Lectures On Levi Convexity
Of Complex Manifolds And
Cohomology Vanishing Theorems
By
E. Vesentini
Tata Institute Of Fundamental Research, Bombay
1967
Lectures On Levi Convexity
Of Complex Manifolds And
Cohomology Vanishing Theorems
By
E. Vesentini
Notes by
M.S. Raghunathan
No part of this book may be reproduced in any
form by print, microﬁlm or any other means with
out written permission from the Tata Institute of
Fundamental Research, Colaba, Bombay 5
Tata Institute of Fundamental Research,
Bombay
1967
Preface
These are notes of lectures which I gave at the Tata Institute of Funda
mental Research in the Winter 1965.
Most of the material of these notes may be found in the papers [1],
[2] and [16] listed in the Bibliography at the end of this book.
My thanks are due to M.S. Raghunathan for the preparation of these
notes and to R. Narasimhan and M.S. Narasimhan for many helpful sug
gestions.
E. Vesentini
iii
Contents
Preface iii
0 Prerequisites 1
1 Vanishing theorems for hermitian manifolds 7
1 The space [
p,q
. . . . . . . . . . . . . . . . . . . . . . 7
2 The space W
p,q
. . . . . . . . . . . . . . . . . . . . . . 15
3 Wellipticity and a weak vanishing theorem . . . . . . . 25
4 Carleman inequalities . . . . . . . . . . . . . . . . . . . 30
2 Wellipticity on Riemannian manifolds 39
5 Wellipticity on Riemannian manifolds . . . . . . . . . . 39
6 A maximum principle . . . . . . . . . . . . . . . . . . . 44
7 Finite dimensionality of spaces of harmonic forms . . . . 50
8 Orthogonal decomposition in [
q
. . . . . . . . . . . . . 57
3 Local expressions for and the main inequality 61
9 Metrics and connections . . . . . . . . . . . . . . . . . 61
10 Local expressions for ∂, ϑ and . . . . . . . . . . . . . 70
11 The main inequality . . . . . . . . . . . . . . . . . . . . 75
4 Vanishing Theorems 79
12 qcomplete manifolds . . . . . . . . . . . . . . . . . . . 79
13 Holomorphic bundles over qcomplete manifolds . . . . 85
14 Examples of qcomplete manifolds . . . . . . . . . . . . 92
15 A theorem on the supports of analytic functionals . . . . 95
v
Chapter 0
Prerequisites
In this chapter we collect together some well known results which will 1
be used in the course of these lectures.
1. Let X be a connected, paracompact complex manifold of (com
plex) dimension we denote by Ω The sheaf of germs of homo
morphic pfoxes end by ∧
x,q
the sheaf of germs of 0
∞
forms of
type (P, /). Further ¦overset−∂ will denote the exterior diﬀeren
tiation with respect to
−
∂ Then we have one.
Proposition. The sequence
0 →Ω
P
→ A
p,0
¯
∂
→ A
p,1
. . .
¯
∂
→ A
p,n
→ 0
is exact. Similarly if K
p,q
denotes the sheaf of germs of (p, q)
currents, the sequence
0 →Ω
p
→ K
p,c
¯
∂
→ K
p,1
→ . . .
¯
∂
→ K
p,n
→ 0
is exact.
We point out that the exactness of the ﬁrst step of the second se
quence:
0 →Ω
p
→ K
po
1
2 0. Prerequisites
is equivalent to the fact that every
−
∂ closed distribution is a holo
morphic function
1
(For a proof see for instance [5], Expos´ e XVIII, [22]) 2
The sheaves A
p,q
being ﬁne, the cohomology group H
q
(X, Ω
p
) is
canonically isomorphic to the qth cohomology group of either one
of the complexes
0
//
H
◦
(X, A
p,o
)
//
¯
∂
//
H
◦
(X, A
p,1
)
¯
∂
// . . .
¯
∂
//
H
◦
(X, A
p,n
)
//
0
0
//
H
◦
(X, K
p,o
)
//
¯
∂
//
H
◦
(X, K
p,1
)
¯
∂
// . . .
