγ + γ¯ 2 γ γ α. J). X) + B(JZ.γ=1 m ∂hαβ ∂hαβ ¯ ¯ dxγ + dyγ ∂xγ ∂yγ ∧ dzα ∧ d¯β z clearly vanishes at x. that is. Y.6. em . Then dΩ = i α. Z and get: B(X. The condition hαβ = hβ α ¯ ¯ together with Lemma 5. a Theorem 5. JY ) = 0. . HERMITIAN AND KAHLER METRICS 29 We now use dΩ = 0 twice. with rαβ (x) = 0 ¯ ∂rαβ ∂rαβ (x) = (x) = 0 ∂xγ ∂yγ at x.β where o(z) denotes generically a function whose 1–jet vanishes at x.3. this means dΩ = 0. Characterization of K¨hler metrics. for every x ∈ M we take an orthonormal basis of Tx M of the a form {e1 . JZ. . ﬁrst on X. Z) + B(JY.γ aαβ¯ = αβαγ . Suppose that we can ﬁnd holomorphic local coordinates zα = xα + iyα around x ∈ M 1 such that hαβ = 2 δαβ + rαβ . γ . . . Y.β. X. JZ.¨ 5. JZ) + B(Y. Let h be a Hermitian metric on a complex manifold (M 2m . We now look for a local coordinate change in which the fundamental form has vanishing ﬁrst order terms. We put 1 zα = wα + bαβγ wβ wγ . 5. . ∂xα ∂yα The fundamental 2–form Ω can be written as 1 aαβ¯ zγ + o(z) dzα ∧ d¯β . X) + B(Z. Y ) = 0.
this . (which is possible because of (15) which ensures that aαβγ is symmetric in α and γ).4. then from (14) we get bαγβ = −aβαγ = −aαβ¯ . a vector ﬁeld Y is a holomorphic section of T M if and only if it is real holomorphic. γ showing that Ω=i 1 δαβ + o(z) dwα ∧ dwβ . we have to remember that T M is identiﬁed with a complex vector bundle via the complex structure J. By Lemma 2. Moreover. This coordinate change is well–deﬁned locally thanks to the holomorphic version of the local inversion theorem. Our next aim is to ex¯ press the ∂–operator on a Hermitian manifold in terms of the Levi–Civita connection.30 where bαβγ are complex numbers satisfying bαβγ = bαγβ . ¯ Lemma 5. as T M–valued (0. ¯ ¯ γ ¯ 2 If we choose bβγα = −aαβγ .7.7. so 1 ¯ ∂ (f Y )(X) = f ( X Y + J JX Y − J( 2 ¯ ¯ = f ∂ Y (X) + ∂f (X)Y. whence (using Einstein’s summation convention) Ω = i 1 δαβ + aαβγ zγ + aαβ¯ zγ + o(z) dzα ∧ d¯β z γ¯ 2 1 ¯ ¯ = i δαβ + aαβγ wγ + aαβ¯ wγ + o(z) (dwα + bαετ wε dwτ ) ∧ (dwβ + bβετ wε dwτ ) ¯ γ ¯ 2 1 = i δαβ + aαβγ wγ + aαβ¯ wγ + bβγα wγ + bαγβ wγ + o(z) dwα ∧ dwβ . Comparison of the Levi–Civita and Chern connections. In other words. For every section Y of the complex vector bundle T M. where 1 ¯ Proof. the ∂–operator.γ bαβγ wβ dwγ . We ﬁrst recall that ∂f (X) = 2 (X + iJX)(f ).1)–form is given by 1 ¯ ∂ Y (X) = ( X Y + J JX Y − J( Y J)X). Y 1 J)X) + (X(f )Y + JX(f )JY ) 2 ¯ which shows that the above deﬁned operator ∂ satisﬁes the Leibniz rule. We then have dzα = dwα + β. 2 denotes the Levi–Civita connection of any Hermitian metric h on M. a product iX for some X ∈ T M is identiﬁed with JX. ¯ 2 5. In order to do so.
J is parallel with respect to the Chern connection. This proves that ∂ = ∂. Z) > 0 ∀ Z ∈ T M C − {0}. W ∈ T 0. Finally. The Chern connection coincides with the Levi–Civita connection if and only if a h is K¨hler. (3) If hαβ denote the coeﬃcients of a Hermitian metric tensor in some local holomorphic ¯ coordinate system. W ) = 0. if = ¯ then J a a is –parallel.1)–form. Proposition 5. Exercises.1 ¯ = ∂ follows from Lemma 5. as 0. J) two natural linear connections: the Levi–Civita connection and the Chern connection ¯ on T M as Hermitian vector bundle. HERMITIAN AND KAHLER METRICS 31 is equivalent to LY J = 0. Conversely. it is a H–connection since h = 0 and Ω = 0. ∀ Z. the condition 1 1 0. X 5. JX] − J[Y. so h is K¨hler by Theorem 5. which is a complex connection. which means that for every vector ﬁeld X ∈ C ∞ (T M) one has 0 = (LY J)X = LY (JX) − JLY X = [Y. (1) Prove that every Hermitian metric on a 2–dimensional almost complex manifold is K¨hler.0 M and ∀ Z. by Theorem 5.5. ¯ ¯ ¯ thus showing that ∂ Y vanishes for every holomorphic section Y . A Hermitian manifold (M. X] = Y JX − JX Y − J Y X + J X Y = ( Y J)X − JX Y + J X Y = J( X Y + J JX Y − J( Y J)X). Moreover.1 M.5. ¯ ¯ (4) Show that the extension of a Hermitian metric h by C–linearity is a symmetric bilinear tensor satisfying h(Z. Then the Levi–Civita connection is a well–deﬁned complex connection in T M since J = 0.1 = 2 ( X + i JX ) = 2 ( X + J JX ). and conversely. a (2) Prove that the fundamental form of a Hermitian metric is a (1. Proof.8. Thus. h(Z. ¯ ¯ ¯ h(Z.7. W ∈ T M C . ∀ Z. By deﬁnition. any symmetric complex bilinear tensor satisfying this system arises from a Hermitian metric. show that hαβ = hβ α . h.¨ 5. . W ) = h(Z.5 again. W ). suppose that h is K¨hler. W ∈ T 1. Let H := h − iΩ denote the Hermitian structure of T M.
ei ) = i=1 R(Jei . JY. as can be easily seen from the symmetries of the curvature. Ric(X. Z. Y ∈ Tx M. Z. Y. Y. 2m The Ricci tensor of M is deﬁned by or equivalently Ric(X.Y ] Z. T. Y. The curvature tensor. If n ≥ 3. Y ) = i=1 R(ei . W ) + ( Z R)(X. Z− Y XZ − [X. R(X. X. R(X. Z). X. X. T ) + R(Z. it is easy to check that λ (which a priori depends on x) has to be constant on M (see [8]). Z. Y )JZ = JR(X. Z. g) is called Einstein if the Ricci tensor Ric is proportional to the metric tensor g at each point of M Ric(X. Z. The symmetries of the curvature operator then read: • • • • ∀ X. ( X R)(Y. Y. Y )Z := X Y It is sometimes convenient to identify the curvature tensor with the following tensor: R(X. X. X)Y }. Y )Z. the set {Jei } is also an orthonormal basis. Y.1. The curvature tensor of K¨hler manifolds a 6. X. Z. A Riemannian manifold (M n . Y. T ). Z. Jei ) = Ric(X. X. T. T. W ) + ( Y R)(Z. This immediately implies hence Ric(JX. Z. Y )Z. T ) = 0 (ﬁrst Bianchi identity). Y. Let (M n . T ) = R(Z. JZ. JY ) = i=1 –parallel. Z ∈ C ∞ (T M). Y.32 6. the curvature tensor ∀ X. Y ) := Tr{V → R(V. ei ). T ) + R(Y. J) is a K¨hler manifold. Y. Y. R(X. T ). Y ) ∀ X. T ) := h(R(X. Y. W ) = 0 (second Bianchi identity). Y ) = λg(X. Y. Y. T ) = R(JX. ∀ X. Y ). T ) = −R(X. Z. JY. R(X. since for every orthonormal basis {ei }. T. JT ) = R(X. Since J is a has more symmetries: (16) R(X. T. We recall that the Ricci tensor of every Riemannian manifold is symmetric. Y ). h. 2m 2m R(ei . Suppose now that (M 2m . This last equation justiﬁes the following . g) be a Riemannian manifold with Levi–Civita connection and denote by R its curvature tensor: R(X. JX. where ei is a local orthonormal basis of T M. Y. Z ∈ C ∞ (T M). T ∈ T M.
ei ) + R(Y.2. X) + ( Z ρ)(X. Z) ◦ J + ( Y R)(Z. • the cohomology class of ρ is (up to some real multiple) equal to the Chern class of the canonical bundle of M. X.2. Among its properties a which will be proved in the next few sections we mention: • the Ricci form ρ is closed. Jei )) (R(X. Y ). X) ◦ J + ( Z R)(X. J) be a K¨hler manifold with a a Levi–Civita covariant derivative and let (zα ) be a system of local holomorphic coordinates. ¯ • in local coordinates. Jei . Jei . ∀ X. (i) Using the ﬁrst Bianchi identity we get Ric(X. (i) The Ricci tensor of a K¨hler manifold satisﬁes a 1 Ric(X. ei . K¨hler metrics in local coordinates. Y ∈ T M. The Ricci form is one of the most important objects on a K¨hler manifold. Z) = 2(( X ρ)(Y. X. Y ) = i R(ei . Jei )) = i = Tr(R(X.1. Jei ) − R(JY. (ii) From (i) we can write 2ρ(X. Jei ) = i (−R(X. Y. THE CURVATURE TENSOR OF KAHLER MANIFOLDS 33 Definition 6. X. Y ) := Ric(JX. X. 6.¨ 6. ei) = i R(ei . JY ) ◦ J) − Ric(X. Let (M 2m . Proof. ei . We introduce the following local basis of the complexiﬁed tangent space: Zα := ∂ . The Ricci form ρ of a K¨hler manifold is deﬁned by a ρ(X. Let (ei ) be a local orthonormal basis of T M. . Y ) = Tr(R(X. ¯ For the moment being we use the Bianchi identities for the curvature tensor to prove the following Proposition 6. Y ) ◦ J). 2dρ(X. JY. Y. Therefore. ∂ zα ¯ 1 ≤ α ≤ m. JY ) ◦ J). Y ) = Tr(R(X. Y )) = Tr(( X R)(Y. Z) + ( Y ρ)(Z. 2 (ii) The Ricci form is closed. ρ can be expressed as ρ = −i∂ ∂ log det(hαβ ). ∂zα Zα := ¯ ∂ . JY. h. where det(hαβ ) denotes ¯ ¯ the determinant of the matrix (hαβ ) expressing the Hermitian metric. JY. Y ). Y ) ◦ J) = 0 from the second Bianchi identity.
Z δ ) This proves the formulas (19) and Γγ = hγ δ αβ ∂zα . Rαβ¯ .0)–tensor. Rαβ¯δ ¯ ¯ ¯γ ¯ γ ¯ ¯ and δ δ δ δ Rαβγ . we get by conjugation ΓC = ΓCB . . . m}. . ZC . ¯ AB A¯ Since is torsion–free we have ΓC = ΓC . Using the curvature symmetries we ﬁnally see that the only non–vanishing components γ¯ of R are Rαβγ δ . The Christoﬀel symbols are deﬁned ¯ by C ZA ZB = ΓAB ZC . run over the set {1. = Γγ hγ δ . since the metric is Hermitian we have (17) ¯ hαβ = hαβ = 0. γ γ ¯ From the fact that T 1. AB BA and since T 1. . The corresponding coeﬃcients are deﬁned by D R(ZA .0 is –parallel we must have Γγ β = 0. hence RABγδ = ¯ RAB¯δ = 0. Rαβ¯ . Of course. ¯ Using the convention α = α. . ZD ) = hDE RABC . ¯ of the K¨hler metric in these coordinates by a hAB := h(ZA . in order to compute these coeﬃcients we notice that ΓCδ = 0 implies α¯ (18) hence ∂hβ δ ¯ = h( ∂zα Γγ hγ δ = ¯ αβ ∂hβ δ ¯ ∂zα ¯ Zα Z δ = 0. . Rαβγ . ¯ αβ ¯ ¯ ∂hβ δ ¯ Zα Z β .0 M is parallel we immediately get RABδ = RABδ = 0. ¯ ¯ ¯ Let hαβ denote the coeﬃcients of the inverse matrix of (hαβ ). C.34 and we let subscripts A. α¯ ¯ ¯ Now. ¯ . The curvature tensor can be viewed either as (3. ZB . ZB ). . ¯ ¯γ ¯ ¯ γ ¯ ¯ . ¯¯ hβα = hαβ = hβ α . . We denote the components 1. Rαβγ δ . Rαβ¯δ . B. . A¯ These relations show that the only non–vanishing Christoﬀel symbols are Γγ αβ and ¯ Γγ β .1)– or as (4. ZB )ZC = RABC ZD and E RABCD = R(ZA . . m.
d σ∈S d m and 1 ˜ hkj hji = d j=1 m ε(σ)h1σ1 . hmσm = 0 j=1 σ∈Sm . . . . σi =j . αγ γα ∂zγ d ∂zγ ∂zγ This proves the following simple expressions for the Ricci tensor Ricαβ = − ¯ and for the Ricci form (21) Lemma 6. ∂ zβ ¯ therefore (20) δ Rαβγ = − ¯ ∂Γδ αγ . m. . ∂zα ∂ zβ ¯ ¯ ρ = −i∂ ∂ log d. hmσm = d = 1. . d (t) = d i. Then the following formula holds m ∂ 2 log d .j=1 hij (t)hji(t).¨ 6. hmσm . . THE CURVATURE TENSOR OF KAHLER MANIFOLDS 35 From (18) and (19) we obtain δ Rαβγ Zδ = R(Zα . . Recall the deﬁnition of the determinant d= σ∈Sm ε(σ)h1σ1 . σi =j m 1 1 ε(σ)h1σ1 .3. σi =j then we obtain easily m ˜ hij hji = j=1 ε(σ)h1σ1 . Let (hij ) = (hij (t)) be the coeﬃcients of a map h : R → Glm (C) with hij := (hij )−1 and let d(t) denote the determinant of (hij ). hi−1σi−1 hkσi hi+1σi+1 . . . . hmσm . hi−1σi−1 hi+1σi+1 . . . . If we denote 1 ˜ hji := d σ∈S 1 d m ε(σ)h1σ1 . . ∂ zβ ¯ Using this formula we can compute the components of the Ricci tensor: ∂Γα αγ A α Ricγ β = Ricβγ = RAβγ = Rαβγ = − . ¯ ¯ ¯ ¯ ∂ zβ ¯ Let us denote by d the determinant of the matrix (hαβ ). Zβ )Zγ = − ¯ ¯ Zβ ( ¯ Zα Z γ ) = − δ Zβ (Γαγ Zδ ) = − ¯ δ ∂Γαγ Zδ . Proof.3 below and (19) we ¯ get 1 ∂d ∂ log d ¯ ∂h ¯ Γα = Γα = hαδ αδ = = . hmσm = j=1 σ∈Sm . Using the Lemma 6.
0 to 0. We now get m d (t) = σ∈Sm i=1 m m ε(σ)hiσi (t)h1σ1 . Compute the image of the fundamental form through this endomorphism. ˜ This shows that hji = hji are the coeﬃcients of the inverse matrix of h. maps Λ0. 2 In particular. . . hi−1σi−1 hi+1σi+1 . (1) Let S := Tr(Ric) denote the scalar curvature of a K¨hler manifold M with Ricci form ρ. hmσm ε(σ)hij (t)h1σ1 .36 for k = i since in the last sum each term corresponding to a permutation σ is the opposite of the term corresponding to the permutation (ik) ◦ σ. . . Using (17) show that the determinant of the complex m × m matrix (hαβ ) is a positive real number whose square ¯ is equal to the determinant of the real 2m × 2m matrix hij representing the metric in the local coordinate system (xi . .j=1 = = 6. .2 and Λ2. hmσm i=1 j=1 σ∈Sm . . Prove that the Ricci tensors of h and h are equal. yi ). prove the formula: 1 δρ = − JdS. σi =j m m ˜ ji = d d hij (t)h hij (t)hji(t). i. (2) Prove that the curvature of a K¨hler manifold. hi−1σi−1 hi+1σi+1 . . a Using the second Bianchi identity. (4) Let h and h be two K¨hler metrics on some complex manifold (M. Exercises.3. the Ricci form of M is harmonic if and only if the scalar curvature S is constant. (3) Let h be a Hermitian metric on some complex manifold M 2m and let zα = xα + iyα be a local system of holomorphic local coordinates on M. viewed as a symmetric endomorphism a of the space of complex 2–forms. . J) having the same a (Riemannian) volume form.j=1 i. where (ik) denotes the transposition of i and k.
with local trivializations ψj : π −1 Uj → Uj × C∗ given by ψj (z) = ([z]. −i . . : zm ]. . z 2 α=1 m 1 Thus u(z) = 2 z2 is a K¨hler potential for the canonical Hermitian metric on Cm . Examples of K¨hler metrics a 7. . . Let π : Cm+1 − {0} → CPm be the canonical projection π(z0 . The Fubini–Study metric on the complex projective space CPm . Y ∈ T CPm .¨ 7. EXAMPLES OF KAHLER METRICS 37 7. G Cm y yy yy yy yy yy φj y yy yy  yy Uj ⊂ CPm The map fj is clearly holomorphic and a direct calculation yields u ◦ fj (z) = v(z) − log zj 2 . this shows that (fj )∗ (∂ ∂u) = ∂ ∂v for every j. . For every j ∈ {0. . zm ) = [z0 : . The next lemma proves that h deﬁnes a K¨hler metric on CPm . so the tensor h deﬁned by h(X. . Y ) := Ω(X. ¯ ∂zα ∂ zβ ¯ 4 ∂xα ∂yα ∂xβ ∂yβ 2 so by Lemma 5. The tensor h is positive deﬁnite on CPm . As ¯ ¯ ¯ ∂ ∂ log zj 2 = 0. which satisﬁes ¯ (22) π ∗ (Ω) = i∂ ∂v. α) = ([z].2 the fundamental form is Ω=i 1 2 i ¯ dzα ∧ d¯α = ∂ ∂z2 . Its coeﬃcients in holomorphic coordinates are ∂ ∂ ∂ ∂ ∂ ∂ 1 1 .2. a Lemma 7. +i hαβ = h = h = δαβ . Consider the holomorphic atlas (Uj . is symmetric and Hermitian.1. This map is clearly surjective. JY ). m}. we deﬁne fj = φj ◦ π. It is moreover a principal C∗ –ﬁbration. . . ∀ X. Cm+1 − {0} fj π 7. The ﬂat metric on Cm . zj −1 and satisfying ψj ◦ ψk ([z]. a Consider the functions u : Cm → R and v : Cm+1 − {0} → R deﬁned by u(w) = log(1 + w2) and v(z) = log(z2 ). We thus can deﬁne a 2–form Ω on m CP by ¯ ΩUj := i(φj )∗ (∂ ∂u). 1)–form. zk α). zj ). Clearly Ω is a closed real (1. . φj ) on CPm described in the ﬁrst section.1.
