Cattani Kahler Manifolds

2150-4
Summer School and Conference on Hodge Theory and Related Topics
14 June - 2 July, 2010
Introduction to Kähler manifolds
E. Cattani
University of Massachussetts at Amherst
Amherst MA
USA
INTRODUCTION TO KÄHLER MANIFOLDS
SUMMER SCHOOL ON HODGE THEORY
ICTP, JUNE 2010
EDUARDO CATTANI
Preliminary Version
Introduction
These notes are intended to accompany the course Introduction to Complex, Her-
mitian and Kähler Manifolds at the ICTP Summer School on Hodge Theory. Their
purpose is to review the concepts and results in the theory of Kähler manifolds
that both motivate and are at the center of Hodge Theory. Although we have tried
to clearly define the main objects of study we have often referred to the literature
for proofs of the main results. We are fortunate in that there are several excellent
books on this subject and we have freely drawn from them in the preparation of
these notes, which make no claim of originality. The classical references remain the
pioneer books by A. Weil [38], S. S. Chern [10], J. Morrow and K. Kodaira [22, 25],
R. O. Wells [39], S. Kobayashi[21], and P. Griffiths and J. Harris [15]. In these notes
we refer most often to two superb recent additions to the literature: the two-volume
work by C. Voisin [34, 35] and D. Huybrechts’ book [18].
We assume from the outset that the reader is familiar with the basic theory of
smooth manifolds at the level of [2], [23], or [33]. Some of this material as well as
the basics of cohomology theory will be reviewed in the course by Loring Tu. The
book by R. Bott and L. Tu [3] is an excellent introduction to the algebraic topology
of smooth manifolds.
These notes consist of five sections which correspond, roughly, to the five lectures
in the course. There is also an Appendix which collects some results on the linear
algebra of complex vector spaces, Hodge structures, nilpotent linear transformations,
and representations of sl(2, C). There are many exercises interspersed throughout
the text, many of which ask the reader to prove, or complete the proof, of some
result in the notes.
A final version of these notes will be posted after the conclusion of the Summer
School.
1
2 EDUARDO CATTANI
Contents
1. Complex Manifolds 3
1.1. Definition and Examples 3
1.2. Holomorphic Vector Bundles 11
2. Differential Forms on Complex Manifolds 14
2.1. Almost Complex Manifolds 14
2.2. Tangent and Cotangent space 16
2.3. de Rham and Dolbeault Cohomologies 19
3. Hermitian and Kähler metrics 22
3.1. Kähler Manifolds 23
3.2. The Chern Class of a Holomorphic Line Bundle 26
4. Harmonic Forms - Hodge Theorem 28
4.1. Compact Real Manifolds 28
4.2. Compact Hermitian Manifolds 32
5. The Hodge Decompositon Theorem 33
5.1. Kähler identities 34
5.2. Lefschetz Theorems 36
5.3. Hodge-Riemann Bilinear Relations 38
Appendix A. Linear Algebra 41
A.1. Real and Complex Vector Spaces 41
A.2. Hodge structures 45
A.3. Symmetric and Hermitian Forms 47
A.4. Polarized Hodge Structures 48
A.5. The Weight filtration of a nilpotent transformation. 49
A.6. Representations of sl(2, C) 50
A.7. Lefschetz decomposition 52
References 53
KÄHLER MANIFOLDS 3
1. Complex Manifolds
1.1. Definition and Examples. Let U ⊂ Cn be an open subset and f : U → C.
We say that f is holomorphic if and only if it is holomorphic as a function of each
variable separately; i.e. if we fix z = a , = j then the function f (a1 , . . . , zj , . . . , an )
is a holomorphic function of zj . A map F = (f1 , . . . , fn ) : U → Cn is said to be
holomorphic if each component fk = fk (z1 , . . . , zn ) is holomorphic. If we identify
Cn ∼ = R2n , and set zj = xj + iyj , fk = uk + ivk , j, k = 1, . . . , n, then the functions
uk , vk are C ∞ functions of the variables x1 , y1 , . . . , xn , yn and satisfy the Cauchy-
Riemann equations:
∂uk ∂vk ∂uk ∂vk
(1.1) = ; = −
∂xj ∂yj ∂yj ∂xj
Conversely, if (u1 , v1 , . . . , un , vn ) : R2n → R2n is a C ∞ map satisfying the Cauchy-
Riemann equations (1.1) then the map (u1 + iv1 , . . . , un + ivn ) is holomorphic. In
other words a C ∞ map F : U ⊂ R2n → R2n defines a holomorphic map Cn → Cn if
and only if the differential of F , written in terms of the basis
(1.2) {∂/∂x1 , . . . , ∂/∂xn , ∂/∂y1 , . . . , ∂/∂yn }
of the tangent space Tp (R2n ) and the basis {∂/∂u1 , . . . , ∂/∂un , ∂/∂v1 , . . . , ∂/∂vn } of
TF (p) (R2n ) is of the form:

A −B
(1.3) DF (p) =
B A
for all p ∈ U . Thus, it follows from Exercise 36 in Appendix A.1 that F is holomor-
phic if and only if DF (p) defines a C-linear map Cn → Cn .
Exercise 1. Prove that a (2n) × (2n)-matrix is of the form (1.3) if and only if it
commutes with the matrix J:

0 −In
(1.4) J := ,
In 0
Definition 1.1. A complex structure on a topological manifold M consists of a
collection of coordinate charts (Uα , φα ) such that:
i) The sets Uα are an open covering of M .
ii) φα : Uα → Cn is a homeomorphism of Uα onto an open subset of Cn for some
fixed n. We call n the complex dimension of M .
iii) If Uα ∩ Uβ = ∅, the map
(1.5) φβ ◦ φ−1
α : φα (Uα ∩ Uβ ) → φβ (Uα ∩ Uβ )
is holomorphic.
Example 1.2. The most basic example of a complex manifold is Cn or any open
subset of Cn . For any p ∈ Cn , the tangent space Tp (Cn ) ∼
= R2n which is identified,
in the natural way, with C itself.
n
Example 1.3. Since GL(n, C), the set of non-singular n × n matrices with complex
2
coefficients, is an open set in Cn , we may view GL(n, C) as a complex manifold.
4 EDUARDO CATTANI
Example 1.4. The basic example of a compact complex manifold is complex pro-
jective space which we will simply denote by Pn . Recall that:

Pn := Cn+1 \ {0} / ∼ ,
where ∼ is the equivalence relation: z ∼ z if and only if there exists λ ∈ C∗ such that
z = λz; z, z ∈ Cn+1 \ {0}. We denote the equivalence class of a point z ∈ Cn+1 \ {0}
by [z]. The sets
(1.6) Ui := {[z] ∈ Pn : zi = 0}
are open and the maps

z0 zi−1 zi+1 zn
(1.7) φi : Ui → C ;
n
φi ([z]) = ,..., , ,...,
zi zi zi zi
define local coordinates such that the maps
(1.8) φi ◦ φ−1
j : φj (Ui ∩ Uj ) → φi (Ui ∩ Uj )
are holomorphic.
In particular, if n = 1, P1 is covered by two coordinate neighborhoods (U0 , φ0 ),
(U1 , φ1 ) with φi (Ui ) = C. The coordinate changes are given by the maps
φ1 ◦ φ−1 ∗
0 : C →C :
∗
φ1 ◦ φ−1
0 (z) = φ1 ([1, z]) = 1/z.
Thus, this is the usual presentation of the sphere S 2 as the Riemann sphere. We
usually identify U0 ∼
= C and denote the point [0, 1] = ∞.
Exercise 2. Verify that the map (1.8) is holomorphic.
Example 1.5. To each point [z] ∈ Pn we may associate the line spanned by z in
Cn+1 ; hence, we may regard Pn as the space of lines in Cn+1 . This construction
may then be generalized by considering k-dimensional subspaces in Cn . This is the
so-called Grassmann manifold G(k, n). To define the complex manifold structure on
G(k, n), we consider first of all the open set in Cnk :
V (k, n) = {W ∈ M(n × k, C) : rank(W ) = k}.
The Grassmann manifold G(k, n) may then be viewed as the quotient space:
G(k, n) := V (k, n)/ ∼ ,
where W ∼ W if and only if there exists M ∈ GL(k, C) such that W = W · M ;
equivalently, W ∼ W if and only if the column vectors of W and W span the same
k-dimensional subspace Ω ⊂ Cn .
Given an index set I = {1 ≤ i1 < · · · < ik ≤ n} and W ∈ V (k, n), we consider
the k × k matrix WI consisting of the I-rows of W and note that if W ∼ W then,
for every index set I, det(WI ) = 0 if and only if det(WI ) = 0. We then define:
UI := {[W ] ∈ G(k, n) : det(WI ) = 0}
This is clearly an open set in G(k, n) and the map:
φI : UI → C(n−k)k ; φI ([W ]) = WI c · WI−1 ,
where I c denotes the (n − k)-tuple of indices complementary to I. The maps φI

define coordinates in UI and one can easily verify that given index sets I and J, the
maps:
(1.9) φI ◦ φ−1
J : φJ (UI ∩ UJ ) → φI (UI ∩ UJ )
are holomorphic.
Exercise 3. Verify that the map (1.9) is holomorphic.
Exercise 4. Prove that both Pn and G(k, n) are compact.
The notion of a holomorphic map between complex manifolds is defined in a way
completely analogous to that of a smooth map between C ∞ manifolds; i.e. if M and
N are complex manifolds of dimension m and n respectively, a map F : M → N is
said to be holomorphic if for each p ∈ M there exists local coordinate systems (U, φ),
(V, ψ) around p and q = F (p), respectively, such that F (U ) ⊂ V and the map
ψ ◦ F ◦ φ−1 : φ(U ) ⊂ Cm → ψ(V ) ⊂ Cn
is holomorphic. Given an open set U ⊂ M we will denote by O(U ) the ring of
holomorphic functions f : U → C and by O∗ (U ) the nowhere zero holomorphic
functions on U . A map between complex manifolds is said to be biholomorphic if it
is holomorphic and has a holomorphic inverse.
The following result shows a striking difference between C ∞ and complex mani-
folds:
Theorem 1.6. If M is a compact, connected, complex manifold and f : M → C is
holomorphic, then f is constant.
Proof. The proof uses the fact that the Maximum Principle† holds for holomorphic
functions of several complex variables (cf. [34, Theorem 1.21]) as well as the Principle
of Analytic Continuation‡ [34, Theorem 1.22].
Given a holomorphic map F = (f1 , . . . , fn ) : U ⊂ Cn → Cn and p ∈ U , we can
associate to F the C-linear map

∂fi
DF (p) : C → C ; DF (p)(v) =
n n
(p) · v,
∂zj
where v = (v1 , . . . , vn )T ∈ Cn . The Cauchy-Riemann equations imply that if we
regard F as a smooth map F̃ : U ⊂ R2n → R2n then the matrix of the differential
DF̃ (p) : R2n → R2n is of the form (1.3) and, clearly DF (p) is non-singular if and
only if DF̃ (p) is non-singular. In that case, by the Inverse Function Theorem, F̃
has a local inverse G̃ whose differential is given by (DF̃ (p))−1 . By Exercise 1, the
inverse of a non-singular matrix of the form (1.3) is of the same form. Hence, it
follows that G̃ is holomorphic and, consequently, F has a local holomorphic inverse.
We then get:
†
If f ∈ O(U ), where U ⊂ Cn is open, has a local maximum at p ∈ U , then f is constant in a
neighborhood of p
‡
If U ⊂ Cn is a connected open subset and f ∈ O(U ) is constant on an open subset V ⊂ U , then
f is constant on U .
6 EDUARDO CATTANI
Theorem 1.7 (Holomorphic Inverse Function Theorem). Let F : U → V be

a holomorphic map between open subsets U, V ⊂ Cn . If DF (p) is non singular for
p ∈ U then there exists open sets U , V , such that p ∈ U ⊂ U and F (p) ∈ V ⊂ V
and such that F : U → V is a biholomorphic map.
The fact that we have a holomorphic version of the Inverse Function Theorem
means that we may also extend the Implicit Function Theorem or, more generally,
the Rank Theorem:
Theorem 1.8 (Rank Theorem). Let F : U → V be a holomorphic map between
open subsets U ⊂ Cn and V ⊂ Cm . If DF (q) has rank k for all q ∈ U then,
given p ∈ U there exists open sets U , V , such that p ∈ U ⊂ U , F (p) ∈ V ⊂ V ,
F (U ) ⊂ V , and biholomorphic maps φ : U → A, ψ : V → B, where A and B are
open sets of the origin in Cn and Cm , respectively, so that the composition
ψ ◦ F ◦ φ−1 : A → B
is the map (z1 , . . . , zn ) ∈ A → (z1 , . . . , zk , 0, . . . , 0).
Proof. We refer to [2, Theorem 7.1] or [33] for a proof in the C ∞ case which can
easily be generalized to the holomorphic case.
Given a holomorphic map F : M → N between complex manifolds and p ∈ M ,
we may define the rank of F at p as
(1.10) rankp (F ) := rank(D(ψ ◦ F ◦ φ−1 )(φ(p))) ,
for any local-coordinates expression of F around p.
Exercise 5. Prove that rankp (F ) is well-defined by (1.10); i.e. it’s independent of
the choices of local coordinates.
We then have the following consequence of the Rank Theorem:
Theorem 1.9. Let F : M → N be a holomorphic map, let q ∈ F (M ) and let
X = F −1 (q). Suppose rankx (F ) = k for all x in an open set U containing X. Then,
X is a complex manifold and
codim(X) := dim M − dim X = k .
Proof. The Rank Theorem implies that given p ∈ X there exist local coordinates
(U, φ) and (V, ψ) around p and q, respectively such that ψ(q) = 0 and
ψ ◦ F ◦ φ−1 (z1 , . . . , zn ) = (z1 , . . . , zk , 0, . . . , 0).
Hence
φ(U ∩ X) = {z ∈ φ(U ) : z1 = · · · = zk = 0}.
Hence (U ∩ X, p ◦ φ), where p denotes the projection onto the last n − k coordinates
in Cn defines local coordinates on X. It is easy to check that these coordinates are
holomorphically compatible.
Definition 1.10. We will say that N ⊂ M is a complex submanifold if we may cover
M with coordinate patches (Uα , φα ) such that
φα (X ∩ Uα ) = {z ∈ φα (U ) : z1 = · · · = zk = 0}.
for some fixed k. In this case, as we saw above, N has the structure of an (n − k)-
dimensional complex manifold.
Proposition 1.11. There are no compact complex submanifolds of Cn of dimension
greater than zero.
Proof. Suppose M ⊂ Cn is a submanifold. Then, each of the coordinate functions
zi restricts to a holomorphic function on M . But, if M is compact, it follows from
Theorem 1.6 that zi must be locally constant. Hence, dim M = 0.
Remark 1. The above result means that there is no chance for a Whitney Embedding
Theorem in the holomorphic category. One of the major results of the theory of
complex manifolds is the Kodaira Embedding Theorem which gives necessary and
sufficient conditions for a compact complex manifold to embed in Pn .
Example 1.12. Let f : Cn → C be a holomorphic function and suppose Z =
f −1 (0) = ∅. Then we say that 0 is a regular value for f if rankp (f ) = 1 for all p ∈ Z;
i.e. for each p ∈ X there exists some i, i = 1, . . . , n such that ∂f /∂zi (p) = 0. In
this case, Z is a complex submanifold of Cn and codim(Z) = 1. We call Z an affine
hypersurface. More generally, given F : Cn → Cm , we say that 0 is a regular value
if rankp (F ) = m for all p ∈ F −1 (0). In this case or F −1 (0) is either empty or is a
submanifold of Cn of codimension m.
Example 1.13. Let P (z0 , . . . , zn ) be a homogeneous polynomial of degree d. We
set
X := {[z] ∈ Pn : P (z0 , . . . , zn ) = 0}.
We note that while P does not define a function on Pn , the zero locus X is still well
defined since P is a homogeneous polynomial. We assume now that the following
regularity condition holds:

n+1 ∂P ∂P
(1.11) z∈C : (z) = · · · = (z) = 0 = {0};
∂z0 ∂zn
i.e. 0 is a regular value of the map P |Cn+1 {0} . Then X is a hypersurface in Pn .
To prove this we note that the requirements of Definition 1.10 are local. Hence,
it is enough to check that X ∩ Ui is a submanifold of Ui for each i; indeed, an affine
hypersurface. Consider the case i = 0 and let f : U0 ∼ = Cn → C be the function
f (u1 , . . . , un ) = P (1, u1 , . . . , un ). Set u = (u1 , . . . , un ) and ũ = (1, u1 , . . . , un ).
Suppose [ũ] ∈ U0 ∩ X and
∂f ∂f
(u) = · · · = (u) = 0
∂u1 ∂un
then
∂P ∂P
(ũ) = · · · = (ũ) = 0
∂z1 ∂zn
But, since P is a homogeneous polynomial of degree d, it follows from the Euler
identity that:
∂P
0 = d · P (ũ) = (ũ).
∂z0
Hence, by (1.11), we would have ũ = 0, which is impossible. Hence 0 is a regular
value of f and X ∩ U0 is an affine hypersurface.
8 EDUARDO CATTANI
Exercise 6. Let P1 (z0 , . . . , zn ), . . . , Pm (z0 , . . . , zn ) be homogeneous polynomials.