¯
∂
//
H
◦
(X, K
p,n
)
//
0.
2. We need the following theorem due to Leray.
Let G = ¦U
i
¦
i∈I
be a locally ﬁnite open covering of a paracompact
space X. Let / be a sheaf of abelian groups on X such that for
every q > 0, H
q
(U
i
o
∩. . . ∩ U
i
p
, /) = 0 for every set (i
0
, . . . i
p
) of
pelements in I. Then the canonical map
H
q
(G, /) → H
q
(x, /)
is an isomorphic for all q ≥ o
For a proof see for instance [13] 209210
3. The following result enables us to apply the above theorem to
locally free sheaves over homomorphic function on a complex
manifold.
Let X be a (para compact) complex manifold of complex dimen
sion n. Given a vector bundle E an X and any covering G =
¦U
i
¦
iǫI
there is a reﬁnement ¦V
j
¦
jǫJ
such that for j
1
, . . . , j
k
with
V
j1
∩ . . . ∩ V
jk
φ, for 0 ≤ p ≤ n, the sequence
1
The idea of the proof is the following. Let α be any compactly supported function
of class C
k
with k > 0. Since
−
∂(T ∗ α) = 0, T ∗ α is a homomorphic function hence it is
C
∞
. Then T itself is a C
∞
function [30, Vol. II, Theorem XXI, 50]. But
−
∂T = 0. Then
T is holomorphic.
3
0 → H
0
(V
j1,... jk
, Ω
p
⊗
O
E
−
→ H
0
(V
j1... jk
A
−
p1
⊗
O
E
−
)
¯
∂⊗1
−→ . . .
. . .
¯
∂O1
→ H
0
(V
j1... jk
, A
pn
⊗
O
E
−
) → 0
where O is the structure sheaf, is exact, and V
j1... jk
is the intersec 3
tion V
j1
∩. . . ∩V
jk
and E is the locally free O sheaf associated to
E. (Since the sheaves, A
pi
⊗
O
E are ﬁne sheaves, this implies that
H
p
(V
ji
. . .
jk
′ Ω
p
⊗
O
E) = 0
4. Let Π : E → X be a holomorphic vector bundle on the complex
manifold X. Let G = (U
i
)
i∈I
be a covering of X and e
i j
: U
i
∩U
j
→
G = GL(n, C) be transition functions with respect to G
′
such that
¦G, e
i j
¦ deﬁne E. This means that we are given isomorphisms
ϕ
i
: E/U
i
→ U
i
× C
n
,
for i ∈ I such that e
i j
are the maps deﬁned by
(ϕ
j
◦ ϕ
−1
i
)(x, v) = (x, e
i j
(x)(v))
for x ∈ U
i
∩ U
j
. In particular, we have for i, j, k ∈ I
e
ik
(x) = e
i j
(x)e
jk
(x)
provided that x ∈ U
i
∩ U
j
∩ U
k
.
For a vector bundle π : E → X, we denote, as before by E or Ω(E)
the sheaf of germs of holomorphic section of E. Ω(E) is a locally
free sheaf over the sheaf of germs of holomorphic functions on X
(we denote this latter sheaf by O). Then, in the above notation,
Ω(E)
¸
¸
¸
¸
¸
U
i
≃ O
n
We ﬁx moreover the following notation:
Ω
p
(E) = E ⊗
O
Ω
p
, in particular Ω
◦
(E) = Ω(E);
A
pq
(E) = E ⊗
O
A
pq
;
K
pq
(E) = E ⊗
O
K
pq
.
4 0. Prerequisites
Ω
p
(E) (resp. A
pq
(E), resp. K
pq
(E)) is the sheaf of germs of holo 4
morphic Evalued pforms (resp. diﬀerentiable (p, q) forms with
values in E, resp. (p, q) currents with values in E ;
Γ(X, A
pq
(E)) = Global(p, q)C
∞
forms with values in E;
Γ(X, K
pq
(E)) = Global(P, q) currents with values in E.