Y ∈ T C . . . ∀ X. This shows that ker((π∗ )z ) is the complex z line generated by z.38 ˆ Proof. We have ˆ is positive deﬁnite. Y ) := i∂ ∂u(X. . Geometrical properties of the Fubini–Study metric. zm ) = 1 (z0 . zj−1. . h = (φj )∗ (h). We a provide here a more geometrical description of this metric. 0) ∈ Cm for some positive real enough to prove that h number r.3. so it is h. . . . Lemma 7. The K¨hler metric on CPm constructed in this way is called the Fubini–Study metric and is usually a denoted by hF S . . 0. . Now. since Um is a group of holomorphic transformations of to prove that h Cm preserving u. z0 z0 0 Thus v ∈ ker(π∗ )z ⇐⇒ v ∈ ker(f∗ )z ⇐⇒ v = v0 z. . . it is transitive on the unit sphere of Cm . The Fubini–Study metric was deﬁned via its fundamental 2–form. zm ). . z m i=2 ¯ Xi Y i ) i=2 We put j = 0 for simplicity and denote f = f0 . Clearly. ˆ is positive deﬁnite at a point p = (r. the canonical projection π : Cm+1 − {0} → CPm is a submersion. . . which was expressed by local K¨hler potentials. vm ) − 2 (z1 . . and the kernel of its diﬀerential π∗z : Tz (Cm+1 − {0}) → Tπ(z) CPm is the complex line generated by z. it also preserves ˆ Moreover. . The composition fj := φj ◦ π is given by fj (z0 . . We have ¯ ∂ ∂ log(1+z2 ) = ∂( 1 1 ( zi d¯i )) = z 2 1 + z i=1 1 + z2 m m i=1 1 dzi ∧d¯i − z ( zi dzi )∧( ¯ zi d¯i ). . Y ) = ˆ hence h is positive deﬁnite. . zj 2 ¯ Re(X1 Y1 + (1 + r 2 ) (1 + r 2 )2 dzi ∧ d¯i . . . . z (1 + z2 )2 i=1 i=1 m m m At p this 2–form simpliﬁes to 1 dz1 ∧ d¯1 + (1 + r 2 ) z (1 + r 2 )2 which shows that ˆ hp (X. . 7. Let us ﬁx some local holomorphic chart φj : Uj → Cm . Its diﬀerential at z applied to some tangent vector v is 1 v0 f∗ z (v) = (v1 . Proof. Let z ∈ Cm+1 with zj = 0. zj+1 . zm ). . showing that it is the projection to CPm of some symmetric tensor ﬁeld of Cm+1 − {0}. where m m ˆ ˆ ¯ h is the symmetric tensor on C deﬁned by h(X.2. and for dimensional reasons (π∗ )z has to be surjective. For every z ∈ Cm+1 − {0}. JY ). .
¨ 7. EXAMPLES OF KAHLER METRICS
39
Consider the complex orthogonal z ⊥ of z in Cm+1 with respect to the canonical Hermitian metric, i.e. the set
m
z := {y ∈ C
⊥
m+1

zj yj = 0}. ¯
j=0
This deﬁnes a codimension 1 complex distribution D in Cm+1 − {0} with Dz := z ⊥ . Let X → X ⊥ denote the orthogonal projection onto z ⊥ in Tz (Cm+1 −{0}) and deﬁne a bilinear symmetric tensor ˜ h on Cm+1 − {0} by 2 ˜ h(X, Y ) := 2 X ⊥ , Y ⊥ , z where ·, · denotes the canonical Hermitian product. ˜ ˜ Lemma 7.3. The (1,1)–form ϕ(X, Y ) := h(JX, Y ) associated to the tensor h satisﬁes ϕ = ¯ log(z2 ) on Cm+1 − {0}. i∂ ∂
Proof. It is enough to prove this relation at a point p = (r, 0, . . . , 0) ∈ Cm+1 − {0} for some positive real number r because both members are Um+1 –invariant and Um+1 is transitive on spheres. We have ¯ ∂ ∂ log(z2 ) = ∂( 1 1 zi d¯i )) = 2 z ( z2 i=0 z 1 r2 On the other hand, we have at p −iϕ ∂ ∂ , ∂zα ∂ zβ ¯ ∂ ∂ ˜ = −ih i , ∂zα ∂ zβ ¯ 1 r2
1 δ r 2 αβ m m m m
i=0
dzi ∧ d¯i − z
1 zi d¯i ). z zi dzi ) ∧ ( ¯ ( z4 i=0 i=0
m
m
At p this 2–form simpliﬁes to
i=1
dzi ∧ d¯i . z ˜ =h ∂ ∂ , ∂zα ∂ zβ ¯
which vanishes if α = 0 or β = 0 and equals −iϕ =
otherwise. Thus dzi ∧ d¯i . z
i=1
˜ Using this lemma and (22) we see that π ∗ h = h, showing that the Fubini–Study metric hF S on m ˜ CP is given by the projection of the above deﬁned semi–positive symmetric tensor ﬁeld h. Proposition 7.4. The group Um+1 acts transitively by holomorphic isometries on (CPm , hF S ) Proof. For every A ∈ Um+1 , z ∈ Cm+1 − {0} and α ∈ C∗ , we have A(αz) = αA(z), showing that the canonical action of Um+1 on Cm+1 − {0} descends to an action on CPm . For every ˜ A ∈ Um+1 , let A be the corresponding transformation of CPm . Looking at its expression in ˜ the canonical holomorphic charts, it is easy to check that every A acts holomorphically on CPm .
40
˜ In order to check that A preserves the Fubini–Study metric, we ﬁrst use (22) and the relation 2 v ◦ A(z) = log Az = log z2 = v(z) to get ¯ ¯ ¯ ˜ π ∗ (A∗ (Ω)) = A∗ (i∂ ∂v) = i∂ ∂A∗ v = i∂ ∂v = π ∗ Ω. ˜ ˜ Lemma 7.2 shows that π∗ is onto, so π ∗ is injective on exterior forms, hence A∗ (Ω) = Ω. As A ˜ also preserves the complex structure, too, this clearly implies that A is an isometry. We will now use our computations in local coordinates from the previous section in order to show that the Fubini–Study metric is Einstein. Since there exists a transitive isometric action on CPm , it is enough to check this at some point, say p := [1 : 0 : . . . : 0] ∈ CPm . From Lemma 7.1 we see that the fundamental form is given in the local chart φ0 by i Ω= 1 + z2
m
i=1
i dzi ∧ d¯i − z ( zi dzi ) ∧ ( ¯ zi d¯i ). z (1 + z2 )2 i=1 i=1
m
m
Lemma 7.5. Let dv denote the volume form on Cm i i z z dv := dx1 ∧ dy1 ∧ . . . ∧ dxm ∧ dym = dz1 ∧ d¯1 ∧ . . . ∧ dzm ∧ d¯m . 2 2 Then the fundamental 2–form Ω satisﬁes Ω∧ ...∧Ω = 2m m! dv. (1 + z2 )m+1
Proof. Both terms are clearly invariant by the action of Um on Cm , which is transitive on spheres, so it is enough to prove the equality at points of the form z = (r, 0, . . . , 0), where it is actually obvious. Now, for every Hermitian metric h on Cm with fundamental form ϕ, the determinant d of the matrix (hαβ ) satisﬁes ¯ 1 m ϕ = d2m dv. m! Applying this to our situation and using the lemma above yields d = det(hαβ ) = ¯ 1 , (1 + z2 )m+1
whence log d = −(m + 1) log(1 + z2 ), so from the local formula (21) for the Ricci form we get ¯ ¯ ρ = −i∂ ∂ log d = (m + 1)i∂ ∂ log(1 + z2 ) = (m + 1)Ω, thus proving that the Fubini–Study metric on CPm is an Einstein metric, with Einstein constant m + 1.
¨ 7. EXAMPLES OF KAHLER METRICS
41
7.4. Exercises. (1) A submersion f : (M, g) → (N, h) between Riemannian manifolds is called Riemannian submersion if for every x ∈ M, the restriction of (f∗ )x to the g–orthogonal of the tangent space to the ﬁber f −1 (f (x)) is an isometry onto Tf (x) N. Prove that the restriction of the 1 canonical projection π to S 2m+1 deﬁnes a Riemannian submersion onto (CPm , 2 hF S ). √ √ (2) Show that (CP1 , hF S ) is isometric to the round sphere of radius 1/ 2, S 2 (1/ 2) ⊂ R3 . Hint: Use the fact that a simply–connected manifold with constant positive sectional √ curvature K is isometric to the sphere of radius 1/ K. (3) Show that for every Hermitian tensor h on Cm with fundamental form ϕ, the determinant d of the matrix (hαβ ) satisﬁes ¯ ϕm = d2m m!dv.
.
Part 3 The Laplace operator .
Lemma 8. The lemma below gives a useful method to compute the formal adjoint: Lemma 8. denoted Q and Q . g) be an oriented Riemannian manifold (not necessarily compact) with volume form dv and let E and F be Hermitian vector bundles over M with Hermitian structures denoted by ·. ∞ ∞ for every compactly supported smooth sections α ∈ C0 (E) and β ∈ C0 (F ). There exists at most one formal adjoint for every linear diﬀerential operator. Suppose that P : C ∞ (E) → C ∞ (F ) has two formal adjoints. Let (M n . thus showing that the smooth positive function R(f σ)2 has to vanish identically on M.1. Then their diﬀerence R := Q − Q satisﬁes α. · E and ·.8. Let P : C ∞ (E) → C ∞ (F ) and Q : C ∞ (F ) → C ∞ (E) be linear diﬀerential operators. . Applying the formula above to the compactly supported sections α := R(f σ) and β := f σ of E and F we get 0= M α. Take a positive function f on M such that f ≡ 1 on some open set U containing x and f = 0 outside a compact set. Definition 8. Natural operators on Riemannian and K¨hler manifolds a 8. The operator Q is called a formal adjoint of P if P α. contradicting the fact that its value at x is non–zero. The n − 1–form ω(α. β M F dv = M α. Qβ E )dv = d(ω(α.1. Since R is a diﬀerential operator. By Stokes’ theorem we see that the integral over M of its exterior derivative vanishes. Rβ E dv = M R(f σ)2dv. then Q is the formal adjoint of P . the value of R(σ) at x only depends on the germ of σ at x. · F .2. β ∈ C ∞ (F ). From the above lemma it is immediate to check that P is the formal adjoint of P ∗ and that Q∗ ◦ P ∗ is the formal adjoint of P ◦ Q. The formal adjoint of a linear diﬀerential operator. Suppose that there exists some σ ∈ C ∞ (F ) and some x ∈ M such that R(σ)x = 0. The formal adjoint of an operator P is usually denoted by P ∗ . Let P : C ∞ (E) → C ∞ (F ) and Q : C ∞ (F ) → C ∞ (E) be linear diﬀerential operators. so in particular R(f σ) has compact support and R(f σ)x = R(σ)x = 0. β F − α. β) has compact support for every compactly supported sections α and β. If there exists a section ω ∈ C ∞ (E ∗ ⊗ F ∗ ⊗ Λn−1 M) such that (23) ( P α. Proof. Qβ E dv. ∀ α ∈ C ∞ (E). Rβ M E dv =0 ∞ ∞ ∀ α ∈ C0 (E).3. Proof. ∀ β ∈ C0 (F ). β)).
. τ ∈ Λk−1 M. . ∧ eik  1 ≤ i1 < . . ∗ω. X ∧ τ . let α ∈ Ωp and β ∈ Ωp+1 be smooth forms. With respect to this scalar product. X ω. . . . . . ∀ X ∈ T M. τ := g(ϕ(ω). We denote generically by {e1 . .2. on Λk M. ω ∈ Λk M. τ dv. . Xk ) := ω(X1 . . . g) with volume form dv. . ∗d ∗ β dv. ∗τ = ω. NATURAL OPERATORS ON RIEMANNIAN AND KAHLER MANIFOLDS 45 8. k! which can also be characterized by the fact that the basis is orthonormal.¨ 8. Then we have dα. en } a local orthonormal frame on M parallel in a point and identify vectors and 1–forms via the metric g. . ε(σ)eσ1 ⊗ . . ∗1 = dv. We consider the following weighted tensor product on Λk M: 1 ω. . ⊗ eσk . . < ik ≤ n} It is well–known and easy to check on the local basis above that the following relations are satisﬁed: (25) (26) (27) ∗dv = 1. There is a natural embedding ϕ of Λk M in T M ⊗k given by which in the above local basis reads ϕ(e1 ∧ . ∧ ek ) = σ∈Sk ϕ(ω)(X1. . The Riemannian product g induces a Riemannian product on all tensor bundles. τ = ω. . ∀ω. . We start with an oriented Riemannian manifold (M n . the interior and exterior products are adjoint operators: (24) We deﬁne the Hodge *–operator ∗ : Λk M → Λn−k M by ω ∧ ∗τ := ω. {ei1 ∧ . . τ . ϕ(τ )). τ ∈ Λk M. Xk ). In this way we can write for instance dv = e1 ∧ . The Laplace operator on Riemannian manifolds. . ∗2 = (−1)k(n−k) d= i The exterior derivative d : C ∞ (Λk M) → C ∞ (Λk+1M) ei ∧ ei has a formal adjoint δ : C ∞ (Λk+1 M) → C ∞ (Λk M) satisfying δ = −(−1)nk ∗ d∗ = − ei i ei . To see this. ∧ en . β dv = dα ∧ ∗β = d(α ∧ ∗β) − (−1)p α ∧ d ∗ β = d(α ∧ ∗β) − (−1)p+p(n−p)α ∧ ∗ ∗ d ∗ β = d(α ∧ ∗β) − (−1)np α. .
The Laplace operator ∆ : C ∞ (Λk M) → C ∞ (Λk M) is deﬁned by ∆ := dδ + δd. We deﬁne the following algebraic (real) operators acting on diﬀerential forms: 1 L : Λk M → Λk+2 M. Lemma 8. then Q is the formal adjoint of P . let now a (M . Λ(ω) := 1 2 Jei ei ω. β ∈ C ∞ (F ). L] = JX ∧ . Qβ E = δ(τ (α.46 (28) Using the Hodge *–operator we get the following useful reformulation of Lemma 8. ∞ δ c := − ∗ dc ∗ = − [L. h. and is clearly formally self–adjoint. ∀ α ∈ C ∞ (E). d] = 0 . [Λ. β F so Lemma 8. The Laplace operator on K¨hler manifolds.3 shows that d∗ = (−1)np+1 ∗ d∗ on p + 1–forms. L(ω) := Ω ∧ ω = ei ∧ Jei ∧ ω. [L. After these preliminaries. the following relations hold: a (29) and (30) [Λ.5. The proof is straightforward. β)). Lemma 8. Λ] = 0 and [X .3: if there exists a section τ ∈ C ∞ (E ∗ ⊗ F ∗ ⊗ Λ1 M) such that P α. 8. δ] = dc . J) be an almost Hermitian manifold with fundamental form Ω. m − p)–forms. − α. δ] = 0. We deﬁne the twisted diﬀerential dc : C ∞ (Λk M) → a ∞ k+1 C (Λ M) by dc (ω) := Jei ∧ ei ω i whose formal adjoint is δ : C (Λ c ∞ k+1 M) → C (Λk M) Jei i ei . The following relations hold: (1) The Hodge *–operator maps (p. i These natural operators can be extended to complex–valued forms by C–linearity. Let us now assume that M is K¨hler. 2 i 2m with adjoint Λ satisfying Λ : Λk+2M → Λk M.3. (2) [X . d] = −δ c .4. q)–forms to (m − q. On a K¨hler manifold.
