ESI Lectures in Mathematics and Physics

Werner Ballmann
Lectures on K¨ahler Manifolds
To my wife Helga
Preface
These are notes of lectures on K¨ ahler manifolds which I taught at the Univer-
sity of Bonn and, in reduced form, at the Erwin-Schr¨ odinger Institute in Vi-
enna. Besides giving a thorough introduction into K¨ ahler geometry, my main
aims were cohomology of K¨ ahler manifolds, formality of K¨ ahler manifolds af-
ter [DGMS], Calabi conjecture and some of its consequences, Gromov’s K¨ahler
hyperbolicity [Gr], and the Kodaira embedding theorem.
Let M be a complex manifold. A Riemannian metric on M is called Her-
mitian if it is compatible with the complex structure J of M,
'JX, JY ` = 'X, Y `.
Then the associated differential two-form ω defined by
ω(X, Y ) = 'JX, Y `
is called the K¨ahler form. It turns out that ω is closed if and only if J is
parallel. Then M is called a K¨ ahler manifold and the metric on M a K¨ ahler
metric. K¨ahler manifolds are modelled on complex Euclidean space. Except
for the latter, the main example is complex projective space endowed with the
Fubini–Study metric.
Let N be a complex submanifold of a K¨ ahler manifold M. Since the re-
striction of the Riemannian metric of M to N is Hermitian and its K¨ ahler
form is the restriction of the K¨ ahler form of M to N, N together with the
induced Riemannian metric is a K¨ ahler manifold as well. In particular, smooth
complex projective varieties together with the Riemannian metric induced by
the Fubini–Study metric are K¨ ahlerian. This explains the close connection of
K¨ ahler geometry with complex algebraic geometry.
I concentrate on the differential geometric side of K¨ ahler geometry, except
for a few remarks I do not say much about complex analysis and complex
algebraic geometry. The contents of the notes is quite clear from the table
below. Nevertheless, a few words seem to be in order. These concern mainly
the prerequisites. I assume that the reader is familiar with basic concepts
from differential geometry like vector bundles and connections, Riemannian
and Hermitian metrics, curvature and holonomy. In analysis I assume the
basic facts from the theory of elliptic partial differential operators, in particular
regularity and Hodge theory. Good references for this are for example [LM,
Section III.5] and [Wa, Chapter 6]. In Chapter 8, I discuss Gromov’s K¨ahler
hyperbolic spaces. Following the arguments in [Gr], the proof of the main result
of this chapter is based on a somewhat generalized version of Atiyah’s L
2
-index
theorem; for the version needed here, the best reference seems to be Chapter
13 in [Ro]. In Chapter 7, I discuss the proof of the Calabi conjecture. Without
further reference I use H¨older spaces and Sobolev embedding theorems. This is
standard material, and many textbooks on analysis provide these prerequisites.
iv Preface
In addition, I need a result from the regularity theory of non-linear partial
differential equations. For this, I refer to the lecture notes by Kazdan [Ka2]
where the reader finds the necessary statements together with precise references
for their proofs. I use some basic sheaf theory in the proof of the Kodaira
embedding theorem in Chapter 9. What I need is again standard and can be
found, for example, in [Hir, Section 1.2] or [Wa, Chapter 5]. For the convenience
of the reader, I include appendices on characteristic classes, symmetric spaces,
and differential operators.
The reader may miss historical comments. Although I spent quite some
time on preparing my lectures and writing these notes, my ideas about the
development of the field are still too vague for an adequate historical discussion.
Acknowledgments. I would like to acknowledge the hospitality of the Max-
Planck-Institut f¨ ur Mathematik in Bonn, the Erwin-Schr¨ odinger-Institute in
Vienna, and the Forschungsinstitut f¨ ur Mathematik at the ETH in Z¨ urich,
who provided me with the opportunity to work undisturbed and undistracted
on these lecture notes and other mathematical projects. I am grateful to
Dmitri Anosov, Marc Burger, Thomas Delzant, Beno Eckmann, Stefan Hilde-
brandt, Friedrich Hirzebruch, Ursula Hamenst¨adt, Daniel Huybrechts, Her-
mann Karcher, Jerry Kazdan, Ingo Lieb, Matthias Lesch, Werner M¨ uller,
Joachim Schwermer, Gregor Weingart, and two anonymous referees for very
helpful discussions and remarks about various topics of these notes. I would
like to thank Anna Pratoussevitch, Daniel Roggenkamp, and Anna Wienhard
for their careful proofreading of the manuscript. My special thanks go to Hans-
Joachim Hein, who read many versions of the manuscript very carefully and
suggested many substantial improvements. Subsections 4.6, 6.3, and 7.4 are
taken from his Diplom thesis [Hei]. Finally I would like to thank Irene Zim-
mermann and Manfred Karbe from the EMS Publishing House for their cordial
and effective cooperation.
Contents
1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Differential Forms . . . . . . . . . . . . . . . . . . . . . 2
1.2 Exterior Derivative . . . . . . . . . . . . . . . . . . . . . 4
1.3 Laplace Operator . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Hodge Operator . . . . . . . . . . . . . . . . . . . . . . 9
2 Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Complex Vector Fields . . . . . . . . . . . . . . . . . . . 18
2.2 Differential Forms . . . . . . . . . . . . . . . . . . . . . 21
2.3 Compatible Metrics . . . . . . . . . . . . . . . . . . . . 23
2.4 Blowing Up Points . . . . . . . . . . . . . . . . . . . . . 24
3 Holomorphic Vector Bundles . . . . . . . . . . . . . . . . . . . 29
3.1 Dolbeault Cohomology . . . . . . . . . . . . . . . . . . . 29
3.2 Chern Connection . . . . . . . . . . . . . . . . . . . . . 31
3.3 Some Formulas . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Holomorphic Line Bundles . . . . . . . . . . . . . . . . . 37
4 K¨ ahler Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1 K¨ahler Form and Volume . . . . . . . . . . . . . . . . . 49
4.2 Levi-Civita Connection . . . . . . . . . . . . . . . . . . 52
4.3 Curvature Tensor . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Ricci Tensor . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 Holonomy . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.6 Killing Fields . . . . . . . . . . . . . . . . . . . . . . . . 59
5 Cohomology of K¨ ahler Manifolds . . . . . . . . . . . . . . . . . 62
5.1 Lefschetz Map and Differentials . . . . . . . . . . . . . . 64
5.2 Lefschetz Map and Cohomology . . . . . . . . . . . . . 67
5.3 The dd
c
-Lemma and Formality . . . . . . . . . . . . . . 73
5.4 Some Vanishing Theorems . . . . . . . . . . . . . . . . . 78
6 Ricci Curvature and Global Structure . . . . . . . . . . . . . . 82
6.1 Ricci-Flat K¨ ahler Manifolds . . . . . . . . . . . . . . . . 83
6.2 Nonnegative Ricci Curvature . . . . . . . . . . . . . . . 84
6.3 Ricci Curvature and Laplace Operator . . . . . . . . . . 86
7 Calabi Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.1 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . 93
7.3 Existence . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.4 Obstructions . . . . . . . . . . . . . . . . . . . . . . . . 99
vi Contents
8 K¨ ahler Hyperbolic Spaces . . . . . . . . . . . . . . . . . . . . . 105
8.1 K¨ahler Hyperbolicity and Spectrum . . . . . . . . . . . 108
8.2 Non-Vanishing of Cohomology . . . . . . . . . . . . . . 112
9 Kodaira Embedding Theorem . . . . . . . . . . . . . . . . . . . 117
9.1 Proof of the Embedding Theorem . . . . . . . . . . . . 119
9.2 Two Applications . . . . . . . . . . . . . . . . . . . . . . 123
Appendix A Chern–Weil Theory . . . . . . . . . . . . . . . . . . . . 124
A.1 Chern Classes and Character . . . . . . . . . . . . . . . 130
A.2 Euler Class . . . . . . . . . . . . . . . . . . . . . . . . . 135
Appendix B Symmetric Spaces . . . . . . . . . . . . . . . . . . . . . 137
B.1 Symmetric Pairs . . . . . . . . . . . . . . . . . . . . . . 141
B.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 148
B.3 Hermitian Symmetric Spaces . . . . . . . . . . . . . . . 156
Appendix C Remarks on Differential Operators . . . . . . . . . . . 162
C.1 Dirac Operators . . . . . . . . . . . . . . . . . . . . . . 165
C.2 L
2
-de Rham Cohomology . . . . . . . . . . . . . . . . . 167
C.3 L
2
-Dolbeault Cohomology . . . . . . . . . . . . . . . . . 168
Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
1 Preliminaries
In this chapter, we set notation and conventions and discuss some preliminaries.
Let M be a manifold
1
of dimension n. A coordinate chart for M is a tuple
(x, U), where U ⊂ M is open and x: U → R
n
is a diffeomorphism onto its
image. As a rule, we will not refer to the domain U of x. The coordinate frame
of a coordinate chart x consists of the ordered tuple of vector fields
X
j
=

∂x
j
, 1 ≤ j ≤ n. (1.1)
We say that a coordinate chart x is centered at a point p ∈ M if x(p) = 0.
Let E →M be a vector bundle over M. We denote the space of sections of E
by E(E) or E(M, E). More generally, for U ⊂ M open, we denote by E(U, E) the
space of sections of E over U. Furthermore, we denote by E
c
(U, E) ⊂ E(U, E)
the subspace of sections with compact support in U.
As long as there is no need to specify a name for them, Riemannian metrics
on E are denoted by angle brackets '· , · `. Similarly, if E is complex, Hermitian
metrics on E are denoted by parentheses (· , · ). As a rule we assume that
Hermitian metrics on a given bundle E are conjugate linear in the first variable
and complex linear in the second. The induced Hermitian metric on the dual
bundle E

will then be complex linear in the first and conjugate linear in the
second variable. The reason for using different symbols for Riemannian and
Hermitian metrics is apparent from (1.12) below.
If E is a complex vector bundle over M and (· , · ) is a Hermitian metric on
E, then
(σ, τ)
2
=

M
(σ, τ) (1.2)
is a Hermitian product on E
c
(M, E). We let L
2
(E) = L
2
(M, E) be the comple-
tion of E
c
(M, E) with respect to the Hermitian norm induced by the Hermitian
product in (1.2) and identify L
2
(M, E) with the space of equivalence classes of
square-integrable measurable sections of E as usual
2
.
Let g = ' · , ·` be a Riemannian metric on M and ∇ be its Levi-Civita con-
nection. By setting g(X)(Y ) := g(X, Y ), we may interpret g as an isomorphism
TM →T

M. We use the standard musical notation for this isomorphism,
v

(w) = 'v, w` and 'ϕ

, w` = ϕ(w), (1.3)
where v, w ∈ TM and ϕ ∈ T

M have the same foot points. It is obvious that
(v

)

= v and (ϕ

)

= ϕ.
For a vector field X, the Lie derivative L
X
g of g measures how much g
varies under the flow of X. It is given by
(L
X
g)(Y, Z) = '∇
Y
X, Z` +'Y, ∇
Z
X`. (1.4)
1
Unless specified otherwise, manifolds and maps are assumed to be smooth.
2
We use similar terminology in the case of real vector bundles and Riemannian metrics.
2 Lectures on K¨ ahler Manifolds
In fact, by the product rule for the Lie derivative,
X'Y, Z` = L
X
(g(Y, Z)) = (L
X
g)(Y, Z) + g(L
X
Y, Z) +g(Y, L
X
Z).
We say that X is a Killing field if the flow of X preserves g. By definition, X
is a Killing field iff L
X
g = 0 or, equivalently, iff ∇X is skew-symmetric.
1.5 Exercises. Let X be a Killing field on M.
1) For any geodesic c in M, the composition X ◦ c is a Jacobi field along
c. Hint: The flow of X consists of local isometries of M and gives rise to local
geodesic variations of c with variation field X ◦ c.
2) For all vector fields Y, Z on M

2
X(Y, Z) +R(X, Y )Z = 0.
Hint: For Y = Z, this equation reduces to the Jacobi equation.
The divergence div X of a vector field X on M measures the change of the
volume element under the flow of X. It is given by
div X = tr ∇X. (1.6)
In terms of a local orthonormal frame (X
1
, . . . , X
n
) of TM, we have
div X =
¸
'∇
Xj
X, X
j
`. (1.7)
By (1.4), we also have
tr L
X
g = 2 div X. (1.8)
1.9 Divergence Formula. Let G be a compact domain in M with smooth
boundary ∂G and exterior normal vector field ν along ∂G. Then

G
div X =

∂G
'X, ν`.
1.1 Differential Forms. We let A

(M, R) := Λ

(T

M) be the bundle of
(multilinear) alternating forms on TM (with values in R) and A

(M, R) :=
E(A

(M, R)) be the space of differential forms on M. We let A
r
(M, R) and
A
r
(M, R) be the subbundle of alternating forms of degree r and the subspace
of differential forms on M of degree r, respectively.
The Riemannian metric on M induces a Riemannian metric on A

(M, R).
Similarly, the Levi-Civita connection induces a connection
ˆ
∇ on A

(M, R),
compare the product rule 1.20 below, and
ˆ
∇ is metric with respect to the
induced metric on A

(M, R).
Recall the interior product of a tangent vector v with an alternating form
ϕ with the same foot point, vϕ = ϕ(v, . . . ), that is, insert v as first variable.
There are the following remarkable relations between ∧ and ,
'v

∧ ϕ, ψ` = 'ϕ, vψ` (1.10)
1 Preliminaries 3
and the Clifford relation
v

∧ (wϕ) +w(v

∧ ϕ) = 'v, w`ϕ, (1.11)
where v and w are tangent vectors and ϕ and ψ alternating forms, all with the
same foot point.
We let A

(M, C) := A

(M, R)⊗
R
C be the bundle of R-multilinear alternat-
ing forms on TM with values in C and A

(M, C) := E(A

(M, C)) be the space
of complex valued differential forms on M. Such forms decompose, ϕ = ρ +iτ,
where ρ and τ are differential forms with values in R as above. We call ρ the
real part, ρ = Re ϕ, and τ the imaginary part, τ = Imϕ, of ϕ and set ϕ = ρ−iτ.
Via complex multilinear extension, we may view elements of A

(M, C) as C-
multilinear alternating forms on T
C
M = TM ⊗
R
C, the complexified tangent
bundle.
We extend Riemannian metric, wedge product, and interior product com-
plex linearly in the involved variables to T
C
M and A

(M, C), respectively. We
extend
ˆ
∇ complex linearly to a connection on A

(M, C),
ˆ
∇ϕ :=
ˆ
∇Re ϕ +
i
ˆ
∇Imϕ. Equations 1.10 and 1.11 continue to hold, but now with complex tan-
gent vectors v, w and C-valued forms ϕ and ψ. There is an induced Hermitian
metric
(ϕ, ψ) := 'ϕ, ψ` (1.12)
on A

(M, C) and a corresponding L
2
-Hermitian product on A

c
(M, C) as in
(1.2).
Let E → M be a complex vector bundle. A differential form with values
in E is a smooth section of A

(M, E) := A

(M, C) ⊗ E, that is, an element
of A

(M, E) := E(A

(M, E)). Locally, any such form is a linear combination
of decomposable differential forms ϕ ⊗σ with ϕ ∈ A

(M, C) and σ ∈ E(M, E).
We define the wedge product of ϕ ∈ A

(M, C) with α ∈ A

(M, E) by
ϕ ∧ α :=
¸
(ϕ ∧ ϕ
j
) ⊗σ
j
, (1.13)
where we decompose α =
¸
ϕ
j
⊗σ
j
. More generally, let E

and E

be further
complex vector bundles over M and µ: E⊗E

→E

be a morphism. We define
the wedge product of differential forms α ∈ A

(M, E) and α

∈ A

(M, E

) by
α ∧
µ
α

:=
¸
j,k

j
∧ ϕ

k
) ⊗µ(σ
j
⊗σ

k
) ∈ A(M, E

), (1.14)
where we write α =
¸
ϕ
j
⊗σ
j
and α

=
¸
ϕ

k
⊗ σ

k
. If α and α

are of degree
r and s, respectively, then
(α ∧
µ
α

)(X
1
, . . . , X
r+s
)
=
1
r!s!
¸
ε(σ)µ{α(X
σ(1)
, . . . , X
σ(r)
) ⊗α

(X
σ(r+1)
, . . . , X
σ(r+s)
)}. (1.15)
4 Lectures on K¨ ahler Manifolds
This formula shows that the wedge products in (1.13) and (1.14) do not depend
on the way in which we write α and α

as sums of decomposable differential
forms.
1.16 Exercise. Let E be a complex vector bundle over M and µ and λ be
composition and Lie bracket in the associated vector bundle End E of endo-
morphisms of E. Let α, α

, and α

be differential forms with values in EndE.
Then
(α ∧
µ
α

) ∧
µ
α

= α ∧
µ


µ
α

)
and
α ∧
λ
α

= α ∧
µ
α

−(−1)
rs
α


µ
α,
if α and α

have degree r and s, respectively.
1.2 Exterior Derivative. We refer to the beginning of Appendix C for some
of the terminology in this and the next subsection. Let E be a complex vector
bundle over M and D be a connection on E. For a differential form α with
values in E and of degree r, we define the exterior derivative of α (with respect
to D) by
d
D
α :=
¸
(dϕ
j
⊗σ
j
+ (−1)
r
ϕ
j
∧ Dσ
j
), (1.17)
where we decompose α =
¸
ϕ
j
⊗ σ
j
and where d denotes the usual exterior
derivative. That this is independent of the decomposition of α into a sum of
decomposable differential forms follows from
d
D
α(X
0
, . . . , X
r
) =
¸
j
(−1)
j
D
Xj
(α(X
0
, . . . ,
ˆ
X
j
, . . . , X
r
))
(1.18)
+
¸
j<k
(−1)
j+k
α([X
j
, X
k
], X
0
, . . . ,
ˆ
X
j
, . . . ,
ˆ
X
k
, . . . , X
r
).
Since d
D
(fα) −fd
D
α = df ∧ α for any function f on M, the principal symbol
σ: T

M ⊗A

(M, E) →A

(M, E) of d
D
is given by σ(ξ ⊗α) = ξ ∧ α.
1.19 Proposition. Let E be a vector bundle over M and D be a connection
on E. Then, for all α ∈ A

(M, E),
d
D
d
D
α = R
D

ε
α,
where we view the curvature tensor R
D
of D as a two-form with values in the
bundle End E of endomorphisms of E and where the wedge product is taken
with respect to the evaluation map ε: End E ⊗E →E.
Via the product rule, A

(M, E) inherits a connection
ˆ
D from
ˆ
∇ and D,
ˆ
D
X
(ϕ ⊗σ) := (
ˆ

X
ϕ) ⊗σ + ϕ ⊗(D
X
σ). (1.20)
1 Preliminaries 5
In terms of
ˆ
D, we have
d
D
α =
¸
X

j

ˆ
D
Xj
α, (1.21)
where (X
1
, . . . , X
n
) is a local frame of TM and (X

1
, . . . , X

n
) is the correspond-
ing dual frame. We emphasize that the curvature tensors of
ˆ
∇ and
ˆ
D act as
derivations with respect to wedge and tensor products.
Let E, E

, E

and µ: E ⊗E

→E

be as in (1.14). Let D, D

and D

be
connections on E, E

, and E

, respectively, such that the product rule
D

X
(µ(σ ⊗τ)) = µ((D

X
σ) ⊗τ) +µ(σ ⊗(D

X
τ)) (1.22)
holds for all vector fields X of M. The induced connections on the bundles of
forms then also satisfy the corresponding product rule,
ˆ
D

X
(α ∧
µ
α

) = (
ˆ
D
X
α) ∧
µ
α

+ (α ∧
µ
(
ˆ
D

X
α

)). (1.23)
For the exterior differential we get
d
D

(α ∧
µ
α

) = (d
D
α) ∧
µ
α

+ (−1)
r
α ∧
µ
(d
D

α

), (1.24)
where we assume that α is of degree r.
Let h = (· , · ) be a Hermitian metric on E. Then h induces a Hermitian
metric on A

(M, E), on decomposable forms given by
(ϕ ⊗σ, ψ ⊗τ) = 'ϕ, ψ`(σ, τ), (1.25)
and a corresponding L
2
-Hermitian product on A

c
(M, E) as in (1.2).
1.26 Exercise. The analogs of (1.10) and (1.11) hold on A

(M, E),
(v

∧ α, β) = (α, vβ) and v

∧ (wα) +w(v

∧ α) = 'v, w`α,
where v, w ∈ TM and α, β ∈ A

(M, E) have the same foot point.
Let D be a Hermitian connection on E. Then the induced connection
ˆ
D on
A

(M, E) as in (1.20) is Hermitian as well.
1.27 Proposition. In terms of a local orthonormal frame (X
1
, . . . , X
n
) of M,
the formal adjoint (d
D
)

of d
D
is given by
(d
D
)

α = −
¸
X
j

ˆ
D
Xj
α.
Proof. Let α and β be differential forms of degree r −1 and r. Let p ∈ M and
choose a local orthonormal frame (X
j
) around p with ∇X
j
(p) = 0. Then, at p,
(d
D
α, β) =
¸
(X

j

ˆ
D
Xj
α, β)
=
¸
(
ˆ
D
Xj
α, X
j
β)
=
¸
X
j
(α, X
j
β) −
¸
(α, X
j

ˆ
D
Xj
β).
The first term on the right is equal to the divergence of the complex vector
field Z defined by (Z, W) = (α, Wβ), see (C.3).
6 Lectures on K¨ ahler Manifolds
Since (d
D
)

(fα) − f(d
D
)

α = −gradfα for any function f on M, the
principal symbol σ: T

M ⊗A

(M, E) →A

(M, E) of (d
D
)

is given by σ(ξ ⊗
α) = −ξ

α.
1.3 Laplace Operator. As above, we let E →M be a complex vector bundle
with Hermitian metric h and Hermitian connection D. We say that a differential
form α with values in E is harmonic if d
D
α = (d
D
)

α = 0 and denote by
H

(M, E) the space of harmonic differential forms with values in E.
The Laplace operator associated to d
D
is

d
D = d
D
(d
D
)

+ (d
D
)

d
D
. (1.28)
Since the principal symbol of a composition of differential operators is the
composition of their principal symbols, the principal symbol σ of ∆
d
D is given
by
σ(ξ ⊗α) = −(ξ ∧ (ξ

α) + ξ

(ξ ∧ α)) = −||ξ||
2
α, (1.29)
by (1.11). In particular, ∆
d
D is an elliptic differential operator.
1.30 Exercise. Assume that M is closed. Use the divergence formula (1.9) to
show that
(∆
d
Dα, β)
2
= (d
D
α, d
D
β)
2
+ ((d
D
)

α, (d
D
)

β)
2
.
Conclude that ϕ is harmonic iff ∆
d
Dα = 0. Compare also Corollary C.22.
Using the Clifford relation 1.11 and the formulas 1.21 and 1.27 for d
D
and
(d
D
)

, a straightforward calculation gives the Weitzenb¨ ock formula

d
Dα = −
¸
j
ˆ
D
2
α(X
j
, X
j
) +
¸
j=k
X

k
∧ (X
j

ˆ
R
D
(X
j
, X
k
)α), (1.31)
where (X
1
, . . . , X
n
) is a local orthonormal frame of TM, (X

1
, . . . , X

n
) is the
corresponding dual frame of T

M, and
ˆ
R
D
denotes the curvature tensor of the
connection
ˆ
D on A

(M, E). Denoting the second term on the right hand side
of (1.31) by Kα and using (C.8), we can rewrite (1.31) in two ways,

d
Dα = −tr
ˆ
D
2
α +Kα =
ˆ
D

ˆ
Dα +Kα. (1.32)
Assume now that D is flat. If (Φ
j
) is a parallel frame of E over an open
subset U of M, then over U, α ∈ A(M, E) can be written as a sum α =
¸
ϕ
j
⊗Φ
j
and
d
D
α =
¸
(dϕ
j
) ⊗Φ
j
and (d
D
)

α =
¸
(d

ϕ
j
) ⊗Φ
j
. (1.33)
For that reason, we often use the shorthand dα and d

α for d
D
α and (d
D
)

α if
D is flat. Then we also have d
2
= (d

)
2
= 0, see Proposition 1.19. The latter
1 Preliminaries 7
implies that the images of d and d

are L
2
-perpendicular and that the Laplace
operator is a square,

d
= (d +d

)
2
. (1.34)
The fundamental estimates and regularity theory for elliptic differential oper-
ators lead to the Hodge decomposition of A

(M, E), see for example Section
III.5 in [LM] or Chapter 6 in [Wa]:
1.35 Theorem (Hodge Decomposition). If M is closed and D is flat, then
A

(M, E) = H

(M, E) +d(A

(M, E)) +d

(A

(M, E)),
where the sum is orthogonal with respect to the L
2
-Hermitian product on A

(M, E).
In particular, if M is closed and D is flat, then the canonical map to coho-
mology,
H

(M, E) →H

(M, E), (1.36)
is an isomorphism of vector spaces. In other words, each de Rham cohomology
class of M with coefficients in E contains precisely one harmonic representative.
In particular, dimH

(M, E) < ∞. The most important case is E = C.
1.37 Remark. It is somewhat tempting to assume that the ring structure
of H

(M, C) is also represented by H

(M, C). However, this only happens in
rare cases. As a rule, the wedge product of harmonic differential forms is not
a harmonic differential form anymore. In fact, it is a specific property of the
K¨ ahler form of a K¨ ahler manifold that its wedge product with a harmonic form
gives a harmonic form, see Theorem 5.25. Compare also Remark 1.42 below.
In the above discussion, we only considered complex vector bundles. There
is a corresponding theory in the real case, which we will use in some instances.
1.38 Exercise. Show that for ϕ ∈ A
1
(M, C), the curvature term K in the
Weitzenb¨ ock formula (1.32) is given by Kϕ = (Ric ϕ

)

, where Ric denotes the
Ricci tensor of M. In other words, ∆
d
ϕ =
ˆ


ˆ
∇ϕ + (Ric ϕ

)

.
The equation in Exercise 1.38 was observed by Bochner (see also [Ya]).
In the following theorem we give his ingenious application of it. Bochner’s
argument can be used in many other situations and is therefore named after
him. Let b
j
(M) be the j-th Betti number of M, b
j
(M) = dim
R
H
j
(M, R) =
dim
C
H
j
(M, C).
1.39 Theorem (Bochner [Boc]). Let M be a closed and connected Riemannian
manifold with non-negative Ricci curvature. Then b
1
(M) ≤ n with equality if
and onlyif M is a flat torus. If, in addition, the Ricci curvature of M is positive
in some point of M, then b
1
(M) = 0.
8 Lectures on K¨ ahler Manifolds
Proof. Since M is closed, we can represent real cohomology classes of dimension
one uniquely by harmonic one-forms, by the (identical) version of Theorem 1.35
for real vector bundles. If ϕ is such a differential form, then
0 =

M
'∆
d
ϕ, ϕ` =

M
||
ˆ
∇ϕ||
2
+

'Ric ϕ

, ϕ

`,
by Exercise 1.38. By assumption, the integrand in the second integral on the
right is non-negative. It follows that ϕ is parallel, therefore also the vector field
ϕ

, and that Ric ϕ

= 0. The rest of the argument is left as an exercise.
1.40 Remarks. 1) The earlier theorem of Bonnet–Myers makes the stronger
assertion that the fundamental group of a closed, connected Riemannian man-
ifold with positive Ricci curvature is finite.
2) Let M be a closed and connected Riemannian manifold with non-negative
Ricci curvature. If the Ricci curvature of M is positive in some point of M,
then the Riemannian metric of M can be deformed to a Riemannian metric of
positive Ricci curvature [Au1] (see also [Eh]).
3) The complete analysis of the fundamental groups of closed and con-
nected Riemannian manifolds with non-negative Ricci curvature was achieved
by Cheeger and Gromoll [CG1], [CG2]. Compare Subsection 6.1.
1.41 Exercise. Conclude from the argument in the proof of Theorem 1.39 that
a closed, connected Riemannian manifold M with non-negative Ricci curvature
is foliated by a parallel family of totally geodesic flat tori of dimension b
1
(M).
Hint: The space p of parallel vector fields on M is an Abelian subalgebra of
the Lie algebra of Killing fields on M. The corresponding connected subgroup
of the isometry group of M is closed and Abelian and its orbits foliate M by
parallel flat tori.
1.42 Remark (and Exercise). The curvature operator
ˆ
R is the symmetric
endomorphism on Λ
2
(TM) defined by the equation
'
ˆ
R(X ∧ Y ), U ∧ V ` := 'R(X, Y )V, U`. (1.43)
Gallot and Meyer showed that the curvature term in the Weitzenb¨ ock formula
(1.32) for ∆
d
on A

(M, R) is positive or non-negative if
ˆ
R > 0 or
ˆ
R ≥ 0,
respectively, see [GM] or (the proof of) Theorem 8.6 in [LM]. In particular,
if M is closed with
ˆ
R > 0, then b
r
(M) = 0 for 0 < r < n, by Hodge theory
as in (1.36) and the Bochner argument in Theorem 1.39. If M is closed with
ˆ
R ≥ 0, then real valued harmonic forms on M are parallel, again by the Bochner
argument. Since the wedge product of parallel differential forms is a parallel
differential form, hence a harmonic form, this is one of the rare instances where
the wedge product of harmonic forms is harmonic (albeit for a trivial reason),
compare Remark 1.37. If M is also connected, then a parallel differential form
on M is determined by its value at any given point of M and hence b
r
(M) ≤

n
r

for 0 < r < n. Equality for any such r implies that M is a flat torus.
1 Preliminaries 9
1.4 Hodge Operator. Suppose now that M is oriented, and denote by vol
the oriented volume form of M. Then the Hodge operator ∗ is defined
3
by the
tensorial equation
∗ϕ ∧ ψ = 'ϕ, ψ` vol. (1.44)
By definition, we have
∗ 1 = vol, ∗ vol = 1 and ∗ ∗ =
¸
(−1)
r(n−r)
P
r
, (1.45)
where P
r
: A

(M, R) →A
r
(M, R) is the natural projection.
Let ϕ and ψ be differential forms with compact support and of degree r and
r − 1, respectively. Since d ∗ ϕ is a differential form of degree n − r + 1 and
r −r
2
is even, we have
∗ ∗ d ∗ ϕ = (−1)
(n−r+1)(r−1)
d ∗ ϕ = (−1)
nr+n−r+1
d ∗ ϕ.
Hence
d(∗ϕ ∧ ψ) = d ∗ ϕ ∧ ψ + (−1)
n−r
∗ ϕ ∧ dψ
= (−1)
nr+n−r+1
∗ (∗ d ∗ ϕ) ∧ ψ + (−1)
n−r
∗ ϕ ∧ dψ
= (−1)
n−r
{(−1)
nr+1
'∗ d ∗ ϕ, ψ` +'ϕ, dψ`} vol.
(1.46)
By Stokes’ theorem, the integral over M of the left hand side vanishes. We
conclude that on differential forms of degree r
d

= (−1)
nr
∗ d ∗ . (1.47)
It is now easy to check that
∗ ∆
d
= ∆
d
∗ . (1.48)
If M is closed, then ∗ maps harmonic forms to harmonic forms, by Exercise 1.30
and Equation 1.48. Hence, for closed M,
∗: H
r
(M, R) →H
n−r
(M, R) (1.49)
is an isomorphism. This is Poincar´e duality on the level of harmonic forms.
Extend ∗ complex linearly to A

(M, C). Let E be a vector bundle over M
and E

be the dual bundle of E. Assume that E is endowed with a Hermitian
metric. Via h(σ)(τ) = (σ, τ) view the Hermitian metric of E as a conjugate
linear isomorphism h: E →E

. We obtain a conjugate linear isomorphism
∗ ⊗h: A

(M, E) →A

(M, E

), (1.50)
where ∗ϕ := ∗ϕ. We have
((∗ ⊗h)α) ∧
ε
β = (α, β) vol, (1.51)
3
Note that the definition here differs from the standard one, the definition here gives a
more convenient sign in (1.47).
10 Lectures on K¨ ahler Manifolds
where ε: E

⊗ E → C is the evaluation map. Note that ((∗ ⊗ h)α) ∧
ε
β is
complex valued.
Let D

be the induced connection on E

. With respect to D and D

, h is
a parallel morphism from E to E

,
(D

X
(h(σ)))(τ) = X(σ, τ) −(σ, D
X
τ) = (D
X
σ, τ) = h(D
X
σ)(τ).
The induced (conjugate) Hermitian metric h

on E

is given by
(h(σ), h(τ)) := (σ, τ). (1.52)
Note that h

is complex linear in the first and conjugate linear in the second
variable. The connection D

is Hermitian with respect to h

,
X(h(σ), h(τ)) = X(σ, τ) = (D
X
σ, τ) + (σ, D
X
τ)
= (h(D
X
σ), h(τ)) + (h(σ), h(D
X
τ))
= (D

(h(σ)), h(τ)) + (h(σ), D

X
(h(τ)).
Via ξ(h

(η)) = (ξ, η) we consider h

as a conjugate linear isomorphism from
E

to E. We have
(σ, h

(h(τ))) = h(σ)(h

(h(τ))) = (h(σ), h(τ)) = (σ, τ),
and hence h

h = id. It follows that
(∗ ⊗h

)(∗ ⊗h) = (∗ ⊗ h)(∗ ⊗h

) = (−1)
r(n−r)
(1.53)
on forms of degree r and n − r. Using (1.24) with D

the usual derivative of
functions and computing as in (1.46), we get
(d
D
)

= (−1)
nr
(∗ ⊗h

) d
D

(∗ ⊗h). (1.54)
This implies that the corresponding Laplacians, for simplicity denoted ∆, sat-
isfy
∆(∗ ⊗h) = (∗ ⊗h) ∆. (1.55)
If M is closed, then ∗ ⊗ h maps harmonic forms to harmonic forms, by Ex-
ercise 1.30 and Equation 1.55. Hence, for closed M, ∗ ⊗ h induces conjugate
linear isomorphisms
H
r
(M, E) →H
n−r
(M, E

). (1.56)
This is Poincar´e duality for vector bundle valued harmonic forms.
2 Complex Manifolds
Let V be a vector space over R. A complex structure on V is an endomorphism
J : V →V such that J
2
= −1. Such a structure turns V into a complex vector
space by defining multiplication with i by iv := Jv. Vice versa, multiplication
by i in a complex vector space is a complex structure on the underlying real
vector space.
2.1 Example. To fix one of our conventions, we discuss the complex vector
space C
m
explicitly. Write a vector in C
m
as a tuple
(z
1
, . . . , z
m
) = (x
1
+iy
1
, . . . , x
m
+iy
m
)
and identify it with the vector (x
1
, y
1
, . . . , x
m
, y
m
) in R
2m
. The corresponding
complex structure on R
2m
is
J(x
1
, y
1
, . . . , x
m
, y
m
) = (−y
1
, x
1
, . . . , −y
m
, x
m
).
We will use this identification of C
m
with R
2m
and complex structure J on
R
2m
without further reference.
Let M be a smooth manifold of real dimension 2m. We say that a smooth
atlas A of M is holomorphic if for any two coordinate charts z : U →U

⊂ C
m
and w: V → V

⊂ C
m
in A, the coordinate transition map z ◦ w
−1
is holo-
morphic. Any holomorphic atlas uniquely determines a maximal holomorphic
atlas, and a maximal holomorphic atlas is called a complex structure. We say
that M is a complex manifold of complex dimension m if M comes equipped
with a holomorphic atlas. Any coordinate chart of the corresponding com-
plex structure will be called a holomorphic coordinate chart of M. A Riemann
surface or complex curve is a complex manifold of complex dimension 1.
Let M be a complex manifold. Then the transition maps z ◦ w
−1
of holo-
morphic coordinate charts are biholomorphic. Hence they are diffeomorphisms
and the determinants of their derivatives, viewed as R-linear maps, are posi-
tive. It follows that a holomorphic structure determines an orientation of M,
where we choose dx
1
∧ dy
1
∧ · · · ∧ dx
m
∧ dy
m
as orientation of C
m
, compare
Example 2.1.
We say that a map f : M → N between complex manifolds is holomorphic
if, for all holomorphic coordinate charts z : U → U

of M and w: V → V

of
N, the map w ◦ f ◦ z
−1
is holomorphic on its domain of definition. We say
that f is biholomorphic if f is bijective and f and f
−1
are holomorphic. An
automorphism of a complex manifold M is a biholomorphic map f : M →M.
To be consistent in what we say next, we remark that open subsets of
complex manifolds inherit a complex structure. For an open subset U of a
complex manifold M, we denote by O(U) the set of holomorphic functions
f : U →C, a ring under pointwise addition and multiplication of functions.
12 Lectures on K¨ ahler Manifolds
The inverse mapping and implicit function theorem also hold in the holo-
morphic setting. Corresponding to the real case, we have the notions of holo-
morphic immersion, holomorphic embedding, and complex submanifold. The
discussion is completely parallel to the discussion in the real case.
Of course, complex analysis is different from real analysis. To state just
one phenomenon where they differ, by the maximum principle a holomorphic
function on a closed complex manifold is locally constant. In particular, C
m
does not contain closed complex submanifolds (of positive dimension).
2.2 Examples. 1) Let U ⊂ C
m
be an open subset. Then M together with the
atlas consisting of the one coordinate chart id: U →U is a complex manifold.
2) Riemann sphere. Consider the unit sphere
S
2
= {(w, h) ∈ C R | ww +h
2
= 1}.
Let N = (0, 1) and S = (0, −1) be the north and south pole of S
2
, respectively.
The stereographic projections π
N
: S
2
\ {N} → C and π
S
: S
2
\ {S} → C are
given by π
N
(w, h) = (1 −h)
−1
w and π
S
(w, h) = (1 +h)
−1
w, respectively. The
transition map π
S
◦ π
−1
N
: C \ {0} → C \ {0} is given by (π
S
◦ π
−1
N
)(z) = 1/z.
It is smooth, and thus π
N
and π
S
define a smooth atlas of S
2
. However, it
is not holomorphic. We obtain a holomorphic atlas by replacing π
S
by its
complex conjugate, π
S
. Then the transition map is (π
S
◦ π
−1
N
)(z) = 1/z, and
hence the atlas of S
2
consisting of π
N
and π
S
is holomorphic. The Riemann
sphere is S
2
together with the complex structure determined by this atlas. It
is a consequence of the uniformization theorem that this complex structure on
S
2
is unique up to diffeomorphism. As we will see, the Riemann sphere is
biholomorphic to the complex line CP
1
, described in the next example.
3) Complex projective spaces. As a set, complex projective space CP
m
is the
space of all complex lines in C
m+1
. For a non-zero vector z = (z
0
, . . . , z
m
) ∈
C
m+1
, we denote by [z] the complex line generated by z and call (z
0
, . . . , z
m
)
the homogeneous coordinates of [z]. For 0 ≤ j ≤ m, we let U
j
= {[z] ∈ CP
m
|
z
j
= 0}. Each [z] in U
j
intersects the affine hyperplane {z
j
= 1} in C
m+1
in
exactly one point. We use this to obtain a coordinate map
a
j
: U
j
→C
m
, a
j
([z]) =
1
z
j
(z
0
, . . . , ˆ z
j
, . . . , z
m
),
where the hat indicates that z
j
is to be deleted. By what we said it is clear
that a
j
is a bijection. For j < k, the transition map a
j
◦ a
−1
k
is defined on
{w ∈ C
m
| w
j
= 0} and given by inserting 1 as k-th variable, multiplying the
resulting (m+1)-vector by (w
j
)
−1
, and deleting the redundant j-th variable 1.
Thus the transition maps are holomorphic. It is now an exercise to show that
there is precisely one topology on CP
m
such that the maps a
j
are coordinate
charts and such that CP
m
together with this topology and the atlas of maps
a
j
is a complex manifold of complex dimension m. For m = 1, we speak of the
complex projective line, for m = 2 of the complex projective plane.
2 Complex Manifolds 13
For m ≤ n, the map f : CP
m
→ CP
n
, [z] → [z, 0], is a holomorphic em-
bedding. More generally, if A: C
m+1
→ C
n+1
is an injective linear map, then
the induced map f : CP
m
→ CP
n
, [z] → [Az], is a holomorphic embedding.
Thus we can view CP
m
in many different ways as a complex submanifold of
CP
n
. We also conclude that the group PGl(m+ 1, C) = Gl(m+ 1, C)/C

acts
by biholomorphic transformations on CP
m
. It is known that PGl(m+ 1, C) is
actually equal to the group of biholomorphic transformations of CP
m
.
For the Riemann sphere S
2
as in the previous example, the map f : S
2

CP
1
,
f(p) =


N
(p), 1] if p = N,
[1, π
S
(p)] if p = S,
is well defined and biholomorphic and thus identifies the Riemann sphere with
the complex projective line.
4) Complex Grassmannians. This example generalizes the previous one.
Let V be a complex vector space of dimension n and G
r
V be the space of
r-dimensional complex subspaces of V , where 0 < r < n.
Let M

be the set of linear maps F : C
r
→V of rank r. There is a canonical
projection
π: M

→G
r
V, π(F) = [F] =: imF.
Let B: V →C
n
be an isomorphism. Then the map
M

→C
n×r
, F →Mat(BF),
where Mat(BF) denotes the matrix of the linear map BF : C
r
→ C
n
, is a
bijection onto the open subset of (n r)-matrices of rank r. This turns M

into a complex manifold of dimension nr, and the complex structure on M

does not depend on the choice of B. We write
Mat(BF) =

F
0
F
1

, where F
0
∈ C
r×r
and F
1
∈ C
(n−r)×r
,
and let U
B
be the subset of [F] in G
r
V such that F
0
has rank r. We leave it
as an exercise to the reader to show that
Z
B
: U
B
→C
(n−r)×r
, Z
B
([F]) = F
1
F
−1
0
,
is a well defined bijection and that, for any two isomorphisms B, C: V → C
n
,
the transition map Z
B
◦Z
−1
C
is holomorphic. With the same arguments as in the
previous example we get that G
r
V has a unique topology such that the maps
Z
B
are coordinate charts turning G
r
V into a complex manifold of complex
dimension r(n − r). Moreover, π: M

→ G
r
V is a holomorphic submersion
and, for any isomorphism B: V →C
n
,
ϕ
B
: U
B
→M

, ϕ
B
([F]) = B
−1

1
F
1
F
−1
0

,
14 Lectures on K¨ ahler Manifolds
where 1 stands for the r r unit matrix and where we consider the matrix on
the right as a linear map C
r
→C
n
, is a holomorphic section of π.
The group Gl(r, C) of invertible matrices in C
r×r
is a complex Lie group
(for complex Lie groups, see Example 8 below) and, considered as group of
automorphisms of C
r
, acts on M

on the right,
M

Gl(r, C) →M

, (F, A) →FA.
This action is holomorphic and turns π: M

→ G
r
V into a principal bundle
with structure group Gl(r, C). The complex Lie group Gl(V ) of automorphisms
of V acts on M

and G
r
V on the left,
Gl(V ) M

→M

, (A, F) →AF,
respectively
Gl(V ) G
r
V →G
r
V, (A, [F]) →[AF].
These actions are also holomorphic and π is equivariant with respect to them.
5) Tautological or universal bundle. Let 0 < r < n and M = G
r
V be
the Grassmannian of r-dimensional complex linear subspaces in V as in the
previous example. As a set, the universal bundle over G
r
V is equal to
U
r
V = {(W, w) | W ∈ G
r
V, w ∈ W}.
There is a canonical projection
π: U
r
V →G
r
V, (W, w) →W.
For each W ∈ G
r
V , the bijection π
−1
(W) ÷ (W, w) → w ∈ W turns the fiber
π
−1
(W) into an r-dimensional complex vector space isomorphic to W.
Let B: V → C
n
be an isomorphism and U
B
⊂ G
r
V be as in the previous
example. Define a bijection
Φ
B
: U
B
C
r
→π
−1
(U
B
), Φ
B
([F], v) = ([F], ϕ
B
([F])v),
where ϕ
B
is as in the previous example. For each [F] ∈ U
B
, the map
C
r
→π
−1
([F]), v →Φ
B
([F], v),
is an isomorphism of vector spaces. With arguments similar to the ones in the
previous examples it follows that U
r
V is a complex manifold in a unique way
such that π: U
r
V →G
r
V is a complex vector bundle over G
r
V and such that
the trivializations Φ
B
as above are holomorphic. In particular, the complex
dimension of U
r
V is r(n − r + 1) and π is holomorphic. Moreover, the left
action of Gl(V ) on G
r
V extends canonically to a holomorphic action on U
r
V ,
Gl(V ) U
r
V →U
r
V, (A, (W, w)) →(AW, Aw).
2 Complex Manifolds 15
This action has two orbits: the set of pairs (W, w) with w = 0 and of pairs
(W, 0).
6) Complex tori. Choose an R-basis B = (b
1
, . . . , b
2m
) of C
m
. Let Γ ⊂ C
m
be the lattice consisting of all integral linear combinations of B, a discrete
subgroup of the additive group C
m
. Then Γ acts by translations on C
m
,
(k, z) →t
k
(z) := k +z.
This action is free and properly discontinuous. For each fixed k ∈ Γ, the
translation t
k
: C
m
→ C
m
is biholomorphic. Hence the quotient T = Γ\C
m
inherits the structure of a complex manifold such that the covering map C
m

T is holomorphic. Now T is diffeomorphic to the 2m-fold power of a circle,
hence T with the above complex structure is called a complex torus. We also
note that addition T T →T, (z, w) →z +w, is well defined and holomorphic
and turns T into a complex Lie group as in Example 8 below.
Let T = Γ\C
m
be a complex torus and f : T →T be a biholomorphic map.
Then any continuous lift g : C
m
→C
m
of f is Γ-equivariant and biholomorphic.
Continuity implies that there is a constant C such that |g(z)| ≤ C(1+|z|) for all
z ∈ C
m
. By a standard result from complex analysis, g is affine. Hence f is of
the form f(z) = Az +b with b ∈ T and A ∈ Gl(m, C) such that A(Γ) = Γ. Vice
versa, for any such A ∈ Gl(m, C) and b ∈ T, the map f : T →T, f(z) = Az +b,
is well defined and biholomorphic.
A one-dimensional complex torus is called an elliptic curve. It follows from
the uniformization theorem that any complex curve diffeomorphic to the torus
S
1
S
1
is an elliptic curve. In particular, the complex structure described in
the next example turns S
1
S
1
into an elliptic curve.
7) Hopf manifold (complex structures on S
2m−1
S
1
). Let m ≥ 1 and
z ∈ C be a non-zero complex number of modulus |z| = 1. Then Z acts freely
and properly discontinuously on C
m
\ {0} by (k, v) → z
k
· v. The quotient
M = (C
m
\ {0})/Z is called a Hopf manifold. It is an exercise to show that M
is diffeomorphic to S
2m−1
S
1
. (Hint: Consider the case z = 2 first.) Since
multiplication by z
k
is biholomorphic, M inherits from C
m
\ {0} the structure
of a complex manifold.
A generalization of this example is due to Calabi and Eckmann, who showed
that the product of odd-dimensional spheres carries a complex structure, see
[CE] (Example 2.5 in [KN, Chapter IX]).
8) Complex Lie groups. As an open subset of C
n×n
, the general linear group
G = Gl(n, C) is a complex manifold of complex dimension n
2
, and multipli-
cation G G → G and inversion G → G are holomorphic maps. The special
linear group Sl(n, C) ⊂ Gl(n, C) is a complex Lie subgroup of complex codimen-
sion 1. If V is a complex vector space of dimension n, then any isomorphism
B: V → C
n
identifies the general linear group Gl(V ) of V with Gl(n, C) and
turns Gl(V ) into a complex Lie group, independently of the choice of B, and the
special linear group Sl(V ) is a complex Lie subgroup of complex codimension
1.
16 Lectures on K¨ ahler Manifolds
The unitary group U(n) is not a complex Lie group – recall that its defining
equation is not holomorphic. In fact, the Lie algebra g of a complex Lie group
G is a complex vector space and the adjoint representation Ad of G is a holo-
morphic map into the complex vector space of complex linear endomorphisms
of g. Hence Ad is constant if G is compact. It follows that compact complex
Lie groups are Abelian, that is, complex tori.
For a Lie group G, a complexification of Gconsists of a complex Lie group G
C
together with an inclusion G → G
C
such that any smooth morphism G → H,
where H is a complex Lie group, extends uniquely to a holomorphic morphism
G
C
→ H. Clearly, if G
C
exists, then it is unique up to isomorphism. For
example, Sl(n, C) is the complexification of Sl(n, R). Any connected compact
Lie group has a complexification [Bu, Section 27]; e.g., the complexifications of
SO(n), SU(n), and U(n) are SO(n, C), Sl(n, C), and Gl(n, C), respectively.
9) Homogeneous spaces. We say that a complex manifold M is homogeneous
if the group of automorphisms of M is transitive on M. For example, if G
is a complex Lie group and H is a closed complex Lie subgroup of G, then
the quotient G/H is in a unique way a homogeneous complex manifold such
that the natural left action by G on G/H and the projection G → G/H are
holomorphic. Flag manifolds, that is, coadjoint orbits of connected compact
Lie groups, are homogeneous complex manifolds. In fact, if G is a connected
compact Lie group and G
λ
is the stabilizer of some λ ∈ g

under the coadjoint
representation, then the inclusion of G into its complexification G
C
induces an
isomorphism G/G
λ
→G
C
/P
λ
, where P
λ
is a suitable parabolic subgroup of G
C
associated to λ, a closed complex Lie subgroup of G
C
, see Section 4.12 in [DK].
For example, U(n)/T = Gl(n, C)/B, where T is the maximal torus of diagonal
matrices in U(n) and B is the Borel group of all upper triangular matrices in
Gl(n, C), the stabilizer of the standard flag in C
n
. For more on flag manifolds,
see Chapter 8 in [Bes].
10) Projective varieties. A closed subset V ⊂ CP
n
is called a (complex)
projective variety if, locally, V is defined by a set of complex polynomial equa-
tions. Outside of its singular locus, that is, away from the subset where the
defining equations do not have maximal rank, a projective variety is a complex
submanifold of CP
n
. We say that V is smooth if its singular locus is empty.
A well known theorem of Chow says that any closed complex submanifold of
CP
n
is a smooth projective variety, see [GH, page 167].
We say that V is a rational curve if V is smooth and biholomorphic to
CP
1
. For example, consider the complex curve C = {[z] ∈ CP
2
| z
2
0
= z
1
z
2
}
in CP
2
, which is contained in U
1
∪ U
2
. On U
1
∩ C we have z
2
/z
1
= (z
0
/z
1
)
2
,
hence we may use u
1
= z
0
/z
1
as a holomorphic coordinate for C on U
1
∩ C.
Similarly, on U
2
∩ C we have z
1
/z
2
= (z
0
/z
2
)
2
and we may use u
2
= z
0
/z
2
as
a holomorphic coordinate for C on U
2
∩ C. The coordinate transformation on
U
1
∩ U
2
∩ C is u
2
= 1/u
1
. Thus C is biholomorphic to CP
1
and hence is a
rational curve in CP
2
. In this example, the defining equation has degree two.
2 Complex Manifolds 17
The map CP
1
÷ [1, z] → [1, z, z
2
, . . . , z
m
] ∈ CP
m
extends to a holomorphic
embedding of CP
1
into CP
m
, and the maximal degree of the obvious defining
equations of the image is m.
Let M be a complex manifold. We say that a complex vector bundle E →M
is holomorphic if E is equipped with a maximal atlas of trivializations whose
transition functions are holomorphic. Such an atlas turns E into a complex
manifold such that the projection E →M is holomorphic.
2.3 Examples. 1) The tangent bundle TM together with its complex structure
J is a complex vector bundle over M. The usual coordinates for the tangent
bundle have holomorphic transition maps and thus turn TM into a complex
manifold and holomorphic vector bundle over M.
2) If E → M is holomorphic, then the complex tensor bundles associated
to E are holomorphic. For example, the dual bundle E

is holomorphic. Note
however that TM ⊗
R
C is not a holomorphic vector bundle over M in any
natural way.
3) Let E → M be a holomorphic vector bundle and f : N → M be a
holomorphic map. Then the pull back f

E →N is holomorphic.
4) The universal bundle U
r
V →G
r
V is holomorphic.
Let (z, U) be a holomorphic coordinate chart of M. As usual, we let z
j
=
x
j
+iy
j
and write the corresponding coordinate frame as (X
1
, Y
1
, . . . , X
m
, Y
m
).
For p ∈ U, we define a complex structure J
p
on T
p
M by
J
p
X
j
(p) = Y
j
(p), J
p
Y
j
(p) = −X
j
(p). (2.4)
Since the transition maps of holomorphic coordinate charts are holomorphic, J
p
is independent of the choice of holomorphic coordinates. We obtain a smooth
field J = (J
p
) of complex structures on TM.
Vice versa, if M is a smooth manifold of real dimension 2m, then a smooth
field J = (J
p
) of complex structures on TM is called an almost complex struc-
ture of M. An almost complex structure J = J
p
is called a complex structure if
it comes from a complex structure on M as in (2.4) above. Any almost complex
structure on a surface is a complex structure (existence of isothermal coordi-
nates). A celebrated theorem of Newlander and Nirenberg [NN] says that an
almost complex structure is a complex structure if and only if its Nijenhuis
tensor or torsion N vanishes, where, for vector fields X and Y on M,
N(X, Y ) = 2{[JX, JY ] −[X, Y ] −J[X, JY ] −J[JX, Y ]}. (2.5)
For an instructive discussion of the Newlander–Nirenberg theorem and its proof,
see [Ka2, Section 6.3].
2.6 Exercises. 1) N is a tensor. Compare also Exercises 2.15 and 2.32.
2) Let ∇ be a torsion free connection and J be an almost complex structure
on M. Show that
1
2
N(X, Y ) = ∇J(JX, Y ) −J∇J(X, Y ) −∇J(JY, X) +J∇J(Y, X)
18 Lectures on K¨ ahler Manifolds
to conclude that J is a complex structure if ∇J = 0.
2.7 Example. In the normed division algebra Ca of Cayley numbers
4
, consider
the sphere S
6
of purely imaginary Cayley numbers of norm one. For a point
p ∈ S
6
and tangent vector v ∈ T
p
S
6
, define
J
p
v := p · v,
where the dot refers to multiplication in Ca. Since p is purely imaginary of
length one, p · (p · x) = −x for all x ∈ Ca. It follows that J
p
v ∈ T
p
S
6
and that
J
2
p
v = p · (p · v) = −v.
Hence J = (J
p
) is an almost complex structure on S
6
. Since |x · y| = |x| · |y|
for all x, y ∈ Ca and p ∈ S
6
has norm one, J is norm preserving.
By parallel translation along great circle arcs through p, extend v to a vector
field V in a neighborhood of p in S
6
. Then V is parallel at p. Along the great
circle arc cos t · p+sint · u in the direction of a unit vector u in T
p
S
6
, the vector
field JV is given by (cos t · p + sin t · u) · V (cos t · p + sin t · u). Hence
dJV (u) = u · v +p · dV (u) = u · v +p · S(u, v),
where S denotes the second fundamental form of S
6
. For v, w ∈ T
p
S
6
, we get
N(v, w) = 2{(p · v) · w −(p · w) · v −p · (v · w) +p · (w · v)} = 4[p, v, w],
where [x, y, z] = (x · y) · z −x · (y · z) denotes the associator of x, y, z ∈ Ca. We
conclude that N = 0 and hence that J does not come from a complex structure
on S
6
. It is a famous open problem whether S
6
carries any complex structure.
2.8 Exercise. View the sphere S
2
as the space of purely imaginary quaternions
of norm one and discuss the corresponding almost complex structure on S
2
.
2.1 Complex Vector Fields. Let V be a real vector space, and let J be a
complex structure on V . We extend J complex linearly to the complexification
V
C
= V ⊗
R
C of V ,
J(v ⊗α) := (Jv) ⊗α. (2.9)
Then we still have J
2
= −1, hence V
C
is the sum
V
C
= V

⊕V

(2.10)
of the eigenspaces V

and V

for the eigenvalues i and −i, respectively. The
maps
V →V

, v →v

:=
1
2
(v −iJv), V →V

, v →v

:=
1
2
(v +iJv), (2.11)
4
Chapter 15 in [Ad] is a good reference to Cayley numbers.
2 Complex Manifolds 19
are complex linear and conjugate linear isomorphisms, respectively, if we con-
sider V together with J as a complex vector space.
Let M be a smooth manifold with an almost complex structure J and
T
C
M = TM ⊗
R
C be the complexified tangent bundle. As in (2.10), we have
the eigenspace decomposition with respect to J,
T
C
M = T

M ⊕T

M. (2.12)
The decomposition in (2.11) shows that T

M and T

M are smooth subbundles
of the complexified tangent bundle T
C
M. Moreover, by (2.11) the maps
TM →T

M, v →v

, and TM →T

M, v →v

, (2.13)
are complex linear respectively conjugate linear isomorphisms of complex vector
bundles over M, where multiplication by i on TM is given by J.
A complex vector field of M is a section of T
C
M. Any such field can be
written in the form Z = X + iY , where X and Y are vector fields of M, that
is, sections of TM. Complex vector fields act as complex linear derivations on
smooth complex valued functions. We extend the Lie bracket complex linearly
to complex vector fields,
[X +iY, U +iV ] := [X, U] −[Y, V ] +i([X, V ] + [Y, U]). (2.14)
2.15 Exercise. The Nijenhuis tensor associated to J vanishes iff T

M is an
involutive distribution of T
C
M, that is, if [Z
1
, Z
2
] is a section of T

M whenever
Z
1
and Z
2
are. And similarly for T

M. More precisely,
N(X, Y )

= −8[X

, Y

]

and N(X, Y )

= −8[X

, Y

]

,
where X and Y are vector fields on M.
Suppose from now on that M is a complex manifold. Then TM (with com-
plex multiplication defined via J) is a holomorphic vector bundle over M. The
isomorphism TM ÷ v → v

∈ T

M as in (2.13) turns T

M into a holomorphic
vector bundle. The bundle T

M is a smooth complex vector bundle over M,
but not a holomorphic vector bundle in a natural way.
Let (z, U) be a holomorphic coordinate chart for M. Write z
j
= x
j
+ iy
j
and set
Z
j
:=

∂z
j
=
1
2
(X
j
−iY
j
), Z
j
:=

∂z
j
=
1
2
(X
j
+iY
j
), (2.16)
where (X
1
, Y
1
, . . . , X
m
, Y
m
) is the coordinate frame associated to the coordi-
nates (x
1
, y
1
, . . . , x
m
, y
m
). In the notation of (2.13), Z
j
= X

j
and Z
j
= X

j
.
Similarly, any complex vector field Z has components
Z

=
1
2
(Z −iJZ) and Z

=
1
2
(Z +iJZ) (2.17)
20 Lectures on K¨ ahler Manifolds
in T

M and T

M, respectively. We note that (X
1
, . . . , X
m
) is a local holomor-
phic frame for TM considered as a holomorphic vector bundle over M and that
(Z
1
, . . . , Z
m
) is the corresponding local holomorphic frame of T

M.
We say that a real vector field X on M is automorphic if the flow of X
preserves the complex structure J of M. This holds iff the Lie derivative
L
X
J = 0 or, equivalently, iff [X, JY ] = J[X, Y ] for all vector fields Y on M.
The space a(M) of automorphic vector fields on M is a Lie algebra with respect
to the Lie bracket of vector fields.
2.18 Proposition. A real vector field X on M is automorphic iff it is holomor-
phic as a section of the holomorphic vector bundle TM. The complex structure
J turns a(M) into a complex Lie algebra.
Proof. To be automorphic or holomorphic is a local property. Hence we can
check the equivalence of the two properties in holomorphic coordinates (z, U).
Then the vector field is given by a smooth map X: U →C
m
and J is given by
multiplication by i.
Let Y : U →C
m
be another vector field. Then the Lie bracket of X with Y
is given by dX(Y ) −dY (X). Hence we get
[X, iY ] = dX(iY ) −idY (X)
= ∂X(iY ) + ∂X(iY ) −idY (X)
= i∂X(Y ) −i∂X(Y ) −idY (X)
= i∂X(Y ) +i∂X(Y ) −idY (X) −2i∂X(Y )
= i[X, Y ] −2i∂X(Y ).
Hence [X, iY ] = i[X, Y ] iff ∂X(Y ) = 0.
2.19 Proposition. If M is a closed complex manifold. then dima(M) < ∞.
Proof. The space of holomorphic sections of a holomorphic vector bundle E
over M is precisely the kernel of the elliptic differential operator ∂ on the space
of smooth sections of E, compare (3.4).
2.20 Remark. A celebrated theorem of Bochner and Montgomery states that,
for a closed complex manifold M, the group Aut(M, J) of automorphisms
f : M → M is a complex Lie group with respect to the compact-open topol-
ogy
5
and that a(M) is the Lie algebra of Aut(M, J), see [BM]. In particular,
Aut(M, J) is either trivial, or a complex torus, or is not compact, compare
Example 2.2.8.
5
This is a subtle point. Without the assertion about the topology, the theorem is a trivial
consequence of Proposition 2.19. There is a corresponding common misunderstanding in the
case of isometry groups of Riemannian manifolds and of other groups of automorphisms.
2 Complex Manifolds 21
2.2 Differential Forms. As above, let V be a real vector space with complex
structure J. The decomposition of V
C
in (2.10) determines a decomposition of
the space of complex valued alternating forms on V
C
,
Λ
r
V

C
= Λ
r
(V

⊕ V

)

= ⊕
p+q=r

Λ
p
(V

)

⊗Λ
q
(V

)

=: ⊕
p+q=r
Λ
p,q
V

C
. (2.21)
Alternating forms on V
C
correspond to complex multilinear extensions of com-
plex valued alternating forms on V . In this interpretation, Λ
p,q
V

C
corresponds
to complex valued alternating r-forms ϕ on V , r = p +q, such that, for α ∈ C,
ϕ(αv
1
, . . . , αv
r
) = α
p
α
q
ϕ(v
1
, . . . , v
r
), (2.22)
where we view V together with J as a complex vector space as usual. To see
this, write v
j
= v

j
+ v

j
as in (2.11). We call elements of Λ
p,q
V

C
alternating
forms on V of type (p, q). An alternating r-form ϕ of type (p, q) satisfies
(J

ϕ)(v
1
, . . . , v
r
) := ϕ(Jv
1
, . . . , Jv
r
) = i
p−q
ϕ(v
1
, . . . , v
r
), (2.23)
but this does not characterize the type. We note however that a non-zero
alternating r-form ϕ satisfies J

ϕ = ϕ iff r is even and ϕ is of type (r/2, r/2).
Conjugation maps Λ
p,q
V

C
to Λ
q,p
V

C
, and Λ
p,p
V

C
is invariant under conju-
gation. The space of complex (p, p)-forms fixed under conjugation is the space
of real valued forms of type (p, p).
Suppose now that M is a complex manifold with complex structure J.
Note that A

(M, C) = Λ

T

C
M, where T

C
M = T

M ⊗
R
C is the complexified
cotangent bundle. As in (2.21), we have the decomposition
A
r
(M, C) = ⊕
p+q=r
A
p,q
(M, C), (2.24)
where A
p,q
(M, C) := Λ
p,q
T

C
M. The complex line bundle K
M
:= A
m,0
(M, C)
plays a special role, it is called the canonical bundle. Smooth sections of
A
p,q
(M, C) are called differential forms of type (p, q), the space of such dif-
ferential forms is denoted A
p,q
(M, C).
Let z : U →U

be a holomorphic coordinate chart for M. Write z
j
= x
j
+iy
j
and set
dz
j
= dx
j
+idy
j
and dz
j
= dx
j
−idy
j
, (2.25)
differential forms of type (1, 0) and (0, 1), respectively. In terms of these, a
differential form ϕ of type (p, q) is given by a linear combination
¸
J,K
a
JK
dz
J
∧ dz
K
=
¸
J,K
a
JK
dz
i1
∧ · · · ∧ dz
ip
∧ dz
j1
∧ · · · ∧ dz
jq
, (2.26)
where J and K run over multi-indices j
1
< · · · < j
p
and k
1
< · · · < k
q
. Under
a transformation z ◦ w
−1
of holomorphic coordinates of M, we have
dz
j
=
¸
∂z
j
∂w
k
dw
k
and dz
j
=
¸
∂z
j
∂w
k
dw
k
. (2.27)
22 Lectures on K¨ ahler Manifolds
This shows that the natural trivializations by the forms dz
I
as above turn the
bundles A
p,0
(M, C) into holomorphic bundles over M. We also see that the
complex vector bundles A
r
(M, C) and A
p,q
(M, C) are not holomorphic in any
natural way for 0 < r ≤ 2m and 0 < q ≤ m, respectively.
A quick computation gives
dϕ =
¸

X
j
(a
JK
)dx
j
+Y
j
(a
JK
)dy
j

∧ dz
J
∧ dz
K
=
¸

Z
j
(a
JK
)dz
j
+Z
j
(a
JK
)dz
j

∧ dz
J
∧ dz
K
=: ∂ϕ + ∂ϕ.
(2.28)
The type of ∂ϕ is (p + 1, q), the type of ∂ϕ is (p, q + 1), hence they are well
defined, independently of the choice of holomorphic coordinates. Now d = ∂+∂
and d
2
= 0. Hence by comparing types, we get

2
= 0, ∂
2
= 0, ∂∂ = −∂∂. (2.29)
In particular, we get differential cochain complexes
· · ·

−→ A
p,q−1
(M, C)

−→ A
p,q
(M, C)

−→ A
p,q+1
(M, C)

−→ · · · (2.30)
whose cohomology groups H
p,q
(M, C) are called Dolbeault cohomology groups
of M. Their dimensions, h
p,q
(M, C) = dim
C
H
p,q
(M, C), are called Hodge
numbers of M. They are invariants associated to the complex structure of M.
The kernel Ω
p
(M) of ∂ on A
p,0
(M, C) consists precisely of the holomor-
phic sections of the holomorphic vector bundle A
p,0
(M, C). These are called
holomorphic forms of degree p. By definition, Ω
p
(M)

= H
p,0
(M, C). The
alternating sum
χ(M, O) :=
¸
(−1)
p
h
p,0
(M, C) =
¸
(−1)
p
dim
C

p
(M) (2.31)
is called the arithmetic genus of M.
2.32 Exercise. Let M be a smooth manifold with an almost complex structure
J. As in the case of complex manifolds, we have the decomposition
A
r
(M, C) = ⊕
p+q=r
A
p,q
(M, C)
into types. If ϕ is a smooth complex valued function, that is ϕ ∈ A
0
(M, C),
then we can decompose as in the case of complex manifolds,
dϕ = d
1,0
ϕ +d
0,1
ϕ ∈ A
1,0
(M, C) + A
0,1
(M, C),
where d
1,0
= ∂ and d
0,1
= ∂ if J is a complex structure. However, if ϕ is of
type (1, 0), then a new term may arise,
dϕ = d
1,0
ϕ +d
0,1
ϕ +d
−1,2
ϕ ∈ A
2,0
(M, C) + A
1,1
(M, C) + A
0,2
(M, C),
2 Complex Manifolds 23
and similarly for ϕ ∈ A
0,1
(M, C). Show that d
−1,2
is a tensor field and that
ϕ(N(X, Y )) = 8 · d
−1,2
ϕ(X, Y )
for all ϕ ∈ A
1,0
(M, C) and vector fields X, Y on M. Any differential form is
locally a finite sum of decomposable differential forms. Conclude that
dϕ ∈ A
p+2,q−1
(M, C) + A
p+1,q
(M, C) + A
p,q+1
(M, C) + A
p−1,q+2
(M, C)
for any differential form ϕ of type (p, q). Which parts of dϕ are tensorial in ϕ?
2.3 Compatible Metrics. Let M be a complex manifold with corresponding
complex structure J. We say that a Riemannian metric g = '· , · ` is compatible
with J if
'JX, JY ` = 'X, Y ` (2.33)
for all vector fields X, Y on M. A complex manifold together with a compatible
Riemannian metric is called a Hermitian manifold
6
.
Let M be a complex manifold as above. If g is a compatible Riemannian
metric on M, then the complex bilinear extension of g to T
C
M, also denoted g
or '· , · `, is symmetric and satisfies the following three conditions:
'Z
1
, Z
2
` = 'Z
1
, Z
2
`;
'Z
1
, Z
2
` = 0 for Z
1
, Z
2
in T

M; (2.34)
'Z, Z` > 0 unless Z = 0.
Vice versa, a symmetric complex bilinear form '· , · ` on T
C
M satisfying these
three conditions is the complex bilinear extension of a Riemannian metric sat-
isfying (2.33). Then (2.33) also holds for the complex linear extension of J to
T
C
M and complex vector fields X and Y .
2.35 Proposition. Let M be a complex manifold with complex structure J.
Then a Riemannian metric g = '· , · ` is compatible with J iff about each point
p
0
∈ M, there are holomorphic coordinates
z = (z
1
, . . . , z
m
) = (x
1
+iy
1
, . . . , x
m
+iy
m
)
such that the associated coordinate frame (X
1
, Y
1
, . . . , X
m
, Y
m
) satisfies
'X
j
, X
k
`(p
0
) = 'Y
j
, Y
k
`(p
0
) = δ
jk
and 'X
j
, Y
k
`(p
0
) = 0.
The proof of Proposition 2.35 is straightforward and left as an exercise. We
note the following immediate consequence.
6
The usual arguments give the existence of compatible Riemannian metrics.
24 Lectures on K¨ ahler Manifolds
2.36 Corollary. Let M be a Hermitian manifold. Then the type decomposition
A
r
(M, C) = ⊕
p+q=r
A
p,q
(M, C)
is orthogonal with respect to the induced Hermitian metric (ϕ, ψ) = 'ϕ, ψ`. In
particular, ∗ϕ has type (m−q, m−p) if ϕ has type (p, q).
Proof. The first assertion is clear from Proposition 2.35. The second assertion
follows from the first since the volume form has type (m, m)
7
.
2.4 Blowing Up Points. Consider the universal bundle U
1
C
m
over CP
m−1
.
The restriction of the map U
1
C
m
→ C
m
, (L, z) → z, to the open subset
{(L, z) ∈ U
1
C
m
| z = 0} is biholomorphic onto C
m
\ {0}. Thus we can think
of U
1
C
m
as C
m
, where the point 0 is replaced by the set {(L, 0) ∈ U
1
C
m
}.
We identify the latter with CP
m−1
and thus have blown up the point 0 of C
m
to CP
m−1
. A similar construction can be carried out for points in complex
manifolds.
2.37 Remark. For any r > 0, we can identify S
r
= {(L, z) ∈ U
1
C
m
| |z| = r}
with the sphere of radius r in C
m
. Then the projection π: S
r
→CP
m−1
turns
into the Hopf fibration: The fibers of π intersect S
r
in Hopf circles, that is,
the intersections of complex lines in C
m
with the sphere S
r
. Renormalizing the
given Riemannian metric on S
r
(of sectional curvature 1/r
2
) by adding the pull
back of any fixed Riemannian metric g on CP
m−1
, we obtain a Riemannian
metric g
r
on the sphere S
2m−1
. We can think of (S
2m−1
, g
r
) as a collapsing
family of Riemannian manifolds with limit (CP
m−1
, g), as r →0. The differen-
tial geometric significance of this kind of collapse was first recognized by Berger:
The sectional curvature of the collapsing spheres stays uniformly bounded as
r →0, see Example 3.35 in [CE], [Kar, page 221], and [CGr].
Let M be a complex manifold and p be a point in M. The blow up of M
at p replaces p by the space of complex lines in T
p
M. The precise construction
goes as follows. Let z = (z
1
, . . . , z
m
): U → U

be holomorphic coordinates
about p with z(p) = 0. In U

CP
m−1
, consider the set
V = {(z, [w]) | z
i
w
j
= z
j
w
i
for all i, j},
where points in CP
m−1
are given by their homogeneous coordinates, denoted
[w]. On the subset {w
j
= 0} of CP
m−1
, V is defined by the m−1 independent
equations
z
i
=
w
i
w
j
z
j
, i = j,
the other equations follow. Hence the system of equations defining V has
constant rank m−1, and hence V is a complex submanifold of U

CP
m−1
of
dimension m with
S = V ∩ {z = 0}

= CP
m−1
.
7
Recall that dz ∧ dz = −2idx ∧ dy.
2 Complex Manifolds 25
On V \ S, [w] is determined by z, hence
V \ S

= U \ {p}.
More precisely, the canonical map
V \ S →U

\ {0}, (z, [w]) →z,
is biholomorphic. We use it to glue V to M\{p} and obtain a complex manifold
˜
M, the blow up of M at p, together with a holomorphic map
π:
˜
M →M
such that π
−1
(p) = S and π:
˜
M \ S →M \ {p} is biholomorphic.
We consider V as an open subset of
˜
M. Choose ε > 0 such that the image
U

of the holomorphic coordinates z contains the ball of radius ε > 0 about 0
in C
m
. Then the map
V
ε
:= {(z, [w]) ∈ V | |z| < ε} →{(L, z) ∈ U
1
C
m
| |z| < ε},
(z, [w]) →([w], z),
is biholomorphic. Hence a neighborhood of S in
˜
M is biholomorphic to a
neighborhood of the zero-section of U
1
C
m
such that the map σ: V
ε
→CP
m−1
,
σ(z, [w]) = [w], corresponds to the projection. For any r ∈ (0, ε), the set
S
r
= {(z, [w]) ∈ V | |z| = r}
corresponds to the sphere of radius r in C
m
. The fibers of π intersect S
r
in Hopf
circles, that is, the intersections of complex lines in C
m
with S
r
. Thus we can
again think of π: S
r
→ CP
m−1
as the Hopf fibration and of the convergence
S
r
→ S, as r → 0, as the collapse of S
2m−1
to CP
m−1
along Hopf circles;
compare with Remark 2.37 above.
2.38 Exercises. Let
˜
M be the blow up of M at p as above.
1) Let ˆ z = f(z) be other holomorphic coordinates of M about p with ˆ z(p) =
0, and let
ˆ
M be the blow up of M at p with respect to the coordinates ˆ z. Write
f(z) =
¸
f
j
(z)z
j
, where the maps f
j
are holomorphic with f
j
(0) = ∂
j
f(0).
Show that
f(z, [w]) =

f(z),
h ¸
f
j
(z)w
j
i”
extends the identity on M\{p} to a biholomorphic map
˜
M →
ˆ
M. In this sense,
the blow up of M at p does not depend on the choice of centered holomorphic
coordinates.
2) Let
˜
f :
˜
M → N be a holomorphic map, where N is another complex
manifold. If f is constant on S, then there is a holomorphic map f : M → N
such that
˜
f = f ◦ π.
26 Lectures on K¨ ahler Manifolds
2.39 Remark. By replacing centered holomorphic coordinates about p by a
tubular neighborhood of the zero section in the normal bundle, there is a rather
immediate generalization of the blow up of points to a blow up of complex
submanifolds, see e.g. Section 4.6 in [GH] or Section 2.5 in [Hu].
2.40 Examples. 1) Kummer surface. In this example it will be convenient
to enumerate coordinates by subindices. Consider the quotient Q = C
2
/Z
2
,
where Z
2
= {1, −1} acts by scalar multiplication on C
2
. Let q
0
be the image
of 0 under the natural projection C
2
→ Q. Note that away from 0 and q
0
,
respectively, the projection is a twofold covering with holomorphic covering
transformations, turning Q\ {q
0
} into a complex manifold. Via
z
0
= t
1
t
2
, z
1
= t
2
1
, z
2
= t
2
2
,
we can identify Q with the algebraic hypersurface
H = {(z
0
, z
1
, z
2
) ∈ C
3
| z
2
0
= z
1
z
2
} ⊂ C
3
,
which has a singularity at the origin 0. We blow up C
3
at 0 to get
˜
C
3
= {(z, [w]) ∈ C
3
CP
2
| z
i
w
j
= z
j
w
i
}.
We have
π
−1
(H \ {0}) = {(z, [w]) ∈
˜
C
3
| z = 0, w
2
0
= w
1
w
2
}.
In particular, the closure of π
−1
(H \ {0}) in
˜
C
3
is the regular hypersurface
˜
H = {(z, [w]) ∈
˜
C
3
| w
2
0
= w
1
w
2
} ⊂
˜
C
3
.
Thus by blowing up 0 ∈ C
3
, we resolved the singularity of H.
Recall that C = {[w] ∈ CP
2
| w
2
0
= w
1
w
2
} is an embedded CP
1
, com-
pare Example 2.2.10. There is a natural projection
˜
H → C. We choose
(u
j
, z
j
) as holomorphic coordinates for
˜
H over the preimage of U
j
under this
projection. The coordinate transformation over the preimage of U
1
∩ U
2
is
(u
2
, z
2
) = (u
−1
1
, u
2
1
z
1
). The holomorphic cotangent bundle A
1,0
(C, C) has
du
j
as a nowhere vanishing section over U
j
, and du
2
= −u
−2
1
du
1
over U
1

U
2
. We conclude that the identification (u
1
, z
1
du
1
) ↔ (u
1
, z
1
) over U
1
and
(u
2
, −z
2
du
2
) ↔ (u
2
, z
2
) over U
2
establishes a biholomorphic map between
˜
H
and A
1,0
(C, C) = A
1,0
(CP
1
, C).
Let T
4
= Z
4
\C
2
. Scalar multiplication by Z
2
on C
2
descends to an action
of Z
2
on T
4
. This action has 16 fixed points, namely the points with integral
or half-integral coordinates in T
4
. At each fixed point x ∈ T
4
, the action is
locally of the form ±1 · (x + t) = x ± t as above. Thus we can resolve each
of the quotient singularities on T
4
/Z
2
by the above construction and obtain a
compact complex surface, the Kummer surface. For more information on the
Kummer surface we refer to [Jo, Section 7.3].
2 Complex Manifolds 27
2) Dependence of blow-up on points.
8
Let m ≥ 2 and T = Γ\C
m
be a com-
plex torus. Let (p
1
, . . . , p
k
) and (p

1
, . . . , p

l
) be two tuples of pairwise different
points in T. Let M and M

be the blow ups of T in the points p
1
, . . . , p
k
and
p

1
, . . . , p

l
, respectively, and let π: M →T and π

: M

→T be the projections.
Let S
i
and S

j
be the preimages of p
i
and p

j
under π and π

, respectively.
Let f : M → M

be a holomorphic map. Since S
i

= CP
m−1
is simply
connected, the restriction of π

◦ f to S
i
lifts to a holomorphic map S
i
→C
m
.
Now S
i
is a closed complex manifold, hence any such lift is constant. It follows
that π

◦ f maps S
i
to a point in T. In particular, there is a holomorphic map
g : T →T such that π

◦ f = g ◦ π.
Suppose now that f is biholomorphic. Then f is not constant on S
i
(we
assume m ≥ 2). By what we said above, it follows that f maps each S
i
biholomorphically to an S

j
. Thus k = l and, up to renumeration, f(S
i
) = S

i
for all i. Moreover, the induced map g : T →T is biholomorphic with g(p
i
) = p

i
for all i.
Let now k = l = 2, p
1
= p

1
= 0, p
2
= p, p

2
= p

. In Example 2.2.6 above
we showed that any biholomorphic map of T is of the form h(z) = Az +b with
b ∈ T and A ∈ Gl(m, C) such that A(Γ) = Γ. Hence the above g is of the form
g(z) = Az for some A ∈ Gl(m, C) with A(Γ) = Γ and Ap = p

(in T). On
the other hand, there are pairs of points p, p

∈ T \ {0} such that there is no
A ∈ Gl(m, C) with A(Γ) = Γ and Ap = p

. Then by what we just said, the
corresponding blow ups M and M

of T in 0, p and 0, p

, respectively, are not
biholomorphic.
2.41 Exercise. In terms of oriented smooth manifolds, the blow up
˜
M of M at
p ∈ M corresponds to the connected sum M#CP
m
, where CP
m
denotes CP
m
with orientation opposite to the standard one: Choose w/w
0
as a coordinate
about q = [1, 0, . . . , 0] in U
0
= {[w
0
, w] | w
0
= 0} ⊂ CP
m
. Let ε > 0 and
V
ε

˜
M be as above. Set
U

ε
= {[w
0
, w] ∈ CP
m
| |w
0
| < ε|w|}.
Use coordinates z about p as in the definition of blow ups and w/w
0
as above
to define the connected sum M#CP
m
. Then the map f : M#CP
m

˜
M,
f(p

) =

p

if p

∈ M \ {p},
(w
0
w/|w|
2
, [w]) ∈ V
ε
if p

= [w
0
, w] ∈ U

ε
,
is well defined and an orientation preserving diffeomorphism. In particular, as
a smooth manifold, the blow up
˜
M does not depend on the choice of p ∈ M.
We conclude our discussion of blow ups with a fact on the automorphism
group of a compact complex surface which we cannot prove in the framework
8
I owe this example to Daniel Huybrechts.
28 Lectures on K¨ ahler Manifolds
of these lecture notes, but which will provide us with an important example in
Subsection 7.4. For a closed complex manifold M with complex structure J,
denote by Aut(M) = Aut(M, J) the group of biholomorphic transformations of
M, endowed with the compact-open topology, and by Aut
0
(M) the component
of the identity in Aut(M).
2.42 Proposition. Let M be a compact complex surface and π: M
p
→ M
be the blow up of M in p ∈ M. Then automorphisms in Aut
0
(M
p
) leave the
exceptional divisor S = π
−1
(p) invariant and, via restriction to M \ S,
Aut
0
(M
p
)

= {Φ ∈ Aut
0
(M) | Φ(p) = p} =: Aut
0
(M, p).
The action of Φ ∈ Aut
0
(M, p) on S

= P(T
p
M) is induced by dΦ
p
.
The only issue is to show that an automorphism of M
p
in Aut
0
(M
p
) leaves
S invariant. Given some facts about topological intersection properties of ana-
lytic cycles, the proof of this is actually quite simple and geometric, see [GH],
Chapter 4.1. A corresponding statement would be wrong for the full automor-
phism group, since if M itself is already a blow up, there may be automorphisms
permuting the various exceptional divisors.
2.43 Example. Consider the blow up of CP
2
in one or two points, which we
choose to be [1, 0, 0] and [1, 0, 0], [0, 1, 0], respectively. By Proposition 2.42, the
respective automorphism groups are

¸
1 ∗ ∗
0 ∗ ∗
0 ∗ ∗
¸

∈ Gl(3, C)

and

¸
1 0 ∗
0 ∗ ∗
0 0 ∗
¸

∈ Gl(3, C)

. (2.44)
For a blow up of CP
2
in three points, the automorphism group, and hence
the complex structure, clearly depends on the choice of points, i.e., on whether
they are in general position or not.
3 Holomorphic Vector Bundles
Let E → M be a holomorphic vector bundle. For U ⊂ M open, we denote by
O(U, E) the space of holomorphic sections of E over U, a module over the ring
O(U). If (Φ
1
, . . . , Φ
k
) is a holomorphic frame of E over U, then
O(U)
k
÷ ϕ →ϕ
µ
Φ
µ
∈ O(U, E) (3.1)
is an isomorphism.
3.1 Dolbeault Cohomology. We now consider differential forms on M with
values in E. Since M is complex, we can distinguish forms according to their
type as before,
A
r
(M, E) =
¸
p+q=r
A
p,q
(M, E), (3.2)
where A
p,q
(M, E) = A
p,q
(M, C)⊗E. With respect to a local holomorphic frame
Φ = (Φ
j
) of E, a differential form of type (p, q) with values in E is a linear
combination
α = ϕ
j
⊗Φ
j
, (3.3)
where the coefficients ϕ
j
are complex valued differential forms of type (p, q).
The space of such differential forms is denoted A
p,q
(M, E). We define the
∂-operator on differential forms with values in E by
∂α := (∂ϕ
j
) ⊗Φ
j
. (3.4)
Since the transition maps between holomorphic frames of E are holomorphic,
it follows that ∂α is well defined. By definition, ∂α is of type (p, q + 1). We
have ∂∂ = 0 and hence
· · ·

−→ A
p,q−1
(M, E)

−→ A
p,q
(M, E)

−→ A
p,q+1
(M, E)

−→ · · · (3.5)
is a cochain complex. The cohomology of this complex is denoted H
p,∗
(M, E)
and called Dolbeault cohomology of M with coefficients in E. The dimensions
h
p,q
(M, E) := dim
C
H
p,q
(M, E) (3.6)
are called Hodge numbers of M with respect to E.
3.7 Remark. Let Ω
p
(E) be the sheaf of germs of holomorphic differential
forms with values in E and of degree p (that is, of type (p, 0)). Let A
p,q
(E)
be the sheaf of germs of differential forms with values in E and of type (p, q).
Then
0 →Ω
p
(E) →A
p,0
(E)

−→ A
p,1
(E)

−→ A
p,2
(E)

−→ · · ·
is a fine resolution of Ω
p
(E). Hence H
q
(M, Ω
p
(E)) is isomorphic to H
p,q
(M, E).
30 Lectures on K¨ ahler Manifolds
We are going to use the notation and results from Chapter 1. We assume
that M is endowed with a compatible metric in the sense of (2.33) and that
E is endowed with a Hermitian metric. Then the splitting of forms into types
as in (3.2) is perpendicular with respect to the induced Hermitian metric on
A

(M, E).
Suppose that α and β are differential forms with values in E and of type
(p, q) and (p, q − 1), respectively. By Corollary 2.36, (∗ ⊗ h)α is of type (m−
p, m − q). Therefore ((∗ ⊗ h)α) ∧
ε
β is a complex valued differential form of
type (m, m−1). It follows that

((∗ ⊗h)α) ∧
ε
β

= d

((∗ ⊗h)α) ∧
ε
β

.
Since the real dimension of M is 2m, we have (∗ ⊗h

)(∗ ⊗h) = (−1)
r
on forms
of degree r = p +q. Computing as in (1.46), we get
d

((∗ ⊗h)α) ∧
ε
β

= ∂

((∗ ⊗h)α) ∧
ε
β

= (−1)
2m−r
{−((∗ ⊗h

)∂(∗ ⊗h)α, β) + (α, ∂β)} vol.
It follows that


= (∗ ⊗h

)∂(∗ ⊗h). (3.8)
Here the ∂-operator on the right belongs to the dual bundle E

of E. Let


= ∂∂

+ ∂

∂ (3.9)
be the Laplace operator associated to ∂. We claim that


(∗ ⊗h) = (∗ ⊗h) ∆

, (3.10)
where the Laplace operator on the left belongs to E

. In fact,


(∗ ⊗h) = ∂(∗ ⊗h)∂(∗ ⊗h

)(∗ ⊗h) + (∗ ⊗h)∂(∗ ⊗h

)∂(∗ ⊗h)
= (∗ ⊗h)∂(∗ ⊗h

)∂(∗ ⊗h) + (∗ ⊗h)(∗ ⊗h

)∂(∗ ⊗h)∂
= (∗ ⊗h) ∆

.
We denote by H
p,q
(M, E) the space of ∆

-harmonic forms of type (p, q) with
coefficients in E. By (3.10), ∗ ⊗h restricts to a conjugate linear isomorphism
H
p,q
(M, E) →H
m−p,m−q
(M, E

). (3.11)
In the rest of this subsection, suppose that M is closed. Then by Hodge theory,
the canonical projection H
p,q
(M, E) →H
p,q
(M, E) is an isomorphism of vector
spaces. From (3.11) we infer Serre duality, namely that ∗⊗h induces a conjugate
linear isomorphism
H
p,q
(M, E) →H
m−p,m−q
(M, E

). (3.12)
In particular, we have h
p,q
(M, E) = h
m−p,m−q
(M, E

) for all p and q.
3.13 Remark. Equivalently we can say that ∗ ⊗h induces a conjugate linear
isomorphism H
q
(M, Ω
p
(E))

= H
m−q
(M, Ω
m−p
(E

)), see Remark 3.7.
3 Holomorphic Vector Bundles 31
3.2 Chern Connection. Let D be a connection on E and σ a smooth section
of E. Then we can split
Dσ = D

σ +D

σ (3.14)
with D

σ ∈ A
1,0
(M, E) and D

σ ∈ A
0,1
(M, E). Let (Φ
1
, . . . , Φ
k
) be a holo-
morphic frame of E over an open subset U ⊂ M. Then

µ
= θ
ν
µ
Φ
ν
, (3.15)
where θ = (θ
ν
µ
) is the connection form and θ
ν
µ
Φ
ν
is shorthand for θ
ν
µ
⊗Φ
ν
.
3.16 Lemma. D

σ = 0 for all local holomorphic sections σ of E iff D

= ∂.
Then the connection form θ ∈ A
1,0
(U, C
k×k
).
Proof. The assertion follows from comparing types in
D

σ +D

σ = Dσ = dϕ
µ
Φ
µ
+ ϕ
µ

µ
= ((∂ + ∂)ϕ
µ
+ ϕ
ν
θ
µ
ν

µ
,
where σ = ϕ
µ
Φ
µ
in terms of a local holomorphic frame (Φ
µ
) of E.
3.17 Lemma. Let h = (· , · ) be a smooth Hermitian metric on E. If D

= ∂,
then D is Hermitian iff
(σ, D

X
τ) = ∂(σ, τ)(X) or, equivalently, (D

X
σ, τ) = ∂(σ, τ)(X)
for all vector fields X on M and local holomorphic sections σ, τ of E.
Proof. In the sense of 1-forms we write (σ, D

τ) = ∂(σ, τ) for the equality in
the lemma. If this equality holds, then
d(σ, τ) = ∂(σ, τ) + ∂(σ, τ)
= (D

σ, τ) + (σ, D

τ)
= (Dσ, τ) + (σ, Dτ),
where we use D

σ = D

τ = 0. Since E has local holomorphic frames, we
conclude that the equality in the lemma implies that D is Hermitian. The
other direction is similar.
3.18 Theorem. Let E →M be a holomorphic vector bundle and h = (· , · ) be
a smooth Hermitian metric on E. Then there is precisely one connection D on
E such that 1) D is Hermitian and 2) D

= ∂.
Proof. Let (Φ
1
, . . . , Φ
k
) be a holomorphic frame of E over an open subset U ⊂
M and θ = (θ
ν
µ
) be the corresponding connection form as in (3.15). Now 2)
implies that θ is of type (1, 0). By Property 1),
dh
µν
= dh(Φ
µ
, Φ
ν
) = h(θ
λ
µ
Φ
λ
, Φ
ν
) +h(Φ
µ
, θ
λ
ν
Φ
λ
)
= θ
λ
µ
h
λν
+h
µλ
θ
λ
ν
= h
νλ
θ
λ
µ
+h
µλ
θ
λ
ν
.
32 Lectures on K¨ ahler Manifolds
Now dh
µν
= ∂h
µν
+ ∂h
µν
. Since θ is of type (1, 0), we conclude
∂h
µν
= h
µλ
θ
λ
ν
or θ = h
−1
∂h, (3.19)
by comparison of types. This shows that Properties 1) and 2) determine D
uniquely. Existence follows from the fact that the local connection forms θ =
h
−1
∂h above transform correctly under changes of frames. Another argument
for the existence is that for local holomorphic sections σ and τ of E, ∂(σ, τ)
is conjugate O-linear in σ so that the equation in Lemma 3.17 leads to the
determination of the yet undetermined D

.
The unique connection D satisfying the properties in Theorem 3.18 will be
called the Chern connection. It depends on the choice of a Hermitian metric
on E.
3.20 Exercises. 1) Let E → M be a holomorphic vector bundle and h be
a Hermitian metric on E. Let f : N → M be a holomorphic map. Then
the Chern connection of the pull back Hermitian metric f

h on the pull back
f

E →N is the pull back of the Chern connection on E with respect to h.
2) Let E →M be a holomorphic vector bundle and E

→M be a holomor-
phic vector subbundle of E. Let h be a Hermitian metric on E and h

be the
restriction of h to E

. For a section σ of E write σ = σ

+ σ

, where σ

is a
section of E

and σ

is perpendicular to E

. Let D be the Chern connection
on E with respect to h. Then the Chern connection D

of E

with respect to
h

is given by D

σ = (Dσ)

.
Recall that this is the standard recipe of getting a Hermitian connection for
a subbundle of a Hermitian bundle with a Hermitian connection.
In what follows, we use the wedge product as defined in (1.15) and the
discussion in Example 1.16, applied to the trivial bundle E = M C
k
.
3.21 Proposition. Let E →M be a holomorphic vector bundle with Hermitian
metric h and corresponding Chern connection D. Let (Φ
1
, . . . , Φ
k
) be a local
holomorphic frame of E and θ and Θ be the corresponding connection and
curvature form. Then:
1) θ = h
−1
∂h with h = (h
µν
);
2) D

= ∂ + θ and D

= ∂;
3) θ is of type (1, 0) and ∂θ = −θ ∧
µ
θ;
4) Θ = ∂θ and Θ is of type (1, 1);
5) ∂Θ = 0 and ∂Θ = Θ∧
λ
θ = Θ∧
µ
θ −θ ∧
µ
Θ.
3 Holomorphic Vector Bundles 33
In the case of a holomorphic line bundle E →M, a local nowhere vanishing
holomorphic section Φ, and h = (Φ, Φ), Proposition 3.21 gives
θ = ∂ ln h and Θ = ∂∂ ln h. (3.22)
We will use these formulas without further reference.
Proof of Proposition 3.21. The first assertion follows from the proof of Theo-
rem 3.18, the second from Lemma 3.16. As for the third assertion, we have
∂θ = ∂(h
−1
∂h) = −(h
−1
∂hh
−1
) ∧
µ
∂h
= −(h
−1
∂h) ∧
µ
(h
−1
∂h) = −θ ∧
µ
θ.
In particular,
Θ = ∂θ + ∂θ + θ ∧
µ
θ = ∂θ.
Hence ∂Θ = 0 and, by the Bianchi identity dΘ = Θ∧
λ
θ, we conclude that
Θ∧
λ
θ = dΘ = ∂Θ+ ∂Θ = ∂Θ.
3.23 Proposition. Let E →M be a holomorphic vector bundle with Hermitian
metric h and Chern connection D. Let p
0
∈ M and z be holomorphic coordi-
nates about p
0
with z(p
0
) = 0. Then there is a holomorphic frame (Φ
1
, . . . , Φ
k
)
of E about p
0
such that
1) h(z) = 1 +O(|z|
2
);
2) Θ(0) = ∂∂h(0).
Proof. Suppose 1) holds. Then
Θ(0) = ∂θ(0) =

(∂h
−1
) ∧
µ
∂h +h
−1
∂∂h

(0) = ∂∂h(0),
hence 1) implies 2). To show 1), we choose a holomorphic frame (Φ
1
, . . . , Φ
k
)
about p
0
such that h
µν
(0) = δ
µν
. We define a new holomorphic frame
˜
Φ
µ
= Φ
µ
+z
i
a
ν
µi
Φ
ν
with a
ν
µi
= −(∂h
νµ
/∂z
i
)(0). It is easy to check that
˜
h = 1 +O(|z|
2
).
3.24 Example (Tautological bundle). Let M = G
r,n
:= G
r
C
n
and E → M
be the tautological bundle as in Example 2.2.5. The standard Hermitian inner
product on C
n
induces a Hermitian metric on E.
We identify sections of E over an open subset U of M with maps σ: U →C
n
such that σ(p) ∈ p for all p ∈ U. Then a holomorphic frame over an open subset
34 Lectures on K¨ ahler Manifolds
U ⊂ M is given by holomorphic mappings Φ
µ
: U → C
n
, 1 ≤ µ ≤ r, such that

1
(p), . . . , Φ
r
(p)) is a basis of p, for all p ∈ U. For such a frame,
h
µν
= (Φ
µ
, Φ
ν
) =
¸
λ
¯
Φ
λ
µ
Φ
λ
ν
= (
¯
Φ
t
Φ)
µ
ν
with Φ = (Φ
ν
µ
).
If σ: U → C
m
is a section of E and X is a vector field of M over U, then
the derivative dσ(X) of σ in the direction of X need not be a section of E;
that is, dσ
p
(X
p
) need not be an element of p anymore. We obtain the Chern
connection D of E by setting
(D
X
σ)(p) = π
p
(dσ
p
(X
p
)),
where π
p
: C
n
→C
n
is the orthogonal projection of C
n
onto p, compare Exer-
cise 3.20.2 (where the ambient bundle is M C
n
in our case).
Consider the plane p
0
spanned by the first r unit vectors, in homogeneous
coordinates in C
n×r
written as
p
0
=
¸
1
0

∈ M,
where 1 = 1
r
is the r r unit matrix. Then we get a holomorphic parameteri-
zation (the inverse of holomorphic coordinates) of M about p
0
by
C
(n−r)×r
÷ z →
¸
1
z

∈ M.
Moreover, the r columns of the matrix (
1
z
) are a holomorphic frame of E. For
this frame we have
h
µν
= δ
µν
+
¸
λ
z
λ
µ
z
λ
ν
= δ
µν
+ (z
t
z)
µ
ν
.
In particular,
h = 1 +O(|z|
2
).
In z = 0, that is, in p
0
, we have
Θ(0) = ∂∂h(0) = ∂z
t
∧ ∂z = dz
t
∧ dz.
The canonical action of the group U(n) on M is transitive and extends canoni-
cally to an action on E which leaves h invariant. Each element from U(n) acts
biholomorphically on M and E. In particular, choosing p
0
as above means no
restriction of generality.
3 Holomorphic Vector Bundles 35
3.3 Some Formulas. For the following, we refer again to notation and results
introduced in Chapter 1. Let E → M be a holomorphic vector bundle over
M. Let h be a Hermitian metric on E and D be the corresponding Chern
connection. Then we have the associated exterior differential
d
D
: A
r
(M, E) →A
r+1
(M, E), (3.25)
see (1.18) and (1.21). Let (Φ
µ
) be a local holomorphic frame of E and
α = ϕ
µ
Φ
µ
(3.26)
be a differential form with values in E of type (p, q) with p + q = r. By the
characteristic property of the Chern connection, we have
d
D
α = dϕ
µ
⊗Φ
µ
+ (−1)
r
ϕ
µ
∧ DΦ
µ
= ∂
D
α + ∂α, (3.27)
where

D
α := ∂ϕ
µ
⊗Φ
µ
+ (−1)
r
ϕ
µ
∧ DΦ
µ
. (3.28)
Note that ∂
D
α and ∂α are of type (p + 1, q) and (p, q + 1), respectively.
3.29 Proposition. Let E be a holomorphic vector bundle over M with Her-
mitian metric and associated Chern connection D. Then, for any differential
form α with values in E,
(∂
D
)
2
α = (∂)
2
α = 0 and (∂
D
∂ + ∂∂
D
)α = (d
D
)
2
α = R
D

ε
α,
where ε: End(E) ⊗E →E is the evaluation map.
Proof. If α is a differential form with values in E of type (p, q), then d
D
d
D
α =
R
D

ε
α, see Proposition 1.19. By Proposition 3.21, R
D
is of type (1, 1), hence
d
D
d
D
α is of type (p + 1, q + 1). By definition,
d
D
d
D
α = ∂
D

D
α + ∂∂α + (∂
D
∂ + ∂∂
D
)α.
The first two forms on the right hand side vanish since they have type (p+2, q)
and (p, q + 2), respectively.
We assume now that, in addition, M is endowed with a compatible Rie-
mannian metric as in (2.33). Let (X
1
, Y
1
, . . . , X
m
, Y
m
) be a local orthonormal
frame of M with JX
j
= Y
j
. Set
Z
j
=
1
2
(X
j
−iY
j
) and Z
j
=
1
2
(X
j
+iY
j
), (3.30)
and let Z

1
, . . . , Z

n
, Z

1
, . . . , Z

n
be the corresponding dual frame of T

C
M. With
our conventions, we have
Z

j
:= 'Z
j
, · ` =
1
2
Z

j
and Z

j
:= 'Z
j
, · ` =
1
2
Z

j
. (3.31)
36 Lectures on K¨ ahler Manifolds
Since X
j
= Z
j
+Z
j
and Y
j
= i(Z
j
−Z
j
), we get
X

j
=
1
2
(Z

j
+Z

j
) and Y

j
= −
i
2
(Z

j
−Z

j
)
for the dual frame (X

1
, Y

1
, . . . , X

m
, Y

m
) of (X
1
, Y
1
, . . . , X
m
, Y
m
). Therefore,
d
D
=
1
2
¸
¸
(Z

j
+Z

j
) ∧
ˆ
D
Zj+Zj
+ (Z

j
−Z

j
) ∧
ˆ
D
Zj−Zj
¸
=
¸
¸
Z

j

ˆ
D
Zj
+Z

j

ˆ
D
Zj
¸
= ∂
D
+ ∂,
(3.32)
meaning that the sum of the first terms is equal to ∂
D
and the sum of the second
to ∂. To compute the adjoint operator (d
D
)

of d
D
with respect to the induced
Hermitian metric as in (1.25), we note first that the relations in Exercise 1.26
also hold for complex tangent vectors. Since Z

j
= 2Z

j
and Z

j
= 2Z

j
, we get
(d
D
)

= −2
¸
¸
Z
j

ˆ
D
Zj
+Z
j

ˆ
D
Zj
¸
= (∂
D
)

+ ∂

, (3.33)
where we use the first relation from Exercise 1.26 for the first equality and
comparison of types for the second. The Laplace operators associated to ∂ and

D
are defined by


= ∂∂

+ ∂

∂ and ∆

D = ∂
D
(∂
D
)

+ (∂
D
)


D
, (3.34)
respectively. Both preserve the type of forms. Using the second relation from
Exercise 1.26, a straightforward computation gives the following Weitzenb¨ock
formulas,


α = −2
¸
j
ˆ
D
2
α(Z
j
, Z
j
) + 2
¸
j,k
Z

j
∧ (Z
k
(
ˆ
R
D
(Z
k
, Z
j
)α))
= −2
¸
j
ˆ
D
2
α(Z
j
, Z
j
) + 2
¸
j,k
Z
j
(Z

k
∧ (
ˆ
R
D
(Z
k
, Z
j
)α)),
(3.35)


Dα = −2
¸
j
ˆ
D
2
α(Z
j
, Z
j
) + 2
¸
j,k
Z
j
(Z

k
∧ (
ˆ
R
D
(Z
k
, Z
j
)α))
= −2
¸
j
ˆ
D
2
α(Z
j
, Z
j
) + 2
¸
j,k
Z

j
∧ (Z
k
(
ˆ
R
D
(Z
k
, Z
j
)α)),
(3.36)
where the frame (Z
1
, Z
1
, . . . , Z
m
, Z
m
) is as in (3.30) and where
ˆ
R
D
denotes the
curvature tensor of
ˆ
D on A

(M, E).
3.37 Exercises. 1) Show that, for a differential form ϕ ⊗σ of degree r,

D
(ϕ ⊗σ) = ∂ϕ ⊗σ + (−1)
r
ϕ ∧ D

σ,
∂(ϕ ⊗σ) = ∂ϕ ⊗σ + (−1)
r
ϕ ∧ D

σ.
3 Holomorphic Vector Bundles 37
2) Prove Formulas 3.35 and 3.36 and show that
4
¸
ˆ
D
2
α(Z
j
, Z
j
) = tr
ˆ
D
2
α +i
¸
ˆ
R
D
(X
j
, JX
j
)α,
4
¸
ˆ
D
2
α(Z
j
, Z
j
) = tr
ˆ
D
2
α +i
¸
ˆ
R
D
(JX
j
, X
j
)α.
3.4 Holomorphic Line Bundles. The results in this subsection will be
mainly used in Section 9. We continue to assume that M is a complex manifold.
Let E →M be a complex line bundle. Let (U
α
) be an open covering of M such
that, for each α, there is a nowhere vanishing smooth section Φ
α
: U
α
→ E.
If σ is a smooth section of E, then we can write σ = σ
α
Φ
α
over U
α
, where
σ
α
: U
α
→C is a smooth function, the principal part of σ with respect to Φ
α
.
Over intersections U
α
∩ U
β
, we have Φ
β
= t
αβ
Φ
α
, where the transition
functions t
αβ
: U
α
∩ U
β
→C

= C \ {0} are smooth. For the principal parts of
a section σ as above we get
σ
α
= t
αβ
σ
β
. (3.38)
The family (t
αβ
) of transition functions satisfies
t
αα
= t
αβ
t
βα
= t
αβ
t
βγ
t
γα
= 1. (3.39)
We say that with respect to the given covering, (t
αβ
) is a 1-cocycle of smooth
functions with values in C

.
Suppose E

is another complex line bundle over M, and suppose there is an
isomorphism, F : E →E

. After passing to a common refinement if necessary,
suppose that we are given, for each α, a nowhere vanishing smooth section
Φ

α
: U
α
→ E

. As above, we have smooth transition functions (t

αβ
) with
Φ

β
= t

αβ
Φ

α
. Over each U
α
, we also have F(Φ
α
) = s
α
Φ

α
, where s
α
: U
α
→C

is smooth. Comparing coefficients we get
t

αβ
= s
α
t
αβ
s
−1
β
. (3.40)
By definition, this means that the two cocycles (t
αβ
) and (t

αβ
) are cohomolo-
gous.
Vice versa, suppose (U
α
) is an open covering of M and (t
αβ
) a C

-valued 1-
cocycle of smooth functions with respect to (U
α
). On the set of triples (α, p, v)
with p ∈ U
α
and v ∈ C set (α, p, v) ∼ (β, q, w) iff p = q and v = t
αβ
w. By
(3.39), ∼ is an equivalence relation. The set E of equivalence classes [α, p, v]
admits a natural projection to M, π([α, p, v]) = p. The fibers π
−1
(p), p ∈ M,
are complex lines. As in the case of the universal bundles over complex Grass-
mannians, it follows that there is a unique topology on E such that π: E →M
is a smooth complex line bundle with nowhere vanishing smooth sections
Φ
α
: U
α
→E, Φ
α
(p) = [α, p, 1]. (3.41)
38 Lectures on K¨ ahler Manifolds
If another cocycle t

αβ
is cohomologous to the given one, t

αβ
= s
α
t
αβ
s
−1
β
for a
family of smooth functions s
α
: U
α
→C

, then F(Φ
α
) := s
α
Φ

α
gives rise to an
isomorphism F : E →E

.
In the language of sheaves, we have established that the space of isomor-
phism classes of smooth complex line bundles over M is naturally isomorphic
to H
1
(M, E

), where E

is the sheaf over M of germs of smooth functions with
values in C

.
Replacing the word ‘smooth’ in the above discussion by the word ‘holomor-
phic’, we obtain the corresponding results for holomorphic line bundles. In
particular, the space of isomorphism classes of holomorphic line bundles over
M is naturally isomorphic to the Picard group Pic(M) := H
1
(M, O

), where
O

is the sheaf of germs of holomorphic functions with values in C

and where
the group law in Pic(M) corresponds to the tensor product of line bundles.
Note that the tensor product L ⊗ L

of a holomorphic line bundle L over M
with its dual bundle L

is holomorphically isomorphic to the trivial bundle so
that the dual bundle corresponds to the inversein Pic(M).
There is an important connection between complex hypersurfaces and holo-
morphic line bundles. Suppose H is a complex hypersurface. Let (U
α
) be an
open covering of M with defining holomorphic functions f
α
: U
α
→C for H,
H ∩ U
α
= {p ∈ U
α
| f
α
(p) = 0}, (3.42)
such that df
α
(p) = 0 for all p ∈ H ∩ U
α
. Then we have
f
α
= t
αβ
f
β
(3.43)
over intersections U
α
∩U
β
, where the functions t
αβ
: U
α
∩U
β
→C

are holomor-
phic. The family (t
αβ
) is a 1-cocycle of holomorphic functions and, therefore,
defines a holomorphic line bundle E as above such that the sections Φ
α
from
(3.41) are holomorphic and such that the section f of E with f|U
α
= f
α
Φ
α
is well-defined and holomorphic. In this way we associate to H a holomorphic
line bundle E together with a holomorphic section f such that H = {p ∈ M |
f(p) = 0}.
3.44 Remark. More generally, there are one-to-one correspondences between
so-called effective divisors and pairs consisting of a line bundle together with
a holomorphic section (up to isomorphism) respectively divisors and pairs of a
line bundle together with a meromorphic section (up to isomorphism), see for
example Chapitre V in [Wei], Section 1.1 in [GH], or Section 2.3, Proposition
4.4.13, and Corollary 5.3.7 in [Hu].
3.45 Examples. We consider some line bundles over CP
m
. We will use the
open covering of CP
m
by the sets U
i
= {[z] ∈ CP
m
| z
i
= 0}, 0 ≤ i ≤ m.
1) The tautological line bundle U = U
1,m
→ CP
m
, compare also Exam-
ples 2.2.5 and 3.24: Over U
i
,
Φ
i
=

z
0
z
i
, . . . ,
z
i
z
i
, . . . ,
z
m
z
i

3 Holomorphic Vector Bundles 39
is a nowhere vanishing holomorphic section of U over U
i
. Over intersections
U
i
∩ U
j
, we have z
i
Φ
i
= z
j
Φ
j
.
Over U
i
, a section σ of U is of the form σ = σ
i
Φ
i
, where σ
i
is a function
called the principal part of σ with respect to Φ
i
. Over intersections U
i
∩ U
j
,
we have
σ
i
=
z
i
z
j
σ
j
.
2) The canonical line bundle K = A
m,0
(CP
m
, C) → CP
m
: Over U
i
as
above, let w
ij
= z
j
/z
i
and set
ω
i
= (−1)
i
dw
i0
∧ · · · ∧
¯
dw
ii
∧ · · · ∧ dw
im
,
where the hat indicates that the corresponding term is to be deleted. Then ω
i
is a nowhere vanishing holomorphic section of K over U
i
. Over U
i
∩ U
j
,
w
ik
=
z
k
z
j
·
z
j
z
i
=
z
j
z
i
· w
jk
and d
z
j
z
i
= −
z
2
j
z
2
i
d
z
i
z
j
.
Therefore
z
m+1
i
ω
i
= z
m+1
j
ω
j
,
and hence K is isomorphic to U
m+1
, the (m+1)-fold tensor product U⊗· · ·⊗U.
3) A hyperplane in CP
m
is determined by an m-dimensional vector subspace
V ⊂ C
m+1
. We choose the hyperplane H = {[z] ∈ CP
m
| z
0
= 0}. Over U
i
as
above, H is defined by the equation
f
i
([z]) :=
z
0
z
i
= 0.
Over U
i
∩ U
j
,
f
i
=
z
j
z
i
f
j
.
For the holomorphic line bundle over CP
m
associated to H, also denoted by
H, the functions f
i
serve as principal parts of a holomorphic section f of H
which vanishes along the given hyperplane. It follows that in our case, H is
inverse to U, that is, H ⊗U is the trivial bundle.
4) Let M be a complex manifold of complex dimension m and
˜
M be the
blow up of M at a point p ∈ M as in Subsection 2.4. We use the notation
introduced there and describe the holomorphic line bundle L →
˜
M determined
by the hypersurface S = π
−1
(p). To that end, we consider the open covering
of
˜
M consisting of the subsets
W
0
=
˜
M \ S, W
i
= {(z, [w]) ∈ V | w
i
= 0}, 1 ≤ i ≤ m,
of
˜
M and corresponding defining holomorphic functions f
i
: W
i
→C of S,
f
0
= 1 and f
i
(z, [w]) = z
i
, 1 ≤ i ≤ m.
40 Lectures on K¨ ahler Manifolds
These functions transform under the rules
z
j
f
0
= f
j
and z
j
f
i
= z
i
f
j
.
The functions f
i
define a global holomorphic section f of L with S as its set of
zeros. We have f = f
i
Φ
i
, where Φ
i
is a nowhere vanishing holomorphic section
of L over W
i
as in (3.41).
Let M be a complex manifold of complex dimension m and
˜
M be the blow
up of M at a point p ∈ M as in Subsection 2.4. Let L →
˜
M be the holomorphic
line bundle as in Example 3.45.4. In the notation of Subsection 2.4, consider
the map σ: V →CP
m−1
, σ(z, [w]) = [w]. Let H →CP
m−1
be the hyperplane
bundle, see Example 3.45.3.
3.46 Lemma. Let L

be the dual of L. Then
L

|V

= σ

H.
Proof. Realize H as the line bundle over CP
m−1
associated to the hyperplane
{w
1
= 0}. Then the defining equations of H over W
i
are g
i
= w
1
/w
i
= 0.
The functions g
i
transform according to the rule w
i
g
i
= w
j
g
j
, or, equivalently,
z
i
g
i
= z
j
g
j
. Hence σ