There is a oneone correspondence between global sections of
A
pq
(E) and collections (ϕ
i
)
i∈I
where each ϕ
i
is a vector valued
form,
ϕ
i
=
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ϕ
′
i
.
.
.
ϕ
m
i
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
each ϕ
k
i
being a scalar C
∞
(p, q) forms on U
i
such that for x ∈ U
i
∩
U
j
we have ϕ
i
(x) = e
i j
(x)ϕ
j
(x). One can set up a similar oneone
correspondence between sections of K
pq
(E) and families (ϕ
i
)
i∈I
of currents, each ¦ϕ
k
i
¦
i
being deﬁned on U
i
, satisfying e
i j
ϕ
j
= ϕ
i
on U
i
∩ U
j
.
We denote by
¯
∂ the operators
1 ⊗
¯
∂ :E ⊗
O
A
pq
−→ E ⊗ A
p,q+1
and 1 ⊗
¯
∂ :E ⊗
O
K
pq
−→ E ⊗ K
p,q+1
.
Then the sequences
0 →Ω
p
(E) → A
p,o
(E)
¯
∂
→ . . .
¯
∂
→ A
p,n
(E) → 0
and 0 →Ω
p
(E) → K
p,o
(E)
¯
∂
→ . . .
¯
∂
→ K
p,n
(E) → 0
are exact. Moreover the sheaves A
pq
(E) and K
pq
(E) are ﬁne.
Hence if Φ denotes either the family of all closed subsets of X 5
or the family of all compact subsets of X, (more generally, any
paracompactifying family) and we set,
Γ
Φ
(X, A
pq
(E)) =
_
σ¦σ a section of A
pq
(E) over X, Supp σ ∈ Φ
_
,
then H
q
Φ
(X, Ω
p
(E)) ≃
_
ϕ¦ϕ ∈ Γ
Φ
(X, A
p,q
(E));
¯
∂ϕ = 0
_
¦
_
¯
∂Γ
Φ
(X, A
p,q−1
(E))
_
5
and similarly with the obvious notation,
H
q
Φ
(X, Ω
p
(E)) ≃
_
ϕ¦ϕ ∈ Γ
Φ
(X, K
p,q
(E));
¯
∂ϕ = 0
_
¦
_
¯
∂Γ
Φ
(X, K
p,q−1
(E))
_
Chapter 1
Vanishing theorems for
hermitian manifolds
1 The space [
p,q
All manifolds considered are assumed to be connected and paracompact. 6
Let π : E → X be a holomorphic vector bundle on a paracompact
complex manifold X of complex dimension n.
Let G =
_
U
i
_
iǫI
be a locally ﬁnite open coordinate covering of X and
_
e
i j
: U
i
∩ U
j
→ GL(mC)
_
i j∈I×I
transition functions deﬁning E.
A hermitian metric along the ﬁbres of E is a collection
_
h
i
: U
i
→
GL(m, C)
_
iǫI
of C
∞
maps such that for xǫU
i
, h
i
(x) is a positive deﬁnite
hermitian matrix and for xǫU
i
∩ U
j
, h
j
(x) =
t
e
i j
(x)h
i
(x)e
i j
(x).
Lemma 1.1. Every holomorphic vector bundle on a complex manifold
admits a hermitian metric.
Proof. Let
_
ρk
_
k∈I
be a C
∞
partition of unity subordinate to
_
U
k
_
k∈I
(=
G). Let
_
h
o
i
: U
i
→ GL(m, C)
_
i∈I
be any family of C
∞
functions such
that for i ∈ I, x ∈ U
i
, h
o
i
(x) is a positive deﬁnite hermitian matrix. Then
the family
_
h
o
i
_
i∈I
deﬁned by
h
i
(x) =
k
ρ
k
(x)
t
e
ki
(x)h
o
k
(x)e
ki
(x)
7
8 1. Vanishing theorems for hermitian manifolds
is a metric along the ﬁbres of E.