Applying (29) to a (p. ¯ and similarly from (30) we obtain ¯ ¯ (35) [Λ. ∂ ∗ := − ∗ ∂∗ and ¯ ¯ ∂ ∗ : C ∞ (Λp. τ . ∂]∂ = ∂Λ∂ − ∂Λ∂ = 0 ¯ ¯ ∂ ∂ ∗ + ∂ ∗ ∂ = 0. ∂] = 0. where ¯ ∂ ∗ : C ∞ (Λp. The two relations a in (30) follow by the uniqueness of the formal adjoint.q−1M).4 (2) and the fact that J and Ω are parallel we get The second relation in (29) just expresses the fact that the K¨hler form is closed. ¯ [Λ. and similarly ¯ Now. ei ei ] = −[L. H(ω. ¯ ¯ Corresponding to the decomposition d = ∂ + ∂ we have the decomposition δ = ∂ ∗ + ∂ ∗ . [L.q M). [L. ¯ One of the most important features of K¨hler metrics is that these new Laplace operators are a essentially the same as the usual one: Theorem 8.q M) → C ∞ (Λp. ∂] = −i∂ ∗ . ∂] + [Λ. q)–form and projecting onto Λp±1. Our identiﬁcation of T M and T ∗ M via the metric maps (1. [L. [Λ.q M are preserved by the covariant derivative follows easily 1 1 ¯ ∂= (ej + iJej ) ∧ ej and ∂= (ej − iJej ) ∧ ej . a Proof. τ ) := ω. ¯ ¯ Notice that ∂ ∗ and ∂ ∗ are formal adjoints of ∂ and ∂ with respect to the Hermitian product H on complex forms given by (31) We deﬁne the Laplace operators ∆∂ := ∂∂ ∗ + ∂ ∗ ∂ and ¯ ¯¯ ¯ ¯ ∆∂ := ∂ ∂ ∗ + ∂ ∗ ∂. ∂] = 0. ∂ ∗ ] = 0. ∂] = i∂ ∗ .q M) → C ∞ (Λp−1. [Λ. 0)–vectors to (0. On any K¨hler manifold one has ∆ = 2∆∂ = 2∆∂ . Using Lemma 8. ei ] ei = Jei ∧ ei = dc . ∂ ∗ := − ∗ ∂ ∗ . δ] = −[L. ∂ ∗ ] = 0.¨ 8.q±1M then yields ¯ ¯ ¯ (34) [L. ¯ δ c = i(∂ ∗ − ∂ ∗ ). ∂ ∗ ] = i∂.6. NATURAL OPERATORS ON RIEMANNIAN AND KAHLER MANIFOLDS 47 Proof. . ∂ ∗ ] = −i∂. From the fact that Λp. the relation ∂ 2 = 0 together with (35) gives ¯ ¯ ¯ ¯ ¯¯ ¯ ¯ ¯ ¯ −i(∂∂ ∗ + ∂ ∗ ∂) = ∂[Λ. 1)–forms and vice–versa. 2 2 j j From the deﬁnition of dc we then get (32) and by adjunction (33) ¯ dc = i(∂ − ∂). [L.q M and Λp.
Exercises. Prove that ei ∧(ei ω) = kω and ei (ei ∧ ω) = (n − k)ω. a (4) Prove that [J. Λ] = 0 and [J. Λ and J on K¨hler manifolds. (1) Consider the extension of J as derivation J : Λk M → Λk M. ∂]∂ = ∂Λ∂ − ∂ ∂Λ + Λ∂∂ − ∂Λ∂ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯¯ = ∂Λ∂ + ∂∂Λ − Λ∂ ∂ − ∂Λ∂ = [∂. • J(α ∧ β) = J(α) ∧ β + α ∧ J(β) for all forms α ∈ Ωp M and β ∈ Ωk M. L] = 0. ¯ ¯ ¯ ¯ ∆ = (∂ + ∂)(∂ ∗ + ∂ ∗ ) + (∂ ∗ + ∂ ∗ )(∂ + ∂) ¯¯ ¯ ¯ ¯ ¯ ¯ ¯ = (∂∂ ∗ + ∂ ∗ ∂) + (∂ ∂ ∗ + ∂ ∗ ∂) + (∂∂ ∗ + ∂ ∗ ∂) + (∂ ∂ ∗ + ∂ ∗ ∂) = ∆∂ + ∆∂ . Λ] = −i∂ ∗ ∂ − i∂ ∂ ∗ = −i∆∂ . a .48 Thus. a (5) Show that the Laplace operator commutes with L.q M equals to the scalar multiplication by i(q − p). It remains to show the equality ∆∂ = ∆∂ . dc ] = −d on K¨hler manifolds. Show that the following relations hold: • J is skew–Hermitian. (2) Let ω be a k–form on a n–dimensional Riemannian manifold M. • [J. Λ]∂ + ∂[∂. • The restriction of J to Λp. (3) Show that 0 = ddc + dc d = dδ c + δ c d = δδ c + δ c δ = δdc + dc δ on every K¨hler manifold. J(ω) := i Jei ∧ ei ω.4. d] = dc and [J. ∂] + [Λ. 8. which is an easy consequence of (35): ¯ ¯ ¯ ¯ ¯ ¯ −i∆∂ = −i(∂∂ ∗ + ∂ ∗ ∂) = ∂[Λ.
From now on we denote the space of smooth complex–valued k–forms by Ωk M := C ∞ (Λk M ⊗ C) and by Z k (M) the space of closed complex k–forms on M. The space of k–forms decomposes as a direct sum Ωk M = Hk (M.1. 84–100. τ ) := M H(ω.1. ω)dv = M δω2 + dω2dv. A proof can be found in [2]. A form is harmonic if and only if it is closed and δ–closed. showing that dω = 0 and δω = 0. C). C) the space of complex harmonic k–forms on M. g) is a compact oriented Riemannian manifold. HODGE AND DOLBEAULT THEORY 49 9.2. ω ) = M δω 2 dv. Hodge and Dolbeault theory 9. C) := dΩk−1 M The De Rham theorem says that the k th cohomology group of M with complex coeﬃcients is naturally isomorphic to the k th De Rham cohomology group: H k (M. one clearly has dΩk−1 M ⊂ Z k M. where ω ∈ Ωk−1 M. Proof. We deﬁne the De Rham cohomology groups by ZkM k . we can write 0 = (dω. Since the exterior derivative satisﬁes d2 = 0. (Hodge decomposition theorem). C). . H(∆ω. pp. ω)dv = M H(dδω + δdω. forms in the kernel of the Laplace operator: Lemma 9. τ )dv. ω ∈ Ωk+1 M and ω H ∈ Hk (M. Theorem 9. C) := {ω ∈ Ωk M  ∆ω = 0}. In this section we assume that (M n . Since M is compact and d and δ are formally adjoint operators we get 0= M Hk (M. Using Lemma 9. Suppose conversely that ω is harmonic. The Hodge decomposition theorem shows that every k–form ω on M can be uniquely written as ω = dω + δω + ω H . Proof. HDR (M. If ω is closed. We now denote by Hk (M.e.1 it is immediate to check that the three spaces above are orthogonal with respect to the global Hermitian product on Ωk M given by (ω. The hard part of the theorem is to show that the direct sum of these three summands is the whole space Ωk M. showing that the second term in the Hodge decomposition of ω vanishes.9. C) k HDR (M. Hodge theory. ω ) = (dδω . One direction is clear. i. C) ⊕ δΩk+1 M ⊕ dΩk−1 M.
In e particular bk (M) = bn−k (M) for every compact n–dimensional manifold M. Proposition 9. it thus preserves the complex structure J. thus LX ◦ ∗ = ∗ ◦ LX . Finally. The spaces Hk (M. On the other hand. (Poincar´ duality). The natural map f : Hk (M. k The complex dimension bk (M) := dimC (HDR (M. C) are isomorphic. First. C) → HDR (M. This shows that the ﬂow of X preserves the metric and the fundamental 2–form. being parallel. We close this section with the following interesting application of Theorem 9. it commutes with the Hodge *–operator. so in particular their intersection is {0}. As we clearly have d ◦ LX = LX ◦ d.1). so LX Ω is exact. . a Proof.4. Let X be a Killing vector ﬁeld. that is LX g = 0. The kernel of f is zero since the spaces of harmonic forms and exact forms are orthogonal. f is well–deﬁned because every harmonic form is closed (Lemma 9. hence f is surjective.50 k Proposition 9. whence d(LX Ω) = LX (dΩ) = 0 and δ(LX Ω) = LX (δΩ) = 0. The isomorphism is simply given by the Hodge *–operator which sends harmonic k–forms to harmonic n − k–forms. C) given by ω → [ω] is an isomorphism.2. there exists a closed form ω such that [ω] = c. for every De Rham cohomology class c. Proposition 9.2. Proof.3. hence X is real holomorphic. too. (Hodge isomorphism). because Ω is coclosed.5. Proof. Every Killing vector ﬁeld on a compact K¨hler manifold is real holomorphic. so it has to vanish by the easy part of Theorem 9. C)) is called the k th Betti number of M and is a topological invariant in view of De Rham’s theorem. We compute the Lie derivative of the fundamental 2–form with respect to X using Cartan’s formula: LX Ω = d(X Ω) + X dΩ = d(X Ω). Thus LX Ω is harmonic and exact. C) and Hn−k (M. showing that f (ω H ) = [ω H ] = [dω + ω H ] = [ω] = c. since the ﬂow of X is isometric. We have seen that the Hodge decomposition of ω is ω = dω +ω H . we see that LX ◦ δ = δ ◦ LX .
q M.q−1 M ⊂ Z p.the second summand in the Dolbeault decomposition of ω vanishes if and only if ¯ ∂ω = 0.q−1 M. can be found in [2]. As before. ·)dv on Ωp. Dolbeault theory.q M = Hp. We consider the ¯ Dolbeault operator ∂ acting on the spaces of (p. Let (M 2m . The space of (p.q M is ∂–harmonic if and only if ∂ω = 0 and ∂ ∗ ω = 0. pp. 84–100.q+1 M and ω H ∈ Hp.q M) ⊂ Ωp+q M. HODGE AND DOLBEAULT THEORY 51 9.q M := C ∞ (Λp. q)–forms Ωp. A (p. We denote by hp.q−1 M. (Dolbeault isomorphism). ¯ We deﬁne the space Hp.q M := ¯ p. ·) := M H(·. Specializing for q = 0 yields Proposition 9. and a proof for the hard part.2. Let p. q)–forms decomposes as a direct sum ¯ ¯ Ωp.q−1 . The proof is very similar to that of Lemma 9. These are the Hodge numbers associated to the complex structure J of M. A form ω ∈ Ωp.q M H p. q)–form ω on M can be uniquely written as ¯ ¯ ω = ∂ω + ∂ ∗ ω + ω H . This is called the Dolbeault decomposition of ω.q the complex dimension of H p.6 shows that the three spaces above are orthogonal with respect to the global Hermitian product (·. This shows that every (p. since it strongly depends on the complex structure J.q M. Corollary 9. q)–forms on M by As before we have Hp.7. Proof.8. 0)–form on a compact Hermitian manifold is holomorphic if and only if ¯ it is ∂–harmonic.q M. h.1 and is left as an exercise.q+1M ⊕ ∂Ωp.q M ⊕ ∂ ∗ Ωp. where ω ∈ Ωp. Lemma 9. ∂Ω M In contrast to De Rham cohomology.q M of ∂–harmonic (p.9. q)–forms. J) be a compact Hermitian manifold. The map f : Hp. The proof is completely similar to the proof of the Hodge isomorphism.q M  ∆∂ ω = 0}.6. ¯ ¯ ¯ ¯ Lemma 9.q M → H p. the Dolbeault cohomology is no longer a topological invariant of the manifold.q ¯ ¯ ¯ Z M denote the space of ∂–closed (p. (Dolbeault decomposition theorem).q M. ω ∈ Ωp. .9. which consists in showing that the direct sum of the three summands is the whole space Ωp.q M := {ω ∈ Ωp. Since ∂ 2 = 0. Theorem 9. we see that ∂Ωp.q M.q M given by ω → [ω] is an isomorphism. We deﬁne the Dolbeault cohomology groups Z p.
q M and Hm−p.p M are non–zero.1 M. In particular (36) shows that all Betti numbers of odd order are even and all Betti numbers of even order are non–zero.q = hm−p. thus proving that the components of a harmonic form in its type decomposition are all harmonic. ∗ ¯ ¯¯ ¯ ¯ ¯∆∂ (ω) = ∗(∂ ∂ ∗ + ∂ ∗ ∂)ω = ∗(∂∂ ∗ + ∂ ∗ ∂)¯ ω ∗ ∗¯ ¯ ¯ ¯ = − ∗ (∂ ∗ ∂ ∗ + ∗ ∂ ∗ ∂)¯ = ∂ ∂(¯ω) − ∗2 ∂ ∗ ∂ ω ω ¯ ∗ ¯ ¯ ∗¯ 2 ∗¯ ∗ ¯ ∗ ¯ ¯ ∗ ¯¯ ∗ = ∂ ∂(¯ω) − ∂ ∗ ∂ ∗ ω = ∂ ∂(¯ω) + ∂ ∂ (¯ω) = ∆∂ (¯ω).3. Moreover. hp.q q.10. This shows that Moreover.m−q M. it commutes with the complex conjugation (in the general case ¯ we only have that ∆∂ α = ∆∂ α) so the complex conjugation deﬁnes an isomorphism between the ¯ p. (Serre duality).p ≥ 1 ∀ 0 ≤ p ≤ m. .q M to Hm−p. We thus have proved the Proposition 9. hp.p . as ∆∂ is a real operator. a ¯ ¯ Firstly. Proof. the fact that ∆ = 2∆∂ shows that Hp. Exercises.q M invariant. Since Ωm is a non– zero multiple of the volume form. hp. identiﬁed with V ∗ via the metric.52 Proposition 9. ω This shows that ¯ is a (C–anti–linear) isomorphism from Hp. Secondly.11. they are all harmonic since Ω is parallel so Ωp is parallel. since ∆∂ leaves the spaces Ωp.p spaces H M and H M.q M.q M ⊂ Hp+q M. (1) Prove that the complex manifold S 1 × S 2k+1 carries no K¨hler metric for k ≥ 1. the following relations hold bee tween Betti and Hodge numbers on compact K¨hler manifolds: a (36) bk = p+q=k ¯ Hk M = ⊕p+q=k Hp. ¯ ∗ If M is K¨hler much more can be said about Hodge and Betti numbers. ¯ ∗(ω) := ∗¯ . Prove that the Lie algebra extension of an endomorphism A of V to Λk V is given by the formula for every orthonormal basis {ei } of V .6. Consider the composition of the Hodge *–operator with the complex conjugation We have ¯ ∗ : Ωp. we deduce that ∆ has the same property. we deduce that all exterior powers Ωp ∈ Ωp. and a parallel form is automatically harmonic. a (2) Let V be an Euclidean vector space.q = hq. A(ω) := A(ei ) ∧ ei ω.m−q M. Consider now the fundamental form Ω ∈ Ω1. In particular hp. too. The spaces Hp.m−q . In addition to Poincar´ and Serre dualities.q M → Ωm−p. due to Theorem 8. 9.m−q M are isomorphic.q .
Show a that if the scalar curvature of M is constant.9. (4) Show that there exists no global K¨hler potential on a compact K¨hler manifold. . in the sense that there exists a real function u such that ¯ ϕ = i∂ ∂u. h) be a compact K¨hler manifold whose second Betti number is equal to 1. a a (5) Let (M. then the metric h is Einstein. HODGE AND DOLBEAULT THEORY 53 ¯ (3) The global i∂ ∂–Lemma. 1)–form on a compact K¨hler manifold a ¯ M. Prove that ϕ is i∂ ∂–exact. Let ϕ be an exact real (1.
.
Part 4 Prescribing the Ricci tensor on K¨hler manifolds a .
k ωik ∧ ωki . Notice that although the coeﬃcients Rij of R depend on the local basis of sections (σi ).1. Let E → M be a complex vector bundle. This shows that Tr(R ) is closed. To compute this explicitly we use the following summation trick: i.i ωki ∧ ωik = − i.12. being locally exact. Z) Z.k ωik ∧ ωki = k. Recall the formula (9) for the curvature of in terms of : (37) R (σi ) =: Rij σj = (dωij − ωik ∧ ωkj )σj . which means that the integral over CP1 of any representative of this class equals −1. its trace is a well–deﬁned (complex–valued) 2–form on M independent of the chosen basis. where the left term denotes the pull–back in cohomology and f ∗ E is the pull–back bundle deﬁned by f ∗ Ex = Ef (x) ∀ x ∈ M. . • (normalization) The ﬁrst Chern class of the tautological bundle of CP1 is equal to −1 in H 2 (CP1 . where of course the trace of the connection form ω = (ωij ) does depend on the local basis (σi ). The following proposition can be taken as a deﬁnition Proposition 10. σk } are local sections of E which form a basis of each ﬁber over some open set U and the connection forms ωij ∈ Λ1 (U) (relative to the choice of this basis) are deﬁned by σi = ωij ⊗ σj . . The comprehensive theory of Chern classes can be found in [8]. where {σ1 .10. ChernWeil theory. where the ﬁrst equality is given by interchanging the summation indices and the second by the fact that the wedge product is skew–symmetric on 1–forms. where E ⊕ F is the Whitney sum deﬁned as the pull–back of the bundle E × F → M × M by the diagonal inclusion of M in M × M. F over M one has c1 (E ⊕ F ) = c1 (E) + c1 (F ). . Ch. We will now explain the Chern–Weil theory allows one to express the images in real cohomology of the Chern classes of E using the curvature of an arbitrary connection on E. • (Whitney sum formula) For every bundles E. From (37) we thus get (38) Tr(R ) = d( ωii ). which is the only one needed in the sequel.1. Z) called the ﬁrst Chern class of E satisfying the following axioms: • (naturality) For every smooth map f : M → N and complex vector bundle E over N. one has f ∗ (c1 (E)) = c1 (f ∗ E). We will outline here the deﬁnition and properties of the ﬁrst Chern class. . . To every complex vector bundle E over a smooth manifold M one can associate a cohomology class c1 (E) ∈ H 2 (M. and can be computed as Tr(R ) = Rii in the local basis (σi ). Chern classes 10.