Suppose that 0 is a regular value of the map
(P1 , . . . , Pm ) : Cn+1 {0} → Cm .
Prove that
X = {[z] ∈ Pn : P1 ([z]) = · · · = Pm ([z]) = 0}
is a codimension m submanifold of Pn . X is called a complete intersection subman-
ifold.
Example 1.14. Consider the Grassmann manifold G(k, n) and let I1 , . . . , I(n) , de-
k
note all strictly increasing k-tuples I ⊂ {1, . . . , n}. We then define
p : G(k, n) → PN −1 ; p([W ]) = [det(WI1 ), . . . , det(WIN )]
Note that the map p is well defined, since W ∼ W implies that W = W · M with
M ∈ GL(k, C) and then for any index set I, det(WI ) = det(M ) det(WI ). We leave
it to the reader to verify that the map p, which is usually called the Plücker map, is
holomorphic.
Exercise 7. Consider the Plücker map p : G(2, 4) → P5 and suppose that the index
sets I1 , . . . , I6 are ordered lexicographically. Show that p is a 1 : 1 holomorphic map
from G(2, 4) onto the subset
(1.12) X = {[z0 , . . . , z6 ] : z0 z5 − z1 z4 + z2 z3 = 0}.
Prove that X is a hypersurface in P5 . Compute rank[W ] p for [W ] ∈ G(2, 4).
Example 1.15. We may define complex Lie groups in a manner completely analo-
gous to the real, smooth case. A complex Lie group is a complex manifold G with a
group structure such that the group operations are holomorphic. The basic example
of a complex Lie group is GL(n, C). We have already observed that GL(n, C) is an
2
open subset of Cn and the product of matrices is given by polynomial functions,
while the inverse of a matrix is given by rational functions on the entries of the
matrix. Other classical examples include the special linear group SL(n, C) and the
symplectic group Sp(g, C). We recall the definition of the latter. Let J be the matrix
(1.4), then
(1.13) Sp(n, C) := {X ∈ GL(2n, C) : X T · J · X = J}.
We set Sp(n, R) := Sp(n, C) ∩ GL(2n, R) for R = Z, Q, R. More generally, let V
be a real vector space, VC its complexification (cf. (A.1)), and A a non-degenerate,
alternating form on V . We denote by Sp(A, V ) (resp. Sp(A, VC )) the group of
automorphisms of V (resp. VC ) that preserve A. Note that we must have dimR (V ) =
2n and there exists a basis relative to which the matrix of A is given by (1.4).
Example 1.16. Let A be a non-degenerate, alternating form on a 2n-dimensional,
real vector space V . Consider the space:
M = {Ω ∈ G(n, VC ) : A(u, v) = 0 for all u, v ∈ Ω}.
Let {e1 , . . . , en , en+1 , . . . , e2n } be a basis of V in which the matrix of A is as in (1.4).

Then if
W1
Ω = [W ] = ,
W2
where W1 and W2 are n × n matrices, we have that Ω ∈ M if and only if

0 −In W1
W1T , W2T · · = W2T · W1 − W1T · W2 = 0.
In 0 W2
Set I0 = {1, . . . , n}. Every element Ω ∈ M ∩ UI0 may be represented by a matrix
of the form Ω = [In , Z]T with Z T = Z. It follows that M ∩ UI0 is an n(n + 1)/2-
dimensional submanifold. Now, given an arbitrary Ω ∈ M , there exists an element
X ∈ Sp(A, VC ) such that X · Ω = Ω0 , where Ω0 = span(e1 , . . . , en ). Since the
elements of Sp(A, VC ) act by biholomorphisms on G(n, VC ) it follows that M and
n(n + 1)/2 dimensional submanifold of G(n, VC ). Moreover, since M is a closed
submanifold of the compact manifold G(k, n), M is also compact.
We will also be interested in considering the open set D ⊂ M consisting of
(1.14) D = D(V, A) := {Ω ∈ M : i A(w, w̄) > 0 for all 0 = w ∈ Ω}.
It follows that Ω ∈ D if and only if the hermitian matrix

0 −In W1
i · W̄1 , W̄2 ·
T T
· = i(W̄2T · W1 − W̄1T · W2 )
In 0 W2
is positive definite. Note that in particular D ⊂ UI0 and that
(1.15) D ∼
= {Z ∈ M(n, C) : Z T = Z ; Im(Z) = (1/2i)(Z − Z̄) > 0},
where M(n, C) denotes the n × n complex matrices. If n = 1 then M ∼ = C and D is
the upper-half plane. We will call D the generalized Siegel upper-half space.
The elements of the complex lie group Sp(A, VC ) ∼ = Sp(n, C) define biholomor-
phisms of G(n, VC ) preserving M . The subgroup
Sp(A, V ) = Sp(A, VC ) ∩ GL(V ) ∼
= Sp(n, R)
preserves D.
Exercise 8. Prove that relative to the description of D as in (1.15) the action of
Sp(A, V ) is given by generalized fractional linear transformations:

A B
· Z = (A · Z + B) · (C · Z + D)−1 .
C D
Exercise 9. Prove that the action of Sp(A, V ) on D is transitive in the sense that
given any two points Ω, Ω ∈ D there exists X ∈ Sp(A, V ) such that X · Ω = Ω .
Exercise 10. Compute the isotropy subgroup:
K := {X ∈ Sp(A, V ) : X · Ω0 = Ω0 },
where Ω0 = [In , i In ]T .
10 EDUARDO CATTANI
Example 1.17. Let TΛ := C/Λ, where Λ ⊂ Z2 is a rank-two lattice in C; i.e.

Λ = {m ω1 + n ω2 ; m, n ∈ Z},
where ω1 , ω2 are complex numbers linearly independent over R. TΛ is locally diffeo-
morphic to C and since the translations by elements in Λ are biholomorphisms of
C, TΛ inherits a complex structure relative to which the natural projection
πΛ : C → TΛ
is a local biholomorphic map.
It is natural to ask if, for different lattices Λ, Λ , the complex tori TΛ , TΛ are
biholomorphic. Suppose F : TΛ → TΛ is a biholomorphism. Then, since C is the
universal covering of TΛ there exists a map F̃ : C → C such that the diagram:
F̃
C −−−−→ C
⏐ ⏐
πΛ
⏐ ⏐π
Λ’
F
C/Λ −−−−→ C/Λ
commutes. In particular, given z ∈ C, λ ∈ Λ, there exists λ ∈ Λ such that
F̃ (z + λ) = F̃ (z) + λ .
This means that the derivative F̃ must be Λ-periodic and, hence, it defines a holo-
morphic function on C/Λ which, by Theorem 1.6, must be constant. This implies
that F̃ must be a linear map and, after translation if necessary, we may assume
that F̃ (z) = μ · z, μ = a + ib ∈ C. Conversely, any such linear map F̃ induces a
biholomorphic map C/Λ → C/F̃ (Λ). In particular, if {ω1 , ω2 } is a Z-basis of Λ then
Im(ω2 /ω1 ) = 0 and we may assume without loss of generality that Im(ω2 /ω1 ) > 0.
Setting μ = ω2 /ω1 we see that TΛ is always biholomorphic to a torus Tτ associated
with a lattice
{m + nτ ; m, n ∈ Z}
and where Im(τ ) > 0.
Now, suppose the tori Tλ , TΛ are biholomorphic and let {ω1 , ω2 } (resp. {ω1 , ω2 })
be a Z-basis of Λ (resp. Λ ) as above. We have
μ · ω1 = m11 ω1 + m21 ω2 ; μ · ω2 = m12 ω1 + m22 ω2 , mij ∈ Z.
Moreover, m11 m22 −m12 m21 = 1, since F is biholomorphic and therefore F̃ (Λ) = Λ .
Hence
ω1 m11 ω1 + m21 ω2 m11 + m21 τ
τ = = =
ω2 m12 ω1 + m22 ω2 m12 + m22 τ
Consequently, Tτ ∼= Tτ if and only if τ and τ are points in the upper-half plane
congruent under the action of the group SL(2, Z) by fractional linear transformations.
Remark 2. Note that while all differentiable structures on the torus S 1 × S 1 are
equivalent there is a continuous moduli of different complex structures. This is one
of the key differences between real and complex geometry and one which we will
study using Hodge Theory.
1.2. Holomorphic Vector Bundles. We may extend the notion of smooth vector
bundle to complex manifolds and holomorphic maps:
Definition 1.18. A holomorphic vector bundle E over a complex manifold M is a
complex manifold E together with a holomorphic map π : E → M such that:
i) For each x ∈ M , the fiber Ex = π −1 (x) is a complex vector space of dimension
d (the rank of E).
ii) There exists an open covering {Uα } of M and biholomorphic maps
Φα : π −1 (Uα ) → Uα × Cd
such that
(a) p1 (Φα (x)) = x for all x ∈ U , where p1 : Uα ×Cd → Uα denotes projection
on the first factor, and
(b) For every x ∈ Uα the map p2 ◦ Φ|Ex : Ex → Cd is an isomorphism of
C-vectorspaces.
E is called the total space of the bundle and M its base. The covering {Uα } is
called a trivializing cover of M and the biholomorphisms {Φα } local trivializations.
When d = 1 we often refer to E as a line bundle.
We note that as in the case of smooth vector bundles, a holomorphic vector bundle
may be described by transition functions. That is, by a covering of M by open sets
Uα together with holomorphic maps
gαβ : Uα ∩ Uβ → GL(d, C)
such that
(1.16) gαβ · gβγ = gαγ
on Uα ∩ Uβ ∩ Uγ . The maps gαβ are defined by the commutative diagram:
(1.17) π −1 (Uα ∩ Uβ )
mm QQQ
Φβ mmm QQQΦα
mm QQQ
mmm QQQ
v mm
m (id,gαβ )
Q(
(Uα ∩ Uβ ) × Rd / (Uα ∩ Uβ ) × Rd
In particular, a holomorphic line bundle over M is given by a collection {Uα , gαβ },

where Uα is an open cover of M and the {gαβ } are nowhere-zero holomorphic func-
tions defined on Uα ∩ Uβ , i.e. gαβ ∈ O∗ (Uα ∩ Uβ ) satisfying the cocycle condition
(1.16).
Example 1.19. The product M × Cd with the natural projection may be viewed as
vector bundle of rank d over the complex manifold M . It is called the trivial bundle
over M .
Example 1.20. We consider the tautological line bundle over Pn . This is the bundle
whose fiber over a point in Pn is the line in Cn+1 defined by that point. More
precisely, let
T := {([z], v) ∈ Pn × Cn+1 : v = λz, λ ∈ C},
12 EDUARDO CATTANI
and let π : T → Pn be the projection to the first factor. Let Ui be as in (1.6). Then
we can define
Φi : π −1 (Ui ) → Ui × C
by
Φi ([z], v) = vi .
The transition functions gij are defined by the diagram (1.17) and we have
Φi ◦ Φ−1
j ([z], 1) = Φi ([z], (z0 /zj , . . . , 1, . . . , zn /zj )) = ([z], zi /zj ),
with the one in the j-th position. Hence,
gij : Ui ∩ Uj → GL(1, C) ∼
= C∗
is the map [z] → zi /zj . It is common to denote the tautological bundle as O(−1).
Exercise 11. Generalize the construction of the tautological bundle over projective
space to obtain the universal rank k bundle over the Grassmann manifold G(k, n).
Consider the space:
(1.18) U := {(Ω, v) ∈ G(k, n) × Cn : v ∈ Ω},
where we regard Ω ∈ G(k, n) as a k-dimensional subspace of Cn . Prove that U
may be trivialized over the open sets UI defined in Example 1.5 and compute the
transition functions relative to these trivializations.
Let π : E → M be a holomorphic vector bundle and suppose F : N → M is
a holomorphic map. Given a trivializing cover {(Uα , Φa )} of E with transition
functions gαβ : Uα ∩ Uβ → GL(d, C), we define
(1.19) hαβ : F −1 (Uα ) ∩ F −1 (Uβ ) → GL(d, C) ; hαβ := gαβ ◦ F.
It is easy to check that the functions hαβ satisfy the cocycle condition (1.16) and,
therefore, define a holomorphic vector bundle over N denoted by F ∗ (E), and called
the pullback bundle. Note that we have a commutative diagram:
F̄
F ∗ (E) −−−−→ E
⏐ ⏐
(1.20) *
⏐ ⏐π
π
F
N −−−−→ M
L L are their transition functions relative
If L and L are line bundles, and gαβ , gαβ
to a common trivializing cover then the functions:

L
hαβ = gαβ · gαβ
L
satisfy (1.16) and define a new line bundle which we denote by L ⊗ L .

Similarly, the functions
L −1
hαβ = (gαβ )
also satisfy (1.16) and define a bundle, called the dual bundle of L and denoted by
L∗ or L−1 . Clearly L ⊗ L∗ is the trivial line bundle over M . The dual bundle of
the tautological bundle is called the hyperplane bundle over Pn and denoted by H
H ∈ O ∗ (U ∩ U ) defined by
or O(1). Note that the transition functions of H are gij i j
H
(1.21) gij ([z]) := zj /zi .
We may also extend the notion of sections to holomorphic vector bundles:

Definition 1.21. A holomorphic section of a holomorphic vector bundle π : E → M
over an open set U ⊂ M is a holomorphic map:
σ: U → E
such that
(1.22) π ◦ σ = id|U .
The sections of E over U are an O(U )-module which will be denoted by O(U, E).
Clearly, the local sections over U of the trivial line bundle are the ring O(U ).
If π : L → M is a line bundle and gαβ are the transition functions associated to
a trivializing covering (Uα , Φα ), then a section σ : M → L may be described by a
collection of functions fα ∈ O(Uα ) defined by:
σ(x) = fα (x)Φ−1
α (x, 1).
Hence, for x ∈ Uα ∩ Uβ we must have
(1.23) fα (x) = gαβ (x) · fβ (x).
Example 1.22. Let M = Pn and let Ui = {[z] ∈ Pn : zi = 0}. Let P ∈ C[z0 , . . . , zn ]
be a homogeneous polynomial of degree d. For each i = 0, . . . , n define
P (z)
fi ([z]) = ∈ O(Ui ).
zid
We then have in Ui ∩ Uj :
zid · fi ([z]) = P (z) = zjd · fj ([z]),
and therefore
fi ([z]) = (zj /zi )d · fj ([z]).
This means that we can consider the polynomial P (z) as defining a section of the
line bundle over Pn with transition functions
gij = (zj /zi )d ,
that is of the bundle H d = O(d). In fact, it is possible to prove that every global
holomophic section of the bundle O(d) is defined, as above, by a homogeneous poly-
nomial of degree d. The proof of this fact requires Hartogs’ Theorem [18, Proposi-
tion 1.1.14] from the theory of holomorphic functions of several complex variables.
We refer to [18, Proposition 2.4.1].
We note that, on the other hand, the tautological bundle has no non-trivial global
holomorphic sections. Indeed, suppose σ ∈ O(Pn , O(−1)) let denote the global
section of O(1) associated to a non-zero linear form . Then, the map
[z] ∈ Pn → ([z])(σ([z])
defines a global holomorphic function on the compact complex manifold Pn , hence
it must be constant. If that constant is non-zero then both σ and are nowhere-zero
which would imply that both O(−1) and O(1) are trivial bundles. Hence σ must be
identically zero.
14 EDUARDO CATTANI
Note that given a section σ : M → E of a vector bundle E, the zero locus {x ∈

M : σ(x) = 0} is a well defined subset of M . Thus, we may view the projective
hypersurface defined in Example 1.13 by a homogeneous polynomial of degree d as
the zero-locus of a section of O(d).
Remark 3. The discussion above means that one should think of sections of line
bundles as locally defined holomorphic functions satisfying a suitable compatibil-
ity condition. Given a compact, connected, complex manifold, global sections of
holomorphic line bundles (when they exist) often play the role that global smooth
functions play in the study of smooth manifolds. In particular, one uses sections
of line bundles to define embeddings of compact complex manifolds into projective
space. This vague observation will be made precise later in the course.
Given a holomorphic vector bundle π : E → M and a local trivialization
Φ : π −1 (U ) → U × Cd
we may define a basis of local sections of E over U (a local frame) as follows. Let
e1 , . . . , ed denote the standard basis of Cd and for x ∈ U set:
σj (x) := Φ−1 (x, ej ); j = 1, . . . , d
Then σj (x) ∈ O(U, E) and for each x ∈ U the vectors σ1 (x), . . . , σd (x) are a basis
of the d-dimensional vector space Ex (they are the image of the basis e1 , . . . , ed by
a linear isomorphism). In particular, if τ : U → M is a map satisfying (1.22) we can
write:
d
τ (x) = fj (x)σj (x)
j=1
and τ is holomorphic (resp. smooth) if and only if the functions fj ∈ O(U ) (resp.
fj ∈ C ∞ (U )).
Conversely, suppose U ⊂ M is an open set and let σ1 , . . . , σd ∈ O(U, E) be a local
frame; i.e. holomorphic sections such that for each x ∈ U , σ1 (x), . . . , σd (x) is a basis
of Ex , then we may define a local trivialization
Φ : π −1 (U ) → U × Cd
by
Φ(v) := (π(v), (λ1 , . . . , λd )),
where v ∈ π −1 (U ) and

d
v = λj σj (π(v)).
j=1
2. Differential Forms on Complex Manifolds

2.1. Almost Complex Manifolds. Given a complex manifold M and a coordinate
atlas (Uα , φα ) covering M , the fact that the change-of-coordinate maps (1.5) are
holomorphic implies that the matrix of the differential D(φβ ◦ φ−1 α ) is of the form
(1.3). This means that the map
Jp : Tp (M ) → Tp (M )
defined by

