. 21
( 44 .)


more conceptual. We notice that dim H i,j (M ) = dim H j,i (M ), hence
dimC H p (M, C) = 2 i<j dim H i,j (M ).
For the second proof we construct a non-degenerate bilinear form
Q on H p (M, R). It is enough to do this for p ¤ n since

dim H p (M, R) = dim H 2n’p (M, R).

Let ±, β be two closed di¬erential forms representing classes [±], [β]
∈ H p (M, R). We set

±§β§ω n’p .
Q([±], [β]) = ±

This form is skew-symmetric when p is odd and symmetric when p
is even. Assume for a moment that [β] is perpendicular to [±] for
any p-form ±. It follows that M ±§(β§ω n’p ) = 0 for any closed ±.
This can only happen if β § ω n’p represents zero cohomology class in
H 2n’p (M, R). In view of Hard Lefschetz [β] = 0 as well.

As a simple example, let us consider the case of M = C2 /Λ, where
Λ is a complete lattice. We will show how the Hard Lefschetz Theorem
works when applied to H 1 (M, C) ’ H 3 (M, C), where L = c(dz1 §d¯1 +

dz2 §d¯2 ), c = ’ 2 . The basis of H 1 (M, C) is constituted by

dz1 , dz2 , d¯1 , d¯2 .
z z

It is mapped by L to a basis

cdz1 §dz2 §d¯2 , cdz1 §dz2 §d¯1 , cd¯1 §dz2 §d¯2 , cdz1 §d¯2 §d¯1 .
z z z z z z

Let V be a representation of sl(2, C).
DEFINITION 3.6.21 The primitive cohomology P rim’k (V ) ‚ V ’k
consists of those vectors v of veight ’k which satisfy Y v = 0, or, equiv-
alently, X k+1 v = 0.

PROPOSITION 3.6.22 For any V we have

V ’k = P rim’k (V ) • X · P rim’k’2 (V ) • X 2 · P rim’k’4 (V ) • · · · .

Proof. Straightforward.

Now we apply this Proposition to the cohomology of a compact
K¨hler manifold M . We de¬ne

Hprim (M )=Ker[Λ : H k (M, C) ’ H k’2 (M, C)] =

= Ker[Ln’k+1 : H k (M, C) ’ H 2n’k+2 (M, C)].
We get

H k (M, C) = Hprim (M ) • L · Hprim (M ) • L2 · Hprim • · · · .
k k’2 k’4

This allows us to conclude that

dim H n (M, C) ≥ dim H n’2 (M, C) ≥ dim H n’4 (M, C) ≥ · · ·


dim H n’1 (M, C) ≥ dim H n’3 (M, C) ≥ dim H n’5 (M, C) ≥ · · · .

EXAMPLE of a non-closed holomorphic 1-form ω on a compact com-
plex manifold M . This example is also important because it is simple
and it immediately provides us with a compact complex manifold M
which is not K¨hler because we proved before that on K¨hler manifolds
a a
every global holomorphic k-form is in fact closed (and never exact un-
less it is the zero form). Let H be the complex Heisenberg group of
3 — 3 matrices of the following form
« 
¬ ·
0 1 y  , x, y, z ∈ C.
We also think of x, y, z as complex coordinates de¬ning the complex
manifold structure on this complex a¬ne non-commutative Lie group.
The group multiplication is just matrix multiplication and is given by
« «  « 
1xz 1ac 1 a + x c + xb + z
¬ ·¬ · ¬
y + b ·.
0 1 y · 0 1 b = 0 1 
001 001 0 0 1

If such two elements of H commute, then xb = ya and hence the center
of H is isomorphic to C and consists of those matrices which have
x = y = 0.
LEMMA 3.6.23 The form ω = dz ’ ydx is a right-invariant non-
closed holomorphic 1-form on H.
To proof this assertion we simply check that the e¬ect of translation of
ω by an element « 
¬ ·
0 1 b ∈ H
transforms it into
d(z + c + bx) ’ (b + y)d(a + x) = dz + bdx ’ bdx ’ ydx = ω.
Also we see that
dω = ’dy§dx = dx§dy = 0.

Now we de¬ne M as the quotient H/“, where “ is the following
discrete cocompact subgroup in H.
« 
1ac √
¬ ·
“ = { 0 1 b  , a, b, c ∈ Z[ ’1]},

where Z[ ’1] is the ring of so-called Gaussian integers, i.e. the complex

numbers of the form n + ’1m, m, n ∈ Z. To explain the compactnes
of H/“ we just point out the following nice ¬bration
« 
√ ¬ ·
C/Z[ ’1] ’ H/“ 0 1 y
“ “

(C/Z[ ’1])2 (x, y)
Therefore, due to the fact that ω is right-invariant, it descends to a
non-closed holomorphic 1 form on H/“ - a compact complex manifold,
which is therefore not K¨hler .