σi (f (x))) i Tr(R ) 2π σj ) f ∗ (f ∗ σ) := f ∗ ( σ) deﬁnes a connection on f ∗ E (one has to check the classical formulas for basis changes in order to prove that f ∗ is well–deﬁned). R). It is actually easy to see that [Tr(R )] is a purely imaginary class. v ∈ E. more precisely a smooth section in Λ (M) ⊗ End(E). Indeed.3. σj )) = h( σi .1. We have to check that c1 ( ) satisﬁes the three conditions in Proposition 10. Thus Tr(A) is a well–deﬁned 1–form on M and (38) readily implies (39) Tr(R ) = Tr(R ) + dTr(A). in the sense that it has a representative which is a purely imaginary 2–form. then is a basis of local sections of f ∗ E. . The naturality is straightforward. The 2–form Tr(R ) is closed and its cohomology class [Tr(R )] ∈ H 2 (M. the Leibniz rule shows that their diﬀerence A := ˜ − is 1 a zero–order operator. The real cohomology c1 ( ) := is equal to the image of c1 (E) in H 2 (M. v)  x ∈ M. σj ) + h(σi . and with respect to this basis we obviously have f Rij ∗ = f ∗ (Rij ). From (37) we get Rij = dωij − ωik ∧ ωkj = −ωji − ωki ∧ ωjk = −ωji + ωjk ∧ ωki = −Rji . f (x) = π(v)}. C) does not depend on .10. If we start with a local basis {σi } adapted to h. Recall that if f : M → N is smooth and π : E → N is a rank k vector bundle.2. then we have 0 = (δij ) = (h(σi . x → (x. Let class be a connection on a complex bundle E over M. Proof. whence c1 (f ∗ ) = f ∗ (c1 ( )). Theorem 10. let us choose an arbitrary Hermitian structure h on E and take such that h is –parallel. showing that the trace of R is a purely imaginary 2–form. CHERN CLASSES 57 If and ˜ are connections on E. If {σi } is a local basis of sections of E. then f ∗ (E) := {(x. The formula f ∗ σi : M → f ∗ (E). = ωij + ωji. ˜ We thus have proved the following Lemma 10.
¯ This just means ω + ω = d log u. σ)) = h( X σ. X(u) = X(h(σ. For any section σ : CP1 → L of L we denote by σ0 : U0 → C and σ1 : U1 → C the expressions of σ in the standard local trivializations of L. On the other hand. Let L → CP1 be the tautological bundle. given by ψi : π −1 Ui → Ui × C. We now take a particular local holomorphic section σ such that σ0 (z) = 1. σ = ω ⊗ σ. i. We choose a local holomorphic section σ and denote its square norm by u. In polar coordinates z = r cos θ + ir sin θ one can readily compute i ∂2f 1 ∂2f ∂f ¯ ∂∂f = r 2 + dr ∧ dθ. From (38) we get ¯ (40) R = dω = d∂ log u = ∂∂ log u. X σ) = ω(X)u + ω (X)u. ψi (w) = (π(w). We denote by z := φ0 = z0 the holomorphic coordinate on U0 . then we can write: ∀X ∈ T CP1 . ˜ ˜ If {σi }.1 = ∂. {˜j } are local basis of sections of E and F then {σi ⊕ 0.∞)×[0. Let be the Chern connection on L associated to h. whose ﬁrst coordinate is 1. ¯ We thus have to check the following condition: i ¯ ∂∂ log u = −1. 0 ⊕ σj } is a local basis for E ⊕ F σ ˜ ˜ in this basis is a block matrix having R and R ˜ on the principal and the curvature of ⊕ ˜ diagonal. The Hermitian product on C2 induces a Hermitian structure h on L. We ﬁnally check the normalization property. This shows that u = (1. we see ¯ that ω is a (1. 0)–form. σ) + h(σ. σ(z) = (1. If ω is the connection form of with respect to the section σ. z)2 = 1 + z2 . If E and F are complex bundles over M with connections and ˜ then one can deﬁne a connection ⊕ ˜ on E ⊕ F by ( ⊕ ˜ )(σ ⊕ σ )(X) := σ(X) ⊕ ˜ σ (X). Its trace is thus the sum of the traces of R and R . r→∞ 2 1 + r 2 d r [0. we deduce that σ(z) is the unique vector lying on the complex line generated by (z0 . From the deﬁnition of σ0 (as the image of σ through the trivialization ψ0 of L).2π] ∞ i ∂2f ∂f r 2 + dr ∧ dθ 2 ∂r ∂r . wi). z). z1 ) in C2 . 2π CP1 z1 It is clearly enough to compute this integral over U0 := CP1 −{[0 : 1]}. + 2 ∂r r ∂θ2 ∂r Applying this formula to f := log(1 + r 2 ) we ﬁnally get i 2π ¯ ∂∂ log u = CP1 i 2π 1 2 = − ∂f 1 ∂f = lim r ∂r 2 r→∞ ∂r 0 r 2r = − lim = −1.e. Thus ω = ∂ log u.58 The Whitney sum formula is also easy to check. since σ is holomorphic and 0.
(i) Consider any connection in E. . 10. it is enough to check this relation for line bundles E and F . . . we deﬁne the ﬁrst Chern class of M – denoted by c1 (M) – to be the ﬁrst Chern class of the tangent bundle T M. = Any connections E and F on E and F respectively induce a connection on E ⊗ F deﬁned by (σ E ⊗ σ F ) := ( E σ E ) ⊗ σ F + σ E ⊗ ( F σ F ). F be two complex vector bundles over M. . . The corresponding connection forms are then related by ω = ωE + ωF . . then σ := σ1 ∧ . From (37) we then get ˜ Tr(R ) = R = d˜ − ω ∧ ω = d˜ ω ˜ ˜ ω and Tr(R ) = i=j k ˜ ˜ (dωij − ωik ∧ ωkj ) = dωij = dTr(ω) = d˜ . ∧ σi−1 ∧ ( ωij ⊗ σ. ∧ σk is a local non–vanishing section of Λk E. CHERN CLASSES 59 If M is an almost complex manifold. viewed as complex vector bundle: c1 (M) := c1 (T M).10. Properties of the ﬁrst Chern class. where ρ denotes the Ricci form. If σ1 . In the next sections we will see that a representative of the ﬁrst Chern class of a K¨hler manifold a 1 is 2π ρ. . inducing a connection ˜ on Λk E. ˜ ˜ (σ1 ∧ .4. Let ω := (ωij ) and ω be the connection forms of and ˜ relative to these local basis: ˜ σi = ωij ⊗ σj We then compute ˜σ = = = i=j and ˜ σ = ω ⊗ σ. Proposition 10. Proof. (ii) Let us denote by e and f the ranks of E and F . σk denotes a local basis of sections of E. . where E ∗ denotes the dual of E. ∧ σk ) i σ1 ∧ . . Because of the canonical isomorphism Λef (E ⊗ F ) ∼ (Λe E)⊗f ⊗ (Λf F )⊗e . (iii) c1 (E ∗ ) = −c1 (E). Let M be a complex manifold and let E. . j ωij ⊗ σj ) ∧ σi+1 ∧ . . ω i=j thus proving that c1 (E) = c1 (Λ E).2. (i) c1 (E) = c1 (Λk E). where k denotes the rank of E. ∧ σk which proves that ω = Tr(ω). (ii) c1 (E ⊗ F ) = rk(F )c1(E) + rk(E)c1 (F ). .
Y ))∗ . where A∗ ∈ End(E ∗ ) denotes the adjoint of A. ∗ be the induced connection ∗ ∗ X σ )(σ) := X(σ ∗ (σ)) − σ ∗ ( X σ). . But in this case the canonical isomorphism E ⊗ E ∗ C (where C denotes the trivial line bundle) shows that 0 = c1 (C) = c1 (E ⊗ E ∗ ) = c1 (E) + c1 (E ∗ ). 10. since (Λk E)∗ is isomorphic to Λk (E ∗ ).3. R (X. (iii) Again. we can suppose that E is a line bundle. deﬁned by A∗ (σ ∗ )(σ) := −σ ∗ (A(σ)).60 so clearly R = dω = d(ω E + ω F ) = R E +R F . (3) Show that if E is a complex line bundle. Exercises. (1) Consider the change of variables z = r cos θ + ir sin θ. ∂∂f = 2 2 ∂r r ∂θ ∂r (2) Show that the ﬁrst Chern class of a trivial bundle vanishes. there is a canonical isomorphism E ⊗ E ∗ be any connection on a complex bundle E and let (4) Let on the dual E ∗ of E deﬁned by ( Show that ∗ C. Y ) = (R (X. Show that for every function f : U ⊂ C → C the following formula holds: 1 ∂2f ∂f i ∂2f ¯ + r 2 + dr ∧ dθ.
for n = 2m. THE RICCI FORM OF KAHLER MANIFOLDS 61 11. A Riemannian metric is a On –structure. where the multiplication by i corresponds to the tensor J and the Hermitian structure is h − iΩ. v0 ] where u is an arbitrary element of the ﬁber G(M)x . In general.¨ 11. J) be a K¨hler manifold with Ricci form ρ and a canonical bundle K := Λm. The point here is that if G is a closed subgroup of On then there exists at most one torsion–free connection on any G–structure (by the uniqueness of the Levi–Civita connection). . we deﬁne a section in Gl(M) ×ρ V by σ(x) := [u. An orientation on M is a Gl+ (R)–structure. the set {u ∈ Gl(M)  σ = [u.2. v0]} deﬁnes a reduction of the structure group of Gl(M) to G.3. This deﬁnition clearly does not depend on u. the G–structure is geometrical if and only if there exists a torsion–free connection on M with respect to which σ is parallel. G := {g ∈ Gln (R)  ρ(g)(v0 ) = v0 }. the Um structure is geometrical if and only if the tensor deﬁning it (namely J) is parallel with respect to the Levi–Civita connection. which by Theorem 5. As Um = O2m ∩ Glm (C). Let (M 2m .5 just means that h is K¨hler. we will interpret the tangent bundle T M as a complex (actually holomorphic) Hermitian vector bundle over M.2. Let M be an a a n–dimensional manifold and let G be any closed subgroup of Gln (R). if the group G can be deﬁned as the group preserving an element of some Gln representation ρ : Gln (R) → End(V ). Y )ξ = R(X.0 M. let G be given by If G(M) is a G–structure. v0 ]. Definition 11. h. We start by a short review on G– a structures which will help us to characterize K¨hler and Ricci–ﬂat K¨hler metrics. In this setting. We now turn back to our main objects of interest. Y )ξ. From Proposition 5. A topological G–structure on M is a reduction of the principal frame bundle Gl(M) to G.8 we know that the Levi–Civita connection on M coincides with the Chern connection on T M. As before. Conversely. Let us give some examples. then a G–structure is simply a section σ in the associated vector bundle Gl(M) ×ρ V with the same algebraic properties as v0 in the sense that for every x ∈ M there exists u ∈ Gl(M) with σ(x) = [u.1. Lemma 11.1. The Ricci form as curvature form on the canonical bundle. A geometrical G–structure is given by a topological G–structure G(M) together with a torsion–free connection on G(M). where X. K¨hler metrics as geometric U(m)–structures. The curvature R ∈ C ∞ (Λ2 M ⊗ End(T M)) of the Chern connection and the curvature tensor R of the Levi–Civita connection are related by R (X. An almost complex n structure is a Glm (C)–structure. Proposition 11. Y are vector ﬁelds on M and ξ is a section of T M. The Um –structure deﬁned by an almost complex structure J together with a Hermitian metric h on a manifold M is geometrical if and only if the metric is K¨hler. a Proof. The Ricci form of K¨hler manifolds a 11. a 11. To see this.
− (e∗ ∧ e∗ )(X.62 Proof. Ricci–ﬂat K¨hler manifolds.3 we get r ∗ (X. The ﬁve statements below are equivalent: (1) M is Ricci–ﬂat. Y ) = TrR (R(X. h.4 and from Lemma 11.m M. Y ) = iRic(JX. Since we will now use both complex and real traces. Y )) = 2 = −r ∗ (X. Y )). where we used the fact that TrR (AR ◦ J) = 2iTrC (A) for every skew–hermitian endomorphism A.4. Y ) i ei ξ (e∗ ⊗ i ei ξ) = de∗ ⊗ i ei ξ ∗ − e∗ ∧ ej ⊗ i ei ξ ej ei ξ. for simplicity. = (X(Yi ) − Y (Xi ) − ])) ei ξ − (Xi Yj − Xj Yi ) = − [X. We ﬁx some notations: let r and r ∗ be the curvatures of the Chern connections of K := Λm. provided we make explicit the deﬁnition of R . J) be a K¨hler manifold with canonical a a bundle K (endowed with the Hermitian structure induced from the K¨hler metric on T M) and a Ricci form ρ. we will make this explicit by a superscript. Y )) = Tr(R(X. Y )ξ = de∗ (X. Let (M 2m . We suppose. so from the proof of Proposition 10. Y )) = −TrC (R(X. Y ) ◦ J) 2 i (2iTrC (R(X.Y ] ξ + (X(Yi ) − Y (Xi )) ei ξ − Xi Y ei ξ + Yi = − [X. Then R ξ= 2 ξ= Denoting Xi := e∗ (X) and Yi := e∗ (Y ) we then obtain i i R (X. Moreover. Y ).5. Y ) = Tr(R (X. 11.0 M and K ∗ := Λ0. Proof.2 we then obtain i iρ(X. It is easy to check that Λm (T M) is isomorphic to K ∗ . The curvature of the Chern connection of the canonical line bundle K is equal to iρ acting by scalar multiplication on K. Y i ej ej X ei ξ ei ξ We are now ready to prove the following characterization of the Ricci form ρ on K¨hler manifolds: a Proposition 11. They are related by r = −r ∗ (exercise). Y )ξ.3. the connection induced on Λm (T M) with the induced Hermitian structure by the Chern connection on T M is clearly the Chern connection of Λm (T M). The proof is tautological. Y ) = r(X.Y ] ξ − Y X ξ + X Y ξ = R(X. By Proposition 6. (2) The Chern connection of the canonical bundle K is ﬂat. that M is simply connected. Then the previous results can be summarized as follows: Theorem 11. Let {ei } ∗ denote a local basis of vector ﬁelds on M and let {ei } denote the dual local basis of Λ1 M. Y ) i j e∗ ([X. .
3 and Proposition 11. The converse statement is also true if M is compact. for a given K¨hler manifold (M. a parallel section of Λm.4. h. the torsion–free connection deﬁning it is just the Levi–Civita connection. Thm. the holonomy of a connection in G(M) is contained in G. (3) Show that a Um –structure on M deﬁnes an almost complex structure together with a Hermitian metric. if the Riemannian holonomy Hol(M) of M is a subgroup of SUm . In the non–simply connected case. Thus. Show that every G–structure is geometrical. J) is a necessary condition for the existence of a Ricci– ﬂat K¨hler metric on M compatible with J. (4) Show that a geometrical Glm (C)–structure is the same as an integrable almost complex structure. (1) ⇐⇒ (2) is a direct consequence of Proposition 11. we get a geometrical SUm –structure simply by extending the holonomy bundle to SUm . (4) M has a geometrical SUm –structure. the set of frames obtained from u by parallel transport) is a Holu (M)– principal bundle. (3) ⇐⇒ (4) The special unitary group SUm can be deﬁned as the stabilizer of a vector in the canonical representation of Um onto Λm. (1) Let G be a closed subgroup of Gln (R) containing SOn .1) shows that for every ﬁxed frame u. a and will be treated in the next section. if M has a geometrical SUm –structure. and the Levi–Civita connection can be restricted to it.0 M. Notice that by Theorem 10. (5) =⇒ (4) The reduction theorem ([8]. that is. THE RICCI FORM OF KAHLER MANIFOLDS 63 (3) There exists a parallel complex volume form. J).4. there exists a parallel section in Λm. 11. therefore the Riemannian holonomy group is a subgroup of SUm . Hint: Start with a torsion free connection and consider the connection ˜ . Ch. (5) The Riemannian holonomy of M is a subgroup of SUm . the holonomy bundle (that is. Proof. and locally deﬁned otherwise). Now. (2) ⇐⇒ (3) follows from the general principle that a connection on a line bundle is ﬂat if and only if there exists a parallel section (globally deﬁned if π1 (M) = 0. (2) Let M n be a connected diﬀerentiable manifold. Thus. (4) =⇒ (5) If G(M) is a G–structure. 7.0 C. the last 3 statements are only local. Prove that M is orientable if and only if its frame bundle Gln (M) is not connected.4. 2. Exercises.0 M if and only if the geometrical Um –structure deﬁned by the K¨hler metric can be further reduced to a a geometrical SUm –structure. the a vanishing of the ﬁrst Chern class of (M.¨ 11.
there is a free transitive group action of C ∞ (Λ1 M) on the space of Hermitian connections on L. Prove that SUm is equal to the stabilizer in Um of the form dz1 ∧ . (5) Let A be a skew–hermitian endomorphism of Cm and let AR be the corresponding real endomorphism of R2m . (6) The special unitary group SUm is usually deﬁned as the subgroup of Um ⊂ Glm (C) consisting of complex unitary matrices of determinant 1. h) be a complex line bundle with Hermitian structure over some smooth manifold M. . where AX Y = 4 (2J( X J)Y + ( JY J)X + J( Y J)X). Prove that the space of Hermitian connections is an aﬃne space over the real vector space C ∞ (Λ1 M). . (7) Let (L. ∧ dzm . Show that TrR (AR ◦ J) = 2iTrC (A). Use the proof of Lemma 5. (8) If L is a complex line bundle over M. . R) is 2π times the curvature of some connection on L.64 1 deﬁned by ˜ X Y := X Y − AX Y . Equivalently. show that every real closed 2–form in the cohomology i class c1 (L) ∈ H 2 (M.4 to check that A is symmetric if and only if J is integrable.