∂ ∂ ∂ ∂
(2.1) J := ; J := −
∂xj ∂yj ∂yj ∂xj
is well defined, provided that the functions zj = xj + iyj , j = 1, . . . , n, define local
holomorphic coordinates near the point p. We note that J is a (1, 1) smooth tensor
on M such that J 2 = − id and that, for each p ∈ M , Jp defines a complex structure
on the real vector space Tp (M ) (cf.(A.8)).
Definition 2.1. An almost complex structure on a C ∞ (real) manifold M is a (1, 1)
tensor J such that J 2 = − id. An almost complex manifold is a pair (M, J) where
J is an almost complex structure on M . The almost complex structure J is said to
be integrable if M has a complex structure inducing J.
If (M, J) is an almost complex manifold then Jp is a complex structure on M and
therefore by Proposition A.1, M must be even-dimensional. We also have:
Exercise 12. If M has an almost complex structure then M is orientable.
Exercise 13. Let M be an orientable (and oriented) two-dimensional manifold and
let , be a Riemannian metric on M . Given p ∈ M let v1 , v2 ∈ Tp (M ) be a
positively oriented orthonormal basis. Prove that Jp : Tp (M ) → Tp (M ) defined by:
Jp (v1 ) = v2 ; Jp (v2 ) = −v1 ,
defines an almost complex structure on M . Show, moreover, that if , is a
Riemannian metric conformally equivalent to , then the two metrics define the
same almost complex structure.
The discussion above shows that if M is a complex manifold then the operator
(2.1) defines an almost complex structure. It is natural to ask about the converse
of this statement; i.e. when does an almost complex structure J on a manifold
arise from a complex structure? The answer, which is provided by the Newlander-
Nirenberg Theorem, may be stated in terms of the Nijenhuis torsion of J:
Exercise 14. Let J be an almost complex structure on M . Prove that
(2.2) N (X, Y ) = [JX, JY ] − [X, Y ] − J[X, JY ] − J[JX, Y ]
is a (1, 2)-tensor satisfying N (X, Y ) = −N (Y, X). N is called the torsion of J.
Exercise 15. Let J be an almost complex structure on a two-dimensional manifold
M . Prove that N (X, Y ) = 0 for all vector fields X and Y on M .
Theorem 2.2 (Newlander-Nirenberg). Let (M, J) be an almost complex man-
ifold, then M has a complex structure inducing the almost complex structure J if
and only if N (X, Y ) = 0 for all vector fields X and Y on M .
Proof. We refer to [38, Proposition 2], [34, §2.2.3] for a proof in the case when M is
a real analytic manifold.
Remark 4. Note that assuming the Newlander-Nirenberg Theorem, it follows from
Exercise 15 that the almost complex structure constructed in Exercise 13 is inte-
grable. We may explicitly construct the complex structure on M by using local
16 EDUARDO CATTANI
isothermal coordinates. Thus, a complex structure on an oriented, two-dimensional

manifold M is equivalent to a Riemannian metric up to conformal equivalence.
In what follows we will be interested in studying complex manifolds; however, the
notion of almost complex structures gives a very convenient way to distinguish those
properties of complex manifolds that depend only on having a (smoothly varying)
complex structure on each tangent space. Thus, we will not explore in depth the
theory of almost complex manifolds except to note that there are many examples
of almost complex structures which are not integrable, that is, do not come from a
complex structure. One may also ask which even-dimensional orientable manifolds
admit almost complex structures. For example, in the case of a sphere S 2n it was
shown by Borel and Serre that only S 2 and S 6 admit almost complex structures.
This is related to the fact that S 1 , S 3 and S 7 are the only parallelizable spheres. We
point out that while it is easy to show that S 6 has a non-integrable almost complex
structure it is still unknown whether S 6 has a complex structure.
2.2. Tangent and Cotangent space. Let (M, J) be an almost complex manifold
and p ∈ M . Let Tp (M ) denote the tangent space of M . Then Jp defines a com-
plex structure on Tp (M ) and therefore, by Proposition A.1, the complexification:
Tp,C (M ) := Tp (M ) ⊗R C, decomposes as

Tp,C (M ) = Tp (M ) ⊕ Tp (M )

where Tp (M ) = Tp (M ) and Tp (M ) is the i-eigenspace of J acting on Tp,C (M ). More-
over, by Proposition A.2, the map v ∈ Tp (M ) → v − iJp (v) defines an isomorphism
of complex vector spaces (Tp (M ), Jp ) ∼ = Tp (M ).
If J is integrable, then given holomorphic local coordinates {z1 , . . . , zn } around
p, we may consider the local coordinate frame (1.2) and, given (2.1), we have that
the above isomorphism maps:
∂/∂xj → ∂/∂xj − i ∂/∂yj .
We set

∂ 1 ∂ ∂ ∂ 1 ∂ ∂
(2.3) := −i ; := +i .
∂zj 2 ∂xj ∂yj ∂ z̄j 2 ∂xj ∂yj
Then, the vectors ∂/∂zj ∈ Tp (M ) are a basis of the complex subspace Tp (M ).
Remark 5. Given local coordinates (U, {z1 , . . . , zn }) on M , a function f : U → C is

holomorphic if the local coordinates expression f (z1 , . . . , zn ) satisfies the Cauchy-
Riemann equations. This is equivalent to the condition
∂
(f ) = 0,
∂ z̄j
for all j. Moreover, in this case ∂z∂ j (f ) coincides with the partial derivative of f with
respect to zj . This justifies the choice of notation. However, we point out that it
makes sense to consider ∂z∂ j (f ) even if the function f is only a C ∞ function.
We will refer to Tp (M ) as the holomorphic tangent space† of M at p. We note

that if {z1 , . . . , zn } and {w1 , . . . , wn } are local complex coordinates around p then
the change of basis matrix from the basis {∂/∂zj } to the basis {∂/∂wk } is given by
the matrix of holomorphic functions

∂wk
.
∂zj
Thus, the complex vector spaces Tp (M ) define a holomorphic vector bundle T h (M )
over M , the holomorphic tangent bundle.
Example 2.3. Let M be an oriented real surface with a Riemannian metric. Let
(U, (x, y)) be positively-oriented, local isothermal coordinates on M ; i.e. the coordi-
nate vector fields ∂/∂x, ∂/∂y are orthogonal and of the same length. Then z = x+iy
defines complex coordinates on M and the vector field ∂/∂z = ∂/∂x − i ∂/∂y is a
local holomorphic section of the holomorphic tangent bundle of M .
We can now characterize the holomorphic tangent bundle of Pn :
Theorem 2.4. The holomorphic tangent bundle T h Pn is equivalent to the bundle
Hom(T , E/T ),
where E = Pn × Cn+1 is the trivial bundle of rank n + 1 on Pn .
Proof. Consider the projection π : Cn+1 \ {0} → Pn . Given λ ∈ C∗ , let Mλ : Cn+1 \
{0} → Cn+1 \ {0} be the map “multiplication by λ”. Then, for every v ∈ Cn+1 \ {0}
we may identify T (Cn+1 \ {0}) ∼= Cn+1 and we have the following commutative
diagram
Mλ
Cn+1I / Cn+1
II uu
II u
I uu
π∗,v III uuuπ∗,λv
$ uz
T[v] (Pn )
(Pn ) is surjective and its kernel is the line L = C · v.
Now, the map π∗,v : Cn+1 → T[v]
Hence we get a family of linear isomorphisms

pv : Cn+1 /L → T[v] (Pn ); v ∈ L, v = 0
with the relation
pv = λ pλv
We can now define a map
Θ : Hom(T , E/T ) → T h Pn .
Let
ξ ∈ Hom(T , E/T )[z] = Hom(T[z] , (E/T )[z] ) ∼
= Hom(L, Cn+1 /L)
then we set
Θ(ξ) := pv (ξ(v)) for any v ∈ L, v = 0.
†
This construction makes sense even if J is not integrable. In that case we may replace the
coordinate frame (1.2) by a local frame {X1 , . . . , Xn , Y1 , . . . , Yn } such that J(Xj ) = Yj and J(Yj ) =
−Xj .
18 EDUARDO CATTANI
Note that this is well defined since

pλv (ξ(v)) = λ−1 pv (λ ξ(v)) = pv (ξ(v)).
Alternatively one may define
d
Θ(ξ) = |t=0 (γ(t)),
dt
where γ(t) is the holomorphic curve through [z] in Pn defined by
γ(t) := [v + tξ(v)].
One then has to show that this map is well defined. It is straightforward, though
tedious, to verify that Θ is an isomorphism of vector bundles.
Exercise 16. Prove that
T h (G(k, n)) ∼
= Hom(U, E/U),
where U is the universal bundle over G(k, n) defined in Exercise 1.18 and E is the
trivial bundle E = G(k, n) × Cn .
As seen in Appendix A, a complex structure on a vector space induces a dual com-
plex structure on the dual vector space. Thus, the complexification of the cotangent
∗ (M ) decomposes as
space Tp,C
∗
Tp,C (M ) := Tp1,0 (M ) ⊕ Tp0,1 (M ) ; Tp0,1 (M ) = Tp1,0 (M ).
Given local holomorphic coordinates {z1 , . . . , zn }, zj = xj + iyj , the one-forms
dzj := dxj + idyj , dz̄j = dxj − idyj , are the dual coframe to ∂/∂z1 , . . . , ∂/∂ z̄n
and, consequently dz1 , . . . , dzn are a local holomorphic frame of the holomorphic
bundle T 1,0 (M ).
The complex structure on Tp∗ (M ) induces a decomposition of the k-th exterior
product (cf. (A.19)):
∗
Λk (Tp,C (M )) = Λa,b
p (M ),
a+b=k
where
a times b times

1,0
(2.4) Λa,b
p (M ) = T p (M ) ∧ . . . ∧ T 1,0
p (M ) ∧ T 0,1
p (M ) ∧ . . . ∧ T 0,1
p (M ).
In this way, the smooth vector bundle Λk (TC∗ (M )) decomposes as a direct sum of
C ∞ vector bundles

(2.5) Λk (TC∗ (M )) = Λa,b (M ).
a+b=k
We will denote by Ak (U ) (resp. Aa,b (U )) the C ∞ (U ) module of local sections of the

bundle Λk (TC∗ (M )) (resp. Λa,b (M )) over U . We then have

(2.6) Ak (U ) = = Aa,b (U ).
a+b=k
Note that, given holomorphic coordinates {z1 , . . . , zn }, the local differential forms
dzI ∧ dz̄J := dzi1 ∧ · · · dzia ∧ dz̄j1 ∧ · · · dz̄jb ,
where I (resp. J) runs over all strictly increasing index sets 1 ≤ i1 < · · · < ia ≤ n
of length a (resp. 1 ≤ j1 < · · · < jb ≤ n of length b) are a local frame for the bundle
Λa,b (M ).
We note that the bundles Λk,0 (M ) are holomorphic bundles of rank nk . We de-
note them by ΩkM to emphasize that we are viewing them as holomorphic, rather than
smooth, bundles. We denote the O(U )-module of holomorphic sections by Ωk (U ). In
particular Λn,0 (M ) is a holomorphic line bundle over M called the canonical bundle
and usually denoted by KM .
Example 2.5. Let M = P1 . Then as we saw in Example 1.4, M is covered by two
coordinate neighborhoods (U0 , φ0 ), (U1 , φ1 ). The coordinate changes are given by
the maps φ1 ◦ φ−1 ∗
0 : C →C :
∗
w = φ1 ◦ φ−1
0 (z) = φ1 ([1, z]) = 1/z.
This means that the local sections dz, dw of the holomorphic cotangent bundle are
related by
dz = (−1/z)2 dw.
It follows from (1.17) that g01 [z0 , z1 ] = −(z0 /z1 )2 . Hence KP1 ∼
= O(−2) = T 2 .
Exercise 17. Find the transition functions for the holomorphic cotangent bundle
of Pn . Prove that KPn ∼
= O(−n − 1) = T n+1 .
2.3. de Rham and Dolbeault Cohomologies. † We recall that if U ⊂ M is an
open set in a smooth manifold M and Ak (U ) denotes the space of C-valued differ-
ential k-forms on U , then there exists a unique operator, the exterior differential:
d : Ak (U ) → Ak+1 (U ) ; k≥0
satisfying the following properties:
i) d is C-linear;
ii) For f ∈ A0 (U ) = C ∞ (U ), df is the one-form on U which acts on a vector
field X by df (X) := X(f ).
iii) Given α ∈ Ar (U ), β ∈ As (U ), the Leibniz property holds:
(2.7) d(α ∧ β) = dα ∧ β + (−1)r α ∧ dβ;
iv) d ◦ d = 0.
It follows from (ii) above that if {X1 , . . . , Xm } is a local frame on U ⊂ M and
ξ1 , . . . , ξm ∈ A1 (U ) is the dual coframe, then given f ∈ C ∞ (U ) we have
m
df = Xi (f )ξi .
i=1
In particular, if M is a complex manifold and (U, z1 , . . . , zn ) are local coordinates
then for f ∈ C ∞ (U ) we have
n
∂f
n
∂f n
∂f n
∂f
(2.8) df = dxj + dyj = dzj + dz̄j
∂xj ∂yj ∂zj ∂ z̄j
j=1 j=1 j=1 j=1
†
Many of the topics in this section will be studied in greater depth in the courses by L. Tu and
F. Elzein.
20 EDUARDO CATTANI
The properties of the operator d imply that for each open set U in M we have a
complex:
C → C ∞ (U ) → A1 (U ) → · · · → A2n−1 (U ) → A2n (U )
d d d d
(2.9)
The quotients:
ker{d : Ak (U ) → Ak+1 (U )}
(2.10) k
HdR (U, C) :=
d(Ak−1 (U ))
are called the de Rham cohomology groups of U . The elements in
Z k (U ) := ker{d : Ak (U ) → Ak+1 (U )}
are called closed k-forms and the elements in B k (U ) := d(Ak−1 (U )) exact k-forms.
We note that if U is connected then HdR0 (U, C) ∼ C. Unless there is possibility of
=
confusion we will drop the subscript since, in this notes, we will only consider de
Rham cohomology.
Exercise 18. Prove that the closed forms are a subring of the ring of differential
forms and that the exact forms are an ideal in the ring of closed forms. Deduce that
the de Rham cohomology

(2.11) H ∗ (U, C) := H k (U, C),
k≥0
inherits a ring structure:
(2.12) [α] ∪ [β] := [α ∧ β].
This is called the cup product on cohomology.
If F : M → N is a smooth map, then given an open set V ⊂ N , F induces maps
F ∗ : Ak (V ) → Ak (F −1 (V ))
which commute with the exterior differential; i.e. F ∗ is a map of complexes. This
implies that F ∗ defines a map between de Rham cohomology groups:
F ∗ : H k (V, C) → H k (F −1 (V ), C)
which satisfies the chain rule (F ◦ G)∗ = G∗ ◦ F ∗ . Since (id)∗ = id it follows that if
F : M → N is a diffeomorphism then F ∗ : H k (N, C) → H k (M, C) is an isomorphism.
In fact, the de Rham cohomology groups are a (smooth) homotopy invariant:
Definition 2.6. Let f0 , f1 : M → N be smooth maps. We say that f0 is (smoothly)
homotopic to f1 if there exists a smooth map
H: R×M → N
such that H(0, x) = f0 (x) and H(1, x) = f1 (x) for all x ∈ M .
Theorem 2.7. Let f0 , f1 : M → N be smoothly homotopic maps. Then
f0∗ = f1∗ : H k (N, C) → H k (M, C).
Proof. We refer to [3, §4] for a proof of this important result.
Corollary 2.8 (Poincaré Lemma). Let U ⊂ M be a contractible open subset then
H k (U, C) = {0} for all k ≥ 1.
Proof. The result follows from Theorem 2.7 since in a contractible open set the
identity map is homotopic to a constant map.
Hence, if U is contractible, the sequence
0 → C → C ∞ (U ) → A1 (U ) → · · · → A2n−1 (U ) → A2n (U ) → 0
d d d d
(2.13)
is exact.
The exterior differential operator is not of “pure bidegree” relative to the decom-
position (2.6). Indeed, it follows from (2.8) that
d(Aa,b (U )) ⊂ Aa+1,b (U ) ⊕ Aa,b+1 (U ).
¯ where ∂ (resp. ∂)
We write d = ∂ + ∂, ¯ is the component of d of bidegree (1, 0) (resp.
2
(0, 1)). From d = 0 we obtain:
(2.14) ∂ 2 = ∂¯2 = 0 ; ∂ ◦ ∂¯ + ∂¯ ◦ ∂ = 0.
¯
Exercise 19. Generalize the Leibniz property to the operators ∂ and ∂.
It follows from (2.14) that, for each p, 0 ≤ p ≤ n, we get a complex
∂¯ ∂¯ ∂¯ ∂¯
(2.15) Ap,0 (U ) → Ap,1 (U ) → · · · → Ap,n−1 (U ) → Ap,n (U )
called the Dolbeault complex. Its cohomology spaces are denoted by H∂p,q
¯ (U ) and
called the Dolbeault cohomology groups.
Exercise 20. Let α ∈ Ap,q (U ). Prove that ∂α = ∂¯ᾱ. Deduce that a form α
¯
is ∂-closed if and only if ᾱ is ∂-closed. Similarly for ∂-exact forms. Conclude
that via conjugation the study of ∂-cohomology reduces to the study of Dolbeault
cohomology.
Proposition 2.9. Let M be a complex manifold, then H∂p,0 ∼ p
¯ (M ) = Ω (M ).
Proof. Let α ∈ Ap,0 (M ) and suppose ∂α ¯ = 0. Let (U, {z1 , . . . , zn } be local coordi-
nates in M and let α|U = I fI dzI , where I runs over all increasing index sets
{1 ≤ i1 < · · · < ip ≤ n}. Then

n
∂fI
¯ U =
∂α| dz̄j ∧ dzI = 0 ,
∂ z̄j
I j=1
This implies that ∂fI /∂ z̄j = 0 for all I and all j. Hence fI ∈ O(U ) for all I and α
is a holomorphic p-form.
Given a = (a1 , . . . , an ) ∈ Cn and ε = (ε1 , . . . , εn ) ∈ (R>0 ∪ ∞)n we denote by
Δε (a) = {z ∈ Cn : |zi − ai | < εi }
the n-dimesional polydisk. For n = 1, a = 0, ε = 1 we set Δ = Δ1 (0), the unit disk,
and Δ∗ = Δ \ {0} the punctured unit disk. The following result is known as the
¯
∂-Poincaré Lemma:
¯
Theorem 2.10. If q ≥ 1 and α is a (p, q), ∂-closed form on a polydisk Δε (a), then
¯
α is ∂-exact; i.e.
H∂p,q
¯ (Δε (a)) = 0 ; q ≥ 1 .
22 EDUARDO CATTANI
Proof. We refer to [15, Chapter 0] or [18, Corollary 1.3.9] for a proof.