H is inverse to L|V .
3.47 Lemma. The canonical bundles K = A
m,0
(M, C) of M and
˜
K = A
m,0
(
˜
M, C)
of
˜
M are related by
˜
K

= π

K ⊗L
m−1
.
Proof. For W
i
, i ≥ 1, as above, let u
j
= w
j
/w
i
= z
j
/z
i
. Then
(u
1
, . . . , u
i−1
, z
i
, u
i+1
, . . . , u
m
)
are holomorphic coordinates on W
i
. Hence
Ψ
i
= du
1
∧ · · · ∧ du
i−1
∧ dz
i
∧ du
i+1
∧ · · · ∧ du
m
is a nowhere vanishing section of
˜
K over W
i
. Since dz
j
= u
j
dz
i
+z
i
du
j
,
Ψ
i
= z
1−m
i
dz
1
∧ · · · ∧ dz
m
.
Hence we have
z
m−1
i
Ψ
i
= z
m−1
j
Ψ
j
over W
i
∩ W
j
. Similarly, Ψ
0
= dz
1
∧ · · · ∧ dz
m
vanishes nowhere over W
0
∩ W
j
and
Ψ
0
= z
m−1
j
Ψ
j
.
Over V , we view dz
1
∧ · · · ∧ dz
m
also as a nowhere vanishing section of π

K.
Now L is trivial over W
0
=
˜
M \ S with nowhere vanishing section Φ
0
as
above. Moreover, the differential of π induces an isomorphism, also denoted π

,
3 Holomorphic Vector Bundles 41
between K restricted to M \ {p} and
˜
K restricted to W
0
. Hence over W
0
, we
obtain an isomorphism as desired by sending π

σ to (σ◦π)⊗Φ
m−1
0
. Similarly, we
obtain an isomorphism over W
i
, i ≥ 1, by sending Ψ
i
to (dz
1
∧· · ·∧dz
m
)⊗Φ
m−1
i
.
By the choice of sections Ψ
i
, these isomorphisms agree on the intersections
W
i
∩ W
j
, hence define an isomorphism over
˜
M.
4 K¨ahler Manifolds
Let M be a complex manifold with complex structure J and compatible Rie-
mannian metric g = '· , · ` as in (2.33). The alternating 2-form
ω(X, Y ) := g(JX, Y ) (4.1)
is called the associated K¨ ahler form. We can retrieve g from ω,
g(X, Y ) = ω(X, JY ). (4.2)
We say that g is a K¨ ahler metric and that M (together with g) is a K¨ ahler
manifold if ω is closed
9
.
4.3 Remark. View TM together with J as a complex vector bundle over
M, and let h be a Hermitian metric on TM. Then g = Re h is a compatible
Riemannian metric on M and Imh is the associated K¨ ahler form:
g(JX, Y ) = Re h(JX, Y ) = Re h(iX, Y ) = Re(−ih(X, Y )) = Imh(X, Y ).
Vice versa, if g is a compatible Riemannian metric on M and ω is the associated
K¨ ahler form, then h = g +iω is a Hermitian metric on TM.
In terms of holomorphic coordinates z on M and the frames introduced in
(2.13), we have
g
jk
= 'Z
j
, Z
k
` = 0, g
jk
= 'Z
j
, Z
k
` = 0 (4.4)
and
g
jk
= 'Z
j
, Z
k
` = 'Z
k
, Z
j
` = g
kj
. (4.5)
We also have
g
jk
= g
jk
. (4.6)
With
dz
j
dz
k
:= dz
j
⊗dz
k
+dz
k
⊗dz
j
(4.7)
we finally get
g = g
jk
dz
j
dz
k
, ω = ig
jk
dz
j
∧ dz
k
. (4.8)
We call the matrix (g
jk
) the fundamental matrix of g (with respect to the given
coordinates).
9
K¨ahler’s original article [K¨ a1] contains much of what we say in this chapter. Compare
also Bourguignon’s essay in [K¨a2] or his review of K¨ahler’s article in [Bo2].
4 K¨ ahler Manifolds 43
4.9 Remark. In terms of holomorphic coordinates z as above and the usual
notation, we have
g
jk
=
1
4
'X
j
−iY
j
, X
k
+iY
k
`
=
1
2
¸
'X
j
, X
k
` −iω(X
j
, X
k
)
¸
=
1
2
h
jk
,
where the h
jk
are the coefficients of the fundamental matrix of h with respect
to the local complex frame (X
1
, . . . , X
m
) of TM.
4.10 Examples. 1) M = C
m
with the Euclidean metric g. Then g is a K¨ ahler
metric with
g =
1
2
¸
dz
j
dz
j
, ω =
i
2
¸
dz
j
∧ dz
j
. (4.11)
For any lattice Γ ⊂ C
m
, the induced metric on the complex torus Γ\C
m
is a
K¨ ahler metric.
2) Products of K¨ ahler manifolds (endowed with the product complex struc-
ture and the product metric) are K¨ ahler manifolds.
3) A complex submanifold N of a K¨ ahler manifold M is K¨ahlerian with
respect to the induced Riemannian metric since the K¨ ahler form on N is the
restriction of the K¨ ahler form of M. More generally, if N →M is a holomorphic
immersion, then N is K¨ ahlerian with respect to the induced Riemannian metric.
4) When endowed with the induced complex structure and Riemannian
metric, covering spaces of K¨ ahler manifolds and quotients of K¨ahler manifolds
by properly discontinuous and free group actions by holomorphic and isometric
transformations are again K¨ ahler manifolds.
5) The following construction yields the Fubini–Study metric in the case
of complex projective spaces. Let G
r,n
be the Grassmannian of r-planes in
C
n
, compare Examples 2.2.4, B.42, and B.83.1. Let M

⊂ C
n×r
be the open
subset of matrices of rank r and π: M

→G
r,n
be the canonical projection, a
holomorphic principal bundle with structure group Gl(r, C) acting on M

on
the right by matrix multiplication. Let Z be a holomorphic section of π over
an open subset U ⊂ G
r,n
. Define a closed form ω of type (1, 1) on U by
ω = i∂∂ ln det(Z
t
Z). (4.12)
We show first that ω does not depend on the choice of Z. In fact, any other
choice is of the form ZF, where F : U →Gl(r, C) is holomorphic, and then
∂∂ ln det(F
t
Z
t
ZF) = ∂∂ ln det(Z
t
Z)
since ln det F is holomorphic and ln det F
t
antiholomorphic. This shows that ω
is well defined independently of Z, and hence ω is a smooth form of type (1, 1)
on all of G
r,n
. The defining formula for ω shows that ω is closed.
44 Lectures on K¨ ahler Manifolds
In the case of complex projective space, that is, in the case r = 1, a straight-
forward computation gives an explicit formula for ω (and the Fubini–Study
metric). Let U = {[z
0
, z] ∈ CP
n−1
| z
0
= 0}. Then w = (1, z/z
0
) is a section
of π over U and hence
ω = i∂∂ ln det(1 + |w|
2
)
=
i
(1 + |w|
2
)
2

(1 + |w|
2
)
¸
j
dw
j
∧ dw
j

¸
j
w
j
dw
j

¸
j
w
j
dw
j

.
(4.13)
The case n = 2 of complex dimension 1 and the comparison with Formula
4.15 are instructive. In the case 1 < r < n − 1, explicit formulas are more
complicated.
The r-plane spanned by Z ∈ M

is denoted [Z], that is, [Z] = π(Z). We
recall the holomorphic action of Gl(n, C) on G
r,n
, see Example 2.2.4. For A
in Gl(n, C), we set λ
A
([Z]) = [AZ] and show next that λ

A
ω = ω for any A in
U(n). To that end we let Z be a holomorphic section of π over the open subset
U ⊂ G
r,n
. Then A
−1
Z ◦ λ
A
is a section of π over λ
−1
A
(U) and
λ

A
ω = iλ

A
(∂∂ ln det(Z
t
Z))
= i∂∂(λ

A
(ln det(Z
t
Z)))
= i∂∂(lndet((Z
t
A)(A
−1
Z)) ◦ λ
A
) = ω
since λ
A
is holomorphic and A is unitary.
We want to show now that g(X, Y ) = ω(X, JY ) is a K¨ahler metric on G
r,n
.
By our discussion above it remains to show that g is positive definite. Since
the action of U(n) is transitive on G
r,n
and g is invariant under this action, it
suffices to show this in the point p
0
, the plane spanned by the first r unit vectors
in C
n
. About p
0
we have holomorphic coordinates [
1
w
] →w ∈ C
(n−r)×r
, where
1 stands for the r r unit matrix. Hence
ω(0) = i∂∂ ln det(1 +w
t
w)|
w=0
= i∂ tr(((∂w
t
)w)(1 +w
t
w)
−1
)|
w=0
= −i tr ∂w
t
∧ ∂w
= −i tr dw
t
∧ dw = i
¸
j,k
dw
k
j
∧ dw
k
j
.
It follows that
g(0) =
¸
j,k
dw
k
j
dw
k
j
, (4.14)
and hence g is positive definite. In conclusion, g is a K¨ahler metric on G
r,n
with K¨ ahler form ω and the natural action of U(n) on G
r,n
preserves g and ω.
4 K¨ ahler Manifolds 45
6) This example is related to the previous one, compare again Examples B.42
and B.83.1. Let G

r,n
⊂ G
r,n
be the open subset of r-planes on which the
Hermitian form Q
r,n−r
as in (B.49) is negative definite. Write Z ∈ C
n×r
as
Z =

Z
0
Z
1

with Z
0
∈ C
r×r
and Z
1
∈ C
(n−r)×r
.
The columns of Z span an r-plane, [Z], in G

r,n
iff
Z
t
Q
r,n−r
Z = −Z
t
0
Z
0
+Z
t
1
Z
1
is negative definite, that is, iff Z
t
0
Z
0
> Z
t
1
Z
1
in the sense of Hermitian matrices.
Then Z
0
is invertible. Consider the open subset
M

= {Z ∈ C
n×r
| Z
t
0
Z
0
> Z
t
1
Z
1
} ⊂ C
n×r
and the canonical projection π: M

→ G

r,n
, a holomorphic principal bundle
with structure group Gl(r, C) acting on M

on the right by matrix multiplica-
tion. Define a closed form ω of type (1, 1) on G

r,n
by
ω = −i∂∂ ln det(Z
t
0
Z
0
−Z
t
1
Z
1
),
where Z is a local holomorphic section of π. As in the previous example we see
that ω is well defined and positive definite.
Let U(r, n − r) be the group of linear transformations of C
n
preserving
Q
r,n−r
. Then the natural action of U(r, n − r) on G
r,n
leaves G

r,n
invariant
and is transitive on G

r,n
. The stabilizer of the plane p
0
∈ G

r,n
spanned by the
first r unit vectors is U(r) U(n −r). As in the previous example we see that
ω is invariant under the action of U(r, n −r) and that g(X, Y ) = ω(X, JY ) is
a K¨ ahler metric on G

r,n
. The case r = 1 gives complex hyperbolic space.
The principal bundle π: M

→G

r,n
has a global holomorphic section since
Z
0
is invertible for all planes in G

r,n
: If Z is a local holomorphic section, then
W = ZZ
−1
0
does not depend on the choice of Z and W
0
= 1, the r r unit
matrix. The map
G

r,n
→C
(n−r)×r
, [Z] →w := W
1
,
is biholomorphic onto the bounded domain D = {w ∈ C
(n−r)×r
| w
t
w < 1}.
We leave it as an exercise to compute the induced action of U(r, n −r) on D.
In the case of complex hyperbolic space, that is, if r = 1, D is the unit ball
in C
n−1
, and the K¨ ahler form ω on D is given by
ω = −i∂∂(1 −|w|
2
)
=
i
(1 −|w|
2
)
2

(1 −|w|
2
)
¸
j
dw
j
∧ dw
j
+
¸
j,k
w
j
w
k
dw
j
∧ dw
k

.
(4.15)
46 Lectures on K¨ ahler Manifolds
For n = 2, we get the real hyperbolic plane with curvature −2. As in the
previous example, the case 1 < r < n −1 is more complicated and the explicit
formula for ω in the coordinates w does not seem to be of much use.
7) Bergmann metric. We refer to [Wei, pages 57–65] for details of the
following construction. Let M be a complex manifold of dimension m and
H(M) be the space of holomorphic differential forms ϕ of type (m, 0) on M
such that
i
m
2

M
ϕ ∧ ϕ < ∞.
In a first step one shows that H(M) together with the Hermitian product
(ϕ, ψ) := i
m
2

M
ϕ ∧ ψ
is a separable complex Hilbert space. For a unitary basis (ϕ
n
) of H(D), set
θ := i
m
2
¸
ϕ
n
∧ ϕ
n
.
In a second step one shows that this sum converges uniformly on compact
subsets of M and that θ is a real analytic differential form of type (m, m) on
M which does not depend on the choice of unitary basis of H(M).
Let M
0
⊂ M be the open subset of points p ∈ M such that there are an open
neighborhood U of p ∈ M and ϕ
0
, . . . , ϕ
m
∈ H(M) such that ϕ
0
(q) = 0 for
all q ∈ U and such that the holomorphic map z : U →C
m
given by ϕ
j
= z
j
ϕ
0
defines a holomorphic coordinate chart of M on U. We may have M
0
= ∅.
In holomorphic coordinates z on an open subset U of M, write
θ = i
m
2
fdz
1
∧ · · · ∧ dz
m
∧ dz
1
∧ · · · ∧ dz
m
.
Then f is a non-negative real analytic function on U and f > 0 on U ∩ M
0
.
The differential form ω := i∂∂ ln f on U ∩ M
0
does not depend on the choice
of coordinates, hence ω is a real analytic differential form of type (1, 1) on
M
0
. The symmetric bilinear form g(X, Y ) := ω(X, JY ) is a K¨ahler metric on
M
0
with associated K¨ ahler form ω. By construction and definition, M
0
and
θ, hence also ω and g, are invariant under biholomorphic transformations of
M. More generally, any biholomorphic transformation F : M → M

induces
a biholomorphic transformation between M
0
and M

0
such that the pull back
F

θ

= θ.
If the group of biholomorphic transformations of M is transitive on M
0
,
then, by invariance, θ = c vol for some constant c, where vol denotes the volume
form of g. In holomorphic coordinates z, this reads
i
m
fdz
1
∧ dz
1
∧ · · · ∧ dz
m
∧ dz
m
= ci
m
det Gdz
1
∧ · · · ∧ dz
m
∧ dz
m
,
where we use notation as in (4.20) below. Hence f = c det G. From (4.63) below
we conclude that g = −Ric, where Ric denotes the Ricci tensor of g. Hence g
4 K¨ ahler Manifolds 47
is an Einstein metric with Einstein constant −1 if the group of biholomorphic
transformations of M is transitive on M
0
.
If D is a bounded domain in C
m
, then ϕ
0
= dz
1
∧· · · ∧dz
m
and ϕ
j
= z
j
ϕ
0
,
1 ≤ j ≤ m, belong to H(D). Hence D
0
= D; we obtain a K¨ ahler metric g on
D, the Bergmann metric of D.
Let D be as in the previous example. Then up to a scaling factor, the
Bergmann metric of D coincides with the K¨ ahler metric defined there since the
action of U(r, n − r) is transitive on D and preserves both metrics and since
the induced action of the stabilizer U(r) U(n − r) of 0 on the tangent space
T
0
D = C
(n−r)×r
is irreducible. The scaling factor can be determined by the
above observation that the Bergmann metric of D is Einsteinian with Einstein
constant −1.
4.16 Proposition. Let M be a complex manifold with a compatible Rieman-
nian metric g = '· , · ` as in (2.33) and Levi-Civita connection ∇. Then
dω(X, Y, Z) = '(∇
X
J)Y, Z` +'(∇
Y
J)Z, X` +'(∇
Z
J)X, Y `,
2'(∇
X
J)Y, Z` = dω(X, Y, Z) −dω(X, JY, JZ).
Proof. Since M is a complex manifold, we can assume that the vector fields X,
Y , Z, JY , and JZ commute. Then
dω(X, Y, Z) = Xω(Y, Z) +Y ω(Z, X) +Zω(X, Y ),
and similarly for dω(X, JY, JZ). The first equation is now immediate from the
definition of ω and the characteristic properties of ∇. As for the second, we
have
'(∇
X
J)Y, Z` = '∇
X
(JY ), Z` −'J(∇
X
Y ), Z`
= '∇
X
(JY ), Z` +'∇
X
Y, JZ`.
By the Koszul formula and the definition of ω,
2'∇
X
(JY ), Z` = X'JY, Z` +JY 'X, Z` −Z'X, JY `
= Xω(Y, Z) −JY ω(JZ, X) +Zω(X, Y )
and
2'∇
X
Y, JZ` = X'Y, JZ` +Y 'X, JZ` −JZ'X, Y `
= −Xω(JY, JZ) +Y ω(Z, X) −JZω(X, JY ),
where we use that X, Y, Z, JY , and JZ commute.
4.17 Theorem. Let M be a complex manifold with a compatible Riemannian
metric g = '· , · ` as in (2.33) and Levi-Civita connection ∇. Then the following
assertions are equivalent:
48 Lectures on K¨ ahler Manifolds
1. g is a K¨ ahler metric.
2. dω = 0.
3. ∇J = 0.
4. In terms of holomorphic coordinates z, we have
∂g
jk
∂z
l
=
∂g
lk
∂z
j
or, equivalently,
∂g
jk
∂z
l
=
∂g
jl
∂z
k
.
5. The Chern connection of the Hermitian metric h on TM as in Remark 4.3
is equal to the Levi-Civita connection ∇.
6. For each point p
0
in M, there is a smooth real function f in a neighbor-
hood of p
0
such that ω = i∂∂f.
7. For each point p
0
in M, there are holomorphic coordinates z centered at
p
0
such that g(z) = 1 +O(|z|
2
).
A function f as in (6) will be called a K¨ ahler potential, holomorphic coordi-
nates as in (7) will be called normal coordinates at p
0
. The existence of normal
coordinates shows that a K¨ ahler manifold agrees with the model example C
m
in (4.10) up to terms of order two and higher.
Proof of Theorem 4.17. By definition, (1) and (2) are equivalent. The equiva-
lence of (2) and (3) is immediate from Proposition 4.16. The equivalence of the
two assertions in (4) follows from barring the respective equation. The equiv-
alence of (2) with (4)is immediate from d = ∂ + ∂ and the formulas defining
∂ and ∂. The conclusions (6) ⇒ (2) and (7) ⇒ (3) are easy and left as an
exercise.
We show now that (3) is equivalent with (5). Let X be a local holomorphic
vector field and Y be another vector field on M. Then

JY
X = ∇
X
(JY ) + [JY, X] = ∇
X
(JY ) +J[Y, X],
where we use that X is automorphic, see Proposition 2.18. On the other hand,
J∇
Y
X = J(∇
X
Y + [Y, X]) = J∇
X
Y +J[X, Y ].
Hence J is parallel iff ∇X is of type (0, 1) for all local holomorphic vector fields
of M. It is now obvious that (5) ⇒ (3). Vice versa, if J is parallel, then ∇ is
Hermitian with respect to h, hence (3) ⇒ (5) by what we just said.
We show next that (2) ⇒ (6). Since ω is real and dω = 0, we have ω = dα
locally, where α is a real 1-form. Then α = β +
¯
β, where β is a form of type
(1,0). Since ω is of type (1,1), we have
∂β = 0, ∂
¯
β = 0 and ω = ∂β + ∂
¯
β.
4 K¨ ahler Manifolds 49
Hence β = ∂ϕ locally, where ϕ is a smooth complex function. Then
¯
β = ∂ϕ
and hence
ω = ∂∂ϕ + ∂∂ϕ = ∂∂(ϕ −ϕ) = i∂∂f
with f = i(ϕ −ϕ).
We explain now that (4) ⇒ (7). Let z be holomorphic coordinates centered
at p
0
such that g(0) = 1. Then g(z) = 1+O(|z|). We solve for new holomorphic
coordinates ˜ z such that
z
j
:= ˜ z
j
+
1
2
A
j
kl
˜ z
k
˜ z
l
,
where
A
j
kl
= −
∂g
lj
∂z
k
(0).
Applying (4) we get that ˜ g(˜ z) = 1 +O(|˜ z|
2
).
4.18 Remarks. 1) Let J be an almost complex structure on a manifold M
and g be a compatible Riemannian metric on M. If J is parallel with respect
to the Levi-Civita connection of g, then J is a complex structure and hence
M is a K¨ ahler manifold, see Exercise 2.6.2. This generalizes Criterion 3 of
Theorem 4.17.
2) It is immediate that ∇J = 0 iff T

M, T

M are parallel subbundles of
the complexified tangent bundle T
C
M = TM⊗C with respect to the canonical
complex linear extension of ∇ to T
C
M, ∇(X +iY ) = ∇X +i∇Y .
4.1 K¨ahler Form and Volume. One of the most basic features of a K¨ahler
manifold is the intimate connection between its K¨ ahler form and volume. Let
M be a K¨ ahler manifold of complex dimension m with K¨ ahler form ω. There
are the following expressions for ω,
ω = ig
jk
dz
j
∧ dz
k
=
¸
X

j
∧ Y

j
=
i
2
¸
Z

j
∧ Z

j
, (4.19)
where (z
1
, . . . , z
m
) are local holomorphic coordinates, (X
1
, Y
1
, . . . , X
m
, Y
m
) is a
local orthonormal field with JX
j
= Y
j
, and (Z
j
, Z
j
) is the associated complex
frame as in (3.30). In turn, we get the following equivalent expansions of ω
m
,
ω
m
= i
m
m! det Gdz
1
∧ dz
1
∧ · · · ∧ dz
m
∧ dz
m
= m!X

1
∧ Y

1
∧ · · · ∧ X

m
∧ Y

m
= i
m
2
−m
m!Z

1
∧ Z

1
∧ · · · ∧ Z

m
∧ Z

m
= m! vol,
(4.20)
where G = (g
jk
) and where vol denotes the volume form of M.
50 Lectures on K¨ ahler Manifolds
4.21 Remark. A differential two-form ω on a manifold M is called a symplectic
form if dω = 0 and ω is non-degenerate at each point of M, that is, for each
point p ∈ M and non-zero vector v ∈ T
p
M there is a vector w ∈ T
p
M such that
ω(v, w) = 0. A manifold together with a symplectic form is called a symplectic
manifold. If ω is a symplectic form on M, then the real dimension of M is
even, dimM = 2m, and ω
m
is non-zero at each point of M.
The K¨ ahler form of a K¨ ahler manifold is symplectic. Vice versa, the Dar-
boux theorem says that a symplectic manifold (M, ω) admits an atlas of smooth
coordinate charts z : U →U

⊂ R
2m ∼
= C
m
such that z

ω
can
= ω, where
ω
can
=
¸
dx
j
∧ dy
j
=
i
2
¸
dz
j
∧ dz
j
is the K¨ ahler form of the Euclidean metric on C
m
as in Example 4.10.1. Refer-
ences for the basic theory of symplectic forms and manifolds are [Ar, Chapter
8] and [Du, Chapter 3]. Compare also our discussion of symplectic Lagrangian
spaces in Example B.53.
4.22 Exercise (Compare (5.43)). Show that ∗ expω = exp ω, i.e., that

1
j!
ω
j
=
1
(m−j)!
ω
m−j
.
4.23 Theorem. Let M be a K¨ ahler manifold as above.
1) If M is closed, then the cohomology class of ω
k
in H
2k
(M, R) is non-zero
for 0 ≤ k ≤ m. In particular, H
2k
(M, R) = 0 for such k.
2) If N ⊂ M is a compact complex submanifold without boundary of complex
dimension k, then the cohomology class of ω
k
in H
2k
(M, R) and the homology
class of N in H
2k
(M, R) are non-zero.
Proof. Evaluation of ω
m
on the fundamental cycle of M, that is, integration of
ω
m
over M gives

M
ω
m
= m! vol(M) = 0.
This shows the first assertion. As for the second, we note that N with the
induced metric is a K¨ ahler manifold, and the K¨ ahler form is the restriction of
ω to N. Hence integrating ω
k
over N gives k! times the volume of N.
4.24 Example. For m ≥ 2, the Hopf manifold S
2m−1
S
1
has vanishing
second cohomology, hence does not carry a K¨ahler metric.
4.25 Remark. Let (M, ω) be a closed symplectic manifold of (real) dimension
2m. Since ω is closed and ω
m
is non-zero at each point of M, the argument in
the proof of Theorem 4.23 applies and shows that H
2k
(M, R) = 0, 0 ≤ k ≤ m.
It follows that S
2m−1
S
1
does not even carry a symplectic form if m ≥ 2. It
is natural to ask whether there are closed manifolds which carry a symplectic
form but not a K¨ ahler metric. The answer is yes, see Example 5.37.2
4 K¨ ahler Manifolds 51
4.26 Wirtinger Inequality. Let W be a Hermitian vector space with inner
product 'v, w` = Re(v, w) and K¨ahler form ω(u, v) = Im(u, v). Let V ⊂ W be
a real linear subspace of even real dimension 2k and (v
1
, . . . , v
2k
) be a basis of
V . Then

k
(v
1
, . . . , v
2k
)| ≤ k! vol
V
(v
1
, . . . , v
2k
)
with equality iff V is a complex linear subspace of W.
Proof. Observe first that both sides of the asserted inequality are multiplied by
| det A| if we transform (v
1
, . . . , v
2k
) with A ∈ Gl(2k, R) to another basis of V .
Hence we are free to choose a convenient basis of V .
Let v, w ∈ W be orthonormal unit vectors with respect to '· , · `. Then
ω(v, w) = 'Jv, w` ∈ [−1, 1] and ω(v, w) = ±1 iff w = ±Jv. This shows
the assertion in the case k = 1. In the general case we observe first that the
restriction of ω to V is given by a skew-symmetric endomorphismA of V . Hence
there are an orthonormal basis (v
1
, . . . , v
2k
) of V and real numbers a
1
, . . . , a
k
such that
A(v
2j−1
) = a
j
v
2j
and A(v
2j
) = −a
j
v
2j−1
.
By what we said above, |a
j
| ≤ 1 with equality iff v
2j
= ±Jv
2j−1
. Therefore

k
(v
1
, . . . , v
2k
)| = k!|a
1
. . . a
k
| ≤ k! = k! vol
V
(v
1
, . . . , v
2k
)
with equality iff V is a complex linear subspace of W.
4.27 Theorem. Let M be a K¨ ahler manifold as above. Let N ⊂ M be a
compact complex submanifold of real dimension 2k with boundary ∂N (possibly
empty). Let P ⊂ M be an oriented submanifold of dimension 2k and boundary
∂P = ∂N. If N−P is the boundary of a real singular chain, then vol N ≤ vol P
with equality if P is also a complex submanifold.
Proof. Since N − P is the boundary of a real singular chain and ω
k
is closed,
we have

N
ω
k

P
ω
k
= ω
k
(N −P) = 0.
On the other hand, by the Wirtinger inequality 4.26 we have

N
ω
k
= vol N and

P
ω
k
≤ vol P,
and equality holds iff P is a complex submanifold.
It follows that a complex submanifold N ⊂ M is minimal, that is, the trace
of the second fundamental form S of N vanishes. In fact, Theorem 4.27 implies
that N is pluri-minimal in the sense that any complex curve in N is minimal in
52 Lectures on K¨ ahler Manifolds
M. This can also be seen by the following computation: Let X, Y be tangent
vector fields along N. Since N is complex, JY is tangent to N as well and
S(X, JY ) = (∇
X
(JY ))

= (J∇
X
Y )

= J(∇
X
Y )

= JS(X, Y ), (4.28)
where we use that the normal bundle of N is invariant under J. It follows that
S is complex linear in Y , hence, by symmetry, also in X. Therefore
S(X, X) +S(JX, JX) = S(X, X) −S(X, X) = 0,
which is what we wanted to show.
4.2 Levi-Civita Connection. Let M be a K¨ ahler manifold. Recall the com-
plex linear extension of ∇ to T
C
M, ∇(X +iY ) = ∇X +i∇Y . We now extend
∇ also complex linearly in the other variable,

(U+iV )
Z := ∇
U
Z +i∇
V
Z. (4.29)
The extended connection is symmetric with respect to the complex bilinear
extension of the Lie bracket to complex vector fields,

W
Z −∇
Z
W = [W, Z], (4.30)
and metric with respect to the complex bilinear extension of the metric of M
to T
C
M,

Z
'U, V ` = '∇
Z
U, V ` +'U, ∇
Z
V `. (4.31)
We also have

W
Z = ∇
W
Z. (4.32)
Since T

M and T

M are complex subbundles of T
C
M, they stay parallel with
respect to the extension. More precisely, since J is parallel,

(X+iJX)
(Y +iJY ) = (∇
X
Y +∇
Y
X) +iJ(∇
X
Y +∇
Y
X) (4.33)
if [JX, Y ] = 0. Furthermore, if [X, Y ] = [JX, Y ] = 0, then

(X+iJX)
(Y −iJY ) = ∇
(X−iJX)
(Y +iJY ) = 0. (4.34)
With respect to local holomorphic coordinates z of M, we let
Z
j
=

∂z
j
=
1
2


∂x
j
−i

∂y
j

, Z
j
=

∂z
j
=
1
2


∂x
j
+i

∂y
j

, (4.35)
where z
j
= x
j
+ iy
j
, 1 ≤ j ≤ m. By (4.34), the mixed covariant derivatives
vanish,

Zj
Z
k
= ∇
Zj
Z
k
= 0. (4.36)
4 K¨ ahler Manifolds 53
Since T

M and T

M are parallel, we get unmixed Christoffel symbols

Zj
Z
k
= Γ
l
jk
Z
l
and ∇
Zj
Z
k
= Γ
l
jk
Z
l
. (4.37)
All Christoffel symbols of mixed type vanish. Furthermore, by Equation 4.32
we have
Γ
l
jk
= Γ
l
jk
. (4.38)
By (4.31) and (4.36),
Γ
µ
jk
g
µl
= '∇
Zj
Z
k
, Z
l
` =
∂g
kl
∂z
j
Γ
¯ µ
jk
g
¯ µl
= '∇
Zj
Z
k
, Z
l
` =
∂g
kl
∂z
j
=
∂g
kl
∂z
j
.
Thus the Christoffel symbols are given by
Γ
l
jk
= Γ
µ
jk
g
µ¯ ν
g
¯ νl
=
∂g
k¯ ν
∂z
j
g
¯ νl
, (4.39)
where the coefficients g
kl
denote the entries of the inverse of the fundamental
matrix.
4.3 Curvature Tensor. The curvature tensor of the complex bilinear exten-
sion of ∇ to T
C
M is the complex trilinear extension of the usual curvature
tensor R of M to T
C
M. We keep the notation R for the extension. Then
R satisfies the usual symmetries, but now, more generally, for complex vector
fields X, Y , Z, U, V ,
R(X, Y )Z = −R(Y, X)Z,
R(X, Y )Z +R(Y, Z)X +R(Z, X)Y = 0,
(4.40)
and
'R(X, Y )U, V ` = −'R(X, Y )V, U`,
'R(X, Y )U, V ` = 'R(U, V )X, Y `.
(4.41)
We also have the reality condition
R(X, Y )Z = R(X, Y )Z. (4.42)
In addition, since J is parallel,
R(X, Y )JZ = JR(X, Y )Z. (4.43)
54 Lectures on K¨ ahler Manifolds
By (2.33), (4.41), and (4.43),
'R(X, Y )JU, JV ` = 'R(JX, JY )U, V ` = 'R(X, Y )U, V `. (4.44)
Since T

M and T

M are parallel, we have
R(U, V )(T

M) ⊆ T

M and R(U, V )(T

M) ⊆ T

M. (4.45)
With respect to local holomorphic coordinates,
R(Z
j
, Z
k
) = R(Z
j
, Z
k
) = 0. (4.46)
For the mixed terms, we get from (4.45)
R(Z
j
, Z
k
)Z
l
= R
µ
jkl
Z
µ
, R(Z
j
, Z
k
)Z
l
= R
¯ µ
jkl
Z
µ
,
R(Z
j
, Z
k
)Z
l
= R
µ
jkl
Z
µ
, R(Z
j
, Z
k
)Z
l
= R
¯ µ
jkl
Z
µ
.
(4.47)
Equation 4.42 implies that
R
¯ µ
jkl
= R
µ
jkl
and R
¯ µ
jkl
= R
µ
jkl
. (4.48)
By (4.36), we have
R
µ
jkl
= −
∂Γ
µ
jl
∂z
k
and R
µ
jkl
=
∂Γ
µ
kl
∂z
j
. (4.49)
Recall that the sectional curvature determines the curvature tensor. We define
the holomorphic sectional curvature of M to be the sectional curvature of the
complex lines in TM. If X is a non-zero tangent vector of M, then the com-
plex line spanned by X has (X, JX) as a basis over R and the corresponding
holomorphic sectional curvature is
K(X ∧ JX) =
'R(X, JX)JX, X`
|X|
4
. (4.50)
4.51 Proposition. The holomorphic sectional curvature determines R.
Proof. We follow the proof of Proposition IX.7.1 in [KN] and consider the
quadri-linear map
Q(X, Y, U, V ) = 'R(X, JY )JU, V ` +'R(X, JU)JY, V ` +'R(X, JV )JY, U`.
It is immediate from the symmetries of R listed above that Q is symmetric.
Thus polarization determines Q explicitly from its values
Q(X, X, X, X) = 3K(X ∧ JX) · |X|
4
(4.52)
4 K¨ ahler Manifolds 55
along the diagonal. Hence it suffices to show that Q determines R. Now
Q(X, Y, X, Y ) = 2'R(X, JY )JY, X` +'R(X, JX)JY, Y `.
By the second symmetry in (4.40),
'R(X, JX)JY, Y ` = −'R(JX, JY )X, Y ` −'R(JY, X)JX, Y `
= 'R(X, Y )Y, X` +'R(X, JY )JY, X`.
Therefore
Q(X, Y, X, Y ) = 3'R(X, JY )JY, X` +'R(X, Y )Y, X`,
and hence also
Q(X, JY, X, JY ) = 3'R(X, Y )Y, X` +'R(X, JY )JY, X`.
In conclusion,
3Q(X, JY, X, JY ) −Q(X, Y, X, Y ) = 8'R(X, Y )Y, X`. (4.53)
Hence Q determines the sectional curvature, hence R.
4.54 Exercise. The bisectional curvature of X, Y is defined to be 'R(X, JX)JY, Y `.
Show that it deserves its name,
'R(X, JX)JY, Y ` = 'R(X, Y )Y, X` +'R(X, JY )JY, X`.
4.4 Ricci Tensor. The Ricci tensor is defined by
Ric(X, Y ) = tr(Z →R(Z, X)Y ), (4.55)
where we can take the trace of the map on the right hand side as an R-linear
endomorphism of TM or as a C-linear endomorphism of T
C
M. The Ricci tensor
is symmetric and
Ric(X, Y ) = Ric(X, Y ). (4.56)
The associated symmetric endomorphism field of TM or T
C
M is also denoted
Ric, that is, we let 'Ric X, Y ` := Ric(X, Y ).
4.57 Proposition. In terms of an orthonormal frame (X
1
, JX
1
, . . . , X
m
, JX
m
)
of M, the Ricci tensor is given by
Ric X =
¸
R(X
j
, JX
j
)JX.
In particular, Ric JX = J Ric X.
56 Lectures on K¨ ahler Manifolds
Proof. We compute
Ric(X, Y ) =
¸
'R(X
j
, X)Y, X
j
` +
¸
'R(JX
j
, X)Y, JX
j
)
=
¸
'R(X
j
, X)JY, JX
j
` −
¸
'R(JX
j
, X)JY, X
j
`
=
¸
'R(X, X
j
)JX
j
, JY ` +
¸
'R(JX
j
, X)X
j
, JY `
= −
¸
'R(X
j
, JX
j
)X, JY ` =
¸
'R(X
j
, JX
j
)JX, Y `.
We conclude that
Ric(JX, JY ) = Ric(X, Y ). (4.58)
and hence there is an associated real differential form of type (1, 1), the Ricci
form
ρ(X, Y ) := Ric(JX, Y ). (4.59)
In terms of local holomorphic coordinates,
Ric(Z
j
, Z
k
) = Ric(Z
j
, Z
k
) = 0, (4.60)
by (4.47). For the mixed terms, we have
Ric(Z
j
, Z
k
) = R
l
ljk
= −
∂Γ
l
lk
∂z
j
= Ric
jk
= Ric
jk
= Ric(Z
j
, Z
k
). (4.61)
With G = (g
jk
) we get, by (4.39),
∂ ln det G
∂z
j
= tr

∂G
∂z
j
· G
−1
«
= Γ
k
jk
. (4.62)
Therefore,
Ric
jk
= −

2
ln det G
∂z
j
∂z
k
. (4.63)
For the Ricci form we get
ρ = i Ric
jk
dz
j
∧ dz
k
= −i ∂∂ ln det G. (4.64)
This formula shows that ρ is closed. In fact, by the above computation of the
Ricci curvature we have
ρ(X, Y ) = −
¸
'R(X, Y )E
j
, JE
j
` =
¸
'JR(X, Y )E
j
, E
j
`
= i
¸
'R(X, Y )E
j
, E
j
` +i
¸
i'JR(X, Y )E
j
, E
j
`
= i
¸
h(R(X, Y )E
j
, E
j
) = i tr R(X, Y ),
(4.65)
where we consider TM with the given complex structure J as a holomorphic
vector bundle over M with Hermitian metric h = g+iω. Now ∇ is a connection
4 K¨ ahler Manifolds 57
on TM considered as a complex vector bundle, namely the Chern connection
associated to h. By (A.33) and (A.35), the first Chern form of TM associated
to ∇ is given by
c
1
(TM, ∇) =
i

tr R =
1

ρ. (4.66)
This shows again that ρ is closed. It also shows that [ρ/2π] = c
1
(TM) in
H
2
(M, R) and hence that [ρ/2π] lies in the image of H
2
(M, Z) →H
2
(M, R).
The induced curvature on the canonical bundle K
M
= A
m,0
(M, C) is given
by
ˆ
R(X, Y )(dz
1
∧ · · · ∧ dz
m
) = iρ(X, Y ) · dz
1
∧ · · · ∧ dz
m
. (4.67)
To prove this, we use that the curvature
ˆ
R acts as a derivation on A

(M, C),
ˆ
R(X, Y )(dz
1
∧ · · · ∧ dz
m
) =
¸
dz
1
∧ · · · ∧
ˆ
R(X, Y )dz
j
∧ · · · ∧ dz
m
= −tr R(X, Y )(dz
1
∧ · · · ∧ dz
m
).
Now the claim follows from (4.65). We conclude that the first Chern form
c
1
(K
M
,
ˆ
∇) = −
i

tr R = −
1

ρ = −c
1
(TM, ∇). (4.68)
4.5 Holonomy. K¨ ahler manifolds are characterized by the property that their
complex structure is parallel. This links the K¨ ahler condition to holonomy
10
.
Suppose that M is a connected K¨ ahler manifold, and let p be a point in
M. Then the complex structure J
p
turns T
p
M into a complex vector space.
Since J is parallel, the holonomy group Hol M = Hol
p
M of M at p preserves
J
p
. Hence up to a unitary isomorphism of (T
p
M, J
p
) with C
m
, m = dim
C
M,
the holonomy group Hol Mof M is contained in U(m). Vice versa, let M be a
connected Riemannian manifold of dimension n = 2m and p be a point in M.
Suppose that after some identification of T
p
M with R
2m
and of R
2m
with C
m
,
the holonomy group Hol
p
M ⊂ U(m) ⊂ SO(2m). Then the complex structure
J
p
on T
p
M induced from the complex structure on C
m
extends to a parallel
complex structure J on M, and M together with J is a K¨ ahler manifold. We
formulate the result of our discussion in the following somewhat sloppy way.
4.69 Proposition. A connected Riemannian manifold of real dimension 2m
is a K¨ ahler manifold iff its holonomy group is contained in U(m).
If the holonomy along a closed loop c of M at p is given by A on T
p
M, viewed
as a complex vector space, then the induced holonomy on the fiber A
m,0
p
(M, C)
of the canonical bundle is given by the complex determinant det
C
A. Hence
Hol(M) ⊂ SU(m) iff A
m,0
(M, C) has a parallel complex volume form. It also
follows that the reduced holonomy group Hol
0
(M) ⊂ SU(m) iff A
m,0
(M, C) is
flat. By (4.67) this holds iff M is Ricci-flat, that is, if Ric = 0.
10
Excellent references for holonomy are [Bes, Chapter 10], [Br], and [Jo, Chapter 3].
58 Lectures on K¨ ahler Manifolds
4.70 Proposition. A connected Riemannian manifold of real dimension 2m
is a Ricci-flat K¨ahler manifold iff its reduced holonomy group is contained in
SU(m).
4.71 Remark. A connected Riemannian manifold M of real dimension 2m
with holonomy Hol(M) ⊂ SU(m) is called a Calabi–Yau manifold
11
. If the real
dimension of M is 4k, we may also have Hol(M) ⊂ Sp(k) ⊂ SU(m), then M is
called a hyper-K¨ ahler manifold. By Proposition 4.70, Calabi–Yau and hyper-
K¨ ahler manifolds are Ricci-flat K¨ ahler manifolds. For examples of Calabi–Yau
and hyper-K¨ ahler manifolds, see Chapters 6 and 7 in [Jo].
Let M be a simply connected complete Riemannian manifold. Then we
have the de Rham decomposition,
M = M
0
M
1
· · · M
k
, (4.72)
where M
0
is a Euclidean space and M
i
, i ≥ 1, is a simply connected complete
Riemannian manifold with irreducible holonomy
12
. Moreover, this decomposi-
tion is unique up to a permutation of the factors M
i
, i ≥ 1. Hence we have, at
a point p = (p
0
, p
1
, . . . , p
k
) ∈ M,
Hol M = Hol M
1
· · · Hol M
k
. (4.73)
After identifying T
pi
M
i
with the corresponding subspace of T
p
M, we also have
T
p0
M
0
= Fix Hol M, (4.74)
the set of vectors fixed by Hol M, and, for each i ≥ 1,
T
p0
M
0
+T
pi
M
i
= Fix
¸
j=i
Hol M
j
. (4.75)
Suppose now in addition that M is a K¨ ahler manifold. Then the complex
structure J of M is parallel, and hence J
p
A = AJ
p
for all A ∈ Hol M. By
(4.74), we have J
p
T
p0
M
0
= T
p0
M
0
, hence M
0
is a complex Euclidean space.
From (4.75) we conclude that J
p
T
pi
M
i
= T
pi
M
i
for all i ≥ 1. Now the de
Rham decomposition as in (4.72) implies the de Rham decomposition for K¨ ahler
manifolds.
4.76 De Rham Decomposition of K¨ahler Manifolds. Let M be a simply
connected complete K¨ahler manifold. Then
M = M
0
M
1
· · · M
k
,
where M
0
is a complex Euclidean space and M
i
, i ≥ 1, is a simply connected
complete K¨ ahler manifold with irreducible holonomy.
11
There are variations of this definition in the literature.
12
It is understood that M
0
= {0} and k = 0 are possible.
4 K¨ ahler Manifolds 59
4.6 Killing Fields. We start with a discussion of Killing fields on Riemannian
manifolds.
4.77 Proposition (Yano [Ya], Lichnerowicz [Li1]). Let M be Riemannian and
X be a vector field on M. Then
X is a Killing field =⇒ div X = 0 and ∇

∇X = Ric X.
If M is closed, then the converse holds true as well.
Proof. If X is a Killing field, then L
X
g = 0 and hence div X = tr L
X
g/2 = 0,
see (1.8). Furthermore, ∇

∇X = Ric X by Exercise 1.5.2.
Suppose now that M is closed and let ξ = X

. The decomposition of the
bilinear form
ˆ
∇ξ into symmetric and skew-symmetric part is given by
ˆ
∇ξ = (
ˆ
∇ξ)
sym
+ (
ˆ
∇ξ)
skew
=
1
2
L
X
g +
1
2
dξ.
Using that
ˆ


= d

on alternating two-forms, we obtain
ˆ


ˆ
∇ξ =
1
2
ˆ


(L
X
g) +
1
2
d

dξ.
The second ingredient is a simple consequence of the Bochner identity 1.38:


∇X = Ric X ⇐⇒
ˆ


ˆ
∇ξ =
1
2

d
(ξ).
Assume now that div X = 0 and ∇

∇X = Ric X. Since div X = −d

ξ, we
obtain
ˆ


L
X
g = dd

ξ = 0.
It follows that L
X
g is L
2
-perpendicular to im
ˆ
∇. However, adding the term
dξ to L
X
g, which is pointwise and hence L
2
-perpendicular to L
X
g, we get
2
ˆ
∇ξ ∈ im
ˆ
∇. We conclude that L
X
g = 0 and hence that X is a Killing field.

Recall that a vector field X on a complex manifold M is automorphic iff
[X, JY ] = J[X, Y ] for all vector fields Y on M. In Proposition 2.18 we showed
that X is automorphic iff X is holomorphic.
Suppose now that M is a K¨ ahler manifold, and let X be a vector field
on M. Since the Levi-Civita connection is equal to the Chern connection of
the holomorphic vector bundle TM with the induced Hermitian metric, X is
automorphic iff ∇X is a form of type (1, 0) with values in TM. In fact,

JY
X = ∇
X
JY −[X, JY ]
= J∇
X
Y −[X, JY ]
= J∇
Y
X +J[X, Y ] −[X, JY ]
(4.78)
for any vector field Y on M. That is, X is automorphic iff ∇X ◦ J = J ◦ ∇X.
60 Lectures on K¨ ahler Manifolds
4.79 Proposition (Lichnerowicz [Li2]). Let M be a K¨ ahler manifold and X
be a vector field on M. Then
X is automorphic =⇒ ∇

∇X = Ric X.
If M is closed, then the converse holds as well.
Proof. We decompose the one-form ∇X into types,
∇X = ∇
1,0
X +∇
0,1
X =
1
2
(∇X −J ◦ ∇X ◦ J) +
1
2
(∇X +J ◦ ∇X ◦ J).
This decomposition is pointwise and hence L
2
-perpendicular. From (4.78) we
getthat X is automorphic iff ∇
0,1
X = 0.
The key observation is the following:


(J ◦ ∇X ◦ J) = J ◦ ∇

(∇X ◦ J) = −Ric X, (4.80)
where the first equality holds since J is parallel and the second follows from
Proposition 4.57. We conclude that
2∇


0,1
X = ∇

∇X −Ric X.
Hence if X is automorphic, then ∇

∇X = Ric X.
We now repeat the linear algebra argument from the end of the proof of
Proposition 4.77: Suppose that M is closed and that ∇

∇X = Ric X. Then



0,1
X = 0, hence ∇
0,1
X is L
2
-perpendicular to im∇. Adding the term

1,0
X, which is L
2
-perpendicular to ∇
0,1
X, we get ∇X ∈ im∇. Hence

0,1
X = 0.
Comparing Propositions 4.77 and 4.79 we arrive at the final result in this
subsection.
4.81 Theorem. Let M be a closed K¨ ahler manifold. Then Killing fields on
M are automorphic. Vice versa, an automorphic field is a Killing field iff it is
volume preserving.
4.82 Exercise. Find non-automorphic Killing fields on C
m
, m ≥ 2.
4.83 Remark. The description of Killing and automorphic vector fields in
terms of the equation ∇

∇X = Ric X is important for the global geometry of
compact K¨ahler manifolds with Ric ≤ 0 or Ric > 0, compare Chapters 6 and
7.
That Killing fields on a closed K¨ ahler manifold M are automorphic can
also be seen in a more conceptual way: A Killing field preserves the metric of
M. Hence in order to show that it preserves J as well, it is clearly enough to
show that it preserves ω, that is, that L
X
ω = 0. But ω is parallel and hence a
harmonic form.
4 K¨ ahler Manifolds 61
Now every diffeomorphism of M acts on H

(M, R), but isometries of M
act on H

(M, R) as well. Clearly, for isometries the two actions are compatible
with the Hodge isomorphism H

(M, R) → H

(M, R). It follows that homo-
topic isometries induce the same maps on H

(M, R), hence connected groups of
isometries act trivially on H

(M, R). In particular, L
X
α = 0 for all harmonic
forms α and all Killing fields X.
5 Cohomology of K¨ahler Manifolds
Let M be a K¨ ahler manifold. In our applications to cohomology further on
we will assume that M is closed. But for the moment this assumption is not
necessary since all of our computations are of a local nature.
Let E →M be a holomorphic vector bundle over M. Let h be a Hermitian
metric on E and D be the corresponding Chern connection. In what follows,
we use notation and results introduced in Chapter 1.
We consider some endomorphism fields of A

(M, E) and their relation with
exterior differentiation. To that end we recall that each element α of A

(M, E)
can be written uniquely as a sum α =
¸
α
r
, where α
r
has degree r, or as a sum
α =
¸
α
p,q
, where α
p,q
is of type (p, q). The corresponding maps P
r
: α →α
r
and P
p,q
: α →α
p,q
define fields of projections of A

(M, E).
The complex structure J of M acts on A

(M, E) via pull back. The induced
endomorphism field of A

(M, E) is denoted C. Note that C leaves the type of
a form invariant. In terms of the above projection fields, we have
C =
¸
i
p−q
P
p,q
. (5.1)
The adjoint operator is
C

=
¸
i
q−p
P
p,q
= C
−1
. (5.2)
We set W := (∗ ⊗ h

)(∗ ⊗ h). Again, W leaves the type of forms invariant.
Since the real dimension of M is even, we have
W =
¸
(−1)
r
P
r
. (5.3)
The Lefschetz map is defined by
L(α) := ω ∧ α =
¸
X

j
∧ Y

j
∧ α =
i
2
¸
Z

j
∧ Z

j
∧ α, (5.4)
where ω is the K¨ ahler form of M and (X
j
, Y
j
) and (Z
j
, Z
j
) are frames as in
(3.30). The Lefschetz map raises types of forms by (1, 1). To compute the
adjoint operator L

of L we use the first relation in Exercise 1.26
13
. Since
Z

j
= 2Z

j
,
L

α =
¸
Y
j
(X
j
α) =
2
i
¸
Z
j
(Z
j
α). (5.5)
We have vanishing commutators
[L, C] = [L, W] = [L

, C] = [L

, W] = 0, (5.6)
by (5.1) and (5.3), where we note that C and W leave the type of forms invariant
and that L raises and L

lowers the type by (1, 1).
13
The L
2
-adjoint of a field of endomorphisms is the field of adjoint endomorphisms so that
there is no ambiguity here.
5 Cohomology of K¨ ahler Manifolds 63
5.7 Remark. The operators C, W, and L and their adjoints are real, that is,
also defined on A

(M, R) and corresponding bundle valued forms A

(M, E),
where E is a real vector bundle, endowed with a Riemannian metric.
5.8 Proposition. [L

, L] =
¸
(m−r)P
r
.
Proof. By (5.4), (5.5), and Exercise 1.26 (applied to complex tangent vectors),
LL

α =
¸
Z

j
∧ Z

j
∧ (Z
k
(Z
k
α))
= −
¸
j=k
Z

j
∧ (Z
k
(Z

j
∧ (Z
k
α))) +
¸
j=k
Z

j
∧ Z

j
∧ (Z
j
(Z
j
α)).
Vice versa,
L

Lα =
¸
Z
k
(Z
k
(Z

j
∧ Z

j
∧ α))
= −
¸
j=k
Z
k
(Z

j
∧ (Z
k
(Z

j
∧ α))) +
¸
j=k
Z
j
(Z
j
(Z

j
∧ Z

j
∧ α)).
The first terms on the right hand sides of the above formulas coincide, hence
cancel in the commutator [L

, L]. Assume now without loss of generality that
α = Z

I
∧ Z

J
⊗ σ, where I and J are multi-indices. Then the second term on
the right hand side in the first formula counts the number of j with j ∈ I and
j ∈ J. The second term on the right hand side in the second formula counts
the number of j with j / ∈ I and j / ∈ J.
We recall now that the standard basis vectors
X =

0 1
0 0

, Y =

0 0
1 0

, H =

1 0
0 −1

(5.9)
of sl
2
(C) satisfy
[X, Y ] = H, [H, X] = 2X, [H, Y ] = −2Y. (5.10)
It follows that
X →L

, Y →L, H →
¸
(m−r)P
r
(5.11)
extends to a representation of sl
2
(C) on (the fibers of) A

(M, E) and on
A

(M, E). Finite dimensional representations of sl
2
(C) split into the sum of
irreducible ones, where the latter are classified by integers n ≥ 0:
V
n
= Cv +CY v + · · · +CY
n
v, (5.12)
where v ∈ V
n
, v = 0, is primitive, that is, Xv = 0, and where HY
k
v = (2k−n)v,
see e.g. Lecture 11 in [FH]. We note that Hv = −nv.
64 Lectures on K¨ ahler Manifolds
5.13 Exercise. Show that on A
r
(M, E) and for s ≥ 1,
[L
s
, L

] = s(r −m)L
s−1
.
The tensor field L

is parallel and has constant rank on A
r
(M, E). Hence
its kernel A
r
P
(M, E) of primitive forms of degree r is a parallel subbundle of
A
r
(M, E). We denote by A
r
P
(M, E) the space of smooth sections of A
r
P
(M, E).
By Proposition 5.8 and what we said about representations of sl
2
(C), we
have A
r
P
(M, E) = 0 for r > m. Furthermore, if α ∈ A
r
P
(M, E), α = 0, then
L
s
α = 0 for 0 ≤ s ≤ m− r and L
s
α = 0 for s > m− r. It follows that the
parallel tensor field L
s
: A
r
(M, E) →A
r+2s
(M, E) is injective for 0 ≤ s ≤ m−r
and surjective for s ≥ m − r, and similarly for A
r
(M, E). Since any finite
dimensional representation of sl
2
(C) splits into irreducible ones,
A
r
(M, E) = ⊕
s≥0
L
s
A
r−2s
P
(M, E), (5.14)
A
r
(M, E) = ⊕
s≥0
L
s
A
r−2s
P
(M, E), (5.15)
the Lefschetz decompositions of A
r
(M, E) and A
r
(M, E).
Since
¸
(m−r)P
r
preserves types of forms and since L raises and L

lowers
types by (1, 1), we have refined decompositions,
A
p,q
(M, E) = ⊕
s≥0
L
s
A
p−s,q−s
P
(M, E), (5.16)
A
p,q
(M, E) = ⊕
s≥0
L
s
A
p−s,q−s
P
(M, E), (5.17)
the Lefschetz decompositions of A
p,q
(M, E) and A
p,q
(M, E). It is trivial that
A
p,q
(M, E) = A
p,q
P
(M, E) and A
p,q
(M, E) = A
p,q
P
(M, E) (5.18)
if p = 0 or if q = 0.
5.19 Remark. Since the tensor field L is parallel, the Lefschetz decompositions
of A
r
(M, E) and A
p,q
(M, E) are parallel. It follows from Exercise 5.13 that they
are orthogonal as well.
5.1 Lefschetz Map and Differentials. We now compute commutators with
exterior derivatives. It is immediate from (3.32) and (4.19) that
[L, ∂
D
] = [L, ∂] = [L, d
D
] = 0. (5.20)
Hence the commutators of the adjoint operators also vanish,
[L

, (∂
D
)

] = [L

, ∂

] = [L

, (d
D
)

] = 0. (5.21)
5.22 Proposition. We have
[L, (∂
D
)

] = i∂, [L, ∂

] = −i∂
D
, [L

, ∂
D
] = i∂

, [L

, ∂] = −i(∂
D
)

.
Proof. By (3.33) and (4.19),
L(∂
D
)

α = −
i
2
¸
j,k
Z

j
∧ Z

j
∧ (2Z
k

ˆ
D
Z
k
α) =
5 Cohomology of K¨ ahler Manifolds 65
= −i
¸
j=k
Z
k
(Z

j
∧ Z

j

ˆ
D
Z
k
α) +i
¸
j=k
Z

j
∧ (Z
j
(Z

j

ˆ
D
Zj
α))
and
(∂
D
)

Lα = −
i
2
¸
j,k
2Z
k

ˆ
D
Z
k
(Z

j
∧ Z

j
∧ α)
= −i
¸
j,k
Z
k
(Z

j
∧ (Z

j

ˆ
D
Z
k
α)).
In conclusion,
[L, (∂
D
)

]α = i
¸
j
{Z

j
∧ (Z
j
(Z

j

ˆ
D
Zj
α)) +Z
j
(Z

j
∧ (Z

j

ˆ
D
Zj
α))}
= i
¸
j
Z

j

ˆ
D
Zj
α = i∂α.
The proof of the second equation [L, ∂

] = −i∂
D
is similar. The remaining two
equations are adjoint to the first two.
We now discuss the three Laplacians ∆
d
, ∆

, and ∆

D and their relation
with the Lefschetz map.
5.23 Proposition. We have
[L, ∆

] = −iR
D

ε
and [L, ∆

D] = iR
D

ε
,
where R
D

ε
is the operator sending α to R
D

ε
α.
Proof. From Proposition 3.29 we recall that R
D

ε
= ∂
D
∂ +∂∂
D
. By Equation
5.20 and Proposition 5.22, we have
[L, ∆

] = L∂∂

+L∂

∂ −∂∂

L −∂

∂L
= ∂L∂

+ ∂

L∂ −i∂
D
∂ − ∂L∂

−i∂∂
D
−∂

L∂
= −i(∂
D
∂ + ∂∂
D
).
Similarly,
[L, ∆

D] = L∂
D
(∂
D
)

+L(∂
D
)


D
−∂
D
(∂
D
)

L −(∂
D
)


D
L
= ∂
D
L(∂
D
)

+ (∂
D
)

L∂
D
+i∂∂
D
−∂
D
L(∂
D
)

+i∂
D
∂ −(∂
D
)

L∂
D
= i(∂
D
∂ + ∂∂
D
).
5.24 Lemma. We have

D


+ ∂


D
= ∂(∂
D
)

+ (∂
D
)

∂ = 0.
66 Lectures on K¨ ahler Manifolds
Proof. By Proposition 5.22, we have
∂(∂
D
)

+ (∂
D
)

∂ = i∂L

∂ −i∂∂L

+iL

∂∂ −i∂L

∂ = 0
since ∂∂ = 0. Now ∂
D


+ ∂


D
is the adjoint operator of ∂(∂
D
)

+ (∂
D
)

∂.
The lemma follows.
5.25 Theorem. We have


+∆

D = ∆
d
D and ∆

−∆

D = [iR
D

ε
, L

].
In particular, [L, ∆
d
D] = 0 and ∆
d
D preserves the type of forms.
Proof. We compute

d
D = (∂
D
+ ∂)((∂
D
)

+ ∂

) + ((∂
D
)

+ ∂

)(∂
D
+ ∂)
= ∆

D + ∆

+ ∂
D


+ ∂(∂
D
)

+ (∂
D
)

∂ + ∂


D
= ∆

D +∆

,
by (5.24). Now the two last claims follow from Proposition 5.23 and since ∆

D
and ∆

preserve the type of forms.
To prove the second equality, let α and β be differential forms with values
in E, where β has compact support. Then there is no boundary term when
considering integration by parts as in the L
2
-products below. By Equation 5.20
and Proposition 5.22,
(i∂

α, ∂

β)
2
= (L


D
α, ∂

β)
2
−(∂
D
L

α, ∂

β)
2
= (∂L


D
α, β)
2
−(∂∂
D
L

α, β)
2
= (L

∂∂
D
α, β)
2
+ (i(∂
D
)


D
α, β)
2
−(∂∂
D
L

α, β)
2
.
Similarly,
(i∂α, ∂β)
2
= −(∂α, i∂β)
2
= −(∂α, L(∂
D
)

β)
2
+ (∂α, (∂
D
)

Lβ)
2
= −(L

∂α, (∂
D
)

β)
2
+ (L


D
∂α, β)
2
= −(∂L

α, (∂
D
)

β)
2
+ (i(∂
D
)

α, (∂
D
)

β)
2
+ (L


D
∂α, β)
2
= −(∂
D
∂L

α, β)
2
+ (i∂
D
(∂
D
)

α, β)
2
+ (L


D
∂α, β)
2
.
The sum of the terms on the left hand side is equal to (i∆

α, β)
2
, hence the
second equality.
5.26 Corollary. If D is flat, then 2∆

= 2∆

D = ∆
d
.
5 Cohomology of K¨ ahler Manifolds 67
Since L commutes with ∆
d
D and ∆
d
D = (∆
d
D)

, L

commutes with ∆
d
D
as well. Hence the representation of sl
2
(C) on A

(M, E) as in (5.11) induces
a representation of sl
2
(C) on the space H

(M, E) of ∆
d
-harmonic forms. Say
that a harmonic form α is primitive if L

α = 0, and denote by H
r
P
(M, E) ⊂
H
r
(M, E) the space of primitive harmonic forms of degree r. A primitive
harmonic form is primitive at each point of M. This will be important in some
applications further on.
The following results are immediate consequences of the above commu-
tation relations and the Lefschetz decompositions of A
r
(M, E) in (5.15) and
A
p,q
(M, E) in (5.17).
5.27 Hard Lefschetz Theorem. The map L
s
: H
r
(M, E) →H
r+2s
(M, E) is
injective for 0 ≤ s ≤ m−r and surjective for s ≥ m−r, s ≥ 0. Furthermore,
H
r
(M, E) = ⊕
s≥0
L
s
H
r−2s
P
(M, E),
the Lefschetz decomposition of H
r
(M, E).
If E is the trivial line bundle with the standard Hermitian metric, then
R
D
= 0 and hence [iR
D

ε
, L

] = 0. This is the important special case of
differential forms with values in C. More generally, if D is flat, that is, if R
D
=
0, then we also have [iR
D

ε
, L

] = 0 and, in particular, 2∆

= 2∆

D = ∆
d
.
5.28 Theorem. Assume that D is flat. Then
H
r
(M, E) = ⊕
p+q=r
H
p,q
(M, E), H
p,q
(M, E) = ⊕
s≥0
L
s
H
p−s,q−s
P
(M, E),
the Hodge and Lefschetz decompositions of H
r
(M, E) and H
p,q
(M, E).
5.29 Remark. Note that L, L

, and
¸
(m − r)P
r
preserve square integra-
bility of differential forms so that Theorems 5.27 and 5.32 also hold for the
corresponding spaces of square integrable harmonic forms. Moreover, the de-
compositions are pointwise and hence L
2
-orthogonal.
Suppose that D is flat and that p + q ≤ m = dim
C
M. Then, by Theo-
rem 5.28,
H
p,q
(M, E) = ⊕
s≥0
L
s
H
p−s,q−s
P
(M, E)
= H
p,q
P
(M, E) ⊕LH
p−1,q−1
(M, E).
(5.30)
Furthermore, the Lefschetz map L
s
: H
p−s,q−s
P
(M, C) →H
p,q
(M, C) is injective
for all s ≥ 0, by Theorem 5.27.
5.2 Lefschetz Map and Cohomology. If D is flat, then d
D
d
D
= 0 and
we get the associated cohomology H

(M, E). If M is closed, then cohomology
classes in H

(M, E) are uniquely represented by harmonic forms, by Hodge
theory.
68 Lectures on K¨ ahler Manifolds
5.31 Hard Lefschetz Theorem. Assume that M is closed and D is flat.
Then the map L
s
: H
r
(M, E) → H
r+2s
(M, E) is injective for 0 ≤ s ≤ m − r
and surjective for s ≥ m−r. Furthermore, we have the Lefschetz decomposition
H
r
(M, E) = ⊕
s≥0
L
s
H
r−2s
P
(M, E).
We recall that Poincar´e duality also shows that H
r
(M, C)

= H
2m−r
(M, C).
However, the isomorphism in Theorem 5.31 is not obtained using the Hodge
operator but the (m−r)-fold cup product with a cohomology class, the K¨ ahler
class. The existence of a cohomology class with this property is a rather special
feature of closed K¨ahler manifolds.
5.32 Theorem. Assume that M is closed and D is flat. Then we have Hodge
and Lefschetz decompositions,
H
r
(M, E) = ⊕
p+q=r
H
p,q
(M, E), H
p,q
(M, E) = ⊕
s≥0
L
s
H
p−s,q−s
P
(M, E).

Suppose that M is closed and D is flat. Let
h
p,q
P
(M, E) := dimH
p,q
P
(M, E) = dimH
p,q
P
(M, E). (5.33)
If p +q ≤ m, then by Theorem 5.32,
h
p,q
(M, E) =
¸
s≥0
h
p−s,q−s
P
(M, E) = h
p,q
P
(M, E) +h
p−1,q−1
(M, E), (5.34)
see also (5.30) above. Recall that b
r
(M) = dimH
r
(M, C) and define Betti
numbers b
r
(M, E) := dimH
r
(M, E).
5.35 Theorem. If M is closed and D is flat, then
1) b
r
(M, E) =
¸
p+q=r
h
p,q
(M, E);
2) h
p,q
(M, E) = h
m−p,m−q
(M, E

);
3) h
q,p
(M, C) = h
p,q
(M, C) = h
m−p,m−q
(M, C);
4) b
r
(M) is even if r is odd, b
1
(M) = 2h
1,0
(M, C).
In particular, h
1,0
(M, C) is a topological invariant of M.
Proof. Assertion 1) is immediate from Theorem 5.32. Assertion 2) follows from
Serre duality 3.12 as does the second equality in Assertion 3) since the dual
bundle of the trivial line bundle is the trivial line bundle. Since conjugation
commutes with ∆
d
(for complex valued differential forms on any Riemannian
manifold), the map
H
p,q
(M, C) →H
q,p
(M, C), ϕ →ϕ,
is (well defined and) a conjugate linear isomorphism. This shows the first equa-
lity in Assertion 3). The remaining assertions are immediate consequences of
Assertions 1) and 3).
5 Cohomology of K¨ ahler Manifolds 69
5.36 Remarks. 1) The above results motivate a graphic arrangement of the
Dolbeault cohomology groups in a diamond with vertices H
0,0
(M, E) and
H
m,m
(M, E) to the south and north and H
m,0
(M, E) and H
0,m
(M, E) to the
west and east, respectively. This diamond is called the Hodge diamond. The
Lefschetz isomorphism L
r
: H
r
(M, E) →H
2m−r
(M, E) is compatible with the
type decomposition and corresponds to a reflection about the equator of the
Hodge diamond. In the case E = C = E

, Serre duality corresponds to the
reflection about the center. Conjugation corresponds to the reflection about
the vertical axis between north and south pole.
2) If M is a closed complex manifold, then Betti numbers and Euler characte-
ristic of M satisfy the Fr¨ olicher relations
b
r
(M) ≤
¸
p+q=r
h
p,q
(M, C) and χ(M) =
¸
p,q
(−1)
p+q
h
p,q
(M, C),
see [GH, page 444] and [Hir, Theorem 15.8.1]. In the K¨ ahler case, these are
clear from Theorem 5.35.1.
5.37 Examples. 1) We see again that Hopf manifolds M = S
2m−1
S
1
do
not carry K¨ ahler metrics for m ≥ 2 since their first Betti number is odd,
b
1
(M) = 1. It is easy to see that h
1,0
(M, C) = 0, see Lemma 9.4 in Borel’s
Appendix to [Hir]. In Theorem 9.5 loc. cit. Borel determines the Dolbeault
cohomology of Hopf and Calabi–Eckmann manifolds completely.
2) We come back to the question whether there are closed symplectic mani-
folds which do not carry K¨ ahler metrics, compare Remark 4.25. A first example
of this kind is due to Thurston: Let H(R) be the Heisenberg group, that is, the
group of (3 3)-matrices of the form
A(x, y, z) =

¸
1 x z
0 1 y
0 0 1
¸

, x, y, z ∈ R,
and H(Z) be the closed subgroup of matrices in H(R) with x, y, z ∈ Z. The
natural projection H(R) →R
2
, A(x, y, z) →(x, y), descends to a projection of
the quotients
π: N := H(R)/H(Z) →R
2
/Z
2
,
a fiber bundle with fiber a circle R/Z. In particular, N is a closed manifold.
Since H(R) is diffeomorphic to R
3
, H(R) is simply connected. Therefore the
fundamental group of N is isomorphic to H(Z). It follows that the first homol-
ogy H
1
(N, Z) = H(Z)/[H(Z), H(Z)]

= Z
2
, and hence b
1
(N) = 2.
Let M := N S
1
, where the parameter of the factor S
1
is denoted t. In
this notation, the differential form ω = dx ∧ dt + dy ∧ dz is well defined and
symplectic on M. However, M does not carry a K¨ ahler metric since the first
Betti number of M is 3, an odd number. For more on this topic see [TO] and
the survey [BT]. Compare also with the Iwasawa manifold in Example 5.58.
70 Lectures on K¨ ahler Manifolds
3) There are closed complex manifolds satisfying h
p,q
(M, C) = h
q,p
(M, C).
For example, the Iwasawa manifold in Example 5.58 satisfies h
1,0
(M, C) = 3
andh
0,1
(M, C) = 2, see [Zh, Example 8.9]. This is one of the arguments that
show that this manifold does not carry a K¨ ahler metric.
On a closed oriented manifold M of real dimension 2m, the intersection
form S is a bilinear form on the middle cohomology,
S: H
m
(M, Z) H
m
(M, Z) →Z, S(x, y) = 'x ∪ y, [M]`, (5.38)
where [M] denotes the oriented fundamental cycle of M and the angle brackets
denote evaluation. Poincar´e duality implies that S is non-degenerate. If m
is odd, then S is skew-symmetric. If m is even, then S is symmetric. Thus
closed oriented manifolds with real dimension divisible by four come with a
remarkable topological invariant, a non-degenerate integral symmetric bilinear
form on their middle cohomology.
We can coarsen this picture and consider the intersection form on H
m
(M, R).
Via representing closed differential forms, it is then given by
S(ϕ, ψ) =

M
ϕ ∧ ψ. (5.39)
The index of S is called the signature of M, denoted σ(M).
In the case of a closed K¨ahler manifold, there is a variation T of S which is
defined on H
r
(M, R), 0 ≤ r ≤ m. Via representing closed differential forms,
T(ϕ, ψ) :=

M
ω
m−r
∧ ϕ ∧ ψ. (5.40)
We may also view T as a complex bilinear form on H
r
(M, C) = H
r
(M, R) ⊗C.
We have
T(ϕ, ψ) = (−1)
r
T(ψ, ϕ) and T(Cϕ, ψ) = T(ϕ, C

ψ). (5.41)
For the proof of the second equality, we note that since the volume form has
type (m, m), we have T(ϕ, ψ) = 0 if ϕ is of type (p, q) and ψ is of type (p

, q

)
with (p, q) = (q

, p

).
In our discussion below, we represent cohomology classes of M by their
harmonic representatives. It is important that the harmonic representative of
a primitive cohomology class is a differential form which is primitive at each
point of M.
5.42 Theorem (Hodge–Riemann Bilinear Relations). Let M be a closed K¨ahler
manifold. Let ϕ, ψ be primitive cohomology classes of type (p, q) and (p

, q

) and
s, t ≥ 0 satisfy p +q + 2s = p

+q

+ 2t = r ≤ m. Then
T(L
s
ϕ, L
t
ψ) = 0 unless s = t, p

= q, and q

= p; (1)
i
q−p
(−1)
(p+q)(p+q−1)
2
T(L
s
ϕ, L
s
ϕ) > 0 unless ϕ = 0. (2)
5 Cohomology of K¨ ahler Manifolds 71
Proof. If s = t, then (1) is immediate since the volume form has type (m, m).
Since ϕ is primitive, L
m−p−q+1
ϕ = 0. Now m−r + 2s = m−p − q. Hence if
s < t, then
ω
m−r
∧ L
s
ϕ ∧ L
t
ψ = L
m−r+s+t
ϕ ∧ ψ = 0.
The case s > t is similar. This completes the proof of (1).
The main and only point in the proof of (2) is a formula for the ∗-operator
on primitive differential forms. If s ≤ m−p −q, then
∗L
s
ϕ = (−1)
(p+q)(p+q−1)
2
s!
(m−p −q −s)!
L
m−p−q−s
Cϕ. (5.43)
For a proof see e.g. [Wei], Th´eor`eme 2 on page 23.
Since L

is a real operator, ϕ is primitive of degree (q, p). Hence, by (5.43),
i
q−p
T(L
s
ϕ, L
s
ϕ) =

M
ω
m−p−q
∧ Cϕ ∧ ϕ
= (−1)
(p+q)(p+q−1)
2
(m−p −q)!

M
∗ϕ ∧ ϕ
= (−1)
(p+q)(p+q−1)
2
(m−p −q)!

M
|ϕ|
2
.

We discuss a few applications of the Hodge–Riemann bilinear relations. Let
r be even. Then T is symmetric, see (5.41), hence the form T
H
(ϕ, ψ) = T(ϕ, ψ)
is Hermitian on H
r
(M, C). By Theorem 5.42, T
H
is non-degenerate and Hodge
and Lefschetz decomposition are orthogonal with respect to T
H
.
Let p, q, s ≥ 0 be given with p + q + 2s = r. Since r is even, p + q is even
as well, that is, p and q have the same parity. Hence (p +q)
2
is a multiple of 4
and therefore
(−1)
(p+q)(p+q−1)
2
i
q−p
= i
(p+q)(p+q−1)+q−p
= (−1)
p
= (−1)
q
.
From Theorem 5.42.2 we conclude that T
H
is positive definite on L
s
H
p,q
P
(M, C)
if p and q are even and negative definite if p and q are odd. For example, since
dimH
0,0
(M, C) = 1, the index of T
H
on H
1,1
(M, C) is (1, h
1,1
(M, C) − 1),
compare also the discussion of K¨ ahler surfaces below.
5.44 Hodge Index Theorem. Let M be a closed K¨ ahler manifold with even
complex dimension m. Then the signature
σ(M) =
¸
p+q even
(−1)
q
h
p,q
(M, C) =
¸
p,q
(−1)
q
h
p,q
(M, C).
Proof. By Theorem 5.35,
¸
p+q=k
(−1)
q
h
p,q
(M, C) = 0 if k is odd. Hence we
only need to prove the first equality. To that end, we note that by the above
discussion
σ(M) =
¸
p+q≤m
p,q even
h
p,q
P
(M, C) −
¸
p+q≤m
p,q odd
h
p,q
P
(M, C).
72 Lectures on K¨ ahler Manifolds
On the other hand, since p +q ≤ m,
h
p,q
P
(M, C) = h
p,q
(M, C) −h
p−1,q−1
(M, C),
see (5.34). This gives
σ(M) =
¸
p+q≤m
p+q even
(−1)
q
(h
p,q
(M, C) −h
p−1,q−1
(M, C))
The terms h
p,q
(M, C) with p +q even and < m occur twice and with the same
sign. Since h
p,q
(M, C) = h
m−p,m−q
(M, C) and p = m− p, q = m− q mod 2,
we conclude that
σ(M) =
¸
p+q even
(−1)
q
h
p,q
(M, C).
5.45 Remarks. Suppose that the complex dimension m of M is even.
1) If p + q = m then p = q modulo 2 so that we can substitute (−1)
p
for
(−1)
q
in Theorem 5.44.
2) For p fixed, χ
p
(M, C) :=
¸
q
(−1)
q
h
p,q
(M, C) is the Euler characteristic
of the cohomology of the Dolbeault chain complex
· · ·

−→ A
p,q−1
(M, C)

−→ A
p,q
(M, C)

−→ A
p,q+1
(M, C)

−→ · · ·
For p = 0, we get the arithmetic genus χ
0
(M, C) of M as in (2.31). The
Hirzebruch–Riemann–Roch formula expresses the numbers χ
p
(M, C) in terms
of the Chern classes of the tangent bundle TM, viewed as a complex vector
bundle. We get that the sum of these numbers is a topological invariant of M,
¸
χ
p
(M, C) = σ(M).
We point out a few other applications of the Hodge–Riemann bilinear re-
lations 5.42. Let M be a closed K¨ ahler manifold with K¨ ahler form ω. The
space H
1
(M, C) and the dimensions of H
1,0
(M, C) and H
0,1
(M, C) are topo-
logical invariants of M, but the inclusions of H
1,0
(M, C) and H
0,1
(M, C) into
H
1,0
(M, C) depend on the complex structure of M. By the Hodge–Riemann
bilinear relations, the form T is a symplectic form on H
1
(M, C) and the sub-
spaces H
1,0
(M, C) and H
0,1
(M, C) are Lagrangian subspaces; for the latter,
compare Example B.53. The associated Hermitian form T
H
(ϕ, ψ) := iT(ϕ, ψ)
is negative definite on H
1,0
(M, C) and positive definite on H
0,1
(M, C). Thus
we obtain a map, the period map, which associates to a complex structure on
M with a K¨ahler form cohomologous to ω the Lagrangian subspace H
1,0
(M, C)
in the period domain, the Hermitian symmetric space of Lagrangian subspaces
on which T
H
is negative definite, see again Example B.53. In other words, we
obtain a tool to study the space of complex structures on M together with a
K¨ ahler structure with K¨ ahler form cohomologous to ω. For example, let M be
a closed surface of genus g. Then H
1
(M, C)

= C
2g
and T is the intersection
5 Cohomology of K¨ ahler Manifolds 73
form on H
1
(M, C). By the Torelli theorem, the period map is a holomorphic
embedding of Teichm¨ uller space into G

C
(L, g) = Sp(g, R)/ U(g). Note that by
definition, the period map is equivariant with respect to the induced actions of
diffeomorphisms.
Suppose now that M is a closed K¨ahler surface, that is, dim
C
M = 2.
Then the Hodge numbers h
1,0
= h
1,0
(M, C), h
0,1
, h
2,1
, and h
1,2
are topological
invariants of M. We have
2h
2,0
+h
1,1
= b
2
(M) = b
2
and h
1,1
= h
1,1
P
+h
0,0
= h
1,1
P
+ 1.
By the Hodge–Riemann bilinear relations, the Hermitian form T
H
(ϕ, ψ) =
S(ϕ, ψ) on H
2
(M, C) is negative definite on H
1,1
P
(M, C) and positive definite on
the remaining summands in the Hodge decomposition of H
2
(M, C). Hence the
intersection form S of M has signature (b
2
−h
1,1
+1, h
1,1
−1) and the signature
of M is b
2
− 2h
1,1
+ 2. We conclude that the remaining Hodge numbers h
2,0
,
h
0,2
, and h
1,1
are invariants of the smooth structure of M. Furthermore, we
obtain again a period map, in this case from the space of complex structures
on M together with a K¨ ahler structure with K¨ ahler form cohomologous to ω to
the corresponding period domain, the Grassmannian of subspaces of H
2
(M, C)
of dimension h
1,1
P
on which T
H
is negative definite; for the latter, compare
Examples 4.10.6 and B.42. See [Wel, Section V.6], [Jo, Section 7.3], and [CMP]
for more on period maps, period domains, and Torelli theorems.
5.3 The dd
c
-Lemma and Formality. We assume throughout this subsec-
tion that D is flat, that is, that R
D
= 0. Then the connection form θ of D with
respect to a local holomorphic frame is holomorphic, see Proposition 3.21. It
follows that E has parallel local holomorphic frames. If (Φ
1
, . . . , Φ
k
) is such a
frame and σ = ϕ
µ
Φ
µ
a local smooth section of E, then d
D
σ = dϕ
µ
⊗ Φ
µ
and

D
σ = ∂ϕ
µ
⊗ Φ
µ
. Therefore we will omit the superindex D and simply write
d and ∂ instead of d
D
and ∂
D
. From Theorem 5.25 we conclude that

d
= 2∆

= 2∆

.
We introduce the complex differential
d
c
= i(∂ −∂) = C

dC = C
−1
dC, (5.46)
where C is as in (5.1). The two last expressions show that d
c
is real, that is,
that d
c
is well defined on the space A

(M, R) of real valued differential forms
on M. We have
d

c
= C

d

C = C
−1
d

C (5.47)
and
dd
c
= 2i∂∂ = −2i∂∂ = −d
c
d, (5.48)
74 Lectures on K¨ ahler Manifolds
where we use that D is flat. Let ∆
dc
= d
c
d

c
+ d

c
d
c
be the Laplacian of d
c
.
Since M is a K¨ ahler manifold, ∆
d
preserves the type of forms. Therefore

d
= C
−1

d
C = ∆
dc
. (5.49)
5.50 Lemma (dd
c
-Lemma). Suppose that M is closed and D is flat. Let α be
a differential form with values in E with
α = dβ and d
c
α = 0 or, respectively, dα = 0 and α = d
c
β.
Then there is a differential form γ with values in E such that α = dd
c
γ.
Proof. In what follows, δ denotes one of the operators d or d
c
. By (5.49), we
have ∆
d
= ∆
dc
=: ∆. Recall the eigenspace decomposition of the Hilbert space
of (equivalence classes of) square integrable sections,
L
2
(A

(M, E)) = ⊕V
λ
,
where V
λ
⊂ A

(M, E) is the eigenspace of ∆ for the eigenvalue λ. Since δ
commutes with ∆ on A

(M, E), we have δ(V
λ
) ⊂ V
λ
.
Let G: L
2
(A

(M, E)) →L
2
(A

(M, E)) be the Green’s operator associated
to ∆. Recall that
G(α) =

0 if α ∈ V
0
,
λ
−1
α if α ∈ V
λ
with λ > 0.
It follows that G and δ commute on A

(M, E). Since G is self-adjoint, we con-
clude that G and δ

commute as well on A

(M, E). Elliptic regularity implies
that G(A

(M, E)) ⊂ A

(M, E). This is also immediate from the displayed
formula.
The Hodge decomposition theorem asserts that we have an L
2
-orthogonal
sum
A

(M, E) = H

(M, E) ⊕δ(A

(M, E)) ⊕δ

(A

(M, E)),
where H

(M, E) = V
0
= ker ∆ is the space of δ-harmonic forms. Any differen-
tial form β with values in E satisfies
β = Hβ +∆Gβ,
where H denotes the L
2
-orthogonal projection onto H

(M, E). By assumption
and the Hodge decomposition,
Hα = 0
since α ∈ imδ. The latter also implies δα = 0, hence
α = ∆Gα = (δδ

+ δ

δ)Gα = δδ

Gα,
5 Cohomology of K¨ ahler Manifolds 75
where we use that G and δ commute on A

(M, E). In the first case we also
assume that d
c
α = 0, in the second that dα = 0, hence
α = d
c
d

c
Gα and α = dd


in the first and second case, respectively. In both cases we conclude
α = dd

Gd
c
d

c
Gα = dd

d
c
Gd

c
Gα = −dd
c
(d

Gd

c
Gα),
where we note that the commutation relation 5.24 implies that d

d
c
= −d
c
d

.

5.51 Exercises. 1) Calculate that, for functions ϕ,
'dd
c
ϕ, ω` = 2'i∂∂ϕ, ω` = −2g
jk

2
ϕ
∂z
j
∂z
k
= ∆ϕ.
This is a truly remarkable formula: It shows that, for K¨ahler manifolds, the
Laplacian of a function does not involve derivatives of the metric. The for-
mula implies that, for M closed and connected, the kernel of dd
c
: A
0
(M, C) →
A
2
(M, C) consists precisely of the constant functions.
2) If α is a differential form of type (p, q) and α = dβ or α = d
c
β, then
α = dd
c
γ.
3) Formulate and prove a ∂∂-Lemma.
What follows is taken from [DGMS]. Assume throughout that M is closed
and D is flat. Since dd
c
= −d
c
d, the subcomplex A

dc
(M, E) of A

(M, E)
consisting of d
c
-closed differential forms is closed under the exterior differential
d. Let H

dc
(M, E) be the cohomology of the complex (A

dc
(M, E), d).
5.52 Lemma. The inclusion A

dc
(M, E) →A

(M, E) induces an isomorphism
H

dc
(M, E)

=
−−→ H

(M, E).
Proof. We show first that the induced map on cohomology is surjective. Let α
be a closed differential form on M. Let β = d
c
α. Then dβ = −d
c
dα = 0, hence
β = dd
c
γ, by Lemma 5.50, and
d
c
(α +dγ) = β +d
c
dγ = β −dd
c
γ = 0.
Hence the cohomology class of α contains a representative in A

dc
(M, E).
Suppose now that d
c
α = 0 and α = dβ. Then α = d(d
c
γ), by Lemma 5.50.
Now d
c
(d
c
γ) = 0, hence α is a d-coboundary in A

dc
(M, E), and hence the
induced map on cohomology is injective.
Set now A

dc
(M, E) := A

dc
(M, E)/d
c
A

(M, E). Then d induces a differen-
tial d on A

dc
(M, E) and turns the latter into a chain complex. The cohomology
of the complex (A

dc
(M, E), d) will be denoted H

dc
(M, E).
76 Lectures on K¨ ahler Manifolds
5.53 Lemma. The differential d = 0 and the projection A

dc
(M, E) →A

dc
(M, E)
induces an isomorphism
H

dc
(M, E)

=
−−→ H

dc
(M, E).
Proof. Let α ∈ A

dc
(M, E) and set β = dα. Then d
c
β = −dd
c
α = 0, and hence
dα = β = d
c
dγ ∈ imd
c
, by Lemma 5.50. Hence d = 0.
Let α ∈ A

dc
(M, E). Then d
c
α = 0 and hence dα = d
c
dβ for some β. Let
γ = α +d
c
β. Then d
c
γ = 0, hence γ ∈ A

dc
(M, E), dγ = 0, hence γ represents
an element in H

dc
(M, E), and finally α = γ modulo d
c
A

(M, E).
Let now α ∈ A

dc
(M, E) with dα = 0 and such that α = d
c
β. Then α = dd
c
γ
for some γ, hence α is cohomologous to 0 in A

dc
(M, E).
For E the trivial line bundle, the chain complex (A
dc
(M, C), d) is invariant
under the wedge product and the chain complex (A

(M, C), d) therefore also
inherits a product, turning them into graded differential algebras over C, com-
mutative in the graded sense as in the case of differential forms. We use cgda
as a shorthand for commutative graded differential algebra.
A homomorphism between cgda’s A and B (over a field F) is called a quasi-
isomorphism if it induces an isomorphism of their cohomology rings. Then A
and B are called quasi-isomorphic, in letters A ≈ B. A cgda A is called formal
if there is a sequence B
0
= A, B
1
, . . . , B
k
of cgda’s over F such that B
i−1
≈ B
i
,
1 ≤ i ≤ k, and such that the differential of B
k
is trivial. For a closed K¨ahler
manifold M we have established that
A

(M, C) ←A

dc
(M, C) →A

dc
(M, C)/d
c
A

(M, C) (5.54)
are quasi-isomorphisms and that the differential of the latter is trivial.
5.55 Theorem ([DGMS]). If M is a closed K¨ahler manifold, then M is formal
over C, that is, A

(M, C) is a formal commutative differential graded algebra.

Since d
c
is a real operator, the proofs also work in the case of real numbers
and show that A

(M, R) is a formal cgda. In other words, M is also formal
over R.
5.56 Remark (See Remark 1.42). Let M be a closed Riemannian manifold
with curvature operator
ˆ
R ≥ 0. Then real valued harmonic forms on M are
parallel. Since the wedge product of parallel differential forms is parallel, hence
harmonic, we get that H

(M, R) together with wedge product and trivial dif-
ferential is a cgda over R. Moreover, the inclusion
H

(M, R) →A

(M, R)
induces an isomorphism in cohomology, see (1.36). Hence M is formal over R.
5 Cohomology of K¨ ahler Manifolds 77
We refer to [DGMS] for a discussion of the meaning of formality for the
topology of M. In a sense made precise in Theorem 3.3 and Corollary 3.4
in [DGMS], the R-homotopy type of a closed and simply connected K¨ ahler
manifold M is a formal consequence of the real cohomology ring of M.
To give at least one application of Theorem 5.55, we explain one of the
features which distinguish formal cgda’s from general ones. Let A be a cgda
and a, b, c ∈ H

(A) be cohomology classes of degree p, q and r, respectively,
such that a ∪ b = b ∪ c = 0. Let α, β, γ ∈ A be representatives of a, b, c,
respectively. Then α ∧ β = dϕ and β ∧ γ = dψ for ϕ, ψ ∈ A of degree p +q −1
and q +r −1, respectively. The Massey triple product
'a, b, c` := [ϕ ∧ γ −(−1)
p
α ∧ ψ] ∈ H
p+q+r−1
(A). (5.57)
We leave it as an exercise to the reader to show that the Massey triple product
is well defined up to α∧H
q+r−1
(A)+γ ∧H
p+q−1
(A), that it is preserved under
quasi-isomorphisms, and that it vanishes for cgda’s with trivial differential. We
conclude that Massey triple products vanish for formal cgda’s, in particular for
the real cohomology of closed K¨ ahler manifolds.
5.58 Example. Let G = H(C) be the complex Heisenberg group, that is, the
complex Lie group of matrices A(x, y, z) as in Example 5.37.2, but now with
x, y, z in C. Let Γ ⊂ H(C) be the discrete subgroup of A(x, y, z) with x, y, z
in R := Z + iZ. Since Γ acts by biholomorphic maps on G, the Iwasawa
manifold M = Γ\G, is a complex manifold of complex dimension 3. The
homomorphismG →C
2
, A(x, y, z) →(x, y), induces a holomorphic submersion
π: M →R
2
\C
2
with fibers biholomorphic to R\C. In particular, M is a closed
manifold and, more specifically, a torus bundle over a torus.
The differential (1, 0)-forms dx, dy, and dz − xdy on G are left-invariant,
hence they descend to differential (1, 0)-forms a, b, and ϕ on M. We have
da = db = 0 and dϕ = −a∧b. Let α = [a], β = [b] in H
1
(M, C). Since π induces
an isomorphism from H
1
(M, Z) = Γ/[Γ, Γ] = R
2
to H
1
(R
2
\C
2
, Z) = R
2
, we
conclude that (α, ¯ α, β,
¯
β) is a basis of H
1
(M, C). We have
'α, β, β` = [ϕ ∧ b].
Since dϕ = 0, it follows that 'α, β, β` = 0 modulo α∧H
1
(M, C)+β∧H
1
(M, C).
It follows that the Iwasawa manifold is not formal over C. Hence M does not
carry a K¨ ahler metric
14
. Note that the first Betti number of M is 4. It can be
shown that b
2
(M) = 8 and b
3
(M) = 10, so that the Betti numbers of M alone
do not show that M is not K¨ahlerian.
A nilmanifold is a quotient Γ\G, where G is a nilpotent Lie group and Γ is
a discrete subgroup of G. The Iwasawa manifold M in Example 5.58 is a nil-
manifold since H(C) is nilpotent. A theorem of Benson–Gordon and Hasegawa
14
For interesting other arguments, see [GH, page 444] and [Zh, Example 8.9]. See also
[FG].
78 Lectures on K¨ ahler Manifolds
says that a compact nilmanifold Γ\G does not admit a K¨ ahler structure un-
less G is Abelian, see [BeG], [Ha], [TO]. This generalizes our discussion in
Example 5.58.
5.4 Some Vanishing Theorems. Let M be a closed complex manifold. We
say that a differential form ω on M of type (1, 1) is positive and write ω > 0 if
g(X, Y ) := ω(X, JY ) (5.59)
is a Riemannian metric on M. If ω is closed and positive, then g turns M
into a K¨ ahler manifold with K¨ ahler form ω. We say that a cohomology class
c ∈ H
1,1
(M, C) is positive and write c > 0 if c has a positive representative. We
say that a holomorphic line bundle E → M is positive, denoted E > 0, if its
first Chern class c
1
(E) > 0. Negative differential forms and cohomology classes
of type (1, 1) and negative line bundles are defined correspondingly. We have
E > 0 iff the dual bundle E

< 0.
Let E → M be a holomorphic line bundle over M. Let h be a Hermitian
metric on E and D be the associated Chern connection. We note that the
curvature form Θ of D acts by scalar multiplication on sections and is indepen-
dent of the local trivialization. Thus Θ and R
D
are complex valued differential
forms on M. In this sense we have Θ = R
D
.
5.60 Examples. We continue the discussion of Examples 3.45 and consider
some line bundles over CP
m
. As before, we will use the open covering of CP
m
by the sets U
j
= {[z] ∈ CP
m
| z
j
= 0}, 0 ≤ j ≤ m.
1) The tautological bundle U → CP
m
. In the notation of Example 3.45.1,
the canonical Hermitian metric h on U is given by
h(ϕ
j
, ϕ
j
) =
1
z
j
z
j
¸
z
k
z
k
.
In the point z ∈ U
j
with coordinates z
j
= 1 and z
k
= 0 for k = j, the curvature
of the Chern connection D of h is
Θ = ∂∂ ln h(ϕ
j
, ϕ
j
) =
¸
k=j
∂∂(z
k
z
k
) = −
¸
k=j
dz
k
∧ dz
k
,
as we already checked in Example 3.24 in greater generality. Hence, in the
above point z,
c
1
(U, D) = −
i

¸
k=j
dz
k
∧ dz
k
< 0.
Now the homogeneity of U and h under the canonical action of the unitary
group U(m+ 1) implies that c
1
(U, D) < 0 everywhere. Hence U is negative.
2) The canonical line bundle K = A
m,0
(CP
m
, C) → CP
m
: We saw in
Example 3.45.2 that K is isomorphic to U
m+1
. For the curvature of the Chern
5 Cohomology of K¨ ahler Manifolds 79
connection of the canonical metric on K we get (m + 1)Θ, where Θ is the
curvature of U with respect to the Chern connection of its canonical Hermitian
metric. By the previous example, we have i(m+ 1)Θ < 0, hence K < 0.
More generally, for a K¨ ahler manifold M with Ricci form ρ, the curvature of
the canonical bundle K
M
is given by iρ, see (4.67). Since the canonical metric
on CP
m
has positive Ricci curvature, this confirms i(m+ 1)Θ = −ρ < 0.
3) The hyperplane bundle H → CP
m
: H is inverse to U, that is, H is
isomorphic to the dual line bundle U

. It follows that H is positive.
5.61 Lemma. Let E →M be a holomorphic line bundle. Then E > 0 iff there
is a Hermitian metric h on E such that the curvature Θ of its Chern connection
satisfies iΘ > 0, and similarly in the case E < 0.
Proof. Since 2πc
1
(E) = [iΘ], E is positive if E has a Hermitian metric with
iΘ > 0. Suppose now that E > 0. Choose a K¨ahler metric g on M with K¨ ahler
form ω ∈ 2πc
1
(E). This is possible since c
1
(E) > 0. Let h be any Hermitian
metric on E, Θ be the curvature form of its Chern connection, and ϕ be a local
and nowhere vanishing holomorphic section of E (a local frame of E). Then
Θ = ∂∂ ln h
ϕ
,
where h
ϕ
= h(ϕ, ϕ). In particular, ∂Θ = ∂Θ = 0. We also have ∂ω = ∂ω = 0.
Now [iΘ] = 2πc
1
(E) = [ω] by the definition of c
1
(M) and the choice of ω. Since
M is a closed K¨ahler manifold, the dd
c
-Lemma 5.50 applies and shows that
ω −iΘ = i∂∂ψ
for some real function ψ. Let h

= h · e
ψ
. Then the curvature Θ

of the Chern
connection of h

satisfies

= i∂∂ ln h

= iΘ+ ω −iΘ = ω > 0.
5.62 Proposition. Let E → M be a holomorphic line bundle and h be a
Hermitian metric on M such that the curvature Θ of its Chern connection D
satisfies iλΘ > 0 for some λ ∈ R. Then ω := iλΘ is the K¨ ahler form of a
K¨ ahler metric on M, and with respect to this K¨ahler metric


−∆

D =
1
λ
¸
(r −m)P
r
.
Proof. By Theorem 5.25 and Proposition 5.8, we have


−∆

D = [iR
D

ε
, L

] =
1
λ
[ω∧, L

] =
1
λ
[L, L

] =
1
λ
¸
(r −m)P
r
.
5.63 Remark. Proposition 5.62 refines the Nakano inequalities, which usually
go into the proof of Kodaira’s vanishing theorem 5.64, compare [Wei, Proposi-
tion VI.2.5]. We note that these inequalities are special cases of the last two
(groups of) equations in the proof of Theorem 5.25. In [Mo], Moroianu proves
Proposition 5.62 by a direct computation using a Weitzenb¨ ock formula.
80 Lectures on K¨ ahler Manifolds
5.64 Kodaira Vanishing Theorem. Let M be a closed complex manifold
and E →M be a holomorphic line bundle.
1. If E > 0, then H
p,q
(M, E) = 0 for p +q > m.
2. If E < 0, then H
p,q
(M, E) = 0 for p +q < m.
Proof. In view of Serre duality 3.12, the two assertions are equivalent since
E > 0 iff E

< 0. We discuss the second assertion (and a similar argument
shows the first). Let h be a Hermitian metric on E such that the curvature Θ
of its Chern connection satisfies iΘ < 0. Let g be the K¨ ahler metric on M with
K¨ ahler form ω = −iΘ. Let α be a ∆

-harmonic form with values in E and of
type (p, q). Then, by Proposition 5.62,
0 ≤ (m−p −q)||α||
2
2
= −(∆

Dα, α)
2
≤ 0.
5.65 Corollary. Let M be a closed complex manifold and E → M be a holo-
morphic line bundle with K

⊗E > 0. Then H
0,q
(M, E) = 0 for q ≥ 1.
This will be important in the proof of the Kodaira embedding theorem 9.6.
Another application: If K

⊗E > 0, then
χ
0
(M, E) :=
¸
(−1)
q
dimH
0,q
(M, E) = dimH
0,0
(M, E),
the dimension of the space of holomorphic sections of E. Hence if K

⊗E > 0,
then the Hirzebruch–Riemann–Roch formula [LM, Theorem III.13.15] com-
putes the dimension of H
0,0
(M, E).
Proof of Corollary 5.65. We note first that a differential form of type (m, q)
with values in a holomorphic bundle L is the same as a form of type (0, q) in
the bundle K ⊗L. Therefore,
H
0,q
(M, E) = H
0,q
(M, K ⊗K

⊗E) = H
m,q
(M, K

⊗ E) = 0
by Theorem 5.64 since m+q > m.
Let M be a K¨ ahler manifold and E →M be a holomorphic line bundle. Let
h be a Hermitian metric on E and D be the corresponding Chern connection.
Let p ∈ M. Then there exist an orthonormal frame (X
1
, Y
1
, . . . , X
m
, Y
m
) with
Y
j
= JX
j
in (or about) p and real numbers θ
1
, . . . , θ
m
such that the curvature
of D at p is given by
R
D
p
=
1
2
¸
θ
j
Z

j
∧ Z

j
, (5.66)
where the frame (Z
1
, Z
1
, . . . , Z
m
, Z
m
) is as in (3.30) and (Z

1
, Z

1
, . . . , Z

m
, Z

m
)
is the corresponding dual frame. For a local section σ of E about p and a
5 Cohomology of K¨ ahler Manifolds 81
differential form α = Z

J
∧ Z
K
⊗σ, we have
[Z

j
∧ Z

j
∧, L

]α =
¸
k
¸
Z

j
∧ (Z

j
∧ (Z
k
(Z
k
α))) −Z
k
(Z
k
(Z

j
∧ (Z

j
∧ α)))
¸
= Z

j
∧ (Z

j
∧ (Z
j
(Z
j
α))) −Z
j
(Z
j
(Z

j
∧ (Z

j
∧ α)))
=

α if j ∈ J ∩ K;
−α if j / ∈ J ∪ K;
0 otherwise.
(5.67)
It follows that for α as above
[iR
D
p

ε
, L

]α =
1
2

¸
j∈J∩K
θ
j

¸
j / ∈J∪K
θ
j

α. (5.68)
This refines Proposition 5.8.
We interpret (the proof of) Theorem 5.64 as the case of “constant curvature”
R
D
= ∓iω. Theorem 5.25 gives us some room for extensions to pinching
conditions. We prove a simple model result in this direction.
5.69 Theorem. Let M be a closed K¨ ahler manifold with K¨ ahler form ω. Let
E → M be a holomorphic line bundle and h be a Hermitian metric on E with
Chern connection D. Let κ: M →R and ε > 0 and suppose that
|R
D
(X, Y ) +iκω(X, Y )| < ε|X| |Y |
for all vector fields X and Y . Then H
p,q
(M, E) = 0 for all p, q with
ε · (m−|p −q|) < κ · (p +q −m).
We note that there are two cases: κ > 0 for p + q > m and κ < 0 for
p +q < m.
Proof of Theorem 5.69. Let p
0
∈ M. The condition on R
D
implies that the
numbers θ
j
as in (5.66) satisfy |θ
j
− κ(p
0
)| < ε for all j. Let I, J be multi-
indices with |I| = p and |J| = q, and set a = |I ∩ J|. Then

¸
j∈J∩K
θ
j

¸
j / ∈J∪K
θ
j

≥ a(κ(p
0
) −ε) −(m−p −q +a)(κ(p
0
) + ε)
≥ (p +q −m)κ(p
0
) −(m−|p −q|)ε > 0.
Hence ([iR
D
, L

]α, α)
2
> 0 unless α = 0. The rest of the argument is as in
Theorem 5.64.
For more on vanishing theorems, see [Ko2] and [ShS].
6 Ricci Curvature and Global Structure
Let M be a closed Riemannian manifold. Recall that the isometry group of M
is a compact Lie group (with respect to the compact-open topology) with Lie
algebra isomorphic to the Lie algebra of Killing fields on M (with sign of the
Lie bracket reversed).
6.1 Theorem (Bochner [Boc]). Let M be a closed Riemannian manifold with
Ric ≤ 0 and X be a vector field on M. Then X is a Killing field iff X is
parallel.
Proof. It is clear that parallel fields are Killing fields. Vice versa, let X be a
Killing field on M. By Proposition 4.77, we have ∇

∇X = Ric X. Therefore
0 ≤

|∇X|
2
=

'∇

∇X, X` =

'Ric X, X` ≤ 0,
and hence X is parallel.
6.2 Corollary. Let M be a closed and connected Riemannian manifold with
Ric ≤ 0. Then the identity component of the isometry group of M is a torus
of dimension k ≤ dimM, and its orbits foliate M into a parallel family of flat
tori. If M is simply connected or if Ric < 0 at some point of M, then the
isometry group of M is finite.
Proof. The first assertion is clear from Theorem 6.1 since the Lie algebra of the
isometry group is given by the Lie algebra of Killing fields. As for the second
assertion, it suffices to show that M does not carry a non-trivial Killing field.
Let X be such a field. By Theorem 6.1, X is parallel. If M is simply connected,
then X points in the direction of a Euclidean factor R of M, by the de Rham
decomposition theorem. This is in contradiction to the compactness of M. In
any case we note that R(· , X)X = 0 since X is parallel. This implies Ric X = 0
and is in contradiction to Ric < 0.
What we need in the proof of Theorem 6.1 is the inequality Ric ≤ 0 together
with the differential equation ∇

∇X = Ric X. Recall now that automorphic
vector fields on closed K¨ ahler manifolds are characterized by the latter equation,
see Proposition 4.79. Hence we have the following analogue of Theorem 6.1.
6.3 Theorem. Let M be a closed K¨ ahler manifold with Ric ≤ 0 and X be a
vector field on M. Then X is automorphic iff X is parallel.
We recall now again that the automorphism group of a closed complex
manifold M is a complex Lie group with respect to the compact-open topology
with Lie algebra isomorphic to the Lie algebra of automorphic vector fields on
M (again and for the same reason with sign of the Lie bracket reversed), see
[BM].
6 Ricci Curvature and Global Structure 83
6.4 Corollary. Let M be a closed K¨ ahler manifold with Ric ≤ 0. Then the
Lie algebras of parallel, Killing, and automorphic vector fields coincide. The
identity components of the automorphism group and of the isometry group of
M coincide and are complex tori of complex dimension ≤ dim
C
M.
6.5 Remark. By a famous result of Lohkamp [Loh], all smooth manifolds of
dimension ≥ 3 admit complete Riemannian metrics of strictly negative Ricci
curvature. Corollary 6.4 shows that, on closed complex manifolds, there are
obstructions against K¨ ahler metrics of non-positive Ricci curvature.
6.1 Ricci-Flat K¨ahler Manifolds. A celebrated theorem of Cheeger and
Gromoll says that the universal covering space
˜
M of a closed and connected
Riemannian manifold M with Ric ≥ 0 splits isometrically as
˜
M = EN, where
E is a Euclidean space and N is closed and simply connected with Ric ≥ 0, see
[CG1]. The following application is given in [Bea].
6.6 Theorem. Let M be a closed and connected K¨ahler manifold with Ric = 0.
Then there is a finite covering
ˆ
M of M which splits isometrically as
ˆ
M = FN,
where F is a closed flat K¨ahler manifold and N is a closed and simply connected
K¨ ahler manifold with Ric = 0.
6.7 Remark. By the Bieberbach theorem, a closed flat Riemannian manifold
is finitely covered by a flat torus [Wo, Chapter 3].
Proof of Theorem 6.6. According to the de Rham decomposition theorem for
K¨ ahler manifolds 4.76 and the theorem of Cheeger and Gromoll mentioned
above, the universal covering space
˜
M of M decomposes as
˜
M = C
k
N, k ≥ 0,
where N is a closed and simply connected K¨ ahler manifold with Ric = 0. Hence
the isometry group I(
˜
M) of
˜
M is the product of the isometry group of C
k
and
the isometry group I(N) of N. By Corollary 6.2, I(N) is finite.
The fundamental group Γ of M acts isometrically as a group of covering
transformations on
˜
M, hence is a subgroup of I(
˜
M). Consider the projection
of Γ to the factor I(N) of I(
˜
M). Since I(N) is finite, the kernel
ˆ
Γ has finite
index in Γ and gives a finite cover of M as asserted.
6.8 Remark. Let (M, J) be a closed complex manifold with c
1
(M) = 0. Then
for each K¨ ahler form ω of M, there is exactly one Ricci-flat K¨ ahler metric g

on M with K¨ ahler form ω

cohomologous to ω, by the Calabi conjecture 7.1.
Therefore the space of Ricci-flat K¨ ahler metrics on M can be identified with
the cone of K¨ahler forms in H
1,1
(M, R) = H
1,1
(M, C) ∩ H
2
(M, R), an open
subset of H
1,1
(M, R), hence of real dimension h
1,1
(M, C) if non-empty
15
. A
reference for this are Subsection 4.9.2 and Section 6.8 in [Jo], where Joyce
discusses deformations of complex and Calabi–Yau manifolds.
15
We do not elaborate on the topology of spaces of metrics.
84 Lectures on K¨ ahler Manifolds
6.2 Nonnegative Ricci Curvature. Let M be a closed K¨ ahler manifold and
ϕ be a form of type (p, 0). Let (X
1
, Y
1
, . . . , X
m
, Y
m
) be a local orthonormal
frame of M such that JX
j
= Y
j
, and define Z
j
and Z
j
as in (3.30). Using that
ϕ is of type (p, 0) and the formulas (3.32) for ∂ and (3.33) for ∂

, we get

d
ϕ = 2∆

ϕ = 2∂

∂ϕ = −4
¸
Z
j
Z

k

ˆ
D
2
ϕ(Z
j
, Z
k
).
Now D
2
ϕ(Z
j
, Z
k
) is still of type (p, 0), hence the right hand side is equal to
−4
¸
ˆ
D
2
ϕ(Z
j
, Z
j
) = −tr
ˆ
D
2
ϕ −i
¸
ˆ
R(X
j
, Y
j
)ϕ,
where we use Exercise 3.37.2 for the latter equality. In conclusion,

d
ϕ = 2∂

∂ϕ =
ˆ


ˆ
∇ϕ +Kϕ, (6.9)
with curvature term K = −i
¸
ˆ
R(X
j
, Y
j
). Now the curvature acts as a deriva-
tion on A

(M, C). Hence to compute its action on A
p,0
(M, C), p > 0, we first
determine its action on forms of type (1, 0). For X ∈ T