Then for every closed real (1. ω).6 below. Let M m be a compact K¨hler manifold with fundamental form ϕ and Ricci form a ρ. Since [ϕ] = [ϕ1 ] we also have [ϕm ] = [ϕ1 ]. By the Calabi theorem M has a K¨hler metric with positive Ricci curvature.1. Before giving an outline of the proof. ¯ It is easy to check that the Hodge operator acts on Λm. The Calabi conjecture 12. An overview.12.0 simply by scalar multiplication with m(m+1) ε := im (−1) 2 . If the ﬁrst Chern class of a compact K¨hler manifold is positive. The global i∂ ∂–Lemma shows that (41) H= ¯ u ∈ C ∞ (M)  ϕ + i∂ ∂u > 0.3.1 is to reformulate the problem in order to reduce it e a to a so–called Monge–Amp`re equation. Now. we state some corollaries. Proof. ω)dv. Since iρ is the curvature of the canonical bundle KM . If the ﬁrst Chern class of a compact K¨hler manifold vanishes. a a 1 1 we denote by dv := m! ϕm and dv1 := m! ϕm their volume forms and consider the real function f 1 m deﬁned by ef dv = dv1 . then M carries a a Ricci–ﬂat K¨hler metric. that is (42) M ef dv = M dv. ω) ¯ and iρ1 = ∂∂ log g1 (ω. ¯ ¯ ¯ .2. 1)–form ρ1 in the cohomology class 2πc1 (M). a Corollary 12. uϕm = 0 M (this last condition is needed since u is only deﬁned up to a constant). we have the following famous result due to Calabi and Yau Theorem 12. We have seen that the ﬁrst Chern class of any compact K¨hler manifold a 1 is represented by 2π ρ. THE CALABI CONJECTURE 65 12. if g and g1 are K¨hler metrics with K¨hler forms ϕ and ϕ1 in the same cohomology class.1. there exists a unique K¨hler metric with fundamental form ϕ1 in the same cohomology class as ϕ. Conversely. The ﬁrst step in the proof of Theorem 12. for every local holomorphic section ω of KM we have ¯ ¯ (43) iρ = ∂∂ log g(ω. We denote by H the set of K¨hler metrics in the same ¯ cohomology class as ϕ. so the a result follows from Theorem 15. Corollary 12. Let ρ and ρ1 denote the corresponding Ricci forms. We thus have (44) and similarly (45) εω ∧ ω = ω ∧ ∗¯ = g(ω. ω)dv ¯ ω ¯ εω ∧ ω = g1 (ω. then M is a simply connected. whose Ricci form is a exactly ρ1 . ω)dv1 = ef g1 (ω.
The mapping Cal : H → H deﬁned by ¯ (ϕ + i∂ ∂u)m Cal(u) = log ϕm ¯ iρ1 − iρ = ∂ ∂f. em . . We ﬁrst show that Cal is injective.4. . (47) ρ1 = ρ − i∂ ∂f.66 From (43)–(45) we get (46) ¯ This shows that the Ricci form of the modiﬁed K¨hler metric ϕ1 = ϕ + i∂ ∂u can be computed by a the formula ¯ (ϕ + i∂ ∂u)m ¯ . 1 ¯ Using the formula 2i∂ ∂ = ddc and the fact that ϕ and ϕ1 are closed forms we get after multiplication by u m−1 m−1 ¯ 0 = 2iu∂ ∂u ∧ = d ud u ∧ c ϕk 1 k=0 m−1 ∧ϕ m−k−1 = udd u ∧ c k=0 c ϕk ∧ ϕm−k−1 1 m−1 ϕk 1 k=0 ∧ϕ m−k−1 − du ∧ d u ∧ k=0 ϕk ∧ ϕm−k−1 . and since 2–forms commute we obtain 1 m−1 0= ϕm 1 ¯ − ϕ = i∂ ∂u ∧ m k=0 ϕk ∧ ϕm−k−1 . where f = log ϕm ¯ Now given closed real (1. 1 Integrating over M and using Stokes’ theorem yields m−1 (48) 0= k=0 M du ∧ Jdu ∧ ϕk ∧ ϕm−k−1 . the global i∂ ∂–Lemma ¯ shows that there exists some real function f such that ρ1 = ρ − i∂ ∂f . Moreover. Je1 . The Calabi conjecture is then equivalent to the following Theorem 12. 1)–form ρ1 in the cohomology class 2πc1 (M). We denote by H the space of smooth functions on M satisfying this condition. since ϕ1 deﬁnes a K¨hler metric. Jem } orthonormal a with respect to g such that m m ϕ= j=1 ej ∧ Jej and ϕ1 = j=1 aj ej ∧ Jej . there exists a local basis {e1 . If Cal(u) = 0 we have ϕm = ϕm . . is a diﬀeomorphism. . f is unique if we impose the normalization condition (42). 1 Now. . It is clearly enough to show that Cal(u) = 0 and u ∈ H implies u = 0.
We refer the reader to [6] for details.. we compute its diﬀerential at some u ∈ H.0 M on a Hermitian manifold M of complex dimension m. then ρ ∧ ϕm−1 = (m − 1)!S. so the Inverse Function Theorem shows that Cal is a local diﬀeomorphism.jk =j j1 <. we may suppose without loss of generality that u = 0. .. which is the hard part of the theorem. THE CALABI CONJECTURE 67 where aj are strictly positive local functions. follows from a priori estimates. ajk . so u = 0 because the integral of u dv over M vanishes. a . .. (1) Show that ∗ω = im(m+2) ω for all ω ∈ Λm. Thus Cal∗ is bijective.2.12.<jk This shows that the integrand in (48) is strictly positive unless du = 0. This shows easily that for every k m ϕk 1 ∧ϕ m−k−1 = ∗( j=1 bk ej ∧ Jej ).. By changing the reference metric if necessary. (3) Prove that the total volume of a K¨hler metric on a compact manifold only depends on a the cohomology class of its fundamental form. m ϕ From the general elliptic theory we know that the Laplace operator is a bijection of the space of functions with zero integral over M. To prove that it is a local diﬀeomorphism. (4) Show that ∗(ϕm−1 ) = (m − 1)!ϕ on every Hermitian manifold M of complex dimension m. j j In fact one can compute explicitly bk = k!(m − k − 1)! j aj1 . For v ∈ T0 H we compute ¯ d d (ϕ + i∂ ∂tv)m Cal∗ (v) = t=0 (Cal(tv)) = t=0 dt dt ϕm ¯ i∂ ∂v ∧ ϕm−1 ¯ ¯ ¯ ¯ = m = Λ(i∂ ∂v) = −∂ ∗ ∂v = −∆∂ v. where ρ denotes a the Ricci form and S is the scalar curvature of the K¨hler metric deﬁned by ϕ. Thus u is a constant. 12... Exercises. Therefore Cal is injective. prove that if (M. j1 =j. bk > 0. (2) Prove that the mapping ¯ u → ϕ + i∂ ∂u is indeed a bijection from the set deﬁned in (41) to the set of K¨hler metrics in the a cohomology class [ϕ]. which show that Cal is proper. The surjectivity of Cal. ϕ) is K¨hler. Using this.
J) be a compact complex manifold. More precisely one has M Sdv = 2πmc1 (M) ∪ [ϕ]m−1 . . Show that the integral over M of the scalar curvature of a K¨hler metric only depends on the cohomology class of its fundamental a form ϕ.68 (5) Let (M m .
This shows that we may suppose that λ = ε = ±1. a We will treat simultaneously the two cases ε = ±1. so neither does the Ricci tensor. We are looking for a new K¨hler metric g1 with fundamental form ϕ1 and Ricci form ρ1 such that a ρ1 = εϕ1 . Now the previously obtained formula (47) for the Ricci form of the new metric reads ¯ m ¯ log (ϕ + i∂ ∂u) . KAHLER–EINSTEIN METRICS 69 13. K¨hler–Einstein metrics a 13. g) satisfying the Einstein condition Ric = λg. this condition turns out to be also suﬃcient: Theorem 13. In the negative case. We can suppose without loss of generality that this form is equal to ϕ (otherwise we just change the initial K¨hler metric). there exists a positive closed (1.¨ 13. ϕm . in order to emphasize the diﬃculties that show up in the case ε = 1. ¯ From this equation and the global i∂ ∂–Lemma it is clear that there exists a unique function u ∈ H ¯ such that ϕ1 = ϕ + i∂ ∂u. The K¨hler–Einstein condition a reads ε = ±1. We will exclude the case λ = 0 which was treated above. which was deﬁned as a trace. we ﬁrst reformulate the problem. Let (M 2m . As before. (Aubin. we see that a necessary condition for the existence of a K¨hler–Einstein manifold on a given compact K¨hler manifold is that its ﬁrst a a Chern class is deﬁnite (positive or negative). (50) ρ1 = ρ − i∂ ∂ ϕm Using (49) and (50). ρ) be a compact K¨hler manifold a with deﬁnite ﬁrst Chern class c1 (M). J. the curvature tensor does not change. Then [ϕ] = 2πεc1 (M) = a ¯ [ερ]. g. Suppose we have such a metric. ρ = εϕ. so the global i∂ ∂–Lemma shows that there exists some function f with (49) ¯ ρ = εϕ + i∂ ∂f.1. λ ∈ R.1. ρ As the ﬁrst Chern class of M is represented by 2π . 1)–form representing the cohomology class 2πεc1 (M). Yau) A compact K¨hler manifold with negative ﬁrst Chern class admits a a unique K¨hler–Einstein metric with Einstein constant ε = −1. ϕ. If we rescale the metric by a positive constant. The Aubin–Yau theorem. By deﬁnition. We turn our attention to compact K¨hler manifolds a (M. From our choice for ϕ we have [2πϕ] = εc1 (M) = [2περ1 ] = [2πϕ1 ]. the K¨hler–Einstein condition for g1 becomes a (51) ¯ (ϕ + i∂ ∂u)m ¯ ¯ εϕ + i∂ ∂f − i∂ ∂ log = εϕ1 .
2 since the Hessian H u of u is of course negative semi–deﬁnite at a point where u reaches its maximum. The injectivity of Cal− can be proved as follows. then the K¨hler metric ϕ1 := ϕ+i∂ ∂u is K¨hler– a a ∞ ¯ > 0). Einstein (we denote by C+ (M) the space of all smooth functions u on M such that ϕ + i∂ ∂u The Aubin–Yau theorem is therefore equivalent to the fact that the mapping ∞ Calε : C+ (M) → C ∞ (M) Calε (u) := Cal(u) + εu is a diﬀeomorphism. ϕm 1 hence. . the (1. JX) = (X(dc u(JX)) − JX(dc u(X))) 2 2 1 u = (H (X.70 which is equivalent to (52) log ¯ (ϕ + i∂ ∂u)m + εu = f + const. thus proving the injectivity of Cal− . X) + H u (JX. JX)) ≤ 0. u ≥ 0 at each minimum points. 1)–form i∂ ∂u is negative semi–deﬁnite. ϕm ∞ ¯ Conversely. ∞ We have already computed the diﬀerential of Cal at u = 0 applied to some v ∈ T0 C+ (M): ¯ Cal∗ (v) = −∆∂ v. so u ≤ 0 on M. so ﬁnally u = 0 on M. Then log ϕm ϕm 2 1 − u1 = log m − u2 . the surjectivity of Cal− is harder to prove and requires non–trivial analysis (see [6]). if u ∈ C+ (M) satisﬁes this equation. m ϕ ϕ ¯ (ϕ1 + i∂ ∂u)m = u. denoting the diﬀerence u2 − u1 by u: (53) log ¯ At a point where u attains its maximum. Suppose that Cal− (u1 ) = Cal− (u2 ) and denote ¯ ¯ by ϕ1 := ϕ + i∂ ∂u1 and ϕ2 := ϕ + i∂ ∂u2 . Taking into account (53) we see that u ≤ 0 at each of its maximum points. Consequently 1 is a bijection of C ∞ (M) since the self–adjoint elliptic operator v → 2 ∆v + v has obviously no kernel and its index is zero. JX) = 1 1 (ddc u)(X. − Cal∗ (v) = −v − ∆∂ v ¯ As before. Similarly. since we can write (for any vector X parallel at that point) ¯ i∂ ∂u(X.
the harmonic part of ξ. .2. i. where f and h are functions with vanishing integral and ξ H is the harmonic part of ξ in the usual Hodge decomposition. X] for every vector ﬁeld X. 1). JY ) = dξ(X. Finally. Proof. ζ) on a compact K¨hler–Einstein manifold M 2m with positive a scalar curvature S is Killing (resp holomorphic) if and only if ξ = Jdh (resp. Since ξ is holomorphic we have Lξ J = 0. this yields ∆f = m f + c1 and ∆h = m h + c2 . and since ∆ commutes with d and dc . 2m Since S > 0. this shows that there are no harmonic 1–forms on M. The Ricci tensor of M satisﬁes Ric(X) = 2m X for every vector X.¨ 13. dc C ∞ (M) and δΩ2 (M) shows that ξ0 equals ξ H . The form ξ − dc h is closed. together with the fact that dC ∞ (M) and dc C ∞ (M) are L2 –orthogonal. Let M 2m a be a compact K¨hler manifold. Comparing this formula with the Hodge decomposition for ξ and using the fact that harmonic 1–forms are L2 –orthogonal to dC ∞ (M). so taking the scalar product with some JY and skew–symmetrising yields dξ(JX. S S so ∆(df + dc h) = d( m f ) + dc ( m h). Then ξ can be decomposed in a unique manner as ξ = df + dc h + ξ H . Now.3. Let ξ be a vector ﬁeld on M. If we view as usual T M as a holomorphic vector bundle. and the images of d and S S dc are L2 –orthogonal. S Proof. S where f and h have vanishing integrals over M. then the Weitzenb¨ck formula o (see (65) below) yields S ¯ ¯ ξ. KAHLER–EINSTEIN METRICS 71 13. the uniqueness of f and h follows easily from the normalization condition. Next. (54) 2∂ ∗ ∂ξ = ∗ ξ + iρξ = ∗ ξ − Ric(ξ) = ∗ ξ − 2m The Bochner formula (Exercise 3 in the next section) reads S (55) ∆ξ = ∗ ξ + Ric(ξ) = ∗ ξ + ξ. A vector ﬁeld ξ (resp. f and h) are eigenfunctions of the Laplace operator corresponding to the eigenvalue m . Thus JX ξ = J X ξ. ζ = df + dc h) where S h (resp. The global ddc –Lemma shows that dξ = ddc h for some function h. Finally the constants have to vanish because of the normalization condition. we have the following characterization of real holomorphic and Killing vector ﬁelds on compact K¨hler–Einstein manifolds with positive scalar curvature.2. ξ − dc h = df + ξ0 Suppose that ζ is holomorphic. Let ξ be a holomorphic (real) vector ﬁeld with dual 1–form also denoted by ξ. From Lemma 13.2 we then can write ζ as a sum ζ = df + dc h. We start by showing the following a Lemma 13. Holomorphic vector ﬁelds on compact K¨hler–Einstein manifolds. dξ is of type (1. so the Hodge decomposition theorem shows that for some function f and some harmonic 1–form ξ0 . a Lemma 13. JX] = J[ξ.e. subtracting (54) from (55) yields ∆ζ = m ζ. Y ). so [ξ.
and moreover δ anti–commutes with dc . The Lie algebra g(M) of Killing vector ﬁelds on a compact K¨hler–Einstein mana ifold M with positive scalar curvature is a real form of the Lie algebra h(M) of (real) holomorphic vector ﬁelds on M. The codiﬀerential of every Killing vector ﬁeld vanishes. There exist compact K¨hler manifolds with positive ﬁrst Chern class whose Lie algebra of holomora phic vector ﬁelds is not reductive. Therefore such a manifold carries no K¨hler–Einstein metric. a thus showing that Theorem 13. F (ξ1 + iζ1 )]. is reductive. Conversely. Let F : g(M) ⊗ C → h(M) be the linear map given by F (ξ + iζ) := ξ + Jζ. ξ1] = [ξ. ξ1]) = [ξ. ξ1] − [ζ. suppose that ξ = df + dc h and f and h are eigenfunctions of the Laplace operator S S corresponding to the eigenvalue m . so from (55) we get S ξ= m Then (54) shows that ξ is holomorphic. Jζ1 ] + [Jξ. Thus 0 = δξ = δdf . we have where ϕ is the fundamental form of M. showing that df = 0. as well as its complexiﬁcation. Together with Lξ J = 0. ξ1 + iζ1 ]) = [ξ.4. S so ξ = dc h with ∆h = m h. ζ1] + [Jξ. ζ1] + [ζ. The two lemmas above clearly show that F is a vector space isomorphism. If moreover df = 0.72 If ξ is Killing. F is well–deﬁned.e. We are now ready to prove the following result of Matsushima: Theorem 13. this shows that Lξ g = 0. Moreover. Then ∆ξ = m ξ.1 cannot hold in the positive case. i. so F ([ξ + iζ. Since J maps holomorphic vector ﬁelds to holomorphic vector ﬁelds. F is a Lie algebra morphism because Killing vector ﬁelds are holomorphic. In particular h(M) is reductive. it is the direct sum of its center and a semi–simple Lie algebra. Proof. 2m . ξ1] + [Jζ. Lξ ϕ = d(ξ ϕ) + ξ dϕ = d(Jξ) = −ddh = 0. ξ1 ] = [F (ξ + iζ). ζ1 ] + [Jζ. ζ1] + J([ξ. so ξ is Killing. ξ1] + J 2 [ζ. The last statement follows from the fact that the isometry group of M is compact. ζ1] + [Jζ.5. and every Lie algebra of compact type. ∗ ξ+ S ξ. then ξ is holomorphic by Proposition 9.