3.7 Hodge Conjecture
A Hodge structure of weight j (over R) consists of a real vector space
V together with a decomposition of its complexi¬cation
V — C = •p+q=j V p,q
such that V p,q = V q,p .
Obviously, the group H j (M, R) for a compact K¨hler manifold M
has a Hodge structure of weight j. In this situation we would like to
give an intrinsic description of groups H p,q ‚ H p+q (M, C), i.e. such
a description which does not use metric or K¨hler form, but only the
complex structure on M .
LEMMA 3.7.1 The group H p,q (M ) consists of those classes of com-
plex coe¬cient closed p + q-forms on M that have representatives of
type (p, q).
Proof. What really needs to be shown is that if ± is a closed form
on M of pure type (p, q) then there exists a harmonic form β on M
of type (p, q) such that ± ’ β is an exact form. In this case we call
β the harmonic representative of ±. The space Hp+q (M ) of harmonic
forms of total degree p + q is a ¬nite-dimensional Hilbert subspace of
the pre-Hilbert space Ap+q (M ) of closed (p + q)-forms on M . We note
that the orthogonal space to Hj in the space of closed j-forms is the
space of exact j-forms. Indeed if ± is harmonic and β is exact, say
β = dγ, then
±, β = ±, dγ = δ±, γ = 0.
Since the dimension of Hj (M ) is equal to the codimension of the exact
j-forms in the closed j-forms, the statement follows. Thus we can
de¬ne the orthogonal projection Q : Ap+q (M ) ’ Hp+q (M ). Now we
let β = Q(±). Then if ± is closed, ± and β are cohomologous. Since we
have an orthogonal decomposition
Aj (M ) = •p+q=j Ap,q (M )

“ •Qp,q

Hj (M ) = •p+q=j Hp+q (M )

It follows that if ± is of type (p, q) then β is of type (p, q) as well.

It is very important to underline the algebraic structure on coho-
mology of M .
LEMMA 3.7.2 The wedge product of di¬erential forms induces cup-
product on cohomology

H j (M, C) — H k (M, C) ’ H j+k (M, C),

H p,q (M ) — H r,s (M ) ’ H p+r,q+s (M ).
In this way we obtain a bi-graded algebra.
Proof. The statement follows from a simple observation that if ± is a
closed form of type (p, q) and β is a closed form of type (r, s) then ±§β
is a closed form of type (p + r, q + s).

Whenever we encounter some interesting phenomena concerning al-
gebraic structures on cohomology, we wish we were able to give a nice
geometric interpretation to them. In certain situations it is really possi-
ble to do so. For instance one can interpret the fact that when M = CPn
(n > 1) the group H 2p (CPn , C) coincides with its subgroup H p,p just by
simple dimension calculation, or, we can use our previous knowledge
that the form ω§ · · · §ω is harmonic and of type (p, p). In general,
classes of type (p, p) are very important, because they are intimately
related with (complex) analytic cycles (which are complex analytic sub-
varieties). Let X ‚ M be a complex analytic subvariety (possibly sin-
gular) of codimension d (M as before is a compact K¨hler manifold of
dimension n). There is a well-de¬ned cohomology class
[X] ∈ H 2d (M, Z) ’ H 2d (M, C),
which is de¬ned using the Poincar´ duality H i (M, Z)
e H2n’i (M, Z).
The fact that this class is still de¬ned when X is singular is a remarkable
theorem of Bloom-Herrera.
[X] ∈ H d,d (M )

Proof. The Poincar´ duality pairing between H 2d (M, C) and H 2n’2d (M, C)
respects the Hodge decompositions

H 2d (M, C) = •p H p,2d’p

H 2n’2d (M, C) = •q H q,2n’2d’q .
We claim that H p,2d’p is perpendicular to H q,2n’2d’q unless p + q = n.
Indeed, if ±, β are of types (p, 2d ’ p) and (q, 2n ’ 2d ’ q) respectively,
then ±§β is of type (p + q, 2n ’ p ’ q) and falls out of the picture
if p + q = n. Thus we coclude that H p,2d’p pairs non-singularly with
H n’p,n’2d+p .
Let κ be a form of type (r, 2n ’ 2d ’ r). By de¬nition of the duality
pairing we have
[X], [κ] = κ|X .
So, if r = n ’ d then the forms κ restricts to zero on X. Using non-
degeneracy of the pairing we arrive to the fact that [X] is of pure type
(d, d).

The statement converse to the one in Proposition is known as the
Hodge conjecture and is still open after more than 50 years of almost
fruitless attempts. We say that [Y ] ∈ H 2p (M, C) is a Hodge cohomology
class if it is integral (i.e. coming from H 2d (M, Z)) and is purely of type
(d, d).
THEOREM 3.7.4 (Lefschetz). Let X ‚ CPn be a projective mani-
fold. Then any Hodge class κ of degree 2 comes from complex-analytic
subvarieties (possibly singular) X1 , ..., Xk of codimension 1 in X, i.e.
κ = k ±[Xi ].

The proof of this result can be found e.g. in [31]. This theorem estab-
lishes the Hodge conjecture for the case d = 1, but unfortunately this is
the only case proved in general. In its ¬rst form the Hodge conjecture
was formulated as the above theorem when one changes 2 to 2d. But
after Atiyah and Hirzebruch found a counterexample in the 1960™s, the
conjecture was modi¬ed and is known at the present in the following

Hodge Conjecture. Let X be a projective manifold, and let κ be a
Hodge class κ of degree 2d. Then there exist complex-analytic subvari-
eties (possibly singular) X1 , ..., Xk of codimension d in X and an integer
h such that
h·κ= ±[Xi ].


. 21
( 44 .)