Hence, if U = Δε (a) is a polydisk we have exact sequences:
∂¯ ∂¯ ∂¯
(2.16) 0 → Ωp (U ) → Ap,0 (U ) → Ap,1 (U ) → · · · → Ap,n (U ) → 0
Remark 6. It will be shown in the course by L. Tu that:

C if 0 ≤ k = 2 ≤ 2n
(2.17) H k (Pn , C) ∼
=
0 otherwise.

n ∼ C if 0 ≤ p = q ≤ n
(2.18) H∂p,q
¯ (P ) =
0 otherwise.
Also in that course you will see how to realize the de Rham and Dolbeault coho-
mology groups as sheaf cohomology groups. This will show, in particular, that even
though our definition of the de Rham cohomology uses the differentiable structure
they are, in fact, topological invariants. On the other hand, the Dolbeault cohomol-
ogy groups depend essentially on the complex structure. This observation is at the
core of Hodge Theory.
3. Hermitian and Kähler metrics

Definition 3.1. Let M be a complex manifold and J its complex structure. A
Riemannian metric g on M is said to be a Hermitian metric if and only if for each
p ∈ M , the bilinear form gp on the tangent space Tp (M ) is compatible with the
complex structure Jp (cf. (A.22).
We recall from (A.23) in the Appendix that given a bilinear form compatible with
the complex structure we may define a J-invariant alternating form. Thus, given a
Hermitian metric on M we may define a differential two-form ω ∈ A2 (M, C) by:
(3.1) ω(X, Y ) := g(JX, Y ),
where we also denote by g the bilinear extension of g to the complexified tangent
space. By (A.24), we have
(3.2) ω ∈ A1,1 (M ) and ω̄ = ω.
We also recall that by Theorem A.8 every form ω as in (3.2) defines a symmetric
(1, 1) tensor on M compatible with J and a Hermitian form H on the complex vector
space (Tp (M ), J).
We express these objects in local coordinates: let (U, {z1 , . . . , zn }) be local com-
plex coordinates on M , then (3.2) implies that we may write
i
n
(3.3) ω := hjk dzj ∧ dz̄k ; hkj = h̄jk .
2
j,k=1
Hence ω(∂/∂zj , ∂/∂ z̄k ) = (i/2) hjk from which it follows that
ω(∂/∂xj , ∂/∂xk ) = −Im(hjk ).
Moreover, it follows from (A.25) that

g(∂/∂xj , ∂/∂xk ) = ω(∂/∂xj , ∂/∂yk ) = Re(hjk ).
Thus,
H(∂/∂xj , ∂/∂xk ) = hjk ,
where H is the Hermitian form on (Tp (M ), Jp ) defined as in (A.26). Hence g, and
consequently H, is positive definite if and only if the Hermitian matrix (hjk ) is
positive definite. We may then restate Definition 3.1 by saying that a Hermitian
structure is a (1, 1) real form ω as in (3.3) such that the matrix (hjk ) is positive
definite. By abuse of notation we will say that, in this case, the two-form ω is
positive.
3.1. Kähler Manifolds.
Definition 3.2. A Hermitian structure on a manifold M is said to be a Kähler
metric if and only if the two-form ω is closed. We will say that a complex manifold
is Kähler if and only if it admits a Kähler structure and refer to ω as a Kähler form.
Exercise 21. Let (M, ω) be a Kähler manifold. Prove that there exist local coframes
χ1 , . . . , χn ∈ A1,0 (U ) such that

n
ω = (i/2) χj ∧ χ̄j .
j=1
Proposition 3.3. Every Kähler manifold M is symplectic.

Proof. Recall that every complex manifold is orientable and that if {z1 , . . . , zn } are
local coordinates on M then we may assume that the frame
{∂/∂x1 , ∂/∂y1 , . . . , ∂/∂xn , ∂/∂yn }
is positively oriented.
Now, if ω is a Kähler form on M then
n
n
i
ω n
= n! det((hij )) (dzj ∧ dz̄j )
2
j=1

n
= n! det((hij )) (dxj ∧ dyj ),
j=1
since dzj ∧ dz̄j = (2/i)dxj ∧ dyj . Therefore, ω n is never zero.

Exercise 22. Prove that ω n /n! is the volume element of the Riemannian metric g
defined by the Kähler form ω.
Proposition 3.3 immediately gives a necessary condition for a compact complex
manifold to be Kähler:
Proposition 3.4. If M is a compact Kähler manifold then
dim H 2k (M, R) > 0,
for all k = 0, . . . , n.
24 EDUARDO CATTANI
Proof. Indeed, this is true of all compact symplectic manifolds as the forms ω k ,
k = 1, . . . , n, induce non-zero de Rham cohomology classes. Suppose, otherwise,
that ω k = dα, then
ω n = d(ω n−k ∧ α).
But then it would follow from Stokes’ Theorem that

ωn = 0
M
which contradicts the fact that ω n is a non-zero multiple of the volume element.
Remark 7. As we will see below, the existence of a Kähler metric on a manifold
imposes many other topological restrictions beyond those satisfied by symplectic
manifolds. The earliest examples of compact symplectic manifolds with no Kähler
structure are due to Thurston [30]. We refer to [36] for further details.
Example 3.5. Affine space Cn with the Kähler form:
i
n
ω = dzj ∧ dz̄j
2
j=1
is a Kähler manifold. The form ω gives the usual symplectic structure on R2n .
The following theorem states that, locally, a Kähler metric agrees with the Eu-
clidean metric up to order two.
Theorem 3.6. Let M be a complex manifold and g a Kähler metric on M . Then,
given p ∈ M there exist local coordinates (U, {z1 , . . . , zn }) around p such that zj (p) =
0 and
i
n
ω = hjk dzj ∧ dz̄k ,
2
j=1
where the coefficients hjk are of the form
(3.4) hjk (z) = δjk + O(||z||2 ).
Proof. We refer to [34, Proposition 3.14] for a proof.
Example 3.7. We will now construct a Kähler form on Pn . We will do this by
exhibiting a positive, real, closed (1, 1)-form on Pn , the resulting metric is called the
Fubini-Study metric on Pn .
Given z ∈ Cn+1 we denote by
||z||2 = |z0 |2 + · · · + |zn |2 .
Let Uj ⊂ Pn be the open set (1.6) and let ρj ∈ C ∞ (Uj ) be the positive function
||z||2
(3.5) ρj ([z]) := ,
|zj |2
and define ωi ∈ A1,1 (Uj ) by
−1 ¯
(3.6) ωj := ∂ ∂ log(ρj ).
2πi
Clearly, ωj is a real two-form. Moreover, on Uj ∩ Uk , we have

log(ρj ) − log(ρk ) = log |zk |2 − log |zj |2 = log(zk z̄k ) − log(zj z̄j ).
¯
Hence, since ∂ ∂(log(z j z̄j )) = 0, we have that ωj = ωk on Uj ∩ Uk . Thus, the forms
ωj piece together to give us a global, real, (1, 1)-form on Pn :
−1 ¯
(3.7) ω = ∂ ∂ log(||z||2 ).
2πi
Moreover, it is clear from the definition of ω that dω = 0.
It remains to show that ω is positive. We observe first of all that the expression
(3.7) shows that if A is a unitary matrix and μA : Pn → Pn is the biholomorphic map
μA ([z]) := [A · z] then μ∗A (ω) = ω. Hence, since given any two points [z], [z ] ∈ Pn
there exists a unitary matrix such that μa ([z]) = [z ], it suffices to prove that ω
is positive definite at just one point, say [1, 0, . . . , 0] ∈ U0 . In the coordinates
{u1 , . . . , un } in U0 , we have ρ0 (u) = 1 + ||u||2 and therefore:

n
n
¯
∂(log ρ0 (u)) = ρ−1
0 (u) uk ∂¯ūk = ρ−1
0 (u) uk dūk ,
k=1 k=1
⎛ ⎛ ⎞ ⎛ ⎞⎞
i −2
n
n n
ω = ρ (u) ⎝ρ0 (u) duj ∧ dūj + ⎝ ūj duj ⎠ ∧ ⎝ uk dūk ⎠⎠ .
2π 0
j=1 j=1 j=1
Hence, at the origin, we have
i
n
ω= duj ∧ dūj .
2π
j=1
which is a positive form.

The function log(ρj ) in the above proof is called a Kähler potential. As the
following result shows, every Kähler metric may be described by a (local) potential.
Proposition 3.8. Let M be a complex manifold and ω a Kähler form on M . Then
for every p ∈ M there exists an open set U ⊂ M and a real function v ∈ C ∞ (U )
¯
such that ω = i ∂ ∂(v).
Proof. Since dω = 0, it follows from the Poincaré Lemma that in a neighborhood
U of p, ω = dα, where α ∈ A1 (U , R). Hence, we maw write α = β + β̄, where
β ∈ A1,0 (U , R). Now, we can write:
ω = dα = ∂β + ∂β ¯ + ∂ β̄ + ∂¯β̄ ,
but, since ω is of type (1, 1) it follows that
ω = ∂β ¯ + ∂ β̄ and ∂β = ∂¯β̄ = 0.
¯
We may now apply the ∂-Poincaré Lemma to conclude that there exists a neigh-

borhood U ⊂ U around p where β̄ = ∂f ¯ for some (C-valued) C ∞ function f on U .
Hence:
ω = ∂∂¯ f¯ + ∂ ∂f
¯ = ∂ ∂(f¯ − f¯) = 2i ∂ ∂(Im(f
¯ )).

26 EDUARDO CATTANI
Theorem 3.9. Let (M, ω) be a Kähler manifold and suppose N ⊂ M is a complex

submanifold, then (N, ω|N ) is a Kähler manifold.
Proof. Let g be the J-invariant Riemannian metric on M associated with ω. Then
g restricts to an invariant Riemannian metric on N whose associated two-form is
ω|N . Since d(ω|N ) = (dω)|N = 0 it follows that N is a Kähler manifold as well.
It follows from Theorem 3.9 that a necessary condition for a compact complex
manifold M to have an embedding in Pn is that there exists a Kähler metric on M .
Moreover, as we shall see below, for a submanifold of projective space there exists a
Kähler metric whose associated cohomology satisfies a suitable integrality condition.
3.2. The Chern Class of a Holomorphic Line Bundle. The construction of

the Kähler metric in Pn may be further understood in the context of hermitian
metrics on (line) bundles. We recall that a Hermitian metric on a C-vector bundle
π : E → M is given by a positive definite Hermitian form
H p : Ep × Ep → C
on each fiber Ep which is smooth in the sense that given sections σ, τ ∈ Γ(U, E), the
function
H(σ, τ )(p) := Hp (σ(p), τ (p))
is C ∞ on U . In particular, a Hermitian metric on a complex manifold is equivalent
to a Hermitian metric on the holomorphic tangent bundle (or on the complexified
tangent bundle). Using partitions of unity one can prove that every smooth vector
bundle E has a Hermitian metric H.
In the case of a line bundle L, the Hermitian form Hp is completely determined
by the value Hp (v, v) on a non-zero element v ∈ Lp . In particular, if {(Uα , Φα )} is a
cover of M by trivializing neighborhoods of L and σα ∈ O(Uα , L) is the local frame
σα (x) = Φ−1
α (x, 1) ; x ∈ Uα ,
then a Hermitian metric H on L is determined by the collection of positive functions:
ρα := H(σα , σα ) ∈ C ∞ (Uα ).
we note that if Uα ∩ Uβ = ∅ then we have σβ = gαβ · σα and, consequently, the
functions ρα satisfy the compatibility condition:
(3.8) ρβ = |gαβ |2 ρα .
In particular, if L is a holomorphic line bundle then the transition functions gαβ are
holomorphic and we have, as in Example 3.7 that
∂ ∂¯ log(ρα ) = ∂ ∂¯ log(ρβ )
on Uα ∩ Uβ and therefore the form
1 ¯
(3.9) ∂ ∂ log(ρα )
2πi
is a global, real, (1, 1) form on M . The cohomology class
(3.10) [(1/2πi) ∂ ∂¯ log(ρα )] ∈ H 2 (M, R)
is called the Chern class of the vector bundle L and denoted by c(L). In fact, the
factor 1/2π factor is chosen so that the Chern class is actually an integral cohomology
class:
(3.11) c(L) ∈ H 2 (M, Z).
Recall that if gαβ are the transition functions for a bundle L then the functions
−1
gαβ are the transition functions of the dual bundle L∗ . In particular, if ρα are
a collection of positive C ∞ functions defining a Hermitian metric on L then the
functions ρ−1 ∗ ∗ ∗
α define a Hermitian metric, H on L . We call H the dual Hermitian
metric. We then have:
(3.12) c(L∗ ) = −c(L).
Definition 3.10. A holomorphic line bundle L → M over a compact Kähler mani-
fold is said to be positive if and only if there exists a Hermitian metric H on L for
which the (1, 1) form (3.10) is positive. We say that L is negative if its dual bundle
L∗ is positive.
We note that in Example 3.7 we have:
|zk |2 ρk ([z]) = |zj |2 ρj ([z])
on Uj ∩ Uk . Hence
2
zj
ρk ([z]) = ρj ([z])

zk
and, by (3.8), it follows that the functions ρj define a Hermitian metric on the
tautological bundle O(−1). Hence taking into account the sign change in (3.6) it
follows that for the Kähler class of the Fubini-study metric agrees with the Chern
class of the hyperplane bundle O(1).

i ¯
(3.13) c(O(1)) = [ω] = ∂ ∂ log(||z||2 ) .
2π
Hence, the hyperplane bundle O(1) is a positive line bundle. Moreover, if M ⊂ Pn
is a complex submanifold then the restriction of O(1) to M is a positive line bundle
over M . We can now state:
Theorem 3.11 (Kodaira Embedding Theorem). A compact complex manifold
M may be embedded in Pn if and only if it there exists a positive holomorphic line
bundle π : L → M .
We refer to [34, Theorem 7.11], [25, Theorem 8.1], [15], [39, Theorem 4.1], and
[18, §5.3] for various proofs of this theorem.
Remark 8. The existence of a positive holomorphic line bundle π : L → M implies
that M admits a Kähler metric whose Kähler class is integral. Conversely, any
integral cohomology class represented by a closed (1, 1) form is the Chern class of a
line bundle (cf. [10, §6]), hence a compact complex manifold M may be embedded
in Pn if and only if it admits a Kähler metric whose Kähler class is integral.
28 EDUARDO CATTANI
When the Kodaira Embedding Theorem is combined with Chow’s Theorem that
asserts that every analytic subvariety of Pn is algebraic we obtain a characteriza-
tion of complex projective varieties as those compact Kähler manifolds admitting a
Kähler metric whose Kähler class is integral.
4. Harmonic Forms - Hodge Theorem

4.1. Compact Real Manifolds. Throughout §4.1 we will let M denote a compact,
oriented, real, n-dimensional manifold with a Riemannian metric g. We recall that
the metric on the tangent bundle T M induces a dual metric on the cotangent bundle
T ∗ M by the condition that if X1 , . . . , Xn ∈ Γ(U, T M ) are a local orthonormal frame
on U then the dual frame ξ1 , . . . , ξn ∈ A1 (U ) is orthonormal as well. We will denote
the dual inner product by , .
Exercise 23. Verify that this metric on T ∗ M is well-defined. That is, it is inde-
pendent of the choice of local orthonormal frames.
We extend the inner product to the exterior bundles Λr (T ∗M ) by the specification
that the local frame
ξI := ξi1 ∧ · · · ξir ,
where I runs over all strictly increasing index sets {1 ≤ i1 < · · · < ir ≤ n}, is
orthonormal.
Exercise 24. Verify that this metric on T ∗ M is well-defined: i.e., it is independent
of the choice of local orthonormal frames, by proving that:
α1 ∧ · · · ∧ αr , β1 ∧ · · · ∧ βr = det(αi , βj ),
where αi , βj ∈ A1 (U ).
Hint: Use the Cauchy-Binet formula for determinants.
Recall that given an oriented Riemannian manifold, the volume element is defined
as the unique n-form Ω ∈ An (M ) such that
Ω(p)(v1 , . . . , vn ) = 1
for any positively oriented ortonormal basis {v1 , . . . , vn } of Tp (M ). If ξ1 , . . . , ξn ∈
A1 (U ) is a positively oriented orthonormal coframe then
ΩU = ξ1 ∧ · · · ∧ ξn .
Exercise 25. Prove that in local coordinates, the volume element may be written
as √
Ω = G dx1 ∧ · · · ∧ dxn ,
where G = det(gij ) and
gij := g(∂/∂xi , ∂/∂xj ).
We now define the Hodge ∗-operator: Let α1 , . . . , αn ∈ Tp∗ (M ) be a positively
oriented orthonormal basis, I = {1 ≤ i1 < · · · < ir ≤ n}, and I c the complementary
index set. Then we set
(4.1) ∗(αI ) = sign(I, I c ) αI c ,
where sign(I, I c ) is the sign of the permutation {I, I c }, and extend it linearly to an
operator:
(4.2) ∗ : Λr (Tp∗ (M )) → Λn−r (Tp∗ (M ))
Note that since ∗ maps an orthonormal basis to an orthonormal basis, it must
preserve the inner product.
Exercise 26. Verify that ∗ is well defined by proving that for α, β ∈ Λr (Tp∗ (M )):
(4.3) α ∧ ∗β = α, β Ω(p).
Exercise 27. Prove that ∗ is an isomorphism and that ∗2 acting on Λr (Tp∗ (M ))
equals (−1)r(n−r) id.
Suppose now that M is compact. We can then define an L2 -inner product on the
space of r-forms on M by:

(4.4) (α, β) := α ∧ ∗β = α(p), β(p) Ω ; α, β ∈ Ar (M ).
M M
Proposition 4.1. The bilinear form ( , ) is a positive definite inner product on
Ar (M ).
Proof. First of all we check that ( , ) is symmetric:

(β, α) = β ∧ ∗α = (−1) r(n−r)
∗(∗β) ∧ ∗α = ∗α ∧ ∗(∗β) = (α, β).
M M M
Now, given 0 = α ∈ Ar (M ), we have

(α, α) = α ∧ ∗α = α(p), α(p) Ω > 0
M M
since α(p), α(p) is a non-negative function which is not identically zero.
Proposition 4.2. The operator δ : Ar+1 (M ) → Ar (M ) defined by:
(4.5) δ := (−1)nr+1 ∗ d ∗
is the formal adjoint of d, that is:
(4.6) (dα, β) = (α, δβ) ; for all α ∈ Ar (M ), β ∈ Ar+1 (M ).
Proof.