Part 5 Vanishing results .
J). depending only on the curvature of the Chern connection: ¯ ¯ ¯¯ (56) 2(∂ ∗ ∂ + ∂ ∂ ∗ ) = ∗ + R. hF ) are Hermitian bundles.q M ⊗ E) → a ∞ p.14. .0M ⊗ E. o The general principle is the following: let (E. The aim of the next 3 sections is to derive vanishing o results under certain positivity assumptions on the curvature using Weitzenb¨ck techniques. ¯ The relation (57) is more or less tautological. then every holomorphic section of Λp. ∗ and ∗ are given locally by 1 ¯ ¯ (57) ∂ : C ∞ (Λp. Weitzenb¨ck techniques o 14. as well as its prolongation to Λp.q M ⊗ E) σ= σ− ej ej σ.q ∞ 1 ¯ C (Λ M ⊗ E) and Chern connection : C (Λ M ⊗ E) → C (ΛC M ⊗ Λp. taking the scalar product with σ and integrating over M. by a slight abuse of notation. We start with the following technical lemma: Lemma 14. 2 (58) (59) (60) ¯ ∂ ∗ : C ∞ (Λp. hE⊗F (σ E ⊗ σ F . sF ).q M ⊗ E). and if R is strictly positive on Λp. Proof.q−1M ⊗ E) ∗ 1 ¯ ∂ ∗ σ = − (ej + iJej ) 2 ∗ ∗ ej (σ).0 M ⊗ E. where R is a section in End(Λp. : C ∞ (Λ1 M ⊗ Λp. it turns out that the diﬀerence of the diﬀerential operators are the formal adjoints of ∂ ¯ ¯ ¯¯ of order two ∗ and 2(∂ ∗ ∂ + ∂ ∂ ∗ ) is a zero–order operator.0M ⊗ E.q M ⊗ E).q M ⊗ E) → C ∞ (Λp.0M ⊗ E is –parallel.0M ⊗ E.q M ⊗ E) → C ∞ (Λp. If R is a positive operator on Λp.q+1 ∞ p. ω σ. We will use the same symbol H for this Hermitian structure. the wedge product there only concerns the Λp. Of course. If (E.1. sE ⊗ sF ) := hE (σ E . identiﬁed via the metric g with a (0. and denotes the Chern connection of E. If {ej } is a local orthonormal basis in T M (identiﬁed via the metric g with an orthonormal basis of Λ1 M).q+1M ⊗ E) ∂σ = (ej − iJej ) ∧ ej (σ). sE )hF (σ F . 0)–vector.1. ¯ using the fact that ∂ ∗ vanishes identically on Λp. on Λp. This follows by applying (56) to some holomorphic section σ of Λp.q M ⊗ E with respect to which one deﬁnes the adjoint operators above is obtained in this way from the Hermitian structure h of E and the Hermitian structure H of Λp.q M ⊗ E) C ∗ (ω ⊗ σ) = (δω)σ − ej ej : C ∞ (Λp. 1)–form. g. with holomorphic structure ∂ : C ∞ (Λp. then ¯ ¯ ∂.q M given by (31). If ∂ ∗ and ∗ ¯ and . hE ) and (F. The Weitzenb¨ck formula. their tensor product inherits a natural Hermitian structure given by The Hermitian structure. using the deﬁnition of ∂ and the fact that ej − iJej is a (1.q M ⊗ E) → C ∞ (Λp. ∂ ∗ . then this holomorphic bundle has no holomorphic section.q M–part of σ.q M ⊗ E using the Levi–Civita connection on the left–hand side of this tensor product. h) → M be some holomorphic Hermitian bundle ¯ over a compact K¨hler manifold (M 2m .q M ⊗ E) → C ∞ (Λp.
q M ⊗ E induced by the Levi–Civita connection on Λp.¨ 14.q−1M ⊗ E) we deﬁne the 1–form α by 1 α(X) := H((X + iJX) σ. s) and compute −δα = ej (α(ej )) = −H((δω)σ. s) + H(σ. s) + H(σ. Y ) ∈ End(Λp. s) + H((ω(ej ))σ. s). whence ∗ ej σ. For vectors X. then we get at that point: 1 1 −δα = ej (α(ej )) = H((ej + iJej ) ej σ. J) be a holomorphic Hermitian bundle over a K¨hler a ˜ manifold M. h) → (M 2m .q M ⊗ E) and s ∈ C ∞ (Λp. s) + H((ω(ej )) = H( ω σ − (δω)σ. ∂s). ek )σ). WEITZENBOCK TECHNIQUES 75 For σ ∈ C ∞ (Λp. 2 By choosing the local basis {ej } parallel in a point for simplicity. (ej − iJej ) ∧ ej s) 2 2 1 ¯ = H((ej + iJej ) ej σ. s) + H(ω ⊗ σ. ej σ We are now ready for the main result of this section Theorem 14. 2 1 ¯ thus showing that the operator − 2 (ej + iJej ) ej is the formal adjoint of ∂.q M ⊗ E) we C deﬁne the 1–form α by α(X) := H((ω(X))σ. ej ej σ = ej σ σ− − ej ej ej σ. let R(X. ej )σ − (ej − iJej ) ∧ (ek + iJek ) (R(ej . 2 2 1 1 = H((ej + iJej ) ej σ.q M ⊗ E) and s ∈ C ∞ (Λp. ej s) The proof of (59) is similar: for ω ⊗ σ ∈ C ∞ (Λ1 M ⊗ Λp. we apply (59) to some section ∗ (ω ⊗ σ) = (δω)σ − ω σ. Y ∈ T M.q M ⊗ E) be the curvature operator of the tensor product connection on Λp. s) + H((ej + iJej ) σ. g.2. where R is the section of End(Λp.q M ⊗ E) deﬁned by (62) i˜ 1 ˜ R(σ) := R(Jej . Then the following formula holds (61) ¯ ¯ ¯¯ 2(∂ ∗ ∂ + ∂ ∂ ∗ ) = ∗ + R. s). σ = ej ⊗ ej ej σ ej σ ej σ of Λ1 M ⊗ Λp. Let (E. ej s) Finally. 2 2 .q M ⊗ E and get C ek ej ) ej ej σ = (δej ) ej σ − = g( ek ek .q M and the Chern connection on E. ej ) − = −g(ek .
Y )(ξ).2 and (64) yield the desired result. Therefore Theorem 14. since the last term in (62) automatically vanishes. We are now ready to obtain the vanishing results mentioned above. ej ) + ig(Jek . The curvature R of Λp. (57). The proof is a simple computation in a local orthonormal frame parallel at a point. for every section ω ⊗ ξ of Λp. Y )ω) ⊗ ξ + ω ⊗ RE (X. ej ) as endomorphisms of the tangent space of M. ej ) 2 ∗ ¯¯ = − 2∂ ∂ ∗ + R. ej )) ek ej + (ej − iJej ) ∧ (ek + iJek ) 2 1 ∗ = − ig(Jek .1. Y )(ω) = R(X. ej ) 2 i 1 ∗ ˜ (ej − iJej ) ∧ ej ((ek + iJek ) = − g(Jek . . The expression of the curvature term becomes then particularly simple.2.76 Proof. Y )ek ∧ ek ω. Y )(ω ⊗ ξ) = (R(X.3.0 given by ρ(p) ω := ρ(ej ) ∧ ej ω. It is an easy exercise to check that the Riemannian curvature operator acts on forms by R(X.2) (i) we have 2ρ = R(Jej .(58) and (60): ¯ ¯ 2∂ ∗ ∂ = − 1 (ek + iJek ) ek ((ej − iJej ) ∧ ej ) 2 1 = − (ek + iJek ) ((ej − iJej ) ∧ ek ej ) 2 1 = −(g(ek . It is easy to check that this action preserves the space Λp. ˜ Proof. 2 E where R is the curvature of E. ej )ξ. If q = 0. Vanishing results on K¨hler manifolds. ej )R(ek . From Proposition (6. ej ) + 2 2 1 ˜ + (ej − iJej ) ∧ (ek + iJek ) R(ek . Most of the applications will concern the a case q = 0. using Lemma 14.0M ⊗ E decomposes in a sum ˜ (64) R(X.0 . where R is the Riemannian curvature. Proposition 14. Let ρ(p) denote the action of the Ricci–form of M on Λp. ej ) ek ej + (ej − iJej ) ∧ (ek + iJek ) ej 2 1 ˜ + (ej − iJej ) ∧ (ek + iJek ) R(ek . ek ej ek ek ) 14.0M ⊗ E we have i ¯ ¯ (63) 2∂ ∗ ∂(ω ⊗ ξ) = ∗ (ω ⊗ ξ) + i(ρ(p) ω) ⊗ ξ + ω ⊗ RE (Jej .
1)–vector and ω vanishes. 0)–form ω. The interior product of a (0.¨ 14. its extension to (p. we have i ¯ ¯ (65) 0 = 2∂ ∗ ∂ξ = ∗ ξ + R(Jej . Proof. ∀ ω ∈ Ωp M. (1) Show that the extension to Λp M ⊗ C of a positive deﬁnite symmetric endomorphism of T M is positive deﬁnite. Exercises. Let M be a compact K¨hler manifold. Proof. then a every holomorphic form is parallel. If the Ricci curvature of M vanishes. Suppose now that Ric is positive deﬁnite. Let us take p = 0 and E = T 1. Taking the Hermitian product with ω and integrating over M yields the result. hence taking the Hermitian product with ω in (66) and integrating over M yields M  ω2 + H(Ric(ω). Thus. From (63) applied to some holomorphic (p.3. If ξ is a holomorphic vector ﬁeld. We take E to be trivial and apply (63) to some holomorphic (p. using the fact that ρξ = Ric(Jξ) = iRic(ξ) yields 0= M H( ∗ ξ − Ric(ξ). ej )ξ = ∗ ξ + iρ(ξ). .3. if Ric is negative deﬁnite.0 M in Proposition 14. 2 Taking the (Hermitian) scalar product with ξ in this formula and integrating over M. X) < 0 for all non–zero X ∈ T M) then M has no holomorphic vector ﬁeld. showing that ω has to vanish 14.5. 0)–forms on M for p > 0. ω)dv = 0. showing that JX ω = iX ω. If the Ricci curvature of M is positive deﬁnite. 0)–form ω we get (66) 0= ∗ ω + iρ(p) (ω). ξ)dv = M  ξ2 − H(Ricξ. Ric(X.4. We thus get iρ(p) (ω) = iρ(ej ) ∧ ej ω = iρ(Jej ) ∧ Jej ω = −ρ(Jej ) ∧ ej ω = Ric(ω).e. Since ρ = 0 we get 0 = ∗ ω. ξ has to vanish identically. Since Ric is positive. too. then there exist no holomorphic (p. ξ)dv. If the Ricci curvature of M is negative a deﬁnite (i. Let M be a compact K¨hler manifold. Theorem 14. (2) Prove the following real version of the Weitzenb¨ck formula: o ∆ω = ∗ ω + Rω. WEITZENBOCK TECHNIQUES 77 Theorem 14. 0)–forms is positive.
ek )(ω)). (4) Prove that there are no global holomorphic forms on the complex projective space.78 where R is the endomorphism of Ωp M deﬁned by (3) Applying the above identity to 1–forms. ∀ ω ∈ Ω1 M. . prove the Bochner formula ∆ω = ∗ R(ω) := −ej ∧ ek (R(ej . ω + Ric(ω).
In order to get a feeling for this notion. A real (1. 1)–form. The positivity of a holomorphic line bundle is a topological property on K¨hler manifolds: a Lemma 15. In order to state another application of the Weitzenb¨ck foro mula we have to make the following Definition 15. thus showing that L is positive. From the global i∂ ∂–Lemma.2. J) is called positive (resp.1. Suppose. The Hirzebruch–Riemann–Roch formula 15. σ) = −i∂ ∂ log h(σ. negative) (1.15. A negative holomorphic line bundle L over a compact K¨hler manifold has no non–vanishing holomorphic section. Thus K is a negative if and only if the Ricci tensor is positive deﬁnite.1 M ∩ H 2 (M. that c1 (L) is positive. ˜ It is then clear that the curvature of the Chern connection ˜ associated to h := heu satisﬁes for every local holomorphic section σ: ˜ ¯ ˜ ¯ ¯ ¯ iR = −i∂ ∂ log h(σ. negative) deﬁnite. A holomorphic line bundle L over a compact K¨hler manifold M is positive if and a only if its ﬁrst Chern class is positive. R) is called positive (resp. Y ) is positive (resp. Positive line bundles. negative) if there exists a Hermitian structure on L with Chern connection and curvature form R such that iR is a positive (resp.3. From Lemma a 11. A holomorphic line bundle L over a compact complex manifold is called positive (resp. negative) if the symmetric tensor A satisfying A(JX. 2 . notice that the fundamental form of a K¨hler manifold is a positive. ej )ξ. g. there exists a real function u such that iR = ω + i∂ ∂u. conversely. negative) if it can be represented by a positive (resp. One direction is clear from the deﬁnition.4 we know that the canonical bundle K of a K¨hler manifold has curvature iρ. That means that there exists a positive (1. a Theorem 15. Proof. 1)–form ϕ on a complex manifold (M. Y ) := ϕ(X. Proof. we now use the formula (40) which gives the curvature of the Chern connection in terms of the square norm of an arbitrary local holomorphic section σ: ¯ R = −∂ ∂ log h(σ. as well as the Ricci form of a K¨hler manifold with positive Ricci tensor.1. σ). THE HIRZEBRUCH–RIEMANN–ROCH FORMULA 79 15. A cohomology class in H 1. Taking p = 0 and E = L in (63) shows that every holomorphic section ξ of E satisﬁes (67) ¯ ¯ 0 = 2∂ ∗ ∂ξ = ∗ i ξ + R (Jej . σ) − i∂ ∂u = iR − i∂ ∂u = ω. 1)–form ω and a Hermitian structure h on L whose Chern connection has curvature R such that [iR ] = [ω] (the factor 2π can obviously be ¯ ¯ skipped). negative) (1. 1)–form.
E) Hq (E). . and that K is isomorphic to the m + 1st tensor power of the tautological bundle L. k)–forms on M. m The holomorphic Euler characteristic Ξ(M. We have seen that the canonical bundle K is negative. Let E → M be a holomorphic vector bundle over some compact Hermitian manifold M 2m . ¯ ∂Ωq−1 (E) ¯ ∂ ¯ ∂ ¯ ∂ We deﬁne the cohomology groups H q (M.k M ⊗ E) the space of E–valued (0. We denote by Ωk (E) := C ∞ (Λ0.2. too. .q (M. ∂ ∗ ω = 0}.4.80 By hypothesis we have iR (X. The index of the elliptic complex (68) is deﬁned to be the index of this elliptic operator: ¯ ¯ ¯ ¯ ¯ ¯ Ind(∂ + ∂ ∗ ) := dim(Ker(∂ + ∂ ∗ )) − dim(Coker(∂ + ∂ ∗ )). E) := H q (M. we have shown with a direct computation that this last bundle has no holomorphic section. E) are isomorphic with the spaces of harmonic E–valued (0. Consider the following elliptic complex (68) Ω0 (E) → Ω1 (E) → . This result is consistent with our previous calculations on CPm . E) := k=0 (−1)k dim H k (M.0 M ⊗ E). ej ) = − Tr(A) 2 2 2 is a strictly positive function on M. 15. Consequently. E) is deﬁned by Ξ(M. E) := By analogy with the usual (untwisted) case. Λp. Y ). q)–forms: H q (M. ej ) = − A(ej . E) . with A negative deﬁnite. On the other hand. The cohomology groups H q (M. taking the Hermitian product with ξ in (67) and integrating over M shows that ξ has to vanish. ¯ Ker(∂ : Ωq (E) → Ωq+1 E) . The Hirzebruch–Riemann–Roch formula. Theorem 15. and deﬁne the space of harmonic E–valued (0. The analog of the Dolbeault decomposition theorem holds true in this case and as a corollary we have H p. Thus i 1 1 R (Jej . → Ωm (E). Y ) = A(JX. we denote ¯ ¯ For every Hermitian structure on E one can consider the formal adjoint ∂ ∗ of ∂. q)–forms on M by ¯ ¯ Hq (E) := {ω ∈ Ωq (E)  ∂ω = 0. We can view the elliptic complex (68) as an elliptic ﬁrst order diﬀerential operator simply by considering ¯ ¯ ∂ + ∂ ∗ : Ωeven (E) → Ωodd (E). which is thus negative.