M, the dual vector
X

= 'X, · ` is of type (1, 0). For Y ∈ T

M, we get
'KX

, Y

= −i
¸
'
ˆ
R(X
j
, Y
j
)X

, Y

` = −i
¸
'R(X
j
, Y
j
)X, Y `
= −
¸
'R(X
j
, Y
j
)X, JY ` = Ric(X, Y ),
where we recall that '· , · ` denotes the complex bilinear extension of the Rie-
mannian metric of M to the complexified tangent bundle T
C
M. Now at a given
point in M, we may choose the above frame (X
1
, Y
1
, . . . , X
m
, Y
m
) to consist of
eigenvectors of the Ricci tensor, considered as an endomorphism. For multi-
indices J : j
1
< · · · < j
p
and K: k
1
< · · · < k
p
, we then have, at the chosen
point,
'K(Z

j1
∧· · · ∧Z

jp
), Z

k1
∧· · · ∧Z

kp
` =

4
p
¸
Ric(Z

, Z

) if J = K,
0 otherwise,
(6.10)
see (3.31). Hence the curvature term K in the above equation is non-negative
if Ric is non-negative and positive if Ric is positive.
6.11 Theorem (Bochner [Boc]). Let M be a closed, connected K¨ahler manifold
with Ric ≥ 0. Then holomorphic forms of type (p, 0) are parallel; in particular,
h
p,0
(M, C) = h
0,p
(M, C) ≤

m
p

.
If Ric > 0 at some point of M, then M does not have non-trivial holomorphic
forms of type (p, 0) for p > 0, and then h
p,0
(M, C) = h
0,p
(M, C) = 0 for p > 0.
6 Ricci Curvature and Global Structure 85
Proof. If ϕ is a holomorphic form of type (p, 0), p > 0, then ∆
d
ϕ = 2∂

∂ϕ = 0.
Hence
0 =

M
'
ˆ


ˆ
∇ϕ, ϕ` +

M
'Kϕ, ϕ` =

M
|
ˆ
∇ϕ|
2
+

M
'Kϕ, ϕ`.
As we just explained, 'Kϕ, ϕ` ≥ 0 if Ric ≥ 0.
The last assertion in Theorem 6.11 also follows from the Kodaira vanishing
theorem. To see this we note that Ric > 0 implies that the anti-canonical
bundle K

M
→M is positive, see (4.68). On the other hand, if K

M
is positive
and E → M is the trivial complex line bundle or, generalizing Definition 5.59
slightly, a non-negative line bundle, then K

M
⊗ E is positive, and we may
apply Corollary 5.65 to conclude that h
0,q
(M, E) = 0 for q ≥ 1. Observe that
Bochner’s argument above works in this case as well (Exercise).
6.12 Theorem (Kobayashi [Ko1]). Let M be a closed, connected K¨ahler mani-
fold with Ric > 0. Then M is simply connected and the arithmetic genus
χ(M, O) =
¸
(−1)
p
h
0,p
(M, C) = 1.
Proof. Consider the universal covering π:
˜
M → M. With respect to the in-
duced K¨ ahler structure on
˜
M, π is a local isometry. Since the Ricci curvature of
M is positive, the fundamental group of M is finite, by the theorem of Bonnet–
Myers. Hence
˜
M is closed as well. Now the Hirzebruch–Riemann–Roch formula
expresses the arithmetic genus of a closed K¨ahler manifold as an integral over
a universal polynomial in the curvature tensor. Since π is a local isometry, this
implies
χ(
˜
M, O) = |Γ| · χ(M, O),
where Γ denotes the fundamental group of M. By Theorem 6.11,
χ(
˜
M, O) = χ(M, O) = 1,
hence |Γ| = 1, and hence M is simply connected.
In his article [Ko1], Kobayashi observed the application of the Kodaira van-
ishing theorem mentioned before Theorem 6.12 and conjectured that closed
and connected K¨ ahler manifolds with positive first Chern class are simply con-
nected, compare (4.68). Now the Calabi conjecture implies that closed K¨ahler
manifolds with positive first Chern class carry K¨ahler metrics with positive
Ricci curvature, see Corollary 7.3. Hence they are in fact simply connected, by
Kobayashi’s original Theorem 6.12.
86 Lectures on K¨ ahler Manifolds
6.3 Ricci Curvature and Laplace Operator. Let M be a closed, con-
nected Riemannian manifold of real dimension n. Denote by ∆ the Laplace
operator on functions on M. If Ric ≥ λ > 0, then the first non-zero eigenvalue
λ
1
of ∆ satisfies
λ
1

n
n −1
λ. (6.13)
Moreover, equality holds iff M is a round sphere of radius

(n −1)/λ. The
inequality (6.13) is due to Lichnerowicz [Li3], the equality discussion to Obata
[Ob]; see also [BGM, Section III.D]. The next theorem deals with a remarkable
improvement of the above inequality in the K¨ ahler case.
6.14 Theorem (Lichnerowicz). Let M be a closed K¨ ahler manifold with Ric ≥
λ > 0. Then the first non-zero eigenvalue λ
1
of ∆ satisfies
λ
1
≥ 2λ.
Equality implies that the gradient field X = gradϕ of any eigenfunction ϕ for
λ
1
is automorphic with Ric X = λX.
Proof. Let ϕ ∈ E(M, R) be any non-constant eigenfunction, so that ∆ϕ = µϕ
for some µ > 0. Then ξ := dϕ satisfies ∆
d
ξ = µξ. Hence, by the Bochner
identity 1.38, X := ξ

= gradϕ = 0 fulfills


∇X = (µ −2λ)X + (2λX −Ric X).
In the notation introduced there, (4.80) implies



0,1
X =
1
2
(µ −2λ)X + (λX −Ric X).
Therefore, by the assumption Ric ≥ λ,
0 ≤ |∇
0,1
X|
2
2
= '∇
0,1
X, ∇X`
2
= '∇


0,1
X, X`
2

1
2
(µ −2λ)|X|
2
2
,
where the index 2 indicates L
2
-inner products and norms. We conclude that
µ ≥ 2λ for all non-trivial eigenvalues µ of ∆. Equality implies that ∇
0,1
X = 0,
that is, X is automorphic, and that Ric X = λX.
6.15 Exercise. Discuss Theorem 6.14 in the case of the round S
2
(and observe
thatn/(n −1) < 2 if n > 2). Compare with Theorem 6.16.
Let M be a closed K¨ ahler manifold. Let a be the complex Lie algebra of
automorphic vector fields on M and k ⊂ a be the real Lie subalgebra of Killing
fields. The next theorem completes the equality discussion in Theorem 6.14 in
the case where M is a K¨ ahler–Einstein manifold.
6 Ricci Curvature and Global Structure 87
6.16 Theorem (Matsushima). Let M be a closed K¨ ahler–Einstein manifold
with Einstein constant λ > 0. Then we have an isomorphism
grad: {ϕ ∈ E(M, R) | ∆ϕ = 2λϕ} =: E

→Jk, ϕ →gradϕ.
Proof. We note first that grad is injective on the space of smooth functions on
M with mean zero.
By the proof of Theorem 7.43 below, Jk consists precisely of automorphic
gradient vector fields. Hence by the equality case in Theorem 6.14, grad maps
E

to Jk. By the above remark, grad is injective on E

.
Vice versa, let X in Jk. Write X = gradϕ, where ϕ is a smooth function
on M with mean zero. Now X is automorphic, hence ∇

∇X = Ric X = λX,
by Proposition 4.79. Hence ∆ϕ = 2λϕ, by Exercise 6.17 and since ϕ and ∆ϕ
have mean zero.
6.17 Exercise (Ricci Equation). Let M be a Riemannian manifold and ϕ be
a smooth function on M. Show that
grad∆ϕ = (∇

∇)(gradϕ) + Ric gradϕ.
We recall that, on functions, ∆ = ∇

∇, where we write ∇ instead of grad.
Thus the formula computes the commutator [∇, ∇

∇]. Derive an analogous
formula for metric connections on vector bundles.
7 Calabi Conjecture
In this section we discuss the Calabi conjecture and present a major part of its
proof
16
. We let M be a closed and connected K¨ ahler manifold with complex
structure J, Riemannian metric g, K¨ ahler form ω, and Ricci form ρ. Recall
that the Ricci form of any K¨ ahler metric on M is contained in 2πc
1
(M).
7.1 Theorem (Calabi–Yau). Let ρ

∈ 2πc
1
(M) be a closed real (1, 1)-form.
Then there is a unique K¨ ahler metric g

on M with K¨ ahler form ω

cohomolo-
gous to ω and with Ricci form ρ

.
This was conjectured by Calabi, who also proved uniqueness of the solution
g

[Ca1], [Ca2]. Existence was proved by Yau [Ya2], [Ya3], [Ka1]. One of the
most striking immediate applications of Theorem 7.1 concerns the existence of
Ricci-flat K¨ ahler metrics:
7.2 Corollary. If c
1
(M) = 0, then M has a unique Ricci-flat K¨ ahler metric
g

with K¨ ahler form ω

cohomologous to ω.
As a consequence, M is finitely covered by a product of a simply connected
closed complex manifold with vanishing first Chern class and a complex torus
(where one of the factors might be of dimension 0), see Theorem 6.6.
Recall Definition 5.59 of positive and negative cohomology classes of type
(1, 1). Say that a closed complex manifold M is a Fano manifold if c
1
(M) is
positive or, equivalently, if the anti-canonical bundle K

M
is ample
17
. Since
their first Chern class is represented by their Ricci form, closed K¨ ahler mani-
folds with positive Ricci curvature are Fano manifolds. For example, Hermitian
symmetric spaces of compact type are Fano manifolds. Theorem 7.1 implies
that Fano manifolds are characterized by positive Ricci curvature:
7.3 Corollary. If M is a Fano manifold, then M carries K¨ahler metrics with
positive Ricci curvature.
Proof. By assumption, M has a positive closed differential form ω of type (1, 1).
Then g(X, Y ) = ω(X, JY ) is a K¨ ahler metric on M with K¨ ahler form ω. Choose
a further positive differential form ρ

∈ 2πc
1
(M), for example ρ

= ω. By
Theorem 7.1, there is a unique K¨ ahler metric g

with K¨ ahler form cohomologous
to ω and Ricci form ρ

. Hence the Ricci curvature of g

is positive.
The argument for Corollary 7.3 works equally well in the case c
1
(M) < 0.
In this case, however, we have the stronger Theorem 7.14 below.
7.4 Examples. 1) Let M ⊂ CP
n
be the (regular) set of zeros of a homogeneous
polynomial of degree d. Then c
1
(M) = (n + 1 −d)c, where c is the restriction
16
My main sources for this are [Au3] and [Jo]. Other good references are [Bo1], [Si1], [Ti3].
17
For the equivalence, see Exercise 9.5 and the Kodaira Embedding Theorem 9.6.
7 Calabi Conjecture 89
of the first Chern class of the hyperplane bundle H over CP
n
to M, see [Hir,
p.159]. By Example 5.60.3, c > 0. We conclude that c
1
(M) > 0 if d ≤ n,
c
1
(M) = 0 if d = n + 1, and c
1
(M) < 0 if d > n + 1.
2) Let M
k
be the blow up of CP
2
in k ≥ 0 points in (very) general position.
Then c
1
(M
k
) > 0 for 0 ≤ k ≤ 8. Moreover, besides CP
1
CP
1
, these are the
only closed complex surfaces with positive first Chern class [Hit], [Ya1]. The
rich geometry of these surfaces is discussed in [Va, Lectures 13–15].
From these examples one gets the impression that Fano manifolds might be
rare and special. In fact, from Corollary 7.3 and Theorems 6.11 and 6.12 we
conclude that Fano manifolds are simply-connected and that they do not admit
any holomorphic p-forms for p > 0. Moreover, for each m ≥ 1, there are only
finitely many diffeomorphism types of Fano manifolds of complex dimension m,
for example 104 for m = 3, see [De].
Before we go into the details of the proof of Theorem 7.1, we reformulate
it into a more convenient analytic problem and discuss another question, the
existence of K¨ahler–Einstein metrics.
Recall that ω
m
= m! vol
g
, where vol
g
denotes the oriented volume form of
g, see (4.20). Hence if ω

is cohomologous to ω, then g

and g have the same
volume. For such an ω

given, there is a smooth function f : M →R with

)
m
= e
f
ω
m
. (7.5)
Then the volume constraint turns into

e
f
d vol
g
= vol(M, g). (7.6)
In other words, e
f
has mean 1. With (4.64) we get
ρ

= −i∂∂ ln det G

= −i∂∂ ln(e
f
det G) = ρ −i∂∂f. (7.7)
Vice versa, for any closed real (1, 1)-form ρ

cohomologous to ρ, there is a
smooth function f : M →R such that
ρ

−ρ = −
1
2
dd
c
f = −i∂∂f, (7.8)
by the dd
c
-Lemma 5.50 and Exercise 5.51. Moreover, f is unique up to an
additive constant. Thus in our search for the K¨ ahler form ω

we have replaced
the Ricci form ρ

, which depends on second derivatives of the metric, by the
function f, which depends only on ω

itself.
For any K¨ ahler form ω

cohomologous to ω, there is a smooth real function
ϕ such that
ω

−ω =
1
2
dd
c
ϕ = i∂∂ϕ, (7.9)
90 Lectures on K¨ ahler Manifolds
again by the dd
c
-Lemma 5.50. Vice versa, any such form ω

is cohomologous
to ω. Thus we seek ϕ with
e
f
ω
m
= (ω

)
m
= (ω +i∂∂ϕ)
m
(7.10)
and
ω

= ω +i∂∂ϕ > 0. (7.11)
By the latter we mean that g

(X, Y ) = ω

(X, JY ) is a K¨ahler metric on M.
In a first step we show that ω

> 0 if (7.10) holds. Consider the Hermitian
metric on T

M given by
(Z, W) = 'Z, W`,
where we note that this Hermitian metric is conjugate linear in the second
variable. With respect to a coordinate frame (Z
1
, . . . , Z
m
) of T

M as in (2.13),
its fundamental matrix is (g
jk
). We write the Hermitian metric (· , · )

on T

M
induced by g

as (· , · )

= (A· , · ), where A is a Hermitian field of endomorphisms
of T

M. Since e
f
is non-zero everywhere, (7.10) and (4.20) imply that det A =
e
f
= 0 at each point of M. We want to show that at each point of M all
eigenvalues of A are positive. To that end we note that with respect to local
coordinates
M(ϕ) := e
f
= det A = det(g
jk
)
−1
det

g
jk
+

2
ϕ
∂z
j
∂z
k
!
. (7.12)
It follows that all eigenvalues of A are positive at a point where ϕ attains a
minimum. Such points exist since M is compact. Suppose now that there is a
point p ∈ M such that A has a negative eigenvalue at p. Then on a path from
p to a point where ϕ attains a minimum, there is a point where A has 0 as
an eigenvalue. In such a point, det A = 0 in contradiction to (7.10). We have
achieved the following reformulation of Theorem 7.1:
Given a smooth function f on M such that e
f
has mean 1, there is a smooth
real function ϕ on M such that
ln M(ϕ) = f. (7.13)
This is a non-linear partial differential equation of Monge–Amp`ere type.
The solution ϕ of (7.13) is unique up to constant functions, see Proposition 7.25.
Since ω

= ω + i∂∂ϕ is cohomologous to ω, the assumption on the mean of
e
f
is automatically fulfilled. In other words, we may delete the normalizing
assumption on the mean and ask for a solution of (7.13) up to an additive
constant f +c instead. This point of view will be useful further on.
Before we discuss the proof of Assertion 7.13, we discuss a related problem,
namely the question of the existence of K¨ahler–Einstein metrics.
7 Calabi Conjecture 91
7.14 Theorem (Aubin–Calabi–Yau). Let M be a closed complex manifold
with negative first Chern class. Then up to a scaling constant, M has a unique
K¨ ahler–Einstein metric (with negative Einstein constant ).
Theorem 7.14 was conjectured by Calabi and proved independently by
Aubin [Au2] and Yau [Ya2], [Ya3]. The case of positive first Chern class is
complicated, we discuss it briefly in Subsection 7.4 below. The case of Ricci-
flat K¨ ahler metrics is treated in Corollary 7.2.
We now reformulate Theorem 7.14 into an analytic problem similar to (7.13)
above. Suppose c
1
(M) < 0 and choose a constant λ < 0. Let ω be a positive real
differential form of type (1, 1) with λω ∈ 2πc
1
(M). Then g(X, Y ) = ω(X, JY )
is a K¨ ahler metric on M with K¨ ahler form ω.
Let g

be a K¨ ahler–Einstein metric on M with K¨ ahler form ω

and Einstein
constant λ. Then λω

= ρ

∈ 2πc
1
(M), hence ω

is cohomologous to ω, and
therefore ω

−ω = i∂∂ϕ. Since the Ricci form ρ ∈ 2πc
1
(M), there is a smooth
real function f on M, unique up to an additive constant, such that ρ − λω =
i∂∂f. We want to solve ρ

= λω

. The latter gives
−i∂∂ ln M(ϕ) = ρ

−ρ = λ(ω

−ω) −i∂∂f = λi∂∂ϕ −i∂∂f, (7.15)
and hence, by Exercise 5.51,
ln M(ϕ) = −λϕ +f + const. (7.16)
Replacing ϕ by const/λ+ϕ, the constant on the right side vanishes. Vice versa,
suppose ϕ solves (7.16). Then the above computation shows that ω

= ω+i∂∂ϕ
solves ρ

= λω

. We explain after Equation 7.18 below why ω+i∂∂ϕ is positive.
In conclusion, Theorem 7.14 is a consequence of the following assertion:
Given λ < 0 and a smooth function f on M, there is a unique smooth real
function ϕ on M such that
ln M(ϕ) = −λϕ +f. (7.17)
Both equations, (7.13) and (7.17), are of the form
ln M(ϕ) = f(p, ϕ), (7.18)
where f = f(p, ϕ) is a smooth real function
18
. Since exp(f(p, ϕ)) > 0, the
same argument as further up shows that for any smooth solution ϕ of (7.18),
ω + i∂∂ϕ is a positive (1, 1)-form. With this question out of the way, we can
now concentrate on the solvability of our equations. If not specified otherwise,
inner products and ∗-operator are taken with respect to the metric g.
18
It is understood that f(p, ϕ) = f(p, ϕ(p)).
92 Lectures on K¨ ahler Manifolds
7.1 Uniqueness. Let p ∈ M. Given a second K¨ ahler metric g

about p, there
exists a centered holomorphic chart z = (z
1
, . . . , z
m
) about p with
g
p
=
1
2
¸
dz
j
p
dz
j
p
, g

p
=
1
2
¸
a
j
dz
j
p
dz
j
p
, (7.19)
where a
1
, . . . , a
m
are positive numbers. Correspondingly,
ω
p
=
i
2
¸
dz
j
p
∧ dz
j
p
, ω

p
=
i
2
¸
a
j
dz
j
p
∧ dz
l
p
. (7.20)
The volume forms are related by
vol

(p) = (
¸
a
j
) · vol(p). (7.21)
We also see that ω
m−j−1
p
∧ (ω

p
)
j
is a positive linear combination of the forms
i
m−1
2
m−1
dz
1
p
∧ dz
1
p
∧ · · · ∧

dz
k
p
∧ dz
k
p
∧ · · · ∧ dz
m
p
∧ dz
m
p
, (7.22)
where behatted terms are to be deleted.
7.23 Lemma. With respect to the ∗-operator of g, ∗(ω
m−j−1
∧ (ω

)
j
) is a
positive (1, 1)-form.
Proof. Immediate from (7.22).
7.24 Lemma. Let ϕ be a smooth real function on M. If η is a differential
form of degree 2m−2, then
dϕ ∧ d
c
ϕ ∧ η = ∗η(gradϕ, J gradϕ) vol . (1)
In particular,
dϕ ∧ d
c
ϕ ∧ ω
m−1
=
1
m
| gradϕ|
2
ω
m
. (2)
Note that if η is of type (m − 1, m − 1) and ∗η is non-negative (of type
(1, 1)), then ∗η(gradϕ, J gradϕ) ≥ 0.
Proof of Lemma 7.24. We have d
c
ϕ = −dϕ◦ J. Hence (d
c
ϕ)

= J · gradϕ, and
therefore
dϕ ∧ d
c
ϕ ∧ η = 'dϕ ∧ d
c
ϕ, ∗η` vol
= 'd
c
ϕ, gradϕ∗η` vol
= ∗η(gradϕ, J gradϕ) vol.
This proves the first equation. The second equation follows from the first and
Equations 4.20 and 4.22.
7 Calabi Conjecture 93
We now come to the uniqueness of solutions of (7.18). It is related to the
dependence of f on the variable ϕ.
7.25 Proposition. Suppose ϕ
1
and ϕ
2
solve (7.18).
1) If f is weakly monotonically increasing in ϕ, then ϕ
1
−ϕ
2
is constant.
2) If f is strictly monotonically increasing in ϕ, then ϕ
1
= ϕ
2
.
In particular, solutions to Equation 7.17 are unique and solutions to Equa-
tion 7.13 are unique up to an additive constant.
Proof of Proposition 7.25. Set ω
1
= ω +i∂∂ϕ
1
and ω
2
= ω +i∂∂ϕ
2
. Then

ω
m
1
=

ω
m
=

ω
m
2
,
where we integrate over all of M. Hence
0 = 2

M

m
1
−ω
m
2
)
=

dd
c

1
−ϕ
2
) ∧ (ω
m−1
1
+ ω
m−2
1
∧ ω
2
+ · · · + ω
m−1
2
).
Applying Stokes’ theorem and Lemmas 7.23 and 7.24 we get
0 =

d((ϕ
1
−ϕ
2
)d
c

1
−ϕ
2
) ∧ (ω
m−1
1
+ ω
m−2
1
∧ ω
2
+ · · · + ω
m−1
2
))
=

d(ϕ
1
−ϕ
2
) ∧ d
c

1
−ϕ
2
) ∧ (ω
m−1
1
+ ω
m−2
1
∧ ω
2
+ · · · + ω
m−1
2
)
+


1
−ϕ
2
)(ω
m
1
−ω
m
2
)

1
m

| grad(ϕ
1
−ϕ
2
)|
2
g1
ω
m
1
+


1
−ϕ
2
)(e
f(p,ϕ1)
−e
f(p,ϕ2)

m
,
where the index g
1
means that we take the norm with respect to the metric
g
1
. If f is weakly monotonically increasing in ϕ, then both terms on the right
hand side are non-negative.
7.2 Regularity. Let U ⊂ C
2
(M) be the open subset of functions ϕ such that
ω +i∂∂ϕ is positive. Let F: U →C
0
(M) be the functional
F(ϕ) = ln M(ϕ) −f(p, ϕ), (7.26)
where f = f(p, ϕ) is a smooth real function. We want to study the solvability of
the equation F(ϕ) = 0 using the continuity method. To that end, the regularity
of solutions will play an essential role in Subsection 7.3 since we will need a
priori estimates of their higher order derivatives, compare (7.40) and (7.41).
94 Lectures on K¨ ahler Manifolds
For ϕ ∈ U and ω
ϕ
= ω +i∂∂ϕ, a continuous differential form, we let g
ϕ
be
the associated continuous Riemannian metric as in (4.2). Since ω
ϕ
is of type
(1, 1), g
ϕ
is compatible with the complex structure of M. We call

ϕ
= −2g
jk
ϕ

2
∂z
j
∂z
k
, (7.27)
the Laplacian of g
ϕ
, compare Exercise 5.51.1. Since ϕ ∈ U, ∆
ϕ
is elliptic.
For ϕ ∈ U and ψ ∈ C
2
(M), the directional derivative of F at ϕ ∈ U in the
direction of ψ is
(F(ϕ +tψ))

|
t=0
= −
1
2

ϕ
ψ − f
ϕ
(p, ϕ) · ψ, (7.28)
where f
ϕ
denotes the partial derivative of f in the direction of the variable ϕ.
It follows that F is continuously differentiable with derivative dF(ϕ) · ψ given
by the right hand side of (7.28). The same statements also hold for F when
considered as a functional from U ∩ C
k+2+α
(M) to C
k+α
(M), where k ≥ 0 is
an integer and α ∈ (0, 1).
7.29 Proposition. If ϕ ∈ U ∩ C
2+α
(M) solves F(ϕ) = 0, then ϕ is smooth.
Proof. Since f is smooth and ϕ is C
2+α
, the function
det(g
jk
)
−1
det(g
jk
+ ϕ
jk
) −f(p, ϕ)
is C
2+α
in p ∈ M and the (free) variables ϕ
jk
. By (7.28), ϕ is an elliptic
solution of the equation F(ϕ) = 0. Hence ϕ is C
4+α
and, recursively, C
k+α
for
any k ≥ 6, by the regularity theory for elliptic solutions of partial differential
equations
19
.
7.3 Existence. From now on, we only consider Equations 7.13 and 7.17,
F(ϕ) = ln M(ϕ) + λϕ −f = 0 on U ∩ C
2+α
(M), (7.30)
where U is as above, α ∈ (0, 1), f is a given smooth function on M, and λ = 0
in the case of (7.13) and λ < 0 in the case of (7.17). We also recall that in the
case of λ = 0, we are only interested in f modulo an additive constant.
7.31 Remark. In the case of complex dimension m = 1, Equation 7.30 be-
comes
e
f−λϕ
g = g −
1
2
∆ϕ.
In this case, the arguments below are much easier. It is a good exercise to follow
the argument below in this more elementary case. This is still a non-trivial case,
the uniformization of closed Riemann surfaces.
19
In [Ka2], Kazdan gives a nice exposition (with references) of this theory.
7 Calabi Conjecture 95
For the proof of the existence of solutions we apply the method of continuity.
We let 0 ≤ t ≤ 1 and replace f by tf in (7.30). That is, we consider the equation
F
t
(ϕ) = ln M(ϕ) + λϕ −tf = 0 on U ∩ C
2+α
(M). (7.30.t)
We let T
α
be the set of t ∈ [0, 1] for which a C
2+α
-solution of Equation 7.30.t
exists (modulo a constant function if λ = 0). The constant function ϕ = 0
solves 7.30.0. Hence T
α
is not empty. The aim is now to show that T
α
is open
and closed. We distinguish between the cases λ < 0 and λ = 0.
We start with the openness and let t ∈ T
α
and ϕ be a solution of Equation
7.30.t. Now
dF
t
(ϕ) = dF(ϕ) = −
1
2

ϕ
+ λ (7.32)
defines a self-adjoint operator on L
2
(M, g
ϕ
). If λ < 0, then the kernel of dF
t
(ϕ)
on L
2
(M) is trivial and hence dF
t
(ϕ) is an isomorphism from its domain H
2
(M)
of definition onto L
2
(M), by the spectral theorem. Therefore, by the regularity
theory for linear partial differential operators, dF
t
(ϕ): C
2+α
(M) →C
α
(M) is
an isomorphism as well. Hence by the implicit function theorem, there is a
unique solution ϕ
s
of Equation 7.30.s which is close to ϕ
t
in C
2+α
(M), for all
s close enough to t. Hence T
α
is open.
In the case λ = 0, we only need to solve modulo constant functions. We let
Z ⊂ L
2
(M) be the space of functions with g
ϕ
-mean zero, a closed complement
of the constant functions in L
2
(M). Since constant functions are smooth, Z ∩
H
2
(M) and Z ∩C
k+α
(M) are closed complements of the constant functions in
H
2
(M) and C
k+α
(M), for all k ≥ 0. By the spectral theorem, dF
t
(ϕ) = −∆
ϕ
/2
is surjective from H
2
(M) onto Z. Hence dF
t
(ϕ): C
2+α
(M) → Z ∩ C
α
(M) is
surjective, by the regularity theory for elliptic differential operators. Hence
by the implicit function theorem, there is a family of functions ϕ
s
, for s close
enough to t, such that ϕ
s
solves Equation 7.30.s up to constant functions.
Hence T
α
is open in this case as well.
We now come to the heart of the matter, the closedness of T
α
. We need to
derive a priori estimates of solutions ϕ of Equation 7.30.t, or, what will amount
to the same, of Equation 7.30.
7.33 Lemma. Let λ < 0 and suppose that ln M(ϕ) = −λϕ +f. Then
|ϕ|


1
|λ|
|f|

.
Proof. Let p ∈ M be a point where ϕ achieves its maximum. Then M(ϕ)(p) ≤
1, by (7.12), and hence λϕ(p) ≥ f(p). Since λ < 0,
ϕ ≤ ϕ(p) ≤
1
λ
f(p) ≤
1
|λ|
|f|

.
96 Lectures on K¨ ahler Manifolds
If ϕ achieves a minimum at p, then M(ϕ)(p) ≥ 1, and hence λϕ(p) ≤ f(p).
Therefore
ϕ ≥ ϕ(p) ≥
1
λ
f(p) ≥ −
1
|λ|
|f|

.
The case λ = 0 is much harder.
7.34 Lemma. Suppose that ϕ is a smooth function on M with ln M(ϕ) = f
and mean 0 with respect to g. Then |ϕ|

≤ C, where C is a constant which
depends only on M, g, and an upper bound for C(f) := |1 −exp◦f|

.
It will become clear in the proof that the constant C is explicitly computable
in terms of C(f) and dimension, volume, and Sobolev constants of M.
Proof of Lemma 7.34. We let ϕ be a smooth function on M with ln M(ϕ) = f.
Recall that the latter holds iff (ω

)
m
= e
f
ω
m
. We have
(1 −e
f

m
= ω
m
−(ω

)
m
= (ω −ω

) ∧ (ω
m−1
+ ω
m−2
∧ ω

+ · · · + (ω

)
m−1
)
= −
1
2
dd
c
ϕ ∧ (ω
m−1
+ ω
m−2
∧ ω

+ · · · + (ω

)
m−1
).
For p ≥ 2, Stokes’ theorem gives

M
d{ϕ|ϕ|
p−2
d
c
ϕ ∧ (ω
m−1
+ · · · + (ω

)
m−1
)} = 0,
where we note that ϕ|ϕ|
p−2
is C
1
for p ≥ 2. Now
d(ϕ|ϕ|
p−2
) = (p −1)|ϕ|
p−2
dϕ,
hence
(p −1)

M
|ϕ|
p−2
dϕ∧d
c
ϕ∧(ω
m−1
+· · · +(ω

)
m−1
) = 2

M
(1 −e
f
)ϕ|ϕ|
p−2
ω
m
.
By Lemma 7.24, the left hand side is equal to
p −1
m

M
|ϕ|
p−2
(| gradϕ|
2
+Φ)ω
m
,
where Φ ≥ 0. From
1
4
p
2
|ϕ|
p−2
| gradϕ|
2
= | grad|ϕ|
p/2
|
2
and ω
m
= m! vol
g
, we conclude that, for p ≥ 2,
| grad|ϕ|
p/2
|
2
2

mp
2
2(p −1)
C(f)|ϕ|
p−1
p−1
, (7.35)
7 Calabi Conjecture 97
where the index refers to the corresponding integral norm.
By the Sobolev embedding theorem, there is a constant C
1
with
|ϕ|2m+2
m
≤ C
1
· (|ϕ|
2
+| gradϕ|
2
). (7.36)
From now on we assume that ϕ has mean 0 with respect to g. Then we
have
|ϕ|
2
≤ C
2
· | gradϕ|
2
, (7.37)
where 1/C
2
2
is the first eigenvalue of the Laplacian of (M, g).
By (7.35), applied in the case p = 2, we have
| gradϕ|
2
2
≤ 2mC(f)|ϕ|
1
.
Now |ϕ|
1
≤ vol(M)
1/2
· |ϕ|
2
. Hence, by (7.37),
| gradϕ|
2
2
≤ 2mC(f)C
2
vol(M)
1/2
| gradϕ|
2
.
With C
3
:= 2mC
2
vol(M)
1/2
, we get
| gradϕ|
2
≤ C
3
C(f) and |ϕ|
2
≤ C
2
C
3
C(f),
where we apply (7.37) for the second estimate. Hence, by (7.36),
|ϕ|2m+2
m
≤ C
1
(C
2
+ 1)C
3
C(f). (7.38)
We now apply an iteration argument to get the desired uniform bound for ϕ.
With α = (m+ 1)/m and p ≥ 2α, we have
|ϕ|
p

=

M
|ϕ|

1/α
=

M
|ϕ|
p
2

2

= ||ϕ|
p/2
|
2

≤ 2C
2
1
·

|ϕ|
p
p
+| grad|ϕ|
p/2
|
2
2

≤ 2C
2
1
·

|ϕ|
p
p
+
mp
2
2(p −1)
C(f)|ϕ|
p−1
p−1

≤ 2C
2
1
·

|ϕ|
p
p
+
mp
2
2(p −1)
C(f)(vol M)
1/p
|ϕ|
p−1
p

≤ C
4
p · max{1, |ϕ|
p
p
}.
By Equation 7.38, there is a constant C ≥ 1 such that
|ϕ|

≤ C · (C
4
2α)

m+1

.
We choose C such that
C · (C
4
p)

m+1
p
≥ 1
98 Lectures on K¨ ahler Manifolds
for all p ≥ 2α and assume recursively that
|ϕ|
p
≤ C · (C
4
p)

m+1
p
for some p ≥ 2α. Since mα = m+ 1, we then get that
|ϕ|


≤ C
α
4
p
α
C

(C
4
p)
−(m+1)α
= C

(C
4
p)
−m−1
and hence that
|ϕ|

≤ C(C
4
p)

m+1

.
Since |ϕ|
p
→|ϕ|

as p →∞, we conclude that |ϕ|

≤ C.
For λ ≤ 0 fixed, Lemmas 7.33 and 7.34 give an priori estimate on the
C
0
-norm of solutions ϕ of Equation 7.30,
|ϕ|

≤ C(|f|

, M, g), (7.39)
where we assume that the mean of ϕ is 0 if λ = 0. This is the main a priori
estimate. There are two further a priori estimates. They are less critical
20
and
hold, in particular, for positive λ as well. However, their derivation is space
and time consuming and will therefore not be presented here. At this stage, the
reader should be well prepared to turn to the literature to get the remaining
details.
Let 0 < β < 1. The first of the two remaining a priori estimates concerns
the g-Laplacian of ϕ,
|∆ϕ|

≤ C

(|ϕ|

, |f|

, |∆f|

, |λ|, M, g), (7.40)
see [Au2], [Ya3] or [Au3, §7.10]. By the definition of the metric g
ϕ
associated
to ϕ, (7.40) gives a uniform upper bound for g
ϕ
against the given metric g.
Now the volume elements of g
ϕ
and g only differ by the factor exp◦f, see (7.5),
hence (7.40) gives also a lower bound for g
ϕ
against g. In particular, (7.40)
gives estimates between norms associated to g and g
ϕ
.
The second a priori estimate concerns the g-covariant derivative of dd
c
ϕ,
|∇dd
c
ϕ|

≤ C

(|∆ϕ|

, |f|
C
3
(M)
, |λ|, M, g), (7.41)
[Ca3], [Au2], [Ya3] or [Au3, §7.11]. In the derivation of this latter estimate, one
actually estimates ∇dd
c
ϕ first in the C
0
-norm associated to g
ϕ
. By what we
said above, this is equivalent to the estimate in (7.41). The derivation of this
estimate is, among others, a true fight against notation and in our sources, the
authors refer to the original articles cited above.
From (7.41) and Exercise 5.51.1 we conclude that
|∇∆ϕ|

≤ C

(|∆ϕ|

, |f|
C
3
(M)
, |λ|, M, g). (7.42)
20
An examination of the case m = 1 is recommended.
7 Calabi Conjecture 99
Hence ∆ϕ is uniformly bounded in the C
β
-norm associated to the fixed metric
g. Using the Schauder estimates, we conclude that ϕ is bounded in the C
2+β
-
norm associated to g, where the bound now also depends on β.
Note that in our application of the continuity method, we consider the
family of functions tf, 0 ≤ t ≤ 1, in place of the original f, whose norms are
bounded by the corresponding norms of f.
We are ready for the final step in the proof of Theorems 7.1 and 7.14. It
remains to show that T
α
is closed. Let 0 < α < β < 1. Let t
n
∈ T
α
and suppose
t
n
→t ∈ [0, 1]. Let ϕ
n
be a solution of Equation 7.30.t
n
. In the case λ = 0, we
assume that ϕ
n
has mean 0. Then by the above, there is a uniform bound on
the C
2+β
-norm of the functions ϕ
n
. Now the inclusion C
2+β
(M) →C
2+α
(M)
is compact. Hence by passing to a subsequence if necessary, ϕ
n
converges in
C
2+α
(M). The limit ϕ = limϕ
n
solves 7.30.t. Hence T
α
is closed.
7.4 Obstructions. In Corollary 7.2 and Theorem 7.14, we obtained existence
of K¨ ahler–Einstein metrics on closed K¨ahler manifolds assuming non-positve
first Chern class. In the case of positive first Chern class, the arguments break
down to a large extent. When applying the continuity method as explained in
the previous subsection, it is no longer possible to show openness or closedness
of T
α
along the lines of what is said there.
Concerning the openness of T
α
, the operator ∆
ϕ
−2λ might have a kernel for
λ > 0. This problem can be overcome [Au4]: In the notation of Subsection 7.3,
we need to solve the Monge–Amp`ere equation ln M(ϕ) +λϕ−f = 0 with λ > 0
and a given smooth real function f. In the λ ≤ 0 case, we replaced f with tf,
so that we obtained a (trivially) solvable equation for t = 0. Here the trick
is to keep f, but to replace ϕ with tϕ. Then the equation for t = 0 is highly
non-trivial, but solvable according to our previous results for λ = 0. A simple
calculation shows that if ω
t
is a solution, then the Ricci form ρ
t
of ω
t
is given
by
ρ
t
= λtω
t
+ λ(1 −t)ω
0
,
therefore Ric
t
> λtg
t
for t < 1 and Ric
1
= λg
1
. Hence the linearized operator

t
− 2λt is invertible for all t ∈ [0, 1) for which a solution exists, see Theo-
rem 6.14. This is enough for the continuity method to work, provided we can
ensure that T
α
is closed. Observe that if the manifold admits automorphic
vector fields, the linearized operator has a kernel for t = 1, see Theorem 6.16.
As for the closedness of T
α
, the a priori estimates (7.40) and (7.41) do
not depend on the sign of λ and continue to hold. However, the remaining
C
0
-estimate (7.40) is critical and might fail. Indeed, there are closed complex
manifolds with positive first Chern class which do not admit K¨ahler–Einstein
metrics at all. As we will see, this can be deduced from special properties
of their automorphism groups, a kind of guiding theme: Non-discrete groups
of automorphisms yield obstructions to the existence of K¨ ahler–Einstein met-
rics. Even though a general criterion for the solvability of the K¨ahler–Einstein
100 Lectures on K¨ ahler Manifolds
problem cannot possibly be based on the automorphism group alone – mani-
folds with trivial and non-trivial automorphism groups can, but need not be
obstructed –, the underlying ideas related to holomorphic actions of complex
Lie groups seem to be at the heart of the (still conjectural) full solution.
In the remainder of this subsection, we will treat the two classical obstruc-
tions against positive K¨ ahler–Einstein metrics that rely on the automorphism
group.
7.43 Theorem (Matsushima [Ma1]). Let M be a closed K¨ahler–Einstein mani-
fold. Let a be the complex Lie algebra of automorphic vector fields on M and
k ⊂ a be the real Lie subalgebra of Killing fields. Then a = k + Jk. If the
Einstein constant of M is positive, then a = k ⊕Jk.
7.44 Exercise. Verify this explicitly for the Fubini–Study metric on CP
m
.
Proof of Theorem 7.43. Let X be a vector field on M and ξ := X

. By Hodge
theory, ξ = η + ζ, where η ∈ imd and ζ ∈ ker d

. In other words, Y := η

is a
gradient field and Z := ζ

is volume preserving (recall that div Z = −d

ζ).
Assume now that X is automorphic. Our first goal is to prove that Y and Z
are automorphic as well. Now Proposition 4.79 and the Bochner identity 1.38
imply that
dd

η +d

dζ = ∆
d
η +∆
d
ζ = ∆
d
ξ = 2 Ric ξ = 2λξ = 2λη + 2λζ,
where λ is the Einstein constant of M. On the other hand, η ∈ imd and ζ ∈
ker d

, so dd

η = 2λη and d

dζ = 2λζ according to the Hodge decomposition
theorem, and hence ∆η = 2 Ric η and ∆ζ = 2 Ric ζ. Again by Proposition 4.79,
Y and Z are automorphic.
Now Z is a volume preserving automorphic field, therefore a Killing field,
by Theorem 4.81. Hence ∇Y is symmetric and ∇Z is skew-symmetric. Let
s be the real vector space of all holomorphic vector fields Y such that ∇Y
is symmetric. By Exercise 7.45.1, multiplication by J yields an isomorphism
k →s of real vector spaces. Hence a = k +Jk.
Now k ∩ s is the space of parallel holomorphic vector fields. If λ > 0, the
Ricci curvature of M is positive, and then M does not carry parallel vector
fields. Hence a = k ⊕Jk if λ > 0.
7.45 Exercise. Let M be a K¨ ahler manifold and X be a real vector field on
M.
1) If X is holomorphic, then JX is a Killing field if and only if ∇X is
symmetric.
2) If ∇X is symmetric, then JX is volume preserving.
3) Visualize 1) and 2) pictorially if dim
C
M = 1.
A Lie algebra is called reductive if it is isomorphic to the direct sum of
its center and a semi-simple Lie algebra. A connected Lie group is called
7 Calabi Conjecture 101
reductive if its Lie algebra is reductive. We note the following consequence of
Theorem 7.43.
7.46 Corollary. If M is a closed K¨ahler–Einstein manifold, then the compo-
nent of the identity of the automorphism group of M is reductive.
Proof. Let a and k be as above. Since k is the Lie algebra of a compact Lie
group, k = z ⊕ [k, k], where z is the center of k. Moreover, the commutator
subalgebra [k, k] of k is semi-simple. Now k ∩Jk consists of parallel vector fields,
see Exercise 7.45.1, hence k ∩ Jk ⊂ z. Therefore
a = k +Jk = (z +Jz) ⊕([k, k] +J[k, k]).
Since a = k +Jk, the center of a is z +Jz, and [k, k] +J[k, k] is semi-simple.
7.47 Example. Let M be the once blown up CP
2
as in Example 2.43. The
Lie algebra a of the automorphism group of M consists of the matrices

¸
0 ∗ ∗
0 ∗ ∗
0 ∗ ∗
¸

∈ C
3×3
.
The center of a is trivial. However, a is not semi-simple since the space of
matrices in a with zero entries in second and third row is an Abelian ideal.
Hence a is not reductive, and hence M does not carry K¨ ahler–Einstein metrics.
By an analogous argument, the blow up of CP
2
in two points does not carry
a K¨ ahler–Einstein metric either; we leave this as an exercise. The blow up of
CP
2
in one or two points carries K¨ ahler metrics with Ric > 0, see [Hit], [Ya1].
Moreover, the blow up of CP
2
in one point carries Riemannian metrics with
Ric > 0 and sectional curvatures ≥ 0 [Che] and Einstein metrics with Ric > 0
which are conformal to K¨ ahler metrics [PP].
7.48 Remark. For closed K¨ ahler manifolds of constant scalar curvature, we
still have a = k+Jk, see [Li2]. Hence the argument in the proof of Corollary 7.46
applies and shows that the identity components of the automorphism groups
of such manifolds are also reductive. In particular, the once and twice blown
up CP
2
do not even admit K¨ ahler metrics of constant scalar curvature.
We now discuss a second obstruction to the existence of positive K¨ahler–
Einstein metrics which is also related to the automorphism group. Let M be a
closed K¨ ahler manifold with K¨ ahler form ω and Ricci form ρ.
7.49 Theorem (Futaki [Fu]). Assume that ω ∈ 2πc
1
(M) and write ρ − ω =
i∂∂F
ω
with F
ω
∈ E(M, R). Then the map
F
ω
: a →R, F
ω
(X) :=

M
X(F
ω

m
,
does not depend on the choice of the K¨ahler form ω ∈ 2πc
1
(M).
102 Lectures on K¨ ahler Manifolds
We remark that the function F
ω
in the definition of the map F
ω
exists, and
is unique up to a constant, since ρ ∈ 2πc
1
(M) ∈ H
1,1
(M, R), see Exercise 5.51.
The map F = F
ω
, with ω ∈ 2πc
1
(M) an arbitrary K¨ ahler form, is called
the Futaki invariant of M. If ω is a K¨ ahler–Einstein metric, ω = ρ, then F
ω
is
constant and the Futaki invariant F
ω
vanishes on a. The Futaki invariant of
both the once and the twice blown up CP
2
is non-zero, see [Ti3, Section 3.2].
Hence we conclude again that these two do not admit K¨ ahler–Einstein metrics.
Let H
1
→ CP
1
and H
2
→ CP
2
be the hyperplane bundles. Let M be the
total space M of the projective bundle of the product H
1
H
2
→CP
1
CP
2
, a
closed and complex manifold of complex dimension 4. Futaki [Fu, §3] shows that
M is a Fano manifold with reductive automorphism group but non-zero Futaki
invariant. Hence the Matsushima criterion for the non-existence of K¨ ahler–
Einstein metrics does not apply in this case, whereas the Futaki obstruction
does.
Proof of Theorem 7.49. We let ω
t
, −ε < t < ε, be a smooth family of K¨ahler
forms in 2πc
1
(M). The goal is to show that

t

F
ωt
(X)

= 0.
By Exercise 5.51, ω
t
− ω
0
= i∂∂Φ
t
and ρ
t
−ω
t
= i∂∂F
t
for families of smooth
functions Φ
t
and F
t
. From the proof of the dd
c
-Lemma 5.50 we see that these
families of functions can be chosen to be smooth in t. As in (7.15), we compute
−i∂∂ ln M(Φ
t
) = ρ
t
−ρ
0
= i∂∂(F
t
+ Φ
t
−F
0
),
hence −lnM(Φ
t
) = F
t

t
−F
0
+ constant.
From now on, we mostly suppress the parameter t in our notation and
indicate differentiation with respect to t by a dot or by ∂
t
. With ϕ =
˙
Φ and
f =
˙
F we get ˙ ω = i∂∂ϕ and ˙ ρ − ˙ ω = i∂∂f. The displayed equation above gives
∆ϕ = 2f + 2ϕ,
compare Exercise 5.51. By Exercise 7.50, we have
2∂
t
ω
m
t
= 2m˙ ω ∧ ω
m−1
= 2mi∂∂ϕ ∧ ω
m−1
= −∆ϕ · ω
m
.
Thus
2∂
t

F
ωt
(X)

= 2

M
{(Xf)ω
m
+ (XF)∂
t
ω
m
}
=

M
¸
X(∆ϕ −2ϕ) −(XF)∆ϕ
¸
ω
m
.
7 Calabi Conjecture 103
From now on, we suppress the measure of integration since it is always ω
m
.
Because X is holomorphic,

M
X(∆ϕ) =

M
'X, grad∆ϕ`
=

M
{Ric(X, gradϕ) +'X, ∇

∇(gradϕ)`}
=

M
{Ric(X, gradϕ) +'∇

∇X, gradϕ`}
= 2

M
Ric(X, gradϕ),
by Exercise 6.17 and Proposition 4.79. It follows that

M
X(∆ϕ −2ϕ) = 2

M
¸
Ric(X, gradϕ) −'X, gradϕ`
¸
= 2

M
¸
ρ(X, J gradϕ) −ω(X, J gradϕ)
¸
= 2

M
(i∂∂F)(X, J gradϕ)
=

M
{∇
2
F(gradϕ, X) +∇
2
F(J gradϕ, JX)},
because of ρ −ω = i∂∂F and Exercise 7.50. Now
div((XF) gradϕ) = 'grad(XF), gradϕ` −(XF)∆ϕ,
hence

M
(XF) ∆ϕ =

M
'grad(XF), gradϕ`
=

M
¸

2
F(gradϕ, X) +'∇
grad ϕ
X, gradF`
¸
.
Since X is automorphic and J is parallel,
'∇
grad ϕ
X, gradF` = '∇
J grad ϕ
X, J gradF` = −'∇
J grad ϕ
JX, gradF`.
Adding up, we get
2∂
t

F
ωt
(X)

=

M
¸

2
F(J gradϕ, JX) +'∇
J grad ϕ
JX, gradF`
¸
=

M
(J gradϕ)(JX(F)) = −

M
(JX(F)) div(J gradϕ).
Now div J gradϕ = 0, see Exercise 7.45.2, hence ∂
t

F
ωt
(X)

= 0.
104 Lectures on K¨ ahler Manifolds
7.50 Exercise. Let M be a K¨ ahler manifold and ϕ be a smooth function on
M. Show that
2mi∂∂ϕ ∧ ω
m−1
= −∆ϕ · ω
m
, (1)
2i∂∂ϕ(X, JY ) = ∇
2
ϕ(X, Y ) +∇
2
ϕ(JX, JY ). (2)
The complex Hessian of ϕ is defined by Hess
c
ϕ(X, Y ) = i∂∂ϕ(X, JY ). Write
down Hess ϕ = ∇
2
ϕ and Hess
c
ϕ in K¨ ahler normal coordinates.
7.51 Remark. Calabi generalized the Futaki invariant to general closed K¨ahler
manifolds M: Fix a K¨ ahler metric ω on M and write ρ
ω
= γ
ω
+i∂∂F
ω
, where γ
ω
is the unique ω-harmonic form in 2πc
1
(M). Then the Calabi–Futaki invariant
CF
ω
: a →R, CF
ω
(X) :=

M
X(F
ω

m
, (7.52)
does not depend on the choice of ω in its K¨ ahler class. It vanishes if the
K¨ ahler class of ω contains a K¨ahler metric with constant scalar curvature. The
previously defined Futaki invariant is the special case where ω ∈ 2πc
1
(M) (if the
latter is positive definite). Our proof of Theorem 7.49 follows Futaki’s original
argument. For a discussion of the more general Calabi–Futaki invariant, see
for example Chapter 2.I in [Bes], Section 3.1 in [Ti3], or Section 1.6 in [Bo3].
It follows from work of Siu, Tian, and Yau that among the Fano surfaces
as in Example 7.4.2, only the blow ups of CP
2
in one or two points do not
admit K¨ ahler–Einstein metrics [Si2], [Ti1], [Ti2], [TY]. These are precisely the
ones whose automorphism group is not reductive and whose Futaki invariants
do not vanish. In higher dimensions, the situation is much more complicated.
There are examples, where the C
0
-estimates in the continuity method could be
verified. For example, the Fermat hypersurfaces
z
d
0
+ · · · +z
d
n
= 0
in CP
n
carry K¨ ahler–Einstein metrics if (n + 1)/2 ≤ d ≤ n, see [Na], [Si2],
[Ti1] and compare Example 7.4.1. For surveys on K¨ ahler–Einstein metrics, we
refer to [Bo3] and [Ti3]. More recent references are [Bi] and [Th]. All these
references contain extensive bibliographies as well.
8 K¨ahler Hyperbolic Spaces
A question attributed to Chern asks whether the Euler characteristic of a closed
Riemannian manifold M of dimension 2m satisfies
(−1)
m
χ(M) > 0 (8.1)
if the sectional curvature K of M is negative. The answer is yes if m ≤ 2, see
[Chr] for the case m = 2. Dodziuk and Singer remarked that Atiyah’s L
2
-index
theorem implies (8.1) if the space of square integrable harmonic forms on the
universal covering space of M vanishes in degree = m and does not vanish in
degree = m. In [Gr], Gromov proves this and more in the case where M is
a K¨ ahler manifold; in this section we are concerned with his arguments and
results.
Let M be a complete and connected Riemannian manifold and E be a flat
Hermitian vector bundle over M. Choose an origin o ∈ M and let r = d(o, · )
be the distance to o. Let α be a differential form on M with values in E and
µ: R
+
→ R
+
be a non-decreasing function. We say that α is O(µ(r)) if there
is a constant c such that
|α| ≤ cµ(cr +c) +c, (8.2)
where we consider the norm |α| = |α(p)| as a function on M. Following Gromov,
we say that α is d(O(µ(r))) if there is a differential form β on M with α = dβ
and such that β is O(µ(r)). We also write d(bounded) instead of d(O(1)), that
is, in the case where β can be chosen to be uniformly bounded.
8.3 Examples. 1) Let M = R
n
with Euclidean metric and origin 0. Let α be
the volume form of M, α = dx
1
∧ · · · ∧ dx
n
. Let β be an (n − 1)-form with
dβ = α. Then on B = B(r, 0),
vol
n
B =

B
α =

∂B
β ≤ max{|β(p)| | p ∈ ∂B} vol
n−1
∂B.
It follows that
max{|β(p)| | p ∈ ∂B} ≥
vol
n
B
vol
n−1
∂B
=
r
n
.
Hence α is not d(bounded). In a similar way one can treat each of the differ-
ential forms dx
I
, where I is a multi-index.
2) Let α be a bounded 1-form on M and f be a smooth function with
df = α. Then
|df| = |α| ≤ |α|

:= sup{|α(p)| | p ∈ M}.
Hence f is Lipschitz continuous with Lipschitz constant |α|

if |α|

< ∞.
106 Lectures on K¨ ahler Manifolds
8.4 Proposition. Let M be complete and simply connected and α be a differ-
ential form on M of degree k with |α|

< ∞ and dα = 0.
1) If K ≤ 0, then α = d(O(r)).
2) If K ≤ −c
2
< 0 and k ≥ 2, then α = d(bounded).
Proof. Since M is complete and simply connected with non-positive sectional
curvature, the theorem of Hadamard–Cartan implies that exp
o
: T
o
M → M is
a diffeomorphism. Hence we may define
τ
t
(p) := exp
o
(t exp
−1
o
(p)),
where 0 ≤ t ≤ 1. Comparison with Euclidean space shows that the differential
of τ
t
satisfies

t∗
v| ≤ t|v|.
The usual proof of the Poincar´e lemma provides a (k −1)-form β with dβ = α.
In Riemannian normal coordinates, β is given by
β(x) = r

1
0
t
k−1
(να)(tx) dt = r

1
0
τ

t
(να)(x) dt,
where r = |x| and ν is the radial normal field. By the above estimate


t
(να)| ≤ t
k−1
|να| ≤ t
k−1
|α|

,
and hence
|β(p)| ≤
r
k
|α|

,
where p = exp(x). Note that this estimate is optimal, see Example 8.3.1 above.
Suppose now that the sectional curvature of M is negative, K ≤ −c
2
< 0.
Comparison with hyperbolic space shows that, for v perpendicular to ν,

t∗
v| ≤
sinh(ctr)
sinh(cr)
|v|,
where r is the distance of the foot point of v to o. Hence
|β(p)| ≤ r

1
0

sinh(ctr)
sinh(cr)

k−1
dt · |α|

=

r
0

sinh(cs)
sinh(cr)

k−1
ds · |α|

0
−∞
(e
(k−1)cs
) ds · |α|

=
1
(k −1)c
|α|

,
therefore α = d(bounded).
8 K¨ ahler Hyperbolic Spaces 107
8.5 Remark. The estimates in the proof are explicit. Note also that the lift of
a differential form on a compact manifold to any covering manifold is uniformly
bounded.
8.6 Proposition. Let (G, K) be a Riemannian symmetric pair such that M =
G/K is a symmetric space of non-compact type. If α is a G-invariant differen-
tial form on M, then α is parallel and hence closed. If α is of positive degree,
then the restriction of α to maximal flats in M vanishes and α = d(bounded).
This result seems to be well known to experts, I learned it from Anna
Wienhard. The proof (and even the statement) of Proposition 8.6 requires
more on symmetric spaces than we develop in Appendix B. Therefore we only
give a brief sketch of the argument. Good references for symmetric spaces of
non-compact type are Chapter 2 in [Eb] and Chapter VI in [Hel].
Sketch of proof of Proposition 8.6. We can assume that (G, K) is an effective
pair. By Remark B.36, G contains the connected component of the identity of
the isometry group of M. By Theorem B.24.3, α is parallel and hence closed.
Let F ⊂ M be a maximal flat and p ∈ F. There is a basis of T
p
F such
that the reflections of T
p
F about the hyperplanes perpendicular to the vectors
of the basis are realized by the differentials of isometries g ∈ G fixing p. Since
α is invariant under G, it follows easily that α|T
p
F = 0.
Let c: R → M be a regular unit speed geodesic and b: M → R be the
Busemann function centered at c(∞) with b(c(t)) = −t. Let V = −gradb and
(f
t
) be the flow of V . Then, for each p ∈ M, c
p
(t) = f
t
(p), t ∈ R, is the unit
speed geodesic through p asymptotic to c. In particular, c
p
is also regular and,
therefore, contained in a unique maximal flat F
p
. The flats F
p
constitute a
smooth foliation F of M by totally geodesic Euclidean spaces.
For any unit tangent vector u of M perpendicular to F, 'R(u, V )V, u` is
negative and, by homogeneity, bounded away from 0 by a negative constant
which does not depend on u. In particular, the Jacobi fields of geodesic vari-
ations by geodesics c
p
and perpendicular to F decay uniformly exponentially.
Since α vanishes along F, it follows easily that
γ = −


0
F

t
αdt
is a well defined smooth differential form on M such that L
V
γ = α and dγ = 0.
With β = V γ we get
α = L
V
γ = d(V γ) +V dγ = dβ.
Now γ is uniformly bounded and |V | = 1, hence α = d(bounded).
8.7 Definition (Gromov). Let M be a K¨ ahler manifold with K¨ ahler form
ω. We say that M is K¨ ahler-hyperbolic if the K¨ ahler form ˜ ω of the universal
covering space
˜
M of M is d(bounded).
108 Lectures on K¨ ahler Manifolds
Recall that the cohomology class of the K¨ ahler form of a closed K¨ahler mani-
fold M is non-trivial. Hence closed K¨ahler manifolds with finite fundamental
group are not K¨ ahler-hyperbolic. On the other hand, if the sectional curvature
of M is strictly negative, then M is K¨ ahler-hyperbolic, by Proposition 8.4. By
Proposition 8.6, Hermitian symmetric spaces of non-compact type are K¨ ahler
hyperbolic as well. These do not have strictly negative curvature if their rank,
that is, the dimension of their maximal flats, is larger than 1.
K¨ ahler-hyperbolic K¨ ahler manifolds are the main topic in Gromov’s article
[Gr]. Cao and Xavier [CX] and independently Jost and Zuo [JZ] observed that
the arguments of Gromov concerned with the vanishing of square integrable
harmonic forms work also under the weaker assumption that the K¨ ahler form
is d(O(µ)), where
¸

1
1/µ(n) = ∞. Note that this assumption is fulfilled if M
is simply connected with non-positive sectional curvature, by Proposition 8.4.1.
We include this extension into our discussion, see Theorem 8.9 below.
For the rest of this section we assume that M is a complete and connected
K¨ ahler manifold with K¨ ahler form ω and complex dimension m. We let E →M
be a flat Hermitian vector bundle. Declaring parallel sections as holomorphic
turns E into a holomorphic vector bundle such that the Chern connection of
the given Hermitian metric is the given flat connection.
Without further notice we use notation and results from Appendix C. We
do not assume throughout that differential forms are smooth and indicate their
regularity when appropriate.
8.1 K¨ahler Hyperbolicity and Spectrum. One of the remarkable proper-
ties of K¨ ahler manifolds is that the Lefschetz map commutes with the Laplacian:
If α is a harmonic form with values in E, then ω ∧ α is harmonic as well. This
is one of the cornerstones in the arguments which follow.
8.8 Lemma. Suppose that ω is d(O(µ(r))), where
¸

1
1/µ(n) = ∞. Let α be
a closed and square integrable differential form with values in E. Then
d

(ω ∧ α) = 0 =⇒ ω ∧ α = 0.
Proof. Choose an origin o ∈ M, and let r : M →R be the distance to o. Since
M is complete and connected, the sublevels of r are relatively compact and
exhaust M. Choose cut off functions ϕ
n
: M → R, n ≥ 1, with ϕ
n
(p) = 1 if
r(p) ≤ n, ϕ
n
(p) = 0 if r(p) ≥ n + 1 and |dϕ
n
| ≤ c
0
.
Let ω = dη with |η| = O(µ(r)). Since dα = 0 = d

(ω ∧ α) and ϕ
n
η is
smooth with compact support,
0 =

M
(d

(ω ∧ α), ϕ
n
η ∧ α) =

M
(ω ∧ α, d(ϕ
n
η ∧ α))
=

An
(ω ∧ α, dϕ
n
∧ η ∧ α) +

M
(ω ∧ α, ϕ
n
ω ∧ α),
8 K¨ ahler Hyperbolic Spaces 109
where A
n
:= {p ∈ M | n ≤ r(p) ≤ n + 1} contains the support of dϕ
n
. The
L
2
-norm of ω∧α is finite since ω is parallel and, therefore, uniformly bounded.
Hence the second term on the right tends to the square of the L
2
-norm of ω∧α
as n →∞. Now

An

(ω ∧ α, dϕ
n
∧ η ∧ α)

≤ (cµ(cn +c) +c) · |ω|

An
|α|
2
.
Hence the assertion follows if the limes inferior of the sequence on the right
hand side is 0. If the latter would not hold, then there would be an ε > 0 such
that
|α|
2
2
=

M
|α|
2
=
¸
n

An
|α|
2

¸
ε
(cµ(cn +c) +c) · |ω|

= ∞.
This would be in contradiction to the assumption that α is square integrable.

8.9 Vanishing Theorem (Cao–Xavier [CX], Jost–Zuo [JZ]). If ω is d(O(µ)),
where
¸

1
1/µ(n) = ∞, then
H
k
2
(M, E) = 0 for k = dim
C
M.
Proof. For k < m this follows from Theorem C.49 and Lemma 8.8. For k > m,
vanishing follows from Poincar´e duality.
8.10 Main Lemma. Suppose ω = dη with |η|

< ∞. Let α ∈ L
2
(A
k
(M, E))
be in the domain of ∆
d,max
and suppose that k = m or that α ⊥ H
m
2
(M, E).
Then
(∆
d,max
α, α)
2
≥ λ
2
|α|
2
2
with λ
2
= c(m)/|η|
2

> 0.
Proof. Assume first that k = m+s > m. The s-th power of the Lefschetz map
defines a parallel field of isomorphisms A
m−s
(M, E) → A
m+s
(M, E), see the
discussion in Section 5. In particular, there is β ∈ L
2
(A
m−s
(M, E)) such that
α = ω
s
∧ β.
Since E is flat, ∆
d
= (d + d

)
2
, hence ∆
d,max
= ∆
d,min
, by Theorem C.19.
Hence we can assume that α and β are smooth with compact support. Let
θ := η ∧ ω
s−1
∧ β.
Then
dθ = ω
s
∧ β −η ∧ ω
s−1
∧ dβ =: α −α

.
Now |θ| ≤ |η||ω
s−1
||β|, hence
|θ|
2
≤ |η|


s−1
||β|
2
,
where we note that |ω
s−1
| is constant. Since L
±s
is parallel, there is a positive
constant c
0
such that |L
s
γ| ≤ c
0
|γ| for all γ ∈ A
m−s
(M, E) and |L
−s
γ| ≤ c
0
|γ|
110 Lectures on K¨ ahler Manifolds
for all γ ∈ A
m+s
(M, E). Since L
s
commutes with ∆
d
, we have L
s

d
β = ∆
d
α,
hence
|dβ|
2
2
≤ (∆
d
β, β)
2
≤ |∆
d
β|
2
|β|
2
≤ c
2
0
|∆
d
α|
2
|α|
2
.
We conclude that

|
2
≤ c
0
|η|


s−1
||∆
d
α|
1/2
2
|α|
1/2
2
.
We also have
(d

α, θ)
2
≤ |d

α|
2
|θ|
2
≤ (∆
d
α, α)
1/2
2
|θ|
2
≤ |η|


s−1
||∆
d
α|
1/2
2
|α|
1/2
2
|β|
2
≤ c
0
|η|


s−1
||∆
d
α|
1/2
2
|α|
3/2
2
.
Therefore
|α|
2
2
= (α, dθ + α

)
2
= (α, dθ)
2
+ (α, α

)
2
≤ (d

α, θ)
2
+|α|
2

|
2
≤ 2c
0
|η|


s−1
||∆
d
α|
1/2
2
|α|
3/2
2
.
Since ∆
d,max
is self-adjoint and non-negative, this proves the asserted inequality
in the case k > m. By applying Poincar´e duality we conclude that it also holds
in the case k < m.
We note that for a differential form α of pure degree, we have
α ∈ dom(d +d

)
max
iff α ∈ domd
max
∩ domd

max
.
Since (d +d

)
max
= (d +d

)
min
, the above inequality shows that, for k = m,
|d
max
α|
2
2
+|d

max
α|
2
2
≥ λ
2
|α|
2
2
for any differential k-form α ∈ dom(d +d

)
max
. Now imd ⊂ domd
max
, hence
domd
max
= domd
max
∩ (ker d

max
+ imd)
= domd
max
∩ ker d

max
+ imd.
Since imd ⊂ ker d
max
, it follows that the image of d
max
on differential forms of
degree m−1 is equal to the image of its restriction to
domd
max
∩ ker d

max
∩ L
2
(A
m−1
(M, E)).
By the above estimate we have
|d
max
α|
2
2
≥ λ
2
|α|
2
2
8 K¨ ahler Hyperbolic Spaces 111
for any α in the latter space. In particular, the image of d
max
on differential
forms of degree m−1 is closed. Therefore it contains imd ∩ L
2
(A
m
(M, E)).
There is a similar discussion for d

max
on forms of degree m+ 1. Hence we
can represent any γ ∈ dom∆
d,max
∩L
2
(A
m
(M, E)) perpendicular to H
m
2
(M, E)
as
γ = d
max
α +d

max
β,
where d
max
α ∈ imd ⊥ imd

÷ d

max
β and
d

max
α = d
max
β = 0, |d
max
α|
2
≥ λ
2
|α|
2
, |d

max
β|
2
≥ λ
2
|β|
2
.
We have
γ ∈ dom∆
d,max
= dom∆
d,min
⊂ dom(d +d

)
min
= dom(d +d

)
max
.
Considering degrees, we get
d
max
α ∈ domd

max
, d

max
β ∈ domd
max
,
and hence α, β ∈ dom∆
d,max
. In conclusion,
|γ|
2
2
= |d
max
α|
2
2
+|d

max
β|
2
2
= (∆
d,max
α, α)
2
+ (∆
d,max
β, β)
2
≤ |∆
d,max
α|
2
|α|
2
+|∆
d,max
β|
2
|β|
2
≤ λ
−2
· (|∆
d,max
α|
2
2
+|∆
d,max
β|
2
2
)
= λ
−2
· (|d

max
γ|
2
2
+|d
max
γ|
2
2
) = λ
−2
· (∆γ, γ)
2
.
Let E

, E

→M be Hermitian vector bundles over M and D be a differential
operator of first order from E(E

) to E(E

). Let F →M be a further Hermitian
vector bundle with Hermitian connection ∇
21
. Define
S
F
: E

⊗T

M ⊗F →E

⊗F, S
F
(α ⊗dϕ ⊗σ) = (S(dϕ)α) ⊗σ, (8.11)
where S is the principal symbol of D, see (C.10). The twist of D by ∇ is the
differential operator D

: E(E

⊗F) →E(E

⊗F) defined by
D

(α ⊗σ) = (Dα) ⊗σ +S
F
(α ⊗∇σ). (8.12)
8.13 Example. For D = d +d

, we have
S
F
(α ⊗∇σ) :=
¸
{(X

j
∧ α) ⊗∇
Xj
σ −(X
j
α) ⊗∇
Xj
σ},
where (X
j
) is a local orthonormal frame of TM.
21
We are short of symbols since D is in use already.
112 Lectures on K¨ ahler Manifolds
If ∇, ∇
0
are Hermitian connections on F and A = ∇−∇
0
, then
(D

−D

0
)(α ⊗σ) = S
F
(α ⊗Aσ) =: S
A
(α ⊗σ) (8.14)
is of order 0, a potential perturbing D

0
. We apply this in the case where
F = M C is the trivial line bundle over M with the canonical Hermitian
metric and canonical flat and Hermitian connection ∇
0
. Let ζ be a real valued
differential one-form on M. Then
∇ = ∇
0
+A, (8.15)
where
Aσ := iζ ⊗σ, (8.16)
is a connection form for F. Since iζ is purely imaginary, ∇ is Hermitian. The
curvature form of ∇ is idζ. Vice versa, if ∇ is a Hermitian connection on a
Hermitian line bundle F →M and the curvature form of ∇ is exact, say equal
to idζ, then ∇
0
= ∇− iζ ⊗ id is a flat Hermitian connection (and F is trivial
if M is simply connected).
8.17 Lemma. Suppose ω = dη with |η|

< ∞. Let D = ∂ +∂

on A

(M, E)
and λ be as in Lemma 8.10. Let F = M C be the trivial line bundle over M
with ∇
0
and ∇ = ∇
0
+iζ ⊗id as above. Assume that |S
A
β| ≤ c · |β| for some
positive constant c < λ/

2. Then
ker D
max
= 0 =⇒ ker D

max
= 0.
Proof. Assume that ker D
max
= 0. Since ∇
0
is flat, we have
|D

0
β|
2
2
= (∆

β, β)
2
=
1
2
(∆
d
β, β)
2

λ
2
2
|β|
2
2
for all β ∈ A

c
(M, E ⊗F), by Lemma 8.10. Hence, for all such β,
|D

β|
2
= |D

0
β +S
A
β|
2
≥ |D

0
β|
2
−|S
A
β|
2
≥ (λ/

2 −c)|β|
2
.
Now D

min
= D

max
, see Theorem C.19. Hence ker D

max
= 0 as asserted.
8.2 Non-Vanishing of Cohomology. We start with a little detour (and a
corresponding change in notation) and discuss Atiyah’s L
2
-index theorem [At].
Let
˜
M →M be a normal Riemannian covering, where M and
˜
M are connected
and M is closed. Let
˜
E →
˜
M be the pull back of a Hermitian vector bundle
E →M. Then the group Γ of covering transformations of
˜
M →M acts on
˜
E,
hence on sections of
˜
E by γ · σ := γ ◦ σ ◦ γ
−1
.
8 K¨ ahler Hyperbolic Spaces 113
Let H ⊂ L
2
(
˜
M,
˜
E) be a closed subspace and (Φ
n
) be an orthonormal basis
of H. Then the function
˜
f :
˜
M →[0, ∞],
˜
f(p) =
¸

n
(p)|
2
, (8.18)
does not depend on the choice of (Φ
n
). It follows that
˜
f is Γ-invariant if H is,
and then
˜
f is the lift of a function f on M and we set
dim
Γ
H =

M
f, (8.19)
the Γ-dimension of H. It is important that dim
Γ
H = 0 iff H = 0.
Let E
±
→ M be Hermitian vector bundles and D: E(E
+
) → E(E

) be a
differential operator. Let
˜
E
±
and
˜
D be the pull backs of E
±
and D to
˜
M.
Then the kernels of
˜
D and its formal adjoint
˜
D

are Γ-invariant. By definition,
the Γ-index of
˜
D is
ind
Γ
˜
D = dim
Γ
ker
˜
D −dim
Γ
ker
˜
D

. (8.20)
The Γ-index of
˜
D is well-defined if one of the dimensions on the right is finite.
8.21 Theorem (Atiyah [At]). If D is elliptic, then the Γ-dimensions on the
right hand side of (8.20) are finite and
ind
Γ
˜
D = ind D.
Theorem 8.21 applies in the case of non-positively curved closed K¨ ahler
manifolds and shows that their Euler characteristic has the right sign. More
precisely, we have the following result.
8.22 Theorem (Cao–Xavier [CX], Jost–Zuo [JZ]). Let M be a closed K¨ ahler
manifold and
˜
M → M be its universal cover. If the K¨ ahler form ˜ ω of
˜
M is
d(O(µ)), where
¸

1
1/µ(n) = ∞, then
(−1)
m
χ(M, E) ≥ 0.
Proof. By Hodge theory, χ(M, E) = ind D, where D = d+d

from A
even
(M, E)
to A
odd
(M, E). We have
˜
D = d+d

, but now from A
even
(
˜
M,
˜
E) to A
odd
(
˜
M,
˜
E).
By Theorem 8.21, ind D = ind
Γ
˜
D, where Γ is the fundamental group of M.
By Theorem 8.9, there are no non-trivial square integrable harmonic forms on
˜
M of degree different from the middle dimension. Hence
(−1)
m
ind
Γ
˜
D = dim
Γ
H
m
(M, E) ≥ 0.
Gromov uses an extended version of (part of) Theorem 8.21. He considers
the twist of the pull back of a specific bundle E with the trivial complex line
114 Lectures on K¨ ahler Manifolds
bundle endowed with a Hermitian connection ∇ which is not invariant under
Γ, whereas the curvature of ∇ still is.
For the reader’s convenience, but without proof, we state a more general
version of Gromov’s extension. Let
˜
E →
˜
M be a Hermitian vector bundle of
rank r, not necessarily the pull back of a Hermitian vector bundle over M. Let
G be an extension of Γ by a compact group K,
1 −→K −→G
ρ
−→Γ −→1. (8.23)
Suppose that G acts by Hermitian isomorphisms on
˜
E such that
π ◦ g = ρ(g) ◦ π for all g ∈ G, (8.24)
where π denotes the projection of
˜
E. Then G acts on L
2
(
˜
M,
˜
E) by
g · σ := g ◦ σ ◦ ρ(g)
−1
. (8.25)
Let H be a G-invariant subspace of L
2
(
˜
M,
˜
E). Then the function
˜
f as in
(8.18) is Γ-invariant, and we define the G-dimension of H by the integral of the
corresponding function f on M as in (8.19).
Let
˜
E
±
be Hermitian vector bundles with Hermitian G-actions satisfying
(8.24) (where now π denotes the projections of
˜
E
±
). Suppose that
˜
D is a
differential operator from E(
˜
E
+
) to E(
˜
E

) commuting with the actions of G,
˜
D(g · σ) = g · (
˜
Dσ). (8.26)
Then the kernels of
˜
D and
˜
D

are invariant under G, and we define the G-index
of
˜
D as in (8.20). The case considered in Theorem 8.21 corresponds to the
trivial extension G = Γ. The extension we state concerns the case where
˜
E =
˜
E
+

˜
E

is a graded Dirac bundle and
˜
D
+
: E(
˜
E
+
) → E(
˜
E

) the associated
Dirac operator in the sense of Gromov and Lawson, see Subsection C.24.
8.27 Theorem. Suppose that
˜
E is a graded Dirac bundle and that the Hermi-
tian action of G on
˜
E satisfies (8.24) and leaves invariant connection, Clifford
multiplication, and splitting
˜
E =
˜
E
+

˜
E

of
˜
E. Then the associated Dirac
operator
˜
D satisfies (8.26), the G-dimensions of ker
˜
D
+
and ker
˜
D

are finite,
the canonical index form ˜ α of
˜
D
+
is the pull back of a form α on M, and
ind
G
(
˜
D
+
) =

M
α.
About the proof. It seems that the arguments in Section 13 of [Ro] generalize
to the setting of Theorem 8.27.
8.28 Remark. We will need Theorem 8.27 for the Hirzebruch–Riemann–Roch
formula, that is, for the twisted Dolbeault operator ∂ + ∂

(times

2) on the
Dirac bundle
A
p,∗
(M,
˜
E) = A
0,∗
(M, C) ⊗A
p,0
(M, C) ⊗
˜
E,
8 K¨ ahler Hyperbolic Spaces 115
with splitting into forms with ∗ even and odd, respectively, see Example C.27.2.
In this case, the canonical index form is
Td(M) ∧ ch(A
p,0
(M, C) ⊗
˜
E) = Td(M) ∧ ch(A
p,0
(M, C)) ∧ ch(
˜
E),
where Td and ch denote Todd genus and Chern character, see Theorem III.13.15
in [LM].
Before returning to the discussion of K¨ ahler manifolds, we discuss a situation
where a group G as above arises in a natural way. To that end, suppose in
addition that M is simply connected. Let F →
˜
M be a Hermitian line bundle
with Hermitian connection ∇such that the curvature form ω of ∇is Γ-invariant.
8.29 Lemma. Let p ∈ M, γ ∈ Γ, and u: F
p
→ F
γp
be a unitary map. Then
there is a unique lift of γ to an isomorphism g of F over γ with g
p
= u and
such that g preserves Hermitian metric and connection of F.
Sketch of proof. Let q ∈ M and y ∈ F
q
. Choose a piecewise smooth path c
in M from p to q. Let x be the element in F
p
which is parallel to y along c.
Set g(y) = z, where z is parallel to ux along γ ◦ c. The dependence of parallel
translation on curvature together with the Γ-invariance of ω shows that z only
depends on the homotopy class of c. Since M is simply connected, g is well
defined. By definition, u preserves parallel translation, hence ∇. Since ∇ is
Hermitian, g preserves the Hermitian metric of F.
It follows from Lemma 8.29 that there is an extension
1 →U(1) →G →Γ →1 (8.30)
of Γ by U(1) which lifts the action of Γ on M to an action of G on F as in
(8.24). Furthermore, the action of G preserves metric and connection of F.
We now return to the discussion of K¨ ahler-hyperbolic K¨ ahler manifolds. We
let M be a closed and connected K¨ ahler manifold and E →M be a flat vector
bundle. We let
˜
M → M be the universal covering of M and
˜
E → M be the
pull back of E.
8.31 Main Theorem. If ˜ ω = dη with |η|

< ∞, then
H
p,q
2
(
˜
M,
˜
E) = 0 if p +q = m.
Proof. Let F =
˜
M C be the trivial line bundle with canonical Hermitian
metric and flat connection ∇
0
, and consider the Hermitian connections

t
= ∇
0
+itη.
The curvature form of ∇
t
is itdη = it˜ ω, where ˜ ω denotes the K¨ahler form of
˜
M with respect to the induced K¨ ahler metric. Since ˜ ω is Γ-invariant, we get
116 Lectures on K¨ ahler Manifolds
extensions G
t
of Γ by U(1) with induced actions on F as in Lemma 8.29 above.
For each fixed t, consider the twist D

t
of the Dolbeault operator D = ∂ +∂

on A
p,∗
(M,
˜
E) by F with connection ∇
t
. With respect to the splitting into
forms with ∗ even and odd, the index form of (D

t
)
+
is
Td(M) ∧ ch(A
p,0
(M, C)) ∧ ch(
˜
E) ∧ ch(F),
where ch F = exp(−tω/2π), see (A.44) and Remark 8.28 above. Hence the
index form is a power series in t, where the coefficient of t
m
in degree dim
R
M
is
rk E
m!(−2π)
m

m
p

· ˜ ω
m
.
By Theorem 8.27, the index ind
Gt
((D

t
)
+
) is the integral of the index form
over M. It follows that ind
Gt
((D

t
)
+
) is a polynomial in t, where the coefficient
of t
m
is non-zero
22
. Hence the zeros of this polynomial are isolated.
On the other hand, H
p,q
2
(
˜
M,
˜
E) = 0 would imply that ind
Gt
(D

t
) = 0 for
all t sufficiently small, by Lemma 8.17.
8.32 Theorem. If ˜ ω = dη with |η|

< ∞, then
(−1)
m
χ(M) =
¸
p+q=m
dim
Γ
H
p,q
2
(
˜
M,
˜
E) > 0,
where Γ denotes the fundamental group of M.
Proof. This is immediate from Theorems 8.21, 8.9, and 8.31, compare the proof
of Theorem 8.22.
22
This is in contrast to the case of compact manifolds, where the index of elliptic differential
operators is integral and would not depend on t.
9 Kodaira Embedding Theorem
Let M be a closed complex manifold of complex dimension m. Recall that a
holomorphic line bundle E → M is positive or negative if its first Chern class
c
1
(E) has a positive or negative representative, respectively, see Definition 5.59.
It is immediate that a holomorphic line bundle is positive iff its dual is negative.
In Examples 5.60 we showed that the tautological bundle U →CP
m
is negative
and that the hyperplane bundle H →CP
m
is positive.
If M admits a positive holomorphic line bundle, then a positive representa-
tive of its first Chern class is a K¨ ahler metric on M. In this sense the topic of
this section belongs to K¨ ahler geometry.
9.1 Remark. We say that a cohomology class in H
2
(M, C) is integral if it is in
the image of the canonical morphism H
2
(M, Z) →H
2
(M, C). If M is a K¨ ahler
manifold, then a cohomology class in H
2
(M, C) is the first Chern class of a
holomorphic line bundle iff it is of type (1, 1) and integral, see e.g. Proposition
III.4.6 in [Wel]. It follows that a closed complex manifold M admits a positive
holomorphic line bundle iff M has a K¨ ahler metric g with integral K¨ ahler class
[ω]. Then M is called a Hodge manifold.
Closed oriented surfaces are Hodge manifolds: Let S be such a surface and
g be a Riemannian metric on S. Then the rotation by a positive right angle, in
the tangent spaces of S, is a parallel complex structure on S and turns S into
a K¨ ahler manifold. Since S is of real dimension two, H
2
(S, R)

= R. Now the
image of H
2
(S, Z) in H
2
(S, R) is a lattice, hence a proper rescaling of g has an
integral K¨ahler class. Hence S is a Hodge manifold. By the latter argument we
also get that a closed K¨ahler manifold M with first Betti number b
1
(M) = 1 is
a Hodge manifold.
The pull back of a positive holomorphic line bundle along a holomorphic
immersion is positive. In particular, if M admits a holomorphic immersion into
CP
n
for some n ≥ m, then M admits a positive holomorphic line bundle. Vice
versa, Hodge conjectured and Kodaira proved that M admits a holomorphic
embedding into CP
n
for some n ≥ m if M has a positive line bundle. We
present Kodaira’s proof of his embedding theorem; our source is Section VI.4
in [Wel].
Kodaira’s proof uses the following general construction which associates to
a holomorphic line bundle E → M a holomorphic map from an open part M

of M to a complex projective space: Let O(M, E) be the space of holomorphic
sections of E and M

be the open set of points p ∈ M such that σ(p) = 0
for some σ ∈ O(M, E). Since M is compact, O(M, E) is of finite dimension.
Consider the evaluation map
ε: E

→O(M, E)

, ε(ϕ)(σ) = ϕ(σ(p)), (9.2)
where p is the foot point of ϕ. Clearly ε is holomorphic. For ϕ ∈ E

non-zero,
118 Lectures on K¨ ahler Manifolds
ε(ϕ) = 0 precisely for p ∈ M

. Hence we obtain a holomorphic map
k: M

→P(O(M, E)

), k(p) = [ε(ϕ)], (9.3)
where ϕ is any non-zero element of the fiber of E

over p.
There is a somewhat less abstract way of describing k. Let B = (σ
0
, . . . , σ
k
)
be a basis of O(M, E) and ϕ be a nowhere vanishing holomorphic section of
E

over an open subset U of M. Then the holomorphic map
k
B
: U →CP
k
, k
B
(p) = [ϕ ◦ σ
0
, . . . , ϕ ◦ σ
k
], (9.4)
does not depend on the choice of ϕ. Hence k
B
is well defined on M

. It is clear
that k
B
corresponds to k in the coordinates of O(M, E)

defined by the basis
B.
We say that a holomorphic line bundle E → M is very ample if, in the
above construction, M

= M and k (or k
B
) is an embedding. We say that E
is ample if some positive power of E is very ample.
9.5 Exercise. Show that E

is isomorphic to the pull back under k of the tau-
tological bundle over P(O(M, E)

). Conclude that ample bundles are positive.
Kodaira proves the following more precise version of the embedding theo-
rem.
9.6 Kodaira Embedding Theorem. Let M be a closed complex manifold
and E →M be a holomorphic line bundle. If E is positive, then E is ample.
Let E be a holomorphic line bundle over M and p be a point in M. Let
ϕ: U → E

be a nowhere vanishing holomorphic section of E

, where U is an
open subset of M containing p. Then the 1-jet of a smooth section σ of E at
p with respect to ϕ is
J
1
ϕ,p
(σ) := ((ϕ ◦ σ)(p), (ϕ ◦ σ)

(p)) ∈ C ⊕(T

p
M ⊗C), (9.7)
where (ϕ◦ σ)

denotes the derivative of ϕ◦ σ. The proof that positive holomor-
phic line bundles are ample relies on the following lemma.
9.8 Lemma. Let E be a line bundle over M. Suppose that
1) for any two different points p, q ∈ M, the map
J
0
p,q
: O(M, E) →E
p
⊕E
q
, J
0
p,q
(σ) = (σ(p), σ(q)),
is surjective and that
2) for any point p ∈ M and section ϕ about p as above, the map
J
1
ϕ,p
: O(M, E) →C ⊕(T

p
M ⊗C)
is surjective. Then E is very ample.
9 Kodaira Embedding Theorem 119
Proof. We first observe that M

= M, by Assumption 1). For a basis B =

0
, . . . , σ
k
) of O(M, E), we consider k
B
: M →CP
k
as above.
Let p, q be different points of M. By Assumption 1), there are global holo-
morphic sections σ
p
and σ
q
of E with
σ
p
(p) = 0, σ
p
(q) = 0 and σ
q
(q) = 0, σ
q
(p) = 0.
Then σ
p
and σ
q
are linearly independent. If we choose B with σ
0
= σ
p
and
σ
1
= σ
q
, then clearly k
B
(p) = k
B
(q). It follows that k
B
is injective, for any
choice of basis B.
Let p be a point in M and ϕ be a holomorphic section of E

in an open
neighborhood U of p as above. By Assumption 2), there are holomorphic
sections σ
0
, . . . , σ
m
in O(M, E) such that J
1
ϕ,p

0
) = (1, 0) and J
1
ϕ,p

i
) =
(0, β
i
), where (β
1
, . . . , β
m
) is a basis of T

p
M ⊗ C. Since J
1
ϕ,p
is a linear map,
σ
0
, . . . , σ
m
are linearly independent in O(M, E). Therefore we can complete
them to a basis B of O(M, E). With respect to affine coordinates (z
1
, . . . , z
k
)
on {z
0
= 0} ⊂ CP
k
, we get
dk
B
(p) = ((β
1
, . . . , β
m
, (ϕ ◦ σ
m+1
)

(p), . . . , (ϕ ◦ σ
k
)

(p)).
Since (β
1
, . . . , β
m
) are linearly independent, dk
B
(p) has rank m. Hence k
B
is
an immersion. Now M is compact, hence k
B
is an embedding.
9.1 Proof of the Embedding Theorem. We use notation and results from
Subsection 3.4, in particular Examples 3.45 and Lemmas 3.46 and 3.47. We
denote the blow up of M in a point p by M
p
and use an index p for objects
associated to M
p
. For points p = q in M, we let M
pq
= (M
p
)
q
and π
pq
: M
pq

M be the canonical projection. There is a natural identification of M
pq
with
M
qp
such that π
pq
= π
qp
. We let L
pq
be the holomorphic line bundle associated
to the hypersurface π
−1
pq
(p) ∪ π
−1
pq
(q).
9.9 Lemma. Let E, F be holomorphic line bundles over M and suppose E > 0.
Then given k
0
≥ 1 there is n
0
≥ 0 such that
π

p
E
n
⊗π

p
F ⊗(L

p
)
k
> 0 and π

pq
E
n
⊗π

pq
F ⊗(L

pq
)
k
> 0
for all p = q ∈ M, n ≥ n
0
, and k ∈ {1, . . . , k
0
}.
Proof. Choose a Hermitian metric on E such that the curvature of its Chern
connection satisfies iΘ
E
> 0, see Lemma 5.61. Fix a Hermitian metric on F
and denote the curvature of its Chern connection by Θ
F
. The Chern connection
of the tensor product E
1
⊗ E
2
of two Hermitian holomorphic vector bundles
E
1
, E
2
→M with induced Hermitian metric
(e
1
⊗e
2
, e

1
⊗e

2
) = (e
1
, e

1
) (e
2
, e

2
)
120 Lectures on K¨ ahler Manifolds
is given by the usual product rule, hence its curvature is the sum of the cur-
vatures of the Chern connections on E
1
and E
2
, Θ = Θ
1
+ Θ
2
. Since M is
compact and iΘ
E
> 0, there is an n
0
> 0 such that i(n
0
Θ
E
+ Θ
F
) > 0. In
particular, E
n
⊗F > 0 for all n ≥ n
0
.
Let p ∈ M and consider the projection π
p
: M
p
→M. The Chern connection
of the pull back of a Hermitian metric on the pull back of a holomorphic vector
bundle is the pull back of the Chern connection of the original Hermitian metric
on the original bundle. Hence the Chern connection of the pull back metric
on π

p
E
n
⊗ π

p
F has curvature Θ
n
equal to the pull back of the curvature on
E
n
⊗F. Therefore iΘ
n
(v, Jv) > 0 for v not in the kernel of π
p∗
, that is, except
for v = 0 or v tangent to S
p
.
Positive powers of L

p
will bring positivity everywhere. To get n
0
indepen-
dent of the chosen point p ∈ M, we proceed as follows: Choose holomorphic
coordinates z : U → U

on M, where U

contains the ball of radius 4 about 0
in C
m
. Set
U
i
= {p ∈ U | |z(p)| < i}.
Fix a cut off function χ on M with χ = 1 on U
2
and χ = 0 on M \ U
3
.
For p in the closure U
1
of U
1
, we use z − z(p) as holomorphic coordinates
in the construction of the blow up M
p
. Via the canonical projection π
p
, we
identify M
p
\ π
−1
p
(U
1
) with M \ U
1
. In this sense, we view L

p
restricted to
M
p
\ π
−1
p
(U
1
) as a holomorphic line bundle over M \ U
1
. The holomorphic
section Φ

0
dual to the holomorphic section Φ
0
as in Example 3.45.4 turns it
into a trivial bundle. We define a Hermitian metric h

p
on this part of L

p
by associating length 1 to Φ

0
. Then Φ

0
is parallel with respect to the Chern
connection of h

p
, and hence the curvature of the latter vanishes identically.
Since the hyperplane bundle H →CP
m−1
is positive, there is a Hermitian
metric h on H such that the curvature of its Chern connection satisfies iΘ
h
>
0. Over π
−1
p
(U
4
), we have L

p
= σ

p
H, where σ
p
: π
−1
p
(U
4
) → CP
m−1
is the
canonical projection. Therefore we view h

p
= σ

p
h as a Hermitian metric on
that part of L

p
. The curvature Θ

p
of the Chern connection of h

p
is the pull
back of Θ
h
, and hence iΘ

p
≥ 0 and iΘ

p
> 0 on the tangent bundle of S
p
. We
now let
h
p
= (χ ◦ π
p
)h

p
+ (1 −χ ◦ π
p
)h

p
.
Then the curvature Θ
p
of the Chern connection of h
p
vanishes on M
p

−1
p
(U
3
)
and Θ
p
= Θ

p
on π
−1
(U
2
). Since σ
p
(q) is smooth in p ∈ U
1
and q ∈ U
3
\ U
2
,
the family of Hermitian metrics h

p
, p ∈ U
1
, is uniformly bounded on U
3
\ U
2
,
together with derivatives of any order. It follows that Θ
p
is uniformly bounded
on U
3
\ U
2
, independently of p ∈ U
1
.
The curvature of the Chern connection of the induced Hermitian metric on
E
n
⊗F ⊗(L

)
k
is Θ = Θ
n
+kΘ
p
. Clearly iΘ is positive outside U
3
\ U
2
. If we
increase the above n
0
if necessary, the positivity of iΘ
n
outweighs the possible
negativity of kiΘ
p
on U
3
\ U
2
. By the uniform boundedness of Θ
p
, n
0
can be
9 Kodaira Embedding Theorem 121
chosen independently of p ∈ U
1
. Now M is compact, hence can be covered by
a finite number of sets U
1
as in the argument.
The argument in the case of two points p = q in a set U
1
as above is similar.
We use the same construction, but now with two pull backs of H corresponding
to the two points p and q. For p and q uniformly far apart, we can apply the
argument for the case of one point twice.
We now come to the final steps of the proof of the embedding theorem. We
will assume some elementary facts from sheaf theory. For a sheaf F over M
and an open subset U of M, we denote by F(U) the space of sections of F over
U. For p ∈ M, we denote by F
p
the stalk of F at p.
Let O be the structure sheaf of M, that is, the sheaf of rings of germs of
holomorphic functions on M. For p ∈ M, let ´
p
and ´
2
p
be the subsheaves of
ideals in O of germs of holomorphic functions which vanish at p respectively
vanish at p at least of second order. Then we have a short exact sequence
0 →´
2
p
→O →O/´
2
p
→0. (9.10)
For q = p, the stalk (O/´
2
p
)
q
= 0. Moreover, the map
J
1
p
: O
p
→C ⊕(T

p
M ⊗C), J
1
p
(f) = (f(p), f

(p)), (9.11)
induces an isomorphism
(O/´
2
p
)
p
→C ⊕(T

p
M ⊗C). (9.12)
Let F →M be a holomorphic line bundle. Denote by O(F) the sheaf of germs
of holomorphic sections of F. Since O(F) is a locally free sheaf of modules over
the structure sheaf O of M, we obtain an induced short exact sequence
0 →O(F) ⊗´
2
p
→O(F) →O(F) ⊗(O/´
2
p
) →0, (9.13)
where the tensor product is taken over O. We may view O(F) ⊗ ´
2
p
as sheaf
of germs of holomorphic sections of F which vanish at least of second order at
the distinguished point p. The stalks in q = p are
(O(F) ⊗´
2
p
)
q
= O(F)
q
and (O(F) ⊗(O/´
2
p
))
q
= {0}. (9.14)
To identify the stalk of O(F) ⊗ (O/´
2
p
) at p, we observe that the map J
1
ϕ,p
from Lemma 9.8 is also defined on the stalk O(F)
p
and that it induces an
isomorphism
(O(F) ⊗(O/´
2
p
))
p
→C ⊕(T

p
M ⊗C). (9.15)
9.16 Lemma. Let E →M be a positive holomorphic line bundle. In Lemma 9.9,
let F be the canonical bundle of M, F = K, and k
0
= m+1. Then E
n
, n ≥ n
0
,
satisfies the assumptions of Lemma 9.8 and hence E is ample.
122 Lectures on K¨ ahler Manifolds
Proof. The proof of Assumption 1) in Lemma 9.8 is similar to and easier than
the proof of Assumption 2). Therefore we concentrate on Assumption 2) and
leave the proof of Assumption 1) to the reader.
Let p ∈ M. By (9.14), the restriction map induces an isomorphism from
the space H
0
(M, O(E
n
) ⊗(O/´
2
p
)) of global holomorphic sections of O(E
n
) ⊗
(O/´
2
p
) onto the stalk (O(E
n
) ⊗(O/´
2
p
))
p
. Hence it remains to show that
O(M, E
n
) = H
0
(M, O(E
n
)) →H
0
(M, O(E
n
) ⊗(O/´
2
p
)) (9.17)
is surjective, by (9.15). Since points in M have codimension m and do not
correspond to line bundles if m > 1, we will need to pass to blow ups of M to
show the surjectivity in (9.17).
Let
˜
M = M
p
be the blow up of M in p. Let π:
˜
M → M be the canonical
projection and
˜
E = π

E. Denote by
˜
O the structure sheaf of
˜
M and by ´
S
and ´
2
S
the subsheaves of
˜
O of germs of holomorphic functions which vanish
along S respectively vanish at least of second order along S. Composition with
π induces injective morphisms π

: O →
˜
O and π

µ
: O(E
n
) →O(
˜
E
n
). Now E
n
is trivial in a neighborhood of p in M, and the pull back of a local trivialization
defines a trivialization of
˜
E
n
in a neighborhood of S. From the coordinate
description of
˜
M in Subsection 2.4 it is then clear that a holomorphic section σ
of E
n
vanishes of at least second order at p iff σ ◦ π vanishes of at least second
order about S. It follows that π induces morphisms
π

λ
: O(E
n
) ⊗´
2
p
→O(
˜
E
n
) ⊗ ´
2
S
and
π

ρ
: O(E
n
) ⊗(O/´
2
p
) →O(
˜
E
n
) ⊗(
˜
O/´
2
S
).
Hence we get a commutative diagram
0 −−−−→ O(
˜
E
n
) ⊗´
2
S
−−−−→ O(
˜
E
n
) −−−−→ O(
˜
E
n
) ⊗
˜
O/´
2
S
−−−−→ 0
¸

π

λ
¸

π

µ
¸

π

ρ
0 −−−−→ O(E
n
) ⊗´
2
p
−−−−→ O(E
n
) −−−−→ O(E
n
) ⊗O/´
2
p
−−−−→ 0
(9.18)
of short exact sequences of sheaves.
Clearly π

λ
and π

µ
are injective. We claim that they are also surjective. This
is clear in the case m = 1 since then blowing up is trivial and π is biholomorphic.
For m > 1, a section ˜ σ of
˜
E
n
over
˜
M restricts to a section σ
p
of
˜
E
n
over
˜
M\S =
M\{p}. By Hartogs’ theorem [GH, page 7], σ
p
extends to a holomorphic section
σ of E
n
over M. By the continuity of σ and ˜ σ, we have π

µ
σ = ˜ σ. Moreover, if
˜ σ vanishes of order at least 2 along S, then also σ at p. Hence π

λ
and π

µ
are
surjective, hence isomorphic. The five lemma implies that π

ρ
is an isomorphism
as well.
9 Kodaira Embedding Theorem 123
A fortiori, the vertical morphisms in (9.18) induce isomorphisms between
the corresponding cohomology groups in the long exact sequences associated to
the horizontal short exact sequences. Thus by (9.17), it remains to show that
H
0
(
˜
M, O(
˜
E
n
)) →H
0
(
˜
M, O(
˜
E
n
) ⊗(
˜
O/´
2
S
))
is surjective. By the long exact sequence, this follows if H
1
(
˜
M, O(
˜
E
n
) ⊗´
2
S
) =
0.
To show this we observe first that ´
S
= O(L

). In the notation of Exam-
ple 3.45.4, this follows in W
i
, i ≥ 1, since there f ∈ ´
S
can be expressed as
f = f
i
z
i
with f
i
holomorphic. That is, the functions z
i
serve as a frame over
W
i
and f
i
z
i
= f = f
j
z
j
over W
i
∩ W
j
. Now
˜
E
n

˜
K

⊗(L

)
2
> 0 by Lemmas
3.46 and 3.47, hence
H
1
(
˜
M, O(
˜
E
n
⊗(L

)
2
)) = 0
by the Kodaira vanishing theorem 5.65.
9.2 Two Applications. The Kodaira embedding theorem has many appli-
cations. We discuss two immediate ones.
9.19 Corollary. Let M be a closed complex manifold and M
p
be the blow up
of M in a point p ∈ M. If M admits a holomorphic embedding into a complex
projective space, then M
p
as well.
Proof. Suppose f : M →CP
n
is a holomorphic embedding. Then the pull back
E → M of the hyperplane bundle H over CP
n
is positive. By Lemma 9.9,
π

p
E ⊗ (L

)
k
is positive for k sufficiently large, where π
p
: M
p
→ M is the
natural map. Hence M
p
admits a holomorphic embedding into some complex
projective space, by Theorem 9.6.
9.20 Corollary. Let M be a closed complex manifold and
˜
M →M be a finite
covering of M. If M admits a holomorphic embedding into a complex projective
space, then
˜
M as well.
Proof. Suppose f : M → CP
n
is a holomorphic embedding. Then the pull
back E → M of the hyperplane bundle H over CP
n
is positive, therefore the
pull back
˜
E →
˜
M of E is positive as well. Hence
˜
M admits a holomorphic
embedding into some complex projective space, by Theorem 9.6.
9.21 Remark. For a finite covering
˜
M → M as above, averaging leads to
the converse assertion: If
˜
M admits a holomorphic embedding into a complex
projective space, then also M, see [GH, page 192].
To complete the picture, we quote the theorem of Chow which says that
a complex analytic submanifold of a complex projective space is a smooth
projective variety, see for example [GH, page 167].
Appendix A Chern–Weil Theory
In this appendix, no K¨ ahler classes are in the way and, therefore, connection
and curvature forms are denoted by the more standard ω and Ω, respectively.
Let V
1
, . . . , V
k
and V be vector spaces over the field F ∈ {R, C} and
Φ: V
1
· · · V
k
→V (A.1)
be a k-linear map. Let Λ

F
R
n
be the algebra of alternating forms on R
n
with
values in F. The map

1
, v
1
, . . . , ϕ
k
, v
k
) →(ϕ
1
∧ · · · ∧ ϕ
k
) ⊗Φ(v
1
, . . . , v
k
), (A.2)
where ϕ
i
∈ Λ

F
R
n
and v
i
∈ V
i
, is linear in each of its 2k arguments. Therefore
it gives rise to a k-linear map
Φ
Λ
: (Λ

F
R
n
⊗V
1
) · · · (Λ

F
R
n
⊗V
k
) →Λ

F
R
n
⊗V. (A.3)
We view Λ

F
R
n
⊗ V
i
and Λ

F
R
n
⊗ V as the space of alternating forms on R
n
with values in V
i
and V , respectively. If α
1
, . . . , α
k
are alternating forms on R
n
with values in V
1
, . . . , V
k
and degrees d
1
, . . . , d
k
, respectively, and if we express
α
i
=
¸
ϕ
ij
⊗v
ij
with ϕ
ij
∈ Λ

F
R
n
and v
ij
∈ V
i
, then
Φ
Λ

1
, . . . , α
k
) =
¸
(j1,...,j
k
)

1j1
∧ · · · ∧ ϕ
kj
k
) ⊗Φ(v
1j1
, . . . , v
kj
k
). (A.4)
The degree of Φ
Λ

1
, . . . , α
k
) is d = d
1
+· · · +d
k
. For x
1
, . . . , x
d
∈ R
n
, we have
Φ
Λ

1
, . . . , α
k
)(x
1
, . . . , x
d
)
(A.5)
=
1
d
1
! . . . d
k
!
¸
σ
ε(σ)Φ(α
1
(x
σ(1)
, . . . , x
σ(d1)
), . . . , α
k
(x
σ(d−d
k
+1)
, . . . , x
σ(d)
)).
This is immediate from (A.4) and the definition of the wedge product. It is
also clear from this formula that we are discussing a generalization of the wedge
product introduced in (1.15).
Let G be a Lie group, and suppose that G acts linearly on the vector
spaces V
1
, . . . , V
k
, and V (via given representations, which we suppress in the
notation). Then G acts also on Λ

F
R
n
⊗V ,
g(ϕ ⊗v) := ϕ ⊗(gv), (A.6)
and similarly for Λ

F
R
n
⊗ V
i
. We say that Φ (from (A.1)) is equivariant (with
respect to these G-actions) if
Φ(gv
1
, . . . , gv
k
) = gΦ(v
1
, . . . , v
k
) (A.7)
Appendix A Chern–Weil Theory 125
for all g ∈ G and v
i
∈ V
i
, 1 ≤ i ≤ k. We say that Φ is invariant if it is
equivariant with respect to the trivial action on V . That is, if (A.7) holds
without the factor g on the right hand side. If Φ is equivariant or invariant,
then the induced map Φ
Λ
is equivariant or invariant as well. This is immediate
from (A.4) and (A.6).
A.8 Examples. 1) Via the adjoint representation, G acts on its Lie algebra
g. The Lie bracket [A, B] defines an equivariant bilinear map λ: g g →g.
2) The adjoint representation of the general linear group G = Gl(m, F) on
its Lie algebra g = gl(m, F) = F
m×m
is given by conjugation, Ad
g
A = gAg
−1
.
The product AB of matrices defines an equivariant bilinear map µ: g g →g.
The trace tr: g → F is an invariant linear map since tr(gAg
−1
) = tr A for all
g ∈ G and A ∈ g. Via polarization, the determinant gives rise to an invariant
m-linear map g · · · g →F, see below.
Let P →M be a principal bundle with structure group G. Let E = P
G
V
and E
i
= P
G
V
i
be the vector bundles associated to the representations of
G on V and V
i
, 1 ≤ i ≤ k. Recall that E consists of equivalence classes of
tuples [f, v] with relation [fg, v] = [f, gv] for all f ∈ P, g ∈ G, and v ∈ V ,
and similarly for E
i
. Suppose that Φ is equivariant. Then Φ induces a field of
morphisms
E
1
· · · E
k
→E,
([f, v
1
], . . . , [f, v
k
]) →[f, Φ(v
1
, . . . , v
k
)].
(A.9)
Let α
i
∈ A

(M, E
i
) be differential forms on M with values in E
i
, 1 ≤ i ≤ k.
With respect to a local section τ : U →P of P, we can write (or identify)
α
i
=
¸
ϕ
ij
⊗[τ, v
ij
] =
¸
[τ, ϕ
ij
⊗v
ij
], (A.10)
where the ϕ
ij
are differential forms on U with values in F and the v
ij
are
smooth maps from U to V
i
. We call (α
i
)
τ
=
¸
ϕ
ij
⊗ v
ij
the principal part of
α
i
with respect to τ. If Φ is equivariant, then Φ
Λ
is equivariant as well, and
we obtain an induced field of morphisms on differential forms,
A

(M, E
1
) · · · A

(M, E
k
) →A

(M, E),

1
, . . . , α
k
) →[τ, Φ
Λ
((α
1
)
τ
, . . . , (α
k
)
τ
)] (A.11)
=
¸
[τ, (ϕ
1j1
∧ · · · ∧ϕ
kj
k
)⊗Φ(v
1j1
, . . . , v
kj
k
)],
where the degree is additive in the degrees of the arguments. In the case where
V = F with trivial action of G, we view A

(M, E) = A

(M, F), the space of
differential forms on M with values in F.
Although some of the following holds in greater generality, we now specialize
to the case where V
1
= · · · = V
k
= g with the adjoint action of G and V = F
with the trivial G-action. That is, we consider an invariant k-linear map
Φ: g · · · g →F. (A.12)
126 Lectures on K¨ ahler Manifolds
Recall that for A, B ∈ g,
λ(A, B) := (Ad(e
tA
)B)

t=0
= [A, B]. (A.13)
The induced map on alternating forms associates to alternating forms α and β
with values in g an alternating form
λ
Λ
(α, β) = α ∧
λ
β (A.14)
with values in F, where the notation is from (A.3) on the left hand side and
from (1.15) on the right, compare also Exercise 1.16. On decomposable forms
α = ϕ ⊗A and β = ψ ⊗B with A, B ∈ g,
α ∧
λ
β = (ϕ ∧ ψ) ⊗[A, B]. (A.15)
A.16 Lemma. For A, B
1
, . . . , B
k
∈ g,
¸
i
Φ(B
1
, . . . , A∧
λ
B
i
, . . . , B
k
) = 0.
Proof. Since Φ is invariant,
Φ(Ad(e
tA
)B
1
, . . . , Ad(e
tA
)B
k
) = Φ(B
1
, . . . , B
k
).
Differentiation of this equation yields the claim.
A.17 Corollary. Let α, β
1
, . . . , β
k
be alternating forms with values in g and
of degrees d, d
1
, . . . , d
k
, respectively. Then, with s(i) := (d
1
+ · · · +d
i−1
)d,
¸
i
(−1)
s(i)
Φ
Λ

1
, . . . , α ∧
λ
β
i
, . . . , β
k
) = 0.
Proof. By multilinearity, it is sufficient to consider decomposable forms. For
these, the claim is immediate from (A.4) and Lemma A.16. The sign arises
from moving the differential form part of α to the leading position.
We return to the principal bundle P →M with structure group G and let D
be a connection on P. We interpret the curvature R of D as a differential two-
form on M with values in P
G
g. With respect to a local section τ : U →P,
we denote the principal parts of D and R by ω and Ω, differential one- and
two-forms on U with values in g. By the above, we obtain a differential 2k-form
Φ(P, R) on M with values in F by setting
Φ(P, R) := [τ, Φ
Λ
(Ω, . . . , Ω)]. (A.18)
A.19 Fundamental Lemma. For the curvature R of a connection D on P,
the differential form Φ(P, R) is closed and its cohomology class in H
2k
(M, F)
does not depend on the choice of D.
Appendix A Chern–Weil Theory 127
Proof. Fix a local section τ of P. In the notation of (A.14) and (1.15), the
Bianchi identity says dΩ = Ω ∧
λ
ω, where λ is the Lie bracket of g. By Exer-
cise 1.16, Ω ∧
λ
ω = −ω ∧
λ
Ω. Hence
d(Φ(P, R)) =
¸
[τ, Φ
Λ
(Ω, . . . , dΩ, . . . , Ω)]
=
¸
[τ, Φ
Λ
(Ω, . . . , Ω ∧
λ
ω, . . . , Ω)] = 0,
by (A.17). This shows the first assertion.
Let D
0
and D
1
be connections on P. Then D
t
= (1 −t)D
0
+ tD
1
is also a
connection on P. The principal part of D
t
with respect to τ is
ω
t
= (1 −t)ω
0
+tω
1
= ω
0
+tβ,
where the difference β = ω
1
−ω
0
is a one-form with values in g. We recall the
structure equation

t
= dω
t
+
1
2
ω
t

λ
ω
t
.
By Exercise 1.16, ω

t

λ
ω
t
= ω
t

λ
ω

t
, where the prime indicates differentiation
with respect to t. Hence

t
= dω

t
+
1
2
ω

t

λ
ω
t
+
1
2
ω
t

λ
ω

t
= dω

t
+ ω

t

λ
ω
t
= dβ + β ∧
λ
ω
t
.
With the Bianchi identity and Corollary A.17,
d

¸
i
Φ
Λ
(Ω
t
, . . . , β, . . . , Ω
t
)

=
¸
j<i
Φ
Λ
(Ω
t
, . . . , dΩ
t
, . . . , β, . . . , Ω
t
) +
¸
i
Φ
Λ
(Ω
t
, . . . , dβ, . . . , Ω
t
)

¸
j>i
Φ
Λ
(Ω
t
, . . . , β, . . . , dΩ
t
, . . . , Ω
t
)
=
¸
j<i
Φ
Λ
(Ω
t
, . . . , Ω
t

λ
ω
t
, . . . , β, . . . , Ω
t
)

¸
i
Φ
Λ
(Ω
t
, . . . , β ∧
λ
ω
t
, . . . , Ω
t
)

¸
j>i
Φ
Λ
(Ω
t
, . . . , β, . . . , Ω
t

λ
ω
t
, . . . , Ω
t
)
+
¸
i
Φ
Λ
(Ω
t
, . . . , Ω

t
, . . . , Ω
t
)
= (Φ
Λ
(Ω
t
, . . . , Ω
t
))

,
128 Lectures on K¨ ahler Manifolds
the principal part of Φ(P, R
t
)

. Hence
Φ(P, R
1
) −Φ(P, R
0
) = dT, (A.20)
where T is the transgression form,
T = [τ,
¸
i

1
0
Φ
Λ
(Ω
t
, . . . , β, . . . , Ω
t
)dt]. (A.21)
Note that β is the principal part of B = D
1
−D
0
, a one-form on M with values
in P
G
g. Hence T is well defined on M.
Since the curvature form of a connection D has degree two, Φ(P, R) = 0 if
there is a pair i < j such that Φ is alternating in the corresponding variables,
Φ(A
1
, . . . , A
i
, . . . , A
j
, . . . , A
m
) = −Φ(A
1
, . . . , A
j
, . . . , A
i
, . . . , A
m
).
In the applications of Theorem A.19, we are therefore only interested in the
case where Φ is symmetric and invariant.
Let S
k
(g) be the space of symmetric k-linear forms on g with values in F.
Let F[g] be the algebra of polynomials on g with values in F and F
k
[g] be the
subspace of homogeneous polynomials of degree k. The evaluation along the
diagonal gives rise to an isomorphism
S
k
(g) →F
k
[g],
˜
Φ →Φ, where Φ(A) :=
˜
Φ(A, . . . , A). (A.22)
The inverse of this isomorphism goes under the name polarization. Since the
isomorphism is equivariant with respect to the natural G-actions, G-invariant
symmetric multilinear forms correspond exactly to G-invariant polynomials.
In terms of a given basis (B
1
, . . . , B
m
) of g, a homogeneous polynomial Φ
of degree k on g can be written as
Φ(A) =
¸
µ
f
µ
a
µ(1)
. . . a
µ(k)
, (A.23)
where µ runs over all non-decreasing maps µ: {1, . . . , k} → {1, . . . , m}, the
coefficients f
µ
are in F and the a
j
are the coordinates of A with respect to the
chosen basis, A =
¸
a
j
B
j
. Then the polarization of Φ is
˜
Φ(A
1
, . . . , A
k
) =
1
k!
¸
µ,σ
f
µ
a
σ(1)µ(1)
. . . a
σ(k)µ(k)
, (A.24)
where now, in addition, σ runs over all permutations of {1, . . . , k} and the a
ij
are the coordinates of the vector A
i
. By multilinearity and (A.23),
˜
Φ(A, . . . , A) =
¸
j1,...,j
k
˜
Φ(B
j1
, . . . , B
j
k
)a
j1
. . . a
j
k
=
¸
µ
f
µ
a
µ(1)
. . . a
µ(k)
. (A.25)
Appendix A Chern–Weil Theory 129
We let Λ
even
F
R
n
:= ⊕Λ
2i
F
R
n
and consider
α =
¸
ϕ
j
⊗B
j
∈ Λ
even
F
R
n
⊗g. (A.26)
Since Λ
even
F
R
n
is a commutative algebra, (A.25) implies
˜
Φ
Λ
(α, . . . , α) =
¸
j1,...,j
k
˜
Φ(B
j1
, . . . , B
j
k

j1
∧ · · · ∧ ϕ
j
k
=
¸
f
µ
ϕ
µ(1)
∧ · · · ∧ ϕ
µ(k)
=: Φ
Λ
(α) ∈ Λ
even
F
R
n
.
(A.27)
The definition of Φ
Λ
(α) is very natural, and we could have started with it. On
the other hand, the detour over polarization shows that we are in the framework
of what we discussed before.
Equation A.27 implies that for homogeneous polynomials Φ and Ψ,
(Φ · Ψ)
Λ
(α) = Φ
Λ
(α) ∧ Ψ
Λ
(α). (A.28)
In particular, linear extension to the algebra F[g] of polynomials gives rise to
an algebra morphism
F[g] →Λ
even
F
R
n
, Φ →Φ
Λ
(α). (A.29)
If α is a sum of forms of strictly positive (and even) degrees, then the natural
extension of this map from F[g] to the larger algebra F[[g]] of formal power
series in g over F is well defined. For this we note that Φ
Λ
(α) has degree kd if
Φ is homogeneous of degree k and α is of pure degree d and that alternating
forms on R
n
of degree > n vanish. In both cases, F[g] or F[[g]], if Φ is invariant
under the natural action of G, then also Φ
Λ
(α).
We return again to the principal bundle P with structure group G. Denote
by F[g]
G
⊂ F[g] and F[[g]]
G
⊂ F[[g]] the subalgebra of G-invariant polynomials
and G-invariant formal power series, respectively. Let α ∈ A
even
(M, P
G
g).
By what we said above, α gives rise to a morphism of commutative algebras,
the Weil homomorphism
F[g]
G
→A
even
(M, F), Φ →Φ(P, α), (A.30)
where Φ(P, α) = [τ, Φ
Λ

τ
)] with respect to a local section τ of P, compare
(A.11). Moreover, if α is a sum of forms of strictly positive (and even) degrees,
then the natural extension to F[[g]]
G
is well defined.
A.31 Main Theorem. Let P →M be a principal bundle with structure group
G. Let D be a connection on P and R be the curvature of D. Then the Weil
homomorphism
F[[g]]
G
→A
even
(M, F), Φ →Φ(P, R),
is a morphism of commutative algebras. Moreover,
130 Lectures on K¨ ahler Manifolds
1. Φ(P, R) is closed and its cohomology class in H
even
(M, F) does not depend
on the choice of D;
2. if P

→ M

is a G-principal bundle with connection D

and F : P

→ P
is a morphism over a smooth map f : M

→ M with D

= F

D, then
Φ(P

, R

) = f

Φ(P, R).
Proof. The first part is clear from the discussion leading to (A.30). Assertion
(1) follows from Lemma A.19. Assertion (2) is immediate from the definitions.