(dα, β) = dα ∧ ∗β = d(α ∧ ∗β) − (−1) r
α∧d ∗β
M M

M
= −(−1)r (−1)r(n−r) α ∧ ∗(∗ d ∗ β) = α ∧ ∗δβ

M M
= (α, δβ).

Remark 9. Note that if dim M is even then δ = − ∗ d ∗ independently of the degree
of the form. Since we will be interested in applying these results in the case of
complex manifolds which, as real manifolds, are even-dimensional we will make that
assumption from now on.
30 EDUARDO CATTANI
We now define the Laplace-Beltrami operator of (M, g) by

Δ : Ar (M ) → Ar (M ) ; Δ(α) := dδα + δdα.
Proposition 4.3. The operators d, δ, ∗ and Δ satisfy the following properties:
i) Δ is self-adjoint; i.e. (Δα, β) = (α, Δβ).
ii) [Δ, d] = [Δ, δ] = [Δ, ∗] = 0.
iii) Δ(α) = 0 if and only if dα = δα = 0.
Proof. We leave the first two items as exercises. Note that given operators D1 , D2 ,
the bracket [D1 , D2 ] = D1 ◦ D2 − D2 ◦ D1 . Thus, ii) states that the Laplacian Δ
commutes with d, δ, and ∗.
Clearly, if dα = δα = 0 we have Δα = 0. Conversely, uppose α ∈ Ar (M ), and
Δα = 0 then
0 = (Δα, α) = (dδα + δdα, α) = (δα, δα) + (da, da).
Hence dα = δα = 0.
Definition 4.4. A form α ∈ Ar (M ) is said to be harmonic if Δα = 0 or, equiva-
lently, if α is closed and co-closed, i.e. δα = 0.
Exercise 28. Let M be a compact, connected, oriented, Riemannian manifold.
Show that the only harmonic functions on M are the constant functions.
Exercise 29. Let α ∈ Ar (M ) be closed. Show that ∗α is closed if and only if α is
harmonic.
The following result shows that harmonic forms are very special within a given
de Rham cohomology class:
Proposition 4.5. A closed r-form α is harmonic if and only if ||α||2 is a local
minimum within the de Rham cohomology class of α. Moreover, in any given de
Rham cohomology class there is at most one harmonic form.
Proof. Let α ∈ Ar (M ) be such that ||α||2 is a local minimum within the de Rham
cohomology class of α. Then, for every β ∈ Ar−1 (M ), the function ν(t) := ||α +
t dβ||2 has a local minimum at t = 0. In particular,
ν (0) = 2(a, dβ) = 2(δα, β) = 0 for all β ∈ Ar−1 (M ).
Hence, δα = 0 and α is harmonic. Now, if α is harmonic, then
||α + dβ||2 = ||α||2 + ||dβ||2 + 2(α, dβ) = ||α||2 + ||dβ||2 ≥ ||α||2
and equality holds only if dβ = 0. This proves the uniqueness statement.
Hodge’s theorem asserts that, in fact, every de Rham cohomology class contains
a (unique) harmonic form. More precisely,
Theorem 4.6 (Hodge Theorem). Let Hr (M ) denote the vector space of harmonic
r-forms on M . Then:
i) Hr (M ) is finite dimensional for all r.
ii) We have the following decomposition of the space of r forms:

Ar (M ) = Δ(Ar (M )) ⊕ Hr (M )
= dδ(Ar (M )) ⊕ δd(Ar (M )) ⊕ Hr (M )
= d(Ar−1 (M )) ⊕ δ(Ar+1 (M )) ⊕ Hr (M ).
Proof. The proof of this theorem involves the theory of elliptic differential operators
on a manifold. We refer to [15, Chapter 0], [37, Chapter 6] and [39, Chapter 4] for
proofs of this important result.
Since d and δ are formal adjoints of each other it follows that

(ker(d), Im(δ)) = (ker(δ), Im(d)) = 0
and, consequently, if α ∈ Z r (M ) and we write
α = dβ + δγ + μ ; β ∈ Ar−1 (M ), γ ∈ Ar+1 (M ), μ ∈ Hr (M ),
then
0 = (α, δγ) = (δγ, δγ)
and therefore δγ = 0. Hence, [α] = [μ]. By the uniqueness statement in Proposi-
tion 4.5 we get:
(4.7) H r (M, R) ∼
= Hr (M ).
Corollary 4.7. Let M be a compact, oriented, n-dimensional manifold. Then
H r (M, R) is finite-dimensional for all r.
Corollary 4.8 (Poincaré Duality). Let M be a compact, oriented, n-dimensional
manifold. Then the bilinear pairing:

(4.8) : H r (M, R) × H n−r (M, R) → R
M

that maps (α, β) → M α ∧ β is non-degenerate. Hence
n−r ∗
H (M, R) ∼
= H r (M, R).
Proof. We may assume without loss of generality that M is a Riemannian mani-
fold. Then, the Hodge star operator commutes with the Laplacian and defines an
isomorphism:
Hr (M ) ∼
= Hn−r (M ).
Hence if 0 = α ∈ Hr (M ) we have ∗α ∈ Hn−r (M ) and

α ∧ ∗α = (α, α) = 0.
M

Exercise 30. Prove that the pairing (4.8) is well-defined.
32 EDUARDO CATTANI
4.2. Compact Hermitian Manifolds. We now consider the case of a compact,

complex, n-dimensional manifold M with a Hermitian metric. First of all, we will be
interested in considering complex valued forms Ar (M, C). We extend the ∗-operator
by linearity to Λr (Tp∗ (M ) ⊗R C) and extending the inner product , on Λr (Tp∗ (M )
to the complexification as a Hermitian inner product , h we get:
α ∧ ∗β̄ = α, βh Ω(p)

for α, β ∈ Λr (Tp∗ (M )⊗R C). Hence we get a positive definite Hermitian inner product
on Ak (M, C) by

(4.9) h
(α, β) := α, β Ω =
h
α ∧ ∗β̄.
M M
Exercise 31. Prove that ∗ maps (p, q)-forms to forms of bidegree (n − q, n − p).
Exercise 32. Let M be a compact, complex, n-dimensional manifold and α ∈
A2n−1 (M, C). Prove that

∂α = ¯ = 0.
∂α
M M
Proposition 4.9. The operator ∂∗:= − ∗ ∂¯ ∗ (resp. ∂¯∗ := − ∗ ∂ ∗) is the formal

¯ relative to the Hermitian inner product ( , )h . The operator
adjoint of ∂ (resp. ∂)
∗ ¯∗
∂ (resp. ∂ ) is of bidegree (−1, 0) (resp. (0, −1)).
¯ the proof
Proof. Given Exercise 32 and the Leibniz property for the operators ∂, ∂,
of the first statement is analogous to that of Proposition 4.2. The second statement
follows from Exercise 31. The details are left as an exercise.
We can now define Laplace-Beltrami operators:

(4.10) Δ∂ = ∂∂ ∗ + ∂ ∗ ∂ ; Δ∂¯ = ∂¯∂¯∗ + ∂¯∗ ∂¯
Note that the operators Δ∂ and Δ∂¯ are of bidegree (0, 0); i.e. they map forms
of bidegree (p, q) to forms of the same bidegree. In particular, if α ∈ Ak (U ) is
decomposed according to (2.6) as
α = αk,0 + αk−1,1 + · · · + α0,k ,

then Δ∂¯(α) = 0 if and only if Δ∂¯(αp,q ) = 0 for all p, q.
The operators Δ∂ and Δ∂¯ are elliptic and, consequently, the Hodge Theorem
remains valid for them. Thus if we set:
(4.11) H∂p,q
¯ (M ) := {α ∈ A (M ) : Δ∂¯(α) = 0},
p,q
we have
H∂p,q ∼ p,q
(4.12) ¯ (M ) = H (M ).
5. The Hodge Decompositon Theorem

In this section we will show that on a compact Kähler manifold the Laplacians Δ
and Δ∂¯ are multiples of each other. Indeed, we have:
Theorem 5.1. Let M be a compact Kähler manifold. Then:
(5.1) Δ = 2Δ∂¯.
This theorem has a remarkable consequence: Suppose α ∈ Hk (M ) is decomposed
according to (2.6) as
α = αk,0 + αk−1,1 + · · · + α0,k ,
then since Δ = 2Δ∂¯, the form α is Δ∂¯-harmonic and, consequently, the components
αp,q are Δ∂¯-harmonic and, hence, Δ-harmonic as well. Therefore, if we set for
p + q = k:
(5.2) Hp,q (M ) := Hk (M, C) ∩ Ap,q (M ),
we get

(5.3) Hk (M, C) ∼
= Hp,q (M ).
p+q=k
Moreover, since Δ is a real operator, it follows that

(5.4) Hq,p (M ) = Hp,q (M ).
If we combine these results with the Hodge Theorem we get:
Theorem 5.2 (Hodge Decomposition Theorem). Let M be a compact Kähler
manifold and let H p,q (M ) be the space of de Rham cohomology classes in H p,q (M, C)
that have a representative of bidegree (p, q). Then,
(5.5) H p,q (M ) ∼
= H (M ) ∼
p,q
∂¯
= Hp,q (M ),
and

(5.6) H k (M, C) ∼
= H p,q (M ).
p+q=k
Moreover, H q,p (M ) = H p,q (M ).

Remark 10. In view of Definition A.3, Theorem 5.2 may be restated as: The
de Rham cohomology groups H k (M, R) have a Hodge structure of weight k with
(H(M, C))p,q ∼ p,q
= H∂¯ (M ).
We will denote by hp,q = dimC H p,q (M ). These are the so-called Hodge numbers of
M . Note that the Betti numbers bk , that is, the dimension of the k-th cohohomology
space are given by:

(5.7) bk = hp,q .
p+q=k
In particular, the Hodge Decomposition Theorem implies a new restriction on the

cohomology of a compact Kähler manifold:
Corollary 5.3. The odd Betti numbers of a compact Kähler manifold are even.
34 EDUARDO CATTANI
Proof. This assertion follows from (5.7) together with the fact that hp,q = hq,p .
Remark 11. Thurston’s examples of complex symplectic manifolds with no Kähler
structure are manifolds which do not satisfy Corollary 5.3.
Remark 12. As pointed out in Example 18, the de Rham cohomology H ∗ (M, C) is
an algebra under the cup product. We note that the Hodge decomposition (5.6) is
compatible with the algebra structure in the sense that

(5.8) H p,q ∪ H p ,q ⊂ H p+p ,q+q .
This additional topological restriction for a compact, complex, symplectic manifold
to have a Kähler metric has been successfully exploited by C. Voisin [36] to obtain
remarkable examples of non-Kähler, symplectic manifolds.
Let M be a compact, n-dimensional Kähler manifold and X ⊂ M a complex
submanifold of codimension k. We may define a linear map:

2(n−k)
(5.9) :H (M, C) → C ; [α] → α|X .
X X
This map defines an element in (H 2(n−k) (M, C))∗
and, therefore, by Corollary 4.8
a cohomology class ηX ∈ H 2k (M, C) defined by the property that for all [α] ∈
H 2(n−k) (M, C):

(5.10) α ∧ ηX = α|X .
M X
The class ηX is called the Poincaré dual of X and one can show that:
(5.11) ηX ∈ H k,k (M ) ∩ H 2k (M, Z).
One can also prove that the construction of the Poincaré dual may be extended to
singular analytic subvarieties (cf. [15, 18]).
Conjecture 5.4 (Hodge Conjecture). Let M be a smooth, projective manifold.
Then
H k,k (M, Q) := H k,k (M ) ∩ H 2k (M, Q)
is generated, as a Q-vector space, by the Poincaré duals of analytic subvarieties of
M.
The Hodge Conjecture is one of the remaining six Clay Millenium Problems [11].
5.1. Kähler identities.
Definition 5.5. Let (M, ω) be an n-dimensional, compact, Kähler manifold. We
define:
(5.12) Lω : Ak (M, C) → Ak+2 (M, C) ; Lω (α) = ω ∧ α ,

Y : A∗ (M, C) → A∗ (M, C), where A∗ (M, C) = 2nk=0 A (M, C) by
k
(5.13) Y (α) = (n − k)α for α ∈ Ak (M, C) ,

and we let N+ denote the formal adjoint of Lω with respect to the Hermitian form
( , )h ; i.e. if α ∈ Ak (M, C),
N+ (α) = (−1)k ∗ Lω ∗ α.
Remark 13. If there is no chance of confusion we will write L for Lω . It is clear,

however, that this operator, as well as its formal adjoint depend on the choice of a
Kähler form ω.
Exercise 33. Prove that if α ∈ Ak (M, C) and β ∈ Ak+2 (M, C), then
(L(α), β)h = (−1)k (α, ∗L ∗ β)h
Proposition 5.6. The operators L, Y , N+ satisfy the following properties.
i) L, Y , and N+ are real.
ii) L is an operator of bidegree (−1, −1); i.e.
L(Ap,q (M )) ⊂ Ap−1,q−1 (M ).
iii) Y is an operator of bidegree (0, 0) and N+ has bidegree (1, 1).
iv) They satisfy the commutativity relations:
[Y, L] = −2L ; [Y, N+ ] = 2N+ ; [N+ , L] = Y .
Proof. All of the listed properties, with the exception of the last commutation re-
lation, are easy to verify. We note that the three operators are C ∞ (U ) linear and
hence are defined pointwise. Thus, the verification of the last identity is a purely
linear algebra statement. We refer to [18, Proposition 1.2.26] for the details.
As noted above, the operators L, Y , N+ are defined pointwise and, because of iv)
in Proposition 5.6 they define an sl2 -triple on the exterior algebra Λ∗ (Tp∗ (M )). Thus,
their action on Λ∗ (Tp∗ (M )) is described by Theorem A.15 and Proposition A.17. Al-
though the vector space A∗ (M, C) is infinite dimensional we still obtain the Lefschetz
decomposition from its validity at each point p ∈ M :
Theorem 5.7. Let (M, ω) be an n-dimensional Kähler manifold and let be such
that 0 ≤ ≤ n. Set
(5.14) P (M, C) := {α ∈ A (M, C) : Ln−+1 (α) = 0}.
Then
i) P (M, C) = {α ∈ A (M, C) : N+ (α) = 0}.
ii) A (M, C) = P (M, C) ⊕ L(A−2 (M, C)).
The following result describe the Kähler identities which describe the commuta-
tivity relations among the differential operators d, ∂, ∂¯ and the Lefschetz operators
L, Y, N+ .
Theorem 5.8 (Kähler identities). Let (M, ω) be a compact, Kähler manifold.
Then the following identities hold:
¯ L] = [∂ ∗ , N+ ] = [∂¯∗ , N+ ] = 0.
i) [∂, L] = [∂,
ii) [∂¯∗ , L] = i∂ ; [∂ ∗ , L] = −i∂¯ ; [∂,

¯ N+ ] = i∂ ∗ ; [∂, N+ ] = −i∂¯∗
Proof. We refer to [18, Proposition 3.1.12] for a full proof. It is clear that it suffices
to prove one of the identities in each item since the others follow from conjugation
or the adjoint property of these operators. We have
[∂, L]α = ∂(Lα) − L(∂α) = ∂(ω ∧ α) − ω ∧ ∂α = (∂ω) ∧ α = 0,
36 EDUARDO CATTANI
¯ which implies that each summand vanishes.

since 0 = dω = ∂ω + ∂ω
The second item is harder to prove and follows from a very clever use of the
Lefschetz decomposition and the fact that {L, Y, N+ } are an sl2 -triple.
Remark 14. An alternative way of proving the second set of identities makes use
of the fact hat these identities are local and only involve the coefficients of the
Kähler metric up to first order. On the other hand, Theorem 3.6 asserts that a
Kähler metric agrees with the standard Hermitian metric on Cn . Thus it suffices to
verify the identities in that case. This is done by a direct computation. This is the
approach in [15] and [34, Proposition 6.5].
We now show how Theorem 5.1 follows from the Kähler identities: Note first of
all that ii) in Theorem 5.8 yields:
i(∂ ∂¯∗ + ∂¯∗ ∂) = ∂[N+ , ∂] + [N+ , ∂]∂ = ∂N+ ∂ − ∂N+ ∂ = 0.
Therefore,
Δ∂ = ∂∂ ∗ + ∂ ∗ ∂ = i ∂[N+ , ∂]
¯ + i [N+ , ∂]∂
¯
= i (∂N+ ∂¯ − ∂ ∂N
¯ + + N+ ∂∂¯ − ∂N¯ + ∂)

= i ([∂, N+ ]∂¯ + N+ ∂ ∂)
¯ − ∂ ∂N
¯ + + N+ ∂∂ ¯ − (∂[N ¯ +)
¯ + , ∂] + ∂∂N

= i N+ (∂ ∂¯ + ∂∂)
¯ + (∂ ∂¯ + ∂∂)N
¯ ¯ ¯∗ ¯∗ ¯
+ − i (∂ ∂ + ∂ ∂)
= Δ∂¯.
These two identities together give the assertion of Theorem 5.1, that is:
Δ = 2Δ∂ = 2Δ∂¯.
5.2. Lefschetz Theorems. We will now show how the Kähler identities imply that
the Laplace-Beltrami operator Δ commutes with the sl2 -representation and conse-
quently, we get a (finite-dimensional) sl2 -representation on the space of harmonic
forms H∗ (M ).
Theorem 5.9. Let (M, ω) be a Kähler manifold. Then, Δ commutes with L, N+
and Y .
Proof. Clearly [Δ, L] = 0 if and only if [Δ∂ , L] = 0. We have:
[Δ∂ , L] = [∂∂ ∗ + ∂ ∗ ∂, L]
= ∂∂ ∗ L − L∂∂ ∗ + ∂ ∗ ∂L − L∂ ∗ ∂
= ∂ ([∂ ∗ , L] + L∂ ∗ ) − L∂∂ ∗ + ([∂ ∗ , L] + L∂ ∗ ) ∂ − L∂ ∗ ∂
= −i∂ ∂¯ − i∂∂ ¯
= 0
A similar argument yields that [Δ, N+ ] = 0. Clearly, [Δ, Y ] = 0.
We can now define an sl2 -representation on the de Rham cohomology of a compact
Kähler manifold:
Theorem 5.10. The operators L, Y , and N+ define a real representation of sl(2, C)
on the de Rham cohomology H ∗ (M, C). Moreover, these operators commute with the
Weil operators of the Hodge structures on the subspaces H k (M, R).
Proof. This is a direct consequence of Theorem 5.9. The last statement follows from
the fact that L, Y , and N+ are of bidegree (1, 1), (0, 0) and (−1, −1), respectively.