In particular, the holomorphic tangent bundle H → X admits a 7
hermitian metric. Let G =
_
U
i
_
i∈I
be a covering of X by means of
coordinate open sets. Let Z
1
i
, . . . , Z
n
i
be any system of coordinates in U
i
.
Then H → X is deﬁned with respect to G by means of the transition
functions J
i j
: U
i
∩ U
j
→ GL(n, C) deﬁned by
J
i j
(x) =
_
¸
¸
¸
¸
_
∂Z
α
j
∂Z
ρ
l
β
(x)
_
¸
¸
¸
¸
_
αβ
for x ∈ U
i
∩ U
j
.
A hermitian metric on H is then a family of C
∞
hermitian positive
deﬁnite matrixvalued functions g
i
: U
i
→ GL(n, C) such that
g =
t
J
i j
g
i
J
i j
Identifying hermitian matrices with hermitian bilinear forms on the tan
gent spaces through the basis
∂
∂Z
1
. . .
∂
∂Z
n
of the holomorphic tangent
space, we may write g
i
as
αβ
g
iα
¯
β
dZ
α
i
d
¯
Z
β
α
.
A hermitian metric on X deﬁnes on X (regarded as a diﬀerentiable
manifold), a Riemannian metric: in fact, if z
α
i
= x
α
i
+ iy
α
i
, then we have
ds
2
=
g
iαβ
dZ
α
i
dI
β
i
=
Re g
iαβ
(dx
α
i
dx
β
i
+ dy
α
i
dy
β
i
)
in the coordinate open set U
i
, in terms of the local real coordinates x
i
, y
i
.
Since X carries a complex structure, it has a canonical orientation de
ﬁned by this structure. Hence X has a canonical structure of an oriented
Riemannian manifold. We denote by ℓ the volume form on X with re
spect to this oriented Riemannian structure.
Let C
r
(X), 0 ≤ r ≤ 2n (n= complex dimension of X) denote the 8
space of complex valued exterior diﬀerential forms of degree r. It is
a module over the algebra, F, (over the complex numbers) of complex
valued C
∞
function on X
1. The space [
p,q
9
Deﬁnition 1.1. The “star operator” ∗ : C
r
(X) → C
m−r
(X) on an ori
ented Riemannian manifold X of (real) dimension m is deﬁned by the
formula
(∗ϕ)(t
1
, . . . , t
m−r
).ℓ = ϕ ∧ τ(t
1
) ∧ . . . ∧ τ(t
m−r)
where ϕ ∈ C
r
(X), t
1
, . . . , t
m−r
are tangent vectors to X at a point P and τ
is the canonical isomorphism of the Fmodule of tangent vector ﬁelds on
X onto the Fmodule of 1forms. Clearly * is a real operator, i.e. ∗ϕ =
∗ϕ (the bar denoting conjugation of complex numbers). Furthermore we
have
Lemma 1.2. Let (X, g) be an oriented Riemannian manifold of dimen
sion m. Then
∗ ∗ ϕ = (−1)
r(m−1)
ϕ for ϕ ∈ C
r
(X). (1 I)
Proof. ([18]) Let x ∈ X and t
1
, . . . , t
m
be an orthonormal basis of the
tangent space at x with respect to the Riemannian metric. Then it is
enough to check that
∗ ∗ (τ(t
i
1
) ∧ . . . ∧ τ(t
ir
)) = (−1)
r(m−1)
τ(t
i
1
) ∧ . . . ∧ τ(t
i
r
)
for every 0 < i
1
< i
2
< . . . < i
r
≤ m, or again that for every sequence
0 < j
1
< . . . < j
r
≤ m,
(∗∗)ττ(t
i
1
) ∧ . . . ∧ τ(t
i
r
)(t
j
1
, . . . t
j
r
)
=
_
τ(t
i
1
) ∧ . . . ∧ τ(t
i
r
)
_
(t
j
1
, . . . t
j
r
) (−1)
r(m−1)
.