A.1 Chern Classes and Character. Let G = Gl(r, C) and g = gl(r, C) =
C
r×r
. We get invariant homogeneous polynomial Φ
k
of degree k on gl(r, C) by
setting
det(tI +A) =
r
¸
k=0
t
r−k
Φ
k
(A). (A.32)
If A is diagonalizable, then Φ
k
(A) is the k-th elementary symmetric function
of the eigenvalues of A. In particular, Φ
0
= 1, Φ
1
= tr, and Φ
r
= det. We note
that Φ
k
(A) is real if A is real.
Let E → M be a complex vector bundle of rank r and P be the bundle of
frames of E. Then the structure group of P is G = Gl(r, C) and P
G
g =
End E. Let D be a connection on E, or, what amounts to the same, on P.
Then the k-th Chern form of E with respect to D is defined as
c
k
(E, D) := Φ
k

P,
i

R
«
, (A.33)
where R is the curvature of D. The total Chern form of E with respect to D is
c(E, D) = det

I +
i

R
«
= 1 +c
1
(E, D) + · · · +c
r
(E, D), (A.34)
where r is the rank of E. The corresponding cohomology classes
c
k
(E) = [c
k
(E, D)] ∈ H
2k
(M, C) and
c(E) = 1 +c
1
(E) + · · · +c
r
(E) ∈ H

(M, C)
(A.35)
are called the Chern classes and the total Chern class of E, respectively.
By definition, Chern classes of isomorphic bundles are equal. If E is trivial
and D is the trivial connection on E, then c
k
(E, D) = 0 for k > 0, and hence
c(E) = 1. Therefore we view the Chern classes as a measure of the deviation
of E from being trivial.
For a complex manifold M with complex structure J, we consider TM
together with J as a complex vector bundle and use the shorthand c
k
(M) and
c(M) instead of c
k
(TM) and c(TM), respectively.
Appendix A Chern–Weil Theory 131
A.36 Proposition. If D is compatible with a Hermitian metric on E, then
the Chern forms c
k
(E, D) are real. In particular, c
k
(E) ∈ H
2k
(M, R) ⊂
H
2k
(M, C).
Proof. With respect to a local orthonormal frame of E, that is, a local section
of P, the principal part Ω of the curvature of D is skew-Hermitian, Ω = −Ω
t
.
Hence
c(E, D) = det
Λ

I +
i


«
= det
Λ

I −
i


«
= c(E, D),
where we use that transposition does not change the determinant.
A.37 Remark. There is a topological construction of Chern classes in H

(M, Z),
see [MS]. The Chern classes we discuss here are their images under the natu-
ral map H

(M, Z) → H

(M, R). In particular, the Chern classes here do not
contain as much information as the ones coming from the topological construc-
tion. On the other hand, the construction here gives explicit differential forms
representing the Chern classes, an advantage in some of the applications in
differential geometry.
A.38 Proposition. The Chern forms satisfy the following properties:
1. (Naturality ) c(f

E, f

D) = f

(c(E, D)).
2. (Additivity ) c(E

⊕E

, D

⊕D

) = c(E

, D

) ∧ c(E

, D

).
3. c
j
(E

, D

) = (−1)
j
c
j
(E, D), where E

= Hom(E, C) is the dual bundle
with induced connection D

.
The corresponding properties of Chern classes follow as a corollary.
Proof of Proposition A.38. The first assertion is immediate fromf


D
= Ω
f

D
.
With respect to local frames of E

and E

over an open subset U ⊂ M, the
principal part of the curvature of D = D

⊕D

is
Ω =

0
0 Ω

.
Over U we have
c(E

⊕E

, D

⊕D

) = det
Λ

I +
i

0
0 I +
i

= det
Λ

I +
i

«
∧ det
Λ

I +
i

«
= c(E

, D

) ∧ c(E

, D

),
compare (A.27). This proves the second assertion. With respect to a local
frame, −Ω
t
is the principal part of the curvature of D

. Hence the last assertion.

132 Lectures on K¨ ahler Manifolds
A.39 Examples. 1) If the bundle E → M is trivial and D is the trivial
connection on E, then c(E, D) = 1. In particular, c(E) = 1. More generally, if
E has a flat connection, then c(E) = 1.
2) If E has rank r and splits as E = E

⊕E

with E

trivial of rank s, then
c
j
(E) = 0 for j > r −s.
3) Let S be the oriented surface of genus g, endowed with a Riemannian
metric and the corresponding Levi-Civita connection D. The rotation J = J
p
by a positive right angle in T
p
S, p ∈ S, is a parallel field of complex structures
which turns S into a K¨ ahler manifold of complex dimension 1 and TS into a
complex line bundle over S. Let p ∈ S and (v, w) be an oriented orthonormal
basis of T
p
S. Then w = Jv and K(p) = −'R(v, w)v, w`, and hence

p
(v, w) = −i K(p),
where K is the Gauss curvature of S. Therefore
c
1
(TS, D) =
K

dA,
where dA denotes the (oriented) area form. The Gauss–Bonnet formula gives

S
c
1
(TS, D) = χ(S) = Euler number of S.
We make this more explicit in the case of S = CP
1
= S
2
. On the open subset
U = {[z
0
, z
1
] ∈ S | z
1
= 0} of S, consider the coordinates x + iy = z := z
0
/z
1
.
Let X = ∂/∂x and Y = ∂/∂y be the corresponding coordinate vector fields.
The complex structure on S turns TS into a complex line bundle. The standard
metric of constant curvature 1 is the real part of the Hermitian metric on TS
which, over U, is given by
h(X, X) =
4
(1 + |z|
2
)
2
. (A.40)
By Proposition 3.21.1, the connection form of the Chern connection D is
ω = h
−1
∂h = (1 + |z|
2
)
2

(1 + |z|
2
)
−2

=
−2zdz
1 + |z|
2
.
By Proposition 3.21.4, the principal part of the curvature is
Ω = ∂ω =
2dz ∧ dz
(1 + |z|
2
)
2
.
Hence the first Chern form
c
1
(TS, D) =
i dz ∧ dz
π(1 + |z|
2
)
2
=
2dx ∧ dy
π(1 + |z|
2
)
2
.
Appendix A Chern–Weil Theory 133
It follows that

S
c
1
(TS, D) =

U
c
1
(TS, D)
=
2
π


0


0
ρ dρdϕ
(1 + ρ
2
)
2
= 4


0
ρ dρ
(1 + ρ
2
)
2
= 2


1
du
u
2
= 2,
where we substitute x + iy = ρ exp(iϕ) and u = 1 + ρ
2
, respectively. As
expected, the right hand side is the Euler number of CP
1
= S
2
.
4) In Example 3.24, we consider the canonical metric of the tautological
bundle U over the complex Grassmannian G
k,n
and compute the principal
part of the curvature of the Chern connection D. In the case U →G
1,2
= CP
1
and with respect to the coordinates z as in the previous example, we obtain
Ω(0) = −dz ∧ dz.
It follows that

CP
1
c
1
(U, D) = −1 and that c
1
(CP
1
) = −2c
1
(U).
5) Let π: S
2m+1
→ CP
m
be the natural projection, x ∈ S
2m+1
, and p =
π(x). Then the differential of π at x identifies the orthogonal complement p

with the tangent space of CP
m
at p. However, this isomorphism depends on
the choice of p.The ambiguity is resolved by passing from p

to Hom(p, p

). In
other words, TCP
m ∼
= Hom(U, U

), where U

is the orthogonal complement of
the tautological bundle U →CP
m
in the trivial bundle C
m+1
CP
m
→CP
m
with respect to the standard Hermitian metric.
By the naturality of Chern classes, the first Chern class c
1
(U) has value −1
on the complex line CP
1
⊂ CP
m
. Now Hom(U, U)

= CCP
m
, the trivial line
bundle over CP
m
. Hence
TCP
m
⊕C = Hom(U, U

⊕U) = Hom(U, C
m+1
)

= U

⊕· · · ⊕U

,
where we have m+ 1 summands on the right hand side. Therefore
c(CP
m
) = c(TCP
m
⊕C) = (1 −c
1
(U))
m+1
= (1 +a)
m+1
,
where a is the generator of H
2
(CP
m
, C) which has value 1 on the complex line
CP
1
⊂ CP
m
. In particular,
c
1
(CP
m
) = (m+ 1)a. (A.41)
It follows that c
m
(CP
m
)[CP
m
] = m+ 1, the Euler number of CP
m
.
A.42 Remark. If E is a complex vector bundle of rank r over a manifold M,
then c
r
(E) is the Euler class of E, considered as an oriented real vector bundle
of rank 2r, see Proposition A.59 below.
134 Lectures on K¨ ahler Manifolds
On gl(r, C) = C
r×r
, consider the invariant formal power series
Φ(A) := tr exp A = r + tr A +
1
2
tr A
2
+ · · · . (A.43)
Let E → M be a complex vector bundle of rank r and P be the bundle of
frames of E. Let D be a connection on E and R be the curvature of D. Then
the cocycle
ch(E, D) := Φ

P,
i

R
«
∈ A
even
(M, C) (A.44)
is called the Chern character of E with respect to D. Its cohomology class
Φ(E) in H
even
(M, C) is called the Chern character of E. If A is diagonal,
A = D(a
1
, . . . , a
r
), then
tr A = a
1
+ · · · +a
r
= σ
1
(a
1
, . . . , a
r
),
tr A
2
= a
2
1
+ · · · a
2
r
= σ
1
(a
1
, . . . , a
r
)
2
−2σ
1
(a
1
, . . . , a
r
),
where σ
1
, σ
2
, . . . denotes the sequence of elementary symmetric functions. Since
diagonal matrices are dense in C
r×r
, we conclude that
tr A = Φ
1
(A), and
1
2
tr A
2
=
1
2
Φ
1
(A)
2
−Φ
2
(A), (A.45)
where Φ
1
and Φ
2
are as in (A.32). By the fundamental theorem on sym-
metric polynomials, the symmetric function tr D(a
1
, . . . , a
r
)
k
can be expressed
uniquely as a polynomial in σ
1
, . . . , σ
r
. It follows that ch(E, D) can be ex-
pressed as a polynomial in the Chern forms c
k
(E, D). From what we noted
above, we have
ch(E, D) = r +c
1
(E, D) +

1
2
c
1
(E, D)
2
−c
2
(E, D)
«
+ · · · . (A.46)
We leave it as an exercise to compute the term of degree 6 in this formula.
A.47 Exercise. Follow the argument in Proposition A.38.2 and show that
ch(E

⊕E

, D

⊕D

) = ch(E

, D

) + ch(E

, D

).
Recall that the derived action of the Lie algebra gl(r, C) gl(s, C) on C
r
⊗C
s
is given by (A, B)(x, y) = (Ax) ⊗ y +x ⊗(By) and show that
ch(E

⊗E

, D

⊗D

) = ch(E

, D

) ∧ ch(E

, D

).
It follows that the Chern character induces a homomorphism from the Grothen-
dieck ring K(M) generated by equivalence classes of complex vector bundles
over M to the even cohomology ring H
even
(M, C).
Appendix A Chern–Weil Theory 135
A.2 Euler Class. On gl(2r, R) = R
2r×2r
, consider the homogeneous polyno-
mial
Pf(A) :=
1
2
r
r!
¸
σ
ε(σ)a
σ(1)σ(2)
. . . a
σ(2r−1)σ(2r)
(A.48)
of degree r, the Pfaffian of A. It is easy to show that
Pf(BAB
t
) = Pf(A) det B. (A.49)
Therefore the Pfaffian defines an invariant polynomial on so(2r), the Lie algebra
of the group of rotations SO(2r). The normal form on so(2r) is
A =

¸
¸
A
1
.
.
.
A
r
¸

with A
j
=

0 −a
j
a
j
0

. (A.50)
For such a matrix A, we have Pf(A) = (−1)
r
a
1
. . . a
r
. By invariance, we
conclude that on so(2r),
Pf(A)
2
= det A. (A.51)
Let E → M be an oriented real vector bundle of rank 2r with a Riemannian
metric g. Then the principal bundle P of oriented orthonormal frames of E has
structure group SO(2r). If D is a connection on P, that is, a metric connection
on E, and R denotes the curvature of D, then the associated form
χ(E, g, D) := Pf

P,
1

R
«
(A.52)
is called the Euler form of E with respect to g and D. It is clear that
χ(f

E, f

g, f

D) = f

χ(E, g, D) (A.53)
and that
χ(E ⊕E

, g ⊕g

, D ⊕D

) = χ(E, g, D) ∧ χ(E

, g

, D

). (A.54)
Let g

be another Riemannian metric on E and write g

= g(B· , · ) where B is
a field of symmetric and positive definite endomorphisms of E. Let C =

B,
then g

(x, y) = g(Cx, Cy), and hence C is an isomorphism of E with C

g = g

.
It follows that χ(E, g

, C

D) = χ(E, g, D). We conclude that the cohomology
class χ(E) of χ(E, g, D) in H
2r
(M, R) does not depend on g and D. It is called
the Euler class of E.
A.55 Remark. If E has a Riemannian metric g with a flat metric connection
D, then χ(E, g, D) = 0 and hence also χ(E) = 0. However, there are vector
bundles E with flat non-metric connections such that χ(E) = 0, see for example
[MS, 312 pp]. In fact, in the definition of the Euler form we need a Riemannian
metric and a corresponding metric connection on E.
136 Lectures on K¨ ahler Manifolds
Our next aim is to compare Euler class and Chern classes. We identify C
r
with R
2r
via the correspondence
(z
1
, . . . , z
r
) ←→(x
1
, y
1
, . . . , x
r
, y
r
),
where z
j
= x
j
+ iy
j
. This induces an inclusion U(r) → SO(2r) of the unitary
group and a corresponding inclusion u(r) → so(2r), A → A
R
, of Lie algebras.
Under this inclusion, diagonal matrices
A =

¸
¸
ia
1
.
.
.
ia
r
¸

¸
¸
A
1
.
.
.
A
r
¸

= A
R
, (A.56)
where A
j
is as above. For such a diagonal A, det(iA) = (−1)
r
a
1
. . . a
r
=
Pf(A
R
). By invariance,
det(iA) = Pf(A
R
) (A.57)
for all A ∈ u(r).
Suppose now that E →M is a complex vector bundle of rank r. Endow E
with a Hermitian metric h and a corresponding Hermitian connection D. If we
consider E as an oriented real vector bundle of rank 2r, the real part g = Re h
of the Hermitian metric defines a Riemannian metric on E and D is metric
with respect to g.
Let ω and Ω be the connection and curvature forms of D with respect to
a local unitary frame of E. If we consider E as an oriented real vector bundle
and choose D as a metric connection for the Riemannian metric g as above,
then ω
R
and Ω
R
are the connection and curvature forms of D with respect to
the induced oriented and orthonormal frame of E. By (A.33) and (A.57),
c
r
(E, D) = det

i


«
= Pf

1


R
«
= χ(E, g, D). (A.58)
A.59 Proposition. For a complex vector bundle E →M of rank r, χ(E) =
c
r
(E).
Proposition A.59 and Equations A.53, A.54 imply that the Euler class of
(the tangent bundle of) an oriented closed manifold M, applied to the fun-
damental class of M, gives the Euler number of M. This is the celebrated
Chern–Gauß–Bonnet formula.
Appendix B Symmetric Spaces
In this appendix we collect a few of the basic facts about symmetric spaces.
Good references for unproved statements and further reading are [Bor], [Hel],
[Wo].
We say that a Riemannian manifold M is a symmetric space if M is con-
nected and if, for all p ∈ M, there is an isometry s
p
of M with
s
p
(p) = p and ds
p
(p) = −id . (B.1)
We then call s
p
the geodesic reflection at p.
B.2 Exercise. Let M be a connected Riemannian manifold and f
1
, f
2
be
isometries of M. Suppose there is a point p ∈ M with f
1
(p) = f
2
(p) and
df
1
(p) = df
2
(p). Then f
1
= f
2
.
We say that a Riemannian manifold M is locally symmetric if each point
p ∈ M has a neighborhood U in M with an isometry s
p
: U → U such that
s
p
(p) = p and ds
p
(p) = −id.
B.3 Exercises. 1) Let M be a symmetric space and
˜
M → M be a covering.
Show that
˜
M with the induced Riemannian metric is also a symmetric space.
2) Let M = M
1
M
2
be a Riemannian product. Show that M is symmetric
if and only if M
1
and M
2
are symmetric.
3) Prove the corresponding assertions in the case of locally symmetric
spaces.
B.4 Proposition. Let M be a locally symmetric space and F be a tensor field
on M of type (k, l). If F is invariant under the local isometries s
p
, p ∈ M,
then F = 0 if k +l is odd.
For example, if M is locally symmetric, then the curvature tensor is parallel,
∇R = 0. In fact, it is not hard to see that M is locally symmetric if and only if
the curvature tensor of M is parallel. Another application of Proposition B.4:
If M is locally symmetric and J is an almost complex structure on M which is
invariant under the local isometries s
p
, p ∈ M, then J is parallel.
Proof of Proposition B.4. Let p ∈ M, v
1
, . . . , v
k
∈ T
p
M, and ϕ
1
, . . . , ϕ
l
∈ T

p
M.
Since s
p
(p) = p, ds
p
(p) = −id, and (s

p
F)
p
= F
p
, we conclude
F
p
(v
1
, . . . , v
k
, ϕ
1
, . . . , ϕ
l
) = F
p
(s
p∗
v
1
, . . . , s
p∗
v
k
, s

p
ϕ
1
, . . . , s

p
ϕ
l
)
= F
p
(−v
1
, . . . , −v
k
, −ϕ
1
, . . . , −ϕ
l
)
= (−1)
k+l
F
p
(v
1
, . . . , v
k
, ϕ
1
, . . . , ϕ
l
).
The fundamental theorem about locally symmetric spaces is as follows:
138 Lectures on K¨ ahler Manifolds
B.5 Theorem (Cartan). Let M and N be locally symmetric Riemannian
spaces, where M is simply connected and N is complete. Let p ∈ M and q ∈ N,
and let A: T
p
M →T
q
N be a linear isomorphism preserving inner products and
curvature tensors,A

g
N
q
= g
M
p
and A

R
N
q
= R
M
p
. Then there is a unique local
isometry F : M →N with F(p) = q and F

(p) = A.
As for a proof, see for example the more general Theorems 1.1.36 in [CEb]
or III.5.1 in [Sa]
23
.
B.6 Corollary. Let M be a complete and simply connected locally symmetric
space. Then M is a symmetric space. More precisely, the differentials of the
isometries of M fixing a point p in M are given precisely by the orthogonal
transformations of T
p
M preserving the curvature tensor R
M
p
.
This is one of the most important immediate applications of Theorem B.5.
Another immediate application asserts that a simply connected symmetric
space is determined, up to isometry, by the inner product and the curvature
tensor at a point (using Propositions B.7 and B.8 below). This implies, for ex-
ample, the uniqueness of the model Riemannian manifolds of constant sectional
curvature.
B.7 Proposition. If M is symmetric, then M is homogeneous.
Proof. Let c: [a, b] → M be a geodesic in M, and let m be the midpoint
of c. Then the geodesic reflection s
m
of M at m reflects c about m, hence
s
m
(c(a)) = c(b).
Since M is connected, any two points p, q ∈ M can be connected by a
piecewise geodesic c. If c: [a, b] →M is such a curve, then there is a subdivision
a = t
0
< t
1
< · · · < t
k
= b of [a, b] such that c|[t
i−1
, t
i
] is a geodesic, 1 ≤ i ≤ k.
By the first part of the proof, there is an isometry f
i
of M mapping c(t
i−1
) to
c(t
i
). Hence f
k
◦ · · · ◦ f
1
is an isometry of M mapping p = c(a) to q = c(b).

B.8 Proposition. If M is homogeneous, then M is complete.
Proof. Let p ∈ M. Then there is an ε > 0 such that the closed ball B(p, ε) of
radius ε about p is compact. Since M is homogeneous, the closed ball B(q, ε)
of radius ε about any point q ∈ M is compact. It follows that M is complete.

Let M be a symmetric space. Let c: [a, b] → M be a piecewise smooth
curve. An isometry f : M → M is called transvection along c if f(c(a)) = c(b)
and if df(c(a)) is equal to parallel translation along c from c(a) to c(b).
Let p be a point in M. Let c: R → M be a geodesic through p. Let s
t
be
the geodesic reflection about c(t) and f
t
= s
t/2
◦ s
0
. Then f
t
is an isometry of
M with f
t
(c(τ)) = c(τ +t) for all τ ∈ R.
23
Both references contain also introductions to symmetric spaces.
Appendix B Symmetric Spaces 139
Let X be a parallel vector field along c. Then ds
0
◦X is parallel along s
0
◦c.
Since s
0
(c(s)) = c(−s) and ds
0
(p) = −id, we have
ds
0
X(s) = −X(−s).
Similarly,
ds
t/2
X(s) = −X(t −s).
We conclude that, for all τ ∈ R,
df
t
X(τ) = X(τ +t).
Hence df
t
(c(τ)) is parallel translation along c from c(τ) to c(τ + t). Hence f
t
is a transvection along c; more precisely, it is the transvection along c|[τ, τ +t]
for any τ ∈ R. It follows that the family (f
t
) is a smooth one-parameter group
of isometries of M which shifts c and whose derivatives correspond to parallel
translation along c. We call it the one-parameter group of transvections along
c. Associated to this family we have the Killing field
X(p) = ∂
t
(f
t
(p))|
t=0
.
We call X the infinitesimal transvection along c.
B.9 Lemma. Let c: R →M be a geodesic with c(0) = p. Then a Killing field
X on M is the infinitesimal transvection along c if and only if X(p) = c

(0)
and ∇X(p) = 0.
Proof. Let X be the infinitesimal transvection along c. Then X(p) = c

(0)
since f
t
(p) = c(t). As for the covariant derivative of X in p, let v ∈ T
p
M and
σ = σ(τ) be a smooth curve with σ(0) = p and σ

(0) = v. Then

v
X = ∇
τ

t
(f
t
(σ(τ)))|
t=τ=0
= ∇
t

τ
(f
t
(σ(τ)))|
τ=t=0
= ∇
t
(df
t
v)|
t=0
= 0
since df
t
is parallel translation along c. Since M is connected, a Killing field
on M is determined by its value and covariant derivative at one point.
B.10 Proposition. Let M be a symmetric space, p be a point in M. Let
c: [a, b] →M be a geodesic loop at p. Then c is a closed geodesic, c

(a) = c

(b).
Proof. Let v ∈ T
p
M and X be the infinitesimal transvection along the geodesic
through p with initial velocity v. Then X(p) = v and ∇X(p) = 0, by Lemma B.9.
Since X is a Killing field, X◦c is a Jacobi field along c, see Exercise 1.5. Recall
that for any two Jacobi fields V, W along c, the function 'V

, W` − 'V, W

` is
constant, where the prime indicates covariant derivative along c. We apply this
to the Jacobi fields X ◦ c and tc

,
'(X ◦ c)

, tc

` −'X ◦ c, c

` = const.
140 Lectures on K¨ ahler Manifolds
Since c(a) = c(b) = p and ∇X(p) = 0, we get
'X(p), c

(a)` = 'X(p), c

(b)`.
Since X(p) = v ∈ T
p
M was arbitrary, we conclude that c

(a) = c

(b).
B.11 Corollary. The fundamental group of a symmetric space is Abelian.
Proof. Let M be a symmetric space and p be a point in M. Shortest loops in
homotopy classes of loops at p are geodesic and hence closed geodesics through
p, by Proposition B.10. The geodesic reflection at p maps closed geodesics
through p into their inverses. Hence s
p
induces the inversion on π
1
(M, p). Now
a group is Abelian if its inversion is a homomorphism.
B.12 Definition. Let M be a symmetric space. We say that M is of compact
type or, respectively, non-compact type if the Ricci curvature of M is positive
or, respectively, negative. We say that M is of Euclidean type if M is flat, that
is, if the sectional curvature of M vanishes.
If M is a symmetric space of compact type, then M is compact and the
fundamental group of M is finite, by the theorem of Bonnet–Myers. Flat tori
and their Riemannian covering spaces are symmetric spaces of Euclidean type.
B.13 Proposition. If M is a symmetric space of non-compact type, then M
does not have non-trivial closed geodesics.
Proof. Let c: R → M be a closed geodesic of unit speed. Set p = c(0) and
u = c

(0). Since M is of non-compact type, Ric(u, u) < 0. Hence the symmetric
endomorphismR(· , u)u of T
p
M has a negative eigenvalue κ and a corresponding
unit eigenvector v perpendicular to u. Let V be the parallel vector field along
c with V (0) = v. Since R is parallel, R(V, c

)c

= κV along c. Therefore
J = cosh(

−κt)V is the Jacobi field along c with J(0) = v and J

(0) = 0.
Let X be the infinitesimal transvection through p with X(p) = v, that is,
along the geodesic determined by v. Then Y = X ◦ c is a Jacobi field along c
with the same initial conditions as J: Y (0) = v and Y

(0) = 0, hence Y = J.
Now as the restriction of a vector field on M to c, Y is periodic along c. On
the other hand, J is definitely not periodic. Contradiction.
B.14 Corollary. Let M be a symmetric space of non-compact type. Then:
1) M is simply connected.
2) If G denotes the component of the identity of the group of isometries of
M and K ⊂ G the stabilizer of a point p ∈ M, then K is connected.
Proof. Since shortest loops in homotopy classes of loops at p are geodesic loops,
the first assertion is immediate from Propositions B.10 and B.13. As for the
second, we note that the last piece of the long exact homotopy sequence of the
fibration G →M, g →gp, reads
1 = π
1
(M) →π
0
(K) →π
0
(G) = 1.
Appendix B Symmetric Spaces 141
B.15 Remark. In Remark B.29 we show that a symmetric space has non-
negative sectional curvature iff its Ricci curvature is non-negative and, simi-
larly, that it has non-positive sectional curvature iff its Ricci curvature is non-
positive. In particular, a symmetric space is of Euclidean type if and only if
its Ricci curvature vanishes. Furthermore, if M is a symmetric space of non-
compact type, then M is diffeomorphic to R
n
, n = dimM, by the theorem
of Hadamard–Cartan and Corollary B.14.1. It follows that, for K and G as
in Corollary B.14.2, K is a deformation retract of G. This improves Corol-
lary B.14.2.
B.1 Symmetric Pairs. Let (G, K) be a pair consisting of a Lie group G and
a closed subgroup K. We say that (G, K) is a symmetric pair if M = G/K is
connected and if there is an involutive automorphism σ: G →G with
F
0
⊂ K ⊂ F, (B.16)
where F = {g ∈ G | σ(g) = g} and F
0
denotes the component of the identity of
F. In what follows, (G, K) is a symmetric pair with involutive automorphism
σ.
Let g be the Lie algebra of left-invariant vector fields of G, via evaluation
identified with the tangent space T
e
G of G at the neutral element e ∈ G.
Denote by σ

the differential of σ at e, and let
k = {X ∈ g | σ

X = X}, p = {X ∈ g | σ

X = −X}, (B.17)
the eigenspaces of σ

for the eigenvalues 1 and −1, respectively. Since σ

is
involutive,
g = k + p. (B.18)
Since F
0
⊂ K ⊂ F, k is the Lie algebra of K. Since K ⊂ F, σ(kgk
−1
) =
kσ(g)k
−1
for all k ∈ K and all g ∈ G, and hence σ

commutes with all Ad
k
,
k ∈ K. Therefore
Ad
k
(k) ⊂ k and Ad
k
(p) ⊂ p for all k ∈ K. (B.19)
Furthermore,
[k, k] ⊂ k, [k, p] ⊂ p, [p, p] ⊂ k. (B.20)
In particular, k is perpendicular to p with respect to the Killing form of g. As
for the proof of (B.20), the first and second inclusion follow from (B.19) above.
For the proof of the third, let X, Y ∈ p. Since σ

is an automorphism of g, we
have
σ

[X, Y ] = [σ

X, σ

Y ] = [−X, −Y ] = [X, Y ]
and hence [X, Y ] ∈ k. Note that the first two inclusions also follow by a similar
argument.
142 Lectures on K¨ ahler Manifolds
Since we divide by K on the right, G acts from the left on M. We denote
by gp ∈ M the image of p ∈ M under left multiplication by g ∈ G. Sometimes
it will be convenient to distinguish between the element g ∈ G and the diffeo-
morphism of M given by left multiplication with g. Then we will use λ
g
to
denote the latter, λ
g
(p) := gp.
B.21 Exercise. Let M be a manifold and G be a Lie group which acts on M.
For X ∈ g, define a smooth vector field X

on M by
X

(p) = ∂
t
(e
tX
(p))|
t=0
.
By definition, exp(tX) is the flow of X

. Show that [X, Y ]

= −[X

, Y

].
We denote by o = [K] the distinguished point of M and by π: G →M the
canonical map, π(g) = [gK] = λ
g
(o). We use π

= dπ(e): p →T
o
M to identify
p and T
o
M. An easy computation shows that for all k ∈ K
π

◦ Ad
k
= dλ
k
(o) ◦ π

. (B.22)
Thus with respect to the identification p

= T
o
M via π

, the isotropy represen-
tation of K on T
o
M corresponds to the restriction of the adjoint representation
of K to p.
We say that a symmetric pair is Riemannian if there is an inner product
'·, ·` on p which is invariant under Ad
K
, that is, under all Ad
k
, k ∈ K. Note
that such inner products are in one-to-one correspondence with G-invariant
Riemannian metrics on M.
B.23 Example. Let M be a symmetric space and o ∈ M be a preferred
origin. Let G be the group Iso(M) of all isometries of M or the component
of the identity in Iso(M). Let K be the stabilizer of o in G. Then (G, K) is
a Riemannian symmetric pair with respect to the involution σ of G given by
conjugation with the geodesic reflection s
o
of M in o. The pull back of the
inner product in T
o
M to p is invariant under Ad
K
and turns (G, K) into a
Riemannian symmetric pair with G/K = M.
B.24 Theorem. Let (G, K) be a Riemannian symmetric pair with correspond-
ing involution σ of G and inner product '· , · ` on p. Let M = G/K and endow
M with the G-invariant Riemannian metric corresponding to '· , · `. Then we
have:
1. M is a symmetric space. The geodesic symmetry s at o is s([gK]) =
[σ(g) · K]. In particular, s ◦ λ
g
= λ
σ(g)
◦ s.
2. For X ∈ p, the curve e
tX
(o), t ∈ R, is the geodesic through o with initial
velocity π

X, and left multiplication by e
tX
, t ∈ R, is the one-parameter
group of transvections along this geodesic.
Appendix B Symmetric Spaces 143
3. With respect to the identification p

= T
o
M, Ad
K
-invariant tensors on p
correspond to G-invariant tensor fields on M and these are parallel.
4. With respect to the identification p

= T
o
M, the curvature tensor R and
Ricci tensor Ric of M at o are given by
R(X, Y )Z = −[[X, Y ], Z] = −[[X

, Y

], Z

](p),
Ric(X, Y ) = −
1
2
B(X, Y ),
where B denotes the Killing form of g and X

, Y

, Z

are associated to
X, Y, Z as in Exercise B.21.
5. With respect to the identification p

= T
o
M, a subspace q ⊂ p is tangent to
a totally geodesic submanifold of M through o if and only if [[q, q], q] ⊂ q.
If the latter inclusion holds, then N = exp(q)(o) is such a submanifold
and is a symmetric space in the induced Riemannian metric.
B.25 Remarks. 1) We can rewrite the first formula in B.24.4 as follows:
R(X, Y ) = −ad
[X,Y ]
or R(· , Y )Z = −ad
Z
◦ ad
Y
,
where X, Y, Z ∈ p and both maps are considered on p.
2) The totally geodesic submanifold in B.24.5 need not be closed. A good
example for this is a non-rational line through 0 in a flat torus.
Proof of Theorem B.24. It is easy to see that s([gK]) := [σ(g) · K] defines an
involutive smooth map s of M with s ◦ λ
g
= λ
σ(g)
◦ s. Since s is involutive, s
is a diffeomorphism of M. Moreover s(o) = o and ds(o) = −id.
Let p = g(o) ∈ M and u ∈ T
p
M. Choose v ∈ T
o
M with dλ
g
(v) = u. By
the G-invariance of the metric, we have |u| = |dλ
g
(v)| = |v| and hence
|ds(u)| = |ds(dλ
g
(v))| = |dλ
σ(g)
(ds(v))| = | −dλ
σ(g)
(v)| = |v| = |u|.
It follows that s is an isometry. This completes the proof of (1).
Let X ∈ p and X

be the corresponding Killing field of M as in Exer-
cise B.21. Since X ∈ p, we have
e
tX
(s(p)) = s(e
−tX
(p))
for all p ∈ M, hence s

X

= −X

. Let u ∈ T
o
M. Since s is an isometry, we
get
−∇
u
X

= ds(∇
u
X

) = ∇
ds(u)
(s

X

) = ∇
u
X

,
and hence ∇X

(o) = 0. Hence X

is the infinitesimal transvection with
X

(o) = π

X. This proves (2).
It is clear that Ad
K
-invariant tensors on p correspond to G-invariant ten-
sor fields on M. Now the one-parameter groups (exp(tX)), X ∈ p, are the
144 Lectures on K¨ ahler Manifolds
transvections along the corresponding geodesics through o. It follows that G-
invariant tensor fields are parallel at o and hence parallel everywhere in M.
Hence (3).
Let X, Y, Z ∈ p and X

, Y

, Z

be the corresponding Killing fields of M as
in Exercise B.21. Then X

Y

, and Z

are parallel at o. Let u = Z

(o) = π

Z
and compute ∇
u
[X

, Y

] using the differential equation for Killing fields. This
gives the first equation in (4). In the proof of the second we can assume that
X = Y , by the symmetry of the Ricci tensor. Since X ∈ p, we have ad X(k) ⊂ p
and adX(p) ⊂ k. Therefore tr ad
2
X|k = tr ad
2
X|p, and the claim about the
Ricci tensor follows.
It remains to prove (5). Note that for a totally geodesic submanifold N
through the origin o, R(u, v)w ∈ T
o
N for all u, v, w ∈ T
o
N. Hence the necessity
of the condition on q is immediate from the formula for R in (3).
Suppose now that [[q, q], q] ⊂ q. Let l = [q, q]. Then l is a Lie subalgebra of k
and h = l+q is a Lie subalgebra of g. Let H be the corresponding connected Lie
subgroup of G. Then the orbit N of o under H, N = H(o), is a submanifold of
M. Now N is totally geodesic at o since N = exp(q)(o) locally about o. Since
H is transitive on N and H acts isometrically on M, N is totally geodesic
everywhere. Since M is a complete Riemannian manifold and N is connected
we have N = exp(q)(o). In particular, N is complete. Since N is invariant
under s, s|N is the geodesic symmetry of N in o. Since N is homogeneous, N
is symmetric.
B.26 Remark (Coverings). The groups K with F
0
⊂ K ⊂ F correspond to
Riemannian coverings
G/F
0
→G/K →G/F.
For example, if M = S
n
⊂ R
n+1
, the unit sphere of dimension n, then the
component G of the identity in Iso(M) is equal to SO(n+1). If we let s be the
reflection in a chosen unit vector v ∈ S
n
and σ be conjugation with s, then the
stabilizer K of v is isomorphic to SO(n) and K = F
0
= F. If we pass to the
quotient RP
n
keeping G, then we have K = F. A similar phenomenon occurs
for some of the other symmetric spaces of compact type.
It is remarkable that the fixed point set of an involution of a compact
and simply connected Lie group is connected, see [Hel, Theorem VII.7.2].
Thus the symmetric space associated to a Riemannian symmetric pair (G, K)
with G compact and simply connected is simply connected. For example,
the unit sphere S
n
is associated to the Riemannian symmetric pair (Spin(n +
1), Spin(n)). Real projective space RP
n
is a homogeneous space of Spin(n+1),
but is not associated to a Riemannian symmetric pair (G, K) with G simply
connected.
B.27 Remark (Effective Pairs). Let (G, K) be a Riemannain symmetric pair
and N ⊂ G be the normal subgroup of elements acting trivially on M = G/K,
N = {g ∈ G | g acts as identity on M}.
Appendix B Symmetric Spaces 145
Then N ⊂ K, hence the involution σ of G factors over G/N. Since M is
connected, an isometry of M = G/K fixing the preferred origin o of M is
determined by its differential at o, see Exercise B.2. Hence
N = {k ∈ K | Ad
k
is the identity on p}.
In particular, the adjoint action of K on p factors over K/N. It follows that
(G/N, K/N) is a Riemannian symmetric pair with (G/N)/(K/N) = G/K = M
and Lie algebra g/n

= k/n + p.
We say that the Riemannian symmetric pair (G, K) is effective or, respec-
tively, infinitesimally effective if N is trivial or, respectively, a discrete subgroup
of G. The pair (G, K) is effective iff G is a subgroup of the isometry group
of M. The above pair (G/N, K/N) is the effective pair associated to (G, K).
Thus it becomes plausible that, for many purposes, it is sufficient to consider
effective or infinitesimally effective pairs.
B.28 Remark (Killing Form). Let (G, K) be an infinitesimally effective Rie-
mannian symmetric pair with K/N compact, where N is as in Remark B.27.
Let X ∈ k. From (B.20) we know that ad
X
(k) ⊂ k and ad
X
(p) ⊂ p. Hence
B(X, X) = tr(ad
X
|k)
2
+ tr(ad
X
|p)
2
= B
K
(X, X) + tr(ad
X
|p)
2
,
where B is the Killing form of G and B
K
the Killing form of K. Since N ⊂ K
is a discrete subgroup and K/N is compact, B
K
≤ 0 and hence B(X, X) ≤
tr(ad
X
|p)
2
. On the other hand, since N is discrete and X ∈ k, ad
X
|p = 0 for
X = 0, compare Exercise B.2. Furthermore, ad
X
is skew-symmetric on p with
respect to the given inner product on p since the adjoint representation of K on
p preserves this product. Now the square of a skew-symmetric endomorphism
has negative trace unless the endomorphism vanishes. Therefore B is negative
definite on k. Hence G is semi-simple if and only if the Ricci curvature of
M = G/K is non-degenerate.
B.29 Remark (Curvature). Let (G, K) be an infinitesimally effective Rieman-
nian symmetric pair with K/N compact, where N is as in Remark B.27. Since
the Ricci tensor of M = G/K is parallel, there are real numbers λ
i
and pairwise
perpendicular parallel distributions E
i
on M such that Ric X = λ
i
X for all X
tangent to E
i
. Identify p = T
o
M and let X, Y ∈ p be tangent to E
i
and E
j
,
respectively. Then, by Theorem B.24.4,
B([X, Y ], [X, Y ]) = −B([X, [X, Y ]], Y ) = 2 Ric([X, [X, Y ]], Y )
= 2λ
j
'[X, [X, Y ]], Y ` = −2λ
j
'R(Y, X)X, Y `,
where we use that Y is tangent to E
j
. By the same computation,
B([Y, X], [Y, X]) = −2λ
i
'R(X, Y )Y, X`.
146 Lectures on K¨ ahler Manifolds
Hence B([X, Y ], [X, Y ]) = 0 if λ
i
= λ
j
or if λ
i
= 0 or λ
j
= 0. Since B is
negative definite on k, this implies that the curvature tensor R vanishes on the
kernel of Ric. For λ
i
= 0 and X, Y ∈ p both tangent to E
i
, we get
'R(X, Y )Y, X` = −
1

i
B([X, Y ], [X, Y ]).
In particular, the sectional curvature of a symmetric space is non-negative
respectively non-positive iff its Ricci curvature is non-negative respectively non-
positive.
By definition, M is of compact or, respectively, non-compact type iff all
λ
i
> 0 or, respectively, all λ
i
< 0. In particular, a symmetric space of non-
compact type is a complete and simply connected Riemannian manifold of
non-positive sectional curvature, compare Corollary B.14 and Remark B.15.
If M is simply connected, the distributions E
i
give rise to a splitting of M
as a Riemannian product with factors M
i
tangent to E
i
. For each i, M
i
is
a symmetric Einstein space with Einstein constant λ
i
. Applying this to the
universal covering space of M, we see that we can renormalize the metric of M
without destroying the geometry of M so that the constants λ
i
are in {±1, 0}.
For example, if M is of compact type, we can renormalize the metric so that
M becomes an Einstein space with Einstein constant 1.
B.30 Remark (Curvature Operator). Let (G, K) be a Riemannian symmetric
pair. Suppose that the inner product on p is the restriction of an Ad
G
-invariant
bilinear form
24
, also denoted '· , · `, on g. Then all ad
X
, X ∈ g, are skew-
symmetric with respect to '· , · `, and then, with respect to the identification
p = T
o
M,
'R(X, Y )V, U` = −'[[X, Y ], V ], U` = '[X, Y ], [U, V ]`. (B.31)
Recall the definition of the curvature operator
ˆ
R in (1.43) and define a mor-
phism
F : Λ
2
p →k, F(X ∧ Y ) := [X, Y ]. (B.32)
By (B.31), we have
'
ˆ
R(X ∧ Y ), U ∧ V ` = 'F(X ∧ Y ), F(U ∧ V )`. (B.33)
In particular, if '· , · ` is an inner product on g, then
'R(X, Y )Y, X` = |[X, Y ]|
2
and
ˆ
R = F

F ≥ 0. (B.34)
Compare Remarks 1.42 and 5.56 for important consequences of
ˆ
R ≥ 0.
24
For example, a negative or positive multiple of the Killing form of G if (G, K) is infini-
tesimally effective and M = G/K is of compact or non-compact type, respectively.
Appendix B Symmetric Spaces 147
B.35 Remark (De Rham Decomposition). Let M be a simply connected sym-
metric space. If N is a factor in the de Rham decomposition of M, then N is
tangent to one of the distributions E
i
as in Remark B.29 (but TN might be
smaller than E
i
). From Remark B.29 we conclude that N is an Einstein space.
Let λ be the Einstein constant of N. We have:
1) If λ < 0, then N is a symmetric space of non-compact type.
2) If λ = 0, then N is a Euclidean space.
3) If λ > 0, then M is of compact type.
In particular, a compact symmetric space is of compact type if and only if
its fundamental group is finite.
B.36 Remark (Holonomy). Let (G, K) be a Riemannian symmetric pair and
set M = G/K. Then π: G →M is a principal bundle with structure group K.
The tangent bundle TM is (canonically isomorphic to) the bundle G
K
p asso-
ciated to the adjoint representation of K on p. The left-invariant distribution
on G determined by p is a principal connection which induces the Levi-Civita
connection on M. In particular, the holonomy group Hol(M) of M at o = [K]
is contained in the image of K under the adjoint representation on p.
Assume now that (G, K) is infinitesimally effective with K/N compact,
where N is as in Remark B.27. Suppose furthermore that the Ricci curvature
of M is non-degenerate. By (B.20), h = [p, p] + p is an ideal in g. By Remark
B.29, B|k is negative definite. By assumption, B|p is non-degenerate. Moreover,
k and p are perpendicular with respect to B, by (B.20). In particular, G is
semi-simple and B|h is non-degenerate with p ⊂ h. Hence the B-perpendicular
complement h

of h is contained in k and h ∩ h

= 0. Since h is an ideal in g,
h

is an ideal as well. Hence [h, h

] = 0. We conclude that for X ∈ h

,
e
tX
e
Y
e
−tX
= e
Y
for all Y ∈ p and t ∈ R. Since exp(tX) ∈ K, this implies that exp(tX) acts as
the identity on M. Hence X = 0 since (G, K) is infinitesimally effective. We
conclude that g = h, that is k = [p, p].
For X, Y ∈ p, [X, Y ] ∈ k and the curvature tensor is R(X, Y ) = −ad
[X,Y ]
on p. Moreover, the endomorphisms R(X, Y ) are in the Lie algebra of the
holonomy group Hol(M). Hence, under the above assumptions on (G, K), the
adjoint image of the component K
0
of the identity of K is equal to the reduced
holonomy group of M. In particular, if (G, K) is effective and M is of compact
or non-compact type, then K
0
is equal to the reduced holonomy group of M.
If, moreover, M is simply connected and G is connected, then K is equal to
the holonomy group of M at the preferred origin and hence G is equal to the
component of the identity of the isometry group of M. Then parallel tensors
on M are also G-invariant, compare Theorem B.24.3 for the converse assertion.
B.37 Exercise (Dual Pairs). Let (G, K) be a Riemannian symmetric pair and
148 Lectures on K¨ ahler Manifolds
g = k +p be the associated decomposition of the Lie algebra g of G. Show that
[X, Y ]

:=

[X, Y ] if X ∈ k,
−[X, Y ] if X, Y ∈ p
defines a (new) Lie bracket on g. Note that this change corresponds to passing
from g to g

= k +ip in the complexification g
C
of g. We say that a Riemannian
symmetric pair (G

, K

) is dual to (G, K) if the corresponding decomposition
of the Lie algebra g

of G

is up to isomorphism given by g

= k + p with Lie
bracket [ · , · ]

. We also say that the corresponding symmetric spaces are dual.
Show:
1) Under duality, the curvature tensor changes sign and compact type cor-
responds to non-compact type.
2) For each symmetric space M, there is a simply connected symmetric
space M

dual to M and M

is unique up to isometry.
3) If M and N are simply connected dual symmetric spaces and K re-
spectively L denotes the group of isometries fixing a point p ∈ M respec-
tively q ∈ N, then there is an orthogonal transformation A: T
p
M →T
q
N with
A

R
N
q
= −R
M
p
and, for any such transformation, L = AKA
−1
.
B.2 Examples. This subsection is devoted to examples which are relevant
in our discussion. To keep the presentation transparent, we divide them into
three classes, Examples B.38, B.42, and B.53.
B.38 Example. Let G = Sl(n, R) with involution σ(A) := (A
t
)
−1
. The set of
fixed points of σ is K = SO(n). The map
G/K →R
n
2
, AK →AA
t
,
identifies G/K with the space of positive definite symmetric (n n)-matrices
with determinant 1. We view these as the space of normalized inner products
on R
n
, or, more geometrically via their unit balls, as the space of ellipsoids in
R
n
with volume equal to the volume of the Euclidean unit ball. We have
g = sl(n, R) = {A ∈ R
n
2
| tr A = 0},
k = so(n) = {U ∈ sl(n, R) | U
t
= −U},
p = {X ∈ sl(n, R) | X
t
= X}.
The symmetric bilinear form
'A, B` = tr AB (B.39)
on R
n
2
is invariant under the adjoint action of Gl(n, R). Its restriction to p
is positive definite and turns (G, K) into a Riemannian symmetric pair. Since
Appendix B Symmetric Spaces 149
'· , · ` is Ad
G
-invariant, each ad
A
, A ∈ g, is skew-symmetric with respect to
'· , · `. Therefore we have, for X, Y, Z ∈ p,
'R(X, Y )Y, X` = '[X, Y ], [X, Y ]` ≤ 0, (B.40)
where we note that '· , · ` is negative definite on k.
The complexification is g
C
= sl(n, C) = {A ∈ C
n
2
| tr A = 0}. We see that
k +ip = su(n) = {U ∈ sl(n, C) | U
t
= −U}.
Hence (SU(n), SO(n)) is a dual pair to (Sl(n, R), SO(n)), where the involution
on SU(n) is given by the passage to the conjugate matrix. See also Exer-
cise B.59.
B.41 Exercises. 1) Discuss (Gl(n, R), O(n)) as a symmetric pair.
2) Show that Sl(2, R)/ SO(2) with the above Riemannian metric has con-
stant negative curvature −2 and construct isometries to some (other) models
of the hyperbolic plane (with curvature −2).
B.42 Example. We show that the Grassmann manifold G
F
(p, q) of p-planes
in F
p+q
together with a natural Riemannian metric is a Riemannian symmetric
space, where F = R, C, or H. We also discuss the dual symmetric space
G

F
(p, q). These symmetric spaces are also discussed in Examples 2.2.4, 4.10.5,
4.10.6, and B.83.1.
Let G = O(p+q), U(p+q), and Sp(p+q) for F = R, C, and H, respectively
25
.
The natural action of G on G
F
(p, q) is transitive
26
. The stabilizer of the p-plane
F
p
{0} ⊂ F
p+q
is K = O(p)O(q), U(p)U(q), and Sp(p)Sp(q), respectively,
hence G
F
(p, q) = G/K.
Let S be the reflection of F
p+q
about F
p
{0}. Then conjugation with S
is an involutive automorphism of G with K as its set of fixed points. Hence
(G, K) is a symmetric pair.
We write square (p + q)-matrices as blocks of four matrices corresponding
to the preferred decomposition F
p+q
= F
p
F
q
. In this notation
k =

U 0
0 V

U
t
= −U, V
t
= −V

, (B.43)
p =

0 −X
t
X 0

X ∈ F
q×p


= F
q×p
. (B.44)
We write

A 0
0 B

=: D(A, B) and

0 −X
t
X 0

=: P(X). (B.45)
25
For the definition of Sp(n), see the discussion after (B.69).
26
We multiply vectors in H
p+q
with scalars from the right and with matrices from the left.
150 Lectures on K¨ ahler Manifolds
For D(A, B) ∈ K, D(U, V ) ∈ k, and P(X), P(Y ) ∈ p, we have
Ad
D(A,B)
P(X) = P(BXA
t
),
[D(U, V ), P(X)] = P(V X −XU),
[P(X), P(Y )] = D(Y
t
X −X
t
Y, Y X
t
−XY
t
).
There is an Ad
G
-invariant inner product on g,
'A, B` =
1
2
Re tr(A
t
B). (B.46)
Its restriction to p turns (G, K) into a Riemannian symmetric pair. With
respect to the identification P of F
q×p
with p above, the curvature tensor of
G
F
(p, q) is given by
R(X, Y )Z = XY
t
Z +ZY
t
X −Y X
t
Z −ZX
t
Y. (B.47)
By the Ad
G
-invariance of the inner product on g,
'R(X, Y )Y, X` = |[X, Y ]|
2
≥ 0. (B.48)
Hence the sectional curvature of G
F
(p, q) is non-negative.
Inside G
F
(p, q), we consider the open subset G

F
(p, q) of p-planes on which
the non-degenerate form
Q
p,q
(x, y) = −
¸
j≤p
x
j
y
j
+
¸
j>p
x
j
y
j
(B.49)
on F
p+q
is negative definite. Let G

= O(p, q), U(p, q), and Sp(p, q), respec-
tively, be the group of linear transformations of F
p+q
preserving this form.
Then G

is transitive on G

F
(p, q) and G

F
(p, q) = G

/K, where K is as above.
Conjugation with the reflection S of F
p+q
about F
p
{0} is an involutive
automorphism of G

which has K as its set of fixed points. Hence (G

, K) is
a symmetric pair as well. Now we have
p =

0 X
t
X 0

X ∈ F
q×p


= F
q×p
, (B.50)
whereas K and k are as above. We write

0 X
t
X 0

= P

(X).
Then
[D(U, V ), P

(X)] = P

(V X −XU),
[P

(X), P

(Y )] = D(X
t
Y −Y
t
X, XY
t
−Y X
t
).
Appendix B Symmetric Spaces 151
We see that with respect to the identifications P and P

of p with F
q×p
above,
the Lie bracket [k, p] remains the same, but the Lie bracket [p, p] changes sign.
That is, G
F
(p, q) and G

F
(p, q) are dual symmetric spaces in the sense of Ex-
ercise B.37. There is a corresponding change in sign for curvature tensor and
sectional curvature.
B.51 Exercise. Show that the Riemannian metrics on G
F
(p, q) and G

F
(p, q)
are Einsteinian with Einstein constant k(p +q + 2) −4 and −k(p +q + 2) + 4,
respectively, where k = dim
R
F.
B.52 Exercises. 1) Find other representations of Grassmannians as homoge-
neous spaces, G
F
(p, q) = G/K, and discuss which of the corresponding pairs
(G, K) are Riemannian symmetric pairs.
2) Replacing orthogonal groups by special orthogonal groups, we obtain
the Grassmann manifold G
o
R
(p, q) of oriented p-planes in R
p+q
as homogeneous
space G/K, where G = SO(p+q) and K = SO(p)SO(q). The same involution
σ as above, namely conjugation with the reflection S, leaves G invariant; with
respect to it k and p remain the same as before. Keeping the formula for the
inner product on p, the formula for the curvature tensor remains the same.
The natural “forget the orientation” map G
o
R
(p, q) → G
R
(p, q) is a twofold
Riemannian covering. If p +q ≥ 3, then G
o
R
(p, q) is simply connected.
3) A complex subspace of dimension p in C
n
is also a real subspace of
R
2n
= C
n
of dimension 2p, and similarly with H
n
. Discuss the corresponding
inclusions of Grassmann manifolds.
B.53 Example. Let V be a vector space over a field F. A symplectic form on
V is an alternating two-form ω on V such that for all non-zero v ∈ V there is
a vector w ∈ V with ω(v, w) = 0. A symplectic vector space is a vector space
together with a symplectic form. The standard example is F
2n
with the form
ω((x, y), (u, v)) =
¸
(x
µ
v
µ
−y
µ
u
µ
).
As in this example, the dimension of a finite-dimensional symplectic vector
space V is even, dimV = 2n, and V has a basis (e
1
, . . . , e
n
, f
1
, . . . , f
n
) such
that
ω(e
µ
, e
ν
) = ω(f
µ
, f
ν
) = 0 and ω(e
µ
, f
ν
) = δ
µν
,
a symplectic basis of V . The choice of a symplectic basis identifies V with the
standard symplectic vector space F
2n
.
Let V be a symplectic vector space over a field F with symplectic form
ω. An endomorphism A: V → V is called symplectic if the pull back A

ω =
ω. Since ω is non-degenerate, symplectic endomorphisms are invertible, hence
automorphisms. Under composition, the set of all symplectic automorphisms
of V is a group, denoted Sp(V ) and, respectively, Sp(n, F) if V = F
2n
. Let
J =

0 1
−1 0

∈ Gl(2n, F) (B.54)
152 Lectures on K¨ ahler Manifolds
be the fundamental matrix of the standard symplectic form ω in F
2n
, where
here and below scalars represent the corresponding multiple of the unit matrix
or block. Then a linear map g ∈ Gl(2n, F) is in Sp(n, F) if and only if g
t
Jg = J.
Writing g as a matrix of (n n)-blocks,
g =

A B
C D

, (B.55)
the condition g
t
Jg = J is equivalent to the conditions
A
t
C = C
t
A, B
t
D = D
t
B, and A
t
D −C
t
B = 1. (B.56)
It follows that the group of g ∈ Sp(n, F) with B = C = 0 is an embedded
general linear group,
Gl(n, F) →Sp(n, F), A →

A 0
0 (A
t
)
−1

. (B.57)
This induces the diagonal embedding of O(n, F) into O(2n, F).
A subspace L of V is isotropic if ω(v, w) = 0 for all v, w ∈ L. A subspace
of V is called a Lagrangian subspace if it is a maximal isotropic subspace. If
the dimension of V is 2n, then Lagrangian subspaces of V have dimension n. If
(e
1
, . . . , e
n
, f
1
, . . . , f
n
) is a symplectic basis of V , then the subspace L
0
spanned
by e
1
, . . . , e
n
is Lagrangian. We denote by G(L, V ) and G
F
(L, n) the space of
Lagrangian subspaces of V and F
2n
, respectively.
Let F = R and identify R
2n
= C
n
by writing (x, y) = x +iy = z. We have
'v, w` = Re'v, w` +iω(v, w),
where the left hand side is the standard Hermitian form on C
n
and where
Re'v, w` and ω(v, w) are the Euclidean inner product and the standard sym-
plectic form on R
2n
, respectively. The real subspace L
0
spanned by the unit
vectors e
1
, . . . , e
n
in C
n
is Lagrangian. If L is any other Lagrangian subspace
of V and (b
1
, . . . , b
n
) is an orthonormal basis of L, then (b
1
, . . . , b
n
) is a unitary
basis of C
n
with its standard Hermitian form 'v, w`. Therefore the natural
action of U(n) on C
n
induces a transitive action on G
R
(L, n) with stabilizer
O(n) of L
0
, hence
G
R
(L, n) = U(n)/ O(n). (B.58)
Let S: C
n
→ C
n
be the real automorphism S(z) = z corresponding to the
reflection S(x, y) = (x, −y) of R
2n
. Then we have σ(A) := SAS
−1
= A for any
A ∈ U(n). Hence the involution σ on U(n) has O(n) as its set of fixed points,
turning (U(n), O(n)) into a Riemannian symmetric pair. Thus G
R
(L, n) is a
symmetric space of dimension n(n + 1)/2 and a totally geodesic submanifold
of the Grassmannian G
R
(n, n).
Appendix B Symmetric Spaces 153
B.59 Exercise. We say that a Lagrangian subspace L of V = C
n
(as above)
is special if the n-form dz
1
∧ · · · ∧ dz
n
is real valued on L. Show that the
above Lagrangian subspace L
0
is special, that the space G
R
(SL, n) of all spe-
cial Lagrangian subspaces of V is given by the submanifold SU(n)/ SO(n) of
G
R
(L, n) = U(n)/ O(n), and that G
R
(SL, n) is totally geodesic in G
R
(L, n).
Under the above identification R
2n
= C
n
, A ∈ O(2n) is complex linear iff
the pull back A

ω = ω. In other words,
U(n) = O(2n) ∩ Sp(n, R). (B.60)
Write g ∈ U(n) as g = A + iB with real (n n)-matrices A and B. Then this
identification reads
U(n) ÷ A +iB ←→

A −B
B A

∈ O(2n) ∩ Sp(n, R), (B.61)
where A
t
B = B
t
A and A
t
A+B
t
B = 1. In particular, O(n) = O(n, R) ⊂ U(n)
is diagonally embedded in O(2n) ∩ Sp(n, R), compare with (B.57) above.
The symmetric space Gl(n, R)/ O(n) is dual to U(n)/ O(n) = G
R
(L, n) in
the sense of Exercise B.37. Motivated by the example of the Grassmannians,
we would like to see if we can view Gl(n, R)/ O(n) as the open set G

R
(L, n) ⊂
G
R
(L, n) of Lagrangian subspaces on which the symmetric form Q
n,n
from
(B.49) is negative definite.
To that end, let O(n, n) be the orthogonal group of Q
n,n
. Clearly g ∈
Gl(2n, R) belongs to O(n, n) iff
A
t
B = C
t
D, A
t
A −C
t
C = 1, and D
t
D−B
t
B = 1, (B.62)
where we write g as a matrix of blocks as in (B.55). Let g ∈ O(n, n) ∩Sp(n, R).
By (B.56) and (B.62),
B = (A
t
)
−1
C
t
D = CA
−1
D. (B.63)
Writing D = AX, we compute
(1 +C
t
C)X = A
t
AX = A
t
D = 1 +C
t
B = 1 +C
t
CX. (B.64)
Hence X = 1, and hence D = A and B = C. We conclude that the intersection
O(n, n) ∩ Sp(n, R) consists of the matrices

A B
B A

(B.65)
with A
t
B = B
t
A and A
t
A − B
t
B = 1. We also see that the subgroup of
matrices in O(n, n) ∩ Sp(n, R) with B = 0 is a diagonally embedded O(n).
154 Lectures on K¨ ahler Manifolds
Let L
0
∈ G

R
(L, n) be the Lagrangian subspace spanned by the first n unit
vectors. Let L ∈ G

R
(L, n) be another Lagrangian subspace. Let (b
1
, . . . , b
n
)
be an orthogonal basis of L, that is, Q
n,n
(b
j
, b
k
) = −δ
jk
. Let A and B be
the (n n)-matrices with columns the first n and the last n coordinates of the
vectors b
j
, respectively. Since L is Lagrangian and (b
1
, . . . , b
n
) is an orthogonal
basis of L, the corresponding matrix g as in (B.65) belongs to O(n, n)∩Sp(n, R).
Since g maps L
0
to L, we conclude that O(n, n) ∩ Sp(n, R) is transitive on
G

R
(L, n). The stabilizer of L
0
is the subgroup of matrices in O(n, n)∩Sp(n, R)
with B = 0, hence
G

R
(L, n) = (O(n, n) ∩ Sp(n, R))/ O(n). (B.66)
Let
k :=
1

2

1 1
−1 1

∈ O(2n) ∩ Sp(n, R). (B.67)
We have
k Gl(n, R)k
−1
= O(n, n) ∩ Sp(n, R),
k O(n)k
−1
= O(n),
(B.68)
where we consider O(n) and Gl(n, R) as embedded subgroups in Sp(n, R) as
above. We conclude that conjugation with k identifies Gl(n, R)/ O(n) with
(O(n, n) ∩ Sp(n, R))/ O(n) = G

R
(L, n). (B.69)
Let F = C and identify C
2n
= H
n
by writing z ∈ H
n
as z = x + jy with
x, y ∈ C
2n
, where we multiply with scalars from the right. We have z = x −jy
and jx = xj. For z = x +jy and w = u +jv in H
n
, we get
'z, w` = 'x −jy, u +jv`
= ('x, u` +'y, v`) +j('x, v` −'y, u`)
= ((x, y), (u, v)) +jω((x, y), (u, v)),
where the first term on the right corresponds to the standard Hermitian form
and ω to the standard symplectic form on C
2n
. The group of quaternionic
(nn)-matrices preserving the form 'z, w` on H
n
is called the symplectic group,
denoted Sp(n). Considered as linear maps on C
2n
, elements of Sp(n) preserve
the Hermitian form and the symplectic form on C
2n
. For any unit quaternion
a,
'az, aw` = 'za, aw` = 'z, w`,
so that left multiplication by a defines an element of Sp(n).
The complex subspace spanned by the unit vectors e
1
, . . . , e
n
in H
n ∼
= C
2n
is Lagrangian. If L ⊂ C
2n
is any Lagrangian subspace of C
2n
and (b
1
, . . . , b
n
)
Appendix B Symmetric Spaces 155
is a unitary basis of L, then 'b
µ
, b
ν
` = δ
µν
. Hence mapping the standard basis
(e
1
, . . . , e
n
) of H
n
to (b
1
, . . . , b
n
) defines an element of Sp(n). It follows that
G
C
(L, n) = Sp(n)/ U(n). (B.70)
The automorphism s(x + jy) = x − jy of H induces a complex automorphism
S: H
n
→ H
n
, defined by the analogous formula S(x + jy) = x − jy. We
see that S corresponds to the reflection S(x, y) = (x, −y) of C
2n
. Since s is
an automorphism of H, we have SAS
−1
= σ(A), where σ(A) is the matrix
obtained by applying s to the entries of A. Clearly σ is an involution of Sp(n)
with U(n) as its set of fixed points, turning (Sp(n), U(n)) into a Riemannian
symmetric pair. Thus G
C
(L, n) is a symmetric space of dimension n(n + 1)
and a totally geodesic submanifold of G
C
(n, n). Since the center of U(n) is not
discrete, G
C
(L, n) is a Hermitian symmetric space, see Proposition B.86.
Under the identification C
2n
= H
n
, A ∈ U(2n) is quaternion linear if and
only if the pull back A

ω = ω. In other words,
Sp(n) = U(2n) ∩ Sp(n, C). (B.71)
Write g ∈ Sp(n) as g = A+jB with complex (n n)-matrices A and B. Then
this identification reads
Sp(n) ÷ A +jB ←→

A −B
B A

∈ U(2n) ∩ Sp(n, C), (B.72)
where A
t
B = B
t
A and A
t
A + B
t
B = 1. In particular, U(n) ⊂ Sp(n) is
diagonally embedded in U(2n) with blocks A and A on the diagonal.
Again we want to identify the dual symmetric space Sp(n, R)/ U(n) with
the open subset G

C
(L, n) ⊂ G
C
(L, n) of Lagrangian subspaces on which the
Hermitian form Q
n,n
from (B.49) is negative definite, where U(n) = O(2n) ∩
Sp(n, R) as in (B.60).
We proceed as in the real case. First of all we note that g ∈ Gl(2n, C)
belongs to the unitary group U(n, n) of Q
n,n
iff
A
t
B = C
t
D, A
t
A −C
t
C = 1, and D
t
D−B
t
B = 1, (B.73)
where we write g as a matrix of blocks as in (B.55). Let g ∈ U(n, n) ∩Sp(n, C).
By (B.56) and (B.73),
B = (A
t
)
−1
C
t
D = CA
−1
D. (B.74)
Writing D = AX, we compute
(1 +C
t
C)X = A
t
AX = A
t
D = 1 +C
t
CX, (B.75)
156 Lectures on K¨ ahler Manifolds
hence X = 1, and hence D = A and C = B. We conclude that the intersection
U(n, n) ∩ Sp(n, C) consists of the matrices

A B
B A

(B.76)
with A
t
B = B
t
A and A
t
A−B
t
B = 1. The subgroup of g ∈ U(n, n) ∩Sp(n, C)
with B = 0 is a diagonally embedded U(n) as above. As in the real case, we
get
G

C
(L, n) = (U(n, n) ∩ Sp(n, C))/ U(n). (B.77)
Let
k :=
1

2

1 i
i 1

∈ U(2n) ∩ Sp(n, C). (B.78)
We have k
−1
= k and hence
kk
−1
= k
2
=

0 i
i 0

. (B.79)
Let g ∈ Sp(n, C). Then k
−1
gk ∈ Sp(n, R) iff k
−1
gk = k
−1
gk, that is, iff
kk
−1
g = gkk
−1
. The latter holds iff D = A and C = B, that is, iff g ∈
U(n, n) ∩ Sp(n, C). We conclude that
k Sp(n, R)k
−1
= U(n, n) ∩ Sp(n, C). (B.80)
We also have
k(O(2n) ∩ Sp(n, R))k
−1
= U(n), (B.81)
where we consider U(n) on the right as an embedded subgroup in Sp(n, C) as
above. We finally conclude that conjugation with k identifies Sp(n, R)/ U(n)
with
(U(n, n) ∩ Sp(n, C))/ U(n) = G

C
(L, n). (B.82)
Since the center of U(n) is not discrete, G

C
(L, n) is a Hermitian symmetric
space, see Remark B.87. Moreover, G

C
(L, n) is biholomorphic to the Siegel
upper half plane. For more on this, see e.g. [Bu, Section 31] or [Hel, Exercise
VIII.B].
B.3 Hermitian Symmetric Spaces. A Hermitian symmetric space is a
symmetric space M together with a parallel almost complex structure on M
which is compatible with the Riemannian metric on M in the sense of (2.33).
Since parallel almost complex structures are complex structures, any such struc-
ture turns M into a K¨ahler manifold.
Symmetric spaces of non-compact type are simply connected (Corollary B.14),
but there are symmetric spaces of compact type with non-trivial fundamental
Appendix B Symmetric Spaces 157
group. However, Hermitian symmetric spaces of compact type are Fano mani-
folds and therefore simply connected, by Theorem 6.12
27
.
Let (G, K) be a Riemannian symmetric pair. Then by Theorem B.24.3,
compatible parallel complex structures on M = G/K correspond to complex
structures on p which preserve the inner product on p and are invariant under
the adjoint action of K on p.
B.83 Examples. 1) Complex Grassmannians. We follow the notation in Ex-
ample B.42 and consider
G
C
(p, q) = U(p +q)/ U(p) U(q) = G/K,
the Grassmannian of complex p-planes in C
p+q
, endowed with the Riemannian
metric introduced in (B.46). We define a complex structure on p by
JP(X) := P(iX),
where X ∈ C
q×p
and P(X) ∈ p is defined as in (B.45). For D(A, B) ∈ K
we have Ad
D(A,B)
P(X) = P(BXA
t
), hence J commutes with Ad
K
, turning
G
C
(p, q) into a Hermitian symmetric space.
The curvature tensor of G
C
(p, q) is given in (B.48). For p = 1, that is, in
the case of complex projective space, we get
R(X, Y )Y = 'Y, Y `X −'X, Y `Y + 3'X, JY `JY.
Thus the holomorphic sectional curvature of complex projective space is con-
stant equal to 4 and the range of its sectional curvature is [1, 4]. We recommend
Karcher’s article [Kar] for a nice exposition of the elementary geometry of com-
plex projective spaces.
To compare the Riemannian metric on G
C
(p, q) here with the one introduced
in Example 4.10, we note that both are invariant under the action of U(p +q).
Thus it suffices to compare them at the plane spanned by the first p unit vectors.
For X ∈ p, we have
e
tX
=

1 −tX
t
tX 1

+o(t).
Comparing with (4.14), we get that the Riemannian metric there is twice the
Riemannian metric discussed here.
The discussion in the case of the dual Grassmannians G

C
(p, q) is similar.
The case p = 1 is of particular importance, U(1, m)/ U(1) U(m) =: CH
m
is
called complex hyperbolic space. It is dual to complex projective space. With
respect to the given normalization of the Riemannian metric, the range of its
sectional curvature is [−4, −1].
27
There are more direct proofs for the latter assertion which do not rely on Kobayashi’s
theorem, see e.g. [Hel, Theorem VIII.4.6].
158 Lectures on K¨ ahler Manifolds
2) Complex quadric. Consider the Grassmannian manifold G
o
R
(2, n − 2) of
oriented real 2-planes in R
n
as in Exercise B.52.2,
G
o
R
(2, n −2) = SO(n)/ SO(2) SO(n −2).
There is a complex structure J on p which is invariant under the adjoint repre-
sentation of SO(2) SO(n−2): Identify R
(n−2)×2 ∼
= C
n−2
, that is, think of the
two columns of X ∈ R
(n−2)×2
as real and imaginary part of a vector in C
n−2
,
and multiply this vector by i. Thus G
o
R
(2, n−2) together with the Riemannian
metric induced by the inner product on p and the parallel complex structure
induced by J is a K¨ ahler manifold.
We can think of G
o
R
(2, n −2) as the complex quadric
Q = {[z] ∈ CP
n−1
| z
2
1
+ · · · +z
2
n
= 0}. (B.84)
To that end, extend the action of SO(n) on R
n
complex linearly to C
n
to
realize SO(n) as a subgroup of U(n). Let p
0
= [1, i, 0, . . . , 0] ∈ Q. The orbit of
p
0
under the induced action of SO(n) on CP
n−1
is equal to Q. The stabilizer
of p
0
is SO(2) SO(n−2), thus Q = SO(n)/ SO(2) SO(n−2) = G
o
R
(2, n−2).
To compare the metric chosen above with the one on CP
n−1
, we let
A =

A
2
1
n−2

with A
2
=
1

2

1 i
i 1

,
where 1
r
denotes the unit matrix of size r r. Since A is unitary, A induces
an isometry of CP
n−1
. Moreover, A maps p
1
= [1, 0, . . . , 0] to p
0
. For X ∈ p,
we get
e
tX
p
0
= e
tX
Ap
1
= A(A
−1
e
tX
A)p
1
.
We have
A
−1
e
tX
A =

1
2
−tA
t
2
X
t
tXA
2
1
n−2

+o(t).
Since the speed of the curve e
tX
p
0
at t = 0 is equal to the speed of the curve
(A
−1
e
tX
A)p
1
at t = 0, we get that the Riemannian metric induced fromCP
n−1
,
normalized as in Example 1 above, is equal to the chosen metric on G
o
R
(2, n−2).
Intersecting two-planes with the unit sphere S
n−1
, we identify G
o
R
(2, n−2)
with the space of oriented great circles on S
n−1
. In particular, G
o
R
(2, 1) is a
two-sphere or, what amounts to the same, a complex projective line. Thus the
quadric Q ⊂ CP
2
in (B.84) is a complex projective line.
B.85 Remark. Let M be a simply connected Hermitian symmetric space with
complex structure J, and let p be a point in M. Then J
p
is the differential of an
isometry j
p
of M fixing p, by Corollary B.6 and (4.43). By Exercise B.2, j
p
is in
the center of the group of holomorphic isometries of M fixing p. Moreover, since
J
p
is skew-symmetric and orthogonal, the endomorphisms cos x · id + sinx · J
p
of T
p
M preserve metric and curvature tensor. Hence, by Corollary B.6, j
p
belongs to the component of the identity of the stabilizer of p in the group of
holomorphic isometries of M.
Appendix B Symmetric Spaces 159
Recall the definition of infinitesimally effective from Remark B.27.
B.86 Proposition. Let (G, K) be an infinitesimally effective Riemannian sym-
metric pair with G and K compact and connected. Assume that G is semi-
simple and that the adjoint representation of K on p is irreducible. Then
M = G/K admits a G-invariant complex structure compatible with the Rie-
mannian metric on M if and only if the center of K is not discrete, that is, if
and only if K is not semi-simple. If the latter holds, then M is a Hermitian
symmetric space of compact type, hence simply connected. If, moreover, (G, K)
is effective, then the center Z
K
of K is a circle and J is (any) one of the two
elements of order 4 in Z
K
.
Proof. Since G is compact and semi-simple, M is a symmetric space of compact
type. Furthermore, π
1
(G) is finite and π
2
(G) = 0. The final part of the long
exact homotopy sequence associated to the fiber bundle G →G/K = M is
0 = π
2
(G) →π
2
(M) →π
1
(K) →π
1
(G) →π
1
(M) →0.
It follows that π
1
(M) is finite. Furthermore, π
2
(M) is infinite if and only
if π
1
(K) is infinite. Since M is compact with finite fundamental group, the
universal covering space of M is compact. Hence π
2
(M) is finitely generated,
and hence H
2
(M, R) = 0 if and only if π
2
(M) is infinite.
If M is a compact Hermitian symmetric space, then M is K¨ahlerian and
hence H
2
(M, R) = 0. By what we just said, this implies that π
1
(K) is infinite.
Since K is compact and connected, this happens precisely if the center of K is
infinite or, equivalently, if the center of K is not discrete.
Vice versa, assume that the center of K is infinite. Since (G, K) is infinites-
imally effective and K is compact, the normal subgroup N as in Remark B.27
is finite. Therefore the center of K/N is infinite as well. Hence we may pass to
the quotient pair (G/N, K/N), which is also a Riemannian symmetric pair. In
conclusion, we may assume that the adjoint representation of K on p is faithful.
Let Z ⊂ End(p) be the space of all endomorphisms of p commuting with all
Ad
k
, k ∈ K. Since the adjoint representation of K on p is irreducible, Z is an
associative division algebra over R, hence Z

= R, C, or H.
Since the adjoint representation of K on p is faithful, K ⊂ End(p). Hence
the center Z
K
of K is a subgroup of Z
×
. Since K is compact, Z
K
is a compact
subgroup of Z
×
. Hence if Z
K
is not finite, Z

= C or Z

= H. Since maximal
Abelian subgroups of H
×
are isomorphic to C
×
, it follows that Z
K
is a circle.
If J ∈ Z
K
is one of the two elements of order 4, then J
2
= −1 and J is an
Ad
K
-invariant complex structure on p, compatible with the inner product on p
since Z
K
⊂ K. Thus J turns M = G/K into a Hermitian symmetric space of
compact type. In particular, M is simply connected. By Remark B.85, there
are no other but the above two choices for an invariant complex structure on
M.
160 Lectures on K¨ ahler Manifolds
B.87 Remark. Let (G, K) and (G

, K

) be dual symmetric pairs as in Exer-
cise B.37 with K = K

. Then a complex structure J on p is invariant under
the adjoint representation of K and compatible with the inner product on p if
and only if the corresponding complex structure J

on p

is invariant under the
adjoint representation of K

and compatible with the inner product on p

.
For example, let M be a symmetric space of non-compact type, G be the
component of the identity in the group of isometries of M and K be the stabi-
lizer of a point in M. Then M is simply connected and given by the symmetric
pair (G, K). Let M

be the simply connected dual symmetric space of M, G

be the group of isometries of M

and K

be the stabilizer of a point in M

.
Then M

is of compact type and G

is compact and semi-simple. Up to iso-
morphism, p

= ip and hence K = K

by Theorem B.5 and Theorem B.24.4.
Hence M is Hermitian iff M

is Hermitian iff K = K

is not semi-simple. In
other words, the criterion of Proposition B.86 also applies to symmetric spaces
of non-compact type.
B.88 Exercises. 1) Interpret O(2n)/ U(n) as space of complex structures on
R
2n
which preserve the Euclidean norm and discuss (O(2n), U(n)) as a Rie-
mannian symmetric pair.
2) Determine the space of all complex structures on R
2n
as a homogeneous
space G/K. Interpret Sp(n, R)/ U(n) as a space of complex structures on R
2n
.
Let D ⊂ C
m
be a bounded domain. We say that D is symmetric if, for all
p ∈ M, there is a biholomorphic transformation s
p
: D →D such that s
p
(p) = p
and ds
p
(p) = −id. Recall that the Bergmann metric of D is invariant under
all biholomorphic transformations of D, see Example 4.10.7. In particular,
transformations s
p
as above are geodesic reflections. It follows that symmetric
bounded domains with the Bergmann metric are Hermitian symmetric spaces.
It then also follows that they are Einstein spaces with Einstein constant −1,
see Example 4.10.7, and, hence, that they are of non-compact type.
Vice versa, for any Hermitian symmetric space M of non-compact type,
there is a symmetric bounded domain D and a biholomorphic map M → D
which is, up to rescaling, an isometry between M and D, equipped with the
Bergmann metric, see Theorem VIII.7.1 in [Hel]. We have seen this explicitly
28
in the case of the dual complex Grassmannians G

r,n
, see Example 4.10.6.
We say that D ⊂ C
m
is a tube domain if D = R
m
+iΩ, where Ω ⊂ R
m
is an
open and convex subset such that, for all y ∈ Ω and t > 0, ty ∈ Ω and such that
Ω does not contain complete lines. For example, the upper half plane is a tube
domain. Each tube domain in C
m
is biholomorphic to a bounded domain in
C
m
, but the converse does not hold. The article [Ma2] by Matsushima contains
a nice introduction into tube domains.
28
We did not identify the Bergmann metric there. The irreducibility of the isotropy repre-
sentation for the two metrics under consideration shows, however, that they must be multiples
of each other.
Appendix B Symmetric Spaces 161
We say that a Hermitian symmetric space is of tube type if it is biholomorphic
to a tube domain. The hyperbolic plane is of tube type. The dual Lagrangian
Grassmannians G

C
(L, n) = Sp(n, R)/ U(n) (as in Example B.53) are of tube
type, see Exercise VIII.B in [Hel]. See [BIW], [KW], [Wi] for more information
on Hermitian symmetric spaces of tube type.
Appendix C Remarks on Differential Operators
Let M be a Riemannian manifold. For a vector bundle E →M, we denote by
E(M, E) and E
c
(M, E) the space of smooth sections of E and smooth sections
of E with compact support, respectively. We also use the shorthand E(E) and
E
c
(E) for these spaces. If E is Hermitian, we denote by L
2
(E) = L
2
(M, E) the
Hilbert space of square integrable measurable sections σ of E, endowed with
the L
2
-inner product from (1.2) and corresponding L
2
-norm |σ|
2
.
Let E, F →M be Hermitian vector bundles and D be a differential operator
of order m ≥ 0 from (smooth sections of) E to (smooth sections of) F,
D: E(M, E) →E(M, F). (C.1)
We also view D as an unbounded operator on L
2
(M, E) with domain E
c
(M, E).
The formal adjoint D

of D is a differential operator of order m from F to E
and satisfies, by definition,
(Dσ, τ)
2
= (σ, D

τ)
2
(C.2)
for all σ ∈ E
c
(M, E) and τ ∈ E
c
(M, F). We say that D is formally self-adjoint
or formally symmetric if E = F and D = D

.
By the divergence formula 1.9, an equivalent characterization of D

is that,
for all smooth sections σ of E and τ of F, there is a complex vector field
Z = X +iY on M with
(Dσ, τ) = (σ, D

τ) + div Z, (C.3)
where the divergence div Z := div X + i div Y . This is the way we will deter-
mine adjoint operators. The discussion in the case of real vector bundles and
Riemannian metrics is similar.
C.4 Example. Let E be a Hermitian vector bundle over M and suppose that
E is endowed with a Hermitian connection D. We view D as a differential
operator,
D: E(E) →E(T

C
M ⊗E). (C.5)
Note that F = T

C
M ⊗ E inherits a Hermitian metric from the Hermitian
metrics on T

C
M and E and, by the product rule (1.20), a metric connection
ˆ
D
from
ˆ
∇ and D. Let σ be a section of E and τ be a section of T

C
M ⊗ E, that
is, a 1-form with values in E. Let p ∈ M and (X
1
, . . . , X
n
) be an orthonormal
frame of M about p with ∇X
j
(p) = 0. Using the latter and that D is metric,
we have at p
(Dσ, τ) =
¸
(D
Xj
σ, τ(X
j
)) (C.6)
=
¸

X
j
(σ, τ(X
j
)) −(σ, (
ˆ
D
Xj
τ)(X
j
))

= −
¸
(σ, (
ˆ
D
Xj
τ)(X
j
)) + div Z,
Appendix C Remarks on Differential Operators 163
where Z is the complex vector field defined by the property (Z, W) = (σ, τ(W)).
We conclude that
D

τ = −
¸
(
ˆ
D
Xj
τ)(X
j
) = −tr
ˆ
Dτ. (C.7)
The same formula holds in the Riemannian context. In both cases, Riemannian
or Hermitian,
D

Dσ = −
¸
D
2
σ(X
j
, X
j
) = −tr D
2
σ. (C.8)
The principal symbol S
D
of a differential operator D as above associates to
each ξ ∈ T

M a homomorphism S
D
(ξ): E
p
→F
p
, where p is the foot point of
ξ. By definition,
S
D
(dϕ)σ =
1
m!
m
¸
j=0
(−1)
j

m
j

ϕ
j
D(ϕ
m−j
σ) (C.9)
for all smooth functions ϕ on M and sections σ of E. The principal symbol is
homogeneous of order m in ξ, S
D
(tξ) = t
m
S
D
(ξ). For example, if D has order
one, then S
D
is tensorial in ξ = dϕ and
S
D
(dϕ)σ = D(ϕσ) −ϕD(σ). (C.10)
If E = F, then the right hand side of (C.9) can be written as an m-fold
commutator,
S
D
(dϕ)σ =
1
m!
[[. . . [[D, ϕ], ϕ], . . . , ϕ], ϕ]σ, (C.11)
where ϕ is viewed as the operator which multiplies sections with ϕ. Recall also
that D is elliptic if S
D
(ξ) is invertible for all non-zero ξ ∈ T

M.
C.12 Exercises. 1) Compute S
D
in terms of local trivializations of E.
2) Show that S
D
∗(ξ) = (−1)
m
S
D
(ξ)

for all ξ ∈ T

M.
3) Let D be a connection on E, viewed as a differential operator as in (C.5).
Show that S
D
(dϕ)σ = dϕ ⊗σ.
The maximal extension D
max
of D is the adjoint operator of D

in the
sense of Hilbert spaces
29
. By definition, σ ∈ L
2
(M, E) belongs to the domain
domD
max
of D
max
iff there is a section σ

∈ L
2
(M, E) such that

, τ)
2
= (σ, D

τ)
2
for all τ ∈ E
c
(M, F), (C.13)
and then D
max
σ := σ

. The operator D
max
is closed and extends D. In
particular, D is closable. Furthermore,
L
2
(M, E) = ker D
max
+ imD

, (C.14)
an L
2
-orthogonal decomposition of L
2
(M, E), where we note that ker D
max
is
a closed subspace of L
2
(M, E).
29
Section III.5 in [Kat] is a good reference for the functional analysis we need.
164 Lectures on K¨ ahler Manifolds
C.15 Exercise. Let σ ∈ E(M, E). Then if σ and Dσ are square integrable,
then σ ∈ domD
max
and D
max
σ = Dσ.
The smallest closed extension D
min
of D is called the minimal extension of
D. The domain domD
min
of D
min
consists of all σ ∈ L
2
(M, E) such that there
is a sequence (σ
n
) in domD = E
c
(M, E) such that σ
n
→ σ in L
2
(M, E) and
the sequence (Dσ
n
) is convergent in L
2
(M, F), and then D
min
σ := limDσ
n
.
By definition, the graph of D
min
is the closure of the graph of D and
D ⊂ D
min
⊂ D
max
. (C.16)
The left inclusion is always strict. By definition, imD
min
⊂ imD, hence
imD
min
= imD. (C.17)
The minimal extension D
min
can also be characterized as the Hilbert space
adjoint of the maximal extension (D

)
max
.
C.18 Exercises. 1) Show that domD
min
endowed with the graph norm of D,
|σ|
D
:= (|σ|
2
2
+ |D
min
σ|
2
2
)
1/2
,
is a Hilbert space, the Sobolev space associated to D.
2) Assuming D = D

, show that
D
a
= D
max
, D
aa
= D
min
, D
aaa
= D
max
,
where the superscript a denotes the Hilbert space adjoint.
We say that a differential operator D is essentially self-adjoint if it is for-
mally self-adjoint and D
min
= D
max
.
C.19 Theorem (Chernoff [Chf]). Let D be elliptic, formally self-adjoint, and
of order one with S
D
uniformly bounded. If M is complete and connected, then
any power D
k
, k ≥ 1, of D is essentially self-adjoint.
The elegant proof of Theorem C.19 in [Chf] uses an existence and uniqueness
result for the associated wave equation. In [LM, Theorem II.5.7], the case k = 1
is treated by a cut-off argument and by using an appropriate parametrix for D.
C.20 Exercise. Let M be complete and connected and D be elliptic of order
one with S
D
uniformly bounded. By passing to the elliptic and formally self-
adjoint operator

0 D

D 0

,
conclude that A
min
= A
max
, where A is any product with alternating factors
D and D

.
Appendix C Remarks on Differential Operators 165
C.21 Proposition. Let D be elliptic of order one with S
D
uniformly bounded.
If M is complete and connected and σ ∈ dom(D

D)
max
, then σ ∈ domD
max
and
(D

Dσ, σ)
2
= |Dσ|
2
2
.
Proof. By Exercise C.20, D
max
= D
min
and (D

D)
max
= (D

D)
min
. Let (σ
j
)
be a sequence of smooth sections of E with compact support such that (σ
j
)
and ((D

D)σ
j
) are Cauchy sequences in L
2
(M, E). We have
|Dσ
j
−Dσ
k
|
2
2
= ((D

D)(σ
j
−σ
k
), (σ
j
−σ
k
))
2
≤ |(D

D)(σ
j
−σ
k
)|
2
|(σ
j
−σ
k
)|
2
,
hence (Dσ
j
) is a Cauchy sequence as well, and hence dom(D

D)
min
⊂ domD
min
.
Clearly the asserted formula holds for any σ
j
, hence also in the limit.
C.22 Corollary. Let D be elliptic of order one with S
D
uniformly bounded,
and let σ ∈ E(M, E) be square integrable. If M is complete and connected, then
D

Dσ = 0 ⇐⇒ Dσ = 0.
C.1 Dirac Operators. Among the elliptic differential operators of order one,
Dirac operators in the sense of Gromov and Lawson [GL], [LM] play a central
role. We say that a Hermitian vector bundle E over a Riemannian manifold M
with Hermitian connection ∇ is a Dirac bundle if it is endowed with a field of
morphisms
TM ⊗E →E, X ⊗σ →Xσ = X · σ,
called Clifford multiplication, such that
XY σ = −Y Xσ −2'X, Y `σ,
(Xσ, τ) = −(σ, Xτ),

X
(Y σ) = (∇
X
Y )σ +Y (∇
X
σ).
(C.23)
The associated Dirac operator D: E(E) →E(E) is then defined by
Dσ =
¸
X
j

Xj
σ, (C.24)
where (X
j
) is a local orthonormal frame of M. We leave it as an exercise to
show that D is formally self-adjoint with principal symbol
S
D
(dϕ)σ = gradϕ · σ. (C.25)
We see that S
D
is uniformly bounded. Hence if M is complete and connected,
then any power D
k
, k ≥ 1, of D is essentially self-adjoint, by Theorem C.19.
We also note that
X · (X · σ) = −|X|
2
σ,
hence D is elliptic.
166 Lectures on K¨ ahler Manifolds
C.26 Exercise. Let E be a Dirac bundle over M. Let F be a Hermitian vector
bundle over M, endowed with a Hermitian connection, also denoted ∇. Then
E ⊗ F with the induced connection and Clifford multiplication X · (σ ⊗τ) :=
(X· σ) ⊗τ is also a Dirac bundle, the twist of E by F. Compare the associated
Dirac operators of E and F with the twist construction in (8.12).
Let E be a Dirac bundle and µ: E → E be a parallel unitary involution
anticommuting with Clifford multiplication, µ(Xσ) = −Xµ(σ). Then the asso-
ciated eigenbundles E
±
for the eigenvalues ±1 of µ are parallel and turn E into
a graded Dirac bundle, E = E
+
⊕ E

. Furthermore, D restricts to operators
D
±
: E(E
±
) →E(E

). We leave it as an exercise to show that (D
+
)

= D

.
C.27 Examples. 1) The morphism TM ⊗A

(M, E) →A

(M, E),
X · α := X

∧ α −Xα,
satisfies the axioms for Clifford multiplication in (C.23), see (1.10) and (1.11),
and d+d

is the associated Dirac operator, see (1.21) and Proposition 1.27. We
note that, as a Dirac bundle, A

(M, E) is the twist of A

(M, C) with E in the
sense of Exercise C.26. The splitting A

(M, E) = A
even
(M, E) ⊕ A
odd
(M, E)
turns A

(M, E) into a graded Dirac bundle.
2) Let M be a complex manifold with complex structure J, endowed with a
compatible Riemannian metric g. Let E →M be a holomorphic vector bundle
with Hermitian metric h and associated Chern connection D. For X ∈ TM,
let
Z
X
:=
1
2
(X −iJX) and Z

X
:= 'Z
X
, · ` ∈ A
0,1
(M, C),
compare Subsection 3.3. Then the morphism TM⊗A
p,∗
(M, E) →A
p,∗
(M, E),
X · α :=

2Z

X
∧ α −

2Z
X
α (C.28)
satisfies the axioms for Clifford multiplication in (C.23).
Let (X
1
, Y
1
, . . . , X
m
, Y
m
) be a local orthonormal frame of M with JX
j
= Y
j
.
By (3.32) and (3.33),
∂ =
¸

Z

Xj

ˆ
D
Xj
+Z

Yj

ˆ
D
Yj

,


= −
¸

Z
Xj

ˆ
D
Xj
+Z
Yj

ˆ
D
Yj

,
where we use that iZ

X
= Z

JX
and iZ
X
= −Z
JX
. It follows that

2(∂ + ∂

)
is the Dirac operator associated to the Clifford multiplication in (C.28).
We note that, as a Dirac bundle, A
p,∗
(M, E) is the twist of A
0,∗
(M, C)
with A
p,0
(M, E) = A
p,0
(M, C) ⊗ E in the sense of Exercise C.26. As in the
previous example, the splitting A
p,∗
(M, E) = A
p,even
(M, E) ⊕ A
p,odd
(M, E)
turns A
p,∗
(M, E) into a graded Dirac bundle.
Appendix C Remarks on Differential Operators 167
C.2 L
2
-de Rham Cohomology. Let E →M be a Hermitian vector bundle
with a flat Hermitian connection D. The exterior differential d = d
D
is a dif-
ferential operator of order one on A

(M, E). By (C.14), we have L
2
-orthogonal
decompositions
L
2
(A

(M, E)) = ker d
max
+ imd

= ker d

max
+ imd, (C.29)
where d

max
stands for (d

)
max
. Since E is flat, d
2
= 0 and hence
imd ⊂ imd
max
⊂ ker d
max
, imd

⊂ imd

max
⊂ ker d

max
. (C.30)
It follows that ker d
max
contains the closure of imd
max
, where we recall that
ker d
max
and ker d

max
are closed subspaces of L
2
(A

(M, E)).
We let H

2
(M, E) := ker d
max
∩ ker d

max
be the space of square integrable
harmonic forms (with values in E). If α is a harmonic form, then α satisfies
(d + d

)α = 0 weakly. Since d + d

is an elliptic differential operator, see e.g.
Exercise C.27.1, it follows that α is smooth and satisfies (d + d

)α = 0 in the
classical sense.
C.31 Theorem (Hodge Decomposition). We have L
2
-orthogonal decomposi-
tions
ker d
max
= H

2
(M, E) + imd, ker d

max
= H

2
(M, E) + imd

.
Proof. Let α ∈ H

2
(M, E) and β ∈ A

c
(M, E). Then (α, dβ)
2
= (d

max
α, β)
2
=
0, hence α is perpendicular to imd. Vice versa, if α ∈ ker d
max
is L
2
-perpendicular
to imd, then d

max
α = 0, and hence α ∈ H

2
(M, E). The proof of the second
equality is similar.
By the first inclusion in (C.30), there is a cochain complex
· · ·
dmax
−−−→ C
k−1
dmax
−−−→ C
k
dmax
−−−→ C
k+1
dmax
−−−→ · · · (C.32)
with C
l
:= domd
max
∩ L
2
(A
l
(M, E)). We let
Z
k
(M, E) = ker{d
max
: C
k
→L
2
(A
k+1
(M, E))},
B
k
(M, E) = im{d
max
: C
k−1
→L
2
(A
k
(M, E))}.
(C.33)
The L
2
-de Rham cohomology
30
of M consists of the quotients
H
k
2
(M, E) = Z
k
(M, E)/B
k
(M, E). (C.34)
The image B
k
(M, E) need not be closed, but its closure B
k
(M, E) is contained
in ker d
max
since the latter is closed in L
2
(A

(M, E)). We define the reduced
L
2
-de Rham cohomology of M by
H
k
2,red
(M, E) := Z
k
(M, E)/B
k
(M, E). (C.35)
We note the following consequences of Exercise C.20 and Corollary C.22.
30
[Car] and [Lot] contain discussions of L
2
-de Rham cohomology.
168 Lectures on K¨ ahler Manifolds
C.36 Proposition. If M is complete and connected, then the natural projec-
tion H

2
(M, E) →H

2,red
(M, E) is an isomorphism.
C.37 Proposition. Let α ∈ A

(M, E) be square integrable. If M is complete
and connected and ∆
d
α = 0, then dα = d

α = 0.
Suppose now that M is oriented. Since ∗ ⊗ h commutes with the Laplace
operator ∆
d
, it induces conjugate linear isomorphisms
H
r
2
(M, E) →H
n−r
2
(M, E

), (C.38)
compare (1.56). In particular, Poincar´e duality holds for reduced L
2
-cohomology
if M is oriented, complete, and connected.
C.3 L
2
-Dolbeault Cohomology. Let M be a complex manifold with com-
plex structure J, endowed with a compatible Riemannian metric g. Let E →M
be a holomorphic vector bundle with Hermitian metric h and associated Chern
connection D. By (C.14), we have L
2
-orthogonal decompositions
L
2
(A
∗,∗
(M, E)) = ker ∂
max
+ im∂

= ker ∂

max
+ im∂. (C.39)
As in the case of the exterior differential d,
im∂ ⊂ im∂
max
⊂ ker ∂
max
, im∂

⊂ im∂

max
⊂ ker ∂

max
. (C.40)
We let H
∗,∗
2
(M, E) := ker ∂
max
∩ ker ∂

max
be the space of square integrable ∂-
harmonic forms (with values in E). If α is a ∂-harmonic form, then α satisfies
(∂ + ∂

)α = 0 weakly. Since ∂ + ∂

is an elliptic differential operator, see e.g.
Exercise C.27.2, it follows that α is smooth and satisfies (∂ + ∂

)α = 0 in the
classical sense.
C.41 Theorem (Hodge Decomposition). We have L
2
-orthogonal decomposi-
tions
ker ∂
max
= H
∗,∗
2
(M, E) + im∂, ker ∂

max
= H
∗,∗
2
(M, E) + im∂

.
By the first inclusion in (C.40), there are cochain complexes
· · ·
∂max
−−−→ C
p,q−1
∂max
−−−→ C
p,q
∂max
−−−→ C
p,q+1
∂max
−−−→ · · · (C.42)
with C
r,s
:= dom∂
max
∩ L
2
(A
r,s
(M, E)). We let
Z
p,q
(M, E) = ker{∂
max
: C
p,q
→L
2
(A
p,q+1
(M, E))},
B
p,q
(M, E) = im{∂
max
: C
p,q−1
→L
2
(A
p,q
(M, E))}.
(C.43)
Appendix C Remarks on Differential Operators 169
The L
2
-Dolbeault cohomology of M consists of the quotients
H
p,q
2
(M, E) := Z
p,q
(M, E)/B
p,q
(M, E). (C.44)
The reduced L
2
-Dolbeault cohomology of M is given by
H
p,q
2,red
(M, E) := Z
p,q
(M, E)/B
p,q
(M, E). (C.45)
We again note consequences of Exercise C.20 and Corollary C.22.
C.46 Proposition. If M is complete and connected, then the natural projec-
tion H
∗,∗
2
(M, E) →H
∗,∗
2,red
(M, E) is an isomorphism.
C.47 Proposition. Let α ∈ A
∗,∗
(M, E) be square integrable. If M is complete
and connected and ∆

α = 0, then ∂α = ∂

α = 0.
Since ∗ ⊗ h commutes with the Laplace operator ∆

, it induces conjugate
linear isomorphisms
H
p,q
2
(M, E) →H
m−p,m−q
2
(M, E

), (C.48)
compare (3.11). In particular, Serre duality holds for reduced L
2
-Dolbeault
cohomology if M is complete and connected.
Suppose now in addition that M is a K¨ ahler manifold and that E is flat.
Then ∆
d
= 2∆