Corollary 5.11 (Hard Lefschetz Theorem). Let (M, ω) be an n-dimensional,
compact Kähler manifold. For each k ≤ n the map
(5.15) Lkω : H n−k (M, C) → H n+k (M, C)
is an isomorphism.
Proof. This follows from the results in A.5. Indeed, we know from (A.34) that the
weight filtration of Lω is given by:

Wk = Wk (Lω ) = Ej (Y ).
j≤k
Hence, we have using (5.13):

GrkW ∼
= Ek (Y ) = H n−k (X, C),
and the result follows from the definition of the weight filtration in Proposition A.12.

We note, in particular that for j ≤ k ≤ n, the maps
Lj : H n−k (M, C) → H n−k+2j (M, C)
are injective. This observation together with the Hard Lefschetz Theorem imply
further cohomological restrictions on a compact Kähler manifold:
Theorem 5.12. The Betti and Hodge numbers of a compact Kähler manifold satisfy:
i) bn−k = bn+k ; hp,q = hq,p = hn−q,n−p = hn−p,n−q .
ii) b0 ≤ b2 ≤ b4 ≤ · · ·
iii) b1 ≤ b3 ≤ b5 ≤ · · ·
In both cases the inequalities continue up to, at most, the middle degree.
Definition 5.13. Let (M, ω) be a compact, n-dimensional Kähler manifold. For
each k = p + q ≤ n, we define the primitive cohomology spaces:
(5.16) H0p,q (M ) := ker{Ln−k+1
ω : H p,q (M ) → H n−q+1,n−p+1 (M )}

(5.17) H0k (M ) := H0p,q (M ).
p+q=k
We now have
Theorem 5.14 (Lefschetz Decomposition). Let (M, ω) be an n-dimensional,
compact Kähler manifold. For each k = p + q ≤ n, we have
(5.18) H p,q (M ) = H0p,q (M ) ⊕ Lω (H p−1,q−1 (M )),
(5.19) H k (M, C) = H0k (M, C) ⊕ Lω (H k−2 (M, C)).

38 EDUARDO CATTANI
5.3. Hodge-Riemann Bilinear Relations. The following result, whose proof

may be found in [18, Proposition 1.2.31] relates the Hodge star operator with the
sl2 -action.
Proposition 5.15. Let α ∈ P k (M, C), then:
j!
(5.20) ∗Lj (α) = (−1)k(k+1)/2 · Ln−k−j (C(α)),
(n − k − j)!
where C is the Weil operator in Ak (M, C).
Definition 5.16. Let (M, ω) be an n-dimensional, compact, Kähler manifold. Let
k be such that 0 ≤ k ≤ n. We define a bilinear form
Qk = Q : H k (M, C) × H k (M, C) → C,

(5.21) Qk (α, β) := (−1) k(k+1)/2
α ∧ β ∧ ω n−k .
M
Exercise 34. Prove that Q is well defined independent of our choice of represen-
tative in the cohomology class. This justifies our using simply α to denote the
cohomology class [α].
Theorem 5.17. The bilinear form Q satisfies the following properties:
i) Qk is symmetric if k is even and skew-symmetric if k is odd.
ii) Q(Lω α, β) + Q(α, Lω β) = 0; we say that Lω is an infinitesimal isomorphism
of Q.

iii) Q(H p,q (M ), H p ,q (M )) = 0 unless p = q and q = p.
p,q
iv) If 0 = α ∈ H0 (M ) then
(5.22) Q(Cα, ᾱ) > 0.
Proof. The first statement is clear. For the second note that the difference between
the two terms is the preceding sign which changes as we switch from k + 2 to k. The
third assertion follows from the fact that the integral vanishes unless the bidegree
of the integrand is (n, n) and, for that to happen, we must have p = q and q = p.
Therefore, we only need to show the positivity condition iv). Let α ∈ H0p,q (M ).
It follows from Proposition 5.15 that
(−1)k(k+1)/2 ω n−k ∧ ᾱ = ∗−1 (n − k)! C(ᾱ).
On the other hand, on H k (M ), we have C 2 = (−1)k id = ∗2 and therefore:

Q(Cα, ᾱ) = α ∧ ∗ᾱ = (α, α)h > 0.
M

Properties iii) and iv) in Theorem 5.17 are called the first and second Hodge-
Riemann bilinear relations. In view of Definition A.9 we may say that the Hodge-
Riemann bilinear relations amount to the statement that the Hodge structure in the
primitive cohomology H0k (M, R) is polarized by the “intersection” form Q defined
by (5.21).
Example 5.18. Let X = Xg denote a compact Riemann surface of genus g. Then

we know that H 1 (X, Z) ∼
= Z2g . The Hodge decomposition in degree 1 is of the form:
H 1 (X, C) = H 1,0 (X) ⊕ H 1,0 (X),
where H 1,0 (X) consists of the one-forms on X which, locally, are of the form f (z) dz,
with f (z) is holomorphic. The form Q on H 1 (X, C) is alternating and given by:

Q(α, β) = α ∧ β.
The Hodge-Riemann bilinear relations then take the form: Q(H 1,0 (X), H 1,0 (X)) = 0
and, since H01,0 (X) = H 1,0 (X),

iQ(α, ā) = i α ∧ ᾱ > 0
X
if α is a non-zero form in H 1,0 (X). Note that, locally,
iα ∧ ᾱ = i|f (z)|2 dz ∧ dz̄ = 2|f (z)|2 dx ∧ dy,
so both bilinear relations are clear in this case. We note that it follows that H 1,0 (X)
defines a point in the complex manifold D = D(H 1 (X, R), Q) defined in Exam-
ple 1.16.
Example 5.19. Suppose now that (M, ω) is a compact, connected, Kähler surface
and let us consider the Hodge structure in the middle cohomology H 2 (X, R). We
have the Hodge decomposition:
H 2 (X, C) = H 2,0 (X) ⊕ H 1,1 (X) ⊕ H 0,2 (X) ; H 0,2 (X) = H 0,2 (X).
Moreover, H02,0 (X) = H 2,0 (X) while
H 1,1 (X) = H01,1 (X) ⊕ Lω H 0,0 (X) = H01,1 (X) ⊕ C · ω,
and
H01,1 (X) = {α ∈ H 1,1 (X) : [ω ∧ α] = 0}.
The intersection form on H 2 (X, R) is given by

Q(α, β) = − α ∧ β,
X
and the second Hodge-Riemann bilinear relation become:

α ∧ ᾱ > 0 if 0 = α ∈ H 2,0 (X),
X

ω 2 > 0,
X
β ∧ β̄ < 0, if 0 = β ∈ H01,1 (X).

X
We note that the first two statements are easy to verify, but that is not the case
with the last one. We point out that the integration form I(α, β) = −Q(α, β) has
index (+, · · · , +, −) in H 1,1 (X)∩H 2 (X, R); i.e. I is a hyperbolic symmetric bilinear
form. Such forms satisfy the reverse Cauchy-Schwarz inequality:
(5.23) I(α, β)2 ≥ I(α, α) · I(β, β),
40 EDUARDO CATTANI
provided that I(α, α) > 0.

The inequality (5.23) is called Hodge’s inequality and plays a central role in the
study of algebraic surfaces. Via Poincaré duals it may be interpreted as an inequal-
ity between intersection indexes of curves in an algebraic surface or, in other words,
about the number of intersection points between two curves. If the ambient surface is
an algebraic torus, X = C∗ × C∗ , then a curve zero-locus of a Laurent polynomial in
two variables and a classical result of Bernstein-Kushnirenko-Khovanskii shows that,
generically on the coefficients of the polynomials, the intersection indexes may be
computed combinatorially from the Newton polytope of the defining polynomials (cf.
Khovanskii’s Appendix in [6] for a full account of this circle of ideas). This relation-
ship between the Hodge inequality and combinatorics led Khovanskii and Teissier
[29] to give (independent) proofs of the classical Alexandrov-Fenchel inequality for
mixed volumes of polytopes using the Hodge inequality and set the basis for a fruitful
interaction between algebraic geometry and combinatorics. In particular, motivated
by problems in convex geometry, Gromov [16] stated a generalization of the Hard
Lefschetz Theorem, Lefschetz decomposition and Hodge-Riemann bilinear relation
to the case of “mixed” Kähler forms. We give a precise statement in the case of the
Hard Lefschetz Theorem and refer to [31, 32, 13, 9] for further details.
A Kähler class is a real, (1, 1) form satisfying a positivity condition. Those forms
define a cone K ⊂ H 1,1 (M ) ∩ H 2 (M, R). We have:
Theorem 5.20 (Mixed Hard Lefschetz Theorem). Let M be a compact, n-
dimensional, Kähler manifold. Let ω1 , . . . , ωk ∈ K, 1 ≤ k ≤ n. Then the map
Lω1 · · · Lωk : H n−k (M, C) → H n+k (M, C)
is an isomorphism.
As mentioned above this result was originally formulated by Gromov who proved
it in the (1, 1) (note that the operators involved preserve the Hodge decomposition).
Later, Timorin [31, 32] proved it in the linear algebra case and in the case of simplicial
toric varieties. Dinh and Nguyen [13] proved it in the form stated above. In [9] the
author gave a proof in the context of variations of Hodge structure which unifies
those previous results as well as similar results in other contexts [19, 4].
Appendix A. Linear Algebra

A.1. Real and Complex Vector Spaces. Here we will review some basic facts
about finite-dimensional real and complex vector spaces that are used throughout
these notes.
We begin by recalling the notion of “complexification” of a real vector space.
Given a vector space V over R we denote by
(A.1) VC := V ⊗R C.
We can formally write v ⊗ (a + ib) = av + b(v ⊗ i), a, b ∈ R, and setting iv := v ⊗ i
we may write VC = V ⊕ iV . Scalar multiplication by complex numbers is then given
by:
(a + ib)(v1 + iv2 ) = (av1 − bv2 ) + i(av2 + bv1 ) ; v1 , v2 ∈ V ; a, b ∈ R.
Note that dimR V = dimC VC and, in fact, if {e1 , . . . , en } is a basis of V (over R)
then {e1 , . . . , en } is also a basis of VC (over C). Clearly (Rn )C ∼
= Cn .
The usual conjugation of complex numbers induces a conjugation operator on VC :
σ(v ⊗ α) := v ⊗ ᾱ ; v ∈ V, α ∈ C,
or, formally, σ(v1 + iv2 ) = v1 − iv2 , v1 , v2 ∈ V . Clearly for w ∈ VC , we have that
w ∈ V if and only if σ(w) = w. If there is no possibility of confusion we will write
σ(w) = w̄, w ∈ VC .
Conversely, if W is a complex vector space over C, then W = VC for a real vector
space V if and only if W has a conjugation σ; i.e. a map σ : W → W such that σ 2
is the identity, σ is additive-linear, and
σ(αw) = ᾱσ(w) ; w ∈ W ; α ∈ C.
The set of fixed points V := {w ∈ W : σ(w) = w} is a real vector space and W = VC .
We call V a “real form” of W .
If V, V are real vector spaces we denote by HomR (V, V ) the vector space of
R-linear maps from V to V . It is easy to check that

(A.2) HomR (V, V ) C ∼
= HomC (VC , VC )
and that if σ, σ are the conjugation operators on VC and VC respectively, then the
conjugation operator on HomC (VC , VC ) is given by:

(A.3) σHom (T ) = σ ◦ T ◦ σ ; T ∈ HomR (V, V ) C ,
or in more traditional notation:
(A.4) T̄ (w) = T (w̄) ; w ∈ VC .
Thus, the group of real automorphisms of V may be viewed as the subgroup:
GL(V ) = {T ∈ GL(VC ) : σ ◦ T = T ◦ σ} ⊂ GL(VC ).
If we choose V = R, then (A.2) becomes
(A.5) (V ∗ )C ∼
= (VC )∗ ,
42 EDUARDO CATTANI
where, as always, V ∗ = HomR (V, R) and (VC )∗ = HomC (VC , C) are the dual vector
spaces. Thus, we may drop the parenthesis and write simply VC∗ . Note that for
α ∈ VC∗ , its conjugate ᾱ is defined by
ᾱ(w) = α(w̄) ; w ∈ VC .
We may similarly extend the notion of complexification to the tensor products
a times b times

(A.6) T a,b
(V ) := V ⊗ · · · ⊗ V ⊗ V ∗ ⊗ · · · ⊗ V ∗ ,
and to the exterior algebra: Λr (V ∗ ) and we have

T a,b (V ) ∼
= T a,b (VC ) ; (Λr (V ∗ ))C ∼
= Λr (VC∗ ).
C
In particular, a tensor B ∈ T 0,2 (V ), which defines a bilinear form
B: V × V → R
may be viewed as an element in T 0,2 (VC ) and defines a bilinear form
B : VC × VC → C
satisfying B̄ = B. Explicitly, given v1 , v2 ∈ V we set:
B(iv1 , v2 ) = B(v1 , iv2 ) = iB(v1 , v2 )
and extend linearly. A bilinear form B : VC × VC → C is real if and only if
B(w, w ) = B(w̄, w̄ ),
for all w, w ∈ VC . Similarly, thinking of elements α ∈ Λr (V ∗ ) as alternating multi-
linear maps
r times

α: V × · · · × V → R
we may view them as alternating multilinear maps
α : VC × · · · × VC → R
satisfying
α(w1 , . . . , wr ) = α(w̄1 , . . . , w̄r ),
for all w1 , . . . , wr ∈ VC .
On the other hand, given a C-vector space W we may think of it as a real vector
space simply by “forgetting” that we are allowed to multiply by complex numbers
and restricting ourselves to multiplication by real numbers (this procedure is called
“restriction of scalars”). To remind ourselves that we are only able to multiply by
real numbers we write W R when we are thinking of W as a real vector space. Note
that
dimR (W R ) = 2 dimC (W ),
and that if {e1 , . . . , en } is a C-basis of W then {e1 , . . . , en , ie1 , . . . , ien } is a basis of
W R.
It is now natural to ask when a real vector space V is obtained from a complex
vector space W by restriction of scalars. Clearly, a necessary condition is that
dimR (V ) be even. But, there is additional structure on V = W R coming from the
fact that W is a C-vector space. Indeed, multiplication by i in W induces an R-linear

map:
(A.7) J : WR → WR ; J(w) := i w,
satisfying J 2 = −I, where I denotes the identity map.
Conversely, let V be a 2n-dimensional real vector space and J : V → V a linear
map such that J 2 = −I. Then we may define a C-vector space structure on V by:
(A.8) (a + ib) ∗ v := a v + b J(v).
We say that J is a complex structure on V and we will often denote by (V, J)
the complex vector space consisting of the points in V endowed with the scalar
multiplication† (A.8).
Exercise 35. Let V be a real vector space and J : V → V a linear map such that
J 2 = −I. Prove that there exists a basis {e1 , . . . , en , f1 , . . . , fn } of V such that the
matrix of J in this basis is of the form:

0 −In
(A.9) J = ,
In 0
where In denotes the (n × n)-identity matrix.
Proposition A.1. Let V be a real vector space. Then the following are equivalent:
i) V has a complex structure J.
ii) The complexification VC admits a decomposition
(A.10) VC = W + ⊕ W −
where W± ⊂ VC are complex subspaces such that W± = W∓ .
Proof. Suppose J : V → V is a linear map such that J 2 = −I. Then we may
extend J to a map J : VC → VC . Now, endomorphisms of complex vector spaces are
diagonalizable and since J 2 = −I, the only possible eigenvalues for J are ±i. Let
W± denote the ±i-eigenspace of J. We then have:
VC = W + ⊕ W − .
Suppose now that w ∈ W± , then since J is a real map we have
J(w̄) = J(w) = ±i w = ∓i w.
Hence W± = W∓ and we obtain the decomposition (A.10).
Conversely, given the decomposition (A.10) we define a linear map J : VC → VC
by the requirement that J(w) = ±iw if w ∈ W± . It is easy to check that J 2 = −I
and that the assumption that W± = W∓ implies that J is a real map; i.e. J¯ = J.
Proposition A.2. Let V be a real vector space with a complex structure J. Then,
the map φ : (V, J) → W+ defined by φ(v) = v − iJ(v) is an isomorphism of complex
vector spaces.
†
We will use ∗ to denote complex multiplication in (V, J) to distinguish from the notation λv,
λ ∈ C which is traditionally used to represent the point (v ⊗ λ) ∈ VC . We will most often identify
(V, J) with a complex subspace of VC as in Proposition A.1 and therefore there will be no chance
of confusion.
44 EDUARDO CATTANI
Proof. We verify first of all that φ(v) ∈ W+ ; that is J(φ(v)) = iφ(v):