If E is the trivial line bundle. Of course. k=0 For a proof of the Hirzebruch–Riemann–Roch formula see [3]. so there exists a global holomorphic (m. so hp.p (M). or π1 (M) = Z2 and M carries no global holomorphic (m.5 shows that there is no holomorphic (p. so h0. 0)–form even if M is not simply connected. a Proof. thus Ξ(M) = 1. a ˜ By Myers’ Theorem. then Hol(M) = SUm . Td(M) = π ∗ Td(M). The ﬁrst one a is a theorem due to Kobayashi: Theorem 15. By Theorem 11.15. then either M is simply connected. Since M is K¨hler we have hp. E) is simply denoted by Ξ(M) := m (−1)k h0.0 (M) = h0. where k denotes the number of sheets of the covering. by naturality. Let M be the universal covering of M. so M is simply connected. ˜ Proof. both concerning the fundamental group of K¨hler manifolds under suitable positivity assumptions of the Ricci tensor. THE HIRZEBRUCH–RIEMANN–ROCH FORMULA 81 and is nothing else but the index of the elliptic complex (68).5. The only thing that we will use in the sequel is that they satisfy the naturality axiom with respect to pull–backs. we have. 0)–form on M. Theorem 14. Our second application concerns Ricci–ﬂat K¨hler manifolds.0 (M) = 0 for p > 0.k (M). the holomorphic Euler characteristic Ξ(M. A compact K¨hler manifold with positive deﬁnite Ricci tensor is simply connected. every holomorphic . A compact K¨hler manifold M with Hol0 (M) = SUm is called Calabi–Yau manifold. If m is even. By Theorem 14. where Td(M) is the Todd class of the tangent bundle of M and ch(E) is the Chern character of E. ˜ ˜ if π : M → M denotes the covering projection. 0)–form. E) = M Td(M)ch(E).5. the fundamental group of M is ﬁnite. Theorem 15. 168) ˜ shows that M is compact (having irreducible holonomy). [1]. too.0 (M) = 1. The Cheeger–Gromoll theorem (cf. p. Now. This shows that k = 1. Let M 2m be a Calabi–Yau manifold. the holomorphic functions are just the constants. Theorem 15.5. and an easy exercise shows that for every top degree form ω on M one has ˜ M π∗ω = k M ω. a compact K¨hler a a 2m manifold M is Ricci–ﬂat if and only if the restricted holonomy group Hol0 (M) is a subgroup of a SUm .7. If m is odd. Let M be the universal cover of M. (Hirzebruch–Riemann–Roch formula) The holomorphic Euler characteristic of E can be computed as follows Ξ(M.6. The Todd class and the Chern character are characteristic classes of the corresponding vector bundles that we will not deﬁne explicitly. Applying the previous argument to M we get Ξ(M ) = 1. ˜ ˜ which is therefore compact. We will give two applications of the Riemann–Roch formula.
one for p = 0 and one for p = m. (2) Prove that the operator is elliptic. we necessarily have hm.3. Ξ(M) = kΞ(M). Thus ˜ Ξ(M ) = 0 2 for m odd for m even ˜ Moreover. so M carries no global holomorphic (m. If m is even. Hint: Start by showing that to any open cover {Ui } of M one can associate a closed cover {Cj } such that for every j there exists some i with Cj ⊂ Ui and such that the interiors of Cj and Ck are disjoint for every j = k. Prove that for every top degree form ω on M one has ˜ M H m−q (M. ˜ (4) Let π : M → M be a k–sheet covering projection between compact oriented manifolds. where k is the order of the fundamental group of M. E). In this last case. in the sense that its principal symbol applied to any non–zero real 1–form is an isomorphism. It is easy to check that SUm has only two invariant one–dimensional complex subspaces on (p. so M has a global holomorphic (m.0 M = 1. 0)–form. 0)–form. . (1) Prove the Kodaira–Serre duality: H q (M. 0)–forms. hence hm. and thus corresponds to a ﬁxed point of the holonomy representation.0 M = 0.82 form on M is parallel. Exercises. then either M is simply connected. (5) Show that the representation of SUm on Λp Cm has no invariant one–dimensional subspace for 1 ≤ p ≤ m − 1. This shows that Ξ(M) = 0 for m odd. E) for every holomorphic vector bundle E over a compact Hermitian manifold M. or k = 2 and Ξ(M) = 1. 15. E ∗ ⊗ KM ) ¯ ¯ ∂ + ∂ ∗ : Ωeven (E) → Ωodd (E) π∗ω = k M ω. (3) Prove that the index of the above deﬁned operator is equal to the holomorphic Euler characteristic Ξ(M.
Further vanishing results 16. ek )ω = −Jej ∧ (ek + iJek ) R(ej . el . Thus Now.k M ⊗ L. es )ej ∧ es ∧ ek el ω +R(ej . es . The ﬁrst term of the curvature operator R applied to some section ω ⊗ ξ ∈ Ω0.∗ M ⊗ K 2 is √ ¯ just the spin bundle of M and the operator 2(∂ + ∂) is just the Dirac operator. ek )ω. Let L be a holoo a morphic Hermitian line bundle over some K¨hler manifold M 2m with scalar curvature S.1. ∀ ω ∈ Ω0. el . el . ek . using (69) twice we get (ej − iJej ) ∧ (ek + iJek ) R(ej . 0)–vector and a (0. ek . es )ej ∧ ek ∧ es el ω. 2 In order to compute the second curvature term we make use of the following algebraic result R1 (ω ⊗ ξ) := (ej − iJej ) ∧ (ek + iJek ) R(ej . k)–forms. es )ej ∧ es ∧ ek el ω = R(ej .k M ⊗ E can be computed as follows i˜ i R(Jej . ek )ω = Jej ∧ (ek + iJek ) R(Jej . ek . ek . Proof. k)–form vanishes we obtain (69) X ω = iJX ω The forms R(ej . The Schr¨dinger–Lichnerowicz formula for K¨hler manifolds. el .k M. Jel . The Riemannian curvature operator satisﬁes for every (0. FURTHER VANISHING RESULTS 83 16. From the ﬁrst Bianchi identity we then obtain . ek )ω. k)–form ω. By changing ej to Jej and then ek to Jek we get ej ∧ (ek + iJek ) R(ej . ej )ω ⊗ ξ 2 2 1 = i(ρ(k) ω) ⊗ ξ − α(Jej . es )ej ∧ ek ∧ es el ω = −R(ej . ek )ω = 2ej ∧ (ek + iJek ) R(ej . R(ej . ek .16. The reader familiar with spin geometry will notice that in this case Λ0. ek )ω = 4iρ(k) ω Lemma 16. R(ej . ek . Jes )ej ∧ ek ∧ es el ω = −R(ej . ek )ej ∧ es ∧ ek el ω = −R(ej . ej )(ω ⊗ ξ) = 2(ρ(k) ω) ⊗ ξ + iα(Jej . We a would like to compute the curvature term in the Weitzenb¨ck formula on sections of Λ0. es )ej ∧ es ∧ el ek ω so this expression vanishes. ek )ω are still (0. ej )ω ⊗ ξ. el . Since the interior product of a (1. el .1. o and to show that this term becomes very simple in the case where L is a square root of the canon1 ical bundle. ek )ω = 4ej ∧ ek R(ej . Let us denote by iα the curvature of the Chern connection of L. Jek )ω = −iJej ∧ (ek + iJek ) R(ej . since the connection preserves the type decomposition of forms.
k)–form ω we have as before (ej − iJej ) ∧ (ek + iJek ) α(ej . 2 The formulas above show that the curvature term in the Weitzenb¨ck formula on Ω0.2. ⊕ Λ0. ek )ω = 4ej ∧ ek α(ej . ek )ω ⊗ ξ) 2 = −2iρ(k) (ω) ⊗ ξ + 2iα(k) (ω) ⊗ ξ. . el )ej ∧ el ω = 4iRic(ej . ek )(ω ⊗ ξ)) 2 1 = − (ej − iJej ) ∧ (ek + iJek ) ((R(ej . es )ej ∧ es ∧ ek el ω = 0. el . ej )ω ⊗ ξ = ω ⊗ ξ. Suppose that the curvature of the line bundle L satisﬁes 1 RL := iα = iρ. Then. ek .m M) ⊗ K 2 1 1 . ej )ω ⊗ ξ 2 −2iρ(k) (ω) ⊗ ξ + 2iα(k) (ω) ⊗ ξ 1 S = − ρ(Jej . ek )ω) ⊗ ξ + iα(ej . in the sense that L has a Hermitian structure h such that a KM is isomorphic to L ⊗ L with the induced tensor product Hermitian structure. (Schr¨dinger–Lichnerowicz formula). ek )ω = 4ej ∧ ek R(ej . (ej − iJej ) ∧ (ek + iJek ) R(ej . 1)–form α and (0. For every (1. if Ψ is a section of the complex vector bundle ΣM := (Λ0. es )ej ∧ ek (es ∧ el ω) = −4Ric(ej . ek . Let L = K 2 be a square root of the canono ical bundle of a K¨hler manifold M. ek )ω = 2ej ∧ (ek + iJek ) α(ej .84 whence Finally we get R(ej . ek )ω = −4α(k) (ω). 4 4 This proves the Theorem 16. Jel )ej ∧ el ω = 4iρ(k) ω.0 M ⊕ . . The second term in (62) thus reads 1 ˜ R2 (ω ⊗ ξ) := − (ej − iJej ) ∧ (ek + iJek ) (R(ej .k M ⊗ L o satisﬁes 1 R(ω ⊗ ξ) = (R1 + R2 )(ω ⊗ ξ) = iρ(k) (ω) ⊗ ξ − α(Jej . el . ek )ω = 4R(ej .
o not necessarily K¨hler). Note that. We apply the Weitzenb¨ck formula to some section ω ⊗ ξ of Λp. ej )ξ + 2iJej ∧ ek R(ej . ek )ω can be simpliﬁed as follows: Jej ∧ ek R(ej . Let M 2m be a compact K¨hler manifold and let a L be a positive line bundle over M. ej )(ω ⊗ ξ) + 2iJej ∧ ek R(ej . ek )el ∧ ek el ω = −ρ(ω) − R(ej .q M ⊗ L) the extension of 1. and it has important applications in geometry and topology (see [4]. ek )(ω ⊗ ξ) ˜ ˜ = iR(Jej . ek )(ω ⊗ ξ) ˜ −iej ∧ Jek R(ej .p M ⊗ L∗ and then take the complex ˜ conjugate. ej )(ω ⊗ ξ) − (ej − iJej ) ∧ (ek + iJek ) R(ej . a 16. From the deﬁnition. We now compute this curvature term. ˜ (ω ⊗ ξ) + R(ω ⊗ ξ). el )Jej ∧ el ω − Jej ∧ R(ej .q M ⊗ L) → C ∞ (Λp+1. ej )(ω ⊗ ξ) + iJej ∧ ek R(ej . es )Jej ∧ es ∧ ek el ω. FURTHER VANISHING RESULTS 85 and D := √ ¯ ¯ 2(∂ + ∂ ∗ ) is the Dirac operator on ΣM.2. whilst ¯ ∂ is an intrinsic operator. ∂ depends of course on the Hermitian structure on L.16. The same computation actually yields the dual formula (71) 2(∂ ∗ ∂ + ∂∂ ∗ )(ω ⊗ ξ) = ∗ Alternatively. ek )(ω ⊗ ξ) +2iJej ∧ ek ω ⊗ R (ej . We consider the K¨hler a metric on M whose fundamental form is just iR . el . Subtracting these two equations yields (72) ¯ ¯ ¯¯ ˜ 2(∂ ∗ ∂ + ∂ ∂ ∗ )(ω ⊗ ξ) = 2(∂ ∗ ∂ + ∂∂ ∗ )(ω ⊗ ξ) + (R − R)(ω ⊗ ξ). we denote by ∂ : C ∞ (Λp. ek )ω ⊗ ξ . 4 The Schr¨dinger–Lichnerowicz formula is valid in a more general setting (on all spin manifolds. ek )ω = Jej ∧ ek R(ej . By a slight abuse of language.0 to forms. the following formula holds D2Ψ = ∗ Ψ+ S Ψ. one can apply (70) to a section ω ⊗ ξ ∗ of Λq. [9]). ek )ω ⊗ ξ + 2(p + q)ω ⊗ ξ. ek )(ω ⊗ ξ) 2 1 ˜ + (ej + iJej ) ∧ (ek − iJek ) R(ej . the expression Jej ∧ ek R(ej . The Kodaira vanishing theorem. ek )(ω ⊗ ξ) 2 ˜ ˜ = iR(Jej . 1 ˜ ˜ ˜ (R − R)(ω ⊗ ξ) = iR(Jej . we know that L carries a Hermitian structure whose Chern connection has curvature R with iR > 0. ek . ek )ξ = 2iρ(ω) ⊗ ξ − 2mω ⊗ ξ + 2iJej ∧ ek R(ej . ek )el ∧ el ω = −Ric(ej . On the other hand.q M ⊗ L: o (70) ¯ ¯ ¯¯ 2(∂ ∗ ∂ + ∂ ∂ ∗ )(ω ⊗ ξ) = ∗ (ω ⊗ ξ) + R(ω ⊗ ξ). = 2iρ(ω) ⊗ ξ + iω ⊗ R (Jej .
(Kodaira vanishing theorem). a . ·. ek )ω = −ρ(ω). el . If σ is a harmonic L–valued form. ¯ ¯ ∂σ2 + ∂ ∗ σ2 dv = M ∂σ2 + ∂ ∗ σ2 + (p + q − m)σ2 dv.q (M. es )Jej ∧ es ∧ ek el ω +R(ej .86 and from the Bianchi identity 2R(ej . Jej . ek . After taking the Hermitian product with ω⊗ξ (which we denote by σ for simplicity) and integrating over M we get (73) M This shows that Jej ∧ ek R(ej . where the last expression vanishes because R(·. one has H p. J·. ek . thus proving the Theorem 16. ej . so from the previous calculation we get ¯ ¯ ¯¯ 2(∂ ∗ ∂ + ∂ ∂ ∗ )(ω ⊗ ξ) = 2(∂ ∗ ∂ + ∂∂ ∗ )(ω ⊗ ξ) + 2(p + q − m)(ω ⊗ ξ). es )Jej ∧ es ∧ ek el ω = −R(el . the left hand side term in (73) vanishes. es )ej ∧ es ∧ ek el ω = 0. es )Jej ∧ es ∧ el ek ω = R(el . es )Jej ∧ es ∧ ek el ω = R(ej .3. L) = 0 whenever p + q > m. If L is a positive holomorphic line bundle on a compact K¨hler manifold M. ek . ek . ·) is symmetric in the last two arguments. el . ek . el .
Part 6 Calabi–Yau manifolds .
. g) be a locally irreducible Riemannian manifold. It is clear that a metric is hyperk¨hler if and only if it is K¨hler with respect to two anti–commuting a a complex structures. and if J1 is diﬀerent from J and −J. Clearly we have A B ¯t Spk = M = ¯ ¯ ∈ M2k (C) M M = I2k . so the parallel skew–symmetric endomorphism 1 I := √ (J1 + JJ1 J) 4 − α2 deﬁnes a complex structure anti–commuting with J. z2 ) = (−¯2 . A Riemannian manifold (M n . so J1 = ±J. In order to obtain the classiﬁcation (up to ﬁnite coverings) a of compact Ricci–ﬂat K¨hler manifolds. g) is called hyperk¨hler if there exist three coma plex structures I. Therefore we have α2 < 4. which was excluded in the hypothesis. with respect to which g is K¨hler. −B A It is tautological that a 4k–dimensional manifold is hyperk¨hler if and only if the bundle of a orthonormal frames has a reduction to Spk . α2 = αId2 = JJ1 + J1 J2 ≤ 2(JJ1 2 + J1 J2 ) ≤ 4J2 J1 2 = 4. a Proof.17. g) is hyperk¨hler. then (M. J. z2 ) → z1 + jz2 . We denote by I. iz2 ) J(z1 . j and k respectively. which correspond to the following endomorphisms of C2k : I(z1 . The equality case can only hold if JJ1 = βJ1 J for some real number β.2. We then compute using (74) (J1 + JJ1 J)2 = (α2 − 4)Id. Hyperk¨hler manifolds. so by local irreducibility it has to be constant: (74) JJ1 + J1 J = αId. which also commute with J (and thus also with K).1. z2 ) = (−i¯2 . z z Let us denote by Spk the group of unitary transformations of C2k (that is. Let (M n . a Consider the identiﬁcation of C2k with Hk given by (z1 . The endomorphism JJ1 +J1 J is symmetric and parallel on M. preserving the canonical Hermitian product and commuting with I). If g is K¨hler with a respect to two complex structures J and J1 . Together with (74) this shows that JJ1 = γId for some real number γ. From the Cauchy–Schwartz inequality we get α ∈ R. where the norm considered here is the operator norm. K on M satisfying K = IJ such that g is a K¨hler metric with respect to a each of these complex structures.1. z2 ) = (iz1 . i¯1 ). z1 ) z ¯ K(z1 . J and K the right product on Hk with i. In the irreducible case. we make the following a Definition 17. Ricci–ﬂat K¨hler metrics a 17. this can be weakened as follows: Proposition 17.