. Moreover, the Lefschetz map L commutes with ∆

.
C.49 Theorem. Suppose that M is a complete and connected K¨ahler manifold
and that E is flat. Then the Lefschetz map L
s
: H
k
2
(M, E) → H
k+2s
2
(M, E) is
injective for 0 ≤ s ≤ m−k and surjective for s ≥ m−k ≥ 0. Furthermore,
H
r
2
(M, E) = ⊕
p+q=r
H
p,q
2
(M, E), H
p,q
2
(M, E) = ⊕
s≥0
L
s
H
p−s,q−s
2,P
(M, E),
the Hodge and Lefschetz decompositions of H
r
2
(M, E) and H
p,q
2
(M, E).
Literature
[Ad] J. F. Adams, Lectures on exceptional Lie groups. With a foreword by J. Pe-
ter May, edited by Zafer Mahmud and Mamoru Mimura, Chicago Lectures
in Mathematics, University of Chicago Press, Chicago, IL, 1996.
[Ar] V. I. Arnold, Mathematical methods of classical mechanics. Translated from
the Russian by K. Vogtmann and A. Weinstein, Grad. Texts in Math. 60,
Springer-Verlag, New York–Heidelberg, 1978.
[At] M. F. Atiyah, Elliptic operators, discrete groups and von Neumann algebras.
Colloque Analyse et Topologie en l’Honneur de Henri Cartan (Orsay, 1974);
Ast´erisque 32–33 (1976), 43–72.
[Au1] T. Aubin, M´etriques riemanniennes et courbure. J. Differential Geometry
4 (1970), 383–424.
[Au2] T. Aubin,
´
Equations du type Monge-Amp`ere sur les vari´et´es k¨ahleriennes
compactes. C. R. Acad. Sci. Paris S´er. A-B 283 (1976), no. 3, Aiii, A119–
A121.
[Au3] T. Aubin, Nonlinear analysis on manifolds. Monge-Amp`ere equations.
Grundlehren Math. Wiss. 252, Springer-Verlag, New York, 1982.
[Au4] T. Aubin, R´eduction du cas positif de l’´equation de Monge-Amp`ere sur
les vari´et´es k¨ahl´eriennes compactes ` a la d´emonstration d’une in´egalit´e. J.
Funct. Anal. 57 (1984), 143–153.
[BT] I. K. Babenko & I. A. Taimanov, On the formality problem for symplectic
manifolds. Global differential geometry, the mathematical legacy of Alfred
Gray (Bilbao, 2000), Contemp. Math. 288, Amer. Math. Soc., Providence,
R.I., 2001, 1–9.
[Bea] A. Beauville, Vari´et´es K¨ahleriennes dont la premi`ere classe de Chern est
nulle. J. Differential Geom. 18 (1983), 755–782 (1984).
[BeG] C. Benson & C. S. Gordon, K¨ahler and symplectic structures on nilmani-
folds. Topology 27 (1988), 513–518.
[BGM] M. Berger, P. Gauduchon, & E. Mazet, Le spectre d’une vari´et´e riemanni-
enne. Lecture Notes in Math. 194, Springer-Verlag, Berlin–New York, 1971.
[Bes] A. L. Besse, Einstein manifolds. Ergeb. Math. Grenzgeb. 10, Springer-
Verlag, Berlin, 1987.
[Bi] O. Biquard, M´etriques k¨ahl´eriennes ` a courbure scalaire constante.
S´eminaire Bourbaki, Expos´e No. 938, to be published in Ast´erisque.
[Boc] S. Bochner, Vector fields and Ricci curvature. Bull. Amer. Math. Soc. 52
(1946), 776–797.
[BM] S. Bochner & D. Montgomery, Groups on analytic manifolds. Ann. of Math.
48 (1947), 659–669.
Literature 171
[Bor] A. Borel, Semisimple groups and Riemannian symmetric spaces. Texts
Read. Math. 16, Hindustan Book Agency, New Delhi, 1998.
[Bo1] J. P. Bourguignon et al., Premi`ere classe de Chern et courbure de Ricci:
preuve de la conjecture de Calabi. Lectures presented at a S´eminaire held at
the Centre de Math´ematiques de l’
´
Ecole Polytechnique, Palaiseau, March–
June 1978, Ast´erisque 58, Soci´et´e Math´ematique de France, Paris, 1978,
169 pp.
[Bo2] J. P. Bourguignon, Eugenio Calabi and K¨ahler metrics. Manifolds and ge-
ometry (Pisa, 1993), Sympos. Math. XXXVI, Cambridge University Press,
Cambridge, 1996, 61–85.
[Bo3] J. P. Bourguignon, M´etriques d’Einstein-K¨ahler sur les vari´et´es de Fano:
obstructions et existence (d’apr`es Y. Matsushima, A. Futaki, S. T. Yau,
A. Nadel et G. Tian). S´eminaire Bourbaki, Expos´e No. 830; Ast´erisque 245
(1997), 277–305.
[Br] R. Bryant, Recent advances in the theory of holonomy. S´eminaire Bourbaki,
Expos´e No. 861; Ast´erisque 266 (2000), 351–374.
[Bu] D. Bump, Lie groups. Grad. Texts in Math. 225, Springer-Verlag, New York,
2004.
[BIW] M. Burger, A. Iozzi & A. Wienhard, Hermitian symmetric spaces and
K¨ahler rigidity. Preprint, ETH Z¨ urich, 2005.
[Ca1] E. Calabi, The space of K¨ahler metrics. Proceedings of the International
Congress of Mathematicians (Amsterdam, 1954), Vol. 2, Erven P. Noordhoff
N.V., Groningen; North-Holland Publishing Co., Amsterdam 1957, 206–
207.
[Ca2] E. Calabi, On K¨ahler manifolds with vanishing canonical class. Algebraic
geometry and topology, A symposium in honor of S. Lefschetz, Princeton
University Press, Princeton, N.J., 1957, 78–89.
[Ca3] E. Calabi, Improper affine hyperspheres of convex type and a generalization
of a theorem by K. J¨orgens. Michigan Math. J. 5 (1958), 105–126.
[CE] E. Calabi & B. Eckmann, A class of compact, complex manifolds which are
not algebraic. Ann. of Math. 58 (1953), 494–500.
[CX] J. Cao & F. Xavier, K¨ahler parabolicity and the Euler number of compact
manifolds of non-positive sectional curvature. Math. Ann. 319 (2001), 483–
491.
[CMP] J. Carlson, S. M¨ uller-Stach & C. Peters, Period mappings and period do-
mains. Cambridge Stud. Adv. Math. 85, Cambridge University Press, Cam-
bridge, 2003.
[Car] G. Carron, Formes harmoniques L
2
sur les vari´et´es non-compactes. Rend.
Mat. Appl. (7) 21 (2001), 87–119.
[Che] J. Cheeger, Some examples of manifolds of nonnegative curvature. J. Dif-
ferential Geometry 8 (1973), 623–628.
172 Literature
[CEb] J. Cheeger & D. G. Ebin, Comparison theorems in Riemannian geome-
try. North-Holland Mathematical Library 9, North-Holland Publishing Co.,
Amsterdam–Oxford; American Elsevier Publishing Co., Inc., New York,
1975.
[CG1] J. Cheeger & D. Gromoll, The splitting theorem for manifolds of nonnega-
tive Ricci curvature. J. Differential Geometry 6 (1971/72), 119–128.
[CG2] J. Cheeger & D. Gromoll, On the structure of complete manifolds of non-
negative curvature. Ann. of Math. 96 (1972), 413–443.
[CGr] J. Cheeger & M. Gromov, Collapsing Riemannian manifolds while keeping
their curvature bounded. I. J. Differential Geom. 23 (1986), 309–346.
[Chr] S.-S. Chern, On curvature and characteristic classes of a Riemann manifold.
Abh. Math. Semin. Univ. Hamb. 20 (1955), 117–126 .
[Chf] P. R. Chernoff, Essential self-adjointness of powers of generators of hyper-
bolic equations. J. Functional Analysis 12 (1973), 401–414.
[De] O. Debarre, Vari´et´es de Fano. S´eminaire Bourbaki, Expos´e No. 827;
Ast´erisque 245 (1997), 197–221.
[DGMS] P. Deligne, P. Griffiths, J. Morgan & D. Sullivan, Real homotopy theory of
K¨ahler manifolds. Invent. Math. 29 (1975), 245–274.
[Du] J. J. Duistermaat, Fourier integral operators. Progr. Math. 130, Birkh¨ auser,
Boston, MA, 1996.
[DK] J. J. Duistermaat and J. A. C. Kolk, Lie Groups. Universitext, Springer-
Verlag, Berlin, 2000.
[Eb] P. Eberlein, Geometry of nonpositively curved manifolds. Chicago Lectures
in Mathematics, University of Chicago Press, Chicago, IL, 1996.
[Eh] P. Ehrlich, Metric deformations of curvature. I. Local convex deformations.
Geom. Dedicata 5 (1976), 1–23.
[FG] M. Fern´andez & A. Gray, The Iwasawa manifold. Differential geometry
(Pe˜ n´ıscola, 1985), Lecture Notes in Math. 1209, Springer-Verlag, Berlin,
1986, 157–159.
[FH] W. Fulton & J. Harris, Representation theory. A first course. Grad. Texts
in Math. 129, Springer-Verlag, New York, 1991.
[Fu] A. Futaki, An obstruction to the existence of K¨ahler-Einstein metrics. In-
vent. Math. 73 (1983), 437–443.
[GM] S. Gallot & D. Meyer, Op´erateur de courbure et laplacien des formes
diff´erentielles d’une vari´et´e riemannienne. J. Math. Pures Appl. 54 (1975),
259–284.
[GH] P. Griffiths & J. Harris, Principles of algebraic geometry. Pure Appl. Math.,
Wiley-Interscience, New York, 1978.
[Gr] M. Gromov, K¨ahler hyperbolicity and L
2
-Hodge theory. J. Differential
Geom. 33 (1991), 263–292.
Literature 173
[GL] M. Gromov & H.B. Lawson Jr., Positive scalar curvature and the Dirac
operator on complete Riemannian manifolds. Inst. Hautes
´
Etudes Sci. Publ.
Math. 58 (1983), 83–196.
[Ha] K. Hasegawa, Minimal models of nilmanifolds. Proc. Amer. Math. Soc. 106
(1989), 65–71.
[Hei] H.-J. Hein, The Ricci flow on K¨ ahler manifolds. Diplomarbeit, Rheinische
Friedrich-Wilhelms-Universit¨at Bonn, Bonn, 2006.
[Hel] S. Helgason, Differential Geometry and Symmetric Spaces. Pure Appl.
Math. XII, Academic Press, New York–London, 1962.
[Hir] F. Hirzebruch, Topological methods in algebraic geometry. Appendix one by
R. L. E. Schwarzenberger, appendix two by A. Borel, reprint of the 2nd,
corr. print. of the 3rd ed. 1978, Classics Math., Springer-Verlag, Berlin,
1995.
[Hit] N. Hitchin, On the curvature of rational surfaces. Differential geometry
(Stanford Univ., Stanford, Calif., 1973), Part 2, Proc. Sympos. Pure Math.
XXVII, Amer. Math. Soc., Providence, R.I., 1975, 65–80.
[Hu] D. Huybrechts, Complex geometry. An introduction. Universitext, Springer-
Verlag, Berlin, 2005.
[JZ] J. Jost & K. Zuo, Vanishing theorems for L
2
-cohomology on infinite cover-
ings of compact K¨ahler manifolds and applications in algebraic geometry.
Comm. Anal. Geom. 8 (2000), 1–30.
[Jo] D. D. Joyce, Compact manifolds with special holonomy. Oxford Math.
Monographs, Oxford University Press, Oxford, 2000.
[K¨a1] E. K¨ahler,
¨
Uber eine bemerkenswerte Hermitesche Metrik. Abh. Math.
Semin. Hamb. Univ. 9 (1932), 173-186.
[K¨a2] E. K¨ahler, Mathematische Werke. Mathematical works. Edited by Rolf
Berndt and Oswald Riemenschneider, Walter de Gruyter, Berlin, 2003.
[Kar] H. Karcher, Riemannian comparison constructions. Global differential ge-
ometry, MAA Stud. Math. 27, Math. Assoc. America, Washington, DC,
1989, 170–222.
[Kat] T. Kato, Perturbation theory for linear operators. Reprint of the 1980 edi-
tion, Classics Math., Springer-Verlag, Berlin, 1995.
[Ka1] J. Kazdan, A remark on the preceding paper of S. T. Yau, “On the Ricci
curvature of a compact K¨ ahler manifold and the complex Monge-Amp`ere
equation. I”. Comm. Pure Appl. Math. 31 (1978), 413–414.
[Ka2] J. Kazdan, Applications of Partial Differential Equations to Problems in
Geometry. Lecture Notes, University of Pennsylvania, Philadelphia, Pa,
2004.
[Ko1] S. Kobayashi, On compact K¨ahler manifolds with positive definite Ricci
tensor. Ann. of Math. 74 (1961), 570–574.
[Ko2] S. Kobayashi, Differential geometry of complex vector bundles. Publ. Math.
Soc. Japan 15, Kanˆo Memorial Lectures 5, Princeton University Press,
Princeton, N.J.; Iwanami Shoten, Tokyo, 1987.
174 Literature
[KN] S. Kobayashi & K. Nomizu, Foundations of differential geometry. Vol. II,
reprint of the 1963 original, Wiley Classics Lib., John Wiley & Sons, Inc.,
New York, 1996.
[KW] A. Kor´anyi & J. A. Wolf, Realization of hermitian symmetric spaces as
generalized half-planes. Ann. of Math. 81 (1965), 265–288.
[LM] H. B. Lawson & M.-L. Michelsohn, Spin geometry. Princeton Math. Ser.
38, Princeton University Press, Princeton, N.J., 1989.
[Li1] A. Lichnerowicz, Transformation infinit´esimales conformes de certaines
vari´et´es riemanniennes compactes. C. R. Acad. Sci. Paris 241 (1955), 726–
729.
[Li2] A. Lichnerowicz, Sur les transformations analytiques des vari´et´es
k¨ahl´eriennes compactes. C. R. Acad. Sci. Paris 244 (1957), 3011–3013.
[Li3] A. Lichnerowicz, G´eom´etrie des groupes de transformations. Travaux et
Recherches Math´ematiques III, Dunod, Paris, 1958.
[Loh] J. Lohkamp, Curvature h-principles. Ann. of Math. 142 (1995), 457–498.
[Lot] J. Lott, L
2
-cohomology of geometrically infinite hyperbolic 3-manifolds.
Geom. Funct. Anal. 7 (1997), 81–119.
[Ma1] Y. Matsushima, Sur la structure du groupe d’hom´eomorphismes analytiques
d’une certaine vari´et´e k¨ahl´erienne. Nagoya Math. J. 11 (1957), 145–150.
[Ma2] Y. Matsushima, On tube domains. Symmetric spaces (Short Courses, Wash-
ington Univ., St. Louis, Mo., 1969–1970), Pure and Appl. Math. 8, Dekker,
New York, 1972, 255–270.
[MS] J. Milnor & J. Stasheff, Characteristic classes. Ann. of Math. Stud. 76,
Princeton University Press, Princeton, N.J.; University of Tokyo Press,
Tokyo, 1974.
[Mo] A. Moroianu, K¨ahler geometry. Lecture Notes 2004.
[Na] A. M. Nadel, Multiplier ideal sheaves and K¨ahler-Einstein metrics of posi-
tive scalar curvature. Ann. of Math. 132 (1990), 549–596.
[NN] A. Newlander & L. Nirenberg, Complex analytic coordinates in almost com-
plex manifolds. Ann. of Math. 65 (1957), 391–404.
[Ob] M. Obata, Certain conditions for a Riemannian manifold to be isometric
with a sphere. J. Math. Soc. Japan 14 (1962), 333–340.
[PP] D. N. Page & C. N. Pope, Inhomogeneous Einstein metrics on complex line
bundles. Classical Quantum Grav. 4 (1987), 213–225.
[Ro] J. Roe, Elliptic operators, topology and asymptotic methods. Pitman Res.
Notes Math. Ser. 179, Longman S&T, Harlow, 1988.
[Sa] T. Sakai, Riemannian geometry. Translated from the 1992 Japanese original
by the author. Transl. Math. Monogr. 149, Amer. Math. Soc., Providence,
R.I., 1996.
[ShS] B. Shiffman & A. Sommese, Vanishing theorems on complex manifolds.
Progr. Math. 56, Birkh¨auser, Boston, MA, 1985.
Literature 175
[Si1] Y. T. Siu, Lectures on Hermitian-Einstein metrics for stable bundles and
K¨ahler-Einstein metrics. DMV Seminar 8, Birkh¨auser Verlag, Basel, 1987.
[Si2] Y. T. Siu, The existence of K¨ahler-Einstein metrics on manifolds with pos-
itive anticanonical line bundle and a suitable finite symmetry group. Ann.
of Math. 127 (1988), 585–627.
[Th] R. P. Thomas, Notes on GIT and symplectic reduction for bundles and
varieties. Preprint, arXiv:math.AG/0512411.
[Ti1] G. Tian, On K¨ahler-Einstein metrics on certain K¨ahler manifolds with
C1(M) > 0. Invent. Math. 89 (1987), 225–246.
[Ti2] G. Tian, On Calabi’s conjecture for complex surfaces with positive first
Chern class. Invent. Math. 101 (1990), 101–172.
[Ti3] G. Tian, Canonical metrics in K¨ ahler geometry. Notes taken by Meike
Akveld, Lectures Math. ETH Z¨ urich, Birkh¨auser Verlag, Basel, 2000.
[TY] G. Tian & S. T. Yau, K¨ahler-Einstein metrics on complex surfaces with
C1 > 0. Comm. Math. Phys. 112 (1987), 175–203.
[TO] A. Tralle & J. Oprea, Symplectic manifolds with no K¨ahler structure. Lec-
ture Notes in Math. 1661, Springer-Verlag, Berlin, 1997.
[Va] R. Vakil, Complex algebraic surfaces. Lecture Notes, Stanford University,
2002.
[Wa] F. W. Warner, Foundations of differentiable manifolds and Lie groups.
Scott, Foresman and Co., Glenview, Ill.-London, 1971.
[Wei] A. Weil, Introduction `a l’´etude des vari´et´es k¨ahl´eriennes. Publications de
l’Institut de Math. de l’Universit´e de Nancago, VI. Actualit´es Sci. Ind. no.
1267, Hermann, Paris 1958.
[Wel] R. O. Wells, Differential analysis on complex manifolds. Prentice-Hall Series
in Modern Analysis, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1973.
[Wi] A. K. Wienhard, Bounded cohomology and geometry. Dissertation, Rheinis-
che Friedrich-Wilhelms-Universit¨at Bonn, Bonn 2004, Bonner Mathematis-
che Schriften 368, Universit¨at Bonn, Mathematisches Institut, Bonn, 2004.
[Wo] J. A. Wolf, Spaces of constant curvature. Fourth edition, Publish or Perish,
Inc., Houston, TX, 1977.
[Ya] K. Yano, On harmonic and Killing vector fields. Ann. of Math. 55 (1952),
38–45.
[Ya1] S. T. Yau, On the curvature of compact Hermitian manifolds. Invent. Math.
25 (1974), 213–239.
[Ya2] S. T. Yau, Calabi’s conjecture and some new results in algebraic geometry.
Proc. Nat. Acad. Sci. U.S.A. 74 (1977), 1798–1799.
[Ya3] S. T. Yau, On the Ricci curvature of a compact K¨ ahler manifold and the
complex Monge-Amp`ere equation. I. Comm. Pure Appl. Math. 31 (1978),
339–411.
[Zh] F. Zheng, Complex differential geometry. AMS/IP Stud. Adv. Math. 18,
Amer. Math. Soc., Providence, R.I.; International Press, Boston, MA, 2000.
Index
almost complex structure, 17, 18, 22,
49
ample, 118
arithmetic genus, 22, 85
automorphic, 20, 59, 60, 82, 100
automorphism, 11, 20, 28, 82, 99, 101,
104
Bergmann metric, 46, 47, 160
Betti number, 7, 68
bisectional curvature, 55
blow up, 25, 28, 39, 101, 123
Calabi conjecture, 83, 85, 88
Calabi–Futaki invariant, 104
Calabi–Yau manifold, 58, 83
canonical bundle, 21, 39, 40, 57, 78
Chern
character, 134
class, 85, 88, 89, 91, 99, 130, 131
connection, 31–33, 35, 48, 79, 108
form, 57, 130, 131
Clifford
multiplication, 165
relation, 3
collapse, 24
compatible, 23, 42
complex
curve, 11
differential, 73
Hessian, 104
hyperbolic space, 45, 157
Lie group, 15
manifold, 11
projective space, 12, 43, 117, 123,
157
quadric, 158
structure, 11, 17, 160
surface, 27, 89
torus, 15
vector field, 18, 19
curvature operator, 8, 76, 146
dd
c
-Lemma, 74, 79
de Rham decomposition, 58, 147
Dirac
bundle, 114, 165
operator, 114, 165
Dolbeault
cohomology, 22, 29
operator, 116
dual, 148, 151, 160
effective, 145, 159
elliptic, 163
essentially self-adjoint, 164
Euler
class, 135
form, 135
exterior derivative, 4
Fano
manifold, 88, 157
surface, 104
formal, 76
adjoint, 5
formal adjoint, 162
Fr¨ olicher relations, 69
Fubini–Study metric, 43, 100
fundamental matrix, 42
Futaki invariant, 102
Grassmann manifold, 13, 43, 133,
149–157
harmonic, 6, 67, 108, 167, 168
Heisenberg group, 69, 77
Hermitian
manifold, 23
symmetric space, 108, 156–161
INDEX 177
Hodge
decomposition, 7, 67, 68, 167–169
diamond, 69
index theorem, 71
manifold, 117
number, 22, 29
operator, 9
Hodge–Riemann bilinear relations,
70–72
holomorphic, 11
form, 22, 84
line bundle, 37, 79–80, 118
sectional curvature, 54
vector bundle, 17, 29, 31, 62, 108
holonomy, 57, 147
Hopf manifold, 15, 50, 69
hyper-K¨ ahler manifold, 58
hyperplane bundle, 39, 79, 117
invariant, 125
Iwasawa manifold, 70, 77
K¨ ahler
form, 42, 49
manifold, 42, 57
metric, 42, 48
potential, 48
K¨ ahler–Einstein
manifold, 87, 100, 101
metric, 91, 99, 102
K¨ ahler-hyperbolic, 107
Killing
field, 2, 59, 60, 82, 100
form, 145
Kodaira
embedding theorem, 118
vanishing theorem, 80
Kummer surface, 26
Lagrangian subspace, 152
Laplace operator, 6, 36, 65–69, 75, 86
Lefschetz
decomposition, 64, 67, 68, 169
map, 62, 65
maximal extension, 163
minimal, 51
extension, 164
Monge–Amp`ere equation, 90, 99
Nijenhuis tensor, 17, 19
normal coordinates, 48
period map, 72
Picard group, 38
positive, 78
primitive, 63, 71
principal symbol, 163
reductive, 100, 101
Ricci
equation, 87
form, 56
Ricci-flat, 57, 58, 83, 88
Riemann sphere, 12
Serre duality, 30, 68, 80, 169
signature, 70, 71
symmetric
pair, 107, 141
space, 137
symplectic
form, 50, 151
manifold, 50, 69
vector space, 151
tautological bundle, 14, 33, 38, 78, 117,
133
transgression, 128
transvection, 139
tube domain, 160
type
of differential form, 21, 29
of symmetric space, 140
universal bundle, 14
wedge product, 3
Weil homomorphism, 129
Weitzenb¨ ock formula, 6, 7, 36
Wirtinger inequality, 51

To my wife Helga

Preface
These are notes of lectures on K¨hler manifolds which I taught at the Univera sity of Bonn and, in reduced form, at the Erwin-Schr¨dinger Institute in Vio enna. Besides giving a thorough introduction into K¨hler geometry, my main a aims were cohomology of K¨hler manifolds, formality of K¨hler manifolds afa a ter [DGMS], Calabi conjecture and some of its consequences, Gromov’s K¨hler a hyperbolicity [Gr], and the Kodaira embedding theorem. Let M be a complex manifold. A Riemannian metric on M is called Hermitian if it is compatible with the complex structure J of M , JX, JY = X, Y . Then the associated differential two-form ω defined by ω(X, Y ) = JX, Y is called the K¨hler form. It turns out that ω is closed if and only if J is a parallel. Then M is called a K¨hler manifold and the metric on M a K¨hler a a metric. K¨hler manifolds are modelled on complex Euclidean space. Except a for the latter, the main example is complex projective space endowed with the Fubini–Study metric. Let N be a complex submanifold of a K¨hler manifold M . Since the rea striction of the Riemannian metric of M to N is Hermitian and its K¨hler a form is the restriction of the K¨hler form of M to N , N together with the a induced Riemannian metric is a K¨hler manifold as well. In particular, smooth a complex projective varieties together with the Riemannian metric induced by the Fubini–Study metric are K¨hlerian. This explains the close connection of a K¨hler geometry with complex algebraic geometry. a I concentrate on the differential geometric side of K¨hler geometry, except a for a few remarks I do not say much about complex analysis and complex algebraic geometry. The contents of the notes is quite clear from the table below. Nevertheless, a few words seem to be in order. These concern mainly the prerequisites. I assume that the reader is familiar with basic concepts from differential geometry like vector bundles and connections, Riemannian and Hermitian metrics, curvature and holonomy. In analysis I assume the basic facts from the theory of elliptic partial differential operators, in particular regularity and Hodge theory. Good references for this are for example [LM, Section III.5] and [Wa, Chapter 6]. In Chapter 8, I discuss Gromov’s K¨hler a hyperbolic spaces. Following the arguments in [Gr], the proof of the main result of this chapter is based on a somewhat generalized version of Atiyah’s L2 -index theorem; for the version needed here, the best reference seems to be Chapter 13 in [Ro]. In Chapter 7, I discuss the proof of the Calabi conjecture. Without further reference I use H¨lder spaces and Sobolev embedding theorems. This is o standard material, and many textbooks on analysis provide these prerequisites.

2] or [Wa. Friedrich Hirzebruch. Gregor Weingart. For this. and Anna Wienhard for their careful proofreading of the manuscript. Ursula Hamenst¨dt. Acknowledgments. and two anonymous referees for very helpful discussions and remarks about various topics of these notes. the Erwin-Schr¨dinger-Institute in u o Vienna. and the Forschungsinstitut f¨r Mathematik at the ETH in Z¨rich. I include appendices on characteristic classes. I would like to acknowledge the hospitality of the MaxPlanck-Institut f¨r Mathematik in Bonn. 6. I am grateful to Dmitri Anosov. Marc Burger. Werner M¨ller. Matthias Lesch. I need a result from the regularity theory of non-linear partial differential equations.4 are taken from his Diplom thesis [Hei]. My special thanks go to HansJoachim Hein. for example. Subsections 4. in [Hir. Daniel Roggenkamp. Section 1.iv Preface In addition.3. my ideas about the development of the field are still too vague for an adequate historical discussion. Ingo Lieb. . Chapter 5]. For the convenience of the reader. I refer to the lecture notes by Kazdan [Ka2] where the reader finds the necessary statements together with precise references for their proofs. Jerry Kazdan. What I need is again standard and can be found. Hera mann Karcher. and differential operators. I use some basic sheaf theory in the proof of the Kodaira embedding theorem in Chapter 9. Finally I would like to thank Irene Zimmermann and Manfred Karbe from the EMS Publishing House for their cordial and effective cooperation.6. The reader may miss historical comments. u u who provided me with the opportunity to work undisturbed and undistracted on these lecture notes and other mathematical projects. Beno Eckmann. Although I spent quite some time on preparing my lectures and writing these notes. u Joachim Schwermer. Daniel Huybrechts. Thomas Delzant. and 7. I would like to thank Anna Pratoussevitch. Stefan Hildebrandt. who read many versions of the manuscript very carefully and suggested many substantial improvements. symmetric spaces.

. . . . . . . . . 4. . . . . . . . . a 4. .3 Laplace Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Existence . . . . . . . . . . . 4 5 Cohomology of K¨hler Manifolds . . . . . . . . . . .1 Ricci-Flat K¨hler Manifolds . . . . . . . . . . . . . . . . . . .4 Ricci Tensor . . . . . . . . . . . . .2 Nonnegative Ricci Curvature . .Contents 1 Preliminaries . . . . . . . . . . . . . . . . .3 Compatible Metrics .4 Hodge Operator . . . . .3 Ricci Curvature and Laplace Operator Calabi 7. . . . . .2 Differential Forms . . . . . . .1 K¨hler Form and Volume a 4. . . . . . . . . Uniqueness . . . . . . . . . . 6. . . . . . . . .1 Dolbeault Cohomology . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . 5. . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Killing Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K¨hler Manifolds . . . . . . . . . . . . . . . . . . . . . .2 Lefschetz Map and Cohomology 5. . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . .5 Holonomy . . Regularity . . . . . .4 Holomorphic Line Bundles . . . . . . . . . . . . . . . . . .3 7. . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . . . . . 3 Holomorphic Vector Bundles . . . . . . . . . . . . . . . . . . . . . .2 Levi-Civita Connection . . . . . . . . . . . . . .1 Differential Forms 1. . . . . . . . . .1 Lefschetz Map and Differentials . . . Obstructions . 1 2 4 6 9 11 18 21 23 24 29 29 31 35 37 42 49 52 53 55 57 59 62 64 67 73 78 82 83 84 86 88 92 93 94 99 2 Complex Manifolds . . . . . . . . . . . . . . 7 . .2 7. .2 Chern Connection . . 2. . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . .1 Complex Vector Fields 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Some Formulas .2 Exterior Derivative 1. . . . . . . . . . . . . . . . . . . . . 6 Ricci Curvature and Global Structure . . . . . . . . . . . . . .4 Conjecture . . . . . . . . .3 The ddc -Lemma and Formality . . . . .1 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Curvature Tensor . . . . 3. . .4 Blowing Up Points . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . 2. . 5. . . . . . . 3. . . a 6. 3. a 5. . . . . . . . . . . . . . . . . . . . . . . . . . .4 Some Vanishing Theorems . . . . . . . .

. . .2 Examples . . . . . . . . . . . . .2 Two Applications . . . . . . Appendix C Remarks on Differential Operators C. . A. . B. . . . . . . . . . . . . . . . .1 Dirac Operators . . . a 8. . . .2 Euler Class . . . . . . . . . . . . B. . . .3 Hermitian Symmetric Spaces . . C. . A. . . . . . . . . . . . . . . . . . . . . . . . .1 Chern Classes and Character . . . . . . . B. . 9. . . . . . . . . .1 Symmetric Pairs . . . . . . Index . . . . . 9. . . . . . .2 Non-Vanishing of Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .vi 8 Contents K¨hler Hyperbolic Spaces . . . . . .1 K¨hler Hyperbolicity and Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 L2 -de Rham Cohomology . . . . . . .3 L2 -Dolbeault Cohomology . . . . . . . . . . . . . . . . . . . . . . a 8. . . . . . . . . . . . . . . . . .1 Proof of the Embedding Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Literature . Kodaira Embedding Theorem . . . . . . . . . . . . . . C. . . . . . . . . . . Appendix B Symmetric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 108 112 117 119 123 124 130 135 137 141 148 156 162 165 167 168 170 176 9 Appendix A Chern–Weil Theory . . . . . . . . . . . . . . . .

E) the space of sections of E over U . More generally. By setting g(X)(Y ) := g(X. we denote by Ec (U. Hermitian metrics on E are denoted by parentheses (·. We use the standard musical notation for this isomorphism. manifolds and maps are assumed to be smooth. Let M be a manifold1 of dimension n. For a vector field X. we set notation and conventions and discuss some preliminaries. E). · be a Riemannian metric on M and be its Levi-Civita connection. E). E) ⊂ E(U. . then (σ.1) We say that a coordinate chart x is centered at a point p ∈ M if x(p) = 0. w ∈ T M and ϕ ∈ T ∗ M have the same foot points. As a rule we assume that Hermitian metrics on a given bundle E are conjugate linear in the first variable and complex linear in the second. we will not refer to the domain U of x. If E is a complex vector bundle over M and (·. for U ⊂ M open. The reason for using different symbols for Riemannian and Hermitian metrics is apparent from (1. It is obvious that (v ) = v and (ϕ ) = ϕ.2) is a Hermitian product on Ec (M. Let E → M be a vector bundle over M . · . (1. w and ϕ . U ). (1. Y ).12) below. We denote the space of sections of E by E(E) or E(M. E) be the completion of Ec (M. τ )2 = M (σ. E) the subspace of sections with compact support in U . Riemannian metrics on E are denoted by angle brackets ·.1 Preliminaries In this chapter. ·). we denote by E(U. if E is complex.3) where v. v (w) = v. The coordinate frame of a coordinate chart x consists of the ordered tuple of vector fields Xj = ∂ . ·) is a Hermitian metric on E. use similar terminology in the case of real vector bundles and Riemannian metrics. As a rule. E) with respect to the Hermitian norm induced by the Hermitian product in (1. Z) = 1 Unless 2 We Y X. We let L2 (E) = L2 (M. (1. ZX . Let g = ·.2) and identify L2 (M.4) specified otherwise. As long as there is no need to specify a name for them. The induced Hermitian metric on the dual bundle E ∗ will then be complex linear in the first and conjugate linear in the second variable. τ ) (1. ∂xj 1 ≤ j ≤ n. where U ⊂ M is open and x : U → Rn is a diffeomorphism onto its image. we may interpret g as an isomorphism T M → T ∗ M . E) with the space of equivalence classes of square-integrable measurable sections of E as usual2 . w = ϕ(w). It is given by (LX g)(Y. A coordinate chart for M is a tuple (x. Furthermore. Z + Y. the Lie derivative LX g of g measures how much g varies under the flow of X. Similarly.

LX Z). ). this equation reduces to the Jacobi equation. (1.10) . ψ = ϕ. Hint: For Y = Z. We say that X is a Killing field if the flow of X preserves g. The Riemannian metric on M induces a Riemannian metric on A∗ (M. 2) For all vector fields Y. respectively. R) and Ar (M. Z) + g(LX Y.7) (1. R) := Λ∗ (T ∗ M ) be the bundle of (multilinear) alternating forms on T M (with values in R) and A∗ (M. iff X is skew-symmetric. the Levi-Civita connection induces a connection ˆ on A∗ (M. The divergence div X of a vector field X on M measures the change of the volume element under the flow of X. R) := E(A∗ (M.5 Exercises. . . and ˆ is metric with respect to the induced metric on A∗ (M. equivalently. Recall the interior product of a tangent vector v with an alternating form ϕ with the same foot point. X Y.20 below. Z)) = (LX g)(Y. Z = LX (g(Y. ν . v ψ (1.1 Differential Forms. compare the product rule 1. 1. Hint: The flow of X consists of local isometries of M and gives rise to local geodesic variations of c with variation field X ◦ c. 1.8) tr LX g = 2 div X.4). 1. that is. We let A∗ (M. 1) For any geodesic c in M .9 Divergence Formula. X is a Killing field iff LX g = 0 or. R)) be the space of differential forms on M . . By definition. R). . Xn ) of T M . by the product rule for the Lie derivative. insert v as first variable. Let X be a Killing field on M .6) In terms of a local orthonormal frame (X1 . R). Let G be a compact domain in M with smooth boundary ∂G and exterior normal vector field ν along ∂G. v ϕ = ϕ(v. v ∧ ϕ. Z on M 2 X(Y. Then div X = G ∂G X. It is given by div X = tr X. R) be the subbundle of alternating forms of degree r and the subspace of differential forms on M of degree r. the composition X ◦ c is a Jacobi field along c. Z) + g(Y. We let Ar (M. we also have Xj X.2 ¨ Lectures on Kahler Manifolds In fact. Similarly. Y )Z = 0. There are the following remarkable relations between ∧ and . . Z) + R(X. we have div X = By (1. Xj . . R). . (1.

respectively. Xr+s ) 1 = ε(σ)µ{α(Xσ(1) . of ϕ and set ϕ = ρ−iτ . all with the same foot point. but now with complex tangent vectors v. We let A∗ (M. (1. E) and α ∈ A∗ (M. Locally. We define the wedge product of ϕ ∈ A∗ (M. More generally. E). an element of A∗ (M. We extend Riemannian metric. C). E) by ϕ ∧ α := (ϕ ∧ ϕj ) ⊗ σj . w and C-valued forms ϕ and ψ. E)). and τ the imaginary part. .12) (ϕ. then ϕk ⊗ σk . (1. Equations 1. E) := A∗ (M. Xσ(r) ) ⊗ α (Xσ(r+1) . . C) := E(A∗ (M. Such forms decompose. If α and α are of degree (α ∧µ α )(X1 . (1. . let E and E be further complex vector bundles over M and µ : E ⊗E → E be a morphism. . that is. E ) by α ∧µ α := j. ψ on A∗ (M. C) := A∗ (M. τ = Im ϕ. ρ = Re ϕ.1 Preliminaries 3 and the Clifford relation v ∧ (w ϕ) + w (v ∧ ϕ) = v. There is an induced Hermitian metric (1. C) with α ∈ A∗ (M. A differential form with values in E is a smooth section of A∗ (M. Xσ(r+s) )}. the complexified tangent bundle. w ϕ. E ). wedge product.11) where v and w are tangent vectors and ϕ and ψ alternating forms. C) and σ ∈ E(M. any such form is a linear combination of decomposable differential forms ϕ ⊗ σ with ϕ ∈ A∗ (M. . and interior product complex linearly in the involved variables to TC M and A∗ (M. . Via complex multilinear extension. We call ρ the real part. ˆ ϕ := ˆ Re ϕ + i ˆ Im ϕ.15) r!s! .k (ϕj ∧ ϕk ) ⊗ µ(σj ⊗ σk ) ∈ A(M. . (1. C). C) ⊗ E. we may view elements of A∗ (M.14) where we write α = ϕj ⊗ σj and α = r and s. respectively. R) ⊗R C be the bundle of R-multilinear alternating forms on T M with values in C and A∗ (M.10 and 1. C) as in c (1. We extend ˆ complex linearly to a connection on A∗ (M. Let E → M be a complex vector bundle. We define the wedge product of differential forms α ∈ A∗ (M.11 continue to hold. . ψ) := ϕ. where ρ and τ are differential forms with values in R as above. C) as Cmultilinear alternating forms on TC M = T M ⊗R C. ϕ = ρ + iτ . E) := E(A∗ (M. C) and a corresponding L2 -Hermitian product on A∗ (M. .13) where we decompose α = ϕj ⊗ σj . . C)) be the space of complex valued differential forms on M . .2). .

1. Let E be a complex vector bundle over M and D be a connection on E. Let α. . we define the exterior derivative of α (with respect to D) by (dϕj ⊗ σj + (−1)r ϕj ∧ Dσj ). . Since dD (f α) − f dD α = df ∧ α for any function f on M . A∗ (M. X0 . 1. if α and α have degree r and s. . (1.14) do not depend on the way in which we write α and α as sums of decomposable differential forms. E). . . .13) and (1. . . Then. For a differential form α with values in E and of degree r.4 ¨ Lectures on Kahler Manifolds This formula shows that the wedge products in (1. . . . That this is independent of the decomposition of α into a sum of decomposable differential forms follows from dD α(X0 . Xk ]. .17) dD α := where we decompose α = ϕj ⊗ σj and where d denotes the usual exterior derivative.16 Exercise. Xr )) + j<k (1.20) . Xj . .19 Proposition. . ˆ Via the product rule. 1. . E) of dD is given by σ(ξ ⊗ α) = ξ ∧ α. (1. . the principal symbol σ : T ∗ M ⊗ A∗ (M. . . α . E) inherits a connection D from ˆ and D. Let E be a complex vector bundle over M and µ and λ be composition and Lie bracket in the associated vector bundle End E of endomorphisms of E. and α be differential forms with values in End E. .2 Exterior Derivative. . Xr ) = j ˆ (−1)j DXj (α(X0 . dD dD α = RD ∧ε α. respectively. for all α ∈ A∗ (M. . Let E be a vector bundle over M and D be a connection on E. Xr ). Then (α ∧µ α ) ∧µ α = α ∧µ (α ∧µ α ) and α ∧λ α = α ∧µ α − (−1)rs α ∧µ α. ˆ DX (ϕ ⊗ σ) := ( ˆ X ϕ) ⊗ σ + ϕ ⊗ (DX σ). Xk . . . where we view the curvature tensor RD of D as a two-form with values in the bundle End E of endomorphisms of E and where the wedge product is taken with respect to the evaluation map ε : End E ⊗ E → E. E) → A∗ (M.18) (−1) j+k ˆ ˆ α([Xj . . Xj . We refer to the beginning of Appendix C for some of the terminology in this and the next subsection.

see (C. . (1.1 Preliminaries 5 ˆ In terms of D. E) have the same foot point. 1. respectively. Xj β) − ˆ (α. β ∈ A∗ (M. D and D be connections on E. Let D.27 Proposition. w α. Let E.24) DX (µ(σ ⊗ τ )) = µ((DX σ) ⊗ τ ) + µ(σ ⊗ (DX τ )) (1. we have dD α = ∗ ˆ Xj ∧ DXj α. c (v ∧ α. E .2). (1. such that the product rule holds for all vector fields X of M .23) (1. W β). The induced connections on the bundles of forms then also satisfy the corresponding product rule. on decomposable forms given by and a corresponding L2 -Hermitian product on A∗ (M. . . E). Let h = (·. 1. (dD α. E). For the exterior differential we get ˆ ˆ ˆ DX (α ∧µ α ) = (DX α) ∧µ α + (α ∧µ (DX α )).22) where we assume that α is of degree r. . E . and E . Then h induces a Hermitian metric on A∗ (M. E) as in (1.3). β) = (α. . W ) = (α. Proof. Xj β) The first term on the right is equal to the divergence of the complex vector field Z defined by (Z. τ ). . ψ ⊗ τ ) = ϕ. .25) dD (α ∧µ α ) = (dD α) ∧µ α + (−1)r α ∧µ (dD α ).11) hold on A∗ (M. . (1. Then the induced connection D on A (M. We emphasize that the curvature tensors of ˆ and D act as derivations with respect to wedge and tensor products. v β) (ϕ ⊗ σ. . Xj (α. . Xn ) is the correspondˆ ing dual frame. Let p ∈ M and choose a local orthonormal frame (Xj ) around p with Xj (p) = 0. w ∈ T M and α. where v. β) = = = ∗ ˆ (Xj ∧ DXj α. . The analogs of (1. Then. the formal adjoint (dD )∗ of dD is given by (dD )∗ α = − ˆ Xj DXj α. Xj DXj β). ∗ and v ∧ (w α) + w (v ∧ α) = v.14). In terms of a local orthonormal frame (X1 . E and µ : E ⊗ E → E be as in (1.10) and (1. ψ (σ. E) as in (1. . ·) be a Hermitian metric on E. Xn ) of M .21) ∗ ∗ where (X1 . ˆ Let D be a Hermitian connection on E.26 Exercise.20) is Hermitian as well. . at p. Let α and β be differential forms of degree r − 1 and r. Xn ) is a local frame of T M and (X1 . β) ˆ (DXj α.

21 and 1. Xn ) is the ∗ ˆ D denotes the curvature tensor of the corresponding dual frame of T M . Xk )α).9) to show that (∆dD α. As above. (X1 . see Proposition 1. In particular. β)2 = (dD α.31) ∗ ∗ where (X1 . (1. the principal symbol σ : T ∗ M ⊗ A∗ (M. We say that a differential form α with values in E is harmonic if dD α = (dD )∗ α = 0 and denote by H∗ (M. 1.3 Laplace Operator. Compare also Corollary C. and R ˆ connection D on A∗ (M.11). . . The Laplace operator associated to dD is ∆dD = dD (dD )∗ + (dD )∗ dD . . The latter . . (1. E).30 Exercise. Xj ) + j j=k ∗ ˆ Xk ∧ (Xj RD (Xj . we can rewrite (1. . E) of (dD )∗ is given by σ(ξ ⊗ α) = −ξ α. we let E → M be a complex vector bundle with Hermitian metric h and Hermitian connection D.31) by Kα and using (C.8). E) can be written as a sum α = ϕj ⊗ Φj and dD α = (dϕj ) ⊗ Φj and (dD )∗ α = (d∗ ϕj ) ⊗ Φj .28) Since the principal symbol of a composition of differential operators is the composition of their principal symbols.19. (1. (dD )∗ β)2 .11 and the formulas 1. then over U . dD β)2 + ((dD )∗ α. Use the divergence formula (1. Assume that M is closed. (1.32) Assume now that D is flat. ∆dD is an elliptic differential operator.6 ¨ Lectures on Kahler Manifolds Since (dD )∗ (f α) − f (dD )∗ α = − grad f α for any function f on M . Then we also have d2 = (d∗ )2 = 0. Denoting the second term on the right hand side of (1.31) in two ways. 1.33) For that reason. If (Φj ) is a parallel frame of E over an open subset U of M . . a straightforward calculation gives the Weitzenb¨ck formula D ∗ ∆dD α = − ˆ D2 α(Xj .29) by (1. E) → A∗ (M. . . ˆ ˆ ˆ ∆dD α = − tr D2 α + Kα = D∗ Dα + Kα. α ∈ A(M. we often use the shorthand dα and d∗ α for dD α and (dD )∗ α if D is flat. (1.22. Conclude that ϕ is harmonic iff ∆dD α = 0.27 for dD and o (d ) . Xn ) is a local orthonormal frame of T M . the principal symbol σ of ∆dD is given by σ(ξ ⊗ α) = −(ξ ∧ (ξ α) + ξ (ξ ∧ α)) = −||ξ||2 α. E) the space of harmonic differential forms with values in E. Using the Clifford relation 1.

E) + d(A∗ (M. the wedge product of harmonic differential forms is not a harmonic differential form anymore. 1. As a rule.32) is given by Kϕ = (Ric ϕ ) . In fact. the curvature term K in the Weitzenb¨ck formula (1.35 Theorem (Hodge Decomposition). In particular. However. The most important case is E = C. If M is closed and D is flat. 1. The equation in Exercise 1. In particular. see for example Section III. where the sum is orthogonal with respect to the L2 -Hermitian product on A∗ (M. It is somewhat tempting to assume that the ring structure of H ∗ (M.36) H∗ (M.38 Exercise. Show that for ϕ ∈ A1 (M. E)).5 in [LM] or Chapter 6 in [Wa]: 1. In the following theorem we give his ingenious application of it. then the canonical map to cohomology.38 was observed by Bochner (see also [Ya]). In the above discussion. C).34) ∆d = (d + d∗ )2 . E) → H ∗ (M. . C) is also represented by H∗ (M. E)) + d∗ (A∗ (M. then A∗ (M. Let M be a closed and connected Riemannian manifold with non-negative Ricci curvature. which we will use in some instances. E). R) = dimC H j (M. this only happens in rare cases.1 Preliminaries 7 implies that the images of d and d∗ are L2 -perpendicular and that the Laplace operator is a square. ∆d ϕ = ˆ ∗ ˆ ϕ + (Ric ϕ ) . E). C). (1. in addition. bj (M ) = dimR H j (M. (1. dim H ∗ (M. There is a corresponding theory in the real case. Then b1 (M ) ≤ n with equality if and onlyif M is a flat torus. each de Rham cohomology class of M with coefficients in E contains precisely one harmonic representative. In other words.39 Theorem (Bochner [Boc]). the Ricci curvature of M is positive in some point of M . we only considered complex vector bundles.25. E) < ∞. if M is closed and D is flat. it is a specific property of the K¨hler form of a K¨hler manifold that its wedge product with a harmonic form a a gives a harmonic form. Let bj (M ) be the j-th Betti number of M . 1.42 below.37 Remark. then b1 (M ) = 0. C). The fundamental estimates and regularity theory for elliptic differential operators lead to the Hodge decomposition of A∗ (M. If. Bochner’s argument can be used in many other situations and is therefore named after him. In other words. where Ric denotes the o Ricci tensor of M . E) = H∗ (M. is an isomorphism of vector spaces. Compare also Remark 1. E). see Theorem 5.

(1. The corresponding connected subgroup of the isometry group of M is closed and Abelian and its orbits foliate M by parallel flat tori. then the Riemannian metric of M can be deformed to a Riemannian metric of positive Ricci curvature [Au1] (see also [Eh]). hence a harmonic form. The curvature operator R is the symmetric endomorphism on Λ2 (T M ) defined by the equation ˆ R(X ∧ Y ). Equality for any such r implies that M is a flat torus. If ϕ is such a differential form.39. see [GM] or (the proof of) Theorem 8. respectively.39 that a closed.6 in [LM]. . then a parallel differential form on M is determined by its value at any given point of M and hence br (M ) ≤ n r for 0 < r < n. In particular. connected Riemannian manifold M with non-negative Ricci curvature is foliated by a parallel family of totally geodesic flat tori of dimension b1 (M ). It follows that ϕ is parallel.38. 1.42 Remark (and Exercise).43) Gallot and Meyer showed that the curvature term in the Weitzenb¨ck formula o ˆ ˆ (1. 1. by Hodge theory as in (1. U . Conclude from the argument in the proof of Theorem 1. then 0= M ∆d ϕ. If M is closed with ˆ R ≥ 0.41 Exercise. by Exercise 1. by the (identical) version of Theorem 1. the integrand in the second integral on the right is non-negative. 3) The complete analysis of the fundamental groups of closed and connected Riemannian manifolds with non-negative Ricci curvature was achieved by Cheeger and Gromoll [CG1]. U ∧ V := R(X. ϕ = M || ˆ ϕ||2 + Ric ϕ .1. ˆ if M is closed with R > 0. Since the wedge product of parallel differential forms is a parallel differential form. If M is also connected.36) and the Bochner argument in Theorem 1. R) is positive or non-negative if R > 0 or R ≥ 0.8 ¨ Lectures on Kahler Manifolds Proof. If the Ricci curvature of M is positive in some point of M . again by the Bochner argument. By assumption.32) for ∆d on A∗ (M.35 for real vector bundles. Compare Subsection 6. Hint: The space p of parallel vector fields on M is an Abelian subalgebra of the Lie algebra of Killing fields on M . Y )V.37. we can represent real cohomology classes of dimension one uniquely by harmonic one-forms. this is one of the rare instances where the wedge product of harmonic forms is harmonic (albeit for a trivial reason). connected Riemannian manifold with positive Ricci curvature is finite. 1) The earlier theorem of Bonnet–Myers makes the stronger assertion that the fundamental group of a closed. compare Remark 1. [CG2]. then br (M ) = 0 for 0 < r < n. and that Ric ϕ = 0. The rest of the argument is left as an exercise. therefore also the vector field ϕ . ˆ 1. 2) Let M be a closed and connected Riemannian manifold with non-negative Ricci curvature. ϕ . Since M is closed.40 Remarks. then real valued harmonic forms on M are parallel.

51) that the definition here differs from the standard one. Via h(σ)(τ ) = (σ.30 and Equation 1. Hence. E ∗ ). Suppose now that M is oriented. (1. R) → Hn−r (M.1 Preliminaries 9 By definition.50) (1.48) If M is closed. the definition here gives a more convenient sign in (1. for closed M . we have ∗ 1 = vol. This is Poincar´ duality on the level of harmonic forms. We obtain a conjugate linear isomorphism ∗ ⊗ h : A∗ (M. Let ϕ and ψ be differential forms with compact support and of degree r and r − 1. R) is the natural projection.47) (1. we have ∗ ∗ d ∗ ϕ = (−1)(n−r+1)(r−1) d ∗ ϕ = (−1)nr+n−r+1 d ∗ ϕ. τ ) view the Hermitian metric of E as a conjugate linear isomorphism h : E → E ∗ . by Exercise 1. β) vol.46) ∗ d ∗ ϕ. We have ((∗ ⊗ h)α) ∧ε β = (α. 1. C). R) → Ar (M. where ∗ϕ := ∗ϕ.4 Hodge Operator. . ∗ : Hr (M. We conclude that on differential forms of degree r d∗ = (−1)nr ∗ d ∗ . E) → A∗ (M. dψ } vol. the integral over M of the left hand side vanishes. Hence d(∗ϕ ∧ ψ) = d ∗ ϕ ∧ ψ + (−1)n−r ∗ ϕ ∧ dψ = (−1) n−r = (−1)nr+n−r+1 ∗ (∗ d ∗ ϕ) ∧ ψ + (−1)n−r ∗ ϕ ∧ dψ {(−1) nr+1 (1. Since d ∗ ϕ is a differential form of degree n − r + 1 and r − r2 is even. ψ + ϕ. (1.45) where Pr : A∗ (M. and denote by vol the oriented volume form of M . Then the Hodge operator ∗ is defined3 by the tensorial equation ∗ϕ ∧ ψ = ϕ.44) ∗ vol = 1 and ∗∗= (−1)r(n−r) Pr .47). ψ vol. then ∗ maps harmonic forms to harmonic forms. Let E be a vector bundle over M and E ∗ be the dual bundle of E.49) is an isomorphism. 3 Note (1. Assume that E is endowed with a Hermitian metric. respectively. e Extend ∗ complex linearly to A∗ (M. (1. By Stokes’ theorem.48. It is now easy to check that ∗ ∆d = ∆d ∗ . R) (1.

Let D∗ be the induced connection on E ∗ . (1.53) on forms of degree r and n − r. The connection D∗ is Hermitian with respect to h∗ . ∗ (DX (h(σ)))(τ ) = X(σ. h is a parallel morphism from E to E ∗ . satisfy ∆ (∗ ⊗ h) = (∗ ⊗ h) ∆. E) → Hn−r (M. Note that ((∗ ⊗ h)α) ∧ε β is complex valued. Via ξ(h∗ (η)) = (ξ.10 ¨ Lectures on Kahler Manifolds where ε : E ∗ ⊗ E → C is the evaluation map. then ∗ ⊗ h maps harmonic forms to harmonic forms.54) If M is closed.52) Note that h∗ is complex linear in the first and conjugate linear in the second variable. and hence h∗ h = id.30 and Equation 1.55. h(τ )) + (h(σ). τ ) = (DX σ. DX (h(τ )). h(DX τ )) ∗ = (D∗ (h(σ)). h(τ )) = X(σ. τ ) + (σ. τ ) − (σ. τ ) = h(DX σ)(τ ). DX τ ) = (h(DX σ).56) Hr (M. With respect to D and D∗ .55) . by Exercise 1. It follows that (∗ ⊗ h∗ )(∗ ⊗ h) = (∗ ⊗ h)(∗ ⊗ h∗ ) = (−1)r(n−r) (1. Using (1. for simplicity denoted ∆. E ∗ ). The induced (conjugate) Hermitian metric h∗ on E ∗ is given by (h(σ). h∗ (h(τ ))) = h(σ)(h∗ (h(τ ))) = (h(σ). η) we consider h∗ as a conjugate linear isomorphism from E ∗ to E. Hence. DX τ ) = (DX σ. τ ).24) with D the usual derivative of functions and computing as in (1. e This implies that the corresponding Laplacians. (1. This is Poincar´ duality for vector bundle valued harmonic forms. ∗ (1. τ ). we get (dD )∗ = (−1)nr (∗ ⊗ h∗ ) dD (∗ ⊗ h). ∗ ⊗ h induces conjugate linear isomorphisms (1. h(τ )) = (σ. X(h(σ). for closed M .46). h(τ )) + (h(σ). h(τ )) := (σ. We have (σ.

. multiplication by i in a complex vector space is a complex structure on the underlying real vector space. 2. A complex structure on V is an endomorphism J : V → V such that J 2 = −1.1. To be consistent in what we say next. . . . y 1 . . The corresponding complex structure on R2m is J(x1 . Let M be a complex manifold. . . xm ). xm + iy m ) and identify it with the vector (x1 . Any coordinate chart of the corresponding complex structure will be called a holomorphic coordinate chart of M . . the map w ◦ f ◦ z −1 is holomorphic on its domain of definition. . Hence they are diffeomorphisms and the determinants of their derivatives. We say that a smooth atlas A of M is holomorphic if for any two coordinate charts z : U → U ⊂ Cm and w : V → V ⊂ Cm in A. Any holomorphic atlas uniquely determines a maximal holomorphic atlas. We say that a map f : M → N between complex manifolds is holomorphic if. . the coordinate transition map z ◦ w−1 is holomorphic. An automorphism of a complex manifold M is a biholomorphic map f : M → M . . y 1 . For an open subset U of a complex manifold M . Then the transition maps z ◦ w−1 of holomorphic coordinate charts are biholomorphic. . . . Let M be a smooth manifold of real dimension 2m. A Riemann surface or complex curve is a complex manifold of complex dimension 1. for all holomorphic coordinate charts z : U → U of M and w : V → V of N . viewed as R-linear maps. To fix one of our conventions. . where we choose dx1 ∧ dy 1 ∧ · · · ∧ dxm ∧ dy m as orientation of Cm . We say that M is a complex manifold of complex dimension m if M comes equipped with a holomorphic atlas. x1 . compare Example 2. We say that f is biholomorphic if f is bijective and f and f −1 are holomorphic. we remark that open subsets of complex manifolds inherit a complex structure. and a maximal holomorphic atlas is called a complex structure. . xm . Write a vector in Cm as a tuple (z 1 . We will use this identification of Cm with R2m and complex structure J on R2m without further reference. −y m . . z m ) = (x1 + iy 1 . .2 Complex Manifolds Let V be a vector space over R.1 Example. It follows that a holomorphic structure determines an orientation of M . . xm . are positive. we denote by O(U ) the set of holomorphic functions f : U → C. Vice versa. y m ) in R2m . . Such a structure turns V into a complex vector space by defining multiplication with i by iv := Jv. . we discuss the complex vector space Cm explicitly. a ring under pointwise addition and multiplication of functions. y m ) = (−y 1 .

. . z j . the transition map aj ◦ a−1 is defined on k {w ∈ Cm | wj = 0} and given by inserting 1 as k-th variable. we denote by [z] the complex line generated by z and call (z 0 . respectively. However. The Riemann sphere is S 2 together with the complex structure determined by this atlas. multiplying the resulting (m + 1)-vector by (wj )−1 . z m ). . . complex analysis is different from real analysis. Let N = (0. . By what we said it is clear that aj is a bijection. h) = (1 − h)−1 w and πS (w. −1) be the north and south pole of S 2 . described in the next example. complex projective space CP m is the space of all complex lines in Cm+1 . Cm does not contain closed complex submanifolds (of positive dimension). As we will see. and hence the atlas of S 2 consisting of πN and πS is holomorphic. we have the notions of holomorphic immersion. . To state just one phenomenon where they differ. Corresponding to the real case. We obtain a holomorphic atlas by replacing πS by its −1 complex conjugate. The −1 −1 transition map πS ◦ πN : C \ {0} → C \ {0} is given by (πS ◦ πN )(z) = 1/z. . . Then M together with the atlas consisting of the one coordinate chart id : U → U is a complex manifold. 3) Complex projective spaces. Each [z] in Uj intersects the affine hyperplane {z j = 1} in Cm+1 in exactly one point.12 ¨ Lectures on Kahler Manifolds The inverse mapping and implicit function theorem also hold in the holomorphic setting. . 2. . In particular. and deleting the redundant j-th variable 1. 2) Riemann sphere. Of course. For j < k. we let Uj = {[z] ∈ CP m | z j = 0}. ˆ zj where the hat indicates that z j is to be deleted. z m ) ∈ Cm+1 . . and thus πN and πS define a smooth atlas of S 2 . πS . . aj ([z]) = 1 0 (z . Consider the unit sphere S 2 = {(w. For m = 1. and complex submanifold. we speak of the complex projective line. 1) and S = (0. holomorphic embedding. . 1) Let U ⊂ Cm be an open subset. As a set. h) = (1 + h)−1 w. The stereographic projections πN : S 2 \ {N } → C and πS : S 2 \ {S} → C are given by πN (w. z m ) the homogeneous coordinates of [z]. h) ∈ C × R | ww + h2 = 1}. The discussion is completely parallel to the discussion in the real case. Thus the transition maps are holomorphic. for m = 2 of the complex projective plane. it is not holomorphic. We use this to obtain a coordinate map a j : U j → Cm . respectively. the Riemann sphere is biholomorphic to the complex line CP 1 .2 Examples. For a non-zero vector z = (z 0 . . by the maximum principle a holomorphic function on a closed complex manifold is locally constant. It is a consequence of the uniformization theorem that this complex structure on S 2 is unique up to diffeomorphism. . For 0 ≤ j ≤ m. It is smooth. Then the transition map is (πS ◦ πN )(z) = 1/z. . . It is now an exercise to show that there is precisely one topology on CP m such that the maps aj are coordinate charts and such that CP m together with this topology and the atlas of maps aj is a complex manifold of complex dimension m.

There is a canonical projection π : M ∗ → Gr V. C) = Gl(m + 1. the map f : S 2 → 1 CP . where Mat(BF ) denotes the matrix of the linear map BF : Cr → Cn . −1 ZB ([F ]) = F1 F0 . With the same arguments as in the previous example we get that Gr V has a unique topology such that the maps ZB are coordinate charts turning Gr V into a complex manifold of complex dimension r(n − r). [πN (p). 4) Complex Grassmannians. πS (p)] if p = S. and the complex structure on M ∗ does not depend on the choice of B. This example generalizes the previous one. C)/C∗ acts by biholomorphic transformations on CP m . C : V → Cn . f (p) = [1. ϕB ([F ]) = B −1 1 −1 . is a holomorphic embedding. More generally. Then the map M ∗ → Cn×r . We leave it as an exercise to the reader to show that ZB : UB → C(n−r)×r . [z] → [z. 0]. It is known that PGl(m + 1. is a bijection onto the open subset of (n × r)-matrices of rank r. For the Riemann sphere S 2 as in the previous example. Let V be a complex vector space of dimension n and Gr V be the space of r-dimensional complex subspaces of V . then the induced map f : CP m → CP n . F1 where F0 ∈ Cr×r and F1 ∈ C(n−r)×r . F1 F0 . F → Mat(BF ). Let M ∗ be the set of linear maps F : Cr → V of rank r. if A : Cm+1 → Cn+1 is an injective linear map. C) is actually equal to the group of biholomorphic transformations of CP m . π(F ) = [F ] =: im F. for any isomorphism B : V → Cn . and let UB be the subset of [F ] in Gr V such that F0 has rank r. Moreover. the map f : CP m → CP n . is a holomorphic embedding. Thus we can view CP m in many different ways as a complex submanifold of CP n . [z] → [Az]. for any two isomorphisms B. π : M ∗ → Gr V is a holomorphic submersion and. Let B : V → Cn be an isomorphism. This turns M ∗ into a complex manifold of dimension nr. 1] if p = N . We also conclude that the group PGl(m + 1. is well defined and biholomorphic and thus identifies the Riemann sphere with the complex projective line. ϕB : UB → M ∗ . We write Mat(BF ) = F0 . −1 the transition map ZB ◦ZC is holomorphic. where 0 < r < n. is a well defined bijection and that.2 Complex Manifolds 13 For m ≤ n.

(A. Define a bijection ΦB : UB × Cr → π −1 (UB ). Gl(V ) × M ∗ → M ∗ . . As a set. These actions are also holomorphic and π is equivariant with respect to them. Let B : V → Cn be an isomorphism and UB ⊂ Gr V be as in the previous example. the bijection π −1 (W ) (W. C) of invertible matrices in Cr×r is a complex Lie group (for complex Lie groups. ΦB ([F ]. v) = ([F ]. w ∈ W }. ϕB ([F ])v). (W. [F ]) → [AF ]. see Example 8 below) and. In particular. Cr → π −1 ([F ]). Aw). Gl(V ) × Ur V → Ur V. C). w)) → (AW. is a holomorphic section of π. is an isomorphism of vector spaces. where ϕB is as in the previous example. (A. There is a canonical projection π : Ur V → Gr V. considered as group of automorphisms of Cr . With arguments similar to the ones in the previous examples it follows that Ur V is a complex manifold in a unique way such that π : Ur V → Gr V is a complex vector bundle over Gr V and such that the trivializations ΦB as above are holomorphic. The complex Lie group Gl(V ) of automorphisms of V acts on M ∗ and Gr V on the left. the left action of Gl(V ) on Gr V extends canonically to a holomorphic action on Ur V . 5) Tautological or universal bundle. w) → W. the map v → ΦB ([F ].14 ¨ Lectures on Kahler Manifolds where 1 stands for the r × r unit matrix and where we consider the matrix on the right as a linear map Cr → Cn . w) → w ∈ W turns the fiber π −1 (W ) into an r-dimensional complex vector space isomorphic to W . M ∗ × Gl(r. This action is holomorphic and turns π : M ∗ → Gr V into a principal bundle with structure group Gl(r. C) → M ∗ . For each [F ] ∈ UB . The group Gl(r. (W. A) → F A. v). F ) → AF. For each W ∈ Gr V . (A. w) | W ∈ Gr V. (F. Moreover. the complex dimension of Ur V is r(n − r + 1) and π is holomorphic. Let 0 < r < n and M = Gr V be the Grassmannian of r-dimensional complex linear subspaces in V as in the previous example. acts on M ∗ on the right. the universal bundle over Gr V is equal to Ur V = {(W. respectively Gl(V ) × Gr V → Gr V.

8) Complex Lie groups. . Then Z acts freely and properly discontinuously on Cm \ {0} by (k. It follows from the uniformization theorem that any complex curve diffeomorphic to the torus S 1 × S 1 is an elliptic curve. A generalization of this example is due to Calabi and Eckmann. the map f : T → T . is well defined and biholomorphic. 6) Complex tori. Hence the quotient T = Γ\Cm inherits the structure of a complex manifold such that the covering map Cm → T is holomorphic.) Since multiplication by z k is biholomorphic. Let m ≥ 1 and z ∈ C be a non-zero complex number of modulus |z| = 1. Choose an R-basis B = (b1 . Let T = Γ\Cm be a complex torus and f : T → T be a biholomorphic map. a discrete subgroup of the additive group Cm . g is affine. . (z. It is an exercise to show that M is diffeomorphic to S 2m−1 × S 1 . and multiplication G × G → G and inversion G → G are holomorphic maps. M inherits from Cm \ {0} the structure of a complex manifold. see [CE] (Example 2. Let Γ ⊂ Cm be the lattice consisting of all integral linear combinations of B. w) with w = 0 and of pairs (W. is well defined and holomorphic and turns T into a complex Lie group as in Example 8 below. z) → tk (z) := k + z. C) ⊂ Gl(n. Continuity implies that there is a constant C such that |g(z)| ≤ C(1+|z|) for all z ∈ Cm . the general linear group G = Gl(n. . who showed that the product of odd-dimensional spheres carries a complex structure. the complex structure described in the next example turns S 1 × S 1 into an elliptic curve. 7) Hopf manifold (complex structures on S 2m−1 × S 1 ). This action is free and properly discontinuous. C) and turns Gl(V ) into a complex Lie group.5 in [KN. Vice versa. Chapter IX]). Hence f is of the form f (z) = Az + b with b ∈ T and A ∈ Gl(m. w) → z + w. C) is a complex manifold of complex dimension n2 . hence T with the above complex structure is called a complex torus. . (k. 0). We also note that addition T × T → T .2 Complex Manifolds 15 This action has two orbits: the set of pairs (W. The quotient M = (Cm \ {0})/Z is called a Hopf manifold. Now T is diffeomorphic to the 2m-fold power of a circle. C) is a complex Lie subgroup of complex codimension 1. b2m ) of Cm . for any such A ∈ Gl(m. The special linear group Sl(n. Then Γ acts by translations on Cm . and the special linear group Sl(V ) is a complex Lie subgroup of complex codimension 1. By a standard result from complex analysis. the translation tk : Cm → Cm is biholomorphic. Then any continuous lift g : Cm → Cm of f is Γ-equivariant and biholomorphic. For each fixed k ∈ Γ. then any isomorphism B : V → Cn identifies the general linear group Gl(V ) of V with Gl(n. (Hint: Consider the case z = 2 first. v) → z k · v. If V is a complex vector space of dimension n. f (z) = Az + b. In particular. A one-dimensional complex torus is called an elliptic curve. C) such that A(Γ) = Γ. independently of the choice of B. As an open subset of Cn×n . C) and b ∈ T . .

Clearly. We say that a complex manifold M is homogeneous if the group of automorphisms of M is transitive on M . a complexification of G consists of a complex Lie group GC together with an inclusion G → GC such that any smooth morphism G → H. if G is a complex Lie group and H is a closed complex Lie subgroup of G. We say that V is a rational curve if V is smooth and biholomorphic to 2 CP 1 . SU(n). U(n)/T = Gl(n. where Pλ is a suitable parabolic subgroup of GC associated to λ. where T is the maximal torus of diagonal matrices in U(n) and B is the Borel group of all upper triangular matrices in Gl(n. locally. Similarly. see Section 4. Hence Ad is constant if G is compact. The coordinate transformation on U1 ∩ U2 ∩ C is u2 = 1/u1 . a closed complex Lie subgroup of GC . where H is a complex Lie group. that is. C). page 167]. consider the complex curve C = {[z] ∈ CP 2 | z0 = z1 z2 } 2 in CP . In fact. C). complex tori. Section 27]. Outside of its singular locus. C). C). which is contained in U1 ∪ U2 . if G is a connected compact Lie group and Gλ is the stabilizer of some λ ∈ g∗ under the coadjoint representation. the defining equation has degree two. if GC exists. Thus C is biholomorphic to CP 1 and hence is a rational curve in CP 2 . Any connected compact Lie group has a complexification [Bu. that is. the stabilizer of the standard flag in Cn . see Chapter 8 in [Bes].g. hence we may use u1 = z0 /z1 as a holomorphic coordinate for C on U1 ∩ C. It follows that compact complex Lie groups are Abelian. For a Lie group G. For example. the Lie algebra g of a complex Lie group G is a complex vector space and the adjoint representation Ad of G is a holomorphic map into the complex vector space of complex linear endomorphisms of g. extends uniquely to a holomorphic morphism GC → H. and Gl(n. then it is unique up to isomorphism. on U2 ∩ C we have z1 /z2 = (z0 /z2 )2 and we may use u2 = z0 /z2 as a holomorphic coordinate for C on U2 ∩ C. e. C) is the complexification of Sl(n. For example.. Flag manifolds. C)/B. then the quotient G/H is in a unique way a homogeneous complex manifold such that the natural left action by G on G/H and the projection G → G/H are holomorphic. We say that V is smooth if its singular locus is empty. respectively. away from the subset where the defining equations do not have maximal rank. A closed subset V ⊂ CP n is called a (complex) projective variety if. V is defined by a set of complex polynomial equations. R).16 ¨ Lectures on Kahler Manifolds The unitary group U(n) is not a complex Lie group – recall that its defining equation is not holomorphic.12 in [DK]. On U1 ∩ C we have z2 /z1 = (z0 /z1 )2 . Sl(n. coadjoint orbits of connected compact Lie groups. 10) Projective varieties. For example. that is. a projective variety is a complex submanifold of CP n . the complexifications of SO(n). For example. then the inclusion of G into its complexification GC induces an isomorphism G/Gλ → GC /Pλ . A well known theorem of Chow says that any closed complex submanifold of CP n is a smooth projective variety. In fact. Sl(n. 9) Homogeneous spaces. In this example. For more on flag manifolds. see [GH. are homogeneous complex manifolds. and U(n) are SO(n. .

X) 2 . we define a complex structure Jp on Tp M by j The map CP 1 [1.15 and 2. Since the transition maps of holomorphic coordinate charts are holomorphic. N (X. Compare also Exercises 2. Note however that T M ⊗R C is not a holomorphic vector bundle over M in any natural way. where.32. Y ) − J(JY. Let M be a complex manifold. . 4) The universal bundle Ur V → Gr V is holomorphic.2 Complex Manifolds 17 2. X) + J J(Y. z 2. (2. Show that 1 N (X. JY ] − [X. U ) be a holomorphic coordinate chart of M . 1) The tangent bundle T M together with its complex structure J is a complex vector bundle over M . we let z j = x + iy j and write the corresponding coordinate frame as (X1 . Xm . For p ∈ U . As usual. An almost complex structure J = Jp is called a complex structure if it comes from a complex structure on M as in (2. 2) If E → M is holomorphic. (2. 1) N is a tensor. 2.3 Examples. We say that a complex vector bundle E → M is holomorphic if E is equipped with a maximal atlas of trivializations whose transition functions are holomorphic. Section 6. Y1 . Y ) − J J(X. . Ym ). z. .5) Jp Yj (p) = −Xj (p). Such an atlas turns E into a complex manifold such that the projection E → M is holomorphic. see [Ka2. The usual coordinates for the tangent bundle have holomorphic transition maps and thus turn T M into a complex manifold and holomorphic vector bundle over M . Any almost complex structure on a surface is a complex structure (existence of isothermal coordinates). A celebrated theorem of Newlander and Nirenberg [NN] says that an almost complex structure is a complex structure if and only if its Nijenhuis tensor or torsion N vanishes. JY ] − J[JX. the dual bundle E ∗ is holomorphic. Y ]}. . z] → [1.4) above. 2) Let be a torsion free connection and J be an almost complex structure on M . then the complex tensor bundles associated to E are holomorphic. 3) Let E → M be a holomorphic vector bundle and f : N → M be a holomorphic map.6 Exercises. . Y ) = J(JX. Y ] − J[X.3]. z m ] ∈ CP m extends to a holomorphic 1 embedding of CP into CP m .4) For an instructive discussion of the Newlander–Nirenberg theorem and its proof. Jp is independent of the choice of holomorphic coordinates. for vector fields X and Y on M . Vice versa. then a smooth field J = (Jp ) of complex structures on T M is called an almost complex structure of M . Then the pull back f ∗ E → N is holomorphic. Let (z. . if M is a smooth manifold of real dimension 2m. For example. . Y ) = 2{[JX. and the maximal degree of the obvious defining equations of the image is m. We obtain a smooth field J = (Jp ) of complex structures on T M . Jp Xj (p) = Yj (p). .

w ∈ Tp S 6 . consider the sphere S 6 of purely imaginary Cayley numbers of norm one. . 2 (2.8 Exercise. 2. extend v to a vector field V in a neighborhood of p in S 6 . By parallel translation along great circle arcs through p. 2. Since p is purely imaginary of length one. 2 15 in [Ad] is a good reference to Cayley numbers. Let V be a real vector space.10) for the eigenvalues i and −i.11) 1 (v − iJv). and let J be a complex structure on V . z] = (x · y) · z − x · (y · z) denotes the associator of x. where S denotes the second fundamental form of S 6 . It is a famous open problem whether S 6 carries any complex structure. w) = 2{(p · v) · w − (p · w) · v − p · (v · w) + p · (w · v)} = 4[p. z ∈ Ca. Hence J = (Jp ) is an almost complex structure on S 6 . v. where the dot refers to multiplication in Ca.18 ¨ Lectures on Kahler Manifolds to conclude that J is a complex structure if J = 0. hence VC is the sum (2. y. Since |x · y| = |x| · |y| for all x. where [x. For v. J is norm preserving. We extend J complex linearly to the complexification VC = V ⊗R C of V . J(v ⊗ α) := (Jv) ⊗ α. Then V is parallel at p.1 Complex Vector Fields. define Jp v := p · v. p · (p · x) = −x for all x ∈ Ca. y. v → v := 4 Chapter Then we still have J 2 = −1. v → v := 1 (v + iJv). In the normed division algebra Ca of Cayley numbers 4 . Hence dJV (u) = u · v + p · dV (u) = u · v + p · S(u. respectively. We conclude that N = 0 and hence that J does not come from a complex structure on S 6 . It follows that Jp v ∈ Tp S 6 and that 2 Jp v = p · (p · v) = −v. Along the great circle arc cos t · p + sin t · u in the direction of a unit vector u in Tp S 6 .9) VC = V ⊕ V of the eigenspaces V and V maps V → V . (2. w]. For a point p ∈ S 6 and tangent vector v ∈ Tp S 6 . The V → V . 2.7 Example. View the sphere S 2 as the space of purely imaginary quaternions of norm one and discuss the corresponding almost complex structure on S 2 . the vector field JV is given by (cos t · p + sin t · u) · V (cos t · p + sin t · u). y ∈ Ca and p ∈ S 6 has norm one. we get N (v. v).

but not a holomorphic vector bundle in a natural way. U ] − [Y. Y ] and N (X. In the notation of (2. by (2. . [X + iY. Let M be a smooth manifold with an almost complex structure J and TC M = T M ⊗R C be the complexified tangent bundle. (2. Complex vector fields act as complex linear derivations on smooth complex valued functions. that is. The isomorphism T M v → v ∈ T M as in (2. V ] + [Y. xm . Suppose from now on that M is a complex manifold. Let (z. And similarly for T M . .17) .13). sections of T M .14) 2. where X and Y are vector fields of M . Any such field can be written in the form Z = X + iY . U ]). ∂z j 2 (2. . where X and Y are vector fields on M . . respectively.12) The decomposition in (2. U + iV ] := [X. The bundle T M is a smooth complex vector bundle over M .2 Complex Manifolds 19 are complex linear and conjugate linear isomorphisms. (2.11) shows that T M and T M are smooth subbundles of the complexified tangent bundle TC M . N (X.13) turns T M into a holomorphic vector bundle. v → v . The Nijenhuis tensor associated to J vanishes iff T M is an involutive distribution of TC M . that is. y 1 . y m ). Z2 ] is a section of T M whenever Z1 and Z2 are. Y1 . V ] + i([X.11) the maps T M → T M. Then T M (with complex multiplication defined via J) is a holomorphic vector bundle over M . ∂z j 2 Zj := ∂ 1 = (Xj + iYj ).10). Ym ) is the coordinate frame associated to the coordinates (x1 . we have the eigenspace decomposition with respect to J. A complex vector field of M is a section of TC M . (2. . and T M → T M. Zj = Xj and Zj = Xj . if [Z1 . Moreover. v → v . . U ) be a holomorphic coordinate chart for M .15 Exercise. As in (2. We extend the Lie bracket complex linearly to complex vector fields. where multiplication by i on T M is given by J. Write z j = xj + iy j and set Zj := ∂ 1 = (Xj − iYj ). . Similarly. Y ] .13) are complex linear respectively conjugate linear isomorphisms of complex vector bundles over M . TC M = T M ⊕ T M. any complex vector field Z has components Z = 1 1 (Z − iJZ) and Z = (Z + iJZ) 2 2 (2. Xm . More precisely. if we consider V together with J as a complex vector space. Y ) = −8[X . Y ) = −8[X .16) where (X1 . .

J). or is not compact. In particular. Proof. the group Aut(M. then dim a(M ) < ∞. Zm ) is the corresponding local holomorphic frame of T M . equivalently. = i∂X(Y ) + i∂X(Y ) − idY (X) − 2i∂X(Y ) Hence [X. The complex structure J turns a(M ) into a complex Lie algebra. Aut(M. . 2. .18 Proposition. or a complex torus. Y ] − 2i∂X(Y ). The space a(M ) of automorphic vector fields on M is a Lie algebra with respect to the Lie bracket of vector fields. Hence we get [X.4). . iY ] = i[X. 2. . Xm ) is a local holomorphic frame for T M considered as a holomorphic vector bundle over M and that (Z1 . Then the Lie bracket of X with Y is given by dX(Y ) − dY (X). Y ] for all vector fields Y on M . . for a closed complex manifold M . Let Y : U → Cm be another vector field.20 ¨ Lectures on Kahler Manifolds in T M and T M .20 Remark. . To be automorphic or holomorphic is a local property. We note that (X1 . J) is either trivial. If M is a closed complex manifold. iff [X. the theorem is a trivial consequence of Proposition 2. Then the vector field is given by a smooth map X : U → Cm and J is given by multiplication by i.2. Proof. There is a corresponding common misunderstanding in the case of isometry groups of Riemannian manifolds and of other groups of automorphisms. compare (3. iY ] = dX(iY ) − idY (X) = ∂X(iY ) + ∂X(iY ) − idY (X) = i∂X(Y ) − i∂X(Y ) − idY (X) = i[X. The space of holomorphic sections of a holomorphic vector bundle E over M is precisely the kernel of the elliptic differential operator ∂ on the space of smooth sections of E. .19.19 Proposition. A real vector field X on M is automorphic iff it is holomorphic as a section of the holomorphic vector bundle T M . . 2. Without the assertion about the topology. Y ] iff ∂X(Y ) = 0. JY ] = J[X. A celebrated theorem of Bochner and Montgomery states that. Hence we can check the equivalence of the two properties in holomorphic coordinates (z. see [BM]. J) of automorphisms f : M → M is a complex Lie group with respect to the compact-open topology5 and that a(M ) is the Lie algebra of Aut(M. .8. This holds iff the Lie derivative LX J = 0 or. compare Example 2. respectively. We say that a real vector field X on M is automorphic if the flow of X preserves the complex structure J of M . U ). 5 This is a subtle point.

We note however that a non-zero alternating r-form ϕ satisfies J ∗ ϕ = ϕ iff r is even and ϕ is of type (r/2. .0 (M. respectively. Write z j = xj +iy j and set (2. q) is given by a linear combination aJK dz J ∧ dz K = aJK dz i1 ∧ · · · ∧ dz ip ∧ dz j1 ∧ · · · ∧ dz jq . C). such that. αvr ) = αp αq ϕ(v1 . we have dz j = ∂z j dwk ∂wk and dz j = ∂z j dwk . vr ).q (M. C) plays a special role. q).25) dz j = dxj + idy j and dz j = dxj − idy j . . . C) = Λ∗ TC M . To see ∗ this. We call elements of Λp. where TC M = T ∗ M ⊗R C is the complexified cotangent bundle.q VC corresponds to complex valued alternating r-forms ϕ on V . ∂wk (2. differential forms of type (1.p VC is invariant under conjugation. . 1). ∗ Λr VC = Λr (V ⊕ V )∗ ∗ = ⊕p+q=r Λp (V )∗ ⊗ Λq (V )∗ =: ⊕p+q=r Λp. .q VC . (2. C).11).21). . Let z : U → U be a holomorphic coordinate chart for M . (2. . . and Λp. As in (2. vr ) := ϕ(Jv1 . p)-forms fixed under conjugation is the space of real valued forms of type (p. C) = ⊕p+q=r Ap. (2.q VC to Λq.27) . a differential form ϕ of type (p.q VC alternating forms on V of type (p. C) := Λp.2 Differential Forms. . r/2).q (M.26) J. we have the decomposition Ar (M. As above. .21) Alternating forms on VC correspond to complex multilinear extensions of com∗ plex valued alternating forms on V . let V be a real vector space with complex structure J. (2. vr ). 0) and (0. r = p + q. In terms of these. .p VC . .24) ∗ where Ap. Λp. .q (M.q TC M . C) are called differential forms of type (p. .2 Complex Manifolds 21 2. The space of complex (p. Under a transformation z ◦ w−1 of holomorphic coordinates of M . ϕ(αv1 . it is called the canonical bundle.23) but this does not characterize the type. write vj = vj + vj as in (2. In this interpretation. the space of such differential forms is denoted Ap. Jvr ) = ip−q ϕ(v1 . . for α ∈ C. ∗ ∗ Note that A∗ (M.10) determines a decomposition of the space of complex valued alternating forms on VC . .22) where we view V together with J as a complex vector space as usual. . An alternating r-form ϕ of type (p. . p). q) satisfies (J ∗ ϕ)(v1 . ∗ ∗ ∗ Conjugation maps Λp. (2. . Smooth sections of Ap.K J.K where J and K run over multi-indices j1 < · · · < jp and k1 < · · · < kq . . q). Suppose now that M is a complex manifold with complex structure J. The complex line bundle KM := Am.q (M. The decomposition of VC in (2.

1 = ∂ if J is a complex structure. C) − Ap. we have the decomposition Ar (M.0 (M.28) =: ∂ϕ + ∂ϕ. C) into types. The = alternating sum χ(M.0 (M.1 (M. C) = ⊕p+q=r Ap. ∂∂ = −∂∂.q (M. C) into holomorphic bundles over M . C).0 = ∂ and d0.q (M. C) + A0. These are called holomorphic forms of degree p. C). By definition. However.0 (M.q (M. If ϕ is a smooth complex valued function.0 ϕ + d0. C) are not holomorphic in any natural way for 0 < r ≤ 2m and 0 < q ≤ m. hp. then we can decompose as in the case of complex manifolds.q (M. C) = (−1)p dimC Ωp (M ) (2. (2. dϕ = d1. q).2 (M. we get ∂ 2 = 0. then a new term may arise. C) − Ap. ∂ 2 = 0. C) are called Dolbeault cohomology groups of M .1 ϕ + d−1. C) + A1.0 (M. q + 1). C).0 (M.0 (M. C).q (M. that is ϕ ∈ A0 (M. C).q−1 (M. independently of the choice of holomorphic coordinates.30) whose cohomology groups H p. As in the case of complex manifolds.0 ϕ + d0. C) − · · · → → → → ∂ ∂ ∂ ∂ (2.q+1 (M.0 (M. They are invariants associated to the complex structure of M .32 Exercise. respectively. The type of ∂ϕ is (p + 1.q (M. C) + A0.2 ϕ ∈ A2. 0). hence they are well defined. are called Hodge numbers of M . O) := (−1)p hp. where d1. . the type of ∂ϕ is (p. Now d = ∂ +∂ and d2 = 0. We also see that the complex vector bundles Ar (M. C) consists precisely of the holomorphic sections of the holomorphic vector bundle Ap. C) = dimC H p.31) is called the arithmetic genus of M .1 ϕ ∈ A1. if ϕ is of type (1.29) In particular. 2. Let M be a smooth manifold with an almost complex structure J.22 ¨ Lectures on Kahler Manifolds This shows that the natural trivializations by the forms dz I as above turn the bundles Ap. we get differential cochain complexes · · · − Ap. dϕ = d1. Ωp (M ) ∼ H p. Their dimensions. C) and Ap. Hence by comparing types. The kernel Ωp (M ) of ∂ on Ap. C).1 (M. A quick computation gives dϕ = = Xj (aJK )dxj + Yj (aJK )dy j ∧ dz J ∧ dz K Zj (aJK )dz j + Zj (aJK )dz j ∧ dz J ∧ dz K (2.

Any differential form is locally a finite sum of decomposable differential forms. Y ) for all ϕ ∈ A1. a symmetric complex bilinear form ·. We say that a Riemannian metric g = ·. Y1 . Show that d−1. Let M be a complex manifold with corresponding complex structure J. also denoted g or ·. Let M be a complex manifold with complex structure J. is symmetric and satisfies the following three conditions: Z1 . A complex manifold together with a compatible Riemannian metric is called a Hermitian manifold 6 . C) for any differential form ϕ of type (p. (2. Conclude that dϕ ∈ Ap+2. Xm . Z > 0 for Z1 . Y on M . C) + Ap+1.33) also holds for the complex linear extension of J to TC M and complex vector fields X and Y .2 is a tensor field and that ϕ(N (X. . C) + Ap. unless Z = 0. . Yk (p0 ) = δjk and Xj . Z2 = Z1 . . q). · is compatible with J iff about each point p0 ∈ M . . . · on TC M satisfying these three conditions is the complex bilinear extension of a Riemannian metric satisfying (2. Y on M . .1 (M. · is compatible with J if JX. 6 The usual arguments give the existence of compatible Riemannian metrics. 2. We note the following immediate consequence. . Y (2.q+2 (M.q−1 (M.0 (M.33). C) and vector fields X.34) Vice versa.q (M.3 Compatible Metrics. . z m ) = (x1 + iy 1 . Z1 . Which parts of dϕ are tensorial in ϕ? 2. · . Let M be a complex manifold as above. Y )) = 8 · d−1. Yk (p0 ) = 0. Then a Riemannian metric g = ·.2 ϕ(X. then the complex bilinear extension of g to TC M .33) for all vector fields X. . C). Z2 = 0 Z. If g is a compatible Riemannian metric on M . Xk (p0 ) = Yj .q+1 (M. . Z2 in T M . there are holomorphic coordinates z = (z 1 .35 is straightforward and left as an exercise. Z2 . The proof of Proposition 2.2 Complex Manifolds 23 and similarly for ϕ ∈ A0. . xm + iy m ) such that the associated coordinate frame (X1 . JY = X. C) + Ap−1. Ym ) satisfies Xj . .35 Proposition. . Then (2.

ψ . w the other equations follow. to the open subset {(L. Hence the system of equations defining V has constant rank m − 1. Let M be a complex manifold and p be a point in M . and hence V is a complex submanifold of U × CP m−1 of dimension m with S = V ∩ {z = 0} ∼ CP m−1 . denoted [w]. C) = ⊕p+q=r Ap. Ar (M. g). 2. In U × CP m−1 . The second assertion follows from the first since the volume form has type (m. C) Proof. = 7 Recall V = {(z. we can identify Sr = {(L. z) ∈ U1 Cm | |z| = r} with the sphere of radius r in Cm .37 Remark. A similar construction can be carried out for points in complex manifolds. j}. Thus we can think of U1 Cm as Cm . 0) ∈ U1 Cm }. We can think of (S 2m−1 .q (M. On the subset {wj = 0} of CP m−1 . [w]) | z i wj = z j wi for all i. The first assertion is clear from Proposition 2. We identify the latter with CP m−1 and thus have blown up the point 0 of Cm to CP m−1 . [Kar. that dz ∧ dz = −2idx ∧ dy.24 ¨ Lectures on Kahler Manifolds 2. . where the point 0 is replaced by the set {(L. z) → z. z) ∈ U1 Cm | z = 0} is biholomorphic onto Cm \ {0}. The precise construction goes as follows. zm ) : U → U be holomorphic coordinates about p with z(p) = 0. page 221]. Let M be a Hermitian manifold. gr ) as a collapsing family of Riemannian manifolds with limit (CP m−1 . Then the projection π : Sr → CP m−1 turns into the Hopf fibration: The fibers of π intersect Sr in Hopf circles. m)7 . and [CGr]. we obtain a Riemannian metric gr on the sphere S 2m−1 . i = j.35. that is. Then the type decomposition is orthogonal with respect to the induced Hermitian metric (ϕ. 2. Consider the universal bundle U1 Cm over CP m−1 . For any r > 0. see Example 3. the intersections of complex lines in Cm with the sphere Sr . ψ) = ϕ. ∗ϕ has type (m − q.36 Corollary. The restriction of the map U1 Cm → Cm . The differential geometric significance of this kind of collapse was first recognized by Berger: The sectional curvature of the collapsing spheres stays uniformly bounded as r → 0. (L.4 Blowing Up Points. consider the set where points in CP m−1 are given by their homogeneous coordinates. In particular. . . . m − p) if ϕ has type (p. as r → 0. The blow up of M at p replaces p by the space of complex lines in Tp M .35 in [CE]. V is defined by the m − 1 independent equations wi z i = j z j . q). . Renormalizing the given Riemannian metric on Sr (of sectional curvature 1/r2 ) by adding the pull back of any fixed Riemannian metric g on CP m−1 . Let z = (z1 .

hence V \ S ∼ U \ {p}. corresponds to the projection. [w]) → z. ˜ ˜ 2) Let f : M → N be a holomorphic map. (z. where the maps fj are holomorphic with fj (0) = ∂j f (0). the blow up of M at p does not depend on the choice of centered holomorphic coordinates. z). We use it to glue V to M \{p} and obtain a complex manifold ˜ M . The fibers of π intersect Sr in Hopf circles. Show that “ h i” f (z. compare with Remark 2. that is. .38 Exercises. together with a holomorphic map ˜ π: M → M ˜ such that π −1 (p) = S and π : M \ S → M \ {p} is biholomorphic. Let M be the blow up of M at p as above. the blow up of M at p. For any r ∈ (0. Then the map Vε := {(z. the canonical map V \ S → U \ {0}. [w]) ∈ V | |z| = r} corresponds to the sphere of radius r in Cm . Write ˆ f (z) = fj (z)z j . as r → 0. Choose ε > 0 such that the image U of the holomorphic coordinates z contains the ball of radius ε > 0 about 0 in Cm . the set Sr = {(z. In this sense. and let M be the blow up of M at p with respect to the coordinates z .37 above. 1) Let z = f (z) be other holomorphic coordinates of M about p with z (p) = ˆ ˆ ˆ 0. as the collapse of S 2m−1 to CP m−1 along Hopf circles. [w]) ∈ V | |z| < ε} → {(L. then there is a holomorphic map f : M → N ˜ such that f = f ◦ π. [w]) → ([w].2 Complex Manifolds 25 On V \ S. Thus we can again think of π : Sr → CP m−1 as the Hopf fibration and of the convergence Sr → S. σ(z. the intersections of complex lines in Cm with Sr . Hence a neighborhood of S in M is biholomorphic to a m neighborhood of the zero-section of U1 C such that the map σ : Vε → CP m−1 . If f is constant on S. [w]) = f (z). = More precisely. ˜ is biholomorphic. [w] is determined by z. ε). is biholomorphic. [w]) = [w]. ˜ 2. where N is another complex manifold. (z. ˜ We consider V as an open subset of M . z) ∈ U1 Cm | |z| < ε}. fj (z)wj ˜ ˆ extends the identity on M \{p} to a biholomorphic map M → M .

g. There is a natural projection H → C. respectively. 2 we can identify Q with the algebraic hypersurface 2 H = {(z0 .10. z1 ) over U1 and ˜ (u2 . [w]) ∈ C3 | z = 0. the action is locally of the form ±1 · (x + t) = x ± t as above.26 ¨ Lectures on Kahler Manifolds 2. The coordinate transformation over the preimage of U1 ∩ U2 is (u2 . C) = A (CP . 2 Recall that C = {[w] ∈ CP 2 | w0 = w1 w2 } is an embedded CP 1 . In this example it will be convenient to enumerate coordinates by subindices. [w]) ∈ C3 × CP 2 | zi wj = zj wi }. Section 7. For more information on the Kummer surface we refer to [Jo.40 Examples. 1) Kummer surface. z2 ) ∈ C3 | z0 = z1 z2 } ⊂ C3 . z2 ) over U2 establishes a biholomorphic map between H 1. which has a singularity at the origin 0. Thus we can resolve each of the quotient singularities on T 4 /Z2 by the above construction and obtain a compact complex surface.5 in [Hu]. Consider the quotient Q = C2 /Z2 . We conclude that the identification (u1 .3]. Section 4. and du2 = −u−2 du1 over U1 ∩ 1 U2 . where Z2 = {1.6 in [GH] or Section 2.2.0 1 and A (C. w0 = w1 w2 }. Note that away from 0 and q0 . C) has 1 1 duj as a nowhere vanishing section over Uj . ˜ In particular. Scalar multiplication by Z2 on C2 descends to an action of Z2 on T 4 . z 1 = t2 .0 1. C). Let T 4 = Z4 \C2 . z1 du1 ) ↔ (u1 . By replacing centered holomorphic coordinates about p by a tubular neighborhood of the zero section in the normal bundle. . z1 .0 (C. there is a rather immediate generalization of the blow up of points to a blow up of complex submanifolds. We have 2 ˜ π −1 (H \ {0}) = {(z. The holomorphic cotangent bundle A1. we resolved the singularity of H. the projection is a twofold covering with holomorphic covering transformations. Let q0 be the image of 0 under the natural projection C2 → Q. This action has 16 fixed points. −1} acts by scalar multiplication on C2 . u2 z1 ). the closure of π −1 (H \ {0}) in C3 is the regular hypersurface 2 ˜ ˜ ˜ H = {(z. [w]) ∈ C3 | w0 = w1 w2 } ⊂ C3 . com˜ pare Example 2. At each fixed point x ∈ T 4 . We choose ˜ over the preimage of Uj under this (uj . 2. zj ) as holomorphic coordinates for H projection. Thus by blowing up 0 ∈ C3 . z2 ) = (u−1 . see e.39 Remark. We blow up C3 at 0 to get ˜ C3 = {(z. namely the points with integral or half-integral coordinates in T 4 . the Kummer surface. −z2 du2 ) ↔ (u2 . Via z 0 = t1 t2 . turning Q \ {q0 } into a complex manifold. 1 z 2 = t2 .

. Then f is not constant on Si (we assume m ≥ 2). . is well defined and an orientation preserving diffeomorphism. Use coordinates z about p as in the definition of blow ups and w/w0 as above m m ˜ to define the connected sum M #CP . Hence the above g is of the form g(z) = Az for some A ∈ Gl(m. . the induced map g : T → T is biholomorphic with g(pi ) = pi for all i. f (p ) = p (w0 w/|w|2 . ˜ 2. Thus k = l and. p ∈ T \ {0} such that there is no A ∈ Gl(m. w] | w0 = 0} ⊂ CP m . if p = [w0 . Let f : M → M be a holomorphic map. C) such that A(Γ) = Γ. . pk ) and (p1 . . 0] in U0 = {[w0 . respectively. p2 = p. [w]) ∈ Vε if p ∈ M \ {p}. are not biholomorphic.2 Complex Manifolds 27 2) Dependence of blow-up on points. respectively. 0. as ˜ a smooth manifold. Set Uε = {[w0 . p1 = p1 = 0. pl . Suppose now that f is biholomorphic.8 Let m ≥ 2 and T = Γ\Cm be a complex torus. Then the map f : M #CP → M . . . where CP denotes CP m with orientation opposite to the standard one: Choose w/w0 as a coordinate about q = [1. It follows that π ◦ f maps Si to a point in T . C) with A(Γ) = Γ and Ap = p . . In particular. hence any such lift is constant. pk and p1 . . p2 = p . p .2. . In particular. . Then by what we just said. Now Si is a closed complex manifold. there are pairs of points p. . the restriction of π ◦ f to Si lifts to a holomorphic map Si → Cm . w] ∈ CP m | |w0 | < ε|w|}. On the other hand. . In terms of oriented smooth manifolds. . w] ∈ Uε . the blow up M does not depend on the choice of p ∈ M . Since Si ∼ CP m−1 is simply = connected. . Moreover. Let (p1 . Let ε > 0 and ˜ Vε ⊂ M be as above. .41 Exercise. Let M and M be the blow ups of T in the points p1 . . the corresponding blow ups M and M of T in 0. In Example 2. . C) with A(Γ) = Γ and Ap = p (in T ). p and 0.6 above we showed that any biholomorphic map of T is of the form h(z) = Az + b with b ∈ T and A ∈ Gl(m. . respectively. there is a holomorphic map g : T → T such that π ◦ f = g ◦ π. Let now k = l = 2. up to renumeration. f (Si ) = Si for all i. pl ) be two tuples of pairwise different points in T . Let Si and Sj be the preimages of pi and pj under π and π . the blow up M of M at m m p ∈ M corresponds to the connected sum M #CP . and let π : M → T and π : M → T be the projections. . We conclude our discussion of blow ups with a fact on the automorphism group of a compact complex surface which we cannot prove in the framework 8I owe this example to Daniel Huybrechts. . By what we said above. it follows that f maps each Si biholomorphically to an Sj .

. and hence the complex structure. Then automorphisms in Aut0 (Mp ) leave the exceptional divisor S = π −1 (p) invariant and. 0]. Consider the blow up of CP 2 in one or two points. 0] and [1.e. .1.42 Proposition. but which will provide us with an important example in Subsection 7. i. which we choose to be [1. 0. p). J) the group of biholomorphic transformations of M . Chapter 4. and by Aut0 (M ) the component of the identity in Aut(M ). Let M be a compact complex surface and π : Mp → M be the blow up of M in p ∈ M . Given some facts about topological intersection properties of analytic cycles. endowed with the compact-open topology. 2. C) and (2. there may be automorphisms permuting the various exceptional divisors.44)     0 ∗ ∗ 0 0 ∗ For a blow up of CP 2 in three points. 2. Aut0 (Mp ) ∼ {Φ ∈ Aut0 (M ) | Φ(p) = p} =: Aut0 (M. via restriction to M \ S. the automorphism group. A corresponding statement would be wrong for the full automorphism group. = The action of Φ ∈ Aut0 (M. since if M itself is already a blow up.28 ¨ Lectures on Kahler Manifolds of these lecture notes.4. = The only issue is to show that an automorphism of Mp in Aut0 (Mp ) leaves S invariant. By Proposition 2. C) . clearly depends on the choice of points.43 Example. denote by Aut(M ) = Aut(M. see [GH]. 0]. [0. 0 ∗ ∗ ∈ Gl(3. the respective automorphism groups are         1 ∗ ∗   1 0 ∗ 0 ∗ ∗ ∈ Gl(3. the proof of this is actually quite simple and geometric. For a closed complex manifold M with complex structure J. p) on S ∼ P (Tp M ) is induced by dΦp . 0.42. on whether they are in general position or not. respectively. 1.

Let Ap. Then is a fine resolution of Ωp (E).q−1 (M. (3. Hence H q (M. 3. q) with values in E is a linear combination α = ϕj ⊗ Φj . we denote by O(U. . E) − · · · → → → → ∂ ∂ ∂ ∂ (3.7 Remark. For U ⊂ M open. we can distinguish forms according to their type as before. The dimensions hp. The space of such differential forms is denoted Ap. a module over the ring O(U ). Ωp (E)) is isomorphic to H p. (3.q p. E) (3. . of type (p. If (Φ1 .2 (E) − · · · ∂ ∂ ∂ (3. Ap.q (M. We now consider differential forms on M with values in E.0 (E) − Ap. then O(U )k is an isomorphism. Since M is complex.q (M.q (M. . a differential form of type (p. We define the ∂-operator on differential forms with values in E by ∂α := (∂ϕj ) ⊗ Φj . E) = p+q=r ϕ → ϕµ Φµ ∈ O(U. 3.3 Holomorphic Vector Bundles Let E → M be a holomorphic vector bundle. The cohomology of this complex is denoted H p. E) and called Dolbeault cohomology of M with coefficients in E.q (M. E).1) where A (M. E) − Ap. 0)). E). E) the space of holomorphic sections of E over U . . (3.4) Since the transition maps between holomorphic frames of E are holomorphic. E) are called Hodge numbers of M with respect to E.1 (E) − Ap. E) := dimC H p.2) Ar (M. E) = A (M.∗ (M.q (M.5) is a cochain complex. → → → 0 → Ωp (E) → Ap. ∂α is of type (p. E) − Ap. q). E). C)⊗E.q+1 (M. By definition. Φk ) is a holomorphic frame of E over U . q + 1). We have ∂∂ = 0 and hence · · · − Ap.q (M.q where the coefficients ϕj are complex valued differential forms of type (p. Let Ωp (E) be the sheaf of germs of holomorphic differential forms with values in E and of degree p (that is.q (E) be the sheaf of germs of differential forms with values in E and of type (p. With respect to a local holomorphic frame Φ = (Φj ) of E.6) . q).3) p.1 Dolbeault Cohomology. it follows that ∂α is well defined.

suppose that M is closed. 3. ∆∂ (∗ ⊗ h) = ∂(∗ ⊗ h)∂(∗ ⊗ h∗ )(∗ ⊗ h) + (∗ ⊗ h)∂(∗ ⊗ h∗ )∂(∗ ⊗ h) = (∗ ⊗ h) ∆∂ .m−q (M. respectively. E) is an isomorphism of vector spaces. we have (∗ ⊗ h∗ )(∗ ⊗ h) = (−1)r on forms of degree r = p + q. q) and (p.30 ¨ Lectures on Kahler Manifolds We are going to use the notation and results from Chapter 1. Here the ∂-operator on the right belongs to the dual bundle E ∗ of E. we get d ((∗ ⊗ h)α) ∧ε β = ∂ ((∗ ⊗ h)α) ∧ε β It follows that = (−1)2m−r {−((∗ ⊗ h∗ )∂(∗ ⊗ h)α.m−q (M. ∂β)} vol.33) and that E is endowed with a Hermitian metric. Suppose that α and β are differential forms with values in E and of type (p.10) (3. We claim that where the Laplace operator on the left belongs to E ∗ .q (M. Then the splitting of forms into types as in (3. In fact. ∆∂ = ∂∂ ∗ + ∂ ∗ ∂ be the Laplace operator associated to ∂.9) ∂ ((∗ ⊗ h)α) ∧ε β = d ((∗ ⊗ h)α) ∧ε β . = (∗ ⊗ h)∂(∗ ⊗ h∗ )∂(∗ ⊗ h) + (∗ ⊗ h)(∗ ⊗ h∗ )∂(∗ ⊗ h)∂ ∆∂ (∗ ⊗ h) = (∗ ⊗ h) ∆∂ . E ∗ ).11) we infer Serre duality. By (3. E ∗ ) for all p and q.36. ∂ ∗ = (∗ ⊗ h∗ )∂(∗ ⊗ h). (3.12) . (3. see Remark 3. We assume that M is endowed with a compatible metric in the sense of (2. Therefore ((∗ ⊗ h)α) ∧ε β is a complex valued differential form of type (m. Ωp (E)) ∼ H m−q (M. q) with coefficients in E. Ωm−p (E ∗ )).q (M. m − q). E) the space of ∆∂ -harmonic forms of type (p.10). (∗ ⊗ h)α is of type (m − p. = H p. E). Let We denote by Hp.q (M.11) In the rest of this subsection. m − 1). ∗ ⊗ h restricts to a conjugate linear isomorphism Hp. Equivalently we can say that ∗ ⊗ h induces a conjugate linear isomorphism H q (M. From (3.q (M.2) is perpendicular with respect to the induced Hermitian metric on A∗ (M. namely that ∗⊗h induces a conjugate linear isomorphism In particular. E ∗ ).m−q (M. E) → H p.q (M. Computing as in (1. we have hp. Then by Hodge theory. By Corollary 2. It follows that Since the real dimension of M is 2m. E) = hm−p. E) → H m−p.46).q (M. E) → Hm−p. (3.8) (3. the canonical projection Hp.13 Remark. q − 1). β) + (α.7.

. Proof. The other direction is similar. E) and D σ ∈ A0. E). τ ) + ∂(σ. .17 Lemma. Let E → M be a holomorphic vector bundle and h = (·. (DX σ. (3. Φν ) = h(θµ Φλ . Dτ ). τ ) = ∂(σ. τ ) = ∂(σ. then D is Hermitian iff (σ. λ λ dhµν = dh(Φµ . τ ) + (σ. Then there is precisely one connection D on E such that 1) D is Hermitian and 2) D = ∂.16 Lemma. where we use D σ = D τ = 0. Φk ) be a holomorphic frame of E over an open subset U ⊂ ν M and θ = (θµ ) be the corresponding connection form as in (3. . Proof. where σ = ϕµ Φµ in terms of a local holomorphic frame (Φµ ) of E. D τ ) = ∂(σ. θν Φλ ) λ λ λ λ = θµ hλν + hµλ θν = hνλ θµ + hµλ θν . τ )(X) or. By Property 1). 3. Let D be a connection on E and σ a smooth section of E.0 (M. D τ ) = (Dσ. µ D σ + D σ = Dσ = dϕµ Φµ + ϕµ DΦµ = ((∂ + ∂)ϕµ + ϕν θν )Φµ . equivalently. Ck×k ). . If this equality holds.0 (U.2 Chern Connection.15) ν ν ν where θ = (θµ ) is the connection form and θµ Φν is shorthand for θµ ⊗ Φν . ·) be a smooth Hermitian metric on E. . . Then ν DΦµ = θµ Φν . If D = ∂. τ ) for the equality in the lemma. Since E has local holomorphic frames. DX τ ) = ∂(σ. Let (Φ1 . Then the connection form θ ∈ A1. Proof. τ ) = (D σ. 3. Let (Φ1 .3 Holomorphic Vector Bundles 31 3. ·) be a smooth Hermitian metric on E. 0). In the sense of 1-forms we write (σ. . Φν ) + h(Φµ . Φk ) be a holomorphic frame of E over an open subset U ⊂ M . τ )(X) for all vector fields X on M and local holomorphic sections σ.1 (M. Now 2) implies that θ is of type (1. D σ = 0 for all local holomorphic sections σ of E iff D = ∂. then d(σ. we conclude that the equality in the lemma implies that D is Hermitian.18 Theorem. Let h = (·. .14) with D σ ∈ A1.15). Then we can split Dσ = D σ + D σ (3. τ ) + (σ. The assertion follows from comparing types in 3. . τ of E.

Let f : N → M be a holomorphic map. Φk ) be a local holomorphic frame of E and θ and Θ be the corresponding connection and curvature form. 3. 4) Θ = ∂θ and Θ is of type (1. applied to the trivial bundle E = M × Ck . Existence follows from the fact that the local connection forms θ = h−1 ∂h above transform correctly under changes of frames.32 ¨ Lectures on Kahler Manifolds Now dhµν = ∂hµν + ∂hµν . 1) Let E → M be a holomorphic vector bundle and h be a Hermitian metric on E. Another argument for the existence is that for local holomorphic sections σ and τ of E. where σ is a section of E and σ is perpendicular to E . .20 Exercises. 3. 0) and ∂θ = −θ ∧µ θ. 2) D = ∂ + θ and D = ∂. Let E → M be a holomorphic vector bundle with Hermitian metric h and corresponding Chern connection D. Since θ is of type (1. It depends on the choice of a Hermitian metric on E. 1).15) and the discussion in Example 1. Recall that this is the standard recipe of getting a Hermitian connection for a subbundle of a Hermitian bundle with a Hermitian connection. Then the Chern connection D of E with respect to h is given by D σ = (Dσ) .16. (3. In what follows. . This shows that Properties 1) and 2) determine D uniquely. .21 Proposition. Let D be the Chern connection on E with respect to h. we use the wedge product as defined in (1. 0). Let h be a Hermitian metric on E and h be the restriction of h to E . τ ) is conjugate O-linear in σ so that the equation in Lemma 3. ∂(σ. . Let (Φ1 . 2) Let E → M be a holomorphic vector bundle and E → M be a holomorphic vector subbundle of E.18 will be called the Chern connection.19) by comparison of types. For a section σ of E write σ = σ + σ . Then: 1) θ = h−1 ∂h with h = (hµν ). we conclude λ ∂hµν = hµλ θν or θ = h−1 ∂h. 5) ∂Θ = 0 and ∂Θ = Θ ∧λ θ = Θ ∧µ θ − θ ∧µ Θ. The unique connection D satisfying the properties in Theorem 3. Then the Chern connection of the pull back Hermitian metric f ∗ h on the pull back f ∗ E → N is the pull back of the Chern connection on E with respect to h. 3) θ is of type (1. .17 leads to the determination of the yet undetermined D .

n := Gr Cn and E → M be the tautological bundle as in Example 2. Then there is a holomorphic frame (Φ1 .18. . We define a new holomorphic frame ˜ Φ µ = Φµ + z i a ν Φ ν µi ˜ with aν = −(∂hνµ /∂z i )(0). . we choose a holomorphic frame (Φ1 . The first assertion follows from the proof of Theorem 3. Let E → M be a holomorphic vector bundle with Hermitian metric h and Chern connection D. Proof of Proposition 3. a local nowhere vanishing holomorphic section Φ.21. .2. As for the third assertion. by the Bianchi identity dΘ = Θ ∧λ θ. the second from Lemma 3. Proposition 3. Then a holomorphic frame over an open subset = −(h−1 ∂h) ∧µ (h−1 ∂h) = −θ ∧µ θ.21 gives θ = ∂ ln h and Θ = ∂∂ ln h. . 3. 2) Θ(0) = ∂∂h(0). It is easy to check that h = 1 + O(|z|2 ). (3.23 Proposition.3 Holomorphic Vector Bundles 33 In the case of a holomorphic line bundle E → M . Φk ) of E about p0 such that 1) h(z) = 1 + O(|z|2 ). Φk ) about p0 such that hµν (0) = δµν . . Proof. We identify sections of E over an open subset U of M with maps σ : U → Cn such that σ(p) ∈ p for all p ∈ U . .5. Hence ∂Θ = 0 and. . µi 3. we have ∂θ = ∂(h−1 ∂h) = −(h−1 ∂hh−1 ) ∧µ ∂h In particular. Suppose 1) holds. The standard Hermitian inner product on Cn induces a Hermitian metric on E. we conclude that Θ ∧λ θ = dΘ = ∂Θ + ∂Θ = ∂Θ. . Let p0 ∈ M and z be holomorphic coordinates about p0 with z(p0 ) = 0. and h = (Φ. Let M = Gr. Θ = ∂θ + ∂θ + θ ∧µ θ = ∂θ. hence 1) implies 2). Then Θ(0) = ∂θ(0) = (∂h−1 ) ∧µ ∂h + h−1 ∂∂h (0) = ∂∂h(0).16. To show 1). Φ). .22) We will use these formulas without further reference.24 Example (Tautological bundle).

choosing p0 as above means no restriction of generality. . Φr (p)) is a basis of p. Consider the plane p0 spanned by the first r unit vectors. where πp : Cn → Cn is the orthogonal projection of Cn onto p. that is. we have Θ(0) = ∂∂h(0) = ∂z t ∧ ∂z = dz t ∧ dz. In z = 0. hµν = (Φµ . such that (Φ1 (p). .34 ¨ Lectures on Kahler Manifolds U ⊂ M is given by holomorphic mappings Φµ : U → Cn . dσp (Xp ) need not be an element of p anymore. for all p ∈ U . For such a frame. We obtain the Chern connection D of E by setting (DX σ)(p) = πp (dσp (Xp )). in p0 . Then we get a holomorphic parameterization (the inverse of holomorphic coordinates) of M about p0 by C(n−r)×r z→ 1 ∈ M. ν In particular. In particular. . µ If σ : U → Cm is a section of E and X is a vector field of M over U . For this frame we have hµν = δµν + λ λ λ zµ zν = δµν + (z t z)µ . . h = 1 + O(|z|2 ). the r columns of the matrix ( z ) are a holomorphic frame of E. Each element from U(n) acts biholomorphically on M and E. z 1 Moreover. The canonical action of the group U(n) on M is transitive and extends canonically to an action on E which leaves h invariant. compare Exercise 3. in homogeneous coordinates in Cn×r written as p0 = 1 ∈ M. that is.20. then the derivative dσ(X) of σ in the direction of X need not be a section of E. Φν ) = λ ¯µ ν ¯ Φλ Φλ = (Φt Φ)µ ν with Φ = (Φν ).2 (where the ambient bundle is M × Cn in our case). . 0 where 1 = 1r is the r × r unit matrix. 1 ≤ µ ≤ r.

3 Holomorphic Vector Bundles

35

3.3 Some Formulas. For the following, we refer again to notation and results introduced in Chapter 1. Let E → M be a holomorphic vector bundle over M . Let h be a Hermitian metric on E and D be the corresponding Chern connection. Then we have the associated exterior differential dD : Ar (M, E) → Ar+1 (M, E), see (1.18) and (1.21). Let (Φµ ) be a local holomorphic frame of E and α = ϕµ Φµ (3.26) (3.25)

be a differential form with values in E of type (p, q) with p + q = r. By the characteristic property of the Chern connection, we have dD α = dϕµ ⊗ Φµ + (−1)r ϕµ ∧ DΦµ = ∂ D α + ∂α, where ∂ D α := ∂ϕµ ⊗ Φµ + (−1)r ϕµ ∧ DΦµ . (3.27) (3.28)

Note that ∂ D α and ∂α are of type (p + 1, q) and (p, q + 1), respectively. 3.29 Proposition. Let E be a holomorphic vector bundle over M with Hermitian metric and associated Chern connection D. Then, for any differential form α with values in E, (∂ D )2 α = (∂)2 α = 0 and (∂ D ∂ + ∂∂ D )α = (dD )2 α = RD ∧ε α,

Proof. If α is a differential form with values in E of type (p, q), then dD dD α = RD ∧ε α, see Proposition 1.19. By Proposition 3.21, RD is of type (1, 1), hence dD dD α is of type (p + 1, q + 1). By definition, dD dD α = ∂ D ∂ D α + ∂∂α + (∂ D ∂ + ∂∂ D )α. The first two forms on the right hand side vanish since they have type (p + 2, q) and (p, q + 2), respectively. We assume now that, in addition, M is endowed with a compatible Riemannian metric as in (2.33). Let (X1 , Y1 , . . . , Xm , Ym ) be a local orthonormal frame of M with JXj = Yj . Set Zj = 1 1 (Xj − iYj ) and Zj = (Xj + iYj ), 2 2 (3.30)

where ε : End(E) ⊗ E → E is the evaluation map.

∗ ∗ ∗ ∗ ∗ and let Z1 , . . . , Zn , Z1 , . . . , Zn be the corresponding dual frame of TC M . With our conventions, we have

Zj := Zj , · =

1 ∗ Z 2 j

and Zj := Zj , · =

1 ∗ Z . 2 j

(3.31)

36

¨ Lectures on Kahler Manifolds

Since Xj = Zj + Zj and Yj = i(Zj − Zj ), we get
∗ Xj =

1 ∗ i ∗ ∗ ∗ (Z + Zj ) and Yj∗ = − (Zj − Zj ) 2 j 2

∗ ∗ ∗ for the dual frame (X1 , Y1∗ , . . . , Xm , Ym ) of (X1 , Y1 , . . . , Xm , Ym ). Therefore,

dD = =

1 2

∗ ∗ ˆ (Zj + Zj ) ∧ DZ ∗ Zj

j +Zj

ˆ ∧ DZj +

∗ Zj

ˆ ∧ DZ

∗ ∗ ˆ + (Zj − Zj ) ∧ DZ

j −Zj

j

=∂

D

+ ∂,

(3.32)

meaning that the sum of the first terms is equal to ∂ D and the sum of the second to ∂. To compute the adjoint operator (dD )∗ of dD with respect to the induced Hermitian metric as in (1.25), we note first that the relations in Exercise 1.26 ∗ ∗ also hold for complex tangent vectors. Since Zj = 2Zj and Zj = 2Zj , we get (dD )∗ = −2 ˆ ˆ Zj DZ + Zj DZj = (∂ D )∗ + ∂ ∗ ,
j

(3.33)

where we use the first relation from Exercise 1.26 for the first equality and comparison of types for the second. The Laplace operators associated to ∂ and ∂ D are defined by ∆∂ = ∂∂ ∗ + ∂ ∗ ∂ and ∆∂ D = ∂ D (∂ D )∗ + (∂ D )∗ ∂ D , (3.34)

respectively. Both preserve the type of forms. Using the second relation from Exercise 1.26, a straightforward computation gives the following Weitzenb¨ck o formulas, ∆∂ α = −2 = −2 ∆∂ D α = −2 = −2 ˆ D2 α(Zj , Zj ) + 2
j j,k ∗ ˆ Zj ∧ (Zk (RD (Zk , Zj )α)) ∗ ˆ Zj (Zk ∧ (RD (Zk , Zj )α)), ∗ ˆ Zj (Zk ∧ (RD (Zk , Zj )α)) ∗ ˆ Zj ∧ (Zk (RD (Zk , Zj )α)),

ˆ D2 α(Zj , Zj ) + 2
j j,k

(3.35)

ˆ D2 α(Zj , Zj ) + 2
j j,k

D α(Zj , Zj ) + 2
j j,k

ˆ2

(3.36)

ˆ where the frame (Z1 , Z1 , . . . , Zm , Zm ) is as in (3.30) and where RD denotes the ˆ on A∗ (M, E). curvature tensor of D 3.37 Exercises. 1) Show that, for a differential form ϕ ⊗ σ of degree r, ∂ D (ϕ ⊗ σ) = ∂ϕ ⊗ σ + (−1)r ϕ ∧ D σ, ∂(ϕ ⊗ σ) = ∂ϕ ⊗ σ + (−1)r ϕ ∧ D σ.