J(φ(v)) = J(v − iJ(v)) = J(v) − iJ 2 (v) = J(v) + iv = i(v − iJ(v)) = iφ(v).
Next we check that the map is C-linear. Let a, b ∈ R, v ∈ V :
φ((a + ib) ∗ v) = φ(av + bJ(v)) = (av + bJ(v)) − iJ(av + bJ(v))
= (av + bJ(v)) + i(bv − aJ(v)) = (a + ib)(v − iJ(v))
= (a + ib)φ(v).
We leave it to the reader to verify that if w ∈ W+ then w = φ( 12 (w + w̄)) and,
therefore, φ is an isomorphism.
Suppose now that (V, J) is a 2n-dimensional real vector space with a complex
structure J and let T ∈ GL(V ). Then T is a complex linear map if and only if
T (iv) = iT (v), i.e. if and only if:
(A.11) T ◦ J = J ◦ T.
Exercise 36. Let V be a real vector space and J : V → V a complex structure on
V . Prove an R-linear map T : V → V is C-linear if and only if the matrix of T ,
written in terms of a basis as in Exercise 35, is of the form:

A −B
(A.12) ,
B A
where A, B are (n × n)-real matrices.
If T ∈ GL(V ) satisfies (A.11) then the extension of T to the complexification VC
and continues to satisfy the commutation relation (A.11). In particular, such a map
T must preserve the eigenspaces of J : VC → VC . Now, if {e1 , . . . , en , f1 , . . . , fn } is a
a basis of V as in Exercise 35, then wi = 12 (ei − iJei ) = 12 (ei − ifi ), i = 1, . . . , n, are
a basis of W+ and the conjugates w̄i = 12 (ei + iJei ) = 12 (ei − ifi ) i = 1, . . . , n, are a
basis of W− . In this basis, the extension of T to VC is written as:

A + iB 0
(A.13) .
0 A − iB
We note in particular that if T ∈ GL(V ) satisfies (A.11) then det(T ) > 0. In-
deed, the determinant is unchanged after complexification and in terms of the basis
{w1 , . . . , wn , w̄1 , . . . , w̄n } the matrix of T is as in (A.13) and we have
(A.14) det(T ) = | det(A + iB)|2 .
If J is a complex structure on the real vector space V then the dual map J ∗ : V ∗ →
V ∗ is a complex structure on the dual space V ∗ . The corresponding decomposition
(A.10) on the complexification VC∗ is given by
(A.15) VC∗ = W+⊥ ⊕ W−⊥ ,
where W±⊥ := {α ∈ VC∗ : α|W± = 0}. Indeed, if α ∈ VC∗ is such that J ∗ (α) = iα and
w ∈ W− then we have:
iα(w) = (J ∗ (α))(w) = α(J(w)) = α(−iw) = −iα(w)
which implies that α(w) = 0. The statement now follows from dimensional reasons.
A.2. Hodge structures. The decomposition of the complexification of a real vector

space defined by a complex structure is a simple but important example of a Hodge
structure.
Definition A.3. A (real) Hodge structure of weight k ∈ Z consists of:
i) A finite-dimensional real vector space V .
ii) A decomposition of the complexification VC as:

(A.16) VC = V p,q ; V q,p = V p,q .
p+q=k
We say that the Hodge structure is rational (resp. integral) if there exists a rational
vector space VQ (resp. a lattice VZ ) such that V = VQ ⊗Q R (resp. V = VZ ⊗Z R).
The following statement is valid for Hodge structures defined over R, Q, or Z.
Proposition A.4. Let V and W be vector spaces with Hodge structures of weight
k, respectively. Then Hom(V, W ) has a Hodge structure of weight − k.
Proof. We set
(A.17) Hom(V, W )a,b := {X ∈ HomC (VC , WC ) : X(V p,q ) ⊂ W p+a,q+b } .
Then Hom(V, W )a,b = 0 unless p + a + q + b = , that is unless a + b = − k. The
rest of the verifications are left to the reader.
In particular, if we choose W = R with the Hodge structure WC = W 0,0 , we see
that if V has a Hodge structure of weight k, then V ∗ has a Hodge structure of weight
−k. Similarly, if V and W have Hodge structures of weight k, respectively, then
V ⊗R W ∼ = HomR (V ∗ , W ) has a Hodge structure of weight k + and

(A.18) (V ⊗ W )a,b = V p,q ⊗C W r,s .
p+r=a
q+s=b
Needless to say, we could take (A.18) as our starting point rather than (A.17). Note
that if V has a Hodge structure of weight k then the tensor product T a,b (V ) defined
in (A.6) has a Hodge structure of weight k(a − b).
Example A.5. Let V be a real vector space with a complex structure J and consider
the complex structure J ∗ on the dual vector space V ∗ . We may define a Hodge
structure of weight 1 on V ∗ by setting (V ∗ )1,0 = W+⊥ and (V ∗ )0,1 = W−⊥ in the
decomposition (A.15). The exterior algebra Λk (V ∗ ) now inherits a Hodge structure
of weight k, where for p + q = k:
p times q times

∗
(A.19) Λ p,q
(V ) = (Λk
(VC∗ ))p,q = (V ∗ )1,0 ∧ . . . ∧ (V ∗ )1,0 ∧ (V ∗ )0,1 ∧ . . . ∧ (V ∗ )0,1 .
There are two alternative ways of describing a Hodge structure on a vector space
V that will be very useful to us.
Definition A.6. A real (resp. rational, integral) Hodge structure of weight k ∈ Z
consists of a real vector space V (resp. a rational vector space VQ , a lattice VZ ) and
a decreasing filtration
· · · F p ⊃ F p−1 · · ·
46 EDUARDO CATTANI
of the complex vector space VC = V ⊗R C (resp. VC = VQ ⊗Q C, VC = VZ ⊗Z C) such

that:
(A.20) VC = F p ⊕ F k−p+1 .
The equivalence of Definitions A.3 and A.6 is easy to verify. Indeed given a
decomposition as in (A.16) we set:

Fp = H a,k−a ,
a≥p
while given a filtration of VC satisfying (A.20) the subspaces

H p,q = F p ∩ F q ; p + q = k.
define a decomposition of VC satisfying (A.16).
In order to state the third definition of a Hodge structure, we need to recall some
basic notions from representation theory. Let us denote by S(R) the real algebraic
group:
a −b
S(R) := ∈ GL(2, R) .
b a
Then C∗ ∼= S(R) via the identification

a −b
z = a + ib → .
b a
The circle group S 1 = {z ∈ C ∗ : |z| = 1} is then identified with the group of
rotations:
cos θ − sin θ
; θ∈R .
sin θ cos θ
Recall that a representation of an algebraic group G defined over the field F =
Q, R, or C, on a F -vector space VF is a group homomorphism ϕ : G → GL(VF ).
Now, if V is a real vector space with a Hodge structure of weight k, then we may
define a representation of S(R) on VC by:

ϕ(z)(v) := z p z̄ q vp,q ,
p+q=k

where v = vp,q is the decomposition of v according to (A.16). We verify that
ϕ(z) ∈ GL(V ), i.e. that ϕ(z) = ϕ(z):

ϕ(z)(v) = ϕ(z)(v̄) = z p z̄ q vq,p
p+q=k
since (v̄)p,q = vq,p . Hence ϕ(z)(v) = ϕ(z)(v). We note also that for λ ∈ R∗ ⊂ C∗ ,
ϕ(λ)(v) = λk v for all v ∈ V .
Conversely, it follows from the representation theory of S(R) that every every finite
dimensional representation of S(R) on a complex vector space splits as a direct sum
of one-dimensional representations where z ∈ S(R) acts as multiplication by z p z̄ q ,
with p, q ∈ Z. Hence, a representation ϕ : S(R) → GL(VC ) defined over R (i.e.
ϕ̄ = ϕ) decomposes VC into subspaces V p,q

(A.21) VC = V p,q ; V q,p = V p,q ,
where ϕ(z) acts as multiplication by z p z̄ q . Note, moreover, that if λ ∈ R∗ ⊂ S(R)

then ϕ(λ) acts on V p,q as multiplication by λp+q . Thus, the following definition is
equivalent to Definitions A.3 and A.6.
Definition A.7. A real Hodge structure of weight k ∈ Z consists of a real vector
space V and a representation ϕ : S(R) → GL(V ) such that ϕ(λ)(v) = λk v for all
v ∈ V and all λ ∈ R∗ ⊂ S(R).
Given a Hodge structure ϕ of weight k on V , the linear operator ϕ(i) : VC → VC
is called the Weil operator and denoted by C. Note that on V p,q , the Weil operator
acts as multiplication by ip−q and, consequently, if J is a complex structure on V
then for the Hodge structure of weight k defined on the exterior product Λk (V ∗ ),
the Weil operator agrees with the natural extension of J ∗ to Λk (V ∗ ).
Exercise 37. Prove that if (V, ϕ), (V , ϕ ) are Hodge structures of weight k and k
respectively, then (V ∗ , ϕ∗ ) and (V ⊗ V , ϕ ⊗ ϕ ) are the natural Hodge structures
on V ∗ and V ⊗ V defined above. Here ϕ∗ : S(R) → GL(V ∗ ) is the representation
ϕ∗ (z) := (ϕ(z))∗ and, similarly, (ϕ ⊗ ϕ )(z) := ϕ(z) ⊗ ϕ (z).
A.3. Symmetric and Hermitian Forms. If V is a real vector space with a com-
plex structure J and B : V × V → R is a bilinear form on V then we say that B is
compatible with J if and only if
(A.22) B(Ju, Jv) = B(u, v) for all u, v ∈ V.
We shall also denote by B the bilinear extension of B to the complexification VC . If
B is symmetric then the bilinear form on VC :
(A.23) ω(u, v) := B(Ju, v)
is alternating; i.e. ω ∈ Λ2 (VC∗ ). Indeed:
ω(u, v) = B(Ju, v) = B(J 2 u, Jv) = −B(u, Jv) = −ω(v, u).
We note that since B is real then so is ω and that, in fact,
(A.24) ω ∈ Λ1,1 (V ∗ ) ∩ Λ2 (VR ).
This follows from considering from the fact that the Weil operator on the Hodge
structure of Λ2 (V ∗ ) agrees with the operator defined by the complex structure J:
(Cω)(u, v) = ω(Ju, Jv) = B(J 2 u, Jv) = −B(u, Jv) = −ω(v, u) = ω(u, v).
Conversely, given an element ω ∈ Λ1,1 (V ∗ ) ∩ Λ2 (VR ), we have ω(Ju, Jv) = ω(u, v)
and we may define a bilinear symmetric form compatible with J by
(A.25) B(u, v) = ω(u, Jv).
We shall also be interested in Hermitian forms on a C-vector space W :
H: W × W → C.
Recall that for such a form,
H(λu + μu , v) = λH(u, v) + μH(u , v) ; u, u , v ∈ W ; λ, μ ∈ C, and
H(v, u) = H(u, v).
48 EDUARDO CATTANI
Now, if H is a Hermitian form and we write:

H(u, v) = S(u, v) − iA(u, v) ; S(u, v), A(u, v) ∈ R.
it is clear that S is a symmetric form and A is an alternating form.
Given a real vector space with a complex structure J and a compatible symmetric
bilinear form B : V ×V → R, we may define a Hermitian form on the complex vector
space (V, J) by:
(A.26) H(u, v) = B(u, v) − iω(u, v),
where ω is as in (A.23). Indeed, we only need to verify that H is C-linear on the
first argument but we have:
H(iu, v) = H(Ju, v) = B(Ju, v) − iω(Ju, v) = ω(u, v) + iB(u, v) = iH(u, v).
We collect these observations in the following:
Theorem A.8. Let V be a real vector space with a complex structure J then the
following data are equivalent:
i) A symmetric bilinear form B on V compatible with J.
ii) An element ω ∈ Λ1,1 (VC∗ ) ∩ Λ2 (V ∗ ).
iii) A Hermitian form H on the complex vector space (V, J)
Moreover, H is positive definite if and only if B is positive definite.
A.4. Polarized Hodge Structures.
Definition A.9. Let (V, ϕ) be a real Hodge structure of weight k. A polarization
of (V, ϕ) is a real† bilinear form Q : V × V → R such that
i) Q(u, v) = (−1)k Q(v, u); i.e., Q is symmetric or skew-symmetric depending
on whether k is even or odd.
ii) The Hodge decomposition is orthogonal relative to the Hermitian form
H : VC × VC → C defined by
(A.27) H(w1 , w2 ) := Q(C w1 , w̄2 ),
where C = ϕ(i) is the Weil operator.
iii) H is positive definite.
Remark 15. Note that if k is even then the Weil operator acts on V p,q as multipli-
cation by ±1 and then it is clear that the form H defined by (A.27) is Hermitian.
Similarly if k is odd since in this case C acts on V p,q as multiplication by ±i. We
also note that (ii) and (iii) above may be restated as follows:

ii’) Q(V p,q , V p ,q ) = 0 if p = k − p.
iii’) ip−q Q(w, w̄) > 0 for all 0 = w ∈ V p,q .
The statements ii’) and iii’) correspond to the Hodge-Riemann bilinear relations in
Theorem 5.17.
†
If (V, ϕ) is a rational (resp. integral) Hodge structure then we require Q to be defined over Q
(resp. over Z).
Example A.10. Let (V, J) be a real vector space with a complex structure and
let B be a positive definite symmetric bilinear form on V compatible with J. Let
VC = W+ ⊕ W− be the induced decomposition and consider the Hodge structure of
weight one on V defined by V 1,0 = W+ . The form Q(u, v) = B(u, Jv) is alternating
and for w = u − iJu, w = u − iJu ∈ W+ we have:
Q(u − iJu, u − iJu ) = Q(u, u ) − Q(Ju, Ju ) − iQ(u, Ju ) − iQ(Ju, u ) = 0.
Similarly, for 0 = u ∈ V :
iQ(u − iJu, u + iJu) = i (Q(u, u) + Q(Ju, Ju) + iQ(u, Ju) − iQ(Ju, u))
= −2Q(u, Ju) = 2B(u, u) > 0 .
Hence Q polarizes the Hodge structure defined by J.
A.5. The Weight filtration of a nilpotent transformation. In this section we
will construct a filtration canonically attached to a nilpotent linear transformation
and study its relationship with representations of the Lie algebra sl(2, R).
Throughout this section N : V → V will be a nilpotent linear transformation of
nilpotency index k; i.e. k the first positive integer such that N k+1 = 0. Given an
integer ≤ k, we will say that a subspace A ⊂ V is a Jordan block of weight m if
A has a basis {f0A , f1A , . . . , fm
A } such that N (f A ) = f A , where we set f A
i i+1 m+1 = 0. It
is convenient to reindex the basis as
eA A
m−2j := fj .
Note that if m is even then the index of eA takes even values from m to −m, while
if m is odd then it takes odd values from m to −m.
It is known from the study of the Jordan Normal Form for a nilpotent transfor-
mation N : V → V of a nilpotency index k, that we may decompose V as a direct
sum:
k
V = Um ,
m=0
where Um is the direct sum of all Jordan blocks of weight m. In fact, these subspaces
Um are unique. Clearly, each Um decomposes further as

m
Um = Um,m−2j ,
j=0
where Um,m−2j is the subspace spanned by all basis vectors eA

m−2j as A runs over all
Jordan blocks of weight m. We now define

k
(A.28) E = E (N ) = Um,
m=0
Theorem A.11. The decomposition (A.28) satisfies:

i) N (E ) ⊂ E−2 .
ii) For ≥ 0, N : E → E− is an isomorphism.
50 EDUARDO CATTANI
Proof. The first statement is clear since for any Jordan block A, N (fjA ) = fj+1A
which implies the assertion. Suppose now that is even. Then, E is spanned by all
basis vectors of the form eA A
= f(m−)/2 where A runs over all Jordan blocks of even
weight m ≥ .
Proposition A.12. Let N be a nilpotent transformation with nilpotency index k.
Then there exists a unique increasing filtration W = W (N ):
(A.29) {0} ⊂ Wk ⊂ W−k+1 ⊂ · · · ⊂ Wk−1 ⊂ Wk = V ,
with the following properties:
i) N (W ) ⊂ W−2 ,
ii) For ≥ 0 : N : GrW
→ Gr− , where Gr := W /W , is an isomorphism.
W W
Moreover, the subspaces W may be expressed in terms of ker(N a ), Im(N b ) and

hence are defined over Q (resp. over R) if N is.
Proof. The existence of W (N ) follows from Theorem A.11 while the uniqueness is a
consequence of the uniqueness properties of the Jordan decomposition. Alternatively
one may give an inductive construction of W (N ) as in [27, Lemma 6.4]. For an
explicit construction involving kernels and images of powers of N we refer to [28].
Example A.13. Suppose k = 1, then the weight filtration is of the form:
{0} ⊂ W−1 ⊂ W0 ⊂ W1 = V.
Since N : V /W0 → W−1 is an isomorphism it follows that
W−1 (N ) = Im(N ) ; W0 (N ) = ker(N ) .
Exercise 38. Prove that if k = 2 the weight filtration:
{0} ⊂ W−2 ⊂ W−1 ⊂ W0 ⊂ W1 ⊂ W2 = V
is given by:
{0} ⊂ Im(N 2 ) ⊂ Im(N ) ∩ ker(N ) ⊂ Im(N ) + ker(N ) ⊂ ker(N 2 ) ⊂ V
A.6. Representations of sl(2, C). We recall that the Lie algebra sl(2, C) consists
of all 2 × 2-complex matrices of trace zero. It has a basis consisting of