× Ms be the De Rham decomposition of M0 . Let now M be an arbitrary compact Ricci–ﬂat K¨hler manifold. . × Mr × Tl )/Γ. .¨ 17. A hyperk¨hler manifold is called strict a a if it is locally irreducible.3. using (22)) that the hyperplane bundle H on CPN has a connection with curvature −iϕ. RICCI–FLAT KAHLER METRICS 89 Lemma 17.168) says that M is isomorphic to a quotient M where M0 is a compact simply connected K¨hler manifold. 17. This shows that every hyperk¨hler manifold is Ricci–ﬂat. Let M0 = M1 × .5. Spk ⊂ SU2k . A well–known result of Chow states that a projective manifold is algebraic.4. Proof. By deﬁnition we have Spk ⊂ U2k . (M0 × Tl )/Γ. A compact complex manifold M with a positive holomorphic line bundle L is projective. Proposition 17.2. Conversely. where Mj are simply connected compact Calabi–Yau manifolds for j ≤ s. Proof. The Berger holonomy theorem then shows that Mj is either Calabi–Yau or strict hyperk¨hler for every j. . J) is called projective if it can be holomorphically embedded in some complex projective space CPN . Every projective manifold has a positive holomorphic line bundle. We thus have the following a Theorem 17. we have the celebrated Theorem 17. p. It is easy to check (e. The Cheeger–Gromoll theorem a ([1]. . so every matrix in Spk is diagonalizable as complex matrix and its eigenvalues are complex numbers of unit norm. . Let ϕ be the fundamental form of the Fubini–Study metric on CPN . deﬁned by a ﬁnite set of homogeneous polynomials in the complex projective space. Tl is a complex torus and Γ is a ﬁnite a group of holomorphic transformations.6. (Kodaira embedding theorem). simply connected compact strict hyperk¨hler manifolds for s + 1 ≤ j ≤ r and Γ is a ﬁnite group of holomorphic a transformations. Projective manifolds. × Ms × Ms+1 .g. A compact Ricci–ﬂat K¨hler manifold M is isomorphic to the quotient a M (M1 × . If v is an eigenvector of some M ∈ Spk with eigenvalue λ ∈ S 1 then ¯ MJv = JMv = Jλv = λJv = λ−1 Jv. . Then Mj are compact Ricci–ﬂat simply connected K¨hler manifolds with irreducible a holonomy for all j. showing that the determinant of M equals 1. that is. A compact complex manifold (M 2m . The restriction of this line bundle to any complex submanifold of CPN is thus positive. A symmetric space which is Ricci–ﬂat is automatically ﬂat. . so the Mj ’s are not symmetric.
so H 1 (M.5 we get h2. A∗ ) is just the set of equivalence classes of complex line bundles over M. Let now M 2m be a Calabi–Yau manifold. let A and A∗ be the sheaves of smooth functions on M with values in C and C∗ respectively. Notice that H 1 (M. if ω is any complex 2–form representing γ in i real cohomology. Z). By the Dolbeault decomposition theorem we obtain that any harmonic 2–form on M is of type (1.90 A proof can be found in [2]. Q) is dense in H 2 (M. Then since [ω] = c1 (L) we get [2πω] = [iR ].0 Cm . A) = 0. thus a positive holomorphic line bundle on M. m > 2. 1). A∗ ) → H 2 (M. there exists a complex line bundle L with c1 (L) = γ. and the space of positive harmonic (1. A) → H 1(M. Proof. 1)–form ω such that [ω] ∈ H 2 (M. . . A) = 0 and H 2 (M. p. M → CPN x → [σ0 (x) : . 1)–forms is open in H1. If the form ω is real and of type (1. : σN (x)] Corollary 17. A∗ ) H 2 (M. σN } such that the holomorphic mapping is an embedding. thus proving that H 1 (M. Z) → H 2 (M. M is then projective. R) = H2 (M. so there ˜ exists some 1–form θ such that 2πω = i(R + dθ). 1). By multiplying with the common denominator. there exists a connection on L such that 2π R = ω. 1 → H 1 (M. . A) → . 1)–part of the connection whose curvature is ω. it admits a partition of unity).1 (M. 0)–forms on M. Z). and the isomorphism above is given by the ﬁrst Chern class. 176. take any ˜ ˜ connection ˜ on L with curvature R . then the complex bundle L has a holomorphic structure. R). R). so by Theorem 14. Clearly the curvature of := ˜ + iθ satisﬁes the desired equation. To see this. Moreover. Since SUm has no ﬁxed point on Λ2.0 (M) = 0. . The main idea is to show that a suitable positive power Lk of L has a basis of holomorphic sections {σ0 . we can ﬁnd a positive harmonic (1. we deduce that there are no parallel (2. . By the Kodaira embedding theorem. Q). Since H 2 (M. Every Calabi–Yau manifold of complex dimension m ≥ 3 is projective. .7. For every compact manifold M. given by the (0. The exact sequence of sheaves ˘ induces an exact sequence in Cech cohomology 0 → Z → A → A∗ → 0 c exp The sheaf A is ﬁne (that is. Consider the fundamental form ϕ of M. . Then the argument above shows that there exists a holomorphic line bundle whose ﬁrst Chern class is ω. The above argument shows that for every integer cohomology class γ ∈ H 2 (M. Z). we may suppose that [ω] ∈ H 2 (M.
A meromorphic function can be always expressed locally as f . fα . Two such collections (Uα . the pole–locus) of h taken with multiplicities. and fα . An analytic hypersurface of M is a subset V ∈ M such that for every x ∈ V there exists an open set Ux ⊂ M containing x and a holomorphic function fx deﬁned on Ux such that V ∩Ux is the zero–set of fx . A divisor D is called eﬀective if all ai ≥ 0 for all i. fx More precisely. which does not depend on x. β. It can be shown that the order of f is a x well–deﬁned positive integer. If V is an irreducible analytic hypersurface. then for every holomorphic function f around x. Then (h) = V o(h. gα ) and (Uβ . gβ ) are equivalent if fα gβ = fβ gα on Uα ∩Uβ for all α. ai ∈ Z. one can write the function h as h = gx near x. with deﬁning function ϕx around some x ∈ V . V ). gα ) where σα is a local holomorphic section of L over Uα and gα is a holomorphic function on Uα . β. CONSTRUCTIONS OF CALABI–YAU MANIFOLDS 91 18. and this is a well–deﬁned integer independent on x. fβ . for every x in M. . and is denoted by o(f. The quotient of any two local deﬁning functions around x is a non–vanishing holomorphic function around x. Let M be a complex manifold. V ). denoted by o(h. A divisor D in a compact complex manifold M is a ﬁnite formal sum with integer coeﬃcients of irreducible analytic hypersurfaces of M. D := i ai Vi . V ).1. V ) − o(gx . such that σα gβ = σβ gα on Uα ∩ Uβ for all α. gα) where {Uα } is an open covering of M. we deﬁne the order of h along V at x to be o(fx . V )V. If V is an irreducible analytic hypersurface containing x. An analytic hypersurface V is called irreducible if it can not be written as the union of two smaller analytic hypersurfaces. where (h)0 and (h)∞ denote the zero–locus (resp. Divisors. g We deﬁne similarly a meromorphic section of a holomorphic line bundle L as an equivalence class of collections (Uα . fα . Constructions of Calabi–Yau manifolds 18.18. β. A meromorphic function h deﬁnes a divisor (h) in a canonical way by (h) := (h)0 − (h)∞ . where f and g are locally deﬁned holomorphic functions. The set of divisors is clearly a commutative group under formal sums. the order of f along V at x is deﬁned to be the largest f positive integer a such that ϕa is holomorphic around x. A meromorphic function on a complex manifold M is an equivalence class of collections (Uα . σα . gα are holomorphic functions deﬁned on Uα such that fα gβ = fβ gα on Uα ∩ Uβ for all α.1. Definition 18. Such an fx is called a local deﬁning function for V near x. Every analytic hypersurface is a ﬁnite union of its irreducible components.
Then the arguments above show that there exists a group homomorphism Notice that the line bundle associated to a divisor (h) is trivial for every meromorphic function h. there are only ﬁnitely many irreducible analytic hypersurfaces along which fx of gx have non–vanishing order. then clearly [−D] = [D]−1 and [D + D ] = [D] ⊗ [D ]. and that the equivalence class of [D] does not depend on the local deﬁning functions hα . Similarly. In this case we write D ≡ D . one can deﬁne the order o(σ.2. ai Vi and fi are local deﬁning functions for Vi near some x ∈ M (of course we can take If D = fi = 1 if Vi does not contain x). Example. and Pic(M) := H 1 (M. if σ is a global meromorphic section of a line bundle L. Line bundles and divisors. V )V. Let H denote the hyperplane {z0 = 0} in CPm and consider the usual open covering z0 Uα = {zα = 0} of CPm . since the transition maps do not vanish. To any divisor D we will associate a holomorphic line bundle [D] on M in the following way. V ) of σ along any irreducible analytic hypersurface V using local trivializations of L. It is easy to check that gαβ are non–vanishing holomorphic hβ functions on Uα ∩ Uβ satisfying the cocycle conditions. This follows directly from the deﬁnition: for any open cover Uα on M. . where h is the meromorphic function deﬁned by σ = σ h. O) the Picard group of equivalence classes of holomorphic line bundles (where O denotes the sheaf of holomorphic functions). then the meromorphic function fiai is called a local deﬁning function for D around x. which are exactly the transition function of the hyperplane line bundle introduced in Section 3. Definition 18. We deﬁne [D] to be the holomorphic line bundle on M with transition functions gαβ := hα . Two divisors D and D are called linearly equivalent if there exists some meromorphic function h such that D = D + (h). one deﬁnes a divisor (σ) on M by (σ) = V o(σ. so they do not contribute to the order. If D and D are divisors. hUα is a local deﬁning [ ] : Div(M) → Pic(M) D → [D]. This clearly does not depend on the chosen trivialization.2. Clearly two meromorphic sections σ and σ of L deﬁne linearly equivalent divisors (σ) = (σ )+(h). which justiﬁes its denomination. The line bundle [H] has thus transition functions gαβ = zα . Then 1 is a local deﬁning function for H on U0 and zα are local deﬁning zβ functions on Uα . Take an open covering Uα of M and local deﬁning meromorphic functions hα for D deﬁned on Uα .92 fx where the above sum is ﬁnite since for every open set Ux where h = gx . We call Div(M) the group of divisors on M. 18. As before.
σα is a deﬁning σβ meromorphic section for (σ) on Uα . so the transition functions for the line bundle [(h)] are equal to 1 on any intersection Uα ∩ Uβ . we get dfα V = (fβ dgαβ + gαβ dfβ )V = gαβ V dfβ V . Thus [ ] descends to a group homomorphism [ ] : Div(M)/≡ −→Pic(M). CONSTRUCTIONS OF CALABI–YAU MANIFOLDS 93 function for the divisor (h) on Uα . fb hb where hα is a local deﬁning meromorphic function for D on Uα .0 1. as fα does not vanish on f Uα .3. Adjunction formulas.161). the normal bundle NV ∗ is the co–kernel of the inclusion i∗ : T 1. the homomorphism [ ] descends to an isomorphism Div(M)/≡ −→ Pic(M).3. We have proved the Theorem 18. Let i : V → M be the inclusion of V into M. Suppose now that [D] = 0 for some divisor D on M.0 ∗ deﬁned as the kernel of the projection i : Λ MV → Λ V . is ∗ 1. the divisor associated to H is just D. Moreover.18. ψα ). We claim that [(σ)] = L. ∼ = . That means that the line bundle [D] is trivial. since fα vanishes along V fβ ∗ which is smooth. the meromorphic section σ deﬁnes meromorphic functions σα on Uα such that gαβ = σα . We will show that the normal and co–normal bundles of V in M can be computed in terms of the divisor V . Let V ⊂ M be a smooth complex hypersurface of a compact complex manifold M. This proves the injectivity of [ ] on isomorphism classes of divisors. thus L is just the line bundle associated to (σ). 0)–forms on M vanishing on V . we have seen that a global meromorphic section σ of L deﬁnes a divisor (σ) on M. Proposition 18. If gαβ denote the transition functions of L with respect to some trivialization (Uα .4.0 V → T 1. since fα = gαβ fβ . Now. so there exists an open cover {Uα } of M and holomorphic non–vanishing functions fα : Uα → C∗ such that hα fα = gαβ = on Uα ∩ Uβ . (First adjunction formula) The restriction to V of the line bundle [V ] associated to the divisor V is isomorphic to the holomorphic normal bundle of V in M: NV = [V ]V . Thus NV is spanned by holomorphic (1. 18. the co–normal bundle NV .0 MV and its dual. By deﬁnition. Let fα be local deﬁning functions for V on some open covering Uα . If the manifold M is projective. Proof. From the deﬁnition. This shows the existence of a α global meromorphic function H on M such that HUα = hα . If L ∈ Pic(M) is a holomorphic line bundle. Moreover. By deﬁnition. we see that dfα V is a non–vanishing local section of NV . the quotients gαβ := fα are the transition functions of [V ] on Uα ∩ Uβ . Every holomorphic line bundle of a projective manifold has a global meromorphic section (see [2] p.
94
∗ Thus the collection (Uα , dfα V ) deﬁnes a global holomorphic section of NV ⊗[V ]V , showing that this ∗ tensor product bundle is trivial. This proves that NV = [−V ]V and consequently NV = [V ]V .
Consider now the exact ﬁbre bundle sequence
∗ 0 → NV → Λ1,0 MV → Λ1,0 V → 0.
Taking the maximal exterior power in this exact sequence yields KM  V so KV This is the second adjunction formula. (KM ⊗ [V ])V .
∗ KV ⊗ NV = KV − [V ]V ,
We will use the following theorem whose proof, based on the Kodaira vanishing theorem, can be found in [2], p. 156. Theorem 18.5. (Lefschetz Hyperplane Theorem). Let V be a smooth analytic hypersurface in a compact complex manifold M 2m such that [V ] is positive. Then the linear maps H i (M, C) → H i(V, C) induced by the inclusion V → M are isomorphisms for i ≤ m − 2 and injective for i = m − 1. If m ≥ 3 then π1 (M) = π1 (V ). Our main application will be the following result on complete intersections in the complex projective space. Theorem 18.6. Let P1 , · · · Pk be homogeneous irreducible relatively prime polynomials in m + 1 variables of degrees d1 , . . . dk . Let N denote the subset in CPm deﬁned by these polynomials: N := {[z0 : . . . : zm ] ∈ CPm  Pi (z0 , . . . , zm ) = 0, ∀1 ≤ i ≤ k}. Then, if N is smooth, we have KN hyperplane divisor in CPm . [qH]N , where q = (d1 + . . . + dk ) − (m + 1) and H is the
Proof. Notice ﬁrst that N is smooth for a generic choice of the polynomials Pi . We denote by Vi the analytic hypersurface in CPm deﬁned by Pi and claim that (75) Vi ∼ di H. = This can be seen as follows. While the homogeneous polynomial Pi is not a well–deﬁned function on CPm , the quotient hi := Pii is a meromorphic function. More precisely, hi is deﬁned by d the collection (Uα , Pii , di ). Clearly the zero–locus of hi is (hi )0 = Vi and the pole–locus is d zα zα (hi )∞ = diH0 , where H0 is just the hyperplane {z0 = 0}. This shows that (hi ) = Vi − di H0 , thus proving our claim. Let now, for i = 1, . . . , k, Ni denote the intersection of V1 , . . . , Vi. Since Ni+1 = Ni ∩ Vi+1 , we have (76) [Ni+1 ]Ni = [di+1 H]Ni .
z0 i
d
z0
18. CONSTRUCTIONS OF CALABI–YAU MANIFOLDS
95
This follows from the fact that if V is an irreducible hypersurface in a projective manifold M and N is any analytic submanifold in M then [V ]N We claim that (77) KNi [ni H]Ni , where ni := (d1 + . . . + di ) − (m + 1). For i = 1 this follows directly from the second adjunction formula together with (75), using the fact that KCPm = [−(m + 1)H]. Suppose that the formula holds for some i ≥ 1. The second adjunction formula applied to the hypersurface Ni+1 of Ni , together with (76) yields Thus (77) is true for every i, and in particular for i = k. This ﬁnishes the proof. KNi+1 = ([Ni+1 ] ⊗ KNi )Ni+1 = ([di+1 H] ⊗ [ni H])Ni+1 = [ni+1 H]Ni+1 . [V ∩ N].
Corollary 18.7. Let d1 , . . . , dk be positive integers and denote their sum by m+1 := d1 +. . .+dk . Suppose that m ≥ k + 3. If P1 , · · · Pk are generic homogeneous irreducible polynomials in m + 1 variables of degrees d1 , . . . dk , then the manifold carries a unique (up to rescaling) Ricci–ﬂat K¨hler metric compatible with the complex structure a m induced from CP . Endowed with this metric, N is Calabi–Yau. Proof. Theorem 18.6 shows that the ﬁrst Chern class of N vanishes. The condition m ≥ k+3 together with Lefschetz Hyperplane Theorem applied inductively to the analytic hypersurfaces Ni ⊂ Ni+1 show that N is simply connected and b2 (N) = b2 (CPm ) = 1, and moreover the restriction of the K¨hler form of CPm to N is a generator of the second cohomology group of N. a The Calabi conjecture shows that there exists a unique Ricci–ﬂat metric on N up to rescaling. If this metric were reducible, we would have at least two independent elements in the second cohomology of N, deﬁned by the K¨hler forms of the two factors. Since b1 (N) = 1 this is a impossible. Thus N is either Calabi–Yau or hyperk¨hler. The latter case is however impossible, a since every compact hyperk¨hler manifold has a parallel (2, 0)–form, thus its second Betti number a cannot be equal to 1. N := {[z0 : . . . : zm ] ∈ CPm  Pi (z0 , . . . , zm ) = 0, ∀1 ≤ i ≤ k}
Interscience Publishers. 97 . [6] D.. 111 (1980). B. Lichnerowicz. Springer. Spineurs harmoniques. 1981. Wiley. Grenzgeb. Uber eine bemerkenswerte Hermitesche Metrik. [4] M. 2000. [9] A. Amsterdam. Principles of Algebraic Geometry. [3] F. H¨rmander. 423–434. C. Einstein Manifolds. Besse. Sci. of Math. An introduction to complex analysis in several variables. 7–9. Berlin. Ergeb. K. 1966. 173–186. [2] P. Jr. Oxford University Press. Gromov. Nomizu. Acad.Bibliography [1] A. Griﬃth. Topological methods in algebraic geometry. 1969. Joyce. K¨hler. Lawson. Abh. Hirzebruch. New York. Hamburg Univ. a [8] S. Harris. Kobayashi. Foundations of Diﬀerential Geometry I. Sem.. J. II. [5] L. ¨ [7] E. Ann.. Math. 9 (1933). third edition. Math. 1978. 1990. Springer. NorthHolland Pubo lishing Co. 10. Paris 257 (1963). New York. Compact Manifolds with Special Holonomy. 1963. R. The classiﬁcation of simply connected manifolds of positive scalar curvature.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.