3 Holomorphic Vector Bundles

37

2) Prove Formulas 3.35 and 3.36 and show that 4 4 ˆ ˆ D2 α(Zj , Zj ) = tr D2 α + i ˆ ˆ D2 α(Zj , Zj ) = tr D2 α + i ˆ RD (Xj , JXj )α, ˆ RD (JXj , Xj )α.

3.4 Holomorphic Line Bundles. The results in this subsection will be mainly used in Section 9. We continue to assume that M is a complex manifold. Let E → M be a complex line bundle. Let (Uα ) be an open covering of M such that, for each α, there is a nowhere vanishing smooth section Φα : Uα → E. If σ is a smooth section of E, then we can write σ = σα Φα over Uα , where σα : Uα → C is a smooth function, the principal part of σ with respect to Φα . Over intersections Uα ∩ Uβ , we have Φβ = tαβ Φα , where the transition functions tαβ : Uα ∩ Uβ → C∗ = C \ {0} are smooth. For the principal parts of a section σ as above we get σα = tαβ σβ . (3.38) The family (tαβ ) of transition functions satisfies tαα = tαβ tβα = tαβ tβγ tγα = 1. (3.39)

We say that with respect to the given covering, (tαβ ) is a 1-cocycle of smooth functions with values in C∗ . Suppose E is another complex line bundle over M , and suppose there is an isomorphism, F : E → E . After passing to a common refinement if necessary, suppose that we are given, for each α, a nowhere vanishing smooth section Φα : Uα → E . As above, we have smooth transition functions (tαβ ) with Φβ = tαβ Φα . Over each Uα , we also have F (Φα ) = sα Φα , where sα : Uα → C∗ is smooth. Comparing coefficients we get tαβ = sα tαβ s−1 . β (3.40)

By definition, this means that the two cocycles (tαβ ) and (tαβ ) are cohomologous. Vice versa, suppose (Uα ) is an open covering of M and (tαβ ) a C∗ -valued 1cocycle of smooth functions with respect to (Uα ). On the set of triples (α, p, v) with p ∈ Uα and v ∈ C set (α, p, v) ∼ (β, q, w) iff p = q and v = tαβ w. By (3.39), ∼ is an equivalence relation. The set E of equivalence classes [α, p, v] admits a natural projection to M , π([α, p, v]) = p. The fibers π −1 (p), p ∈ M , are complex lines. As in the case of the universal bundles over complex Grassmannians, it follows that there is a unique topology on E such that π : E → M is a smooth complex line bundle with nowhere vanishing smooth sections Φα : Uα → E, Φα (p) = [α, p, 1]. (3.41)

.38 ¨ Lectures on Kahler Manifolds If another cocycle tαβ is cohomologous to the given one. 1) The tautological line bundle U = U1..45 Examples.3.24: Over Ui .3. We consider some line bundles over CP m . In this way we associate to H a holomorphic line bundle E together with a holomorphic section f such that H = {p ∈ M | f (p) = 0}. 3. zi zi zi .43) over intersections Uα ∩Uβ . Let (Uα ) be an open covering of M with defining holomorphic functions fα : Uα → C for H. 3. E ∗ ). More generally. see for example Chapitre V in [Wei]. (3. we obtain the corresponding results for holomorphic line bundles. tαβ = sα tαβ s−1 for a β family of smooth functions sα : Uα → C∗ .42) (3. z0 zi zm Φi = . O∗ ). Replacing the word ‘smooth’ in the above discussion by the word ‘holomorphic’. Then we have fα = tαβ fβ H ∩ Uα = {p ∈ Uα | fα (p) = 0}. The family (tαβ ) is a 1-cocycle of holomorphic functions and. We will use the open covering of CP m by the sets Ui = {[z] ∈ CP m | zi = 0}.41) are holomorphic and such that the section f of E with f |Uα = fα Φα is well-defined and holomorphic.. where O∗ is the sheaf of germs of holomorphic functions with values in C∗ and where the group law in Pic(M ) corresponds to the tensor product of line bundles. and Corollary 5..7 in [Hu].. compare also Examples 2. therefore. Note that the tensor product L ⊗ L∗ of a holomorphic line bundle L over M with its dual bundle L∗ is holomorphically isomorphic to the trivial bundle so that the dual bundle corresponds to the inversein Pic(M ).. In particular. then F (Φα ) := sα Φα gives rise to an isomorphism F : E → E . such that dfα (p) = 0 for all p ∈ H ∩ Uα .2. Section 1.4.1 in [GH]...44 Remark. we have established that the space of isomorphism classes of smooth complex line bundles over M is naturally isomorphic to H 1 (M. Proposition 4. where E ∗ is the sheaf over M of germs of smooth functions with values in C∗ . 0 ≤ i ≤ m.5 and 3.13. In the language of sheaves. there are one-to-one correspondences between so-called effective divisors and pairs consisting of a line bundle together with a holomorphic section (up to isomorphism) respectively divisors and pairs of a line bundle together with a meromorphic section (up to isomorphism). defines a holomorphic line bundle E as above such that the sections Φα from (3. or Section 2. Suppose H is a complex hypersurface. There is an important connection between complex hypersurfaces and holomorphic line bundles.. where the functions tαβ : Uα ∩Uβ → C∗ are holomorphic. the space of isomorphism classes of holomorphic line bundles over M is naturally isomorphic to the Picard group Pic(M ) := H 1 (M.m → CP m .

z0 = 0. 1 ≤ i ≤ m. Over intersections Ui ∩ Uj . H is inverse to U . 3) A hyperplane in CP m is determined by an m-dimensional vector subspace V ⊂ Cm+1 .4.3 Holomorphic Vector Bundles 39 is a nowhere vanishing holomorphic section of U over Ui . also denoted by H. Over Ui ∩ Uj . [w]) = zi . To that end. H is defined by the equation fi ([z]) := Over Ui ∩ Uj . that is. ˜ of M and corresponding defining holomorphic functions fi : Wi → C of S. zi and hence K is isomorphic to U m+1 . zi zi zj m+1 m+1 ωi = zj ωj . Over Ui . σi = zj 2) The canonical line bundle K = Am. H ⊗ U is the trivial bundle. zi For the holomorphic line bundle over CP m associated to H. a section σ of U is of the form σ = σi Φi . [w]) ∈ V | wi = 0}. where the hat indicates that the corresponding term is to be deleted. zi zj fj . we consider the open covering ˜ of M consisting of the subsets fi = ˜ W0 = M \ S. Over intersections Ui ∩ Uj . f0 = 1 and fi (z. . let wij = zj /zi and set ωi = (−1)i dwi0 ∧ · · · ∧ dwii ∧ · · · ∧ dwim . the (m+1)-fold tensor product U ⊗· · ·⊗U . Then ωi is a nowhere vanishing holomorphic section of K over Ui . the functions fi serve as principal parts of a holomorphic section f of H which vanishes along the given hyperplane. we have zi σj . we have zi Φi = zj Φj . where σi is a function called the principal part of σ with respect to Φi .0 (CP m . C) → CP m : Over Ui as above. Over Ui as above. We choose the hyperplane H = {[z] ∈ CP m | z0 = 0}. ˜ 4) Let M be a complex manifold of complex dimension m and M be the blow up of M at a point p ∈ M as in Subsection 2. It follows that in our case. 1 ≤ i ≤ m. We use the notation ˜ introduced there and describe the holomorphic line bundle L → M determined −1 by the hypersurface S = π (p). wik = Therefore zk zj zj · = · wjk zj zi zi and d 2 zj zi zj =− 2d . Wi = {(z.

σ(z.4. . ui−1 . . . . the differential of π induces an isomorphism. we view dz1 ∧ · · · ∧ dzm also as a nowhere vanishing section of π ∗ K.4. Proof. Let L → M be the holomorphic line bundle as in Example 3. Moreover. 3. or.40 ¨ Lectures on Kahler Manifolds These functions transform under the rules zj f0 = fj and zj fi = zi fj .41).4.46 Lemma. The functions fi define a global holomorphic section f of L with S as its set of zeros. zi gi = zj gj . [w]) = [w]. zi . see Example 3.0 (M. Let L∗ be the dual of L. C) ˜ are related by of M ˜ = K ∼ π ∗ K ⊗ Lm−1 . Hence we have m−1 m−1 zi Ψi = zj Ψj over Wi ∩ Wj . The canonical bundles K = Am. consider the map σ : V → CP m−1 . also denoted π ∗ . ui+1 . . Since dzj = uj dzi + zi duj . Hence σ ∗ H is inverse to L|V .45. C) of M and K = Am. where Φi is a nowhere vanishing holomorphic section of L over Wi as in (3. The functions gi transform according to the rule wi gi = wj gj . um ) are holomorphic coordinates on Wi .45.47 Lemma. . ˜ is a nowhere vanishing section of K over Wi . i ≥ 1. Over V . Let H → CP m−1 be the hyperplane bundle. . In the notation of Subsection 2. = Proof. Ψ0 = dz1 ∧ · · · ∧ dzm vanishes nowhere over W0 ∩ Wj and m−1 Ψ0 = zj Ψj . as above. Similarly. equivalently. Hence Ψi = du1 ∧ · · · ∧ dui−1 ∧ dzi ∧ dui+1 ∧ · · · ∧ dum 1−m Ψi = zi dz1 ∧ · · · ∧ dzm . Then (u1 . .3. let uj = wj /wi = zj /zi . We have f = fi Φi . Then the defining equations of H over Wi are gi = w1 /wi = 0. Realize H as the line bundle over CP m−1 associated to the hyperplane {w1 = 0}. For Wi . ˜ Now L is trivial over W0 = M \ S with nowhere vanishing section Φ0 as above. ˜ Let M be a complex manifold of complex dimension m and M be the blow ˜ up of M at a point p ∈ M as in Subsection 2. . Then L∗ |V ∼ σ ∗ H. ˜ ˜ 3.0 (M.

i ≥ 1. hence define an isomorphism over M . these isomorphisms agree on the intersections ˜ Wi ∩ Wj . By the choice of sections Ψi . Similarly. we 0 m−1 obtain an isomorphism over Wi . Hence over W0 .3 Holomorphic Vector Bundles 41 ˜ between K restricted to M \ {p} and K restricted to W0 . . we obtain an isomorphism as desired by sending π ∗ σ to (σ◦π)⊗Φm−1 . by sending Ψi to (dz1 ∧· · ·∧dzm )⊗Φi .

We also have gjk = gjk . ω = igjk dz j ∧ dz k . With dz j we finally get g = gjk dz j dz k . Zk = 0 (4. (4. We can retrieve g from ω. (4. 9 K¨hler’s original article [K¨1] contains much of what we say in this chapter. Y ) = Re h(iX.6) (4. Zk = Zk . Y ). Zk = 0. and gjk = Zj . JY ).3 Remark. Y ) = Re h(JX. we have gjk = Zj .4 K¨hler Manifolds a Let M be a complex manifold with complex structure J and compatible Riemannian metric g = ·. Y ) is called the associated K¨hler form. 4.7) dz k := dz j ⊗ dz k + dz k ⊗ dz j . Y ) = ω(X. The alternating 2-form ω(X. Y )) = Im h(X. Y ) := g(JX.13).33). and let h be a Hermitian metric on T M . a g(X. Y ) = Re(−ih(X.2) (4. · as in (2.1) We say that g is a K¨hler metric and that M (together with g) is a K¨hler a a manifold if ω is closed9 . Then g = Re h is a compatible Riemannian metric on M and Im h is the associated K¨hler form: a g(JX.4) (4. then h = g + iω is a Hermitian metric on T M . Compare a a also Bourguignon’s essay in [K¨2] or his review of K¨hler’s article in [Bo2]. a a gjk = Zj . Vice versa. a In terms of holomorphic coordinates z on M and the frames introduced in (2. Zj = gkj .8) We call the matrix (gjk ) the fundamental matrix of g (with respect to the given coordinates). if g is a compatible Riemannian metric on M and ω is the associated K¨hler form. View T M together with J as a complex vector bundle over M .5) (4.

42. B. C) is holomorphic.n . covering spaces of K¨hler manifolds and quotients of K¨hler manifolds a a by properly discontinuous and free group actions by holomorphic and isometric transformations are again K¨hler manifolds.12) We show first that ω does not depend on the choice of Z. the induced metric on the complex torus Γ\Cm is a K¨hler metric. Let Gr.10 Examples. . and hence ω is a smooth form of type (1. Xk + iYk 4 1 1 = Xj . 1) on U by ω = i∂∂ ln det(Z t Z). ω = dz j ∧ dz j .n be the canonical projection.83. More generally. C) acting on M ∗ on the right by matrix multiplication. and B. . a 5) The following construction yields the Fubini–Study metric in the case of complex projective spaces. This shows that ω is well defined independently of Z. . a holomorphic principal bundle with structure group Gl(r. 2 2 For any lattice Γ ⊂ Cm . Xk − iω(Xj . a 2) Products of K¨hler manifolds (endowed with the product complex struca ture and the product metric) are K¨hler manifolds. Let Z be a holomorphic section of π over an open subset U ⊂ Gr.4.¨ 4 Kahler Manifolds 43 4. and then ∂∂ ln det(F t Z t ZF ) = ∂∂ ln det(Z t Z) since ln det F is holomorphic and ln det F t antiholomorphic. Let M ∗ ⊂ Cn×r be the open subset of matrices of rank r and π : M ∗ → Gr.n be the Grassmannian of r-planes in Cn . any other choice is of the form ZF .9 Remark. 2 2 where the hjk are the coefficients of the fundamental matrix of h with respect to the local complex frame (X1 . a 4) When endowed with the induced complex structure and Riemannian metric.n . In fact. Define a closed form ω of type (1. compare Examples 2.2. Then g is a K¨hler metric with i 1 g= (4. . a 4. In terms of holomorphic coordinates z as above and the usual notation. where F : U → Gl(r. if N → M is a holomorphic a immersion. (4. Xk ) = hjk . 1) M = Cm with the Euclidean metric g. we have gjk = 1 Xj − iYj . . a 3) A complex submanifold N of a K¨hler manifold M is K¨hlerian with a a respect to the induced Riemannian metric since the K¨hler form on N is the a restriction of the K¨hler form of M .11) dz j dz j .1. then N is K¨hlerian with respect to the induced Riemannian metric. 1) on all of Gr. Xm ) of T M . The defining formula for ω shows that ω is closed.

4. Y ) = ω(X. see Example 2. z] ∈ CP n−1 | z 0 = 0}. We recall the holomorphic action of Gl(n. (4. a . In the case 1 < r < n − 1. [Z] = π(Z).13) j dwj ∧ dwj − j wj dwj ∧ wj dwj . in the case r = 1.2.n a with K¨hler form ω and the natural action of U(n) on Gr. The r-plane spanned by Z ∈ M ∗ is denoted [Z]. To that end we let Z be a holomorphic section of π over the open subset U ⊂ Gr.n . a By our discussion above it remains to show that g is positive definite. C).44 ¨ Lectures on Kahler Manifolds In the case of complex projective space.15 are instructive.n . explicit formulas are more complicated. the plane spanned by the first r unit vectors 1 in Cn . Then A−1 Z ◦ λA is a section of π over λ−1 (U ) and A λ∗ ω = iλ∗ (∂∂ ln det(Z t Z)) A A = i∂∂(λ∗ (ln det(Z t Z))) A = i∂∂(ln det((Z t A)(A−1 Z)) ◦ λA ) = ω since λA is holomorphic and A is unitary. Since the action of U(n) is transitive on Gr. We want to show now that g(X. that is. JY ) is a K¨hler metric on Gr. z/z 0) is a section of π over U and hence ω = i∂∂ ln det(1 + |w|2 ) i (1 + |w|2 ) = (1 + |w|2 )2 (4. Hence ω(0) = i∂∂ ln det(1 + wt w)|w=0 = i∂ tr(((∂wt )w)(1 + wt w)−1 )|w=0 = −i tr dwt ∧ dw = i It follows that g(0) = j.n .n preserves g and ω. j The case n = 2 of complex dimension 1 and the comparison with Formula 4. Then w = (1. About p0 we have holomorphic coordinates [ w ] → w ∈ C(n−r)×r . it suffices to show this in the point p0 . g is a K¨hler metric on Gr. that is. For A in Gl(n.14) and hence g is positive definite. Let U = {[z 0 .k k dwj k dwj .n and g is invariant under this action. = −i tr ∂wt ∧ ∂w j. a straightforward computation gives an explicit formula for ω (and the Fubini–Study metric). C) on Gr.k k k dwj ∧ dwj . we set λA ([Z]) = [AZ] and show next that λ∗ ω = ω for any A in A U(n). where 1 stands for the r × r unit matrix. In conclusion.

n − r) on Gr. is biholomorphic onto the bounded domain D = {w ∈ C(n−r)×r | wt w < 1}.49) is negative definite. Consider the open subset t t M ∗ = {Z ∈ Cn×r | Z0 Z0 > Z1 Z1 } ⊂ Cn×r and the canonical projection π : M ∗ → G− . The stabilizer of the plane p0 ∈ G− spanned by the r. C) acting on M ∗ on the right by matrix multiplication. The map G− → C(n−r)×r . [Z]. that is. Let G− ⊂ Gr.n [Z] → w := W1 .n−r as in (B.n Hermitian form Qr. iff Z0 Z0 > Z1 Z1 in the sense of Hermitian matrices. that is. the r × r unit matrix. The case r = 1 gives complex hyperbolic space. Y ) = ω(X.n t t ω = −i∂∂ ln det(Z0 Z0 − Z1 Z1 ).n The principal bundle π : M ∗ → G− has a global holomorphic section since r.n and is transitive on G− . n − r) and that g(X. a holomorphic principal bundle r. a r. Define a closed form ω of type (1.¨ 4 Kahler Manifolds 45 6) This example is related to the previous one.n−r Z = −Z0 Z0 + Z1 Z1 t t is negative definite. . Let U(r.n Z0 is invertible for all planes in G− : If Z is a local holomorphic section. Write Z ∈ Cn×r as Z= Z0 Z1 with Z0 ∈ Cr×r and Z1 ∈ C(n−r)×r .n first r unit vectors is U(r) × U(n − r). and the K¨hler form ω on D is given by ω = −i∂∂(1 − |w|2 ) i (1 − |w|2 ) = (1 − |w|2 )2 (4. As in the previous example we see that ω is invariant under the action of U(r.n−r . compare again Examples B. D is the unit ball a in Cn−1 .1.n −1 W = ZZ0 does not depend on the choice of Z and W0 = 1.42 and B. then r. Then the natural action of U(r.83. where Z is a local holomorphic section of π. if r = 1. Then Z0 is invertible.n be the open subset of r-planes on which the r. in G− iff r.n with structure group Gl(r. In the case of complex hyperbolic space.15) j dwj ∧ dwj + j. As in the previous example we see that ω is well defined and positive definite. The columns of Z span an r-plane. 1) on G− by r.n r. We leave it as an exercise to compute the induced action of U(r.n leaves G− invariant r.n t t Z t Qr. n − r) be the group of linear transformations of Cn preserving Qr. n − r) on D. r. JY ) is a K¨hler metric on G− .k wj wk dwj ∧ dwk .

. im M In a first step one shows that H(M ) together with the Hermitian product (ϕ. are invariant under biholomorphic transformations of M . . by invariance. 7) Bergmann metric. the case 1 < r < n − 1 is more complicated and the explicit formula for ω in the coordinates w does not seem to be of much use. We refer to [Wei. where we use notation as in (4. In holomorphic coordinates z on an open subset U of M .63) below we conclude that g = − Ric. From (4. . pages 57–65] for details of the following construction. ψ) := im 2 M ϕ∧ψ is a separable complex Hilbert space. hence ω is a real analytic differential form of type (1. More generally. By construction and definition. 1) on M0 . The differential form ω := i∂∂ ln f on U ∩ M0 does not depend on the choice of coordinates. . Hence f = c det G. M0 and θ. write θ = im f dz 1 ∧ · · · ∧ dz m ∧ dz 1 ∧ · · · ∧ dz m . ϕm ∈ H(M ) such that ϕ0 (q) = 0 for all q ∈ U and such that the holomorphic map z : U → Cm given by ϕj = z j ϕ0 defines a holomorphic coordinate chart of M on U . For a unitary basis (ϕn ) of H(D). We may have M0 = ∅. we get the real hyperbolic plane with curvature −2. any biholomorphic transformation F : M → M induces a biholomorphic transformation between M0 and M0 such that the pull back F ∗ θ = θ.20) below. The symmetric bilinear form g(X. this reads im f dz 1 ∧ dz 1 ∧ · · · ∧ dz m ∧ dz m = cim det Gdz 1 ∧ · · · ∧ dz m ∧ dz m . then. As in the previous example. where vol denotes the volume form of g. hence also ω and g. Then f is a non-negative real analytic function on U and f > 0 on U ∩ M0 . where Ric denotes the Ricci tensor of g. In a second step one shows that this sum converges uniformly on compact subsets of M and that θ is a real analytic differential form of type (m.46 ¨ Lectures on Kahler Manifolds For n = 2. Y ) := ω(X. Let M be a complex manifold of dimension m and H(M ) be the space of holomorphic differential forms ϕ of type (m. JY ) is a K¨hler metric on a a M0 with associated K¨hler form ω. set θ := im 2 ϕn ∧ ϕn . 0) on M such that 2 ϕ ∧ ϕ < ∞. m) on M which does not depend on the choice of unitary basis of H(M ). Let M0 ⊂ M be the open subset of points p ∈ M such that there are an open neighborhood U of p ∈ M and ϕ0 . In holomorphic coordinates z. Hence g 2 . If the group of biholomorphic transformations of M is transitive on M0 . θ = c vol for some constant c.

JZ By the Koszul formula and the definition of ω. we have ( X J)Y. JY ). · as in (2. Y.33) and Levi-Civita connection .16 Proposition. Let M be a complex manifold with a compatible Riemannian metric g = ·. JY. X) + Zω(X. then ϕ0 = dz 1 ∧ · · · ∧ dz m and ϕj = z j ϕ0 . Then dω(X. Z = dω(X. 4. JY. we can assume that the vector fields X. Let D be as in the previous example. JZ = −Xω(JY. Z) + Y ω(Z. Y. Then up to a scaling factor. n − r) is transitive on D and preserves both metrics and since the induced action of the stabilizer U(r) × U(n − r) of 0 on the tangent space T0 D = C(n−r)×r is irreducible. Let M be a complex manifold with a compatible Riemannian metric g = ·. Z is an Einstein metric with Einstein constant −1 if the group of biholomorphic transformations of M is transitive on M0 . Z. 2 X (JY ). Y. Z = = X (JY X (JY ). and similarly for dω(X. JY . Proof.33) and Levi-Civita connection . Z . Hence D0 = D. Z) − dω(X. X J)Y. and JZ commute. Then dω(X. Z + JY X. Z − Z X. X + ( Z J)X.17 Theorem. and JZ commute. JZ + Y X. JZ − JZ X. we obtain a K¨hler metric g on a D. Z = X JY. The scaling factor can be determined by the above observation that the Bergmann metric of D is Einsteinian with Einstein constant −1. X Y. + ( Y J)Z. belong to H(D). Z) − JY ω(JZ. JZ). · as in (2. If D is a bounded domain in Cm . JZ). JY . Z. Y. X) + Zω(X. where we use that X. X) − JZω(X. The first equation is now immediate from the definition of ω and the characteristic properties of . Y . JZ) + Y ω(Z. the Bergmann metric of D. Z − J( ). Y ). Z) = Xω(Y. As for the second.¨ 4 Kahler Manifolds 47 4. Z) = ( 2( X J)Y. JY = Xω(Y. Since M is a complex manifold. Then the following assertions are equivalent: = X Y. Y . 1 ≤ j ≤ m. Z + XY ). Y ) and 2 X Y. the Bergmann metric of D coincides with the K¨hler metric defined there since the a action of U(r. Y .

We show next that (2) ⇒ (6). Vice versa. Then JY X= X (JY ) + [JY.0). we have ∂β = 0. there is a smooth real function f in a neighborhood of p0 such that ω = i∂∂f . Let X be a local holomorphic vector field and Y be another vector field on M . there are holomorphic coordinates z centered at p0 such that g(z) = 1 + O(|z|2 ). The Chern connection of the Hermitian metric h on T M as in Remark 4. g is a K¨hler metric. we have ω = dα ¯ locally. if J is parallel. Since ω is of type (1. where α is a real 1-form. 7.18. J = 0. a 2. hence (3) ⇒ (5) by what we just said. 6. then is Hermitian with respect to h. For each point p0 in M . In terms of holomorphic coordinates z. equivalently. 1) for all local holomorphic vector fields of M . The equivalence of (2) with (4)is immediate from d = ∂ + ∂ and the formulas defining ∂ and ∂. The equivalence of the two assertions in (4) follows from barring the respective equation. Then α = β + β.3 is equal to the Levi-Civita connection . where we use that X is automorphic. X]. holomorphic coordia nates as in (7) will be called normal coordinates at p0 . Proof of Theorem 4. (1) and (2) are equivalent. see Proposition 2. Hence J is parallel iff X is of type (0. dω = 0. We show now that (3) is equivalent with (5). It is now obvious that (5) ⇒ (3). 5. The conclusions (6) ⇒ (2) and (7) ⇒ (3) are easy and left as an exercise. . ∂gjk ∂z l = ∂gjl ∂z k . 4.48 ¨ Lectures on Kahler Manifolds 1. 3. where β is a form of type (1. X]) = J XY + J[X. A function f as in (6) will be called a K¨hler potential.10) up to terms of order two and higher.1). For each point p0 in M .17. The existence of normal coordinates shows that a K¨hler manifold agrees with the model example Cm a in (4. we have ∂gjk ∂z l = ∂glk ∂z j or. On the other hand. By definition. Y ]. ¯ ∂β = 0 ¯ and ω = ∂β + ∂ β. The equivalence of (2) and (3) is immediate from Proposition 4. Since ω is real and dω = 0. J Y X = J( XY + [Y.16. X] = X (JY ) + J[Y.

where ϕ is a smooth complex function. T M are parallel subbundles of the complexified tangent bundle TC M = T M ⊗ C with respect to the canonical complex linear extension of to TC M . ω m = im m! det G dz 1 ∧ dz 1 ∧ · · · ∧ dz m ∧ dz m ∗ ∗ ∗ = m!X1 ∧ Y1∗ ∧ · · · ∧ Xm ∧ Ym ∗ ∗ ∗ ∗ = im 2−m m!Z1 ∧ Z1 ∧ · · · ∧ Zm ∧ Zm = m! vol. Zj ) is the associated complex frame as in (3. (X1 . 4.17. Applying (4) we get that g (˜) = 1 + O(|˜|2 ). . 2) It is immediate that J = 0 iff T M. This generalizes Criterion 3 of a Theorem 4. 1) Let J be an almost complex structure on a manifold M and g be a compatible Riemannian metric on M .2. . ˜ ˜ ˜ 2 kl where Aj = − kl ∂glj ∂z k (0).19) where (z 1 . If J is parallel with respect to the Levi-Civita connection of g. . We solve for new holomorphic coordinates z such that ˜ 1 z j := z j + Aj z k z l .1 K¨hler Form and Volume. Then g(z) = 1+O(|z|). ω = igjk dz j ∧ dz k = ∗ Xj ∧ Yj∗ = i 2 ∗ ∗ Zj ∧ Zj . Xm . . In turn. see Exercise 2. . (4.20) where G = (gjk ) and where vol denotes the volume form of M . Let a M be a K¨hler manifold of complex dimension m with K¨hler form ω. We explain now that (4) ⇒ (7). we get the following equivalent expansions of ω m . (X + iY ) = X + i Y . .6.18 Remarks. . . Let z be holomorphic coordinates centered at p0 such that g(0) = 1. Y1 . One of the most basic features of a K¨hler a a manifold is the intimate connection between its K¨hler form and volume. and (Zj . ˜z z 4.30). z m ) are local holomorphic coordinates. Ym ) is a local orthonormal field with JXj = Yj .¨ 4 Kahler Manifolds 49 ¯ Hence β = ∂ϕ locally. (4. then J is a complex structure and hence M is a K¨hler manifold. . Then β = ∂ϕ and hence ω = ∂∂ϕ + ∂∂ϕ = ∂∂(ϕ − ϕ) = i∂∂f with f = i(ϕ − ϕ). There a a are the following expressions for ω.

for each point p ∈ M and non-zero vector v ∈ Tp M there is a vector w ∈ Tp M such that ω(v. that is. that is.1. It is natural to ask whether there are closed manifolds which carry a symplectic form but not a K¨hler metric. ω) be a closed symplectic manifold of (real) dimension 2m. see Example 5.23 Theorem. Let (M. dim M = 2m. As for the second. then the cohomology class of ω k in H 2k (M. R) = 0 for such k. It follows that S 2m−1 × S 1 does not even carry a symplectic form if m ≥ 2. w) = 0.21 Remark. A manifold together with a symplectic form is called a symplectic manifold.37. the Dara a boux theorem says that a symplectic manifold (M. Since ω is closed and ω m is non-zero at each point of M . In particular. that 1 1 ω m−j . 0 ≤ k ≤ m. integration of ω m over M gives ω m = m! vol(M ) = 0.24 Example. A differential two-form ω on a manifold M is called a symplectic form if dω = 0 and ω is non-degenerate at each point of M .25 Remark.22 Exercise (Compare (5. Compare also our discussion of symplectic Lagrangian spaces in Example B. 2) If N ⊂ M is a compact complex submanifold without boundary of complex dimension k. R) are non-zero. Show that ∗ exp ω = exp ω. Chapter 3]. M This shows the first assertion. For m ≥ 2. i. Let M be a K¨hler manifold as above. the Hopf manifold S 2m−1 × S 1 has vanishing second cohomology. and the K¨hler form is the restriction of a a ω to N .53. Chapter 8] and [Du. Hence integrating ω k over N gives k! times the volume of N . H 2k (M. we note that N with the induced metric is a K¨hler manifold.2 a . Proof.23 applies and shows that H 2k (M. Refera ences for the basic theory of symplectic forms and manifolds are [Ar. The K¨hler form of a K¨hler manifold is symplectic.43)).. R) is non-zero for 0 ≤ k ≤ m. 4. and ω m is non-zero at each point of M .10. then the real dimension of M is even. where = ωcan = dxj ∧ dy j = i 2 dz j ∧ dz j is the K¨hler form of the Euclidean metric on Cm as in Example 4. then the cohomology class of ω k in H 2k (M. a 4. the argument in the proof of Theorem 4. 4. R) and the homology class of N in H2k (M. The answer is yes. ω) admits an atlas of smooth coordinate charts z : U → U ⊂ R2m ∼ Cm such that z ∗ ωcan = ω. ∗ ωj = j! (m − j)! 4. R) = 0.50 ¨ Lectures on Kahler Manifolds 4.e. hence does not carry a K¨hler metric. Vice versa. Evaluation of ω m on the fundamental cycle of M . If ω is a symplectic form on M . a 1) If M is closed.

If N −P is the boundary of a real singular chain. . This shows the assertion in the case k = 1. . Therefore |ω k (v1 . w ∈ W be orthonormal unit vectors with respect to ·. |aj | ≤ 1 with equality iff v2j = ±Jv2j−1 . w = Re(v. .27 implies that N is pluri-minimal in the sense that any complex curve in N is minimal in . . . v). v2k )| = k!|a1 . . On the other hand. v2k ) be a basis of V . Let M be a K¨hler manifold as above. . Then |ω k (v1 . . w ∈ [−1. Let v. .¨ 4 Kahler Manifolds 51 4. . Theorem 4. w) and K¨hler form ω(u. . Hence we are free to choose a convenient basis of V . In the general case we observe first that the restriction of ω to V is given by a skew-symmetric endomorphism A of V . . In fact. . . w) = Jv. Since N − P is the boundary of a real singular chain and ω k is closed. Let V ⊂ W be a a real linear subspace of even real dimension 2k and (v1 . .26 Wirtinger Inequality. v2k ) with A ∈ Gl(2k.27 Theorem. It follows that a complex submanifold N ⊂ M is minimal. . Let P ⊂ M be an oriented submanifold of dimension 2k and boundary ∂P = ∂N . . . we have N ωk − P ω k = ω k (N − P ) = 0. v2k ) of V and real numbers a1 . . . Proof.26 we have ω k = vol N N and P ω k ≤ vol P. ak such that A(v2j−1 ) = aj v2j and A(v2j ) = −aj v2j−1 . . . and equality holds iff P is a complex submanifold. . Then ω(v. . . . . . Let N ⊂ M be a a compact complex submanifold of real dimension 2k with boundary ∂N (possibly empty). v2k )| ≤ k! volV (v1 . that is. . then vol N ≤ vol P with equality if P is also a complex submanifold. . R) to another basis of V . ak | ≤ k! = k! volV (v1 . w) = ±1 iff w = ±Jv. 1] and ω(v. Observe first that both sides of the asserted inequality are multiplied by | det A| if we transform (v1 . by the Wirtinger inequality 4. · . . 4. v2k ) with equality iff V is a complex linear subspace of W . Proof. . Let W be a Hermitian vector space with inner product v. . . v) = Im(u. v2k ) with equality iff V is a complex linear subspace of W . the trace of the second fundamental form S of N vanishes. By what we said above. Hence there are an orthonormal basis (v1 .

since J is parallel. (4. Y ). X) + S(JX. 1 ≤ j ≤ m.33) if [JX. It follows that S is complex linear in Y . Z]. X) − S(X. Y ] = [JX.31) Z U. (4. This can also be seen by the following computation: Let X. Y ] = 0. (4. if [X. which is what we wanted to show. Z V . (X + iY ) = X + i Y . We also have W Z= W Z. JY is tangent to N as well and S(X.32) Since T M and T M are complex subbundles of TC M . (4. they stay parallel with respect to the extension. (4. Zj = ∂ 1 = j ∂z 2 ∂ ∂ +i j j ∂x ∂y .28) where we use that the normal bundle of N is invariant under J.34).35) where z j = xj + iy j . Therefore S(X. (U+iV ) Z := UZ +i V Z. More precisely. 4. We now extend also complex linearly in the other variable.29) The extended connection is symmetric with respect to the complex bilinear extension of the Lie bracket to complex vector fields. Since N is complex. Y be tangent vector fields along N . (X+iJX) (Y + iJY ) = ( XY + Y X) + iJ( XY + Y X) (4. then (X+iJX) (Y − iJY ) = (X−iJX) (Y + iJY ) = 0. hence. Recall the coma plex linear extension of to TC M . JY ) = ( X (JY ))⊥ = (J XY )⊥ = J( XY )⊥ = JS(X. X) = 0. Let M be a K¨hler manifold. Y ] = 0. By (4.52 ¨ Lectures on Kahler Manifolds M . (4. WZ − ZW = [W. we let Zj = ∂ 1 = j ∂z 2 ∂ ∂ −i j j ∂x ∂y . the mixed covariant derivatives vanish. also in X. Furthermore. V = Z U. (4. by symmetry.30) and metric with respect to the complex bilinear extension of the metric of M to TC M . (4.2 Levi-Civita Connection.36) Zj Zk = Z j . V + U. Zk = 0. JX) = S(X.34) With respect to local holomorphic coordinates z of M .

V . The curvature tensor of the complex bilinear extension of to TC M is the complex trilinear extension of the usual curvature tensor R of M to TC M . Y )JZ = JR(X. ∂z j (4.37) All Christoffel symbols of mixed type vanish. (4. (4. 4. Then R satisfies the usual symmetries. Z)X + R(Z.40) . Y )Z = −R(Y. V = − R(X. U .¨ 4 Kahler Manifolds 53 Since T M and T M are parallel.42) R(X. V )X. Y . X)Z. We keep the notation R for the extension. Zl = ∂gkl ∂z j ∂gkl ∂z j = ∂gkl ∂z j . and R(X. for complex vector fields X. Y . Zl = Thus the Christoffel symbols are given by ¯ Γl = Γµ gµ¯ g ν l = jk jk ν ∂gk¯ ν l ν ¯ g . more generally.32 we have Γl = Γl . (4. We also have the reality condition R(X.43) (4. (4. Y )Z = R(X. X)Y = 0. V = R(U. Y )V. U . but now.39) where the coefficients g kl denote the entries of the inverse of the fundamental matrix.41) R(X.31) and (4. Y )Z. Y )Z + R(Y.3 Curvature Tensor. Z. Furthermore. since J is parallel. jk (4. Zj Zk . R(X.38) jk jk By (4. Y )U. R(X. Y )U.36). we get unmixed Christoffel symbols Zj Zk = Γl Z l jk and Zj Z k = Γl Z l . In addition. by Equation 4. Γµ gµl = jk ¯ Γµ gµl = ¯ jk Zj Zk . Y )Z.

we have R(U.49) Recall that the sectional curvature determines the curvature tensor. jkl jkl jkl R µ Zµ . ∂z j (4. (4. X . JX) as a basis over R and the corresponding holomorphic sectional curvature is K(X ∧ JX) = R(X. Since T M and T M are parallel. Y )U.36). V )(T M ) ⊆ T M and R(U.42 implies that ¯ Rµ = Rµ jkl jkl ¯ and Rµ = Rµ .33). X) = 3K(X ∧ JX) · X 4 (4. jkl ¯ R(Zj . V + R(X.52) .51 Proposition. We define the holomorphic sectional curvature of M to be the sectional curvature of the complex lines in T M . JU )JY. and (4. It is immediate from the symmetries of R listed above that Q is symmetric. we get from (4. (4. Zk )Zl = Rµ Zµ . JV )JY. V . R(Zj . jkl µ ¯ (4. Zk ) = 0. Y )JU. U .45) (4.47) (4. U. We follow the proof of Proposition IX.41). Y. X.54 ¨ Lectures on Kahler Manifolds By (2. JV = R(JX. X. R(X. Proof.45) R(Zj . V )(T M ) ⊆ T M.43). X 4 (4. Zk ) = R(Zj .46) R(Zj .1 in [KN] and consider the quadri-linear map Q(X. If X is a non-zero tangent vector of M . For the mixed terms. Zk )Zl = Rµ Zµ .48) By (4.50) 4. Zk )Zl = R jkl Zµ . V = R(X. The holomorphic sectional curvature determines R.44) With respect to local holomorphic coordinates. JX)JX. Thus polarization determines Q explicitly from its values Q(X.7. (4. JY )U. then the complex line spanned by X has (X. V ) = R(X. Zk )Zl = Equation 4. we have Rµ = − jkl ∂Γµ jl ∂z k and Rµ = jkl ∂Γµ kl . JY )JU. V + R(X. R(Zj .

X . JY )JY. X . Y . JXm ) of M . Y )Y. Y. 4. Hence it suffices to show that Q determines R.57 Proposition. Y is defined to be R(X. Y ) = Ric(X. JX)JY. X)Y ). X . X . . JY )JY. JY )JY. Y. Y . By the second symmetry in (4. (4. The bisectional curvature of X. X + R(X. X. JXj )JX. JY. Y ) = tr(Z → R(Z. JY ) − Q(X. Y ). Y − R(JY. Y := Ric(X.54 Exercise. Y ) = 2 R(X. In conclusion. JX)JY.56) The associated symmetric endomorphism field of T M or TC M is also denoted Ric. hence R. JX)JY. X + R(X. X. X + R(X.¨ 4 Kahler Manifolds 55 along the diagonal. JY )X.53) where we can take the trace of the map on the right hand side as an R-linear endomorphism of T M or as a C-linear endomorphism of TC M . we let Ric X. The Ricci tensor is defined by Ric(X. Y ) = 3 R(X. JX1 . that is. .40). X. Hence Q determines the sectional curvature. R(Xj . JY )JY. Y )Y. X. 4. Y = − R(JX. Y. Therefore Q(X. R(X. In terms of an orthonormal frame (X1 .4 Ricci Tensor. Y ). JX)JY. and hence also Q(X. Ric JX = J Ric X. Now Q(X. JY )JY. . X + R(X.55) (4. (4. Y ) = 8 R(X. R(X. the Ricci tensor is given by Ric X = In particular. . Y = R(X. Y )Y. X. Y )Y. The Ricci tensor is symmetric and Ric(X. . JY ) = 3 R(X. JY. X)JX. X + R(X. Y = R(X. X . 3Q(X. Xm . Y )Y. 4. Show that it deserves its name.

JXj )X. (4. X)Xj . Y ). In fact. Y )Ej . We compute Ric(X.63) This formula shows that ρ is closed. X)JY. (4. X)Y. X)Y. JXj ) R(JXj . by (4. Zk ) = Rl =− ∂Γl lk (4. 1).56 Proof. (4.58) R(X. Zk ) = 0. Y ) = = = =− We conclude that ¨ Lectures on Kahler Manifolds R(Xj . Now is a connection . Zk ). Zk ) = Ric(Zj . Y )Ej . Y )Ej . JXj )JX. Ricjk = − For the Ricci form we get ρ = i Ricjk dz j ∧ dz k = −i ∂∂ ln det G. by the above computation of the Ricci curvature we have ρ(X.59) In terms of local holomorphic coordinates. Ej (4. JY + R(Xj . by (4. Y ) = − =i =i R(X. Xj R(JXj . and hence there is an associated real differential form of type (1. Ej + i JR(X.65) h(R(X. where we consider T M with the given complex structure J as a holomorphic vector bundle over M with Hermitian metric h = g +iω.61) With G = (gjk ) we get. JY ) = Ric(X. (4. Ej ) = i tr R(X. jk (4. ∂ ln det G ∂G = tr · G−1 j ∂z ∂z j Therefore.64) „ « = Γk . Y ). Xj + R(Xj . Y )Ej . ∂z j ∂z k (4. Y ). Y . Y )Ej . X)JY.62) ∂ 2 ln det G .47). JEj = R(X. we have Ric(Zj . JXj − R(JXj . Xj )JXj . the Ricci form ρ(X.39). Y ) := Ric(JX. Ej i JR(X. JY R(Xj .60) ljk ∂zj = Ricjk = Ricjk = Ric(Zj . For the mixed terms. JY = Ric(JX. Ric(Zj .

the holonomy group Hol M of M is contained in U(m). the first Chern form of T M associated to is given by 1 i tr R = ρ. a If the holonomy along a closed loop c of M at p is given by A on Tp M .0 (M.68) 4. C) is given by ˆ (4. Y )(dz1 ∧ · · · ∧ dzm ). viewed as a complex vector space. let M be a connected Riemannian manifold of dimension n = 2m and p be a point in M . Jp ) with Cm . that is.67) this holds iff M is Ricci-flat. R). and let p be a point in a M . Hence up to a unitary isomorphism of (Tp M. then the induced holonomy on the fiber Am. the holonomy group Hol M = Holp M of M at p preserves Jp . ˆ R(X. 10 Excellent references for holonomy are [Bes.33) and (A. The induced curvature on the canonical bundle KM = Am.0 (M. Hence Hol(M ) ⊂ SU(m) iff Am. ˆ To prove this. C) p of the canonical bundle is given by the complex determinant detC A. Chapter 10]. namely the Chern connection associated to h. It also shows that [ρ/2π] = c1 (T M ) in H 2 (M. 4.35). This links the K¨hler condition to holonomy10 . [Br]. we use that the curvature R acts as a derivation on A∗ (M. Chapter 3]. ).65).5 Holonomy. R) and hence that [ρ/2π] lies in the image of H 2 (M. We a formulate the result of our discussion in the following somewhat sloppy way. c1 (KM .¨ 4 Kahler Manifolds 57 on T M considered as a complex vector bundle.0 (M. m = dimC M . Now the claim follows from (4.67) R(X. ) = 2π 2π This shows again that ρ is closed. K¨hler manifolds are characterized by the property that their a complex structure is parallel. Then the complex structure Jp on Tp M induced from the complex structure on Cm extends to a parallel complex structure J on M . ˆ ) = − 2π 2π (4. By (4. It also follows that the reduced holonomy group Hol0 (M ) ⊂ SU(m) iff Am. By (A. Then the complex structure Jp turns Tp M into a complex vector space. Vice versa. Y ) · dz1 ∧ · · · ∧ dzm . if Ric = 0. Suppose that after some identification of Tp M with R2m and of R2m with Cm .0 (M. (4. We conclude that the first Chern form 1 i tr R = − ρ = −c1 (T M. and M together with J is a K¨hler manifold. the holonomy group Holp M ⊂ U(m) ⊂ SO(2m).66) c1 (T M. A connected Riemannian manifold of real dimension 2m is a K¨hler manifold iff its holonomy group is contained in U(m). and [Jo. C) is flat. Y )dzj ∧ · · · ∧ dzm = − tr R(X. C) has a parallel complex volume form.69 Proposition. Y )(dz1 ∧ · · · ∧ dzm ) = ˆ dz1 ∧ · · · ∧ R(X. Z) → H 2 (M. Since J is parallel. a Suppose that M is a connected K¨hler manifold. C). Y )(dz1 ∧ · · · ∧ dzm ) = iρ(X. .

we also have (4. M = M0 × M1 × · · · × Mk . Tp0 M0 + Tpi Mi = Fix j=i (4. Then we have the de Rham decomposition. By Proposition 4. and hence Jp A = AJp for all A ∈ Hol M . see Chapters 6 and 7 in [Jo]. Moreover.58 ¨ Lectures on Kahler Manifolds 4. this decomposition is unique up to a permutation of the factors Mi . pk ) ∈ M .74). i ≥ 1. is understood that M0 = {0} and k = 0 are possible. p1 . For examples of Calabi–Yau a a and hyper-K¨hler manifolds. is a simply connected complete K¨hler manifold with irreducible holonomy. we may also have Hol(M ) ⊂ Sp(k) ⊂ SU(m). 4. a 11 There 12 It are variations of this definition in the literature. . Calabi–Yau and hypera K¨hler manifolds are Ricci-flat K¨hler manifolds. . hence M0 is a complex Euclidean space. where M0 is a complex Euclidean space and Mi . If the real dimension of M is 4k. Let M be a simply a connected complete K¨hler manifold. Hence we have. and. By (4. i ≥ 1. .72) implies the de Rham decomposition for K¨hler a manifolds. . for each i ≥ 1.74) Hol Mj . A connected Riemannian manifold M of real dimension 2m with holonomy Hol(M ) ⊂ SU(m) is called a Calabi–Yau manifold11 .73) After identifying Tpi Mi with the corresponding subspace of Tp M . Then the complex a structure J of M is parallel. Hol M = Hol M1 × · · · × Hol Mk . (4. Then a M = M0 × M1 × · · · × Mk . 4. .76 De Rham Decomposition of K¨hler Manifolds. the set of vectors fixed by Hol M . Now the de Rham decomposition as in (4.70. (4. Tp0 M0 = Fix Hol M. a Let M be a simply connected complete Riemannian manifold.75) we conclude that Jp Tpi Mi = Tpi Mi for all i ≥ 1. i ≥ 1.75) Suppose now in addition that M is a K¨hler manifold. is a simply connected complete Riemannian manifold with irreducible holonomy12 . From (4. A connected Riemannian manifold of real dimension 2m is a Ricci-flat K¨hler manifold iff its reduced holonomy group is contained in a SU(m).71 Remark. then M is called a hyper-K¨hler manifold.72) where M0 is a Euclidean space and Mi . we have Jp Tp0 M0 = Tp0 M0 .70 Proposition. at a point p = (p0 .

2. we get 2 ˆ ξ ∈ im ˆ . Suppose now that M is a K¨hler manifold. X is automorphic iff X is a form of type (1. Recall that a vector field X on a complex manifold M is automorphic iff [X. JY ] (4. adding the term dξ to LX g. 0) with values in T M . then LX g = 0 and hence div X = tr LX g/2 = 0. Proof. Let M be Riemannian and X be a vector field on M . The decomposition of the bilinear form ˆ ξ into symmetric and skew-symmetric part is given by ˆ ξ = ( ˆ ξ)sym + ( ˆ ξ)skew = 1 LX g + 1 dξ. Since div X = −d∗ ξ.6 Killing Fields. JY X= =J =J X JY XY for any vector field Y on M . Lichnerowicz [Li1]). 4.5.38: ∗ ∗ X = Ric X.78) X ◦J =J ◦ X. We conclude that LX g = 0 and hence that X is a Killing field. JY ] = J[X. If X is a Killing field. . Since the Levi-Civita connection is equal to the Chern connection of the holomorphic vector bundle T M with the induced Hermitian metric. Then X is a Killing field =⇒ div X = 0 and If M is closed. then the converse holds true as well. In Proposition 2. Y ] − [X.¨ 4 Kahler Manifolds 59 4. Furthermore. 1 X = Ric X ⇐⇒ ˆ ∗ ˆ ξ = ∆d (ξ). It follows that LX g is L2 -perpendicular to im ˆ . That is. 2 Assume now that div X = 0 and ∗ X = Ric X. JY ] Y − [X. In fact. JY ] X + J[X. which is pointwise and hence L2 -perpendicular to LX g. 2 2 The second ingredient is a simple consequence of the Bochner identity 1. we obtain ˆ ∗ ˆ ξ = 1 ˆ ∗ (LX g) + 1 d∗ dξ.8). 2 2 Using that ˆ ∗ = d∗ on alternating two-forms. ∗ X = Ric X by Exercise 1. X is automorphic iff − [X.77 Proposition (Yano [Ya]. We start with a discussion of Killing fields on Riemannian manifolds. Suppose now that M is closed and let ξ = X . see (1. However. Y ] for all vector fields Y on M . we obtain ˆ ∗ LX g = dd∗ ξ = 0.18 we showed that X is automorphic iff X is holomorphic. and let X be a vector field a on M .

which is L2 -perpendicular to 0. Vice versa. 4.1 X. We decompose the one-form X= 1. Then X is automorphic =⇒ ∗ X = Ric X.1 X = 0. But ω is parallel and hence a harmonic form. Hence in order to show that it preserves J as well.79 Proposition (Lichnerowicz [Li2]).1 X= ∗ X − Ric X. Hence 0. an automorphic field is a Killing field iff it is volume preserving. Hence if X is automorphic. then ∗ X = Ric X.77: Suppose that M is closed and that ∗ X = Ric X. Proof.79 we arrive at the final result in this subsection.57. Adding the term 1.0 X. We conclude that 2 ∗ 0. Let M be a closed K¨hler manifold. 1 X ◦ J) + ( X + J ◦ 2 X ◦ J). it is clearly enough to show that it preserves ω. m ≥ 2.0 X into types.60 ¨ Lectures on Kahler Manifolds 4. Comparing Propositions 4. The description of Killing and automorphic vector fields in terms of the equation ∗ X = Ric X is important for the global geometry of compact K¨hler manifolds with Ric ≤ 0 or Ric > 0. Then Killing fields on a M are automorphic. From (4. .77 and 4. then the converse holds as well. that LX ω = 0. (4. The key observation is the following: ∗ (J ◦ X ◦ J) = J ◦ ∗ ( X ◦ J) = − Ric X.1 X is L2 -perpendicular to im . We now repeat the linear algebra argument from the end of the proof of Proposition 4. we get X ∈ im .1 X= 1 ( X −J ◦ 2 This decomposition is pointwise and hence L2 -perpendicular.80) where the first equality holds since J is parallel and the second follows from Proposition 4.1 X = 0. If M is closed. Find non-automorphic Killing fields on Cm . hence 0. That Killing fields on a closed K¨hler manifold M are automorphic can a also be seen in a more conceptual way: A Killing field preserves the metric of M .78) we getthat X is automorphic iff 0.1 X = 0.83 Remark. X+ 0. Then ∗ 0. compare Chapters 6 and a 7.82 Exercise. that is. 4. 4.81 Theorem. Let M be a K¨hler manifold and X a be a vector field on M .

Clearly. R). R) → H ∗ (M. R) as well. for isometries the two actions are compatible with the Hodge isomorphism H∗ (M. LX α = 0 for all harmonic forms α and all Killing fields X. R). but isometries of M act on H∗ (M. It follows that homotopic isometries induce the same maps on H∗ (M. hence connected groups of isometries act trivially on H∗ (M. In particular.¨ 4 Kahler Manifolds 61 Now every diffeomorphism of M acts on H ∗ (M. . R). R).

The complex structure J of M acts on A∗ (M. (5.4) where ω is the K¨hler form of M and (Xj . Since ∗ Zj = 2Zj . E) and their relation with exterior differentiation. W ] = [L∗ .2) ip−q Pp. The Lefschetz map raises types of forms by (1. (5. where we note that C and W leave the type of forms invariant and that L raises and L∗ lowers the type by (1. we have C= The adjoint operator is C∗ = iq−p Pp. The induced endomorphism field of A∗ (M.30). E) can be written uniquely as a sum α = αr . Since the real dimension of M is even. C] = [L∗ . Note that C leaves the type of a form invariant. (5. where αr has degree r.3) i 2 ∗ ∗ Zj ∧ Zj ∧ α.26 13 . we use notation and results introduced in Chapter 1. To compute the adjoint operator L∗ of L we use the first relation in Exercise 1. (5. E) via pull back. we have W = The Lefschetz map is defined by L(α) := ω ∧ α = ∗ Xj ∧ Yj∗ ∧ α = (−1)r Pr . But for the moment this assumption is not necessary since all of our computations are of a local nature.q . In our applications to cohomology further on a we will assume that M is closed. (5.1) and (5.q . Zj ) are frames as in a (3.q : α → αp. Let h be a Hermitian metric on E and D be the corresponding Chern connection. C] = [L. (5.1) We set W := (∗ ⊗ h∗ )(∗ ⊗ h). 1). E). 13 The L2 -adjoint of a field of endomorphisms is the field of adjoint endomorphisms so that there is no ambiguity here. W leaves the type of forms invariant. W ] = 0. q). To that end we recall that each element α of A∗ (M. Yj ) and (Zj .q = C −1 . We consider some endomorphism fields of A∗ (M. where αp. . In what follows. 1). E) is denoted C. The corresponding maps Pr : α → αr and Pp. Let E → M be a holomorphic vector bundle over M . Again. 2 L∗ α = Yj (Xj α) = Zj (Zj α). or as a sum α= αp. In terms of the above projection fields.q define fields of projections of A∗ (M.3).5 Cohomology of K¨hler Manifolds a Let M be a K¨hler manifold.5) i We have vanishing commutators [L.q is of type (p.6) by (5.

12) where v ∈ Vn . where E is a real vector bundle. and Exercise 1. see e. The first terms on the right hand sides of the above formulas coincide. Then the second term on the right hand side in the first formula counts the number of j with j ∈ I and j ∈ J. E). Y ] = −2Y. that is. . Finite dimensional representations of sl2 (C) split into the sum of irreducible ones. H→ (m − r)Pr (5. 1 0 H= 1 0 0 −1 (5. L]. 5.g.5). It follows that X → L∗ . that is. (5. is primitive. R) and corresponding bundle valued forms A∗ (M. endowed with a Riemannian metric. hence cancel in the commutator [L∗ .26 (applied to complex tangent vectors). where the latter are classified by integers n ≥ 0: Vn = Cv + CY v + · · · + CY n v. By (5. L∗ Lα = =− ∗ ∗ Zk (Zk (Zj ∧ Zj ∧ α)) j=k ∗ ∗ Zk (Zj ∧ (Zk (Zj ∧ α))) + j=k ∗ ∗ Zj (Zj (Zj ∧ Zj ∧ α)). [L∗ . Assume now without loss of generality that ∗ ∗ α = ZI ∧ ZJ ⊗ σ. (5.4). v = 0. We note that Hv = −nv. Y ] = H. X] = 2X. also defined on A∗ (M. W . ∗ ∗ Zj ∧ Zj ∧ (Zk (Zk α)) j=k ∗ ∗ Zj ∧ (Zk (Zj ∧ (Zk α))) + j=k ∗ ∗ Zj ∧ Zj ∧ (Zj (Zj α)). / / We recall now that the standard basis vectors X= of sl2 (C) satisfy [X. The second term on the right hand side in the second formula counts the number of j with j ∈ I and j ∈ J. LL∗ α = =− Vice versa. [H. where I and J are multi-indices. Xv = 0. Proof. (5.¨ 5 Cohomology of Kahler Manifolds 63 5. Y → L. L] = (m − r)Pr . The operators C.7 Remark. and L and their adjoints are real. Lecture 11 in [FH].9) extends to a representation of sl2 (C) on (the fibers of) A∗ (M.11) [H. E) and on A∗ (M. and where HY k v = (2k−n)v.10) 0 0 1 .8 Proposition. E). 0 Y = 0 0 .

q (M. ∂ D ] = [L. We now compute commutators with exterior derivatives. E). 1). α = 0.q (M. Since the tensor field L is parallel. E) = ⊕s≥0 L s Ap−s. E) and Ar (M. 5. E) = ⊕s≥0 L r [Ls . and similarly for Ar (M. we have refined decompositions. L∗ ] = s(r − m)Ls−1 .q (M.1 Lefschetz Map and Differentials. and Ap. E) = ⊕s≥0 Ls Ap−s. E) = Ap.19). [L∗ . dD ] = 0. P 5.q (M. Since (m − r)Pr preserves types of forms and since L raises and L∗ lowers types by (1.13 that they are orthogonal as well. A (M. E) → Ar+2s (M. then P P Ls α = 0 for 0 ≤ s ≤ m − r and Ls α = 0 for s > m − r.q Ar−2s (M. Hence its kernel Ar (M.19 Remark.q−s (M. P p.17) (5. E).20) j. E).q (M. (∂ D )∗ ] = i∂. E) and Ap.q (M.8 and what we said about representations of sl2 (C). ∂ D ] = i∂ ∗ . Since any finite dimensional representation of sl2 (C) splits into irreducible ones. Show that on Ar (M. [L∗ . (dD )∗ ] = 0. E) = 0 for r > m. By (3. E).k ∗ ∗ ˆ Zj ∧ Zj ∧ (2Zk DZ α) = k . ∂ ∗ ] = [L∗ . Ar (M.64 ¨ Lectures on Kahler Manifolds 5. E) are parallel. ∂ ∗ ] = −i∂ D . E) and for s ≥ 1. It follows that the parallel tensor field Ls : Ar (M. E) of primitive forms of degree r is a parallel subbundle of P Ar (M. E). if α ∈ Ar (M. P P By Proposition 5. E) = Ap. E) and Ap. the Lefschetz decompositions of Ar (M. Furthermore. E) the space of smooth sections of Ar (M.22 Proposition. E). It is trivial that Ap.q (M. we have Ar (M. P s (5. 5.32) and (4. E). It follows from Exercise 5. Hence the commutators of the adjoint operators also vanish.13 Exercise. E) P (M. E).18) the Lefschetz decompositions of Ap. (5.14) (5. L(∂ D )∗ α = − [L. E). E). (∂ D )∗ ] = [L∗ . We have [L. ∂] = −i(∂ D )∗ .19) that [L. E) = ⊕s≥0 Ls Ar−2s (M. P (5. The tensor field L∗ is parallel and has constant rank on Ar (M. ∂] = [L. [L∗ . E) P if p = 0 or if q = 0.21) (5. We denote by Ar (M.33) and (4. A Ap.16) (5.q (M.15) the Lefschetz decompositions of Ar (M.q−s (M. E). It is immediate from (3. i 2 Proof. E) is injective for 0 ≤ s ≤ m−r and surjective for s ≥ m − r.

From Proposition 3. We have [L. [L. and ∆∂ D and their relation with the Lefschetz map. We now discuss the three Laplacians ∆d . = ∂ D L(∂ D )∗ + (∂ D )∗ L∂ D + i∂∂ D − ∂ D L(∂ D )∗ + i∂ D ∂ − (∂ D )∗ L∂ D . We have ∂ D ∂ ∗ + ∂ ∗ ∂ D = ∂(∂ D )∗ + (∂ D )∗ ∂ = 0. ∆∂ ] = −iRD ∧ε and [L. k ∗ ∗ ∗ ∗ ˆ ˆ {Zj ∧ (Zj (Zj ∧ DZ α)) + Zj (Zj ∧ (Zj ∧ DZ α))} j j =i j ∗ ˆ Zj ∧ DZ α = i∂α. j The proof of the second equation [L. = ∂L∂ ∗ + ∂ ∗ L∂ − i∂ D ∂ − ∂L∂ ∗ − i∂∂ D − ∂ ∗ L∂ Similarly. ∆∂ . ∆∂ D ] = L∂ D (∂ D )∗ + L(∂ D )∗ ∂ D − ∂ D (∂ D )∗ L − (∂ D )∗ ∂ D L = i(∂ D ∂ + ∂∂ D ).k ∗ ∗ ˆ Zk (Zj ∧ (Zj ∧ DZ α)). [L. The remaining two equations are adjoint to the first two. By Equation 5. (∂ D )∗ ]α = i j j. ∆∂ ] = L∂∂ ∗ + L∂ ∗ ∂ − ∂∂ ∗ L − ∂ ∗ ∂L = −i(∂ D ∂ + ∂∂ D ). where RD ∧ε is the operator sending α to RD ∧ε α. ∂ ∗ ] = −i∂ D is similar.k ∗ ∗ ˆ 2Zk DZ (Zj ∧ Zj ∧ α) k = −i In conclusion. 5. 5. ∆∂ D ] = iRD ∧ε . we have [L.20 and Proposition 5. Proof.29 we recall that RD ∧ε = ∂ D ∂ + ∂∂ D .22.¨ 5 Cohomology of Kahler Manifolds 65 = −i and (∂ D )∗ Lα = − j=k ∗ ∗ ˆ Zk (Zj ∧ Zj ∧ DZ α) + i k j=k ∗ ∗ ˆ Zj ∧ (Zj (Zj ∧ DZ α)) j i 2 j.24 Lemma.23 Proposition.

where β has compact support. ∂ ∗ β)2 − (∂ D L∗ α. (i∂α. β)2 + (L∗ ∂ D ∂α. Then there is no boundary term when considering integration by parts as in the L2 -products below. Now the two last claims follow from Proposition 5. β)2 = (L∗ ∂∂ D α. β)2 = −(∂ D ∂L∗ α. We have ∆∂ + ∆∂ D = ∆dD and ∆∂ − ∆∂ D = [iRD ∧ε . L∗ ]. L(∂ D )∗ β)2 + (∂α.20 and Proposition 5. The sum of the terms on the left hand side is equal to (i∆∂ α. β)2 . (∂ D )∗ Lβ)2 = −(L∗ ∂α.25 Theorem. we have ∂(∂ D )∗ + (∂ D )∗ ∂ = i∂L∗ ∂ − i∂∂L∗ + iL∗ ∂∂ − i∂L∗ ∂ = 0 since ∂∂ = 0.26 Corollary. Similarly. (∂ D )∗ β)2 + (i(∂ D )∗ α. 5. By Proposition 5. 5. Now ∂ D ∂ ∗ + ∂ ∗ ∂ D is the adjoint operator of ∂(∂ D )∗ + (∂ D )∗ ∂. β)2 + (i∂ D (∂ D )∗ α. The lemma follows. ∂ ∗ β)2 = (L∗ ∂ D α.24). We compute ∆dD = (∂ D + ∂)((∂ D )∗ + ∂ ∗ ) + ((∂ D )∗ + ∂ ∗ )(∂ D + ∂) = ∆∂ D + ∆∂ + ∂ D ∂ ∗ + ∂(∂ D )∗ + (∂ D )∗ ∂ + ∂ ∗ ∂ D = ∆∂ D + ∆∂ .23 and since ∆∂ D and ∆∂ preserve the type of forms. ∂β)2 = −(∂α. Proof. To prove the second equality.66 ¨ Lectures on Kahler Manifolds Proof. [L. (∂ D )∗ β)2 + (L∗ ∂ D ∂α. β)2 − (∂∂ D L∗ α. β)2 − (∂∂ D L∗ α.22. ∂ ∗ β)2 = (∂L∗ ∂ D α. hence the second equality. i∂β)2 = −(∂α. In particular. . (i∂ ∗ α.22. β)2 . then 2∆∂ = 2∆∂ D = ∆d . (∂ D )∗ β)2 + (L∗ ∂ D ∂α. By Equation 5. β)2 + (i(∂ D )∗ ∂ D α. ∆dD ] = 0 and ∆dD preserves the type of forms. β)2 = −(∂L∗ α. by (5. β)2 . If D is flat. let α and β be differential forms with values in E.

This is the important special case of differential forms with values in C. Assume that D is flat. by Hodge theory. then we also have [iRD ∧ε . Say r that a harmonic form α is primitive if L∗ α = 0. if RD = 0. E) ⊕ LHp−1.27 and 5. Furthermore. the Lefschetz map Ls : HP (M. The map Ls : Hr (M. If D is flat.27 Hard Lefschetz Theorem. then cohomology classes in H ∗ (M. E). E) the space of primitive harmonic forms of degree r. E) is injective for 0 ≤ s ≤ m − r and surjective for s ≥ m − r. (5. E) = ⊕s≥0 Ls HP (M. 5. E). E) = ⊕s≥0 Ls HP (M. E) as in (5. E) and Hp. if D is flat.11) induces a representation of sl2 (C) on the space H∗ (M. The following results are immediate consequences of the above commutation relations and the Lefschetz decompositions of Ar (M. Since L commutes with ∆dD and ∆dD = (∆dD )∗ .29 Remark. Hence the representation of sl2 (C) on A∗ (M. L∗ commutes with ∆dD as well. the decompositions are pointwise and hence L2 -orthogonal. and denote by HP (M.27. Then Hr (M. L∗ ] = 0 and. This will be important in some applications further on. E). . 5. Moreover. E).¨ 5 Cohomology of Kahler Manifolds 67 5.q (M. E) in (5. in particular. p−s. p−s.28 Theorem.q = HP (M. 2∆∂ = 2∆∂ D = ∆d . Note that L. E) = ⊕s≥0 Ls HP (M. E) p. E) of ∆d -harmonic forms.q−1 (M. that is. E) in (5.q (M. If M is closed. then dD dD = 0 and we get the associated cohomology H ∗ (M.q (M.15) and Ap. E). E).2 Lefschetz Map and Cohomology.q−s Furthermore. the Lefschetz decomposition of Hr (M. Then. If E is the trivial line bundle with the standard Hermitian metric. E) ⊂ r H (M. L∗ ] = 0. r−2s Hr (M. E) are uniquely represented by harmonic forms. C) → Hp. L∗ .q−s Hp.q (M. More generally. E) → Hr+2s (M. Suppose that D is flat and that p + q ≤ m = dimC M . s ≥ 0. by Theorem 5. E) = ⊕p+q=r Hp.q (M. by Theorem 5. 5. C) is injective for all s ≥ 0. then RD = 0 and hence [iRD ∧ε .q (M.28.17).30) p−s. the Hodge and Lefschetz decompositions of Hr (M.32 also hold for the corresponding spaces of square integrable harmonic forms. E). and (m − r)Pr preserve square integrability of differential forms so that Theorems 5.q−s Hp. A primitive harmonic form is primitive at each point of M .

Suppose that M is closed and D is flat. E) is injective for 0 ≤ s ≤ m − r and surjective for s ≥ m−r.31 Hard Lefschetz Theorem.0 (M. E) := dim H r (M. we have the Lefschetz decomposition r−2s H r (M. C) → Hq.32. E). a 5. C).q (M. If M is closed and D is flat. p−s.p (M. E) = hm−p. e = However. C) = hm−p. P (5. The existence of a cohomology class with this property is a rather special feature of closed K¨hler manifolds. Assertion 2) follows from Serre duality 3. the map Hp.30) above.m−q (M. Furthermore. H r (M. C) and define Betti numbers br (M. h1. We recall that Poincar´ duality also shows that H r (M. Assertion 1) is immediate from Theorem 5. E) = p+q=r hp. E ∗ ).31 is not obtained using the Hodge operator but the (m − r)-fold cup product with a cohomology class. C) ∼ H 2m−r (M. Recall that br (M ) = dim H r (M. E). 3) hq.q (M. 5.0 (M.q (M.p (M. P P (5. 4) br (M ) is even if r is odd.q−s (M. Assume that M is closed and D is flat.35 Theorem. the K¨hler a class. E). is (well defined and) a conjugate linear isomorphism.12 as does the second equality in Assertion 3) since the dual bundle of the trivial line bundle is the trivial line bundle. C).34) see also (5. hp. 2) hp. E) = s≥0 hp−s. then 1) br (M. the isomorphism in Theorem 5. Then we have Hodge and Lefschetz decompositions. b1 (M ) = 2h1. E). Proof.q (M. then by Theorem 5. Let p.q hp.32. E) + hp−1.q (M. Since conjugation commutes with ∆d (for complex valued differential forms on any Riemannian manifold). C). E) := dim HP (M. Then the map Ls : H r (M. E) = dim HP (M.q (M. C) = hp. E). C). E).q p. The remaining assertions are immediate consequences of Assertions 1) and 3).m−q (M.32 Theorem.q (M.q (M.q−1 (M. ϕ → ϕ. E).68 ¨ Lectures on Kahler Manifolds 5. E) = ⊕s≥0 Ls HP (M. E) = ⊕p+q=r H p. E) = hp.q−s H p.q (M. .33) If p + q ≤ m. This shows the first equality in Assertion 3). Assume that M is closed and D is flat. In particular. E) → H r+2s (M. C) is a topological invariant of M . E) = ⊕s≥0 Ls HP (M.

It follows that the first homology H1 (N. In the K¨hler case. E) and H m.m (M. page 444] and [Hir.1. y. In Theorem 9. 1) We see again that Hopf manifolds M = S 2m−1 × S 1 do not carry K¨hler metrics for m ≥ 2 since their first Betti number is odd.36 Remarks. A first example a of this kind is due to Thurston: Let H(R) be the Heisenberg group.q (M.4 in Borel’s Appendix to [Hir]. where the parameter of the factor S 1 is denoted t. see [GH. z ∈ R.25. see Lemma 9. H(R) is simply connected. z) = 0 1 y  . Theorem 15. and hence b1 (N ) = 2. 0 0 1 and H(Z) be the closed subgroup of matrices in H(R) with x. The natural projection H(R) → R2 . E) to the south and north and H m. x. Compare also with the Iwasawa manifold in Example 5. that is. z) → (x. Therefore the fundamental group of N is isomorphic to H(Z). y. Serre duality corresponds to the reflection about the center. C) = 0. C). N is a closed manifold. This diamond is called the Hodge diamond.0 (M. z ∈ Z. In particular. 1) The above results motivate a graphic arrangement of the Dolbeault cohomology groups in a diamond with vertices H 0. the differential form ω = dx ∧ dt + dy ∧ dz is well defined and symplectic on M . 5. descends to a projection of the quotients π : N := H(R)/H(Z) → R2 /Z2 . A(x. E) → H 2m−r (M. . an odd number. y. a b1 (M ) = 1. E) and H 0. In the case E = C = E ∗ . respectively.58.m (M.35. E) is compatible with the type decomposition and corresponds to a reflection about the equator of the Hodge diamond. 2) We come back to the question whether there are closed symplectic manifolds which do not carry K¨hler metrics. y. The Lefschetz isomorphism Lr : H r (M. It is easy to see that h1. cit. Since H(R) is diffeomorphic to R3 .q (−1)p+q hp. compare Remark 4. Borel determines the Dolbeault cohomology of Hopf and Calabi–Eckmann manifolds completely.8. For more on this topic see [TO] and the survey [BT]. y). However.37 Examples.0 (M.q (M. E) to the west and east. a fiber bundle with fiber a circle R/Z. the group of (3 × 3)-matrices of the form   1 x z A(x. = Let M := N × S 1 .¨ 5 Cohomology of Kahler Manifolds 69 5. M does not carry a K¨hler metric since the first a Betti number of M is 3. 2) If M is a closed complex manifold. In this notation. C) p+q=r and χ(M ) = p. then Betti numbers and Euler characteristic of M satisfy the Fr¨licher relations o br (M ) ≤ hp.5 loc. Conjugation corresponds to the reflection about the vertical axis between north and south pole. Z) = H(Z)/[H(Z). these are a clear from Theorem 5.0 (M. H(Z)] ∼ Z2 .1].

Poincar´ duality implies that S is non-degenerate.q (M. q) and (p . C) = 3 andh0. It is important that the harmonic representative of a primitive cohomology class is a differential form which is primitive at each point of M . denoted σ(M ).58 satisfies h1. In our discussion below. R). If m is even. ϕ) and T (Cϕ. it is then given by S(ϕ. (5.70 ¨ Lectures on Kahler Manifolds 3) There are closed complex manifolds satisfying hp. Thus closed oriented manifolds with real dimension divisible by four come with a remarkable topological invariant. If m e is odd. y) = x ∪ y. ψ) := M ω m−r ∧ ϕ ∧ ψ. p ). R) ⊗ C. S : H m (M. then S is skew-symmetric. Z) × H m (M.38) where [M ] denotes the oriented fundamental cycle of M and the angle brackets denote evaluation. C) = H r (M. Let ϕ. we have T (ϕ. m).0 (M. Then T (Ls ϕ. ψ) = M ϕ ∧ ψ. 5. (5. q) = (q . L ϕ) > 0 s s (1) (−1) (p+q)(p+q−1) 2 unless ϕ = 0. C) = 2. we represent cohomology classes of M by their harmonic representatives. the intersection form S is a bilinear form on the middle cohomology. C). ψ) = T (ϕ.41) For the proof of the second equality. then S is symmetric. Example 8.39) The index of S is called the signature of M . t ≥ 0 satisfy p + q + 2s = p + q + 2t = r ≤ m.p (M. T (L ϕ.42 Theorem (Hodge–Riemann Bilinear Relations).1 (M. Z) → Z. Lt ψ) = 0 i q−p unless s = t.40) We may also view T as a complex bilinear form on H r (M. q) and ψ is of type (p . Via representing closed differential forms. q ) and s. 0 ≤ r ≤ m. For example. We have T (ϕ. the Iwasawa manifold in Example 5. [M ] . ψ) = (−1)r T (ψ. Let M be a closed K¨hler a manifold. T (ϕ. a non-degenerate integral symmetric bilinear form on their middle cohomology. (2) . C ∗ ψ). there is a variation T of S which is a defined on H r (M. In the case of a closed K¨hler manifold. (5. a On a closed oriented manifold M of real dimension 2m.9]. S(x. (5. p = q. C) = hq. see [Zh. This is one of the arguments that show that this manifold does not carry a K¨hler metric. we note that since the volume form has type (m. ψ) = 0 if ϕ is of type (p. R). and q = p. We can coarsen this picture and consider the intersection form on H m (M. Via representing closed differential forms. ψ be primitive cohomology classes of type (p. q ) with (p.

p + q is even as well. s ≥ 0 be given with p + q + 2s = r. p and q have the same parity. Now m − r + 2s = m − p − q.42. see (5. C) − hp. (p+q)(p+q−1) 2 M We discuss a few applications of the Hodge–Riemann bilinear relations. Hence (p + q)2 is a multiple of 4 and therefore (−1) (p+q)(p+q−1) 2 iq−p = i(p+q)(p+q−1)+q−p = (−1)p = (−1)q .41). For a proof see e. we note that by the above discussion hp. ϕ is primitive of degree (q. Hence we only need to prove the first equality. e e Since L∗ is a real operator. Let r be even.43) ω m−p−q ∧ Cϕ ∧ ϕ (p+q)(p+q−1) 2 = (−1) = (−1) (m − p − q)! (m − p − q)! M ∗ϕ ∧ ϕ |ϕ|2 . the index of TH on H 1. By Theorem 5.q odd . Lm−p−q+1 ϕ = 0.¨ 5 Cohomology of Kahler Manifolds 71 The case s > t is similar. σ(M ) = P P p+q≤m p. If s = t. hence the form TH (ϕ. p). (m − p − q − s)! (5. p. iq−p T (Ls ϕ. m). then ω m−r ∧ Ls ϕ ∧ Lt ψ = Lm−r+s+t ϕ ∧ ψ = 0. C) = 0 if k is odd. q. Hence if s < t. compare also the discussion of K¨hler surfaces below.q p. Proof. TH is non-degenerate and Hodge and Lefschetz decomposition are orthogonal with respect to TH . by (5. C). C) is (1. then (1) is immediate since the volume form has type (m. This completes the proof of (1). C).0 (M.q (M.q (−1)q hp. Th´or`me 2 on page 23. Since ϕ is primitive. that is. If s ≤ m − p − q.43). ψ) is Hermitian on H r (M. p+q=k (−1) h (M. Let p. then ∗Ls ϕ = (−1) (p+q)(p+q−1) 2 Proof. [Wei]. The main and only point in the proof of (2) is a formula for the ∗-operator on primitive differential forms. To that end. Then T is symmetric.g. C) if p and q are even and negative definite if p and q are odd.q From Theorem 5. a 5. C) − 1). Then the signature σ(M ) = (−1)q hp.q (M.1 (M.q (M. For example.1 (M. h1.q (M. Let M be a closed K¨hler manifold with even a complex dimension m.35. C) = 1. Hence.q even p+q≤m p. By Theorem 5. ψ) = T (ϕ.44 Hodge Index Theorem. Ls ϕ) = M s! Lm−p−q−s Cϕ.42. Since r is even. C) = p+q even q p. C).2 we conclude that TH is positive definite on Ls HP (M. since dim H 0.

In other words.45 Remarks. compare Example B. C) − Ap. let M be a a a closed surface of genus g. 1) If p + q = m then p = q modulo 2 so that we can substitute (−1)p for (−1)q in Theorem 5.1 (M. C).q+1 (M. By the Hodge–Riemann bilinear relations. C) with p + q even and < m occur twice and with the same sign. C) are Lagrangian subspaces. for the latter. C) are topological invariants of M . see again Example B. C) = hp. the Hermitian symmetric space of Lagrangian subspaces on which TH is negative definite. C) and H 0. C) − hp−1.44. The Hirzebruch–Riemann–Roch formula expresses the numbers χp (M.0 (M. the form T is a symplectic form on H 1 (M.72 ¨ Lectures on Kahler Manifolds On the other hand. ψ) is negative definite on H 1. C) a in the period domain. C) = hm−p. C) = σ(M ). The a a space H 1 (M.q−1 (M.31).q (M.q (M. which associates to a complex structure on M with a K¨hler form cohomologous to ω the Lagrangian subspace H 1.34). C) and H 0.q (M. we obtain a tool to study the space of complex structures on M together with a K¨hler structure with K¨hler form cohomologous to ω.q (M. C) − Ap. hp. C) and the subspaces H 1. C) into H 1.m−q (M.0 (M.1 (M. C) is the Euler characteristic of the cohomology of the Dolbeault chain complex → → → → · · · − Ap.q (M. viewed as a complex vector bundle. we get the arithmetic genus χ0 (M. C) − · · · For p = 0.q (M. C) := q (−1)q hp. Then H 1 (M. For example. We get that the sum of these numbers is a topological invariant of M . ψ) := iT (ϕ. C)) p+q≤m p+q even The terms hp.0 (M. C) in terms of the Chern classes of the tangent bundle T M .0 (M. since p + q ≤ m. P see (5.1 (M.q (M.42. This gives σ(M ) = (−1)q (hp. Thus we obtain a map.q−1 (M. q = m − q mod 2. Since hp.0 (M. C) and H 0. σ(M ) = p+q even 5.q−1 (M.1 (M.q (M. we conclude that (−1)q hp. C) ∼ C2g and T is the intersection = ∂ ∂ ∂ ∂ . C) and positive definite on H 0. but the inclusions of H 1.53. Let M be a closed K¨hler manifold with K¨hler form ω. C). C) and p = m − p. C) of M as in (2. The associated Hermitian form TH (ϕ. C) − hp−1. the period map.0 (M. C). We point out a few other applications of the Hodge–Riemann bilinear relations 5. C) and the dimensions of H 1. χp (M.53. C) depend on the complex structure of M . 2) For p fixed. Suppose that the complex dimension m of M is even. χp (M.

1 on which TH is negative definite. and h1. Section V.1 S(ϕ. C) of dimension h1. that dc is well defined on the space A∗ (M. h0.1 + 2. C) and positive definite on the remaining summands in the Hodge decomposition of H 2 (M. Hence the intersection form S of M has signature (b2 − h1. If (Φ1 .2 are topological invariants of M . C) is negative definite on HP (M.¨ 5 Cohomology of Kahler Manifolds 73 form on H 1 (M. and [CMP] for more on period maps. that is. a Then the Hodge numbers h1. Therefore we will omit the superindex D and simply write d and ∂ instead of dD and ∂ D .25 we conclude that ∆d = 2∆∂ = 2∆∂ . R) of real valued differential forms on M .1 − 1) and the signature of M is b2 − 2h1. Φk ) is such a frame and σ = ϕµ Φµ a local smooth section of E.0 (M. the period map is a holomorphic embedding of Teichm¨ller space into G− (L.46) where C is as in (5.1 + 1. . Section 7. We conclude that the remaining Hodge numbers h2. Note that by u C definition. h1.0 + h1.6 and B. g) = Sp(g.1). . ψ) on H 2 (M. ψ) = 1. and h1. that is. the period map is equivariant with respect to the induced actions of diffeomorphisms.1 = b2 (M ) = b2 and h1. for the latter. in this case from the space of complex structures on M together with a K¨hler structure with K¨hler form cohomologous to ω to a a the corresponding period domain. R)/ U(g).10. We introduce the complex differential dc = i(∂ − ∂) = C ∗ dC = C −1 dC.1 = h1. see Proposition 3. that RD = 0.3]. and Torelli theorems.48) . From Theorem 5.47) c and ddc = 2i∂∂ = −2i∂∂ = −dc d. P P By the Hodge–Riemann bilinear relations.0 = h1. C).2 . then dD σ = dϕµ ⊗ Φµ and ∂ D σ = ∂ϕµ ⊗ Φµ . Then the connection form θ of D with respect to a local holomorphic frame is holomorphic.1 .3 The ddc -Lemma and Formality. C).1 . (5. [Jo. The two last expressions show that dc is real.21. Suppose now that M is a closed K¨hler surface. We have d∗ = C ∗ d∗ C = C −1 d∗ C (5. the Hermitian form TH (ϕ. 5. By the Torelli theorem. Furthermore. dimC M = 2.1 + 1. It follows that E has parallel local holomorphic frames. C).1 are invariants of the smooth structure of M . the Grassmannian of subspaces of H 2 (M. that is. h2. period domains. See [Wel. . h0. .1 + h0. we obtain again a period map. (5. We have 2h2. We assume throughout this subsection that D is flat.0 .0 = h1.42.6]. compare P Examples 4.

E) = H∗ (M. Let α be a differential form with values in E with α = dβ and dc α = 0 or. −1 λ α if α ∈ Vλ with λ > 0. respectively. we conclude that G and δ ∗ commute as well on A∗ (M. E) = V0 = ker ∆ is the space of δ-harmonic forms. Any differential form β with values in E satisfies β = Hβ + ∆Gβ. c c Since M is a K¨hler manifold. Proof. where H∗ (M. This is also immediate from the displayed formula. where Vλ ⊂ A∗ (M. we have ∆d = ∆dc =: ∆. It follows that G and δ commute on A∗ (M. we have δ(Vλ ) ⊂ Vλ .50 Lemma (ddc -Lemma). . E) ⊕ δ(A∗ (M. E)) ⊂ A∗ (M. E) is the eigenspace of ∆ for the eigenvalue λ. Therefore a ∆d = C −1 ∆d C = ∆dc . E). (5. E). Elliptic regularity implies that G(A∗ (M. E)) ⊕ δ ∗ (A∗ (M. hence α = ∆Gα = (δδ ∗ + δ ∗ δ)Gα = δδ ∗ Gα. Since δ commutes with ∆ on A∗ (M. E).49). By (5. E)) → L2 (A∗ (M.49) 5. Then there is a differential form γ with values in E such that α = ddc γ. By assumption and the Hodge decomposition. L2 (A∗ (M. where H denotes the L2 -orthogonal projection onto H∗ (M. Let ∆dc = dc d∗ + d∗ dc be the Laplacian of dc . In what follows. Recall the eigenspace decomposition of the Hilbert space of (equivalence classes of) square integrable sections. Suppose that M is closed and D is flat. Recall that G(α) = 0 if α ∈ V0 . The Hodge decomposition theorem asserts that we have an L2 -orthogonal sum A∗ (M. δ denotes one of the operators d or dc . E)) be the Green’s operator associated to ∆. E). E)) = ⊕Vλ .74 ¨ Lectures on Kahler Manifolds where we use that D is flat. Hα = 0 since α ∈ im δ. The latter also implies δα = 0. ∆d preserves the type of forms. Let G : L2 (A∗ (M. Since G is self-adjoint. E)). dα = 0 and α = dc β. E).

E) induces an isomorphism d ∗ − Hdc (M. E). Since ddc = −dc d. Let α be a closed differential form on M . hence β = ddc γ. by Lemma 5. d Suppose now that dc α = 0 and α = dβ. The inclusion A∗c (M.52 Lemma. d .¨ 5 Cohomology of Kahler Manifolds 75 where we use that G and δ commute on A∗ (M. 3) Formulate and prove a ∂∂-Lemma. Hence the cohomology class of α contains a representative in A∗c (M. Let Hdc (M. In the first case we also assume that dc α = 0. In both cases we conclude α = dd∗ Gdc d∗ Gα = dd∗ dc Gd∗ Gα = −ddc (d∗ Gd∗ Gα). Now dc (dc γ) = 0. E) of A∗ (M. E) −→ H ∗ (M. 1) Calculate that. E). E). ω = 2 i∂∂ϕ. and hence the d induced map on cohomology is injective. the subcomplex A∗c (M. Let β = dc α. 5. We show first that the induced map on cohomology is surjective. c c c where we note that the commutation relation 5. d). Set now A∗c (M. the kernel of ddc : A0 (M. E)/dc A∗ (M. q) and α = dβ or α = dc β. The formula implies that.24 implies that d∗ dc = −dc d∗ . hence α = dc d∗ Gα c and α = dd∗ Gα in the first and second case. E) be the cohomology of the complex (A∗c (M. E) → A∗ (M. and dc (α + dγ) = β + dc dγ = β − ddc γ = 0. Then dβ = −dc dα = 0. d) will be denoted Hdc (M. ddc ϕ. E). in the second that dα = 0. the a Laplacian of a function does not involve derivatives of the metric.51 Exercises. E).50. E). E) := A∗c (M. Then d induces a differend d tial d on A∗c (M. 2) If α is a differential form of type (p.50. ω = −2g jk ∂2ϕ = ∆ϕ. What follows is taken from [DGMS]. E) d consisting of dc -closed differential forms is closed under the exterior differential ∗ d. for K¨hler manifolds. Then α = d(dc γ). The cohomology d ∗ of the complex (A∗c (M. ∂zj ∂zk This is a truly remarkable formula: It shows that. for functions ϕ. C) consists precisely of the constant functions. Assume throughout that M is closed and D is flat. by Lemma 5. ∼ = Proof. respectively. E). for M closed and connected. E) and turns the latter into a chain complex. hence α is a d-coboundary in A∗c (M. then α = ddc γ. E). d 5. C) → A2 (M.

C) → A∗c (M. and hence d dα = β = dc dγ ∈ im dc . Then dc β = −ddc α = 0. by Lemma 5. E). A∗ (M. C)/dc A∗ (M. The differential d = 0 and the projection A∗c (M. R) → A∗ (M. C) d d are quasi-isomorphisms and that the differential of the latter is trivial.42). E). Then dc γ = 0. E) → A∗c (M. E) with dα = 0 and such that α = dc β. C) is a formal commutative differential graded algebra. We use cgda as a shorthand for commutative graded differential algebra. hence α is cohomologous to 0 in A∗c (M. . Let d γ = α + dc β. commutative in the graded sense as in the case of differential forms. . we get that H∗ (M. R) induces an isomorphism in cohomology. 5. E). A cgda A is called formal if there is a sequence B0 = A. Since dc is a real operator. hence harmonic. E) d d induces an isomorphism ∗ ∗ − Hdc (M. E) and set β = dα. see (1. hence γ represents d ∗ an element in Hdc (M. R) together with wedge product and trivial differential is a cgda over R. hence γ ∈ A∗c (M. that is. Let M be a closed Riemannian manifold ˆ with curvature operator R ≥ 0. C) ← A∗c (M. the proofs also work in the case of real numbers and show that A∗ (M. E).54) . E). E) −→ Hdc (M. d For E the trivial line bundle. the chain complex (Adc (M. . 5. A homomorphism between cgda’s A and B (over a field F ) is called a quasiisomorphism if it induces an isomorphism of their cohomology rings. Hence d = 0. dγ = 0.55 Theorem ([DGMS]). Let α ∈ A∗c (M. Let now α ∈ A∗c (M. d) is invariant under the wedge product and the chain complex (A∗ (M. (5. Moreover.50. turning them into graded differential algebras over C. For a closed K¨hler a manifold M we have established that A∗ (M.36). E). Hence M is formal over R. Bk of cgda’s over F such that Bi−1 ≈ Bi . B1 . and such that the differential of Bk is trivial. R) is a formal cgda. . in letters A ≈ B. and finally α = γ modulo dc A∗ (M. the inclusion H∗ (M.53 Lemma. If M is a closed K¨hler manifold. C). ∼ = Proof. Then dc α = 0 and hence dα = dc dβ for some β. Then real valued harmonic forms on M are parallel.56 Remark (See Remark 1. Since the wedge product of parallel differential forms is parallel. M is also formal over R.76 ¨ Lectures on Kahler Manifolds 5. Let α ∈ A∗c (M. 1 ≤ i ≤ k. In other words. Then A and B are called quasi-isomorphic. C). then M is formal a over C. Then α = ddc γ d for some γ. d) therefore also inherits a product.

y. a torus bundle over a torus. respectively. z in R := Z + iZ. Γ] = R2 to H1 (R2 \C2 . The Massey triple product a. q and r. z in C. in particular for the real cohomology of closed K¨hler manifolds. y. The homomorphism G → C2 . 0)-forms a.4 in [DGMS]. a A nilmanifold is a quotient Γ\G. c ∈ H ∗ (A) be cohomology classes of degree p. Example 8. C). see [GH. β. Note that the first Betti number of M is 4. hence they descend to differential (1. It follows that the Iwasawa manifold is not formal over C. we ¯ conclude that (α.55. dy. respectively. β = 0 modulo α∧H 1 (M. β. and dz − xdy on G are left-invariant. Since π induces an isomorphism from H1 (M.57) We leave it as an exercise to the reader to show that the Massey triple product is well defined up to α ∧ H q+r−1 (A) + γ ∧ H p+q−1 (A). C). M is a closed manifold and. it follows that α. c := [ϕ ∧ γ − (−1)p α ∧ ψ] ∈ H p+q+r−1 (A). To give at least one application of Theorem 5. See also [FG]. Let G = H(C) be the complex Heisenberg group. Let α.37. but now with x.58 Example. It can be a shown that b2 (M ) = 8 and b3 (M ) = 10. C)+β ∧H 1 (M. respectively. Z) = R2 . y. β) is a basis of H 1 (M. so that the Betti numbers of M alone do not show that M is not K¨hlerian. the Iwasawa manifold M = Γ\G. We conclude that Massey triple products vanish for formal cgda’s. ψ ∈ A of degree p + q − 1 and q + r − 1.¨ 5 Cohomology of Kahler Manifolds 77 We refer to [DGMS] for a discussion of the meaning of formality for the topology of M . page 444] and [Zh. a 5. induces a holomorphic submersion π : M → R2 \C2 with fibers biholomorphic to R\C. The Iwasawa manifold M in Example 5. γ ∈ A be representatives of a. . Since dϕ = 0.9]. 0)-forms dx. Hence M does not carry a K¨hler metric14 . Then α ∧ β = dϕ and β ∧ γ = dψ for ϕ. b.58 is a nilmanifold since H(C) is nilpotent. β = [ϕ ∧ b]. y. Let α = [a].3 and Corollary 3. (5. We have da = db = 0 and dϕ = −a∧b. that is. we explain one of the features which distinguish formal cgda’s from general ones. such that a ∪ b = b ∪ c = 0. Let Γ ⊂ H(C) be the discrete subgroup of A(x. We have ¯ α. z) as in Example 5. A theorem of Benson–Gordon and Hasegawa 14 For interesting other arguments. β = [b] in H 1 (M. and that it vanishes for cgda’s with trivial differential. y. z) → (x. b. y). Let A be a cgda and a. b. b. In particular. The differential (1. Since Γ acts by biholomorphic maps on G.2. β. is a complex manifold of complex dimension 3. more specifically. c. that it is preserved under quasi-isomorphisms. where G is a nilpotent Lie group and Γ is a discrete subgroup of G. the R-homotopy type of a closed and simply connected K¨hler a manifold M is a formal consequence of the real cohomology ring of M . z) with x. A(x. β. C). the complex Lie group of matrices A(x. α. In a sense made precise in Theorem 3. Z) = Γ/[Γ. and ϕ on M .

the curvature of the Chern connection D of h is Θ = ∂∂ ln h(ϕj . ϕj ) = 1 zj zj zk zk . in the above point z. We note that the curvature form Θ of D acts by scalar multiplication on sections and is independent of the local trivialization. For the curvature of the Chern . Hence. In this sense we have Θ = RD .45. In the notation of Example 3. D) < 0 everywhere.58. [TO].2 that K is isomorphic to U m+1 . Y ) := ω(X. JY ) (5.60 Examples. D) = − dzk ∧ dzk < 0.45. Let E → M be a holomorphic line bundle over M . We say that a holomorphic line bundle E → M is positive. 1) The tautological bundle U → CP m . 1) and negative line bundles are defined correspondingly.59) is a Riemannian metric on M . we will use the open covering of CP m by the sets Uj = {[z] ∈ CP m | zj = 0}. [Ha]. as we already checked in Example 3. We say that a differential form ω on M of type (1.78 ¨ Lectures on Kahler Manifolds says that a compact nilmanifold Γ\G does not admit a K¨hler structure una less G is Abelian.1 (M.0 (CP m . see [BeG].4 Some Vanishing Theorems. the canonical Hermitian metric h on U is given by h(ϕj . 2π k=j Now the homogeneity of U and h under the canonical action of the unitary group U(m + 1) implies that c1 (U. Thus Θ and RD are complex valued differential forms on M . We say that a cohomology class a a c ∈ H 1. We have E > 0 iff the dual bundle E ∗ < 0. 1) is positive and write ω > 0 if g(X. Negative differential forms and cohomology classes of type (1. In the point z ∈ Uj with coordinates zj = 1 and zk = 0 for k = j. This generalizes our discussion in Example 5. 0 ≤ j ≤ m. Hence U is negative. if its first Chern class c1 (E) > 0. 2) The canonical line bundle K = Am. As before. i c1 (U. C) is positive and write c > 0 if c has a positive representative. denoted E > 0. If ω is closed and positive. 5. Let h be a Hermitian metric on E and D be the associated Chern connection. We continue the discussion of Examples 3. 5.24 in greater generality. C) → CP m : We saw in Example 3.1. Let M be a closed complex manifold.45 and consider some line bundles over CP m . ϕj ) = k=j ∂∂(zk zk ) = − k=j dzk ∧ dzk . then g turns M into a K¨hler manifold with K¨hler form ω.

3) The hyperplane bundle H → CP m : H is inverse to U . that is. the ddc -Lemma 5.61 Lemma. where hϕ = h(ϕ. 5. We note that these inequalities are special cases of the last two (groups of) equations in the proof of Theorem 5. Suppose now that E > 0. compare [Wei. It follows that H is positive. Choose a K¨hler metric g on M with K¨hler a a form ω ∈ 2πc1 (E). for a K¨hler manifold M with Ricci form ρ.25.62 Proposition. In [Mo]. the curvature of a the canonical bundle KM is given by iρ. we have ∆∂ − ∆∂ D = 1 1 1 [ω∧.62 by a direct computation using a Weitzenb¨ck formula. ϕ). which usually go into the proof of Kodaira’s vanishing theorem 5.8. we have i(m + 1)Θ < 0.67). and with respect to this K¨hler metric a a 1 (r − m)Pr . H is isomorphic to the dual line bundle U ∗ .63 Remark. this confirms i(m + 1)Θ = −ρ < 0. Proposition VI.50 applies and shows that a for some real function ψ. Let h = h · eψ . Let E → M be a holomorphic line bundle and h be a Hermitian metric on M such that the curvature Θ of its Chern connection D satisfies iλΘ > 0 for some λ ∈ R. λ Proof. Since M is a closed K¨hler manifold. ω − iΘ = i∂∂ψ 5. In particular. Proposition 5.¨ 5 Cohomology of Kahler Manifolds 79 connection of the canonical metric on K we get (m + 1)Θ.64. λ λ λ 5. Now [iΘ] = 2πc1 (E) = [ω] by the definition of c1 (M ) and the choice of ω. ∂Θ = ∂Θ = 0. where Θ is the curvature of U with respect to the Chern connection of its canonical Hermitian metric. Moroianu proves Proposition 5. Since the canonical metric on CP m has positive Ricci curvature. Let h be any Hermitian metric on E. E is positive if E has a Hermitian metric with iΘ > 0. By the previous example.62 refines the Nakano inequalities. This is possible since c1 (E) > 0. o ∆∂ − ∆∂ D = [iRD ∧ε . Proof. By Theorem 5. L∗ ] = [L. L∗ ] = . Let E → M be a holomorphic line bundle. Then ω := iλΘ is the K¨hler form of a a K¨hler metric on M . Since 2πc1 (E) = [iΘ]. and similarly in the case E < 0.25 and Proposition 5. Θ be the curvature form of its Chern connection. Then E > 0 iff there is a Hermitian metric h on E such that the curvature Θ of its Chern connection satisfies iΘ > 0. Then the curvature Θ of the Chern connection of h satisfies iΘ = i∂∂ ln h = iΘ + ω − iΘ = ω > 0. L∗ ] = (r − m)Pr . More generally. hence K < 0.2. Then Θ = ∂∂ ln hϕ . see (4.5]. and ϕ be a local and nowhere vanishing holomorphic section of E (a local frame of E). We also have ∂ω = ∂ω = 0.

K ⊗ K ∗ ⊗ E) = H m. Zm . Therefore. then H p. E) = H 0. 2 5.13. the dimension of the space of holomorphic sections of E. Hence if K ∗ ⊗ E > 0. E) = dim H 0. Let p ∈ M .0 (M.15] computes the dimension of H 0. q) in the bundle K ⊗ L. Z1 . Theorem III. Then. This will be important in the proof of the Kodaira embedding theorem 9.62. the two assertions are equivalent since E > 0 iff E ∗ < 0. Zm ) is the corresponding dual frame. . Let g be the K¨hler metric on M with a K¨hler form ω = −iΘ. Z1 . Let a h be a Hermitian metric on E and D be the corresponding Chern connection.0 (M.12.q (M. Let α be a ∆∂ -harmonic form with values in E and of a type (p. Then there exist an orthonormal frame (X1 .64 Kodaira Vanishing Theorem. q). .q (M. E). H 0.65. then χ0 (M. 2 ∗ ∗ ∗ ∗ where the frame (Z1 . E) = 0 for p + q > m.q (M. E) = 0 for q ≥ 1. 1. .6. E) = 0 for p + q < m. Let M be a closed complex manifold and E → M be a holomorphic line bundle with K ∗ ⊗ E > 0. then the Hirzebruch–Riemann–Roch formula [LM. . If E < 0. Zm ) is as in (3.66) θj Z j ∧ Z j . . . E). We note first that a differential form of type (m. by Proposition 5. Ym ) with Yj = JXj in (or about) p and real numbers θ1 . then H p. In view of Serre duality 3. K ∗ ⊗ E) = 0 by Theorem 5. E) := (−1)q dim H 0. If E > 0. 2.65 Corollary. For a local section σ of E about p and a . θm such that the curvature of D at p is given by 1 D ∗ ∗ Rp = (5. . We discuss the second assertion (and a similar argument shows the first). q) with values in a holomorphic bundle L is the same as a form of type (0. Another application: If K ∗ ⊗ E > 0.q (M. Then H 0.80 ¨ Lectures on Kahler Manifolds 5. Let M be a K¨hler manifold and E → M be a holomorphic line bundle. .q (M. . Zm . Proof.30) and (Z1 . . . 0 ≤ (m − p − q)||α||2 = −(∆∂ D α. Y1 . . α)2 ≤ 0. Let M be a closed complex manifold and E → M be a holomorphic line bundle.q (M.q (M. . . Xm . . Proof of Corollary 5. Let h be a Hermitian metric on E such that the curvature Θ of its Chern connection satisfies iΘ < 0. .64 since m + q > m.

L∗ ]α = 81 k ∗ ∗ ∗ ∗ Zj ∧ (Zj ∧ (Zk (Zk α))) − Zk (Zk (Zj ∧ (Zj ∧ α))) It follows that for α as above ∗ ∗ ∗ ∗ = Zj ∧ (Zj ∧ (Zj (Zj α))) − Zj (Zj (Zj ∧ (Zj ∧ α)))   α if j ∈ J ∩ K. see [Ko2] and [ShS]. We note that there are two cases: κ > 0 for p + q > m and κ < 0 for p + q < m.q (M.64. J be multiindices with |I| = p and |J| = q. L∗ ]α. Y ) + iκω(X.25 gives us some room for extensions to pinching conditions. Then j∈J∩K θj − j ∈J∪K / θj ≥ a(κ(p0 ) − ε) − (m − p − q + a)(κ(p0 ) + ε) ≥ (p + q − m)κ(p0 ) − (m − |p − q|)ε > 0.  (5.69.¨ 5 Cohomology of Kahler Manifolds ∗ differential form α = ZJ ∧ ZK ⊗ σ. α)2 > 0 unless α = 0. /   0 otherwise. The rest of the argument is as in Theorem 5. 5. Let I. Y )| < ε|X| |Y | for all vector fields X and Y . We prove a simple model result in this direction. E) = 0 for all p. . D [iRp ∧ε . q with ε · (m − |p − q|) < κ · (p + q − m). Hence ([iRD . The condition on RD implies that the numbers θj as in (5.68) This refines Proposition 5.69 Theorem. Let a a E → M be a holomorphic line bundle and h be a Hermitian metric on E with Chern connection D. L∗ ]α = 1 2 j∈J∩K θj − θj α. j ∈J∪K / (5. and set a = |I ∩ J|. Let κ : M → R and ε > 0 and suppose that |RD (X. Let p0 ∈ M .64 as the case of “constant curvature” RD = iω. Proof of Theorem 5. For more on vanishing theorems. we have ∗ ∗ [Zj ∧ Zj ∧. Theorem 5. We interpret (the proof of) Theorem 5.66) satisfy |θj − κ(p0 )| < ε for all j.67) = −α if j ∈ J ∪ K.8. Let M be a closed K¨hler manifold with K¨hler form ω. Then H p.

This implies Ric X = 0 and is in contradiction to Ric < 0. This is in contradiction to the compactness of M . Proof. Then the identity component of the isometry group of M is a torus of dimension k ≤ dim M . by the de Rham decomposition theorem. In any case we note that R(·. Then X is automorphic iff X is parallel. Vice versa.2 Corollary. X ≤ 0. X)X = 0 since X is parallel. As for the second assertion.77. If M is simply connected. it suffices to show that M does not carry a non-trivial Killing field. and its orbits foliate M into a parallel family of flat tori. X = Ric X. we have ∗ X = Ric X. It is clear that parallel fields are Killing fields. What we need in the proof of Theorem 6.1 since the Lie algebra of the isometry group is given by the Lie algebra of Killing fields.1 is the inequality Ric ≤ 0 together with the differential equation ∗ X = Ric X.1. Let M be a closed and connected Riemannian manifold with Ric ≤ 0. If M is simply connected or if Ric < 0 at some point of M . see [BM].6 Ricci Curvature and Global Structure Let M be a closed Riemannian manifold. then the isometry group of M is finite. By Theorem 6. The first assertion is clear from Theorem 6. By Proposition 4.79. then X points in the direction of a Euclidean factor R of M . let X be a Killing field on M . Then X is a Killing field iff X is parallel. a see Proposition 4. 6. Hence we have the following analogue of Theorem 6. Recall now that automorphic vector fields on closed K¨hler manifolds are characterized by the latter equation.3 Theorem. . Recall that the isometry group of M is a compact Lie group (with respect to the compact-open topology) with Lie algebra isomorphic to the Lie algebra of Killing fields on M (with sign of the Lie bracket reversed). Let M be a closed Riemannian manifold with Ric ≤ 0 and X be a vector field on M . Let X be such a field. and hence X is parallel. We recall now again that the automorphism group of a closed complex manifold M is a complex Lie group with respect to the compact-open topology with Lie algebra isomorphic to the Lie algebra of automorphic vector fields on M (again and for the same reason with sign of the Lie bracket reversed).1 Theorem (Bochner [Boc]).1. Proof. Therefore 0≤ | X|2 = ∗ X. 6. X is parallel. 6. Let M be a closed K¨hler manifold with Ric ≤ 0 and X be a a vector field on M .

Hence a ˜ ˜ the isometry group I(M ) of M is the product of the isometry group of Ck and the isometry group I(N ) of N . J) be a closed complex manifold with c1 (M ) = 0.2.2 and Section 6. Chapter 3]. . a 6. 6. a ˆ ˆ Then there is a finite covering M of M which splits isometrically as M = F ×N . Consider the projection ˜ ). a 6. a closed flat Riemannian manifold is finitely covered by a flat torus [Wo.8 in [Jo]. Proof of Theorem 6. Then the a Lie algebras of parallel. C) if non-empty 15 . do not elaborate on the topology of spaces of metrics. Let (M.1 (M. R) = H 1.5 Remark. The identity components of the automorphism group and of the isometry group of M coincide and are complex tori of complex dimension ≤ dimC M . R).76 and the theorem of Cheeger and Gromoll mentioned a ˜ above. Then for each K¨hler form ω of M .1 (M.8 Remark.7 Remark. there are obstructions against K¨hler metrics of non-positive Ricci curvature.1 (M.9.1 Ricci-Flat K¨hler Manifolds. hence is a subgroup of I(M ). 6. By a famous result of Lohkamp [Loh].6. an open a subset of H 1. Corollary 6. there is exactly one Ricci-flat K¨hler metric g a a on M with K¨hler form ω cohomologous to ω. and automorphic vector fields coincide. C) ∩ H 2 (M. A reference for this are Subsection 4. a Therefore the space of Ricci-flat K¨hler metrics on M can be identified with a the cone of K¨hler forms in H 1. hence of real dimension h1. on closed complex manifolds. 6.6 Ricci Curvature and Global Structure 83 6. The following application is given in [Bea].4 shows that. by the Calabi conjecture 7. where E is a Euclidean space and N is closed and simply connected with Ric ≥ 0.1 (M. By the Bieberbach theorem. the kernel Γ has finite ˆ of Γ to the factor I(N ) of I(M index in Γ and gives a finite cover of M as asserted.6 Theorem. Let M be a closed and connected K¨hler manifold with Ric = 0. According to the de Rham decomposition theorem for K¨hler manifolds 4. where F is a closed flat K¨hler manifold and N is a closed and simply connected a K¨hler manifold with Ric = 0.1. I(N ) is finite. A celebrated theorem of Cheeger and a ˜ Gromoll says that the universal covering space M of a closed and connected ˜ Riemannian manifold M with Ric ≥ 0 splits isometrically as M = E ×N . Let M be a closed K¨hler manifold with Ric ≤ 0. see [CG1]. 15 We ˜ M = Ck × N. Since I(N ) is finite. By Corollary 6. all smooth manifolds of dimension ≥ 3 admit complete Riemannian metrics of strictly negative Ricci curvature.4 Corollary. The fundamental group Γ of M acts isometrically as a group of covering ˜ ˜ transformations on M . where Joyce discusses deformations of complex and Calabi–Yau manifolds. k ≥ 0. Killing. R). the universal covering space M of M decomposes as where N is a closed and simply connected K¨hler manifold with Ric = 0.