0 1 0 0 1 0
(A.30) n+ := ; n− := ; y :=
0 0 1 0 0 −1
This basis satisfies the commutativity relations:
(A.31) [y, n+ ] = 2n+ ; [y, n− ] = −2n− ; [n+ , n− ] = y ;
A representation ρ of sl(2, C) on a complex vector space VC is a Lie algebra
homomorphism
ρ : sl(2, C) → gl(VC ).
We denote the image of the generators of sl(2, C) by N+ , N− and Y . These elements
satisfy commutativity relations analogous to (A.31). Conversely, given elements
{N+ , N− , Y } ⊂ gl(V ) satisfying the commutation relations from (A.31) we can define
a representation ρ : sl(2, C) → gl(VC ). We will refer to {N+ , N− , Y } as an sl2 -triple.
A representation is called irreducible if V has no proper subspaces invariant under

ρ(sl(2, C)). We say that ρ is real if VC is the complexification of a real vector space
V and ρ(sl(2, R)) ⊂ gl(V ).
Example A.14. For each integer n we may define an irreducible representation ρn
on VC = Cn as follows. Suppose that n = 2k + 1, then we label the standard basis
of V as: {e−2k , e−2k+2 , . . . , e0 , . . . e2k−2 , e2k } and define:
(A.32) Y (e2j ) := 2j · e2j ; N− (e2j ) = e2j−2 ; N+ (e2j ) = μ2j · e2j+2 ,
where the integers μ2j are the unique solution to the recursion equations:
(A.33) μ2j−2 − μ2j = 2j ; μ−2k−2 = 0.
It is easy to check that the first two commutation relations are satisfied. On the
other hand,
[N+ , N− ](e2j ) = N+ (e2j−2 ) − N− (μ2j · e2j−2 ) = (μ2j−2 − μ2j ) · e2j = 2j · e2j = Y (e2j ).
Exercise 39. Find the solution to the equations (A.33).
Exercise 40. Extend the construction of the representation in Example A.14 to the
case n = 2k.
We note that for the representation ρn we have N−n = 0 and N−n−1 = 0; that
is, the index of nilpotency of N− (and of N+ ) is n − 1. At the same time, the
eigenvalues of Y range from −n + 1 to n − 1. We will refer to n − 1 as the weight
of the representation ρn . This notion of weight is consistent with that defined for
Jordan blocks above.
The basic structure theorem about representations of sl(2, C) is the following
Theorem A.15. Every finite dimensional representation of sl(2, C) splits as a di-
rect sum of irreducible representations. Moreover, an irreducible representation of
dimension n is isomorphic to ρn .
Proof. We refer to [39, Chapter V, Section 3] for a proof.
Suppose now that ρ : sl(2, C) → gl(V ) is a representation and set N = N− .

Let k be the unipotency index of N then ρ splits as a direct sum of irreducible
representations of weight at most k and we have:

(A.34) W (N ) = Ej (Y ).
j≤
Indeed, it is enough to check this statement for each irreducible representation and
there the statement is clear.
Exercise 41. Let N ∈ gl(V ) be nilpotent and Y ∈ gl(V ) be semisimple. Then the
following are equivalent.
i) There exists an sl2 -triple {N+ , N− , Y } with N− = N .
ii) [Y, N ] = −2N and the weight filtration of N is given by (A.34).
52 EDUARDO CATTANI
The above exercise implies that if N is a nilpotent element, W its weight filtration
and we {V } is a splitting of the filtration W in the sense that:
(A.35) W = V ⊕ W−1 ,
then if we define Y ∈ gl(V ) by Y (v) = v if v ∈ V , the pair {N, Y } may be extended
to an sl2 -triple. Moreover if N is defined over Q (resp. over R) and the splitting is
defined over Q (resp. over R) so is the sl2 -triple.
Exercise 42. Apply Exercise 41 to prove that if N is a nilpotent transformation,
then there exists an sl2 -triple with N = N− . This is a version of the Jacobson-
Morosov Theorem.
A.7. Lefschetz decomposition. Let N be a nilpotent transformation with nilpo-
tency index k. Then, for any , with 0 ≤ ≤ k we have
N : GrW → Gr−
W
is an isomorphism. We define
(A.36) P := ker{N +1 : GrW → Gr−−2
W
}
and call it the -th primitive space.
Example A.16. For any k we have Pk = GrkW since N k+1 = 0. Suppose k = 1
then, P0 = Gr0W = ker(N )/Im(N ).
Exercise 43. Let k = 2. Prove that P1 = Gr1W but that
P0 = ker(N )/(ker(N ) ∩ Im(N )) ⊂ Gr0N .
Proposition A.17. Let N ∈ gl(V ) be a nilpotent transformation with nilpotency
index k. Then for any sl2 -triple with N− = N we have
P := ker{N+ : GrW → Gr+2
W
}.
Moreover, for every , 0 ≤ ≤ k, we have:
(A.37) GrW = P ⊕ N (Gr+2
W
).
Proof. Let {N+ , N− , Y } with N− = N be an sl2 -triple with N− = N . Then P
is given by all eigenvectors of Y of eigenvalue living in the sum of irreducible
components of the representation of weight and this is are exactly the elements
by N+ . Similarly, it suffices to verify the decomposition (A.37) in each irreducible
component which is easy to do.
The decomposition (A.37), or more precisely, the decomposition obtained from
(A.37) inductively:
(A.38) GrW = P ⊕ N (P+2 ) ⊕ N 2 (P+4 ) + · · ·
is called the Lefschetz decomposition.
Example A.18. The only interesting term in the Lefschetz decomposition for k = 2
occurs for = 0, where, according to Exercises 38 and 43 we get:
Gr0W = P0 ⊕ N (P2 ) = ker(N )/(ker(N ) ∩ Im(N )) ⊕ N (V / ker(N 2 )).
References
[1] Werner Ballmann. Lectures on Kähler manifolds. ESI Lectures in Mathematics and Physics.
European Mathematical Society (EMS), Zürich, 2006.
[2] William M. Boothby. An introduction to differentiable manifolds and Riemannian geometry,
volume 120 of Pure and Applied Mathematics. Academic Press Inc., Orlando, FL, second edi-
tion, 1986.
[3] Raoul Bott and Loring W. Tu. Differential forms in algebraic topology, volume 82 of Graduate
Texts in Mathematics. Springer-Verlag, New York, 1982.
[4] Paul Bressler and Valery A. Lunts. Hard Lefschetz theorem and Hodge-Riemann relations for
intersection cohomology of nonrational polytopes. Indiana Univ. Math. J., 54(1):263–307, 2005.
[5] Jean-Luc Brylinski and Steven Zucker. An overview of recent advances in Hodge theory. In
Several complex variables, VI, volume 69 of Encyclopaedia Math. Sci., pages 39–142. Springer,
Berlin, 1990.
[6] Yu. D. Burago and V. A. Zalgaller. Geometric inequalities, volume 285 of Grundlehren der
Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-
Verlag, Berlin, 1988. Translated from the Russian by A. B. Sosinskiı̆, Springer Series in Soviet
Mathematics.
[7] James Carlson and Phillip Griffiths. What is ... a period domain? Notices Amer. Math. Soc.,
55(11):1418–1419, 2008.
[8] James Carlson, Stefan Müller-Stach, and Chris Peters. Period mappings and period domains,
volume 85 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cam-
bridge, 2003.
[9] Eduardo Cattani. Mixed Lefschetz theorems and Hodge-Riemann bilinear relations. Int. Math.
Res. Not. IMRN, (10):Art. ID rnn025, 20, 2008.
[10] Shiing-shen Chern. Complex manifolds without potential theory (with an appendix on the geom-
etry of characteristic classes). Universitext. Springer-Verlag, New York, second edition, 1995.
[11] Pierre Deligne. The Hodge conjecture. In The millennium prize problems, pages 45–53. Clay
Math. Inst., Cambridge, MA, 2006.
[12] Jean-Pierre Demailly. Complex analytic and algebraic geometry. OpenContent Book available
from http://www-fourier.ujf-grenoble.fr/ demailly/manuscripts/agbook.pdf.
[13] Tien-Cuong Dinh and Viêt-Anh Nguyên. The mixed Hodge-Riemann bilinear relations for
compact Kähler manifolds. Geom. Funct. Anal., 16(4):838–849, 2006.
[14] Phillip Griffiths. Hodge theory and geometry. Bull. London Math. Soc., 36(6):721–757, 2004.
[15] Phillip Griffiths and Joseph Harris. Principles of algebraic geometry. Wiley Classics Library.
John Wiley & Sons Inc., New York, 1994. Reprint of the 1978 original.
[16] M. Gromov. Convex sets and Kähler manifolds. In Advances in differential geometry and topol-
ogy, pages 1–38. World Sci. Publ., Teaneck, NJ, 1990.
[17] Richard Hain. Periods of limit mixed Hodge structures. In Current developments in mathemat-
ics, 2002, pages 113–133. Int. Press, Somerville, MA, 2003.
[18] Daniel Huybrechts. Complex geometry. Universitext. Springer-Verlag, Berlin, 2005. An intro-
duction.
[19] Kalle Karu. Hard Lefschetz theorem for nonrational polytopes. Invent. Math., 157(2):419–447,
2004.
[20] Kazuya Kato and Sampei Usui. Classifying spaces of degenerating polarized Hodge structures,
volume 169 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, 2009.
[21] Shoshichi Kobayashi. Differential geometry of complex vector bundles, volume 15 of Publications
of the Mathematical Society of Japan. Princeton University Press, Princeton, NJ, 1987. Kanô
Memorial Lectures, 5.
[22] Kunihiko Kodaira. Complex manifolds and deformation of complex structures. Classics in Math-
ematics. Springer-Verlag, Berlin, english edition, 2005. Translated from the 1981 Japanese
original by Kazuo Akao.
[23] John M. Lee. Introduction to smooth manifolds, volume 218 of Graduate Texts in Mathematics.
Springer-Verlag, New York, 2003.
54 EDUARDO CATTANI
[24] Andrei Moroianu. Lectures on Kähler geometry, volume 69 of London Mathematical Society
Student Texts. Cambridge University Press, Cambridge, 2007.
[25] James Morrow and Kunihiko Kodaira. Complex manifolds. AMS Chelsea Publishing, Provi-
dence, RI, 2006. Reprint of the 1971 edition with errata.
[26] Chris A. M. Peters and Joseph H. M. Steenbrink. Mixed Hodge structures, volume 52 of Ergeb-
nisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathe-
matics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in
Mathematics]. Springer-Verlag, Berlin, 2008.
[27] Wilfried Schmid. Variation of Hodge structure: the singularities of the period mapping. Invent.
Math., 22:211–319, 1973.
[28] Joseph Steenbrink and Steven Zucker. Variation of mixed Hodge structure. I. Invent. Math.,
80(3):489–542, 1985.
[29] Bernard Teissier. Variétés toriques et polytopes. In Bourbaki Seminar, Vol. 1980/81, volume
901 of Lecture Notes in Math., pages 71–84. Springer, Berlin, 1981.
[30] W. P. Thurston. Some simple examples of symplectic manifolds. Proc. Amer. Math. Soc.,
55(2):467–468, 1976.
[31] V. A. Timorin. Mixed Hodge-Riemann bilinear relations in a linear context. Funktsional. Anal.
i Prilozhen., 32(4):63–68, 96, 1998.
[32] V. A. Timorin. An analogue of the Hodge-Riemann relations for simple convex polyhedra.
Uspekhi Mat. Nauk, 54(2(326)):113–162, 1999.
[33] Loring W. Tu. An introduction to manifolds. Universitext. Springer, New York, 2008.
[34] Claire Voisin. Hodge theory and complex algebraic geometry. I, volume 76 of Cambridge Stud-
ies in Advanced Mathematics. Cambridge University Press, Cambridge, english edition, 2007.
Translated from the French by Leila Schneps.
[35] Claire Voisin. Hodge theory and complex algebraic geometry. II, volume 77 of Cambridge Stud-
ies in Advanced Mathematics. Cambridge University Press, Cambridge, english edition, 2007.
Translated from the French by Leila Schneps.
[36] Claire Voisin. Hodge structures on cohomology algebras and geometry. Math. Ann., 341(1):39–
69, 2008.
[37] Frank W. Warner. Foundations of differentiable manifolds and Lie groups, volume 94 of Grad-
uate Texts in Mathematics. Springer-Verlag, New York, 1983. Corrected reprint of the 1971
edition.
[38] André Weil. Introduction à l’étude des variétés kählériennes. Publications de l’Institut de
Mathématique de l’Université de Nancago, VI. Actualités Sci. Ind. no. 1267. Hermann, Paris,
1958.
[39] Raymond O. Wells, Jr. Differential analysis on complex manifolds, volume 65 of Graduate
Texts in Mathematics. Springer, New York, third edition, 2008. With a new appendix by Oscar
Garcia-Prada.
Department of Mathematics and Statistics, University of Massachusetts Amherst,

Amherst, MA 01003, USA
E-mail address: cattani@math.umass.edu

Cattani Kahler Manifolds

Uploaded by

Copyright:

Available Formats

Cattani Kahler Manifolds

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cattani Kahler Manifolds

Uploaded by

Copyright:

Available Formats

2150-4

Summer School and Conference on Hodge Theory and Related Topics

14 June - 2 July, 2010

Introduction to Kähler manifolds

where I c denotes the (n − k)-tuple of indices complementary to I. The maps φI

Theorem 1.7 (Holomorphic Inverse Function Theorem). Let F : U → V be

Exercise 6. Let P1 (z0 , . . . , zn ), . . . , Pm (z0 , . . . , zn ) be homogeneous polynomials.

Let {e1 , . . . , en , en+1 , . . . , e2n } be a basis of V in which the matrix of A is as in (1.4).

Example 1.17. Let TΛ := C/Λ, where Λ ⊂ Z2 is a rank-two lattice in C; i.e.

In particular, a holomorphic line bundle over M is given by a collection {Uα , gαβ },

satisfy (1.16) and deﬁne a new line bundle which we denote by L ⊗ L .

We may also extend the notion of sections to holomorphic vector bundles:

Note that given a section σ : M → E of a vector bundle E, the zero locus {x ∈

2. Differential Forms on Complex Manifolds

isothermal coordinates. Thus, a complex structure on an oriented, two-dimensional

Remark 5. Given local coordinates (U, {z1 , . . . , zn }) on M , a function f : U → C is

We will refer to Tp (M ) as the holomorphic tangent space† of M at p. We note

Note that this is well deﬁned since

We will denote by Ak (U ) (resp. Aa,b (U )) the C ∞ (U ) module of local sections of the

Proof. We refer to [15, Chapter 0] or [18, Corollary 1.3.9] for a proof.

3. Hermitian and Kähler metrics

Moreover, it follows from (A.25) that

Proposition 3.3. Every Kähler manifold M is symplectic.

since dzj ∧ dz̄j = (2/i)dxj ∧ dyj . Therefore, ω n is never zero.

Clearly, ωj is a real two-form. Moreover, on Uj ∩ Uk , we have

which is a positive form.

Theorem 3.9. Let (M, ω) be a Kähler manifold and suppose N ⊂ M is a complex

3.2. The Chern Class of a Holomorphic Line Bundle. The construction of

4. Harmonic Forms - Hodge Theorem

= −(−1)r (−1)r(n−r) α ∧ ∗(∗ d ∗ β) = α ∧ ∗δβ

We now deﬁne the Laplace-Beltrami operator of (M, g) by

ii) We have the following decomposition of the space of r forms:

Since d and δ are formal adjoints of each other it follows that

4.2. Compact Hermitian Manifolds. We now consider the case of a compact,

α ∧ ∗β̄ = α, βh Ω(p)

Proposition 4.9. The operator ∂∗:= − ∗ ∂¯ ∗ (resp. ∂¯∗ := − ∗ ∂ ∗) is the formal

We can now deﬁne Laplace-Beltrami operators:

α = αk,0 + αk−1,1 + · · · + α0,k ,

5. The Hodge Decompositon Theorem

Moreover, since Δ is a real operator, it follows that

Moreover, H q,p (M ) = H p,q (M ).

In particular, the Hodge Decomposition Theorem implies a new restriction on the

(5.13) Y (α) = (n − k)α for α ∈ Ak (M, C) ,

Remark 13. If there is no chance of confusion we will write L for Lω . It is clear,

ii) [∂¯∗ , L] = i∂ ; [∂ ∗ , L] = −i∂¯ ; [∂,

¯ which implies that each summand vanishes.

Hence, we have using (5.13):

(5.19) H k (M, C) = H0k (M, C) ⊕ Lω (H k−2 (M, C)).

5.3. Hodge-Riemann Bilinear Relations. The following result, whose proof

Example 5.18. Let X = Xg denote a compact Riemann surface of genus g. Then

β ∧ β̄ < 0, if 0 = β ∈ H01,1 (X).

provided that I(α, α) > 0.

Appendix A. Linear Algebra

fact that W is a C-vector space. Indeed, multiplication by i in W induces an R-linear

Proof. We verify ﬁrst of all that φ(v) ∈ W+ ; that is J(φ(v)) = iφ(v):

A.2. Hodge structures. The decomposition of the complexiﬁcation of a real vector

of the complex vector space VC = V ⊗R C (resp. VC = VQ ⊗Q C, VC = VZ ⊗Z C) such

while given a ﬁltration of VC satisfying (A.20) the subspaces

where ϕ(z) acts as multiplication by z p z̄ q . Note, moreover, that if λ ∈ R∗ ⊂ S(R)

Now, if H is a Hermitian form and we write:

α ∧ ∗β̄ = α, βh Ω(p)