Ym ) to consist of eigenvectors of the Ricci tensor. where we recall that ·. Xm . Using that ϕ is of type (p. Y = −i R(Xj . Now the curvature acts as a deriva∗ tion on A (M. For X ∈ T M . Y = −i =− ˆ R(Xj . Y1 .30). C) = 0 for p > 0. (6. Let M be a closed. we then have. (6. p If Ric > 0 at some point of M . at the chosen point.32) for ∂ and (3.31). For multiindices J : j1 < · · · < jp and K : k1 < · · · < kp .11 Theorem (Bochner [Boc]).0 (M. ∗ ∗ ∗ ∗ K(Zj1 ∧ · · · ∧ Zjp ). Zk ) is still of type (p. C) = h0. in particular. .37. hp. 0). . Hence to compute its action on Ap. we get KX . Let M be a closed K¨hler manifold and a ϕ be a form of type (p. Zjµ ) if J = K. Y R(Xj . Y ). p > 0.0 (M. For Y ∈ T M . Zk1 ∧ · · · ∧ Zkp = 4p 0 Ric(Zjµ . . and define Zj and Zj as in (3.84 ¨ Lectures on Kahler Manifolds 6. Now D2 ϕ(Zj . 0). . and then hp. Zk ). 0) for p > 0. Yj )X . 0). connected K¨hler manifold a with Ric ≥ 0. Zj ) = − tr D2 ϕ − i ˆ R(Xj .2 Nonnegative Ricci Curvature. Y1 . then M does not have non-trivial holomorphic forms of type (p. Let (X1 . . we may choose the above frame (X1 . Now at a given point in M . · denotes the complex bilinear extension of the Riemannian metric of M to the complexified tangent bundle TC M . . In conclusion. . 6. . Xm . . Yj )X.p (M. 0) are parallel. C) = h0.33) for ∂ ∗ . Then holomorphic forms of type (p. Yj ). C) ≤ m . C). ∆d ϕ = 2∂ ∗ ∂ϕ = ˆ ∗ ˆ ϕ + Kϕ. JY = Ric(X. considered as an endomorphism. 0) and the formulas (3. Yj )X. C).9) ˆ with curvature term K = −i R(Xj . · is of type (1. we first determine its action on forms of type (1. hence the right hand side is equal to −4 ˆ ˆ D2 ϕ(Zj . the dual vector X = X. where we use Exercise 3.0 (M.p (M. Hence the curvature term K in the above equation is non-negative if Ric is non-negative and positive if Ric is positive. Ym ) be a local orthonormal frame of M such that JXj = Yj . we get ∆d ϕ = 2∆∂ ϕ = 2∂ ∗ ∂ϕ = −4 ∗ ˆ Zj Zk ∧ D2 ϕ(Zj .10) otherwise. Yj )ϕ. see (3. 0).2 for the latter equality.

this implies ˜ χ(M. By Theorem 6. O) = |Γ| · χ(M. ϕ = M | ˆ ϕ|2 + Kϕ. Consider the universal covering π : M → M .p (M. Hence 0= M ˆ ∗ ˆ ϕ. then ∆d ϕ = 2∂ ∗ ∂ϕ = 0. To see this we note that Ric > 0 implies that the anti-canonical ∗ ∗ bundle KM → M is positive. hence |Γ| = 1. the fundamental group of M is finite. ϕ ≥ 0 if Ric ≥ 0. Now the Hirzebruch–Riemann–Roch formula expresses the arithmetic genus of a closed K¨hler manifold as an integral over a a universal polynomial in the curvature tensor. ˜ Proof. Since the Ricci curvature of a M is positive. M As we just explained. by the theorem of Bonnet– ˜ Myers. and hence M is simply connected. On the other hand. see (4.3. p > 0.68).11.65 to conclude that h (M. ˜ χ(M. by Kobayashi’s original Theorem 6. C) = 1. O). and we may 0.11 also follows from the Kodaira vanishing theorem.12 and conjectured that closed and connected K¨hler manifolds with positive first Chern class are simply cona nected. 0). In his article [Ko1]. Kϕ. Hence M is closed as well. Kobayashi observed the application of the Kodaira vanishing theorem mentioned before Theorem 6. With respect to the in˜ duced K¨hler structure on M . O) = (−1)p h0. a non-negative line bundle. Then M is simply connected and the arithmetic genus χ(M.68). E) = 0 for q ≥ 1. Since π is a local isometry. .12. generalizing Definition 5. if KM is positive and E → M is the trivial complex line bundle or. O) = χ(M.6 Ricci Curvature and Global Structure 85 Proof. connected K¨hler mania fold with Ric > 0. Let M be a closed. see Corollary 7. If ϕ is a holomorphic form of type (p. Now the Calabi conjecture implies that closed K¨hler a manifolds with positive first Chern class carry K¨hler metrics with positive a Ricci curvature. compare (4. π is a local isometry.12 Theorem (Kobayashi [Ko1]). ϕ + M Kϕ. Hence they are in fact simply connected.59 ∗ slightly. O) = 1. Observe that Bochner’s argument above works in this case as well (Exercise). where Γ denotes the fundamental group of M . ϕ . then KM ⊗ E is positive.q apply Corollary 5. The last assertion in Theorem 6. 6.

Discuss Theorem 6. Compare with Theorem 6. see also [BGM.3 Ricci Curvature and Laplace Operator. X 2 ≤ 1 (µ − 2λ) X 2 2 2. The inequality (6. The next theorem completes the equality discussion in Theorem 6. connected Riemannian manifold of real dimension n. by the assumption Ric ≥ λ. If Ric ≥ λ > 0.1 X. Let ϕ ∈ E(M. a . Section III. by the Bochner identity 1.80) implies ∗ 0. The next theorem deals with a remarkable improvement of the above inequality in the K¨hler case.1 X 2 2 = 0. Let a be the complex Lie algebra of a automorphic vector fields on M and k ⊂ a be the real Lie subalgebra of Killing fields.14 in the case of the round S 2 (and observe thatn/(n − 1) < 2 if n > 2). 6.1 X = 0. X 2 = ∗ 0. that is. X := ξ = grad ϕ = 0 fulfills ∗ X = (µ − 2λ)X + (2λX − Ric X). the equality discussion to Obata [Ob]. Hence.D]. Let M be a closed K¨hler manifold with Ric ≥ a λ > 0.14 Theorem (Lichnerowicz). where the index 2 indicates L2 -inner products and norms.13) is due to Lichnerowicz [Li3]. 0≤ 0. 2 In the notation introduced there. a 6. Equality implies that 0. (6. Let M be a closed K¨hler manifold.86 ¨ Lectures on Kahler Manifolds 6.13) n−1 Moreover. Then ξ := dϕ satisfies ∆d ξ = µξ. Proof.1 X. R) be any non-constant eigenfunction.1 X= Therefore. Equality implies that the gradient field X = grad ϕ of any eigenfunction ϕ for λ1 is automorphic with Ric X = λX.38. then the first non-zero eigenvalue λ1 of ∆ satisfies n λ1 ≥ λ. Let M be a closed. and that Ric X = λX. equality holds iff M is a round sphere of radius (n − 1)/λ. We conclude that µ ≥ 2λ for all non-trivial eigenvalues µ of ∆. X is automorphic. Denote by ∆ the Laplace operator on functions on M .15 Exercise. (4. 1 (µ − 2λ)X + (λX − Ric X). so that ∆ϕ = µϕ for some µ > 0.16.14 in the case where M is a K¨hler–Einstein manifold. Then the first non-zero eigenvalue λ1 of ∆ satisfies λ1 ≥ 2λ.

by Exercise 6. Then we have an isomorphism grad: {ϕ ∈ E(M. Let M be a Riemannian manifold and ϕ be a smooth function on M . 6. Proof. We recall that. By the proof of Theorem 7. where ϕ is a smooth function on M with mean zero. . Derive an analogous formula for metric connections on vector bundles. by Proposition 4. R) | ∆ϕ = 2λϕ} =: E2λ → Jk.17 Exercise (Ricci Equation).16 Theorem (Matsushima). on functions. Let M be a closed K¨hler–Einstein manifold a with Einstein constant λ > 0.43 below. Thus the formula computes the commutator [ . By the above remark. Now X is automorphic. ϕ → grad ϕ. let X in Jk. Jk consists precisely of automorphic gradient vector fields. grad maps E2λ to Jk. Hence by the equality case in Theorem 6. Hence ∆ϕ = 2λϕ.17 and since ϕ and ∆ϕ have mean zero. hence ∗ X = Ric X = λX. ∆ = ∗ .6 Ricci Curvature and Global Structure 87 6. grad is injective on E2λ .14. Vice versa. Show that grad ∆ϕ = ( ∗ )(grad ϕ) + Ric grad ϕ. Write X = grad ϕ.79. We note first that grad is injective on the space of smooth functions on M with mean zero. ∗ ]. where we write instead of grad.

a 7. For example. where c is the restriction 16 My 17 For main sources for this are [Au3] and [Jo]. there is a unique K¨hler metric g with K¨hler form cohomologous a to ω and Ricci form ρ . Theorem 7. who also proved uniqueness of the solution g [Ca1]. 1)-form.1 concerns the existence of Ricci-flat K¨hler metrics: a a 7. Proof. Then g(X. We let M be a closed and connected K¨hler manifold with complex a structure J. if the anti-canonical bundle KM is ample17 . By assumption. In this case. Hence the Ricci curvature of g is positive. This was conjectured by Calabi. [Ka1]. however. JY ) is a K¨hler metric on M with K¨hler form ω.7 Calabi Conjecture In this section we discuss the Calabi conjecture and present a major part of its proof 16 . equivalently.6.3 works equally well in the case c1 (M ) < 0. Let ρ ∈ 2πc1 (M ) be a closed real (1. Then c1 (M ) = (n + 1 − d)c. then M has a unique Ricci-flat K¨hler metric g with K¨hler form ω cohomologous to ω. [Ca2].1 Theorem (Calabi–Yau). The argument for Corollary 7. If c1 (M ) = 0. 1). Hermitian symmetric spaces of compact type are Fano manifolds. 1) Let M ⊂ CP n be the (regular) set of zeros of a homogeneous polynomial of degree d. Since their first Chern class is represented by their Ricci form. we have the stronger Theorem 7. Riemannian metric g.5 and the Kodaira Embedding Theorem 9. Existence was proved by Yau [Ya2]. 1). One of the most striking immediate applications of Theorem 7. Recall Definition 5. see Theorem 6. Recall a that the Ricci form of any K¨hler metric on M is contained in 2πc1 (M ). M is finitely covered by a product of a simply connected closed complex manifold with vanishing first Chern class and a complex torus (where one of the factors might be of dimension 0). then M carries K¨hler metrics with a positive Ricci curvature. closed K¨hler mania folds with positive Ricci curvature are Fano manifolds. .2 Corollary. and Ricci form ρ. a As a consequence. see Exercise 9.6. K¨hler form ω. Other good references are [Bo1]. By a Theorem 7. for example ρ = ω. Say that a closed complex manifold M is a Fano manifold if c1 (M ) is ∗ positive or. M has a positive closed differential form ω of type (1. the equivalence.3 Corollary. Y ) = ω(X.59 of positive and negative cohomology classes of type (1. 7. If M is a Fano manifold.1.4 Examples. Choose a a a further positive differential form ρ ∈ 2πc1 (M ). [Ya3].14 below. [Si1].1 implies that Fano manifolds are characterized by positive Ricci curvature: 7. Then there is a unique K¨hler metric g on M with K¨hler form ω cohomoloa a gous to ω and with Ricci form ρ . [Ti3].

besides CP 1 × CP 1 . there is a smooth real function a ϕ such that 1 (7. By Example 5. [Ya1]. f is unique up to an additive constant. For any K¨hler form ω cohomologous to ω. Thus in our search for the K¨hler form ω we have replaced a the Ricci form ρ . In other words. see [De].3 and Theorems 6. which depends only on ω itself.50 and Exercise 5. The rich geometry of these surfaces is discussed in [Va.5) Vice versa. Moreover.7 Calabi Conjecture 89 of the first Chern class of the hyperplane bundle H over CP n to M . for example 104 for m = 3. a Recall that ω m = m! volg . g). there is a smooth function f : M → R with (ω )m = ef ω m . there are only finitely many diffeomorphism types of Fano manifolds of complex dimension m.9) ω − ω = ddc ϕ = i∂∂ϕ.159].60.51.6) (7. c > 0. by the function f . For such an ω given. there is a smooth function f : M → R such that 1 ρ − ρ = − ddc f = −i∂∂f.8) by the ddc -Lemma 5. the existence of K¨hler–Einstein metrics. these are the only closed complex surfaces with positive first Chern class [Hit]. for each m ≥ 1. 2) Let Mk be the blow up of CP 2 in k ≥ 0 points in (very) general position. then g and g have the same volume. With (4. Lectures 13–15]. see [Hir. where volg denotes the oriented volume form of g. From these examples one gets the impression that Fano manifolds might be rare and special. p. Before we go into the details of the proof of Theorem 7. 2 .20). for any closed real (1. (7.12 we conclude that Fano manifolds are simply-connected and that they do not admit any holomorphic p-forms for p > 0.3.64) we get ρ = −i∂∂ ln det G = −i∂∂ ln(ef det G) = ρ − i∂∂f.1.7) (7. 1)-form ρ cohomologous to ρ. which depends on second derivatives of the metric. c1 (M ) = 0 if d = n + 1. Moreover. see (4. Then c1 (Mk ) > 0 for 0 ≤ k ≤ 8. from Corollary 7. In fact. Moreover. we reformulate it into a more convenient analytic problem and discuss another question. ef has mean 1. Hence if ω is cohomologous to ω. 2 (7. We conclude that c1 (M ) > 0 if d ≤ n. Then the volume constraint turns into ef d volg = vol(M. and c1 (M ) < 0 if d > n + 1.11 and 6.

50. . To that end we note that with respect to local coordinates M (ϕ) := e = det A = det(gjk ) f −1 ∂2ϕ det gjk + ∂zj ∂zk ! . We have achieved the following reformulation of Theorem 7.13. (7. Zm ) of T M as in (2.10). det A = 0 in contradiction to (7.13).13) This is a non-linear partial differential equation of Monge–Amp`re type. its fundamental matrix is (gjk ).20) imply that det A = ef = 0 at each point of M . In such a point. (7. where we note that this Hermitian metric is conjugate linear in the second variable.12) It follows that all eigenvalues of A are positive at a point where ϕ attains a minimum.10) holds.13) up to an additive constant f + c instead. there is a smooth real function ϕ on M such that ln M (ϕ) = f. We want to show that at each point of M all eigenvalues of A are positive.25. e The solution ϕ of (7. Consider the Hermitian metric on T M given by (Z.10) and (4. the assumption on the mean of ef is automatically fulfilled. Since ef is non-zero everywhere. a . . see Proposition 7. Thus we seek ϕ with ef ω m = (ω )m = (ω + i∂∂ϕ)m and ω = ω + i∂∂ϕ > 0. Suppose now that there is a point p ∈ M such that A has a negative eigenvalue at p. W . we may delete the normalizing assumption on the mean and ask for a solution of (7. With respect to a coordinate frame (Z1 . Y ) = ω (X. Such points exist since M is compact. namely the question of the existence of K¨hler–Einstein metrics.10) (7. (7. W ) = Z. JY ) is a K¨hler metric on M . ·). we discuss a related problem. We write the Hermitian metric (·. In other words. Before we discuss the proof of Assertion 7. ·) = (A·. ·) on T M induced by g as (·. .11) a By the latter we mean that g (X. In a first step we show that ω > 0 if (7. any such form ω is cohomologous to ω. . This point of view will be useful further on. (7.90 ¨ Lectures on Kahler Manifolds again by the ddc -Lemma 5.13) is unique up to constant functions. Since ω = ω + i∂∂ϕ is cohomologous to ω. Vice versa.1: Given a smooth function f on M such that ef has mean 1. Then on a path from p to a point where ϕ attains a minimum. there is a point where A has 0 as an eigenvalue. where A is a Hermitian field of endomorphisms of T M .

and hence. .15) Replacing ϕ by const/λ+ϕ. The case of Ricciflat K¨hler metrics is treated in Corollary 7. Theorem 7. Both equations. With this question out of the way. Let M be a closed complex manifold with negative first Chern class. we discuss it briefly in Subsection 7. If not specified otherwise. there is a unique smooth real function ϕ on M such that ln M (ϕ) = −λϕ + f. (7. inner products and ∗-operator are taken with respect to the metric g. suppose ϕ solves (7. Vice versa. (7. hence ω is cohomologous to ω. there is a smooth real function f on M . such that ρ − λω = i∂∂f . In conclusion.13) above. Suppose c1 (M ) < 0 and choose a constant λ < 0. ϕ(p)).14 is a consequence of the following assertion: Given λ < 0 and a smooth function f on M . The case of positive first Chern class is complicated.17). the same argument as further up shows that for any smooth solution ϕ of (7.51. Since the Ricci form ρ ∈ 2πc1 (M ). ϕ)) > 0. are of the form ln M (ϕ) = f (p. unique up to an additive constant.18) (7. Since exp(f (p.2. by Exercise 5.14 into an analytic problem similar to (7.18). the constant on the right side vanishes. a Theorem 7.14 was conjectured by Calabi and proved independently by Aubin [Au2] and Yau [Ya2].16) (7. JY ) is a K¨hler metric on M with K¨hler form ω. (7. 1)-form. a a Let g be a K¨hler–Einstein metric on M with K¨hler form ω and Einstein a a constant λ.13) and (7.7 Calabi Conjecture 91 7.17) where f = f (p. Then up to a scaling constant. [Ya3]. we can now concentrate on the solvability of our equations. 18 It is understood that f (p. M has a unique K¨hler–Einstein metric (with negative Einstein constant ). We want to solve ρ = λω . 1) with λω ∈ 2πc1 (M ).18 below why ω +i∂∂ϕ is positive. We explain after Equation 7. ϕ) = f (p. Then λω = ρ ∈ 2πc1 (M ). ω + i∂∂ϕ is a positive (1. The latter gives −i∂∂ ln M (ϕ) = ρ − ρ = λ(ω − ω) − i∂∂f = λi∂∂ϕ − i∂∂f. a We now reformulate Theorem 7. Let ω be a positive real differential form of type (1. ln M (ϕ) = −λϕ + f + const. ϕ). and therefore ω − ω = i∂∂ϕ. Then the above computation shows that ω = ω +i∂∂ϕ solves ρ = λω . ϕ) is a smooth real function 18 . Then g(X.4 below. Y ) = ω(X.16).14 Theorem (Aubin–Calabi–Yau).

The second equation follows from the first and Equations 4. m (1) (2) Note that if η is of type (m − 1.19) where a1 . In particular. grad ϕ ∗η vol .92 ¨ Lectures on Kahler Manifolds 7. Given a second K¨hler metric g about p. there a exists a centered holomorphic chart z = (z 1 . 1)). = dc ϕ. m − 1) and ∗η is non-negative (of type (1. z m ) about p with gp = 1 2 j dzp j dzp . Proof of Lemma 7. (7. Hence (dc ϕ) = J · grad ϕ. Let p ∈ M . Let ϕ be a smooth real function on M . . If η is a differential form of degree 2m − 2. J grad ϕ) vol. ωp = i 2 j j dzp ∧ dzp . Proof. Correspondingly. .22. (7. then dϕ ∧ dc ϕ ∧ η = ∗η(grad ϕ. 1)-form. (7.22). Immediate from (7.23 Lemma. . ∗(ω m−j−1 ∧ (ω )j ) is a positive (1. . .20) The volume forms are related by vol (p) = ( aj ) · vol(p).1 Uniqueness.21) m−j−1 We also see that ωp ∧ (ωp )j is a positive linear combination of the forms im−1 1 1 m m k k dz ∧ dzp ∧ · · · ∧ dzp ∧ dzp ∧ · · · ∧ dzp ∧ dzp . With respect to the ∗-operator of g. . and therefore dϕ ∧ dc ϕ ∧ η = dϕ ∧ dc ϕ.20 and 4. J grad ϕ) vol .24 Lemma. J grad ϕ) ≥ 0. am are positive numbers. . ωp = i 2 j l aj dzp ∧ dzp . gp = 1 2 j aj dzp j dzp . This proves the first equation. 7. We have dc ϕ = −dϕ ◦ J. . then ∗η(grad ϕ.24. 2m−1 p where behatted terms are to be deleted. ∗η vol = ∗η(grad ϕ.22) 7. dϕ ∧ dc ϕ ∧ ω m−1 = 1 | grad ϕ|2 ω m . (7.

ϕ1 ) − ef (p. Set ω1 = ω + i∂∂ϕ1 and ω2 = ω + i∂∂ϕ2 . 7.3 since we will need a priori estimates of their higher order derivatives. the regularity of solutions will play an essential role in Subsection 7. . then ϕ1 − ϕ2 is constant.41).26) where f = f (p.17 are unique and solutions to Equation 7.13 are unique up to an additive constant.2 Regularity. where we integrate over all of M .7 Calabi Conjecture 93 We now come to the uniqueness of solutions of (7. then both terms on the right hand side are non-negative. solutions to Equation 7. We want to study the solvability of the equation F (ϕ) = 0 using the continuity method. Let U ⊂ C 2 (M ) be the open subset of functions ϕ such that ω + i∂∂ϕ is positive. compare (7. ϕ) is a smooth real function. It is related to the dependence of f on the variable ϕ.18).23 and 7. In particular. Let F : U → C 0 (M ) be the functional F (ϕ) = ln M (ϕ) − f (p.25. 7. Suppose ϕ1 and ϕ2 solve (7. Applying Stokes’ theorem and Lemmas 7.ϕ2 ) )ω m . Then m ω1 = ωm = m ω2 . where the index g1 means that we take the norm with respect to the metric g1 . (7. ϕ). To that end.24 we get 0= = m−1 m−2 m−1 + ω1 ∧ ω2 + · · · + ω2 )) d((ϕ1 − ϕ2 )dc (ϕ1 − ϕ2 ) ∧ (ω1 m−1 m−2 m−1 d(ϕ1 − ϕ2 ) ∧ dc (ϕ1 − ϕ2 ) ∧ (ω1 + ω1 ∧ ω2 + · · · + ω2 ) + ≥ 1 m m m (ϕ1 − ϕ2 )(ω1 − ω2 ) m | grad(ϕ1 − ϕ2 )|21 ω1 + g (ϕ1 − ϕ2 )(ef (p. 2) If f is strictly monotonically increasing in ϕ.25 Proposition.18). then ϕ1 = ϕ2 . If f is weakly monotonically increasing in ϕ. Hence 0=2 M m m (ω1 − ω2 ) = m−1 m−2 m−1 ddc (ϕ1 − ϕ2 ) ∧ (ω1 + ω1 ∧ ω2 + · · · + ω2 ). 1) If f is weakly monotonically increasing in ϕ. Proof of Proposition 7.40) and (7.

We also recall that in the case of λ = 0. compare Exercise 5. the uniformization of closed Riemann surfaces.31 Remark. In the case of complex dimension m = 1. (7. 1). C k+α for any k ≥ 6. f is a given smooth function on M . the directional derivative of F at ϕ ∈ U in the direction of ψ is 1 (F (ϕ + tψ)) |t=0 = − ∆ϕ ψ − fϕ (p. From now on. we only consider Equations 7.3 Existence. 19 In [Ka2]. a continuous differential form. the function 7. It follows that F is continuously differentiable with derivative dF (ϕ) · ψ given by the right hand side of (7. It is a good exercise to follow the argument below in this more elementary case. Kazdan gives a nice exposition (with references) of this theory.27) the Laplacian of gϕ . we let gϕ be the associated continuous Riemannian metric as in (4. 2 In this case. Proof. Hence ϕ is C 4+α and. Since ϕ ∈ U .1.51. For ϕ ∈ U and ψ ∈ C 2 (M ). Equation 7.13 and 7.17).13) and λ < 0 in the case of (7. 1). ϕ) · ψ.30) where U is as above. the arguments below are much easier.28). α ∈ (0. This is still a non-trivial case. F (ϕ) = ln M (ϕ) + λϕ − f = 0 on U ∩ C 2+α (M ). We call jk ∆ϕ = −2gϕ ∂2 . ∂zj ∂zk (7. The same statements also hold for F when considered as a functional from U ∩ C k+2+α (M ) to C k+α (M ).2). 2 (7. 1).94 ¨ Lectures on Kahler Manifolds For ϕ ∈ U and ωϕ = ω + i∂∂ϕ. By (7. Since ωϕ is of type (1. and λ = 0 in the case of (7. 7.28).17.29 Proposition. 7. Since f is smooth and ϕ is C 2+α .30 becomes 1 ef −λϕ g = g − ∆ϕ.28) where fϕ denotes the partial derivative of f in the direction of the variable ϕ. ϕ is an elliptic solution of the equation F (ϕ) = 0. If ϕ ∈ U ∩ C 2+α (M ) solves F (ϕ) = 0. then ϕ is smooth. det(gjk )−1 det(gjk + ϕjk ) − f (p. where k ≥ 0 is an integer and α ∈ (0. recursively. . ϕ) is C 2+α in p ∈ M and the (free) variables ϕjk . by the regularity theory for elliptic solutions of partial differential equations 19 . we are only interested in f modulo an additive constant. ∆ϕ is elliptic. gϕ is compatible with the complex structure of M .

then the kernel of dFt (ϕ) on L2 (M ) is trivial and hence dFt (ϕ) is an isomorphism from its domain H 2 (M ) of definition onto L2 (M ).30.30.30. Then ϕ ∞ ≤ 1 f |λ| ∞. or.12).33 Lemma. Hence by the implicit function theorem. We need to derive a priori estimates of solutions ϕ of Equation 7. there is a unique solution ϕs of Equation 7. a closed complement of the constant functions in L2 (M ). Z ∩ H 2 (M ) and Z ∩ C k+α (M ) are closed complements of the constant functions in H 2 (M ) and C k+α (M ).s which is close to ϕt in C 2+α (M ).30). That is. In the case λ = 0. 1] for which a C 2+α -solution of Equation 7. we consider the equation Ft (ϕ) = ln M (ϕ) + λϕ − tf = 0 on U ∩ C 2+α (M ). and hence λϕ(p) ≥ f (p). 7.30.t. for all s close enough to t. Therefore.t. what will amount to the same. by the regularity theory for elliptic differential operators. (7. Then M (ϕ)(p) ≤ 1. by (7. the closedness of Tα . We let 0 ≤ t ≤ 1 and replace f by tf in (7.t exists (modulo a constant function if λ = 0). by the regularity theory for linear partial differential operators.7 Calabi Conjecture 95 For the proof of the existence of solutions we apply the method of continuity. We distinguish between the cases λ < 0 and λ = 0. Hence Tα is open in this case as well.30. Hence Tα is open. Hence dFt (ϕ) : C 2+α (M ) → Z ∩ C α (M ) is surjective. gϕ ). We start with the openness and let t ∈ Tα and ϕ be a solution of Equation 7. dFt (ϕ) : C 2+α (M ) → C α (M ) is an isomorphism as well. By the spectral theorem. by the spectral theorem.30.32) 2 defines a self-adjoint operator on L2 (M. Let λ < 0 and suppose that ln M (ϕ) = −λϕ + f . such that ϕs solves Equation 7. Hence Tα is not empty. If λ < 0. The aim is now to show that Tα is open and closed. for all k ≥ 0. We let Z ⊂ L2 (M ) be the space of functions with gϕ -mean zero.t) We let Tα be the set of t ∈ [0. The constant function ϕ = 0 solves 7. Now 1 dFt (ϕ) = dF (ϕ) = − ∆ϕ + λ (7. dFt (ϕ) = −∆ϕ /2 is surjective from H 2 (M ) onto Z.30. Let p ∈ M be a point where ϕ achieves its maximum. for s close enough to t. of Equation 7. there is a family of functions ϕs . ϕ ≤ ϕ(p) ≤ 1 1 f (p) ≤ f λ |λ| ∞. .s up to constant functions. Hence by the implicit function theorem. Since constant functions are smooth. Since λ < 0. we only need to solve modulo constant functions. We now come to the heart of the matter.30. Proof.0.

Stokes’ theorem gives M d{ϕ|ϕ|p−2 dc ϕ ∧ (ω m−1 + · · · + (ω )m−1 )} = 0. 2 For p ≥ 2. It will become clear in the proof that the constant C is explicitly computable in terms of C(f ) and dimension. grad |ϕ|p/2 2 2 ≤ mp2 C(f ) ϕ 2(p − 1) p−1 p−1 . where we note that ϕ|ϕ|p−2 is C 1 for p ≥ 2. where C is a constant which depends only on M . Suppose that ϕ is a smooth function on M with ln M (ϕ) = f and mean 0 with respect to g. hence (p − 1) M |ϕ|p−2 dϕ ∧ dc ϕ ∧ (ω m−1 + · · · + (ω )m−1 ) = 2 M (1 − ef )ϕ|ϕ|p−2 ω m . Recall that the latter holds iff (ω )m = ef ω m .34.96 ¨ Lectures on Kahler Manifolds If ϕ achieves a minimum at p.34 Lemma. Now d(ϕ|ϕ|p−2 ) = (p − 1)|ϕ|p−2 dϕ. then M (ϕ)(p) ≥ 1. and an upper bound for C(f ) := 1 − exp ◦f ∞ .35) . we conclude that. From |ϕ|p−2 (| grad ϕ|2 + Φ)ω m . We have (1 − ef )ω m = ω m − (ω )m = (ω − ω ) ∧ (ω m−1 + ω m−2 ∧ ω + · · · + (ω )m−1 ) 1 = − ddc ϕ ∧ (ω m−1 + ω m−2 ∧ ω + · · · + (ω )m−1 ). 7. We let ϕ be a smooth function on M with ln M (ϕ) = f . Then ϕ ∞ ≤ C. Therefore 1 1 f ∞. M 1 2 p−2 p |ϕ| | grad ϕ|2 = | grad |ϕ|p/2 |2 4 and ω m = m! volg . (7. and hence λϕ(p) ≤ f (p). for p ≥ 2. ϕ ≥ ϕ(p) ≥ f (p) ≥ − λ |λ| The case λ = 0 is much harder. Proof of Lemma 7.24. volume. g. By Lemma 7. the left hand side is equal to p−1 m where Φ ≥ 0. and Sobolev constants of M .

By (7. grad ϕ 2 2 ≤ 2mC(f )C2 vol(M )1/2 grad ϕ 2 . With C3 := 2mC2 vol(M )1/2 . With α = (m + 1)/m and p ≥ 2α. ϕ p }. applied in the case p = 2.38) We now apply an iteration argument to get the desired uniform bound for ϕ. (7. there is a constant C1 with ϕ 2m+2 m ≤ C1 · ( ϕ 2 + grad ϕ 2 ). we have 1/α ϕ p pα = M 2 ≤ 2C1 · 2 ≤ 2C1 · |ϕ|pα ϕ ϕ p p p p = M |ϕ| 2 2α 2 2 p 2 2α = |ϕ|p/2 2 2α mp2 C(f ) ϕ p−1 p−1 2(p − 1) mp2 2 C(f )(vol M )1/p ϕ ≤ 2C1 · ϕ p + p 2(p − 1) ≤ C4 p · max{1. By the Sobolev embedding theorem. by (7. by (7.7 Calabi Conjecture 97 where the index refers to the corresponding integral norm. we have From now on we assume that ϕ has mean 0 with respect to g.35). Hence.38.37). Hence.36). there is a constant C ≥ 1 such that ϕ We choose C such that C · (C4 p)− 2α ≤ C · (C4 2α)− m+1 p m+1 2α . ϕ 2m+2 m ≤ C1 (C2 + 1)C3 C(f ).36) 2 where 1/C2 is the first eigenvalue of the Laplacian of (M. g).37) for the second estimate. ≤ vol(M )1/2 · ϕ 2 . ≥1 . (7.37) grad ϕ Now ϕ 1 2 2 ≤ 2mC(f ) ϕ 1 . where we apply (7. Then we have ϕ 2 ≤ C2 · grad ϕ 2 . p + + grad |ϕ|p/2 p−1 p By Equation 7. (7. we get grad ϕ 2 ≤ C3 C(f ) and ϕ 2 ≤ C2 C3 C(f ).

. f C 3(M) .41) and Exercise 5. There are two further a priori estimates. g). [Ya3] or [Au3.1 we conclude that ∆ϕ 20 An ∞ ≤ C ( ∆ϕ ∞.41). [Ya3] or [Au3. (7. Lemmas 7. The first of the two remaining a priori estimates concerns the g-Laplacian of ϕ. In the derivation of this latter estimate. M. [Au2]. this is equivalent to the estimate in (7.10]. f C 3 (M) . g). their derivation is space and time consuming and will therefore not be presented here. For λ ≤ 0 fixed. From (7. in particular.30. M.40) gives estimates between norms associated to g and gϕ . hence (7. the reader should be well prepared to turn to the literature to get the remaining details.5). §7. By what we said above. They are less critical20 and hold.40) gives a uniform upper bound for gϕ against the given metric g. The derivation of this estimate is. In particular.42) examination of the case m = 1 is recommended. (7. (7. This is the main a priori estimate. we then get that ϕ and hence that ϕ Since ϕ p pα pα pα α ≤ C4 pα C pα (C4 p)−(m+1)α = C pα (C4 p)−m−1 → ϕ ∞ as p → ∞.39) where we assume that the mean of ϕ is 0 if λ = 0. |λ|.41) [Ca3]. g).40) see [Au2]. ϕ ∞ ≤ C( f ∞ . (7. M. At this stage. among others. (7. (7. ∆f ∞ . g).11]. one actually estimates ddc ϕ first in the C 0 -norm associated to gϕ .34 give an priori estimate on the C 0 -norm of solutions ϕ of Equation 7. for positive λ as well.33 and 7. a true fight against notation and in our sources. see (7. The second a priori estimate concerns the g-covariant derivative of ddc ϕ. |λ|.51. §7. Let 0 < β < 1.98 ¨ Lectures on Kahler Manifolds for all p ≥ 2α and assume recursively that ϕ p ≤ C · (C4 p)− m+1 p for some p ≥ 2α. ∆ϕ ∞ ≤C( ϕ ∞. However. Now the volume elements of gϕ and g only differ by the factor exp ◦f .40) gives also a lower bound for gϕ against g. |λ|. Since mα = m + 1. ddc ϕ ∞ ≤ C ( ∆ϕ ∞. By the definition of the metric gϕ associated to ϕ. ∞ ≤ C. f ∞. the authors refer to the original articles cited above. we conclude that ϕ ≤ C(C4 p)− m+1 pα . M.

but to replace ϕ with tϕ. Hence the linearized operator ∆t − 2λt is invertible for all t ∈ [0. We are ready for the final step in the proof of Theorems 7. 0 ≤ t ≤ 1.16. see Theorem 6. therefore Rict > λtgt for t < 1 and Ric1 = λg1 . Now the inclusion C 2+β (M ) → C 2+α (M ) is compact. When applying the continuity method as explained in the previous subsection.30. we assume that ϕn has mean 0.14. the a priori estimates (7. However.3. Concerning the openness of Tα . 7. 1].40) is critical and might fail. the arguments break down to a large extent. In the case λ = 0. In the λ ≤ 0 case. Using the Schauder estimates.14. Hence Tα is closed. Observe that if the manifold admits automorphic vector fields.t. whose norms are bounded by the corresponding norms of f . In the case of positive first Chern class.30. Indeed. where the bound now also depends on β. there is a uniform bound on the C 2+β -norm of the functions ϕn . The limit ϕ = lim ϕn solves 7. it is no longer possible to show openness or closedness of Tα along the lines of what is said there.14. we consider the family of functions tf . This is enough for the continuity method to work. As for the closedness of Tα . It remains to show that Tα is closed. the operator ∆ϕ −2λ might have a kernel for λ > 0. Even though a general criterion for the solvability of the K¨hler–Einstein a .41) do not depend on the sign of λ and continue to hold. Then the equation for t = 0 is highly non-trivial. we obtained existence of K¨hler–Einstein metrics on closed K¨hler manifolds assuming non-positve a a first Chern class. Let tn ∈ Tα and suppose tn → t ∈ [0.40) and (7. the remaining C 0 -estimate (7. Let ϕn be a solution of Equation 7. the linearized operator has a kernel for t = 1. so that we obtained a (trivially) solvable equation for t = 0. see Theorem 6.1 and 7.4 Obstructions. As we will see. ϕn converges in C 2+α (M ). A simple calculation shows that if ωt is a solution. Then by the above. this can be deduced from special properties of their automorphism groups. we replaced f with tf . we conclude that ϕ is bounded in the C 2+β norm associated to g. Hence by passing to a subsequence if necessary. 1) for which a solution exists. Note that in our application of the continuity method. Let 0 < α < β < 1. then the Ricci form ρt of ωt is given by ρt = λtωt + λ(1 − t)ω0 . we need to solve the Monge–Amp`re equation ln M (ϕ) + λϕ − f = 0 with λ > 0 e and a given smooth real function f . Here the trick is to keep f . This problem can be overcome [Au4]: In the notation of Subsection 7. in place of the original f . but solvable according to our previous results for λ = 0.2 and Theorem 7.tn . there are closed complex manifolds with positive first Chern class which do not admit K¨hler–Einstein a metrics at all.7 Calabi Conjecture 99 Hence ∆ϕ is uniformly bounded in the C β -norm associated to the fixed metric g. a kind of guiding theme: Non-discrete groups of automorphisms yield obstructions to the existence of K¨hler–Einstein meta rics. In Corollary 7. provided we can ensure that Tα is closed.

38 imply that dd∗ η + d∗ dζ = ∆d η + ∆d ζ = ∆d ξ = 2 Ric ξ = 2λξ = 2λη + 2λζ. multiplication by J yields an isomorphism k → s of real vector spaces. In the remainder of this subsection. If the Einstein constant of M is positive. the Ricci curvature of M is positive.1. Y and Z are automorphic. Let M be a K¨hler manifold and X be a real vector field on a M. 2) If X is symmetric. Then a = k + Jk. Hence Y is symmetric and Z is skew-symmetric. the underlying ideas related to holomorphic actions of complex Lie groups seem to be at the heart of the (still conjectural) full solution. Again by Proposition 4. In other words. Hence a = k + Jk. where η ∈ im d and ζ ∈ ker d∗ . Let X be a vector field on M and ξ := X . Our first goal is to prove that Y and Z are automorphic as well. By Exercise 7. by Theorem 4.100 ¨ Lectures on Kahler Manifolds problem cannot possibly be based on the automorphism group alone – manifolds with trivial and non-trivial automorphism groups can. 7. Let M be a closed K¨hler–Einstein mania fold.81.43.44 Exercise. then JX is volume preserving. Now Proposition 4. Let s be the real vector space of all holomorphic vector fields Y such that Y is symmetric. ξ = η + ζ.43 Theorem (Matsushima [Ma1]). and hence ∆η = 2 Ric η and ∆ζ = 2 Ric ζ. Assume now that X is automorphic. η ∈ im d and ζ ∈ ker d∗ . A connected Lie group is called . 7. Now k ∩ s is the space of parallel holomorphic vector fields. A Lie algebra is called reductive if it is isomorphic to the direct sum of its center and a semi-simple Lie algebra. so dd∗ η = 2λη and d∗ dζ = 2λζ according to the Hodge decomposition theorem. and then M does not carry parallel vector fields. 7. If λ > 0. where λ is the Einstein constant of M . Hence a = k ⊕ Jk if λ > 0.45. then a = k ⊕ Jk. 3) Visualize 1) and 2) pictorially if dimC M = 1. we will treat the two classical obstructions against positive K¨hler–Einstein metrics that rely on the automorphism a group.45 Exercise. Let a be the complex Lie algebra of automorphic vector fields on M and k ⊂ a be the real Lie subalgebra of Killing fields. Y := η is a gradient field and Z := ζ is volume preserving (recall that div Z = −d∗ ζ).79.79 and the Bochner identity 1. therefore a Killing field. On the other hand. then JX is a Killing field if and only if X is symmetric. Proof of Theorem 7. 1) If X is holomorphic. Now Z is a volume preserving automorphic field. By Hodge theory. Verify this explicitly for the Fubini–Study metric on CP m . but need not be obstructed –.

Let a and k be as above. 0 ∗ ∗ The center of a is trivial. Hence the argument in the proof of Corollary 7. Assume that ω ∈ 2πc1 (M ) and write ρ − ω = i∂∂Fω with Fω ∈ E(M. In particular.46 Corollary.43. Hence a is not reductive. see [Hit]. The blow up of a a CP 2 in one or two points carries K¨hler metrics with Ric > 0. a By an analogous argument. k = z ⊕ [k. we leave this as an exercise.45. the blow up of CP 2 in two points does not carry a K¨hler–Einstein metric either. Proof. the center of a is z + Jz. the once and twice blown a up CP 2 do not even admit K¨hler metrics of constant scalar curvature. Let M be the once blown up CP 2 as in Example 2. Let M be a closed K¨hler manifold with K¨hler form ω and Ricci form ρ. 7. Then the map Fω : a → R. Therefore a = k + Jk = (z + Jz) ⊕ ([k. k] of k is semi-simple. a is not semi-simple since the space of matrices in a with zero entries in second and third row is an Abelian ideal. hence k ∩ Jk ⊂ z. Since k is the Lie algebra of a compact Lie group.49 Theorem (Futaki [Fu]). Fω (X) := X(Fω )ω m .47 Example. where z is the center of k. k] + J[k. 7. R).46 applies and shows that the identity components of the automorphism groups of such manifolds are also reductive. If M is a closed K¨hler–Einstein manifold. see [Li2]. We note the following consequence of Theorem 7. We now discuss a second obstruction to the existence of positive K¨hler– a Einstein metrics which is also related to the automorphism group. we a still have a = k+Jk. k]. Now k ∩ Jk consists of parallel vector fields. Moreover. However. [Ya1].43. a . the blow up of CP 2 in one point carries Riemannian metrics with Ric > 0 and sectional curvatures ≥ 0 [Che] and Einstein metrics with Ric > 0 which are conformal to K¨hler metrics [PP].48 Remark. the commutator subalgebra [k. and [k. M does not depend on the choice of the K¨hler form ω ∈ 2πc1 (M ). For closed K¨hler manifolds of constant scalar curvature. k]). Moreover.1. and hence M does not carry K¨hler–Einstein metrics. Since a = k + Jk.7 Calabi Conjecture 101 reductive if its Lie algebra is reductive. The Lie algebra a of the automorphism group of M consists of the matrices   0 ∗ ∗ 0 ∗ ∗ ∈ C3×3 . a a 7. k] is semi-simple. then the compoa nent of the identity of the automorphism group of M is reductive. a 7. see Exercise 7. k] + J[k.

a Proof of Theorem 7.1 (M. and is unique up to a constant. . with ω ∈ 2πc1 (M ) an arbitrary K¨hler form. we mostly suppress the parameter t in our notation and ˙ indicate differentiation with respect to t by a dot or by ∂t . we have m ˙ 2∂t ωt = 2mω ∧ ω m−1 = 2mi∂∂ϕ ∧ ω m−1 = −∆ϕ · ω m . Futaki [Fu.51. From the proof of the ddc -Lemma 5. hence − ln M (Φt ) = Ft + Φt − F0 + constant.49. Hence the Matsushima criterion for the non-existence of K¨hler– a Einstein metrics does not apply in this case.15). The Futaki invariant of both the once and the twice blown up CP 2 is non-zero. From now on. R). since ρ ∈ 2πc1 (M ) ∈ H 1. a Let H1 → CP 1 and H2 → CP 2 be the hyperplane bundles. Thus 2∂t Fωt (X) = 2 = M M {(Xf )ω m + (XF )∂t ω m } X(∆ϕ − 2ϕ) − (XF )∆ϕ ω m . see [Ti3. then Fω is a constant and the Futaki invariant Fω vanishes on a. whereas the Futaki obstruction does.51. −ε < t < ε. §3] shows that M is a Fano manifold with reductive automorphism group but non-zero Futaki invariant. As in (7. By Exercise 7. we compute −i∂∂ ln M (Φt ) = ρt − ρ0 = i∂∂(Ft + Φt − F0 ). The map F = Fω . a closed and complex manifold of complex dimension 4.2]. ω = ρ. Section 3. If ω is a K¨hler–Einstein metric.102 ¨ Lectures on Kahler Manifolds We remark that the function Fω in the definition of the map Fω exists. ωt − ω0 = i∂∂Φt and ρt − ωt = i∂∂Ft for families of smooth functions Φt and Ft .50. compare Exercise 5. By Exercise 5.51. With ϕ = Φ and ˙ we get ω = i∂∂ϕ and ρ − ω = i∂∂f . The goal is to show that ∂t Fωt (X) = 0. see Exercise 5. We let ωt . be a smooth family of K¨hler forms in 2πc1 (M ). Hence we conclude again that these two do not admit K¨hler–Einstein metrics.50 we see that these families of functions can be chosen to be smooth in t. Let M be the total space M of the projective bundle of the product H1 ×H2 → CP 1 ×CP 2 . is called a the Futaki invariant of M . The displayed equation above gives ˙ ˙ f =F ˙ ∆ϕ = 2f + 2ϕ.

grad ϕ). grad F . Now div((XF ) grad ϕ) = grad(XF ). grad ∆ϕ {Ric(X.17 and Proposition 4. we get 2∂t Fωt (X) = = M 2 M F (J grad ϕ. grad ϕ) + Ric(X. It follows that X(∆ϕ − 2ϕ) = 2 =2 M M M Ric(X. grad ϕ 2 M = F (grad ϕ. because of ρ − ω = i∂∂F and Exercise 7.7 Calabi Conjecture 103 From now on.50. grad F = J grad ϕ X. hence (XF ) ∆ϕ = M M grad(XF ). X) + 2 F (J grad ϕ. . JX) + J grad ϕ JX. grad ϕ) − X. J grad F = − J grad ϕ JX. grad ϕ − (XF )∆ϕ. we suppress the measure of integration since it is always ω m . grad F .45. grad ϕ X. grad F (J grad ϕ)(JX(F )) = − (JX(F )) div(J grad ϕ). X(∆ϕ) = M M X.79. J grad ϕ) (i∂∂F )(X. grad ϕ) + X. grad ϕ ρ(X. JX)}. grad ϕ } =2 by Exercise 6. M ∗ ∗ = M (grad ϕ) } = M X. hence ∂t Fωt (X) = 0. J grad ϕ) =2 M = M { 2 F (grad ϕ. Because X is holomorphic. see Exercise 7. Adding up. X) + grad ϕ X. Since X is automorphic and J is parallel. M Now div J grad ϕ = 0.2. J grad ϕ) − ω(X. {Ric(X.

or Section 1. where γω a is the unique ω-harmonic form in 2πc1 (M ). see [Na]. For a discussion of the more general Calabi–Futaki invariant. These are precisely the a ones whose automorphism group is not reductive and whose Futaki invariants do not vanish. we a refer to [Bo3] and [Ti3]. Section 3.104 ¨ Lectures on Kahler Manifolds 7.4. JY ). [Si2].49 follows Futaki’s original argument.I in [Bes]. JY ) = 2 (1) 2 ϕ(X. only the blow ups of CP 2 in one or two points do not admit K¨hler–Einstein metrics [Si2]. 7. Calabi generalized the Futaki invariant to general closed K¨hler a manifolds M : Fix a K¨hler metric ω on M and write ρω = γω +i∂∂Fω . a [Ti1] and compare Example 7. More recent references are [Bi] and [Th]. All these references contain extensive bibliographies as well. Our proof of Theorem 7. 2i∂∂ϕ(X.6 in [Bo3]. Show that 2mi∂∂ϕ ∧ ω m−1 = −∆ϕ · ω m . see for example Chapter 2. JY ).2.1. Y ) = i∂∂ϕ(X. Y ) + ϕ(JX. [TY]. For surveys on K¨hler–Einstein metrics. Tian. It vanishes if the a K¨hler class of ω contains a K¨hler metric with constant scalar curvature. M (7. There are examples. (2) The complex Hessian of ϕ is defined by Hessc ϕ(X.52) does not depend on the choice of ω in its K¨hler class. The a a previously defined Futaki invariant is the special case where ω ∈ 2πc1 (M ) (if the latter is positive definite). the Fermat hypersurfaces d d z0 + · · · + zn = 0 in CP n carry K¨hler–Einstein metrics if (n + 1)/2 ≤ d ≤ n.51 Remark. [Ti1]. [Ti2]. CF ω (X) := X(Fω )ω m . Then the Calabi–Futaki invariant CF ω : a → R. For example. the situation is much more complicated. Write a down Hess ϕ = 2 ϕ and Hessc ϕ in K¨hler normal coordinates.4. Let M be a K¨hler manifold and ϕ be a smooth function on a M .1 in [Ti3]. It follows from work of Siu. In higher dimensions. where the C 0 -estimates in the continuity method could be verified.50 Exercise. . and Yau that among the Fano surfaces as in Example 7.

Choose an origin o ∈ M and let r = d(o. Let α be a differential form on M with values in E and µ : R+ → R+ be a non-decreasing function. we say that α is d(O(µ(r))) if there is a differential form β on M with α = dβ and such that β is O(µ(r)). 1) Let M = Rn with Euclidean metric and origin 0. 8. 0).2) where we consider the norm |α| = |α(p)| as a function on M .1) if the sectional curvature K of M is negative. The answer is yes if m ≤ 2. Then on B = B(r. Let M be a complete and connected Riemannian manifold and E be a flat Hermitian vector bundle over M . voln B = B α= ∂B β ≤ max{|β(p)| | p ∈ ∂B} voln−1 ∂B. r voln B = . (8. Following Gromov. . α = dx1 ∧ · · · ∧ dxn . Let α be the volume form of M . in the case where β can be chosen to be uniformly bounded. We also write d(bounded) instead of d(O(1)). We say that α is O(µ(r)) if there is a constant c such that |α| ≤ cµ(cr + c) + c. Dodziuk and Singer remarked that Atiyah’s L2 -index theorem implies (8.1) if the space of square integrable harmonic forms on the universal covering space of M vanishes in degree = m and does not vanish in degree = m. in this section we are concerned with his arguments and a results. Gromov proves this and more in the case where M is a K¨hler manifold.3 Examples. Then |df | = |α| ≤ α ∞ := sup{|α(p)| | p ∈ M }.8 K¨hler Hyperbolic Spaces a A question attributed to Chern asks whether the Euler characteristic of a closed Riemannian manifold M of dimension 2m satisfies (−1)m χ(M ) > 0 (8. voln−1 ∂B n It follows that max{|β(p)| | p ∈ ∂B} ≥ Hence α is not d(bounded). In [Gr]. where I is a multi-index. In a similar way one can treat each of the differential forms dxI . see [Chr] for the case m = 2. ∞ Hence f is Lipschitz continuous with Lipschitz constant α if α ∞ < ∞. · ) be the distance to o. 2) Let α be a bounded 1-form on M and f be a smooth function with df = α. Let β be an (n − 1)-form with dβ = α. that is.

e In Riemannian normal coordinates. therefore α = d(bounded).4 Proposition. where r = |x| and ν is the radial normal field. Let M be complete and simply connected and α be a differential form on M of degree k with α ∞ < ∞ and dα = 0. see Example 8. The usual proof of the Poincar´ lemma provides a (k − 1)-form β with dβ = α. 2) If K ≤ −c2 < 0 and k ≥ 2. where p = exp(x). Since M is complete and simply connected with non-positive sectional curvature.1 above. Comparison with hyperbolic space shows that. o where 0 ≤ t ≤ 1. then α = d(bounded). sinh(cr) where r is the distance of the foot point of v to o. ∞. . for v perpendicular to ν. 1) If K ≤ 0. Note that this estimate is optimal.106 ¨ Lectures on Kahler Manifolds 8. Proof.3. the theorem of Hadamard–Cartan implies that expo : To M → M is a diffeomorphism. Hence 1 |β(p)| ≤ r = 0 0 r sinh(ctr) k−1 dt · α ∞ sinh(cr) sinh(cs) k−1 ds · α ∞ sinh(cr) ∞ 0 ≤ −∞ (e(k−1)cs ) ds · α = 1 α (k − 1)c ∞. then α = d(O(r)). K ≤ −c2 < 0. Hence we may define τt (p) := expo (t exp−1 (p)). Comparison with Euclidean space shows that the differential of τt satisfies |τt∗ v| ≤ t|v|. β is given by β(x) = r 0 1 tk−1 (ν α)(tx) dt = r 0 1 τt∗ (ν α)(x) dt. By the above estimate |τt∗ (ν α)| ≤ tk−1 |ν α| ≤ tk−1 α and hence |β(p)| ≤ r α k ∞. |τt∗ v| ≤ sinh(ctr) |v|. Suppose now that the sectional curvature of M is negative.

6. By Remark B. α is parallel and hence closed. In particular. G contains the connected component of the identity of the isometry group of M .¨ 8 Kahler Hyperbolic Spaces 107 8. by homogeneity. Let F ⊂ M be a maximal flat and p ∈ F . Sketch of proof of Proposition 8. Then. K) be a Riemannian symmetric pair such that M = G/K is a symmetric space of non-compact type. By Theorem B.5 Remark. t ∈ R. . There is a basis of Tp F such that the reflections of Tp F about the hyperplanes perpendicular to the vectors of the basis are realized by the differentials of isometries g ∈ G fixing p. 8. We can assume that (G. the Jacobi fields of geodesic variations by geodesics cp and perpendicular to F decay uniformly exponentially.6 requires more on symmetric spaces than we develop in Appendix B. Let M be a K¨hler manifold with K¨hler form a a ω. is the unit speed geodesic through p asymptotic to c. Since α is invariant under G. Since α vanishes along F .6 Proposition. cp is also regular and.3. Now γ is uniformly bounded and |V | = 1.7 Definition (Gromov). In particular.24. Therefore we only give a brief sketch of the argument. hence α = d(bounded). If α is a G-invariant differential form on M .36. The proof (and even the statement) of Proposition 8. If α is of positive degree. K) is an effective pair. for each p ∈ M . it follows easily that α|Tp F = 0. V )V. therefore. Note also that the lift of a differential form on a compact manifold to any covering manifold is uniformly bounded. The flats Fp constitute a smooth foliation F of M by totally geodesic Euclidean spaces. For any unit tangent vector u of M perpendicular to F . then α is parallel and hence closed. then the restriction of α to maximal flats in M vanishes and α = d(bounded). Good references for symmetric spaces of non-compact type are Chapter 2 in [Eb] and Chapter VI in [Hel]. Let V = − grad b and (ft ) be the flow of V . it follows easily that γ=− ∞ 0 Ft∗ α dt is a well defined smooth differential form on M such that LV γ = α and dγ = 0. Let c : R → M be a regular unit speed geodesic and b : M → R be the Busemann function centered at c(∞) with b(c(t)) = −t. cp (t) = ft (p). u is negative and. I learned it from Anna Wienhard. The estimates in the proof are explicit. With β = V γ we get α = LV γ = d(V γ) + V dγ = dβ. R(u. Let (G. This result seems to be well known to experts. bounded away from 0 by a negative constant which does not depend on u. 8. contained in a unique maximal flat Fp . We say that M is K¨hler-hyperbolic if the K¨hler form ω of the universal a a ˜ ˜ covering space M of M is d(bounded).

by Proposition 8. see Theorem 8. For the rest of this section we assume that M is a complete and connected K¨hler manifold with K¨hler form ω and complex dimension m.8 Lemma. is larger than 1. Declaring parallel sections as holomorphic turns E into a holomorphic vector bundle such that the Chern connection of the given Hermitian metric is the given flat connection. Since dα = 0 = d∗ (ω ∧ α) and ϕn η is smooth with compact support. By a Proposition 8. the sublevels of r are relatively compact and exhaust M . ϕn η ∧ α) = (ω ∧ α. Cao and Xavier [CX] and independently Jost and Zuo [JZ] observed that the arguments of Gromov concerned with the vanishing of square integrable harmonic forms work also under the weaker assumption that the K¨hler form a ∞ is d(O(µ)). with ϕn (p) = 1 if r(p) ≤ n. Hence closed K¨hler manifolds with finite fundamental a group are not K¨hler-hyperbolic.108 ¨ Lectures on Kahler Manifolds Recall that the cohomology class of the K¨hler form of a closed K¨hler mania a fold M is non-trivial. where 1 1/µ(n) = ∞. Without further notice we use notation and results from Appendix C.4. ϕn (p) = 0 if r(p) ≥ n + 1 and |dϕn | ≤ c0 . Let α be 1 a closed and square integrable differential form with values in E. 8. Proof. the dimension of their maximal flats. Choose an origin o ∈ M . K¨hler-hyperbolic K¨hler manifolds are the main topic in Gromov’s article a a [Gr]. On the other hand. We do not assume throughout that differential forms are smooth and indicate their regularity when appropriate. Let ω = dη with |η| = O(µ(r)).6. = An M . then M is K¨hler-hyperbolic. d(ϕn η ∧ α)) (ω ∧ α. if the sectional curvature a of M is strictly negative. Choose cut off functions ϕn : M → R. Then d∗ (ω ∧ α) = 0 =⇒ ω ∧ α = 0. Hermitian symmetric spaces of non-compact type are K¨hler a hyperbolic as well. Suppose that ω is d(O(µ(r))). These do not have strictly negative curvature if their rank. dϕn ∧ η ∧ α) + M (ω ∧ α. 8. 0= M (d∗ (ω ∧ α). by Proposition 8. One of the remarkable propera ties of K¨hler manifolds is that the Lefschetz map commutes with the Laplacian: a If α is a harmonic form with values in E. where ∞ 1/µ(n) = ∞. We let E → M a a be a flat Hermitian vector bundle.9 below. Note that this assumption is fulfilled if M is simply connected with non-positive sectional curvature.1 K¨hler Hyperbolicity and Spectrum. n ≥ 1. ϕn ω ∧ α). We include this extension into our discussion. This is one of the cornerstones in the arguments which follow. Since M is complete and connected.1.4. then ω ∧ α is harmonic as well. and let r : M → R be the distance to o. that is.

Then (∆d. ∆d = (d + d∗ )2 . E). Hence the second term on the right tends to the square of the L2 -norm of ω ∧ α as n → ∞.max and suppose that k = m or that α ⊥ H2 (M.49 and Lemma 8. dϕn ∧ η ∧ α) ≤ (cµ(cn + c) + c) · ω ∞ An |α|2 . there is a positive constant c0 such that |Ls γ| ≤ c0 |γ| for all γ ∈ Am−s (M. Hence the assertion follows if the limes inferior of the sequence on the right hand side is 0. Now An (ω ∧ α. If ω is d(O(µ)). For k > m. The L2 -norm of ω ∧ α is finite since ω is parallel and. This would be in contradiction to the assumption that α is square integrable. Suppose ω = dη with η ∞ < ∞.9 Vanishing Theorem (Cao–Xavier [CX]. uniformly bounded. α)2 ≥ λ2 α 2 with λ2 = c(m)/ η 2 > 0. Then dθ = ω s ∧ β − η ∧ ω s−1 ∧ dβ =: α − α . by Theorem C. Proof. therefore. E) → Am+s (M. Since E is flat. vanishing follows from Poincar´ duality. Hence we can assume that α and β are smooth with compact support. hence ≤ η ∞ |ω s−1 | β 2. Assume first that k = m + s > m.max = ∆d. 8. where we note that |ω s−1 | is constant. then k H2 (M. For k < m this follows from Theorem C. hence ∆d.8. 2 ∞ Proof. If the latter would not hold. The s-th power of the Lefschetz map defines a parallel field of isomorphisms Am−s (M. E)) such that α = ω s ∧ β. In particular.max α. E) and |L−s γ| ≤ c0 |γ| . Let α ∈ L2 (Ak (M. then there would be an ε > 0 such that α 2 2 = M |α|2 = n An |α|2 ≥ ε (cµ(cn + c) + c) · ω ∞ = ∞.19.10 Main Lemma. e 8. see the discussion in Section 5. Jost–Zuo [JZ]). there is β ∈ L2 (Am−s (M. E)) m be in the domain of ∆d.¨ 8 Kahler Hyperbolic Spaces 109 where An := {p ∈ M | n ≤ r(p) ≤ n + 1} contains the support of dϕn . Let θ := η ∧ ω s−1 ∧ β. E). θ 2 Now |θ| ≤ |η||ω s−1 ||β|. ∞ where 1 1/µ(n) = ∞. E) = 0 for k = dimC M .min. Since L±s is parallel.

θ)2 ≤ d∗ α ≤ ≤ Therefore α 2 2 2 2 ≤ c0 η ∞ |ω s−1 | ∆d α 1/2 2 α 1/2 2 . θ)2 + α ≤ 2c0 η ∞ |ω s−1 2 α 2 1/2 2 | ∆d α α 3/2 2 . we have Ls ∆d β = ∆d α. we have α ∈ dom(d + d∗ )max 2 2 iff α ∈ dom dmax ∩ dom d∗ . β)2 ≤ ∆d β 2 β 2 ≤ c2 ∆d α 2 α 2 . α)2 1/2 θ 2 = (α. We note that for a differential form α of pure degree. dθ)2 + (α. max + d∗ α max 2 2 Since (d + d∗ )max = (d + d∗ )min . E). By applying Poincar´ duality we conclude that it also holds e in the case k < m. this proves the asserted inequality in the case k > m. E)). θ 2 1/2 1/2 η ∞ |ω | ∆d α 2 α 2 β 2 1/2 3/2 c0 η ∞ |ω s−1 | ∆d α 2 α 2 . max By the above estimate we have dmax α 2 2 ≥ λ2 α 2 2 . hence dom dmax = dom dmax ∩ (ker d∗ + im d) max = dom dmax ∩ ker d∗ + im d. s−1 ≤ (∆d α. α )2 ≤ (d∗ α. Since Ls commutes with ∆d . max Since im d ⊂ ker dmax . it follows that the image of dmax on differential forms of degree m − 1 is equal to the image of its restriction to dom dmax ∩ ker d∗ ∩ L2 (Am−1 (M. dmax α ≥ λ2 α 2 2 for any differential k-form α ∈ dom(d + d∗ )max .110 ¨ Lectures on Kahler Manifolds for all γ ∈ Am+s (M. dθ + α )2 = (α. the above inequality shows that. for k = m. hence dβ 2 ≤ (∆d β. Now im d ⊂ dom dmax . Since ∆d.max is self-adjoint and non-negative. 2 0 We conclude that α We also have (d∗ α.

max α ≤λ −2 −2 2 α 2 + ∆d. − (Xj α) ⊗ Xj σ}. β)2 ≤ ∆d. β ∈ dom ∆d. we have S F (α ⊗ σ) := {(Xj ∧ α) ⊗ Xj σ σ). Let E .10). max We have γ ∈ dom ∆d.max . γ = dmax α 2 + d∗ β 2 2 max 2 = (∆d.min ⊂ dom(d + d∗ )min = dom(d + d∗ )max . max where dmax α ∈ im d ⊥ im d∗ d∗ α = dmax β = 0.max β 2 2 2 β 2 =λ · ( ∆d.max ∩L2 (Am (M. In particular. d∗ β ∈ dom dmax .12) where S is the principal symbol of D. Let F → M be a further Hermitian vector bundle with Hermitian connection 21 . max 2 2 d∗ β and max dmax α 2 ≥ λ2 α 2 .¨ 8 Kahler Hyperbolic Spaces 111 for any α in the latter space.max α. α)2 + (∆d. the image of dmax on differential forms of degree m − 1 is closed. max and hence α.max β. The twist of D by differential operator D : E(E ⊗ F ) → E(E ⊗ F ) defined by D (α ⊗ σ) = (Dα) ⊗ σ + S F (α ⊗ 8. where (Xj ) is a local orthonormal frame of T M . (8. d∗ β max 2 ≥ λ2 β 2 . we get dmax α ∈ dom d∗ . Considering degrees.11) is the (8. Therefore it contains im d ∩ L2 (Am (M. S F (α ⊗ dϕ ⊗ σ) = (S(dϕ)α) ⊗ σ.max α ·( d∗ γ 2 max 2 + ∆d. see (C.13 Example. . 21 We are short of symbols since D is in use already.max = dom ∆d. Define S F : E ⊗ T ∗ M ⊗ F → E ⊗ F. γ)2 .max β + dmax γ 2 ) = λ 2 2 2) −2 · (∆γ. E)). There is a similar discussion for d∗ on forms of degree m + 1. E → M be Hermitian vector bundles over M and D be a differential operator of first order from E(E ) to E(E ). Hence we max m can represent any γ ∈ dom ∆d. In conclusion. For D = d + d∗ . E)) perpendicular to H2 (M. E) as γ = dmax α + d∗ β.

see Theorem C. if is a Hermitian connection on a Hermitian line bundle F → M and the curvature form of is exact.10. Vice versa. Assume that |S A β| ≤ c · |β| for some √ positive constant c < λ/ 2. E) and λ be as in Lemma 8. β)2 = 1 λ2 (∆d β. 0 is flat. Then the group Γ of covering transformations of M → M acts on E. Let F = M × C be the trivial line bundle over M with 0 and = 0 + iζ ⊗ id as above.16) Aσ := iζ ⊗ σ.112 If . a potential perturbing D . Suppose ω = dη with η ∞ < ∞. Let E → M be the pull back of a Hermitian vector bundle ˜ ˜ E → M . We apply this in the case where F = M × C is the trivial line bundle over M with the canonical Hermitian metric and canonical flat and Hermitian connection 0 .2 Non-Vanishing of Cohomology. β)2 ≥ β 2 2 for all β ∈ A∗ (M. where M and M are connected ˜ ˜ and M is closed. Hence ker Dmax = 0 as asserted. Let ζ be a real valued differential one-form on M .10. by Lemma 8.19. We start with a little detour (and a corresponding change in notation) and discuss Atiyah’s L2 -index theorem [At]. Since iζ is purely imaginary. Assume that ker Dmax = 0. 8. Then ker Dmax = 0 =⇒ ker Dmax = 0. (8. E ⊗ F ). Hence. is Hermitian. Then = where is a connection form for F . The curvature form of is idζ. for all such β. ˜ hence on sections of E by γ · σ := γ ◦ σ ◦ γ −1 . 8. say equal to idζ. then (8.15) (8. c D β 2 = D ≥ D 0 0 β + SAβ β 2 2 A 2 − S β √ ≥ (λ/ 2 − c) β 2 . Since D 0 0 + A. then 0 = − iζ ⊗ id is a flat Hermitian connection (and F is trivial if M is simply connected).14) )(α ⊗ σ) = S F (α ⊗ Aσ) =: S A (α ⊗ σ) 0 is of order 0. 0 ¨ Lectures on Kahler Manifolds are Hermitian connections on F and A = (D − D 0 − 0 . Now Dmin = Dmax . Proof. . Let D = ∂ + ∂ ∗ on A∗ (M. ˜ ˜ Let M → M be a normal Riemannian covering.17 Lemma. we have 2 2 β 2 2 = (∆∂ β.

If the K¨hler form ω of M is a ˜ ∞ d(O(µ)). Gromov uses an extended version of (part of) Theorem 8.9. More precisely.21 Theorem (Atiyah [At]). ˜ the Γ-index of D is ˜ ˜ ˜ indΓ D = dimΓ ker D − dimΓ ker D∗ . (8. 8. ind D = indΓ D. 8. where 1 1/µ(n) = ∞. Proof. E) be a closed subspace and (Φn ) be an orthonormal basis of H. E) = ind D. (8. E). ˜ f (p) = |Φn (p)|2 . E) ≥ 0. By Hodge theory.20) are finite and ˜ indΓ D = ind D. By Theorem 8. where D = d+d∗ from Aeven (M.18) ˜ does not depend on the choice of (Φn ).20) ˜ The Γ-index of D is well-defined if one of the dimensions on the right is finite.¨ 8 Kahler Hyperbolic Spaces 113 ˜ ˜ Let H ⊂ L2 (M. Jost–Zuo [JZ]). ˜ ˜ Then the kernels of D and its formal adjoint D∗ are Γ-invariant. E). (8. then the Γ-dimensions on the right hand side of (8.21 applies in the case of non-positively curved closed K¨hler a manifolds and shows that their Euler characteristic has the right sign.21. Theorem 8. Then the function ˜ ˜ f : M → [0. It is important that dimΓ H = 0 iff H = 0. Hence ˜ (−1)m indΓ D = dimΓ Hm (M.21. E) to Aodd (M. we have the following result. but now from Aeven (M. E) ≥ 0. Let E ± → M be Hermitian vector bundles and D : E(E + ) → E(E − ) be a ˜ ˜ ˜ differential operator. By Theorem 8. Let M be a closed K¨hler a ˜ ˜ manifold and M → M be its universal cover. Let E ± and D be the pull backs of E ± and D to M . ∞]. By definition. there are no non-trivial square integrable harmonic forms on ˜ M of degree different from the middle dimension. We have D = d+d∗ . then (−1)m χ(M. E) ˜ ˜ ˜ ˜ ˜ to Aodd (M. He considers the twist of the pull back of a specific bundle E with the trivial complex line . ˜ is the lift of a function f on M and we set and then f dimΓ H = M f.22 Theorem (Cao–Xavier [CX]. It follows that f is Γ-invariant if H is. ˜ where Γ is the fundamental group of M . If D is elliptic. χ(M.19) the Γ-dimension of H.

114

¨ Lectures on Kahler Manifolds

bundle endowed with a Hermitian connection which is not invariant under Γ, whereas the curvature of still is. For the reader’s convenience, but without proof, we state a more general ˜ ˜ version of Gromov’s extension. Let E → M be a Hermitian vector bundle of rank r, not necessarily the pull back of a Hermitian vector bundle over M . Let G be an extension of Γ by a compact group K, 1 −→ K −→ G −→ Γ −→ 1.
ρ

(8.23)

˜ Suppose that G acts by Hermitian isomorphisms on E such that ˜ ˜ ˜ where π denotes the projection of E. Then G acts on L2 (M, E) by ˜ ˜ ˜ Let H be a G-invariant subspace of L2 (M, E). Then the function f as in (8.18) is Γ-invariant, and we define the G-dimension of H by the integral of the corresponding function f on M as in (8.19). ˜ Let E ± be Hermitian vector bundles with Hermitian G-actions satisfying ˜ ˜ (8.24) (where now π denotes the projections of E ± ). Suppose that D is a ˜ ˜ differential operator from E(E + ) to E(E − ) commuting with the actions of G, ˜ ˜ Then the kernels of D and D∗ are invariant under G, and we define the G-index ˜ as in (8.20). The case considered in Theorem 8.21 corresponds to the of D ˜ trivial extension G = Γ. The extension we state concerns the case where E = ˜ ˜ ˜ ˜ ˜ + ⊕ E − is a graded Dirac bundle and D+ : E(E + ) → E(E − ) the associated E Dirac operator in the sense of Gromov and Lawson, see Subsection C.24. ˜ 8.27 Theorem. Suppose that E is a graded Dirac bundle and that the Hermi˜ tian action of G on E satisfies (8.24) and leaves invariant connection, Clifford ˜ ˜ ˜ ˜ multiplication, and splitting E = E + ⊕ E − of E. Then the associated Dirac ˜ ˜ satisfies (8.26), the G-dimensions of ker D+ and ker D− are finite, ˜ operator D ˜ the canonical index form α of D+ is the pull back of a form α on M , and ˜ ˜ indG (D+ ) =
M

π ◦ g = ρ(g) ◦ π

for all g ∈ G,

(8.24)

g · σ := g ◦ σ ◦ ρ(g)−1 .

(8.25)

˜ ˜ D(g · σ) = g · (Dσ).

(8.26)

α.

About the proof. It seems that the arguments in Section 13 of [Ro] generalize to the setting of Theorem 8.27. 8.28 Remark. We will need Theorem 8.27 for the Hirzebruch–Riemann–Roch √ formula, that is, for the twisted Dolbeault operator ∂ + ∂ ∗ (times 2) on the Dirac bundle ˜ ˜ Ap,∗ (M, E) = A0,∗ (M, C) ⊗ Ap,0 (M, C) ⊗ E,

¨ 8 Kahler Hyperbolic Spaces

115

with splitting into forms with ∗ even and odd, respectively, see Example C.27.2. In this case, the canonical index form is ˜ ˜ Td(M ) ∧ ch(Ap,0 (M, C) ⊗ E) = Td(M ) ∧ ch(Ap,0 (M, C)) ∧ ch(E), where Td and ch denote Todd genus and Chern character, see Theorem III.13.15 in [LM]. Before returning to the discussion of K¨hler manifolds, we discuss a situation a where a group G as above arises in a natural way. To that end, suppose in ˜ addition that M is simply connected. Let F → M be a Hermitian line bundle with Hermitian connection such that the curvature form ω of is Γ-invariant. 8.29 Lemma. Let p ∈ M , γ ∈ Γ, and u : Fp → Fγp be a unitary map. Then there is a unique lift of γ to an isomorphism g of F over γ with gp = u and such that g preserves Hermitian metric and connection of F . Sketch of proof. Let q ∈ M and y ∈ Fq . Choose a piecewise smooth path c in M from p to q. Let x be the element in Fp which is parallel to y along c. Set g(y) = z, where z is parallel to ux along γ ◦ c. The dependence of parallel translation on curvature together with the Γ-invariance of ω shows that z only depends on the homotopy class of c. Since M is simply connected, g is well defined. By definition, u preserves parallel translation, hence . Since is Hermitian, g preserves the Hermitian metric of F . It follows from Lemma 8.29 that there is an extension 1 → U(1) → G → Γ → 1 (8.30)

of Γ by U(1) which lifts the action of Γ on M to an action of G on F as in (8.24). Furthermore, the action of G preserves metric and connection of F . We now return to the discussion of K¨hler-hyperbolic K¨hler manifolds. We a a let M be a closed and connected K¨hler manifold and E → M be a flat vector a ˜ ˜ bundle. We let M → M be the universal covering of M and E → M be the pull back of E. 8.31 Main Theorem. If ω = dη with η ˜
p,q ˜ ˜ H2 (M, E) = 0 ∞

< ∞, then

if p + q = m.

˜ Proof. Let F = M × C be the trivial line bundle with canonical Hermitian metric and flat connection 0 , and consider the Hermitian connections
t

=

0

+ itη.

The curvature form of t is itdη = it˜ , where ω denotes the K¨hler form of ω ˜ a ˜ M with respect to the induced K¨hler metric. Since ω is Γ-invariant, we get a ˜

116

¨ Lectures on Kahler Manifolds

extensions Gt of Γ by U(1) with induced actions on F as in Lemma 8.29 above. t For each fixed t, consider the twist D of the Dolbeault operator D = ∂ + ∂ ∗ p,∗ ˜ on A (M, E) by F with connection t . With respect to the splitting into t forms with ∗ even and odd, the index form of (D )+ is ˜ Td(M ) ∧ ch(Ap,0 (M, C)) ∧ ch(E) ∧ ch(F ), where ch F = exp(−tω/2π), see (A.44) and Remark 8.28 above. Hence the index form is a power series in t, where the coefficient of tm in degree dimR M is m rk E · ωm. ˜ m!(−2π)m p By Theorem 8.27, the index indGt ((D )+ ) is the integral of the index form t over M . It follows that indGt ((D )+ ) is a polynomial in t, where the coefficient m 22 of t is non-zero . Hence the zeros of this polynomial are isolated. t p,q ˜ ˜ On the other hand, H2 (M, E) = 0 would imply that indGt (D ) = 0 for all t sufficiently small, by Lemma 8.17. 8.32 Theorem. If ω = dη with η ˜ (−1)m χ(M ) =
p+q=m ∞
t

< ∞, then
p,q ˜ ˜ dimΓ H2 (M, E) > 0,

where Γ denotes the fundamental group of M . Proof. This is immediate from Theorems 8.21, 8.9, and 8.31, compare the proof of Theorem 8.22.

22 This is in contrast to the case of compact manifolds, where the index of elliptic differential operators is integral and would not depend on t.

in the tangent spaces of S. Since S is of real dimension two.4. Recall that a holomorphic line bundle E → M is positive or negative if its first Chern class c1 (E) has a positive or negative representative. then a cohomology class in H 2 (M. respectively. Z) in H 2 (S. hence a proper rescaling of g has an integral K¨hler class. Then the rotation by a positive right angle. In Examples 5. C) is the first Chern class of a holomorphic line bundle iff it is of type (1. The pull back of a positive holomorphic line bundle along a holomorphic immersion is positive. It follows that a closed complex manifold M admits a positive holomorphic line bundle iff M has a K¨hler metric g with integral K¨hler class a a [ω].60 we showed that the tautological bundle U → CP m is negative and that the hyperplane bundle H → CP m is positive. Since M is compact. If M admits a positive holomorphic line bundle. For ϕ ∈ E ∗ non-zero.59. Now the a = image of H 2 (S. Consider the evaluation map ε : E ∗ → O(M.g. C) is integral if it is in a the image of the canonical morphism H 2 (M. . E)∗ . our source is Section VI. if M admits a holomorphic immersion into CP n for some n ≥ m. then M admits a positive holomorphic line bundle. Clearly ε is holomorphic. Hodge conjectured and Kodaira proved that M admits a holomorphic embedding into CP n for some n ≥ m if M has a positive line bundle. see e. see Definition 5. Closed oriented surfaces are Hodge manifolds: Let S be such a surface and g be a Riemannian metric on S. Proposition III. E). It is immediate that a holomorphic line bundle is positive iff its dual is negative. We present Kodaira’s proof of his embedding theorem. O(M.9 Kodaira Embedding Theorem Let M be a closed complex manifold of complex dimension m. R) ∼ R. 1) and integral. If M is a K¨hler manifold. Hence S is a Hodge manifold. (9. In particular. then a positive representative of its first Chern class is a K¨hler metric on M .4 in [Wel]. We say that a cohomology class in H 2 (M.1 Remark. ε(ϕ)(σ) = ϕ(σ(p)). By the latter argument we a also get that a closed K¨hler manifold M with first Betti number b1 (M ) = 1 is a a Hodge manifold. Then M is called a Hodge manifold. R) is a lattice. a 9. In this sense the topic of a this section belongs to K¨hler geometry.6 in [Wel].2) where p is the foot point of ϕ. H 2 (S. Vice versa. Z) → H 2 (M. Kodaira’s proof uses the following general construction which associates to a holomorphic line bundle E → M a holomorphic map from an open part M of M to a complex projective space: Let O(M. C). is a parallel complex structure on S and turns S into a K¨hler manifold. E) is of finite dimension. E) be the space of holomorphic sections of E and M be the open set of points p ∈ M such that σ(p) = 0 for some σ ∈ O(M.

E)∗ ).p : O(M. ϕ ◦ σk ]. The proof that positive holomorphic line bundles are ample relies on the following lemma. in the above construction. It is clear that kB corresponds to k in the coordinates of O(M. Show that E ∗ is isomorphic to the pull back under k of the tautological bundle over P (O(M. If E is positive. E)∗ ). . E) and ϕ be a nowhere vanishing holomorphic section of E ∗ over an open subset U of M . (9. Let E be a holomorphic line bundle over M and p be a point in M .7) where (ϕ ◦ σ) denotes the derivative of ϕ ◦ σ. Hence kB is well defined on M . 9. Hence we obtain a holomorphic map k : M → P (O(M. (ϕ ◦ σ) (p)) ∈ C ⊕ (Tp M ⊗ C). (9. . . σ(q)). kB (p) = [ϕ ◦ σ0 . k(p) = [ε(ϕ)]. . . . Let ϕ : U → E ∗ be a nowhere vanishing holomorphic section of E ∗ .5 Exercise. q ∈ M .q (σ) = (σ(p). where U is an open subset of M containing p.6 Kodaira Embedding Theorem. 9. Then E is very ample. . We say that E is ample if some positive power of E is very ample.p (σ) := ((ϕ ◦ σ)(p).q : O(M. We say that a holomorphic line bundle E → M is very ample if. Let B = (σ0 . M = M and k (or kB ) is an embedding.3) where ϕ is any non-zero element of the fiber of E ∗ over p. Let M be a closed complex manifold and E → M be a holomorphic line bundle. Then the 1-jet of a smooth section σ of E at p with respect to ϕ is 1 ∗ Jϕ. Conclude that ample bundles are positive. the map 0 Jp. . then E is ample. E)∗ defined by the basis B. (9. Kodaira proves the following more precise version of the embedding theorem. Let E be a line bundle over M . E) → C ⊕ (Tp M ⊗ C) is surjective. is surjective and that 2) for any point p ∈ M and section ϕ about p as above.4) does not depend on the choice of ϕ. 0 Jp. 9. Suppose that 1) for any two different points p. Then the holomorphic map kB : U → CP k . σk ) be a basis of O(M.118 ¨ Lectures on Kahler Manifolds ε(ϕ) = 0 precisely for p ∈ M . There is a somewhat less abstract way of describing k.8 Lemma. . the map 1 ∗ Jϕ. E) → Ep ⊕ Eq .

. σm in O(M. Proof. . For points p = q in M . n ≥ n0 . e2 ) . k0 }. . . e1 ⊗ e2 ) = (e1 . . Fix a Hermitian metric on F and denote the curvature of its Chern connection by ΘF . (ϕ ◦ σm+1 ) (p). Since (β1 . Let p be a point in M and ϕ be a holomorphic section of E ∗ in an open neighborhood U of p as above. . and k ∈ {1.45 and Lemmas 3. . σp (q) = 0 and σq (q) = 0. . The Chern connection of the tensor product E1 ⊗ E2 of two Hermitian holomorphic vector bundles E1 . . . Now M is compact. . We denote the blow up of M in a point p by Mp and use an index p for objects associated to Mp . E).1 Proof of the Embedding Theorem. . βm ) is a basis of Tp M ⊗ C. q be different points of M . . dkB (p) has rank m. . We first observe that M = M . .46 and 3. 0) and Jϕ. σm are linearly independent in O(M. There is a natural identification of Mpq with Mqp such that πpq = πqp . By Assumption 1). We let Lpq be the holomorphic line bundle associated −1 −1 to the hypersurface πpq (p) ∪ πpq (q). . βm ) are linearly independent. . Let p. F be holomorphic line bundles over M and suppose E > 0. zk ) on {z0 = 0} ⊂ CP k . . . σ0 . For a basis B = (σ0 . By Assumption 2). With respect to affine coordinates (z1 . If we choose B with σ0 = σp and σ1 = σq . .47. E). we consider kB : M → CP k as above. βi ). . 9. .9 Kodaira Embedding Theorem 119 Proof. . Let E. where (β1 . E). then clearly kB (p) = kB (q). σk ) of O(M. Then σp and σq are linearly independent. . . we let Mpq = (Mp )q and πpq : Mpq → M be the canonical projection. . by Assumption 1). . see Lemma 5. Then given k0 ≥ 1 there is n0 ≥ 0 such that ∗ ∗ πp E n ⊗ πp F ⊗ (L∗ )k > 0 p and ∗ ∗ πpq E n ⊗ πpq F ⊗ (L∗ )k > 0 pq for all p = q ∈ M . there are holomorphic 1 1 sections σ0 . Choose a Hermitian metric on E such that the curvature of its Chern connection satisfies iΘE > 0. We use notation and results from Subsection 3. . hence kB is an embedding. σq (p) = 0. we get dkB (p) = ((β1 . . there are global holomorphic sections σp and σq of E with σp (p) = 0. E) such that Jϕ. in particular Examples 3. . E2 → M with induced Hermitian metric (e1 ⊗ e2 . 9. . e1 ) (e2 . It follows that kB is injective. Since Jϕ. .p (σi ) = ∗ 1 (0.61. (ϕ ◦ σk ) (p)). .p is a linear map. . . βm . Hence kB is an immersion. for any choice of basis B.9 Lemma.p (σ0 ) = (1. Therefore we can complete them to a basis B of O(M. .4.

together with derivatives of any order.120 ¨ Lectures on Kahler Manifolds Fix a cut off function χ on M with χ = 1 on U2 and χ = 0 on M \ U3 . is given by the usual product rule. independently of p ∈ U1 . Then Φ∗ is parallel with respect to the Chern 0 0 connection of hp . Jv) > 0 for v not in the kernel of πp∗ . We define a Hermitian metric hp on this part of L∗ p by associating length 1 to Φ∗ . p ∈ U1 . By the uniform boundedness of Θp . is uniformly bounded on U3 \ U2 . For p in the closure U1 of U1 . that is. Therefore we view hp = σp h as a Hermitian metric on that part of L∗ . Since σp (q) is smooth in p ∈ U1 and q ∈ U3 \ U2 . To get n0 indepenp dent of the chosen point p ∈ M . The holomorphic section Φ∗ dual to the holomorphic section Φ0 as in Example 3. Via the canonical projection πp . Θ = Θ1 + Θ2 . Clearly iΘ is positive outside U3 \ U2 . and hence the curvature of the latter vanishes identically. the family of Hermitian metrics hp . and hence iΘp ≥ 0 and iΘp > 0 on the tangent bundle of Sp .4 turns it 0 into a trivial bundle.45. where U contains the ball of radius 4 about 0 in Cm . It follows that Θp is uniformly bounded on U3 \ U2 . there is a Hermitian metric h on H such that the curvature of its Chern connection satisfies iΘh > −1 ∗ −1 0. we proceed as follows: Choose holomorphic coordinates z : U → U on M . Hence the Chern connection of the pull back metric ∗ ∗ on πp E n ⊗ πp F has curvature Θn equal to the pull back of the curvature on n E ⊗ F . Let p ∈ M and consider the projection πp : Mp → M . We now let hp = (χ ◦ πp )hp + (1 − χ ◦ πp )hp . there is an n0 > 0 such that i(n0 ΘE + ΘF ) > 0. The curvature Θp of the Chern connection of hp is the pull p back of Θh . we have L∗ = σp H. Since the hyperplane bundle H → CP m−1 is positive. Positive powers of L∗ will bring positivity everywhere. If we increase the above n0 if necessary. −1 Then the curvature Θp of the Chern connection of hp vanishes on Mp \ πp (U3 ) −1 and Θp = Θp on π (U2 ). The curvature of the Chern connection of the induced Hermitian metric on E n ⊗ F ⊗ (L∗ )k is Θ = Θn + kΘp . n0 can be . In particular. In this sense. Since M is compact and iΘE > 0. The Chern connection of the pull back of a Hermitian metric on the pull back of a holomorphic vector bundle is the pull back of the Chern connection of the original Hermitian metric on the original bundle. E n ⊗ F > 0 for all n ≥ n0 . Therefore iΘn (v. except for v = 0 or v tangent to Sp . we view L∗ restricted to p −1 Mp \ πp (U1 ) as a holomorphic line bundle over M \ U1 . the positivity of iΘn outweighs the possible negativity of kiΘp on U3 \ U2 . Over πp (U4 ). hence its curvature is the sum of the curvatures of the Chern connections on E1 and E2 . Set Ui = {p ∈ U | |z(p)| < i}. where σp : πp (U4 ) → CP m−1 is the p ∗ canonical projection. we use z − z(p) as holomorphic coordinates in the construction of the blow up Mp . we −1 identify Mp \ πp (U1 ) with M \ U1 .

(9. satisfies the assumptions of Lemma 9. let F be the canonical bundle of M . we obtain an induced short exact sequence 0 → O(F ) ⊗ 2 p → O(F ) → O(F ) ⊗ (O/ 2 p) → 0.8 is also defined on the stalk O(F )p and that it induces an isomorphism ∗ (9. We may view O(F ) ⊗ 2 as sheaf p of germs of holomorphic sections of F which vanish at least of second order at the distinguished point p. the sheaf of rings of germs of holomorphic functions on M . Since O(F ) is a locally free sheaf of modules over the structure sheaf O of M . we can apply the argument for the case of one point twice. For p ∈ M . Then E n . p 9.8 and hence E is ample. .13) where the tensor product is taken over O. let p and 2 be the subsheaves of p ideals in O of germs of holomorphic functions which vanish at p respectively vanish at p at least of second order. F = K. we denote by Fp the stalk of F at p. The argument in the case of two points p = q in a set U1 as above is similar. (9. f (p)). (9. that is. We will assume some elementary facts from sheaf theory. the map 1 Jp (f ) = (f (p). n ≥ n0 . 1 ∗ Jp : Op → C ⊕ (Tp M ⊗ C). hence can be covered by a finite number of sets U1 as in the argument.16 Lemma.9. Let E → M be a positive holomorphic line bundle. The stalks in q = p are (O(F ) ⊗ 2 p )q = O(F )q and (O(F ) ⊗ (O/ 2 p ))q = {0}. Then we have a short exact sequence 0→ For q = p. We use the same construction.9 Kodaira Embedding Theorem 121 chosen independently of p ∈ U1 . the stalk (O/ 2 p )q 2 p → O → O/ 2 p → 0. For p and q uniformly far apart.p p from Lemma 9.11) induces an isomorphism (O/ 2 p )p ∗ → C ⊕ (Tp M ⊗ C).12) Let F → M be a holomorphic line bundle.10) = 0. (9. we denote by F (U ) the space of sections of F over U . Denote by O(F ) the sheaf of germs of holomorphic sections of F . Moreover. Let O be the structure sheaf of M .15) (O(F ) ⊗ (O/ 2 ))p → C ⊕ (Tp M ⊗ C). and k0 = m+ 1.14) 1 To identify the stalk of O(F ) ⊗ (O/ 2 ) at p. we observe that the map Jϕ. Now M is compact. For p ∈ M . but now with two pull backs of H corresponding to the two points p and q. We now come to the final steps of the proof of the embedding theorem. For a sheaf F over M and an open subset U of M . In Lemma 9. (9.

By Hartogs’ theorem [GH. 0 − − → O(E n ) ⊗ −− 2 p ˜ ˜ ˜ − − → O(E n ) − − → O(E n ) ⊗ O/ −− −− π ∗ π ∗  µ  ρ 2 S −−→ 0 −− −−→ 0 −− (9.15). Denote by O the structure sheaf of M and by S ˜ and 2 the subsheaves of O of germs of holomorphic functions which vanish S along S respectively vanish at least of second order along S. O(E n )) → H 0 (M. if ∗ ∗ σ vanishes of order at least 2 along S.17). Now E n is trivial in a neighborhood of p in M . Hence it remains to show that p p O(M. ˜ ˜ Let M = Mp be the blow up of M in p. page 7]. It follows that π induces morphisms ∗ πλ : O(E n ) ⊗ 2 p ˜ → O(E n ) ⊗ 2 S and Hence we get a commutative diagram ˜ 0 − − → O(E n ) ⊗ −− π ∗  λ 2 S ∗ πρ : O(E n ) ⊗ (O/ 2 p) ˜ ˜ → O(E n ) ⊗ (O/ 2 S ). Composition with ∗ ˜ ˜ π induces injective morphisms π ∗ : O → O and πµ : O(E n ) → O(E n ). we will need to pass to blow ups of M to show the surjectivity in (9. . This is clear in the case m = 1 since then blowing up is trivial and π is biholomorphic. we have πµ σ = σ . By the continuity of σ and σ . hence isomorphic.17) is surjective.122 ¨ Lectures on Kahler Manifolds Proof. O(E n ) ⊗ (O/ 2 p )) (9. O(E n ) ⊗ (O/ 2 )) of global holomorphic sections of O(E n ) ⊗ p (O/ 2 ) onto the stalk (O(E n ) ⊗ (O/ 2 ))p . ˜ ˜ ˜ ˜ For m > 1. the restriction map induces an isomorphism from the space H 0 (M.18) − − → O(E n ) − − → O(E n ) ⊗ O/ −− −− 2 p of short exact sequences of sheaves.14). ∗ ∗ Clearly πλ and πµ are injective. by (9.8 is similar to and easier than the proof of Assumption 2). We claim that they are also surjective.4 it is then clear that a holomorphic section σ description of M of E n vanishes of at least second order at p iff σ ◦ π vanishes of at least second order about S. Hence πλ and πµ are ˜ ∗ surjective. By (9. Since points in M have codimension m and do not correspond to line bundles if m > 1. The five lemma implies that πρ is an isomorphism as well. Moreover. σp extends to a holomorphic section ∗ ˜ ˜ σ of E n over M . Let p ∈ M . The proof of Assumption 1) in Lemma 9. Let π : M → M be the canonical ∗ ˜ ˜ ˜ projection and E = π E. then also σ at p. E n ) = H 0 (M. a section σ of E n over M restricts to a section σp of E n over M \S = ˜ M \{p}. Therefore we concentrate on Assumption 2) and leave the proof of Assumption 1) to the reader. and the pull back of a local trivialization ˜ defines a trivialization of E n in a neighborhood of S. From the coordinate ˜ in Subsection 2.

Thus by (9. By Lemma 9. then also M . The Kodaira embedding theorem has many applications. O(E n ) ⊗ 2 ) = S 0. O(E n ⊗ (L∗ )2 )) = 0 by the Kodaira vanishing theorem 5. ˜ 9. see [GH. see for example [GH.6.6. page 192]. For a finite covering M → M as above. ˜ 9.20 Corollary. this follows in Wi .65. i ≥ 1.17). by Theorem 9. where πp : Mp → M is the natural map. therefore the ˜ ˜ ˜ pull back E → M of E is positive as well. hence ˜ ˜ H 1 (M.18) induce isomorphisms between the corresponding cohomology groups in the long exact sequences associated to the horizontal short exact sequences. O(E n ) ⊗ (O/ 2 S )) ˜ ˜ is surjective. ∗ πp E ⊗ (L∗ )k is positive for k sufficiently large. Let M be a closed complex manifold and Mp be the blow up of M in a point p ∈ M . In the notation of Example 3.9 Kodaira Embedding Theorem 123 A fortiori. Then the pull back E → M of the hyperplane bundle H over CP n is positive.19 Corollary. then Mp as well. this follows if H 1 (M.2 Two Applications. By the long exact sequence.9. Suppose f : M → CP n is a holomorphic embedding. We discuss two immediate ones. the vertical morphisms in (9.45. Then the pull back E → M of the hyperplane bundle H over CP n is positive. 9. If M admits a holomorphic embedding into a complex projective ˜ space. Hence M admits a holomorphic embedding into some complex projective space. If M admits a holomorphic embedding into a complex projective space. To show this we observe first that S = O(L∗ ). the functions zi serve as a frame over ˜ ˜ Wi and fi zi = f = fj zj over Wi ∩ Wj . Proof. it remains to show that ˜ ˜ ˜ ˜ ˜ H 0 (M. Hence Mp admits a holomorphic embedding into some complex projective space.46 and 3. Now E n ⊗ K ∗ ⊗ (L∗ )2 > 0 by Lemmas 3. averaging leads to ˜ admits a holomorphic embedding into a complex the converse assertion: If M projective space. To complete the picture. we quote the theorem of Chow which says that a complex analytic submanifold of a complex projective space is a smooth projective variety. Suppose f : M → CP n is a holomorphic embedding. . O(E n )) → H 0 (M. then M as well. 9. That is. Proof.4. page 167].47. Let M be a closed complex manifold and M → M be a finite covering of M . since there f ∈ S can be expressed as f = fi zi with fi holomorphic.21 Remark. by Theorem 9.

xd ∈ Rn . Vk . F g(ϕ ⊗ v) := ϕ ⊗ (gv). . dk . . . .6) and similarly for Λ∗ Rn ⊗ Vi . gvk ) = gΦ(v1 . . . (A. xd ) (A. . .5) 1 ε(σ)Φ(α1 (xσ(1) . αk are alternating forms on Rn with values in V1 . . .. F F F (A.1) be a k-linear map. . . If α1 . Let V1 . . αk ) = (j1 .. . Therefore F it gives rise to a k-linear map ΦΛ : (Λ∗ Rn ⊗ V1 ) × · · · × (Λ∗ Rn ⊗ Vk ) → Λ∗ Rn ⊗ V. . .1)) is equivariant (with F respect to these G-actions) if Φ(gv1 . The map (ϕ1 . . respectively. which we suppress in the notation). . = d1 ! . respectively. . . . vk ) (A. . (A. . We say that Φ (from (A. .4) and the definition of the wedge product. . v1 . . . Let Λ∗ Rn be the algebra of alternating forms on Rn with F values in F. . xσ(d) )). . and if we express αi = ϕij ⊗ vij with ϕij ∈ Λ∗ Rn and vij ∈ Vi . . . . we have ΦΛ (α1 . . ϕk . . αk (xσ(d−dk +1) . Vk and V be vector spaces over the field F ∈ {R. .4) The degree of ΦΛ (α1 . . . . .7) . . and V (via given representations. . vk ). . therefore. Then G acts also on Λ∗ Rn ⊗ V . (A. is linear in each of its 2k arguments. αk ) is d = d1 + · · · + dk . . . dk ! σ This is immediate from (A. . vkjk ). αk )(x1 . vk ) → (ϕ1 ∧ · · · ∧ ϕk ) ⊗ Φ(v1 . . It is also clear from this formula that we are discussing a generalization of the wedge product introduced in (1. .15). . . For x1 . .3) We view Λ∗ Rn ⊗ Vi and Λ∗ Rn ⊗ V as the space of alternating forms on Rn F F with values in Vi and V . no K¨hler classes are in the way and. .Appendix A Chern–Weil Theory In this appendix. Let G be a Lie group.. . then F ΦΛ (α1 . . respectively. xσ(d1 ) ). . . and suppose that G acts linearly on the vector spaces V1 . . connection a and curvature forms are denoted by the more standard ω and Ω. Vk and degrees d1 .jk ) (ϕ1j1 ∧ · · · ∧ ϕkjk ) ⊗ Φ(v1j1 . . . . .2) where ϕi ∈ Λ∗ Rn and vi ∈ Vi . . C} and Φ : V1 × · · · × Vk → V (A. . . . . . . . .. . . . . .

then the induced map ΦΛ is equivariant or invariant as well.9) Let αi ∈ A∗ (M.7) holds without the factor g on the right hand side. vk )]. . F) on its Lie algebra g = gl(m. . if (A. B] defines an equivariant bilinear map λ : g × g → g. We call (αi )τ = αi with respect to τ . and v ∈ V .Appendix A Chern–Weil Theory 125 for all g ∈ G and vi ∈ Vi . . Ek ) → A∗ (M. . we now specialize to the case where V1 = · · · = Vk = g with the adjoint action of G and V = F with the trivial G-action. gv] for all f ∈ P . . In the case where V = F with trivial action of G. v1 ].6). Φ(v1 . where the degree is additive in the degrees of the arguments. (A. That is. The trace tr : g → F is an invariant linear map since tr(gAg −1 ) = tr A for all g ∈ G and A ∈ g. (ϕ1j1 ∧ · · · ∧ϕkjk )⊗Φ(v1j1 . ΦΛ ((α1 )τ . . A∗ (M. then ΦΛ is equivariant as well. we consider an invariant k-linear map Φ : g × · · · × g → F.8 Examples. (αk )τ )] = (A. We say that Φ is invariant if it is equivariant with respect to the trivial action on V . With respect to a local section τ : U → P of P . This is immediate from (A. . Recall that E consists of equivalence classes of tuples [f. 1 ≤ i ≤ k. the determinant gives rise to an invariant m-linear map g × · · · × g → F. Then Φ induces a field of morphisms ([f. g ∈ G. The Lie bracket [A. F) = Fm×m is given by conjugation. vij ] = [τ. see below. (A. A. E). 1) Via the adjoint representation. . E) = A∗ (M. Adg A = gAg −1 . If Φ is equivariant. Let P → M be a principal bundle with structure group G. . Let E = P ×G V and Ei = P ×G Vi be the vector bundles associated to the representations of G on V and Vi . v] with relation [f g. 1 ≤ i ≤ k. . 1 ≤ i ≤ k. . vkjk )]. we can write (or identify) αi = ϕij ⊗ [τ. . and we obtain an induced field of morphisms on differential forms. vk ]) → [f. E1 ) × · · · × A∗ (M.11) [τ.4) and (A. [f. . . . (α1 . ϕij ⊗ vij ]. F).12) . The product AB of matrices defines an equivariant bilinear map µ : g × g → g. . If Φ is equivariant or invariant.10) where the ϕij are differential forms on U with values in F and the vij are ϕij ⊗ vij the principal part of smooth maps from U to Vi . . (A. That is. Although some of the following holds in greater generality. G acts on its Lie algebra g. v] = [f. . and similarly for Ei . E1 × · · · × Ek → E. αk ) → [τ. 2) The adjoint representation of the general linear group G = Gl(m. the space of differential forms on M with values in F. we view A∗ (M. . Ei ) be differential forms on M with values in Ei . Via polarization. Suppose that Φ is equivariant. .

. . . it is sufficient to consider decomposable forms. respectively. . β) = α ∧λ β (A.17 Corollary. we denote the principal parts of D and R by ω and Ω. . Bk ) = 0. B ∈ g.and two-forms on U with values in g.14) with values in F. λ(A. . . .16 Lemma. differential one. We interpret the curvature R of D as a differential twoform on M with values in P ×G g. ΦΛ (Ω. . B].16. . . . compare also Exercise 1. . β1 . . . . On decomposable forms α = ϕ ⊗ A and β = ψ ⊗ B with A. We return to the principal bundle P → M with structure group G and let D be a connection on P . Differentiation of this equation yields the claim. . With respect to a local section τ : U → P . . i (A.18) A. . we obtain a differential 2k-form Φ(P.15) Φ(B1 . A ∧λ Bi . with s(i) := (d1 + · · · + di−1 )d. . i (−1)s(i) ΦΛ (β1 . The sign arises from moving the differential form part of α to the leading position. d1 . . .4) and Lemma A. R) on M with values in F by setting Φ(P. the differential form Φ(P. . . . Φ(Ad(etA )B1 . Bk ∈ g. Since Φ is invariant. B ∈ g. For these. βk be alternating forms with values in g and of degrees d. (A. . . α ∧λ β = (ϕ ∧ ψ) ⊗ [A. Let α. Proof. B1 . . Ad(etA )Bk ) = Φ(B1 .15) on the right. Bk ). Proof. βk ) = 0. By the above. . . By multilinearity.19 Fundamental Lemma.3) on the left hand side and from (1. α ∧λ βi . . For A. For the curvature R of a connection D on P . . the claim is immediate from (A. B) := (Ad(etA )B)t=0 = [A.16. Ω)]. . . dk . Then. B].13) The induced map on alternating forms associates to alternating forms α and β with values in g an alternating form λΛ (α. A. (A. R) := [τ. F) does not depend on the choice of D.126 ¨ Lectures on Kahler Manifolds Recall that for A. . R) is closed and its cohomology class in H 2k (M. . . . . where the notation is from (A. A. . .

Appendix A Chern–Weil Theory

127

Proof. Fix a local section τ of P . In the notation of (A.14) and (1.15), the Bianchi identity says dΩ = Ω ∧λ ω, where λ is the Lie bracket of g. By Exercise 1.16, Ω ∧λ ω = −ω ∧λ Ω. Hence d(Φ(P, R)) = = [τ, ΦΛ (Ω, . . . , dΩ, . . . , Ω)] [τ, ΦΛ (Ω, . . . , Ω ∧λ ω, . . . , Ω)] = 0,

by (A.17). This shows the first assertion. Let D0 and D1 be connections on P . Then Dt = (1 − t)D0 + tD1 is also a connection on P . The principal part of Dt with respect to τ is ωt = (1 − t)ω0 + tω1 = ω0 + tβ, where the difference β = ω1 − ω0 is a one-form with values in g. We recall the structure equation 1 Ωt = dωt + ωt ∧λ ωt . 2 By Exercise 1.16, ωt ∧λ ωt = ωt ∧λ ωt , where the prime indicates differentiation with respect to t. Hence 1 1 Ωt = dωt + ωt ∧λ ωt + ωt ∧λ ωt 2 2 = dωt + ωt ∧λ ωt = dβ + β ∧λ ωt . With the Bianchi identity and Corollary A.17, d
i

ΦΛ (Ωt , . . . , β, . . . , Ωt ) =
j<i

ΦΛ (Ωt , . . . , dΩt , . . . , β, . . . , Ωt ) +
i

ΦΛ (Ωt , . . . , dβ, . . . , Ωt )

− =
j<i

ΦΛ (Ωt , . . . , β, . . . , dΩt , . . . , Ωt )
j>i

ΦΛ (Ωt , . . . , Ωt ∧λ ωt , . . . , β, . . . , Ωt ) ΦΛ (Ωt , . . . , β ∧λ ωt , . . . , Ωt ) ΦΛ (Ωt , . . . , β, . . . , Ωt ∧λ ωt , . . . , Ωt ) ΦΛ (Ωt , . . . , Ωt , . . . , Ωt )
i

− −

i

j>i

+

= (ΦΛ (Ωt , . . . , Ωt )) ,

128

¨ Lectures on Kahler Manifolds

the principal part of Φ(P, Rt ) . Hence Φ(P, R1 ) − Φ(P, R0 ) = dT, where T is the transgression form, T = [τ,
i 0 1

(A.20)

ΦΛ (Ωt , . . . , β, . . . , Ωt )dt].

(A.21)

Note that β is the principal part of B = D1 − D0 , a one-form on M with values in P ×G g. Hence T is well defined on M . Since the curvature form of a connection D has degree two, Φ(P, R) = 0 if there is a pair i < j such that Φ is alternating in the corresponding variables, Φ(A1 , . . . , Ai , . . . , Aj , . . . , Am ) = −Φ(A1 , . . . , Aj , . . . , Ai , . . . , Am ). In the applications of Theorem A.19, we are therefore only interested in the case where Φ is symmetric and invariant. Let Sk (g) be the space of symmetric k-linear forms on g with values in F. Let F[g] be the algebra of polynomials on g with values in F and Fk [g] be the subspace of homogeneous polynomials of degree k. The evaluation along the diagonal gives rise to an isomorphism ˜ Sk (g) → Fk [g], Φ → Φ, ˜ where Φ(A) := Φ(A, . . . , A). (A.22)

The inverse of this isomorphism goes under the name polarization. Since the isomorphism is equivariant with respect to the natural G-actions, G-invariant symmetric multilinear forms correspond exactly to G-invariant polynomials. In terms of a given basis (B1 , . . . , Bm ) of g, a homogeneous polynomial Φ of degree k on g can be written as Φ(A) =
µ

fµ aµ(1) . . . aµ(k) ,

(A.23)

where µ runs over all non-decreasing maps µ : {1, . . . , k} → {1, . . . , m}, the coefficients fµ are in F and the aj are the coordinates of A with respect to the chosen basis, A = aj Bj . Then the polarization of Φ is 1 ˜ Φ(A1 , . . . , Ak ) = k! fµ aσ(1)µ(1) . . . aσ(k)µ(k) ,
µ,σ

(A.24)

where now, in addition, σ runs over all permutations of {1, . . . , k} and the aij are the coordinates of the vector Ai . By multilinearity and (A.23), ˜ Φ(A, . . . , A) =
j1 ,...,jk

˜ Φ(Bj1 , . . . , Bjk )aj1 . . . ajk =
µ

fµ aµ(1) . . . aµ(k) . (A.25)

Appendix A Chern–Weil Theory

129

We let Λeven Rn := ⊕Λ2i Rn and consider F F α= ϕj ⊗ Bj ∈ ΛevenRn ⊗ g. F (A.26)

Since ΛevenRn is a commutative algebra, (A.25) implies F ˜ ΦΛ (α, . . . , α) =
j1 ,...,jk

˜ Φ(Bj1 , . . . , Bjk )ϕj1 ∧ · · · ∧ ϕjk Λeven Rn . F

(A.27)

=

fµ ϕµ(1) ∧ · · · ∧ ϕµ(k) =: ΦΛ (α) ∈

The definition of ΦΛ (α) is very natural, and we could have started with it. On the other hand, the detour over polarization shows that we are in the framework of what we discussed before. Equation A.27 implies that for homogeneous polynomials Φ and Ψ, (Φ · Ψ)Λ (α) = ΦΛ (α) ∧ ΨΛ (α). (A.28)

In particular, linear extension to the algebra F[g] of polynomials gives rise to an algebra morphism F[g] → ΛevenRn , Φ → ΦΛ (α). F (A.29)

If α is a sum of forms of strictly positive (and even) degrees, then the natural extension of this map from F[g] to the larger algebra F[[g]] of formal power series in g over F is well defined. For this we note that ΦΛ (α) has degree kd if Φ is homogeneous of degree k and α is of pure degree d and that alternating forms on Rn of degree > n vanish. In both cases, F[g] or F[[g]], if Φ is invariant under the natural action of G, then also ΦΛ (α). We return again to the principal bundle P with structure group G. Denote by F[g]G ⊂ F[g] and F[[g]]G ⊂ F[[g]] the subalgebra of G-invariant polynomials and G-invariant formal power series, respectively. Let α ∈ Aeven (M, P ×G g). By what we said above, α gives rise to a morphism of commutative algebras, the Weil homomorphism F[g]G → Aeven (M, F), Φ → Φ(P, α), (A.30)

where Φ(P, α) = [τ, ΦΛ (ατ )] with respect to a local section τ of P , compare (A.11). Moreover, if α is a sum of forms of strictly positive (and even) degrees, then the natural extension to F[[g]]G is well defined. A.31 Main Theorem. Let P → M be a principal bundle with structure group G. Let D be a connection on P and R be the curvature of D. Then the Weil homomorphism F[[g]]G → Aeven (M, F), Φ → Φ(P, R),

is a morphism of commutative algebras. Moreover,

For a complex manifold M with complex structure J. C) and c(E) = 1 + c1 (E) + · · · + cr (E) ∈ H ∗ (M. Then the k-th Chern form of E with respect to D is defined as ck (E. D) := Φk P. Let D be a connection on E. 2. D).35) „ « (A. and hence c(E) = 1. We note that Φk (A) is real if A is real. and Φr = det. By definition.33) where R is the curvature of D. Assertion (2) is immediate from the definitions. Φ1 = tr. respectively. R ) = f ∗ Φ(P. F) does not depend on the choice of D.34) are called the Chern classes and the total Chern class of E. we consider T M together with J as a complex vector bundle and use the shorthand ck (M ) and c(M ) instead of ck (T M ) and c(T M ). Therefore we view the Chern classes as a measure of the deviation of E from being trivial. C) = Cr×r . . We get invariant homogeneous polynomial Φk of degree k on gl(r. Chern classes of isomorphic bundles are equal. on P . The corresponding cohomology classes ck (E) = [ck (E. D)] ∈ H 2k (M. Proof.30). or. Let G = Gl(r. D) = det I + 2π where r is the rank of E. D) + · · · + cr (E. R). Then the structure group of P is G = Gl(r. The total Chern form of E with respect to D is i R = 1 + c1 (E.19. then Φ(P . then ck (E. if P → M is a G-principal bundle with connection D and F : P → P is a morphism over a smooth map f : M → M with D = F ∗ D. Assertion (1) follows from Lemma A. respectively. what amounts to the same. Φ(P. (A. C) (A. „ i R . The first part is clear from the discussion leading to (A. Let E → M be a complex vector bundle of rank r and P be the bundle of frames of E. If E is trivial and D is the trivial connection on E. D) = 0 for k > 0. then Φk (A) is the k-th elementary symmetric function of the eigenvalues of A. C) and g = gl(r.130 ¨ Lectures on Kahler Manifolds 1. C) and P ×G g = End E. In particular. c(E. Φ0 = 1. R) is closed and its cohomology class in H even (M. A.32) If A is diagonalizable. 2π « (A. C) by setting r det(tI + A) = k=0 tr−k Φk (A).1 Chern Classes and Character.

Appendix A Chern–Weil Theory

131

A.36 Proposition. If D is compatible with a Hermitian metric on E, then the Chern forms ck (E, D) are real. In particular, ck (E) ∈ H 2k (M, R) ⊂ H 2k (M, C). Proof. With respect to a local orthonormal frame of E, that is, a local section of P , the principal part Ω of the curvature of D is skew-Hermitian, Ω = −Ωt . Hence c(E, D) = detΛ I +

i i Ω = detΛ I − Ω = c(E, D), 2π 2π

«

«

where we use that transposition does not change the determinant. A.37 Remark. There is a topological construction of Chern classes in H ∗ (M, Z), see [MS]. The Chern classes we discuss here are their images under the natural map H ∗ (M, Z) → H ∗ (M, R). In particular, the Chern classes here do not contain as much information as the ones coming from the topological construction. On the other hand, the construction here gives explicit differential forms representing the Chern classes, an advantage in some of the applications in differential geometry. A.38 Proposition. The Chern forms satisfy the following properties: 1. (Naturality ) c(f ∗ E, f ∗ D) = f ∗ (c(E, D)). 2. (Additivity ) c(E ⊕ E , D ⊕ D ) = c(E , D ) ∧ c(E , D ).

3. cj (E ∗ , D∗ ) = (−1)j cj (E, D), where E ∗ = Hom(E, C) is the dual bundle with induced connection D∗ . The corresponding properties of Chern classes follow as a corollary.

Proof of Proposition A.38. The first assertion is immediate from f ∗ ΩD = Ωf ∗ D . With respect to local frames of E and E over an open subset U ⊂ M , the principal part of the curvature of D = D ⊕ D is Ω= Over U we have c(E ⊕ E , D ⊕ D ) = detΛ 0 i 0 I + 2π Ω „ « „ « i i = detΛ I + Ω ∧ detΛ I + Ω 2π 2π I+
i 2π Ω

Ω 0

0 Ω

.

= c(E , D ) ∧ c(E , D ), compare (A.27). This proves the second assertion. With respect to a local frame, −Ωt is the principal part of the curvature of D∗ . Hence the last assertion.

132

¨ Lectures on Kahler Manifolds

A.39 Examples. 1) If the bundle E → M is trivial and D is the trivial connection on E, then c(E, D) = 1. In particular, c(E) = 1. More generally, if E has a flat connection, then c(E) = 1. 2) If E has rank r and splits as E = E ⊕ E with E trivial of rank s, then cj (E) = 0 for j > r − s. 3) Let S be the oriented surface of genus g, endowed with a Riemannian metric and the corresponding Levi-Civita connection D. The rotation J = Jp by a positive right angle in Tp S, p ∈ S, is a parallel field of complex structures which turns S into a K¨hler manifold of complex dimension 1 and T S into a a complex line bundle over S. Let p ∈ S and (v, w) be an oriented orthonormal basis of Tp S. Then w = Jv and K(p) = − R(v, w)v, w , and hence Ωp (v, w) = −i K(p), where K is the Gauss curvature of S. Therefore c1 (T S, D) = K dA, 2π

where dA denotes the (oriented) area form. The Gauss–Bonnet formula gives c1 (T S, D) = χ(S) = Euler number of S.
S

We make this more explicit in the case of S = CP 1 = S 2 . On the open subset U = {[z0 , z1 ] ∈ S | z1 = 0} of S, consider the coordinates x + iy = z := z0 /z1 . Let X = ∂/∂x and Y = ∂/∂y be the corresponding coordinate vector fields. The complex structure on S turns T S into a complex line bundle. The standard metric of constant curvature 1 is the real part of the Hermitian metric on T S which, over U , is given by h(X, X) = 4 . (1 + |z|2 )2 −2zdz . 1 + |z|2 (A.40)

By Proposition 3.21.1, the connection form of the Chern connection D is ω = h−1 ∂h = (1 + |z|2 )2 ∂ (1 + |z|2 )−2 =

By Proposition 3.21.4, the principal part of the curvature is Ω = ∂ω = Hence the first Chern form c1 (T S, D) = i dz ∧ dz 2dx ∧ dy = . π(1 + |z|2 )2 π(1 + |z|2 )2 2dz ∧ dz . (1 + |z|2 )2

Appendix A Chern–Weil Theory

133

It follows that c1 (T S, D) =
S U

c1 (T S, D) 2 π
∞ 0 ∞ 0

=

=4

ρ dρdϕ (1 + ρ2 )2 0 ∞ ρ dρ du =2 = 2, 2 )2 (1 + ρ u2 1

where we substitute x + iy = ρ exp(iϕ) and u = 1 + ρ2 , respectively. As expected, the right hand side is the Euler number of CP 1 = S 2 . 4) In Example 3.24, we consider the canonical metric of the tautological bundle U over the complex Grassmannian Gk,n and compute the principal part of the curvature of the Chern connection D. In the case U → G1,2 = CP 1 and with respect to the coordinates z as in the previous example, we obtain Ω(0) = −dz ∧ dz. It follows that CP 1 c1 (U, D) = −1 and that c1 (CP 1 ) = −2c1 (U ). 5) Let π : S 2m+1 → CP m be the natural projection, x ∈ S 2m+1 , and p = π(x). Then the differential of π at x identifies the orthogonal complement p⊥ with the tangent space of CP m at p. However, this isomorphism depends on the choice of p.The ambiguity is resolved by passing from p⊥ to Hom(p, p⊥ ). In other words, T CP m ∼ Hom(U, U ⊥ ), where U ⊥ is the orthogonal complement of = the tautological bundle U → CP m in the trivial bundle Cm+1 × CP m → CP m with respect to the standard Hermitian metric. By the naturality of Chern classes, the first Chern class c1 (U ) has value −1 on the complex line CP 1 ⊂ CP m . Now Hom(U, U ) ∼ C × CP m , the trivial line = bundle over CP m . Hence T CP m ⊕ C = Hom(U, U ⊥ ⊕ U ) = Hom(U, Cm+1 ) ∼ U ∗ ⊕ · · · ⊕ U ∗ , = where we have m + 1 summands on the right hand side. Therefore c(CP m ) = c(T CP m ⊕ C) = (1 − c1 (U ))m+1 = (1 + a)m+1 , where a is the generator of H 2 (CP m , C) which has value 1 on the complex line CP 1 ⊂ CP m . In particular, c1 (CP m ) = (m + 1)a. It follows that cm (CP m )[CP m ] = m + 1, the Euler number of CP m . A.42 Remark. If E is a complex vector bundle of rank r over a manifold M , then cr (E) is the Euler class of E, considered as an oriented real vector bundle of rank 2r, see Proposition A.59 below. (A.41)

. 1 r where σ1 . the symmetric function tr D(a1 . 2 2 (A. . . . D ) ∧ ch(E . σr . D) + „ 1 c1 (E. C) (A. ar )2 − 2σ1 (a1 .43) Let E → M be a complex vector bundle of rank r and P be the bundle of frames of E. B)(x. D ). If A is diagonal. C) on Cr ⊗ Cs is given by (A. D ). C) × gl(s. ar ). y) = (Ax) ⊗ y + x ⊗ (By) and show that ch(E ⊗ E . Its cohomology class Φ(E) in H even (M. we have ch(E. . . D) = r + c1 (E.47 Exercise. C) = Cr×r . 2 « (A. . ar )k can be expressed uniquely as a polynomial in σ1 . . . and 1 1 tr A2 = Φ1 (A)2 − Φ2 (A). . . . Follow the argument in Proposition A.38. 2 (A. D) := Φ P. . . . ar ). D ) + ch(E . Let D be a connection on E and R be the curvature of D. .2 and show that ch(E ⊕ E . . From what we noted above. A = D(a1 . σ2 . . . It follows that the Chern character induces a homomorphism from the Grothendieck ring K(M ) generated by equivalence classes of complex vector bundles over M to the even cohomology ring H even (M. Since diagonal matrices are dense in Cr×r . D ⊕ D ) = ch(E . D) + · · · . . A. . ar ). C) is called the Chern character of E. . D) can be expressed as a polynomial in the Chern forms ck (E.45) where Φ1 and Φ2 are as in (A. Then the cocycle „ « i R ∈ Aeven (M. . denotes the sequence of elementary symmetric functions. D). . D ⊗ D ) = ch(E . C).44) ch(E.134 ¨ Lectures on Kahler Manifolds On gl(r. Recall that the derived action of the Lie algebra gl(r. we conclude that tr A = Φ1 (A). consider the invariant formal power series Φ(A) := tr exp A = r + tr A + 1 tr A2 + · · · . .46) We leave it as an exercise to compute the term of degree 6 in this formula. tr A2 = a2 + · · · a2 = σ1 (a1 . D)2 − c2 (E. . 2π is called the Chern character of E with respect to D. then tr A = a1 + · · · + ar = σ1 (a1 . By the fundamental theorem on symmetric polynomials.32). It follows that ch(E. . .

D). A. A= . the Lie algebra of the group of rotations SO(2r). aj 0 Ar For such a matrix A.48) Pf(A) := r 2 r! σ of degree r. We conclude that the cohomology class χ(E) of χ(E. Let C = B. By invariance. g . Then the principal bundle P of oriented orthonormal frames of E has structure group SO(2r). D) := Pf P. f ∗ g. However. aσ(2r−1)σ(2r) (A. (A. see for example [MS. D) = 0 and hence also χ(E) = 0. ·) where √ is B a field of symmetric and positive definite endomorphisms of E. g.49) Therefore the Pfaffian defines an invariant polynomial on so(2r). If D is a connection on P . in the definition of the Euler form we need a Riemannian metric and a corresponding metric connection on E. g ⊕ g . that is. then g (x. . g . the Pfaffian of A.55 Remark. D) and that χ(E ⊕ E . g. (A. . R) = R2r×2r . (A. C ∗ D) = χ(E. Cy). consider the homogeneous polynomial 1 ε(σ)aσ(1)σ(2) . g. we have Pf(A) = (−1)r a1 . . The normal form on so(2r) is   A1 0 −aj   . then χ(E. D) in H 2r (M.. and R denotes the curvature of D. 312 pp]. g. It is clear that χ(f ∗ E. then the associated form χ(E. Pf(A)2 = det A. we conclude that on so(2r). D ⊕ D ) = χ(E. g. .52) is called the Euler form of E with respect to g and D. It is called the Euler class of E.51) Let E → M be an oriented real vector bundle of rank 2r with a Riemannian metric g.53) Let g be another Riemannian metric on E and write g = g(B ·. R) does not depend on g and D. g. f ∗ D) = f ∗ χ(E. On gl(2r.50)  with Aj = . there are vector bundles E with flat non-metric connections such that χ(E) = 0.Appendix A Chern–Weil Theory 135 A. (A. In fact. It follows that χ(E. D) ∧ χ(E . and hence C is an isomorphism of E with C ∗ g = g . . D ). a metric connection on E. y) = g(Cx.2 Euler Class. If E has a Riemannian metric g with a flat metric connection D. ar . „ 1 R 2π « (A. It is easy to show that Pf(BAB t ) = Pf(A) det B.54) (A.

z r ) ←→ (x1 . where z j = xj + iy j . Let ω and Ω be the connection and curvature forms of D with respect to a local unitary frame of E. A= (A.33) and (A. This is the celebrated Chern–Gauß–Bonnet formula. diagonal matrices     A1 ia1     . .54 imply that the Euler class of (the tangent bundle of) an oriented closed manifold M ..136 ¨ Lectures on Kahler Manifolds Our next aim is to compare Euler class and Chern classes. iar Ar where Aj is as above. . . Suppose now that E → M is a complex vector bundle of rank r. . By invariance. By (A. ar = Pf(AR ). xr .57). A.59 Proposition. . We identify Cr with R2r via the correspondence (z 1 . χ(E) = cr (E).58) A. (A. .57) det(iA) = Pf(AR ) for all A ∈ u(r). cr (E. . y r ). This induces an inclusion U(r) → SO(2r) of the unitary group and a corresponding inclusion u(r) → so(2r). . det(iA) = (−1)r a1 . g. y 1 . D).53. the real part g = Re h of the Hermitian metric defines a Riemannian metric on E and D is metric with respect to g. . For a complex vector bundle E → M of rank r. Under this inclusion. A → AR . . gives the Euler number of M .56) →  = AR . of Lie algebras.. 2π « (A. For such a diagonal A. Endow E with a Hermitian metric h and a corresponding Hermitian connection D.59 and Equations A. . If we consider E as an oriented real vector bundle of rank 2r. . applied to the fundamental class of M . Proposition A. If we consider E as an oriented real vector bundle and choose D as a metric connection for the Riemannian metric g as above. . then ωR and ΩR are the connection and curvature forms of D with respect to the induced oriented and orthonormal frame of E. . D) = det „ i Ω = Pf 2π « „ 1 ΩR = χ(E.

. . . p ∈ M . Since sp (p) = p. l).3 Exercises. . ϕ1 . s∗ ϕl ) p p = (−1)k+l Fp (v1 . vk ∈ Tp M . there is an isometry sp of M with sp (p) = p and dsp (p) = − id . . Good references for unproved statements and further reading are [Bor].1) We then call sp the geodesic reflection at p.4 Proposition. . Let M be a locally symmetric space and F be a tensor field on M of type (k. if M is locally symmetric. [Hel]. ∗ Proof of Proposition B. then the curvature tensor is parallel. . . . p ∈ M . . . B. If F is invariant under the local isometries sp . and (s∗ F )p = Fp . Show that M 2) Let M = M1 × M2 be a Riemannian product. ˜ with the induced Riemannian metric is also a symmetric space. . we conclude p Fp (v1 . .4: If M is locally symmetric and J is an almost complex structure on M which is invariant under the local isometries sp . . . B. . 3) Prove the corresponding assertions in the case of locally symmetric spaces. . −ϕ1 . for all p ∈ M . . dsp (p) = − id. . . Another application of Proposition B. f2 be isometries of M . . 1) Let M be a symmetric space and M → M be a covering. . . . The fundamental theorem about locally symmetric spaces is as follows: = Fp (−v1 . then F = 0 if k + l is odd. For example. R = 0. vk . Show that M is symmetric if and only if M1 and M2 are symmetric. . We say that a Riemannian manifold M is locally symmetric if each point p ∈ M has a neighborhood U in M with an isometry sp : U → U such that sp (p) = p and dsp (p) = − id. [Wo]. ϕ1 . Then f1 = f2 . . .Appendix B Symmetric Spaces In this appendix we collect a few of the basic facts about symmetric spaces. . Let p ∈ M . .2 Exercise. s∗ ϕ1 . ˜ B.4. ϕl ∈ Tp M . . then J is parallel. . . In fact. ϕl ). . . (B. −vk . Suppose there is a point p ∈ M with f1 (p) = f2 (p) and df1 (p) = df2 (p). it is not hard to see that M is locally symmetric if and only if the curvature tensor of M is parallel. . vk . ϕl ) = Fp (sp∗ v1 . −ϕl ) . v1 . Let M be a connected Riemannian manifold and f1 . . We say that a Riemannian manifold M is a symmetric space if M is connected and if. . . sp∗ vk . and ϕ1 . . .

By the first part of the proof. B. b] → M be a piecewise smooth curve. by the inner product and the curvature tensor at a point (using Propositions B. any two points p. and let m be the midpoint of c. the closed ball B(q. If c : [a. q ∈ M can be connected by a piecewise geodesic c. Let c : [a.1 in [Sa]23 . b] such that c|[ti−1 . where M is simply connected and N is complete. This implies. Let st be the geodesic reflection about c(t) and ft = st/2 ◦ s0 . then there is a subdivision a = t0 < t1 < · · · < tk = b of [a. b] → M be a geodesic in M . ε) of radius ε about any point q ∈ M is compact.7 and B. b] → M is such a curve.5. the differentials of the isometries of M fixing a point p in M are given precisely by the orthogonal M transformations of Tp M preserving the curvature tensor Rp .6 Corollary. Since M is connected. 23 Both references contain also introductions to symmetric spaces. As for a proof. Then ft is an isometry of M with ft (c(τ )) = c(τ + t) for all τ ∈ R. Then the geodesic reflection sm of M at m reflects c about m. Let M be a symmetric space. Hence fk ◦ · · · ◦ f1 is an isometry of M mapping p = c(a) to q = c(b). If M is symmetric. up to isometry. Then there is an ε > 0 such that the closed ball B(p. It follows that M is complete. B.5. Let M and N be locally symmetric Riemannian spaces. 1 ≤ i ≤ k. Then there is a unique local isometry F : M → N with F (p) = q and F (p) = A.5 Theorem (Cartan).8 Proposition.8 below). .A∗ gq = gp and A∗ Rq = Rp . and let A : Tp M → Tq N be a linear isomorphism preserving inner products and N M N M curvature tensors. Let c : [a. hence sm (c(a)) = c(b). Proof. Let c : R → M be a geodesic through p. ε) of radius ε about p is compact. then M is homogeneous. An isometry f : M → M is called transvection along c if f (c(a)) = c(b) and if df (c(a)) is equal to parallel translation along c from c(a) to c(b). see for example the more general Theorems 1. B. the uniqueness of the model Riemannian manifolds of constant sectional curvature. ti ] is a geodesic. Another immediate application asserts that a simply connected symmetric space is determined. then M is complete.1.138 ¨ Lectures on Kahler Manifolds B. This is one of the most important immediate applications of Theorem B. there is an isometry fi of M mapping c(ti−1 ) to c(ti ).36 in [CEb] or III. Since M is homogeneous. Let p ∈ M . Let M be a complete and simply connected locally symmetric space. Then M is a symmetric space. Proof.7 Proposition. Let p be a point in M . for example. If M is homogeneous. More precisely. Let p ∈ M and q ∈ N .

Then c is a closed geodesic. more precisely. Similarly. Proof. where the prime indicates covariant derivative along c. it is the transvection along c|[τ. see Exercise 1. = = = τ ∂t (ft (σ(τ )))|t=τ =0 t ∂τ (ft (σ(τ )))|τ =t=0 t (dft v)|t=0 =0 since dft is parallel translation along c. . Let v ∈ Tp M and X be the infinitesimal transvection along the geodesic through p with initial velocity v. by Lemma B. Then X(p) = c (0) since ft (p) = c(t). B. We apply this to the Jacobi fields X ◦ c and tc . p be a point in M . Let c : R → M be a geodesic with c(0) = p. Then vX We conclude that. Associated to this family we have the Killing field X(p) = ∂t (ft (p))|t=0 . Then ds0 ◦ X is parallel along s0 ◦ c.9. B. Hence dft (c(τ )) is parallel translation along c from c(τ ) to c(τ + t). (X ◦ c) . the function V . Proof. dft X(τ ) = X(τ + t). Then a Killing field X on M is the infinitesimal transvection along c if and only if X(p) = c (0) and X(p) = 0.9 Lemma. dst/2 X(s) = −X(t − s).Appendix B Symmetric Spaces 139 Let X be a parallel vector field along c. W − V. As for the covariant derivative of X in p. c = const. for all τ ∈ R. We call it the one-parameter group of transvections along c. c (a) = c (b). It follows that the family (ft ) is a smooth one-parameter group of isometries of M which shifts c and whose derivatives correspond to parallel translation along c. Since M is connected. let v ∈ Tp M and σ = σ(τ ) be a smooth curve with σ(0) = p and σ (0) = v. W along c. Hence ft is a transvection along c. X ◦ c is a Jacobi field along c. Since X is a Killing field. we have ds0 X(s) = −X(−s).5. Recall that for any two Jacobi fields V. Since s0 (c(s)) = c(−s) and ds0 (p) = − id. We call X the infinitesimal transvection along c. W is constant. Let c : [a.10 Proposition. Let M be a symmetric space. a Killing field on M is determined by its value and covariant derivative at one point. b] → M be a geodesic loop at p. Let X be the infinitesimal transvection along c. tc − X ◦ c. Then X(p) = v and X(p) = 0. τ + t] for any τ ∈ R.

The fundamental group of a symmetric space is Abelian. Let M be a symmetric space and p be a point in M . B. then M does not have non-trivial closed geodesics. Since M is of non-compact type. 2) If G denotes the component of the identity of the group of isometries of M and K ⊂ G the stabilizer of a point p ∈ M . As for the second. Then Y = X ◦ c is a Jacobi field along c with the same initial conditions as J: Y (0) = v and Y (0) = 0. Since R is parallel. Ric(u. We say that M is of compact type or. along the geodesic determined by v. If M is a symmetric space of compact type. Hence the symmetric endomorphism R(·. Contradiction. R(V. On the other hand. We say that M is of Euclidean type if M is flat. respectively.13 Proposition. Now a group is Abelian if its inversion is a homomorphism. Let X be the infinitesimal transvection through p with X(p) = v. Now as the restriction of a vector field on M to c. Let c : R → M be a closed geodesic of unit speed. Proof. we note that the last piece of the long exact homotopy sequence of the fibration G → M . Since shortest loops in homotopy classes of loops at p are geodesic loops.11 Corollary. The geodesic reflection at p maps closed geodesics through p into their inverses. c (a) = X(p). g → gp. by the theorem of Bonnet–Myers. Let M be a symmetric space of non-compact type.10 and B. B.10. Let M be a symmetric space.140 ¨ Lectures on Kahler Manifolds Since c(a) = c(b) = p and X(p) = 0. Let V be the parallel vector field along c with V (0) = v. J is definitely not periodic. Set p = c(0) and u = c (0). Therefore √ J = cosh( −κt)V is the Jacobi field along c with J(0) = v and J (0) = 0. Flat tori and their Riemannian covering spaces are symmetric spaces of Euclidean type. u) < 0. hence Y = J. by Proposition B. that is.13. c (b) . p). Since X(p) = v ∈ Tp M was arbitrary.14 Corollary. that is. B. we get X(p). B. Shortest loops in homotopy classes of loops at p are geodesic and hence closed geodesics through p. if the sectional curvature of M vanishes. then K is connected. Hence sp induces the inversion on π1 (M. respectively. then M is compact and the fundamental group of M is finite. the first assertion is immediate from Propositions B. c )c = κV along c. negative. Then: 1) M is simply connected. Proof. non-compact type if the Ricci curvature of M is positive or. . u)u of Tp M has a negative eigenvalue κ and a corresponding unit eigenvector v perpendicular to u. Y is periodic along c. If M is a symmetric space of non-compact type.12 Definition. reads 1 = π1 (M ) → π0 (K) → π0 (G) = 1. we conclude that c (a) = c (b). Proof.

(G. Since σ∗ is involutive.16) where F = {g ∈ G | σ(g) = g} and F0 denotes the component of the identity of F . g = k + p. B. It follows that.2. (B.29 we show that a symmetric space has nonnegative sectional curvature iff its Ricci curvature is non-negative and. For the proof of the third. K) is a symmetric pair with involutive automorphism σ. (B. Since K ⊂ F . [k. We say that (G. and hence σ∗ commutes with all Adk .20). (B. a symmetric space is of Euclidean type if and only if its Ricci curvature vanishes. k is the Lie algebra of K. Y ] and hence [X. k is perpendicular to p with respect to the Killing form of g. n = dim M . σ∗ Y ] = [−X.19) . k ∈ K.1.14. Furthermore.18) Since F0 ⊂ K ⊂ F . then M is diffeomorphic to Rn . (B. [k.20) and Adk (p) ⊂ p for all k ∈ K.17) the eigenspaces of σ∗ for the eigenvalues 1 and −1. k] ⊂ k. K) is a symmetric pair if M = G/K is connected and if there is an involutive automorphism σ : G → G with F0 ⊂ K ⊂ F.2. via evaluation identified with the tangent space Te G of G at the neutral element e ∈ G. −Y ] = [X. Since σ∗ is an automorphism of g. that it has non-positive sectional curvature iff its Ricci curvature is nonpositive. respectively. In particular. In what follows. Y ] = [σ∗ X. Let (G. [p. by the theorem of Hadamard–Cartan and Corollary B. As for the proof of (B. the first and second inclusion follow from (B. let X. similarly. Note that the first two inclusions also follow by a similar argument. for K and G as in Corollary B.Appendix B Symmetric Spaces 141 B.1 Symmetric Pairs. K is a deformation retract of G.14. Y ∈ p. This improves Corollary B. Y ] ∈ k. p] ⊂ k.19) above. we have σ∗ [X. K) be a pair consisting of a Lie group G and a closed subgroup K. and let k = {X ∈ g | σ∗ X = X}. p] ⊂ p. p = {X ∈ g | σ∗ X = −X}. (B.15 Remark.14. In Remark B. In particular. Therefore Adk (k) ⊂ k Furthermore. if M is a symmetric space of noncompact type. Denote by σ∗ the differential of σ at e. σ(kgk −1 ) = kσ(g)k −1 for all k ∈ K and all g ∈ G. Let g be the Lie algebra of left-invariant vector fields of G.

G acts from the left on M . The geodesic symmetry s at o is s([gK]) = [σ(g) · K]. the isotropy represen= tation of K on To M corresponds to the restriction of the adjoint representation of K to p. π(g) = [gK] = λg (o). Let K be the stabilizer of o in G. (B. K) is a Riemannian symmetric pair with respect to the involution σ of G given by conjugation with the geodesic reflection so of M in o. · on p which is invariant under AdK .23 Example. k ∈ K. define a smooth vector field X ∗ on M by X ∗ (p) = ∂t (etX (p))|t=0 . that is. B. We denote by o = [K] the distinguished point of M and by π : G → M the canonical map. s ◦ λg = λσ(g) ◦ s. K) into a Riemannian symmetric pair with G/K = M . M is a symmetric space. λg (p) := gp. · . is the one-parameter group of transvections along this geodesic. 2. Then we have: 1. Let (G. t ∈ R.24 Theorem. Y ∗ ]. We denote by gp ∈ M the image of p ∈ M under left multiplication by g ∈ G. For X ∈ p. Show that [X. Sometimes it will be convenient to distinguish between the element g ∈ G and the diffeomorphism of M given by left multiplication with g.21 Exercise. K) be a Riemannian symmetric pair with corresponding involution σ of G and inner product ·. Let M = G/K and endow M with the G-invariant Riemannian metric corresponding to ·. Note that such inner products are in one-to-one correspondence with G-invariant Riemannian metrics on M . . For X ∈ g. is the geodesic through o with initial velocity π∗ X. We say that a symmetric pair is Riemannian if there is an inner product ·. Y ]∗ = −[X ∗ . Let M be a manifold and G be a Lie group which acts on M . Then (G. B. An easy computation shows that for all k ∈ K π∗ ◦ Adk = dλk (o) ◦ π∗ . We use π∗ = dπ(e) : p → To M to identify p and To M . In particular. The pull back of the inner product in To M to p is invariant under AdK and turns (G. By definition. Let M be a symmetric space and o ∈ M be a preferred origin. Then we will use λg to denote the latter. · on p.142 ¨ Lectures on Kahler Manifolds Since we divide by K on the right. and left multiplication by etX . under all Adk . t ∈ R. exp(tX) is the flow of X ∗ .22) Thus with respect to the identification p ∼ To M via π∗ . the curve etX (o). B. Let G be the group Iso(M ) of all isometries of M or the component of the identity in Iso(M ).

Hence X ∗ is the infinitesimal transvection with X ∗ (o) = π∗ X. Y ) = − B(X. Let p = g(o) ∈ M and u ∈ Tp M . 4. 5. Z ∗ are associated to X. Proof of Theorem B.24. It is clear that AdK -invariant tensors on p correspond to G-invariant tensor fields on M . then N = exp(q)(o) is such a submanifold and is a symmetric space in the induced Riemannian metric. It follows that s is an isometry. Z ∗ ](p). a subspace q ⊂ p is tangent to = a totally geodesic submanifold of M through o if and only if [[q. Y )Z = −[[X.4 as follows: R(X. Since X ∈ p. are the . we get − u X ∗ = ds( u X ∗ ) = ds(u) (s∗ X ∗ ) = u X ∗ . This proves (2). Let X ∈ p and X ∗ be the corresponding Killing field of M as in Exercise B. This completes the proof of (1). X ∈ p. Let u ∈ To M . 2 where B denotes the Killing form of g and X ∗ . Now the one-parameter groups (exp(tX)). where X. q] ⊂ q. 1 Ric(X. and hence X ∗ (o) = 0. B. With respect to the identification p ∼ To M . Y. Since s is an isometry. Y ). we have etX (s(p)) = s(e−tX (p)) for all p ∈ M . Y ].5 need not be closed. Y ∗ ].25 Remarks. we have u = dλg (v) = v and hence ds(u) = ds(dλg (v)) = dλσ(g) (ds(v)) = − dλσ(g) (v) = v = u . If the latter inclusion holds. Z ∈ p and both maps are considered on p. A good example for this is a non-rational line through 0 in a flat torus. Y )Z = − adZ ◦ adY .21. 2) The totally geodesic submanifold in B. Y ∗ . Moreover s(o) = o and ds(o) = − id.21.24.24. Y. Z] = −[[X ∗ . With respect to the identification p ∼ To M . q]. s is a diffeomorphism of M . hence s∗ X ∗ = −X ∗ . Z as in Exercise B. Y ) = − ad[X. With respect to the identification p ∼ To M . Choose v ∈ To M with dλg (v) = u. 1) We can rewrite the first formula in B. the curvature tensor R and = Ricci tensor Ric of M at o are given by R(X.Appendix B Symmetric Spaces 143 3. AdK -invariant tensors on p = correspond to G-invariant tensor fields on M and these are parallel. By the G-invariance of the metric.Y ] or R(·. It is easy to see that s([gK]) := [σ(g) · K] defines an involutive smooth map s of M with s ◦ λg = λσ(g) ◦ s. Since s is involutive.

Now N is totally geodesic at o since N = exp(q)(o) locally about o. B. Since N is invariant under s. Y. R(u. N = H(o). is a submanifold of M .27 Remark (Effective Pairs). In the proof of the second we can assume that X = Y . K) be a Riemannain symmetric pair and N ⊂ G be the normal subgroup of elements acting trivially on M = G/K.7. Y ∗ ] using the differential equation for Killing fields.26 Remark (Coverings). If we let s be the reflection in a chosen unit vector v ∈ S n and σ be conjugation with s. B. In particular. Spin(n)). The groups K with F0 ⊂ K ⊂ F correspond to Riemannian coverings G/F0 → G/K → G/F. then the component G of the identity in Iso(M ) is equal to SO(n + 1). If we pass to the quotient RP n keeping G. Since N is homogeneous. q]. N is totally geodesic everywhere. Since H is transitive on N and H acts isometrically on M . if M = S n ⊂ Rn+1 . Then l is a Lie subalgebra of k and h = l+q is a Lie subalgebra of g. Z ∗ be the corresponding Killing fields of M as in Exercise B. Y ∗ . .21. see [Hel. Z ∈ p and X ∗ . Real projective space RP n is a homogeneous space of Spin(n + 1). Let l = [q. Therefore tr ad2 X|k = tr ad2 X|p. Hence the necessity of the condition on q is immediate from the formula for R in (3). Note that for a totally geodesic submanifold N through the origin o. Hence (3). Since M is a complete Riemannian manifold and N is connected we have N = exp(q)(o). and the claim about the Ricci tensor follows. then we have K = F . the unit sphere S n is associated to the Riemannian symmetric pair (Spin(n + 1). Thus the symmetric space associated to a Riemannian symmetric pair (G. but is not associated to a Riemannian symmetric pair (G. A similar phenomenon occurs for some of the other symmetric spaces of compact type. This gives the first equation in (4). Theorem VII. Let H be the corresponding connected Lie subgroup of G. For example. N is symmetric. s|N is the geodesic symmetry of N in o. K) with G compact and simply connected is simply connected. Let X. then the stabilizer K of v is isomorphic to SO(n) and K = F0 = F . q] ⊂ q. Then X ∗ Y ∗ . It follows that Ginvariant tensor fields are parallel at o and hence parallel everywhere in M .2].144 ¨ Lectures on Kahler Manifolds transvections along the corresponding geodesics through o. q]. and Z ∗ are parallel at o. N = {g ∈ G | g acts as identity on M }. w ∈ To N . K) with G simply connected. Let (G. Then the orbit N of o under H. For example. Suppose now that [[q. Let u = Z ∗ (o) = π∗ Z and compute u [X ∗ . v. by the symmetry of the Ricci tensor. we have ad X(k) ⊂ p and ad X(p) ⊂ k. the unit sphere of dimension n. N is complete. v)w ∈ To N for all u. It remains to prove (5). Since X ∈ p. It is remarkable that the fixed point set of an involution of a compact and simply connected Lie group is connected.

Y ]]. adX |p = 0 for X = 0. where N is as in Remark B. the adjoint action of K on p factors over K/N . Hence G is semi-simple if and only if the Ricci curvature of M = G/K is non-degenerate. X . Therefore B is negative definite on k. Y ∈ p be tangent to Ei and Ej . K) is effective or. K/N ) is a Riemannian symmetric pair with (G/N )/(K/N ) = G/K = M and Lie algebra g/n ∼ k/n + p. Thus it becomes plausible that. Since the Ricci tensor of M = G/K is parallel. . by Theorem B. [X. The pair (G.29 Remark (Curvature). B([Y.20) we know that adX (k) ⊂ k and adX (p) ⊂ p. Furthermore. B([X.2.24. respectively. X]. Y ) = 2 Ric([X. K) is effective iff G is a subgroup of the isometry group of M . see Exercise B. [Y. Y )Y. Identify p = To M and let X. a discrete subgroup of G. K) be an infinitesimally effective Riemannian symmetric pair with K/N compact. Hence B(X. adX is skew-symmetric on p with respect to the given inner product on p since the adjoint representation of K on p preserves this product. hence the involution σ of G factors over G/N . On the other hand. where we use that Y is tangent to Ej . K/N ) is the effective pair associated to (G. Then. The above pair (G/N. X) = tr(adX |k)2 + tr(adX |p)2 = BK (X. From (B. Y ]) = −B([X. [X.28 Remark (Killing Form).27. Let X ∈ k. Y ]]. Hence N = {k ∈ K | Adk is the identity on p}. X]) = −2λi R(X. where B is the Killing form of G and BK the Killing form of K.2. It follows that (G/N. B. X)X. Now the square of a skew-symmetric endomorphism has negative trace unless the endomorphism vanishes.27. where N is as in Remark B. compare Exercise B. since N is discrete and X ∈ k. Since M is connected. Y ]]. X) + tr(adX |p)2 . K) be an infinitesimally effective Riemannian symmetric pair with K/N compact. B. X) ≤ tr(adX |p)2 . Y ) = 2λj [X. there are real numbers λi and pairwise perpendicular parallel distributions Ei on M such that Ric X = λi X for all X tangent to Ei . an isometry of M = G/K fixing the preferred origin o of M is determined by its differential at o. = We say that the Riemannian symmetric pair (G.Appendix B Symmetric Spaces 145 Then N ⊂ K. infinitesimally effective if N is trivial or. Y . respectively. it is sufficient to consider effective or infinitesimally effective pairs. By the same computation. [X. Y = −2λj R(Y. In particular. for many purposes. respectively. K). Let (G. [X. BK ≤ 0 and hence B(X.4. Let (G. Since N ⊂ K is a discrete subgroup and K/N is compact. Y ].

the distributions Ei give rise to a splitting of M as a Riemannian product with factors Mi tangent to Ei . Mi is a symmetric Einstein space with Einstein constant λi . Y ]) = 0 if λi = λj or if λi = 0 or λj = 0. the sectional curvature of a symmetric space is non-negative respectively non-positive iff its Ricci curvature is non-negative respectively nonpositive. (B. Y )Y. 24 For example. (B. if M is of compact type. [U. U = − [[X. If M is simply connected.31) By (B. K) be a Riemannian symmetric pair. if ·.15. F (X ∧ Y ) := [X. Suppose that the inner product on p is the restriction of an AdG -invariant bilinear form24 . Y ].34) ˆ Compare Remarks 1. and then. all λi < 0. 2λi In particular. respectively. then R(X. F (U ∧ V ) . · . 0}. also denoted ·. R(X. B. U ∧ V = F (X ∧ Y ). Y ]. a symmetric space of noncompact type is a complete and simply connected Riemannian manifold of non-positive sectional curvature. U = [X. Then all adX . non-compact type iff all λi > 0 or. respectively.56 for important consequences of R ≥ 0. Y ]. V ]. K) is infinitesimally effective and M = G/K is of compact or non-compact type. . Y ]. (B. By definition. Y ∈ p both tangent to Ei . For each i. this implies that the curvature tensor R vanishes on the kernel of Ric. In particular.32) ˆ R(X ∧ Y ). we see that we can renormalize the metric of M without destroying the geometry of M so that the constants λi are in {±1. [X.43) and define a morphism F : Λ2 p → k. we have ˆ Recall the definition of the curvature operator R in (1. Y )Y.33) In particular.31). on g. we can renormalize the metric so that M becomes an Einstein space with Einstein constant 1. Y ]. a negative or positive multiple of the Killing form of G if (G. with respect to the identification p = To M . Y ]). · is an inner product on g. Y ]|2 ˆ and R = F ∗ F ≥ 0. are skewsymmetric with respect to ·. [X. Let (G. For example. Y )V.30 Remark (Curvature Operator). respectively. X = |[X. (B. X ∈ g. M is of compact or.146 ¨ Lectures on Kahler Manifolds Hence B([X.14 and Remark B.42 and 5. V ] . Applying this to the universal covering space of M . we get R(X. · . Since B is negative definite on k. X = − 1 B([X. compare Corollary B. For λi = 0 and X.

B. If. 3) If λ > 0. Let λ be the Einstein constant of N . We conclude that g = h.37 Exercise (Dual Pairs). a compact symmetric space is of compact type if and only if its fundamental group is finite. In particular. By Remark B.3 for the converse assertion. Hence [h. h⊥ is an ideal as well. Y ) = − ad[X. Then π : G → M is a principal bundle with structure group K. Hence. In particular. then N is a Euclidean space. We have: 1) If λ < 0. [X. By assumption. if (G.20). etX eY e−tX = eY for all Y ∈ p and t ∈ R. K) is effective and M is of compact or non-compact type. then N is tangent to one of the distributions Ei as in Remark B. K) be a Riemannian symmetric pair and set M = G/K. Assume now that (G. Let M be a simply connected symmetric space. K) is infinitesimally effective with K/N compact. Since h is an ideal in g. In particular. h = [p. Hence the B-perpendicular complement h⊥ of h is contained in k and h ∩ h⊥ = 0.29. Since exp(tX) ∈ K. then K0 is equal to the reduced holonomy group of M . For X. Y ∈ p. Y ) are in the Lie algebra of the holonomy group Hol(M ).27. K). If N is a factor in the de Rham decomposition of M . the adjoint image of the component K0 of the identity of K is equal to the reduced holonomy group of M . moreover. then M is of compact type.36 Remark (Holonomy).20). the holonomy group Hol(M ) of M at o = [K] is contained in the image of K under the adjoint representation on p. The tangent bundle T M is (canonically isomorphic to) the bundle G×K p associated to the adjoint representation of K on p. B. Let (G. By (B.24. compare Theorem B. Y ] ∈ k and the curvature tensor is R(X. K) is infinitesimally effective. Moreover. We conclude that for X ∈ h⊥ . by (B. B|k is negative definite. B|p is non-degenerate.Appendix B Symmetric Spaces 147 B. that is k = [p. Hence X = 0 since (G. The left-invariant distribution on G determined by p is a principal connection which induces the Levi-Civita connection on M . From Remark B. G is semi-simple and B|h is non-degenerate with p ⊂ h. M is simply connected and G is connected. In particular. 2) If λ = 0.29 (but T N might be smaller than Ei ). then N is a symmetric space of non-compact type. the endomorphisms R(X. under the above assumptions on (G. where N is as in Remark B. Then parallel tensors on M are also G-invariant. this implies that exp(tX) acts as the identity on M . p] + p is an ideal in g. Moreover. h⊥ ] = 0. p]. Suppose furthermore that the Ricci curvature of M is non-degenerate. Let (G.Y ] on p.35 Remark (De Rham Decomposition). K) be a Riemannian symmetric pair and .29 we conclude that N is an Einstein space. then K is equal to the holonomy group of M at the preferred origin and hence G is equal to the component of the identity of the isometry group of M . k and p are perpendicular with respect to B.

53. R) with involution σ(A) := (At )−1 . (B. Y ∈ p defines a (new) Lie bracket on g. the curvature tensor changes sign and compact type corresponds to non-compact type. We also say that the corresponding symmetric spaces are dual. Y ] := [X. more geometrically via their unit balls.42. Its restriction to p is positive definite and turns (G. B. for any such transformation. as the space of ellipsoids in Rn with volume equal to the volume of the Euclidean unit ball. R). R) | U t = −U }. Since . or. To keep the presentation transparent. This subsection is devoted to examples which are relevant in our discussion. ·] . R) = {A ∈ Rn | tr A = 0}. 2) For each symmetric space M . identifies G/K with the space of positive definite symmetric (n × n)-matrices with determinant 1. then there is an orthogonal transformation A : Tp M → Tq N with N M A∗ Rq = −Rp and. The symmetric bilinear form A. Let G = Sl(n. K) if the corresponding decomposition of the Lie algebra g of G is up to isomorphism given by g = k + p with Lie bracket [ ·. K) into a Riemannian symmetric pair. Examples B. we divide them into three classes.38.148 ¨ Lectures on Kahler Manifolds g = k + p be the associated decomposition of the Lie algebra g of G. −[X. We have g = sl(n. and B. R) | X t = X}.39) on Rn is invariant under the adjoint action of Gl(n. Show that [X. 3) If M and N are simply connected dual symmetric spaces and K respectively L denotes the group of isometries fixing a point p ∈ M respectively q ∈ N . Y ] if X. B = tr AB 2 2 k = so(n) = {U ∈ sl(n. p = {X ∈ sl(n. Note that this change corresponds to passing from g to g = k + ip in the complexification gC of g.38 Example. We say that a Riemannian symmetric pair (G . there is a simply connected symmetric space M dual to M and M is unique up to isometry. The map G/K → Rn . K ) is dual to (G. 2 AK → AAt .2 Examples. Show: 1) Under duality. L = AKA−1 . The set of fixed points of σ is K = SO(n). B. B. Y ] if X ∈ k. We view these as the space of normalized inner products on Rn .

and Sp(p+q) for F = R. multiply vectors in Hp+q with scalars from the right and with matrices from the left. B) and 0 X −X t 0 =: P (X).59.6. respectively. where the involution on SU(n) is given by the passage to the conjugate matrix. The natural action of G on GF (p. (B.5. 4. X ∈ Fq×p ∼ Fq×p .2. X = [X.44) 0 B =: D(A. R)/ SO(2) with the above Riemannian metric has constant negative curvature −2 and construct isometries to some (other) models of the hyperbolic plane (with curvature −2). 1) Discuss (Gl(n. C. SO(n)) is a dual pair to (Sl(n. · is AdG -invariant. and B. The stabilizer of the p-plane Fp ×{0} ⊂ Fp+q is K = O(p)×O(q). q). q) = G/K. See also Exercise B. Hence (G.43) (B. We see that k + ip = su(n) = {U ∈ sl(n.40) where we note that ·. Then conjugation with S is an involutive automorphism of G with K as its set of fixed points.10. In this notation k= p= We write A 0 25 For 26 We U 0 0 X 0 V −X t 0 U t = −U. see the discussion after (B. hence GF (p. for X. We write square (p + q)-matrices as blocks of four matrices corresponding to the preferred decomposition Fp+q = Fp × Fq .69). C) | U t = −U }.83. Y ] ≤ 0.1. Y. 2) Show that Sl(2. B. Hence (SU(n).Appendix B Symmetric Spaces 149 ·.42 Example. · . U(p+q). where F = R. (B. F 4. and H. Let S be the reflection of Fp+q about Fp × {0}. These symmetric spaces are also discussed in Examples 2. Y ]. C. or H. each adA . respectively25 .10. Let G = O(p+q). B. We show that the Grassmann manifold GF (p. We also discuss the dual symmetric space G− (p.41 Exercises. K) is a symmetric pair. Z ∈ p. A ∈ g. O(n)) as a symmetric pair.4. [X. C) = {A ∈ Cn | tr A = 0}. Y )Y. · is negative definite on k. U(p)×U(q). is skew-symmetric with respect to ·. R). q) is transitive26 . R). . SO(n)). V t = −V . q) of p-planes in Fp+q together with a natural Riemannian metric is a Riemannian symmetric space. Therefore we have. R(X. 2 The complexification is gC = sl(n. and Sp(p)×Sp(q).45) the definition of Sp(n). = (B.

Xt 0 = P − (X). and P (X). P (Y ) ∈ p. q) of p-planes on which F the non-degenerate form Qp. Then G− is transitive on G− (p. P (X)] = P (V X − XU ).B) P (X) = P (BXAt ). XY t − Y X t ). K) into a Riemannian symmetric pair. q). P − (Y )] = D(X t Y − Y t X. Let G− = O(p. q). Its restriction to p turns (G. B = 1 Re tr(At B). .48) (B. where K is as above.q (x. q) is non-negative.46) [P (X). respectively. By the AdG -invariance of the inner product on g.47) Hence the sectional curvature of GF (p. we consider the open subset G− (p. and Sp(p. V ). There is an AdG -invariant inner product on g. B) ∈ K. P − (X)] = P − (V X − XU ). A. the curvature tensor of GF (p. q). Hence (G− . (B.150 ¨ Lectures on Kahler Manifolds For D(A. be the group of linear transformations of Fp+q preserving this form. = (B. Y X t − XY t ). [D(U. V ). we have AdD(A. Inside GF (p. q) and G− (p. Now we have p= 0 X Xt 0 X ∈ Fq×p ∼ Fq×p . q).50) whereas K and k are as above. V ) ∈ k.49) on Fp+q is negative definite. F F Conjugation with the reflection S of Fp+q about Fp × {0} is an involutive automorphism of G− which has K as its set of fixed points. Y )Y. 2 (B. [P − (X). y) = − xj yj + j≤p j>p xj yj (B. Y )Z = XY t Z + ZY t X − Y X t Z − ZX t Y. X = |[X. q) = G− /K. D(U. P (Y )] = D(Y t X − X t Y. K) is a symmetric pair as well. Y ]|2 ≥ 0. R(X. With respect to the identification P of Fq×p with p above. We write 0 X Then [D(U. q) is given by R(X. U(p.

A symplectic form on V is an alternating two-form ω on V such that for all non-zero v ∈ V there is a vector w ∈ V with ω(v. q) of oriented p-planes in Rp+q as homogeneous R space G/K. The choice of a symplectic basis identifies V with the standard symplectic vector space F 2n . y). q) = G/K. and V has a basis (e1 . Sp(n. . Show that the Riemannian metrics on GF (p. That is. and similarly with Hn . An endomorphism A : V → V is called symplectic if the pull back A∗ ω = ω. . fn ) such that ω(eµ . hence automorphisms. with respect to it k and p remain the same as before.Appendix B Symmetric Spaces 151 We see that with respect to the identifications P and P − of p with Fq×p above. . symplectic endomorphisms are invertible. v)) = (xµ vµ − yµ uµ ). F ) (B. denoted Sp(V ) and. . Discuss the corresponding inclusions of Grassmann manifolds. Let V be a vector space over a field F . As in this example. the Lie bracket [k.51 Exercise. q) → GR (p. q) F are Einsteinian with Einstein constant k(p + q + 2) − 4 and −k(p + q + 2) + 4. w) = 0. The standard example is F 2n with the form ω((x. Keeping the formula for the inner product on p.37. fν ) = 0 and ω(eµ . q) and G− (p. GF (p. GF (p. A symplectic vector space is a vector space together with a symplectic form. respectively. p] changes sign. then Go (p.52 Exercises. where k = dimR F. eν ) = ω(fµ . 2) Replacing orthogonal groups by special orthogonal groups. Let J= 0 1 −1 0 ∈ Gl(2n. dim V = 2n. we obtain the Grassmann manifold Go (p. p] remains the same. F ) if V = F 2n . q) is a twofold R Riemannian covering. . .53 Example. Let V be a symplectic vector space over a field F with symplectic form ω. but the Lie bracket [p. respectively. namely conjugation with the reflection S. (u. B. B. leaves G invariant. the dimension of a finite-dimensional symplectic vector space V is even. . fν ) = δµν . 1) Find other representations of Grassmannians as homogeneous spaces.54) . q) is simply connected. If p + q ≥ 3. en . Since ω is non-degenerate. . q) are dual symmetric spaces in the sense of ExF ercise B. f1 . B. Under composition. R 3) A complex subspace of dimension p in Cn is also a real subspace of 2n R = Cn of dimension 2p. a symplectic basis of V . The same involution σ as above. the set of all symplectic automorphisms of V is a group. and discuss which of the corresponding pairs (G. There is a corresponding change in sign for curvature tensor and sectional curvature. K) are Riemannian symmetric pairs. where G = SO(p+q) and K = SO(p)×SO(q). The natural “forget the orientation” map Go (p. q) and G− (p. the formula for the curvature tensor remains the same.

w + iω(v. w = Re v. . . n) = U(n)/ O(n). −y) of R2n . A→ A 0 . D (B. bn ) is an orthonormal basis of L. and At D − C t B = 1. F ). . y) = x + iy = z. y) = (x. w and ω(v. F ) into O(2n. Therefore the natural action of U(n) on Cn induces a transitive action on GR (L. w) are the Euclidean inner product and the standard symplectic form on R2n . w ∈ L. . . where here and below scalars represent the corresponding multiple of the unit matrix or block.57) This induces the diagonal embedding of O(n. Then we have σ(A) := SAS −1 = A for any A ∈ U(n). Writing g as a matrix of (n × n)-blocks. . . f1 . en is Lagrangian. . Thus GR (L. respectively. n). . B t D = Dt B. . n) the space of Lagrangian subspaces of V and F 2n . w) = 0 for all v. Let S : Cn → Cn be the real automorphism S(z) = z corresponding to the reflection S(x. w . O(n)) into a Riemannian symmetric pair. .58) GR (L. . w). n) is a symmetric space of dimension n(n + 1)/2 and a totally geodesic submanifold of the Grassmannian GR (n. We denote by G(L. V ) and GF (L.152 ¨ Lectures on Kahler Manifolds be the fundamental matrix of the standard symplectic form ω in F 2n . en . The real subspace L0 spanned by the unit vectors e1 . respectively. bn ) is a unitary basis of Cn with its standard Hermitian form v. . . then (b1 . F ). (B. We have v. Then a linear map g ∈ Gl(2n. then Lagrangian subspaces of V have dimension n. A subspace of V is called a Lagrangian subspace if it is a maximal isotropic subspace. fn ) is a symplectic basis of V . en in Cn is Lagrangian. then the subspace L0 spanned by e1 . . If L is any other Lagrangian subspace of V and (b1 . turning (U(n). g= A C B . n) with stabilizer O(n) of L0 . F ) → Sp(n. Let F = R and identify R2n = Cn by writing (x. . . F ) with B = C = 0 is an embedded general linear group.56) It follows that the group of g ∈ Sp(n. Gl(n. . . hence (B. where the left hand side is the standard Hermitian form on Cn and where Re v. F ) is in Sp(n. If (e1 . . . If the dimension of V is 2n. .55) the condition g t Jg = J is equivalent to the conditions At C = C t A. Hence the involution σ on U(n) has O(n) as its set of fixed points. . A subspace L of V is isotropic if ω(v. F ) if and only if g t Jg = J. . 0 (At )−1 (B. .

37.62).n . By (B. We conclude that the intersection O(n. (B. and hence D = A and B = C. At A − C t C = 1. n) = U(n)/ O(n). To that end. The symmetric space Gl(n.49) is negative definite. n) iff At B = C t D. and that GR (SL. R). . R)/ O(n) is dual to U(n)/ O(n) = GR (L.56) and (B. Writing D = AX.60) Write g ∈ U(n) as g = A + iB with real (n × n)-matrices A and B. A ∈ O(2n) is complex linear iff the pull back A∗ ω = ω. compare with (B. In particular.Appendix B Symmetric Spaces 153 B.n from (B. R)/ O(n) as the open set G− (L. Then this identification reads U(n) A + iB ←→ A B −B A ∈ O(2n) ∩ Sp(n. Let g ∈ O(n. and Dt D − B t B = 1.57) above. n) ∩ Sp(n. we would like to see if we can view Gl(n.65) with At B = B t A and At A − B t B = 1. R) with B = 0 is a diagonally embedded O(n). R) belongs to O(n.63) Hence X = 1.64) (B. (B. Motivated by the example of the Grassmannians. n) of Lagrangian subspaces on which the symmetric form Qn. n) in the sense of Exercise B. Under the above identification R2n = Cn . O(n) = O(n. B = (At )−1 C t D = CA−1 D. U (n) = O(2n) ∩ Sp(n. Clearly g ∈ Gl(2n. R). R).61) where At B = B t A and At A + B t B = 1. n) be the orthogonal group of Qn. We say that a Lagrangian subspace L of V = Cn (as above) is special if the n-form dz 1 ∧ · · · ∧ dz n is real valued on L.55). R).62) where we write g as a matrix of blocks as in (B. let O(n. R) ⊂ U(n) is diagonally embedded in O(2n) ∩ Sp(n. We also see that the subgroup of matrices in O(n.59 Exercise. n) ⊂ R GR (L. n). we compute (1 + C t C)X = At AX = At D = 1 + C t B = 1 + C t CX. n) ∩ Sp(n. n) is totally geodesic in GR (L. (B. n) of all special Lagrangian subspaces of V is given by the submanifold SU(n)/ SO(n) of GR (L. n) ∩ Sp(n. that the space GR (SL. Show that the above Lagrangian subspace L0 is special. (B. R) consists of the matrices A B B A (B. In other words.

aw = z. . u ) = ((x. bn ) R be an orthogonal basis of L. n)∩Sp(n. where the first term on the right corresponds to the standard Hermitian form and ω to the standard symplectic form on C2n . R) as above. k O(n)k −1 = O(n). elements of Sp(n) preserve the Hermitian form and the symplectic form on C2n . . R))/ O(n) = G− (L.n (bj . Considered as linear maps on C2n . R) as embedded subgroups in Sp(n. that is. bk ) = −δjk . (B. R)k −1 = O(n. .65) belongs to O(n. denoted Sp(n). v − y. . v)) + jω((x. hence G− (L. R) is transitive on G− (L.154 ¨ Lectures on Kahler Manifolds Let L0 ∈ G− (L. n) ∩ Sp(n. n) = (O(n. u + y. R Let 1 k := √ 2 1 1 −1 1 ∈ O(2n) ∩ Sp(n. For any unit quaternion a. . v)). (u. so that left multiplication by a defines an element of Sp(n). The group of quaternionic (n×n)-matrices preserving the form z. . Let L ∈ G− (L. . . For z = x + jy and w = u + jv in Hn . we get z. aw = za. If L ⊂ C2n is any Lagrangian subspace of C2n and (b1 . v ) + j( x. . . n) be another Lagrangian subspace. Since L is Lagrangian and (b1 . R).66) (B.68) where we consider O(n) and Gl(n. . bn ) . n) ∩ Sp(n. (B. We have z = x − jy and jx = xj. n) ∩ Sp(n. the corresponding matrix g as in (B. y). en in Hn ∼ C2n = is Lagrangian. n) ∩ Sp(n. n) ∩ Sp(n. y ∈ C2n . Let (b1 . w = x − jy. We conclude that conjugation with k identifies Gl(n. R). az. we conclude that O(n. y). . . where we multiply with scalars from the right. R) R with B = 0. R). . Let A and B be the (n × n)-matrices with columns the first n and the last n coordinates of the vectors bj . n). Since g maps L0 to L.69) Let F = C and identify C2n = Hn by writing z ∈ Hn as z = x + jy with x. . (u. n) be the Lagrangian subspace spanned by the first n unit R vectors. The complex subspace spanned by the unit vectors e1 .67) We have k Gl(n. respectively. w on Hn is called the symplectic group. u + jv = ( x. Qn. R (B. R)/ O(n) with (O(n. The stabilizer of L0 is the subgroup of matrices in O(n. bn ) is an orthogonal basis of L. R))/ O(n). n). . w .

Under the identification C2n = Hn . we have SAS −1 = σ(A).73). .n iff At B = C t D. R) as in (B.86. We see that S corresponds to the reflection S(x.60). . bν = δµν . . see Proposition B. n) ⊂ GC (L. In other words. C). (B. Writing D = AX. (B. GC (L. At A − C t C = 1. We proceed as in the real case. defined by the analogous formula S(x + jy) = x − jy. −y) of C2n . y) = (x. Let g ∈ U(n. R)/ U(n) with the open subset G− (L.74) . B = (At )−1 C t D = CA−1 D. bn ) defines an element of Sp(n). . Since the center of U(n) is not discrete. (B. U(n) ⊂ Sp(n) is diagonally embedded in U(2n) with blocks A and A on the diagonal.55).75) (B. n). . By (B. (B. n) is a symmetric space of dimension n(n + 1) and a totally geodesic submanifold of GC (n. Hence mapping the standard basis (e1 . where σ(A) is the matrix obtained by applying s to the entries of A. .71) Write g ∈ Sp(n) as g = A + jB with complex (n × n)-matrices A and B. turning (Sp(n). and Dt D − B t B = 1. Since s is an automorphism of H. First of all we note that g ∈ Gl(2n.Appendix B Symmetric Spaces 155 is a unitary basis of L. In particular. en ) of Hn to (b1 . then bµ .56) and (B. A ∈ U(2n) is quaternion linear if and only if the pull back A∗ ω = ω.n from (B. C).73) where we write g as a matrix of blocks as in (B. n) = Sp(n)/ U(n). we compute (1 + C t C)X = At AX = At D = 1 + C t CX.49) is negative definite. U(n)) into a Riemannian symmetric pair. It follows that GC (L. Then this identification reads Sp(n) A + jB ←→ A B −B A ∈ U (2n) ∩ Sp(n. . n) is a Hermitian symmetric space. (B. n) of Lagrangian subspaces on which the C Hermitian form Qn. Clearly σ is an involution of Sp(n) with U(n) as its set of fixed points. Again we want to identify the dual symmetric space Sp(n. n) ∩ Sp(n. C) belongs to the unitary group U(n. Sp(n) = U(2n) ∩ Sp(n. where U(n) = O(2n) ∩ Sp(n. n) of Qn. C). .72) where At B = B t A and At A + B t B = 1. Thus GC (L.70) The automorphism s(x + jy) = x − jy of H induces a complex automorphism S : Hn → Hn .

R)k −1 = U(n. Section 31] or [Hel.87. (B.76) Let with At B = B t A and At A − B t B = 1. n) ∩ Sp(n. Exercise VIII. (B. n) = (U(n. Moreover.79) Let g ∈ Sp(n.156 ¨ Lectures on Kahler Manifolds hence X = 1. C) consists of the matrices A B B A (B. [Bu. C). Then k −1 gk ∈ Sp(n.B]. that is. A Hermitian symmetric space is a symmetric space M together with a parallel almost complex structure on M which is compatible with the Riemannian metric on M in the sense of (2. n) ∩ Sp(n. Since parallel almost complex structures are complex structures. n) is a Hermitian symmetric C space. and hence D = A and C = B. The latter holds iff D = A and C = B. C))/ U(n). We also have k(O(2n) ∩ Sp(n. see Remark B.82) C . a Symmetric spaces of non-compact type are simply connected (Corollary B. R))k −1 = U(n). n) is biholomorphic to the Siegel C upper half plane. The subgroup of g ∈ U(n. G− (L. C). R) iff k −1 gk = k −1 gk. We finally conclude that conjugation with k identifies Sp(n. we get G− (L.78) We have k −1 = k and hence kk −1 = k 2 = 0 i . R)/ U(n) with (U (n. n). any such structure turns M into a K¨hler manifold. iff kk −1 g = gkk −1 .14).81) Since the center of U(n) is not discrete.g.80) (B. C) with B = 0 is a diagonally embedded U(n) as above. (B. G− (L. iff g ∈ U(n. C). (B. C) as above. C).33). that is.3 Hermitian Symmetric Spaces. n) ∩ Sp(n. but there are symmetric spaces of compact type with non-trivial fundamental where we consider U(n) on the right as an embedded subgroup in Sp(n. n) ∩ Sp(n. As in the real case. B. see e. For more on this. We conclude that k Sp(n. We conclude that the intersection U(n. n) ∩ Sp(n. n) ∩ Sp(n. i 0 (B. C))/ U(n) = G− (L.77) C 1 k := √ 2 1 i i 1 ∈ U(2n) ∩ Sp(n.

46). in the case of complex projective space. 27 There are more direct proofs for the latter assertion which do not rely on Kobayashi’s theorem. [Hel. However. etX = tX 1 Comparing with (4. we have 1 −tX t + o(t). We define a complex structure on p by JP (X) := P (iX). Y X − X.B) P (X) = P (BXAt ). U(1. q) here with the one introduced in Example 4.24. JY JY.42 and consider GC (p. hence J commutes with AdK . We recommend Karcher’s article [Kar] for a nice exposition of the elementary geometry of complex projective spaces. To compare the Riemannian metric on GC (p. Thus the holomorphic sectional curvature of complex projective space is constant equal to 4 and the range of its sectional curvature is [1. q) = U(p + q)/ U(p) × U(q) = G/K.48). where X ∈ Cq×p and P (X) ∈ p is defined as in (B. We follow the notation in Example B. we get that the Riemannian metric there is twice the Riemannian metric discussed here. see e. Then by Theorem B.6]. turning GC (p. Let (G.g. It is dual to complex projective space. B.14).45).Appendix B Symmetric Spaces 157 group. For X ∈ p.10. For p = 1.1227 . C The case p = 1 is of particular importance. . the Grassmannian of complex p-planes in Cp+q . For D(A. The discussion in the case of the dual Grassmannians G− (p. q) is given in (B. Y )Y = Y. compatible parallel complex structures on M = G/K correspond to complex structures on p which preserve the inner product on p and are invariant under the adjoint action of K on p. B) ∈ K we have AdD(A. endowed with the Riemannian metric introduced in (B. we note that both are invariant under the action of U(p + q). Hermitian symmetric spaces of compact type are Fano manifolds and therefore simply connected. by Theorem 6. we get R(X. −1]. the range of its sectional curvature is [−4. Y Y + 3 X. q) into a Hermitian symmetric space. With respect to the given normalization of the Riemannian metric. q) is similar. Theorem VIII. 4].83 Examples.4.3. Thus it suffices to compare them at the plane spanned by the first p unit vectors. 1) Complex Grassmannians. that is. m)/ U(1) × U(m) =: CH m is called complex hyperbolic space. The curvature tensor of GC (p. K) be a Riemannian symmetric pair.

6 and (4.52. Moreover. the endomorphisms cos x · id + sin x · Jp of Tp M preserve metric and curvature tensor. 0] ∈ Q. what amounts to the same. by Corollary B. By Exercise B. . n − 2). R Intersecting two-planes with the unit sphere S n−1 .2. n − 2) = SO(n)/ SO(2) × SO(n − 2).84) is a complex projective line.158 ¨ Lectures on Kahler Manifolds 2) Complex quadric. The orbit of p0 under the induced action of SO(n) on CP n−1 is equal to Q. n − 2) as the complex quadric R To that end. we get that the Riemannian metric induced from CP n−1 .2. 1 2 2 Q = {[z] ∈ CP n−1 | z1 + · · · + zn = 0}. 0. . normalized as in Example 1 above. i. a complex projective line. thus Q = SO(n)/ SO(2) × SO(n − 2) = Go (2. jp belongs to the component of the identity of the stabilizer of p in the group of holomorphic isometries of M . think of the = two columns of X ∈ R(n−2)×2 as real and imaginary part of a vector in Cn−2 . For X ∈ p. Consider the Grassmannian manifold Go (2. Go (2. by Corollary B. Since A is unitary.84) where 1r denotes the unit matrix of size r × r. A maps p1 = [1. In particular. Thus Go (2. R (B. . Let M be a simply connected Hermitian symmetric space with complex structure J. 1) is a R two-sphere or. The stabilizer of p0 is SO(2) × SO(n − 2). . . we let A= A2 1n−2 with 1 A2 = √ 2 1 i i . R To compare the metric chosen above with the one on CP n−1 . since Jp is skew-symmetric and orthogonal. . Thus the quadric Q ⊂ CP 2 in (B. n−2). 0] to p0 . n − 2) together with the Riemannian R metric induced by the inner product on p and the parallel complex structure induced by J is a K¨hler manifold. B. we get etX p0 = etX Ap1 = A(A−1 etX A)p1 .85 Remark. n − 2) of R oriented real 2-planes in Rn as in Exercise B. Hence. . a We can think of Go (2. Then Jp is the differential of an isometry jp of M fixing p. extend the action of SO(n) on Rn complex linearly to Cn to realize SO(n) as a subgroup of U(n). We have A−1 etX A = 12 tXA2 −tAt X t 2 + o(t). There is a complex structure J on p which is invariant under the adjoint representation of SO(2) × SO(n − 2): Identify R(n−2)×2 ∼ Cn−2 .6. Let p0 = [1. and multiply this vector by i. Moreover. is equal to the chosen metric on Go (2. Go (2. that is. and let p be a point in M . n − 2) R with the space of oriented great circles on S n−1 . jp is in the center of the group of holomorphic isometries of M fixing p. 1n−2 Since the speed of the curve etX p0 at t = 0 is equal to the speed of the curve (A−1 etX A)p1 at t = 0. . . we identify Go (2. A induces an isometry of CP n−1 . 0.43).

K) is infinitesimally effective and K is compact. then M is a Hermitian symmetric space of compact type. The final part of the long exact homotopy sequence associated to the fiber bundle G → G/K = M is 0 = π2 (G) → π2 (M ) → π1 (K) → π1 (G) → π1 (M ) → 0. If M is a compact Hermitian symmetric space. Let Z ⊂ End(p) be the space of all endomorphisms of p commuting with all Adk . K/N ). Assume that G is semisimple and that the adjoint representation of K on p is irreducible. Since M is compact with finite fundamental group.27 is finite. the universal covering space of M is compact. Hence π2 (M ) is finitely generated. K) is effective. Proof. = Since the adjoint representation of K on p is faithful. Z ∼ C or Z ∼ H. then the center ZK of K is a circle and J is (any) one of the two elements of order 4 in ZK . (G. equivalently. there are no other but the above two choices for an invariant complex structure on M. Since K is compact and connected. M is a symmetric space of compact type. assume that the center of K is infinite. Thus J turns M = G/K into a Hermitian symmetric space of compact type. Since G is compact and semi-simple. R) = 0 if and only if π2 (M ) is infinite. K ⊂ End(p).86 Proposition. Z is an associative division algebra over R. then J 2 = −1 and J is an AdK -invariant complex structure on p. we may assume that the adjoint representation of K on p is faithful. if the center of K is not discrete. Furthermore. and hence H 2 (M. hence simply connected. Therefore the center of K/N is infinite as well. it follows that ZK is a circle. Since maximal = = Abelian subgroups of H× are isomorphic to C× .Appendix B Symmetric Spaces 159 Recall the definition of infinitesimally effective from Remark B. In particular.85. Since K is compact. B. π2 (M ) is infinite if and only if π1 (K) is infinite. π1 (G) is finite and π2 (G) = 0. this implies that π1 (K) is infinite. Hence we may pass to the quotient pair (G/N. ZK is a compact subgroup of Z × . In conclusion. If J ∈ ZK is one of the two elements of order 4. By Remark B. hence Z ∼ R. then M is K¨hlerian and a hence H 2 (M. C. If the latter holds. if and only if K is not semi-simple. By what we just said. that is. It follows that π1 (M ) is finite. Furthermore. Hence if ZK is not finite. K) be an infinitesimally effective Riemannian symmetric pair with G and K compact and connected. Hence the center ZK of K is a subgroup of Z × . compatible with the inner product on p since ZK ⊂ K. . Then M = G/K admits a G-invariant complex structure compatible with the Riemannian metric on M if and only if the center of K is not discrete.27. this happens precisely if the center of K is infinite or. moreover. Since (G. k ∈ K. Since the adjoint representation of K on p is irreducible. If. R) = 0. the normal subgroup N as in Remark B. M is simply connected. which is also a Riemannian symmetric pair. or H. Let (G. Vice versa.

see Example 4. K) and (G . The article [Ma2] by Matsushima contains a nice introduction into tube domains. In particular. Recall that the Bergmann metric of D is invariant under all biholomorphic transformations of D. K).7. We have seen this explicitly28 in the case of the dual complex Grassmannians G− . there is a biholomorphic transformation sp : D → D such that sp (p) = p and dsp (p) = − id. hence.5 and Theorem B. transformations sp as above are geodesic reflections. U(n)) as a Riemannian symmetric pair. G be the group of isometries of M and K be the stabilizer of a point in M . R)/ U(n) as a space of complex structures on R2n .10. The irreducibility of the isotropy representation for the two metrics under consideration shows. In other words. r.37 with K = K . Each tube domain in Cm is biholomorphic to a bounded domain in Cm . for all y ∈ Ω and t > 0. that they are of non-compact type. and. where Ω ⊂ Rm is an open and convex subset such that.7. but the converse does not hold. Let (G. equipped with the Bergmann metric. 1) Interpret O(2n)/ U(n) as space of complex structures on R2n which preserve the Euclidean norm and discuss (O(2n). K ) be dual symmetric pairs as in Exercise B. For example. Then a complex structure J on p is invariant under the adjoint representation of K and compatible with the inner product on p if and only if the corresponding complex structure J on p is invariant under the adjoint representation of K and compatible with the inner product on p . Hence M is Hermitian iff M is Hermitian iff K = K is not semi-simple.160 ¨ Lectures on Kahler Manifolds B. Then M is simply connected and given by the symmetric pair (G. ty ∈ Ω and such that Ω does not contain complete lines. for all p ∈ M .87 Remark. . Interpret Sp(n. the criterion of Proposition B. It follows that symmetric bounded domains with the Bergmann metric are Hermitian symmetric spaces. Then M is of compact type and G is compact and semi-simple. 28 We did not identify the Bergmann metric there. p = ip and hence K = K by Theorem B. Up to isomorphism.24. B.7. 2) Determine the space of all complex structures on R2n as a homogeneous space G/K.4. that they must be multiples of each other.1 in [Hel]. see Theorem VIII. for any Hermitian symmetric space M of non-compact type. an isometry between M and D.10. For example. We say that D is symmetric if. up to rescaling. see Example 4.86 also applies to symmetric spaces of non-compact type.n We say that D ⊂ Cm is a tube domain if D = Rm + iΩ.10.6. see Example 4. Let D ⊂ Cm be a bounded domain. there is a symmetric bounded domain D and a biholomorphic map M → D which is. the upper half plane is a tube domain. It then also follows that they are Einstein spaces with Einstein constant −1. G be the component of the identity in the group of isometries of M and K be the stabilizer of a point in M . Let M be the simply connected dual symmetric space of M .88 Exercises. however. let M be a symmetric space of non-compact type. Vice versa.

Appendix B Symmetric Spaces 161 We say that a Hermitian symmetric space is of tube type if it is biholomorphic to a tube domain. [KW]. R)/ U(n) (as in Example B. [Wi] for more information on Hermitian symmetric spaces of tube type.B in [Hel]. . The dual Lagrangian Grassmannians G− (L.53) are of tube C type. see Exercise VIII. The hyperbolic plane is of tube type. n) = Sp(n. See [BIW].

We say that D is formally self-adjoint or formally symmetric if E = F and D = D∗ . (C. .Appendix C Remarks on Differential Operators Let M be a Riemannian manifold. F ). Xn ) be an orthonormal frame of M about p with Xj (p) = 0. we have at p (Dσ.20). by the product rule (1. Using the latter and that D is metric. ∗ Note that F = TC M ⊗ E inherits a Hermitian metric from the Hermitian ∗ ˆ metrics on TC M and E and. By the divergence formula 1. (C. .9. a 1-form with values in E. Let E. there is a complex vector field Z = X + iY on M with (Dσ. F → M be Hermitian vector bundles and D be a differential operator of order m ≥ 0 from (smooth sections of) E to (smooth sections of) F . We also use the shorthand E(E) and Ec (E) for these spaces.4 Example. (C. This is the way we will determine adjoint operators. we denote by E(M. E) the Hilbert space of square integrable measurable sections σ of E. We also view D as an unbounded operator on L2 (M. . If E is Hermitian. We view D as a differential operator. τ (Xj )) − (σ. τ ) = = =− (DXj σ.5) D : E(E) → E(TC M ⊗ E). ∗ (C. respectively. Let E be a Hermitian vector bundle over M and suppose that E is endowed with a Hermitian connection D. (DXj τ )(Xj )) + div Z. Let p ∈ M and (X1 . by definition.6) . D∗ τ ) + div Z. (Dσ. (DXj τ )(Xj )) ˆ (σ.2) and corresponding L2 -norm σ 2 . we denote by L2 (E) = L2 (M. endowed with the L2 -inner product from (1. E) and Ec (M. C. Let σ be a section of E and τ be a section of T ∗ M ⊗ E.1) for all σ ∈ Ec (M. F ). For a vector bundle E → M . τ ) = (σ.2) D : E(M. E) → E(M. D∗ τ )2 (C.3) where the divergence div Z := div X + i div Y . an equivalent characterization of D∗ is that. τ )2 = (σ. E) with domain Ec (M. The discussion in the case of real vector bundles and Riemannian metrics is similar. that from C is. E). for all smooth sections σ of E and τ of F . a metric connection D ˆ and D. τ (Xj )) ˆ Xj (σ. E) and τ ∈ Ec (M. E) the space of smooth sections of E and smooth sections of E with compact support. . The formal adjoint D∗ of D is a differential operator of order m from F to E and satisfies.

Furthermore. Riemannian or Hermitian. The operator Dmax is closed and extends D. ϕ]σ. if D has order one. (C. W ) = (σ. 1 SD (dϕ)σ = [[. (C. then SD is tensorial in ξ = dϕ and SD (dϕ)σ = D(ϕσ) − ϕD(σ). By definition. In both cases. For example. C. E) belongs to the domain dom Dmax of Dmax iff there is a section σ ∈ L2 (M.13) and then Dmax σ := σ .5). By definition.10) If E = F . 29 Section III. 3) Let D be a connection on E. . τ )2 = (σ. The principal symbol is homogeneous of order m in ξ. τ (W )). F ). ϕ]. Xj ) = − tr D2 σ. (C. . D is closable. . D∗ τ )2 for all τ ∈ Ec (M. (C. . ϕ]. where p is the foot point of ξ. E).9) for all smooth functions ϕ on M and sections σ of E. 2) Show that SD∗ (ξ) = (−1)m SD (ξ)∗ for all ξ ∈ T ∗ M .11) m! where ϕ is viewed as the operator which multiplies sections with ϕ. where we note that ker Dmax is a closed subspace of L2 (M. Recall also that D is elliptic if SD (ξ) is invertible for all non-zero ξ ∈ T ∗ M . In particular. viewed as a differential operator as in (C.5 in [Kat] is a good reference for the functional analysis we need.8) D∗ Dσ = − The principal symbol SD of a differential operator D as above associates to each ξ ∈ T ∗ M a homomorphism SD (ξ) : Ep → Fp .14) an L2 -orthogonal decomposition of L2 (M. The maximal extension Dmax of D is the adjoint operator of D∗ in the sense of Hilbert spaces29 . E) such that (σ . E) = ker Dmax + im D∗ . E). Show that SD (dϕ)σ = dϕ ⊗ σ. D2 σ(Xj . We conclude that D∗ τ = − ˆ ˆ (DXj τ )(Xj ) = − tr Dτ.9) can be written as an m-fold commutator. (C. (C. SD (tξ) = tm SD (ξ). then the right hand side of (C. . L2 (M. 1) Compute SD in terms of local trivializations of E. .Appendix C Remarks on Differential Operators 163 where Z is the complex vector field defined by the property (Z. SD (dϕ)σ = 1 m! m (−1)j j=0 m j ϕ D(ϕm−j σ) j (C. [[D. . σ ∈ L2 (M.12 Exercises. ϕ].7) The same formula holds in the Riemannian context.

and then Dminσ := lim Dσn . By passing to the elliptic and formally selfadjoint operator 0 D∗ . E) such that σn → σ in L2 (M. Let D be elliptic. D 0 conclude that Amin = Amax .16) The left inclusion is always strict. where the superscript a denotes the Hilbert space adjoint. |σ|D := (|σ|2 + |Dminσ|2 )1/2 . Let σ ∈ E(M.7]. formally self-adjoint. By definition. We say that a differential operator D is essentially self-adjoint if it is formally self-adjoint and Dmin = Dmax .5. If M is complete and connected. k ≥ 1. the Sobolev space associated to D. In [LM.164 ¨ Lectures on Kahler Manifolds C. The smallest closed extension Dmin of D is called the minimal extension of D. C. C. (C.15 Exercise. E) such that there is a sequence (σn ) in dom D = Ec (M. the case k = 1 is treated by a cut-off argument and by using an appropriate parametrix for D. Daaa = Dmax .18 Exercises. C. show that Da = Dmax .20 Exercise. of D is essentially self-adjoint. 1) Show that dom Dmin endowed with the graph norm of D. F ). 2) Assuming D = D∗ . By definition. hence (C.17) The minimal extension Dmin can also be characterized as the Hilbert space adjoint of the maximal extension (D∗ )max . The domain dom Dmin of Dmin consists of all σ ∈ L2 (M. E). the graph of Dmin is the closure of the graph of D and D ⊂ Dmin ⊂ Dmax .19 in [Chf] uses an existence and uniqueness result for the associated wave equation. where A is any product with alternating factors D and D∗ . Theorem II. 2 2 is a Hilbert space. Daa = Dmin . then σ ∈ dom Dmax and Dmax σ = Dσ. Then if σ and Dσ are square integrable. im Dmin ⊂ im D. then any power Dk . im Dmin = im D.19 Theorem (Chernoff [Chf]). . Let M be complete and connected and D be elliptic of order one with SD uniformly bounded. The elegant proof of Theorem C. E) and the sequence (Dσn ) is convergent in L2 (M. and of order one with SD uniformly bounded.

We also note that X · (X · σ) = −|X|2 σ.21 Proposition. called Clifford multiplication. C. (C. Hence if M is complete and connected. We say that a Hermitian vector bundle E over a Riemannian manifold M with Hermitian connection is a Dirac bundle if it is endowed with a field of morphisms T M ⊗ E → E. If M is complete and connected. then D∗ Dσ = 0 ⇐⇒ Dσ = 0.22 Corollary. 2 Proof. hence (Dσj ) is a Cauchy sequence as well. Let D be elliptic of order one with SD uniformly bounded. k ≥ 1.19. X ⊗ σ → Xσ = X · σ. XY σ = −Y Xσ − 2 X. If M is complete and connected and σ ∈ dom(D∗ D)max . Xτ ). by Theorem C. hence also in the limit. By Exercise C. [LM] play a central role. such that (Xσ. Among the elliptic differential operators of order one. and hence dom(D∗ D)min ⊂ dom Dmin . Dmax = Dmin and (D∗ D)max = (D∗ D)min .Appendix C Remarks on Differential Operators 165 C. then any power Dk . Clearly the asserted formula holds for any σj .23) The associated Dirac operator D : E(E) → E(E) is then defined by Dσ = Xj Xj σ. hence D is elliptic.20. τ ) = −(σ. and let σ ∈ E(M. Y σ. X (Y σ) = ( X Y )σ + Y ( (C. Let (σj ) be a sequence of smooth sections of E with compact support such that (σj ) and ((D∗ D)σj ) are Cauchy sequences in L2 (M. X σ). (C.1 Dirac Operators. We have |Dσj − Dσk |2 = ((D∗ D)(σj − σk ).25) We see that SD is uniformly bounded. Dirac operators in the sense of Gromov and Lawson [GL]. . Let D be elliptic of order one with SD uniformly bounded. σ)2 = |Dσ|2 . (σj − σk ))2 2 ≤ |(D∗ D)(σj − σk )|2 |(σj − σk )|2 .24) where (Xj ) is a local orthonormal frame of M . E). We leave it as an exercise to show that D is formally self-adjoint with principal symbol SD (dϕ)σ = grad ϕ · σ. C. E) be square integrable. of D is essentially self-adjoint. then σ ∈ dom Dmax and (D∗ Dσ.

E = E + ⊕ E − .28). Then the associated eigenbundles E ± for the eigenvalues ±1 of µ are parallel and turn E into a graded Dirac bundle. endowed with a compatible Riemannian metric g.28) X · α := 2ZX ∧ α − 2ZX α satisfies the axioms for Clifford multiplication in (C. . 2 compare Subsection 3. E) is the twist of A∗ (M. Furthermore. endowed with a Hermitian connection. see (1.11). the twist of E by F . E) = Aeven (M. C. E) into a graded Dirac bundle.∗ (M.12). √ where we use that iZX = ZJX and iZX = −ZJX .32) and (3. Let F be a Hermitian vector bundle over M .33). C). . Compare the associated Dirac operators of E and F with the twist construction in (8.0 (M.odd (M. E).∗ (M. As in the previous example. We note that.27. as a Dirac bundle. Ap.27 Examples.0 (M. ∂= ∂∗ = − ˆ ˆ ZXj ∧ DXj + ZYj ∧ DYj .even (M. · ∈ A0.166 ¨ Lectures on Kahler Manifolds C. Let E be a Dirac bundle over M . µ(Xσ) = −Xµ(σ). .21) and Proposition 1.∗ (M. let 1 ZX := (X − iJX) and ZX := ZX . C) ⊗ E in the sense of Exercise C.10) and (1.3. E) → Ap. E) = Ap. see (1. We note that. . We leave it as an exercise to show that (D+ )∗ = D− . 2) Let M be a complex manifold with complex structure J. . E) ⊕ Ap. E) is the twist of A0. E) into a graded Dirac bundle. satisfies the axioms for Clifford multiplication in (C.∗ (M. as a Dirac bundle.1 (M. E) ⊕ Aodd (M. and d+d∗ is the associated Dirac operator. For X ∈ T M . Then the morphism T M ⊗ Ap.26. Let E be a Dirac bundle and µ : E → E be a parallel unitary involution anticommuting with Clifford multiplication. E) turns Ap. Then E ⊗ F with the induced connection and Clifford multiplication X · (σ ⊗ τ ) := (X · σ) ⊗ τ is also a Dirac bundle. Let E → M be a holomorphic vector bundle with Hermitian metric h and associated Chern connection D.26. 1) The morphism T M ⊗ A∗ (M. By (3.23). Ym ) be a local orthonormal frame of M with JXj = Yj . E). It follows that 2(∂ + ∂ ∗ ) is the Dirac operator associated to the Clifford multiplication in (C. C) with Ap. the splitting Ap. also denoted . Y1 . E) turns A∗ (M. Xm . D restricts to operators D± : E(E ± ) → E(E ).23). ˆ ˆ ZXj DXj + ZYj DYj . Let (X1 .∗ (M. X · α := X ∧ α − X α. A∗ (M.∗ (M.26 Exercise. E) → A∗ (M. √ √ (C. E) = Ap. The splitting A∗ (M. C) with E in the sense of Exercise C.

32) with C l := dom dmax ∩ L2 (Al (M. . we have L2 -orthogonal decompositions L2 (A∗ (M. see e.27. E) = im{dmax : C k−1 → L2 (Ak (M.31 Theorem (Hodge Decomposition).2 L2 -de Rham Cohomology. E)/B k (M. E). E) + im d∗ . it follows that α is smooth and satisfies (d + d∗ )α = 0 in the classical sense. then d∗ α = 0. E). E) is contained in ker dmax since the latter is closed in L2 (A∗ (M. If α is a harmonic form. E))}. The proof of the second max equality is similar.22.1. d = 0 and hence ∗ 2 (C. By (C. max where d∗ max stands for (d )max . E)). ∗ ker d∗ = H2 (M. E) and β ∈ A∗ (M. max max By the first inclusion in (C. E)) = ker dmax + im d∗ = ker d∗ + im d.g.20 and Corollary C.33) (C.red (M. C. E)). dβ)2 = (d∗ α. 30 [Car] and [Lot] contain discussions of L2 -de Rham cohomology. (C. β)2 = c max 0. We define the reduced L2 -de Rham cohomology of M by k H2. E) = Z k (M. Let E → M be a Hermitian vector bundle with a flat Hermitian connection D. hence α is perpendicular to im d. Exercise C.34) The image B k (M. E))}. E). (C. E) := ker dmax ∩ ker d∗ max be the space of square integrable harmonic forms (with values in E). E)). E) := Z k (M.35) We note the following consequences of Exercise C. max ∗ We let H2 (M. im d∗ ⊂ im d∗ ⊂ ker d∗ . E). there is a cochain complex · · · − max C k−1 − max C k − max C k+1 − max · · · −→ − −→ − −→ − −→ − Z k (M. E) = ker{dmax : C k → L2 (Ak+1 (M. max im d ⊂ im dmax ⊂ ker dmax . Then (α. The exterior differential d = dD is a differential operator of order one on A∗ (M. Let α ∈ H2 (M. Since E is flat. E) need not be closed. E)/B k (M. if α ∈ ker dmax is L2 -perpendicular ∗ to im d.29) (C. E). then α satisfies (d + d∗ )α = 0 weakly. We have L2 -orthogonal decompositions ∗ Proof. where we recall that ker dmax and ker d∗ are closed subspaces of L2 (A∗ (M. but its closure B k (M.Appendix C Remarks on Differential Operators 167 C. ∗ ker dmax = H2 (M. E) + im d. Vice versa. B k (M. Since d + d∗ is an elliptic differential operator. We let The L2 -de Rham cohomology30 of M consists of the quotients k H2 (M.30). and hence α ∈ H2 (M. d d d d (C.14).30) It follows that ker dmax contains the closure of im dmax .

g.168 ¨ Lectures on Kahler Manifolds C.40) ∗. Suppose now that M is oriented.38) compare (1. ∗ ∗ im ∂ ∗ ⊂ im ∂max ⊂ ker ∂max .q+1 (M.3 L2 -Dolbeault Cohomology.s := dom ∂max ∩ L2 (Ar.43) ∂ ∂ ∂ ∂ (C. (C. then dα = d∗ α = 0.56). We have L2 -orthogonal decompositions ∗. Since ∗ ⊗ h commutes with the Laplace operator ∆d .q−1 → L2 (Ap. E) be square integrable. E))}.∗ ∗ ker ∂max = H2 (M. C. Exercise C.41 Theorem (Hodge Decomposition). we have L2 -orthogonal decompositions ∗ L2 (A∗. C. endowed with a compatible Riemannian metric g.∗ (M. (C.40). (C. Poincar´ duality holds for reduced L2 -cohomology e if M is oriented. complete.∗ ∗ We let H2 (M. In particular. If M is complete and connected and ∆d α = 0.s (M. E) := ker ∂max ∩ ker ∂max be the space of square integrable ∂harmonic forms (with values in E). Let M be a complex manifold with complex structure J. E)) = ker ∂max + im ∂ ∗ = ker ∂max + im ∂.q → L2 (Ap. then the natural projec∗ ∗ tion H2 (M. E ∗ ). ∗. If α is a ∂-harmonic form.q (M. If M is complete and connected.39) As in the case of the exterior differential d. (C.36 Proposition.q (M.2.14).q+1 −max · · · −→ − −→ − −→ − −→ − with C r.q (M.27. C. Let E → M be a holomorphic vector bundle with Hermitian metric h and associated Chern connection D.q −max C p. E) = im{∂max : C p.red (M. E))}. Let α ∈ A∗ (M.42) . E) + im ∂. E) → H2. Since ∂ + ∂ ∗ is an elliptic differential operator. By (C. see e. E) → H2 (M.q−1 −max C p. E) is an isomorphism. E) + im ∂ ∗ . there are cochain complexes · · · −max C p. We let B p.∗ ker ∂max = H2 (M. By the first inclusion in (C.37 Proposition. E) = ker{∂max : C p. then α satisfies (∂ + ∂ ∗ )α = 0 weakly. and connected. it induces conjugate linear isomorphisms n−r r H2 (M. Z p. E)). it follows that α is smooth and satisfies (∂ + ∂ ∗ )α = 0 in the classical sense. im ∂ ⊂ im ∂max ⊂ ker ∂max .

E) → H2 (M. then ∂α = ∂ ∗ α = 0. E) be square integrable.47 Proposition.q (M.q (M.q−s H2 (M. C.∗ (M. E) → H2.m−q H2 (M. E ∗ ). E) and H2 (M. Suppose that M is a complete and connected K¨hler manifold a k+2s k (M.q p−s.44) The reduced L2 -Dolbeault cohomology of M is given by p. a Then ∆d = 2∆∂ . then the natural projec∗. (C. it induces conjugate linear isomorphisms p. E)/B p.46 Proposition.11). Moreover.q (M. E) = ⊕p+q=r H2 (M. E) → H2 injective for 0 ≤ s ≤ m − k and surjective for s ≥ m − k ≥ 0. C.q (M. E). Furthermore.49 Theorem.q H2. . p. C. p. (C. E). E).q r the Hodge and Lefschetz decompositions of H2 (M.∗ ∗.red (M. Suppose now in addition that M is a K¨hler manifold and that E is flat. E). E) := Z p. If M is complete and connected and ∆∂ α = 0.red (M.q H2 (M. E) is and that E is flat.∗ tion H2 (M. Then the Lefschetz map Ls : H2 (M. Since ∗ ⊗ h commutes with the Laplace operator ∆∂ . E) = ⊕s≥0 Ls H2. E) is an isomorphism.P (M. In particular. the Lefschetz map L commutes with ∆∂ .q m−p. (C. E). E)/B p.q r H2 (M. p. If M is complete and connected. Serre duality holds for reduced L2 -Dolbeault cohomology if M is complete and connected.22. Let α ∈ A∗. E) := Z p.20 and Corollary C.Appendix C Remarks on Differential Operators 169 The L2 -Dolbeault cohomology of M consists of the quotients p.48) compare (3.45) We again note consequences of Exercise C.

Weinstein. Le spectre d’une vari´t´ riemanniee enne. Math. Aubin. & E. New York–Heidelberg. R´duction du cas positif de l’´quation de Monge-Amp`re sur e e e les vari´t´s k¨hl´riennes compactes a la d´monstration d’une in´galit´. Aubin. Providence. [Ar] [At] [Au1] [Au2] [Au3] [Au4] [BT] [Bea] [BeG] [BGM] [Bes] [Bi] [Boc] [BM] . V. Gordon. On the formality problem for symplectic manifolds. 1974). Elliptic operators. Math. 48 (1947). Gauduchon. 194. Montgomery. P. I. 57 (1984). Chicago Lectures in Mathematics. Wiss. Differential Geom. Arnold. Ergeb. J. A119– e A121. Ast´risque 32–33 (1976). R. University of Chicago Press. 252. 1–9. Colloque Analyse et Topologie en l’Honneur de Henri Cartan (Orsay. 938. Expos´ No.. Acad. Math. 2001. Groups on analytic manifolds. 2000). 383–424. Springer-Verlag. Aubin. New York. ´ T. A. Vari´t´s K¨hleriennes dont la premi`re classe de Chern est ee a e nulle. M. Berlin–New York. K. Einstein manifolds. 288. 776–797. Monge-Amp`re equations. e e e S. ee a e ` e e e Funct. 659–669. T. J. Chicago. Topology 27 (1988). Soc. Nonlinear analysis on manifolds. S. of Math. R. Bull. e T. Aubin. A. Vogtmann and A. Vector fields and Ricci curvature. Amer. 43–72. J. Berlin. Contemp. 18 (1983). Besse. 1982. Paris S´r. Bochner & D. 1987. Math. e Grundlehren Math. edited by Zafer Mahmud and Mamoru Mimura. F. Biquard. Babenko & I. no. T. 52 (1946). Beauville. e a e ` S´minaire Bourbaki. Texts in Math. Ann. Grenzgeb. Global differential geometry. O. Berger. K¨hler and symplectic structures on nilmania folds. Lectures on exceptional Lie groups. C. the mathematical legacy of Alfred Gray (Bilbao. Soc. 60. Amer. 1978. Mazet. Springer-Verlag. Bochner. S. M. Mathematical methods of classical mechanics. 3. Anal. L. Aiii. Atiyah. Lecture Notes in Math. 143–153. Differential Geometry e 4 (1970). 1971. Taimanov. Equations du type Monge-Amp`re sur les vari´t´s k¨hleriennes e ee a compactes. IL. A. 10. Translated from the Russian by K. Peter May. I.I. Grad.Literature [Ad] J. 513–518. M´triques k¨hl´riennes a courbure scalaire constante. Sci. With a foreword by J. Adams. A-B 283 (1976). to be published in Ast´risque. M´triques riemanniennes et courbure. F. 1996. discrete groups and von Neumann algebras. SpringerVerlag. Benson & C. C.. 755–782 (1984). Springer-Verlag.

J. of Math. Lefschetz. Period mappings and period dou mains. Ann. Rend. Texts in Math. Soci´t´ Math´matique de France. o E. M. Manifolds and gea ometry (Pisa. 483– 491. M´triques d’Einstein-K¨hler sur les vari´t´s de Fano: e a ee obstructions et existence (d’apr`s Y. S. Cheeger. Borel. Math. 2004. 61–85. Calabi & B. S. 861.J. Calabi. Math. Improper affine hyperspheres of convex type and a generalization of a theorem by K. 16.. Lectures presented at a S´minaire held at e ´ the Centre de Math´matiques de l’Ecole Polytechnique. The space of K¨hler metrics. On K¨hler manifolds with vanishing canonical class. New York. ETH Z¨rich. E. e A. Vol. complex manifolds which are not algebraic. 2. Sympos. 58 (1953). 225. Burger. Peters.Literature [Bor] [Bo1] 171 A. A. N. e Expos´ No. Semisimple groups and Riemannian symmetric spaces. Matsushima. Appl. Mat. Preprint. Differential Geometry 8 (1973). T. Michigan Math. 206– 207. e e D. S´minaire Bourbaki. 105–126. (7) 21 (2001). Bryant. Tian). 1978. Recent advances in the theory of holonomy. 830. J¨rgens. Erven P. J. Proceedings of the International a Congress of Mathematicians (Amsterdam. 319 (2001). Carron. 2003. 87–119. North-Holland Publishing Co. 2005. Hermitian symmetric spaces and K¨hler rigidity. Palaiseau. Groningen. March– e June 1978. Bump. Carlson. Formes harmoniques L2 sur les vari´t´s non-compactes. Eugenio Calabi and K¨hler metrics. ee G. P. Texts Read. Cao & F. a u E. E. Xavier. 5 (1958). S´minaire Bourbaki. J. Princeton. R. 78–89. Math. A class of compact. J. 1993). Paris. Princeton University Press. Some examples of manifolds of nonnegative curvature. P. Ast´risque 58. e ee e 169 pp. Cambridge University Press. Wienhard. M¨ller-Stach & C. Math. A symposium in honor of S.. A. New Delhi.. 1954). Futaki. Premi`re classe de Chern et courbure de Ricci: e preuve de la conjecture de Calabi. J. Noordhoff N. Lie groups. Ann. J. Bourguignon et al.V. Amsterdam 1957. Cambridge Stud. J. Cambridge. 1998. Grad.. Nadel et G. Hindustan Book Agency. Adv. Cambridge. 1996. P. Iozzi & A. Expos´ No. XXXVI. 85. [Bo2] [Bo3] [Br] [Bu] [BIW] [Ca1] [Ca2] [Ca3] [CE] [CX] [CMP] [Car] [Che] . Springer-Verlag. 623–628. 1957. K¨hler parabolicity and the Euler number of compact a manifolds of non-positive sectional curvature. Cambridge University Press. J. Calabi. Eckmann. Algebraic a geometry and topology. Bourguignon. Ast´risque 245 e e e (1997). Ast´risque 266 (2000). 277–305. 351–374. Yau. Calabi. 494–500. Bourguignon.

Abh. On curvature and characteristic classes of a Riemann manifold. Futaki. K¨hler hyperbolicity and L2 -Hodge theory. I. Universitext. J. North-Holland Publishing Co. Debarre. Cheeger & M. Duistermaat. 1–23. J. Springer-Verlag. Gromoll. 1991. Vari´t´s de Fano. Dedicata 5 (1976). ee e e Ast´risque 245 (1997). American Elsevier Publishing Co. Op´rateur de courbure et laplacien des formes e diff´rentielles d’une vari´t´ riemannienne. Ann. M. C. Birkh¨user. Differential Geometry 6 (1971/72). A. New York. Geometry of nonpositively curved manifolds. Wiley-Interscience. Principles of algebraic geometry. J. Deligne. Lecture Notes in Math. J. MA. 157–159. Chernoff. J. J. Texts in Math. Lie Groups. Differential Geom. North-Holland Mathematical Library 9.. Meyer. A. Harris. [FH] [Fu] [GM] [GH] [Gr] . Gallot & D. Inc. Pure Appl. e [CG1] [CG2] [CGr] [Chr] [Chf] [De] [DGMS] P.172 [CEb] Literature J. 827. 197–221. Eberlein. Representation theory. Math. Gromoll. M. Ebin. Cheeger & D. I. Sullivan. of Math. 437–443. 309–346. a Boston. J. 413–443. Chicago. G. Differential geometry a (Pe˜´ nıscola. Math. Math. R. 129. Ina vent. J.. e ee 259–284. University of Chicago Press. Essential self-adjointness of powers of generators of hyperbolic equations. Gray. Functional Analysis 12 (1973). Berlin. Morgan & D. An obstruction to the existence of K¨hler-Einstein metrics. Griffiths. Collapsing Riemannian manifolds while keeping their curvature bounded. P. a [Du] [DK] [Eb] [Eh] [FG] J. Chern. 263–292. Gromov. 20 (1955). P. Geom. Duistermaat and J. Kolk. Progr. Fulton & J. The splitting theorem for manifolds of nonnegative Ricci curvature. Math. SpringerVerlag. 1978. 73 (1983). IL. W. Springer-Verlag. Differential a Geom. 33 (1991). Real homotopy theory of K¨hler manifolds. Math. Amsterdam–Oxford. Pures Appl. The Iwasawa manifold. Comparison theorems in Riemannian geometry. P. 29 (1975).. Invent. 1975. 23 (1986). S´minaire Bourbaki. O. 54 (1975). S. 245–274. Fern´ndez & A. New York. Univ. 2000. 1985). Cheeger & D. 96 (1972). 1986. J. J.. 1996.-S. Chicago Lectures in Mathematics. A first course. P. 117–126 . Math. 1996. Metric deformations of curvature. J. P. Griffiths & J. 1209. Local convex deformations. Fourier integral operators. Semin. Harris. Expos´ No. Gromov. Ehrlich. 130. 119–128. Hamb. Berlin. 401–414. S. Cheeger & D. New York. Grad. On the structure of complete manifolds of nonnegative curvature. J.

corr. Springer-Verlag. Geom. Academic Press. Karcher. 83–196. Amer. Schwarzenberger. Univ. Helgason. DC. Hitchin. K. Berlin. E. F. Abh. Perturbation theory for linear operators. Borel. 65–71. Berlin. The Ricci flow on K¨hler manifolds. I”. Math. Berlin. 1–30. Pa. Hirzebruch. Mathematische Werke. 2000. XII. Kanˆ Memorial Lectures 5. University of Pennsylvania.. “On the Ricci curvature of a compact K¨hler manifold and the complex Monge-Amp`re a e equation.. Walter de Gruyter. Hein. Math. Kazdan. Proc. 2004. Rheinische a Friedrich-Wilhelms-Universit¨t Bonn. Providence. 74 (1961). N.. Philadelphia. Positive scalar curvature and the Dirac ´ operator on complete Riemannian manifolds. Complex geometry. On the curvature of rational surfaces. Diplomarbeit. N. Proc. Edited by Rolf a Berndt and Oswald Riemenschneider. Math. 1962. Riemannian comparison constructions. Yau. 9 (1932). Monographs. Assoc. Part 2. Hasegawa.I. Publ. New York–London. Joyce. L. o Princeton. D. Kato. Inst. Appendix one by R. 1987. Comm. S. Differential geometry (Stanford Univ. Calif. Reprint of the 1980 edition. Differential Geometry and Symmetric Spaces. Kobayashi. Math. 2006. J. Princeton University Press. H. J. R. a Semin. Global differential geometry. [Ha] [Hei] [Hel] [Hir] [Hit] [Hu] [JZ] [Jo] [K¨1] a [K¨2] a [Kar] [Kat] [Ka1] [Ka2] [Ko1] [Ko2] . Sympos.J. Vanishing theorems for L2 -cohomology on infinite coverings of compact K¨hler manifolds and applications in algebraic geometry. Soc. Oxford Math. Topological methods in algebraic geometry.. Publ. Math. T. Gromov & H. 570–574. Lecture Notes. D. 170–222. Jost & K. 1978. 27. appendix two by A. Huybrechts. 1973). D. 1995. Math. of the 3rd ed. 8 (2000). K¨hler. K¨hler. Amer. Japan 15.-J. Differential geometry of complex vector bundles. XXVII.. Classics Math. S. T. 2005. Springer-Verlag. On compact K¨hler manifolds with positive definite Ricci a tensor.. 1995. ¨ E. Berlin. Hautes Etudes Sci. Uber eine bemerkenswerte Hermitesche Metrik. Washington.. Math. Lawson Jr. Oxford. 31 (1978). H. of Math. Classics Math. Soc. Minimal models of nilmanifolds. 413–414. MAA Stud. Stanford. Universitext. Iwanami Shoten. An introduction. E. Kazdan. Tokyo. A remark on the preceding paper of S. Pure Math. 173-186. Pure Appl. a S. reprint of the 2nd.. Bonn. Compact manifolds with special holonomy. America. Hamb. Anal. Applications of Partial Differential Equations to Problems in Geometry. Pure Appl. Math. a Comm. Kobayashi. SpringerVerlag.Literature [GL] 173 M. print. 106 (1989). Mathematical works. Soc. Math. Ann.B. 1975. Zuo. Oxford University Press. J. 58 (1983). 2003. 65–80. 1989.

Geom. ee a e Y. 726– ee 729. Princeton University Press.I. Transformation infinit´simales conformes de certaines e vari´t´s riemanniennes compactes. Newlander & L. A. 11 (1957). Progr. Ann. Kobayashi & K. University of Tokyo Press. 1989. Harlow. Michelsohn. Sur la structure du groupe d’hom´omorphismes analytiques e d’une certaine vari´t´ k¨hl´rienne. Soc. R. Lohkamp. Stud. II. Paris 244 (1957). A. a e A. R.J.. Foundations of differential geometry. Acad. Shiffman & A. 179. New York. 457–498. Complex analytic coordinates in almost complex manifolds. Louis. K¨hler geometry.. N. Nadel. C. Inhomogeneous Einstein metrics on complex line bundles. of Math.-L. 81–119. Nagoya Math. Soc. J. 265–288. MA. Certain conditions for a Riemannian manifold to be isometric with a sphere.. 132 (1990). 213–225. Classical Quantum Grav. Matsushima. B. N. 145–150. Symmetric spaces (Short Courses. Longman S&T.J. topology and asymptotic methods. 255–270. Lott. Spin geometry. Wiley Classics Lib. Washington Univ. Nirenberg. John Wiley & Sons. T. Princeton Math. 549–596. Sci. A. L2 -cohomology of geometrically infinite hyperbolic 3-manifolds. Princeton. C. Translated from the 1992 Japanese original by the author. D. Birkh¨user.. reprint of the 1963 original. G´om´trie des groupes de transformations. Y. 7 (1997). Pitman Res.. Tokyo. On tube domains. Kor´nyi & J. Anal. Ser. 3011–3013. R. Moroianu. Vol. J. Ann. 1958. Ann. Math. Funct. 149. A. Stasheff. Math. Amer.. Japan 14 (1962). Math. 1985. 4 (1987). 1988. J. Dunod. Lichnerowicz. Elliptic operators. Acad. Providence. 81 (1965). Page & C. 391–404. B. Realization of hermitian symmetric spaces as a generalized half-planes. 1969–1970). Mo. New York. Multiplier ideal sheaves and K¨hler-Einstein metrics of posia tive scalar curvature. of Math. Vanishing theorems on complex manifolds. Pure and Appl. 76. N. Inc. Princeton. N. Sur les transformations analytiques des vari´t´s ee k¨hl´riennes compactes. Travaux et e e Recherches Math´matiques III. Roe. Monogr. Ann. Obata. 1996. A. Math. e J. Sommese. M. 65 (1957). Characteristic classes. a [KW] [LM] [Li1] [Li2] [Li3] [Loh] [Lot] [Ma1] [Ma2] [MS] [Mo] [Na] [NN] [Ob] [PP] [Ro] [Sa] [ShS] . Notes Math. A. 8. Boston. Pope. of Math. 333–340. Ser. Matsushima. 1996. of Math. St. J.. Math. Curvature h-principles. a A. Sakai. Riemannian geometry.174 [KN] Literature S. Lawson & M. 56. H. 38. Paris. Lichnerowicz. 142 (1995). Ann. Paris 241 (1955).. Lichnerowicz. Wolf. 1974. M. Nomizu. J. Dekker. Sci. of Math. Transl. Princeton University Press. Lecture Notes 2004. 1972. Milnor & J.

Wells. G. 585–627. Spaces of constant curvature. Prentice-Hall. Leca ture Notes in Math. Houston. Tian. Complex algebraic surfaces. 2000. 127 (1988). 2004. Invent. R. Yau. Stanford University. MA. Math. Glenview.I. 213–239. K. The existence of K¨hler-Einstein metrics on manifolds with posa itive anticanonical line bundle and a suitable finite symmetry group. On Calabi’s conjecture for complex surfaces with positive first Chern class. Englewood Cliffs. Foresman and Co. Fourth edition. Canonical metrics in K¨hler geometry. Ill. de l’Universit´ de Nancago. Nat. O. Ann.Literature [Si1] [Si2] 175 Y. 1987. P. Introduction ` l’´tude des vari´t´s k¨hl´riennes. 38–45. Vakil.J. Boston. R. 1661. S. N. Tian. Phys. Universit¨t Bonn.. Math. Tralle & J. Lectures Math. K¨hler-Einstein metrics on complex surfaces with a C1 > 0. Scott. T. T. Yano. Bounded cohomology and geometry. 89 (1987). Invent.. On the curvature of compact Hermitian manifolds. Basel. Tian. 112 (1987). F. Yau. Ann. Invent. Yau.A. 1971. A. Proc. Symplectic manifolds with no K¨hler structure. Publications de a e ee a e l’Institut de Math. Berlin. 2002. I. Actualit´s Sci. Birkh¨user Verlag.. Publish or Perish. Lectures on Hermitian-Einstein metrics for stable bundles and K¨hler-Einstein metrics. a a Y. 101 (1990). Math. F. Wolf. Siu. 2000. Dissertation. Math. Preprint. of Math. Birkh¨user Verlag. Basel. Hermann. Comm. 101–172. G. S. u a G. 18. 1997. Acad. Pure Appl. T. U. ETH Z¨rich. International Press. 25 (1974). Warner. Siu. 55 (1952). A. Calabi’s conjecture and some new results in algebraic geometry. AMS/IP Stud. e 339–411. no. Bonn. 31 (1978). Rheinische Friedrich-Wilhelms-Universit¨t Bonn. On the Ricci curvature of a compact K¨hler manifold and the a complex Monge-Amp`re equation. arXiv:math. e e 1267. Foundations of differentiable manifolds and Lie groups. R. T. TX. Thomas. Notes on GIT and symplectic reduction for bundles and varieties.. Mathematisches Institut. Complex differential geometry. Soc. W. Ind. DMV Seminar 8. Lecture Notes. Weil. VI. A. Paris 1958. Springer-Verlag. Amer. T. 1798–1799. T. a J. Inc. K. Oprea. Tian & S. Wienhard. G. On harmonic and Killing vector fields. Notes taken by Meike a Akveld. 74 (1977). Math. A. 175–203. Bonner Mathematisa che Schriften 368. Inc. 1977.-London.. Math. Differential analysis on complex manifolds. Adv. of Math.S. Math. Bonn 2004. 1973. Zheng. Providence. Prentice-Hall Series in Modern Analysis. [Th] [Ti1] [Ti2] [Ti3] [TY] [TO] [Va] [Wa] [Wei] [Wel] [Wi] [Wo] [Ya] [Ya1] [Ya2] [Ya3] [Zh] . Comm. On K¨hler-Einstein metrics on certain K¨hler manifolds with a a C1 (M ) > 0. Yau. Sci. R.. S.AG/0512411. 225–246.

67. 18. 134 class. 45. 147 Dirac bundle. 101. 6. 79 de Rham decomposition. 100 fundamental matrix. 21. 156–161 . 99. 40. 117. 23. 168 Heisenberg group. 58. 35. 77 Hermitian manifold. 19 curvature operator. 43. 101. 149–157 harmonic. 20. 118 arithmetic genus. 76 adjoint. 43. 49 ample. 29 operator. 165 Dolbeault cohomology. 11 differential. 39. 39. 83. 57. 22. 151. 60. 42 Futaki invariant. 48. 160 Betti number. 163 essentially self-adjoint. 25. 165 relation. 108 form. 85 automorphic. 42 complex curve. 22. 89 torus. 102 Grassmann manifold. 91. 8. 104 Bergmann metric. 160 surface. 104 formal. 31–33. 13.Index almost complex structure. 88. 104 Calabi–Yau manifold. 28. 28. 83 canonical bundle. 68 bisectional curvature. 157 surface. 76. 135 exterior derivative. 85. 114. 69. 43. 160 effective. 100 automorphism. 11 projective space. 116 dual. 47. 131 Clifford multiplication. 82. 69 o Fubini–Study metric. 145. 17. 24 compatible. 108. 88. 74. 133. 46. 58. 162 Fr¨licher relations. 22. 11. 131 connection. 146 ddc -Lemma. 85. 12. 114. 73 Hessian. 7. 104 hyperbolic space. 15 manifold. 167. 78 Chern character. 123. 148. 157 Lie group. 23 symmetric space. 55 blow up. 135 form. 15 vector field. 130. 57. 159 elliptic. 89. 59. 17. 4 Fano manifold. 18. 82. 164 Euler class. 11. 130. 158 structure. 108. 79. 123 Calabi conjecture. 157 quadric. 165 operator. 20. 5 formal adjoint. 3 collapse. 27. 99. 88 Calabi–Futaki invariant.

87. 11 form. 57. 68. 128 transvection. 108 holonomy. 50. 86 Lefschetz decomposition. 42. 147 Hopf manifold. 107. 145 Kodaira embedding theorem. 9 Hodge–Riemann bilinear relations. 71 symmetric pair. 39. 36. 83. 152 Laplace operator. 38. 37. 80 Kummer surface. 118 vanishing theorem. 70. 64. 125 Iwasawa manifold. 15. 163 minimal. 58 a hyperplane bundle. 17. 67. 164 Monge–Amp`re equation. 84 line bundle. 51 . 129 Weitzenb¨ck formula. 50. 101 Ricci equation. 133 transgression. 69 vector space. 71 principal symbol. 22. 6. 78. 68. 169 map. 77 K¨hler a form. 63. 90. 139 tube domain. 80. 50. 6. 42. 56 Ricci-flat. 87 form. 117.INDEX 177 Hodge decomposition. 14. 60. 62. 29 operator. 163 reductive. 69 hyper-K¨hler manifold. 14 wedge product. 31. 75. 3 Weil homomorphism. 100. 72 Picard group. 82. 151 manifold. 88 Riemann sphere. 48 K¨hler–Einstein a manifold. 26 Lagrangian subspace. 78 primitive. 68. 22. 101 metric. 57 metric. 42. 29. 7. 17. 65 maximal extension. 57. 69 index theorem. 70. 19 normal coordinates. 151 tautological bundle. 59. 107 a Killing field. 79–80. 29 of symmetric space. 117 number. 49 manifold. 79. 100. 36 o Wirtinger inequality. 140 universal bundle. 48 period map. 54 vector bundle. 33. 100 form. 102 K¨hler-hyperbolic. 70–72 holomorphic. 12 Serre duality. 2. 51 extension. 38 positive. 62. 117 invariant. 71 manifold. 21. 118 sectional curvature. 65–69. 7. 91. 58. 137 symplectic form. 160 type of differential form. 167–169 diamond. 67. 99 e Nijenhuis tensor. 169 signature. 30. 48 potential. 141 space. 99.