Suppose now that ( j
1
, ..., j
r
) (i
1
, ..., i
r
). then 9
(∗∗)(τ(t
i
1
) ∧ ..., ∧τ(t
i
r
)(t
j
1
, ..., t
j
r
).l
= ∗
_
τ(t
i
1
) ∧ ... ∧ τ(t
i
r
)
_
∧ τ(t
j
1
) ∧ ∧ τ(t
j
r
)
and the right hand side is zero because,
_
τ(t
i
1
) ∧ ∧ τ(t
i
r
)
_
(t
λ
1
, . . . , t
λ
m−r
)l
10 1. Vanishing theorems for hermitian manifolds
= τ(t
i
1
) ∧ ... ∧ τ(t
i
r
) ∧ τ(t
λ
1
) ∧ ... ∧ τ(t
λ
m−r
)
and here the right side is zero unless
_
i
1
, .., i
r
, λ
1
, .., λ
m−r
_
=
_
1, 2, ..., m
_
.
t
1
, ..., t
m
are so chosen that τ(t
1
) ∧... ∧τ(t
m
) give the natural orienta
tion,
∗(τ(t
i
1
) ∧ ... ∧ τ(t
i
r
))(t
λ
1
, ..., t
λ
m−r
) = 0
if (i
1
, ..., i
r
, λ
1
, ..., λ
m−r
) (1, 2, ..m) and
∗(τ(t
i
1
) ∧ ... ∧ τ(t
i
r
))(t
λ
1
, ...t
m−r
)l =
= τ(t
i
1
) ∧ τ(t
i
r
) ∧ τ(t
λ
1
)...(t
λ
m−r
).
Now, the right side is clearly ε l where ε is the signature of the
permutation
(1, 2, ..., n) (t
i
1
, ..., t
i
r
, t
λ
1
, ...t
λ
m−r
).
It follows that
∗(τ(t
i
1
) ∧ ... ∧ τ(t
i
r
)) = ετ(t
µ
1
) ∧ ... ∧ τ(t
µ
m−r
)
where 0 < µ
1
< . . . < µ
m−r
≤ m are so chosen that 10
_
t
i
1
, ..., t
i
r
, t
µ
1
, .., t
µ
m−r
_
= (1, 2, .., m).
Hence
(∗∗)(τ(t
i
1
)∧), ∧ (t
i
1
, ..., t
i
r
)
= ετ(t
µ
1
) ∧ .. ∧ τ(t
µ
r
) ∧ τ(t
i
1
) ∧ ... ∧ τ(t
i
r
)
= εε
′
ℓ
where ε
′
is the signature of the permutation
(1, 2, ..., m) (t
µ
1
, .., t
µ
r
, t
i
1
, .., t
i
r
).
Hence εε
′
= (−1)
r(m−r)
= (−1)
r(m−1)
. It follows that
(∗∗)(τ(t
i
1
)∧... ∧τ(t
i
)(t
i
1
, .., t
i
r
) = (−1)
r(m−1)
_
τ(t
i
1
) ∧ ... ∧ τ(t
i
r
)
_
(t
i
1
, ..., t
i
r
).
1. The space [
p,q
11
For ( j
1
, j
2
, .., j
r
) (i
1
, .., i
r
) we have already seen
(∗∗)τ(t
i
1
) ∧ .. ∧ τ(t
i
))(t
j
1
, ..., t
j
r
) = 0.
Evidently,
_
τ(t
i
1
) ∧ ... ∧ τ(t
i
)
_
(t
j
1
, .., t
j
r
) = 0.
Hence the lemma.
The isomorphisms τ between the Fmodule of C
∞
tangent vector
ﬁelds on X onto the Emodule C
1
extends to an isomorphism between
the rth exterior power of these modules.
Let U be a coordinate open set on the diﬀerentiable manifold X.
Let x
1
, ..., x
m
be a coordinate system on U. Then for the ﬁxed basis
dx
1
, ..., dx
m
for the space of diﬀerentials at every point of U, any rform 11
ϕ ∈ C
r
can be represented in U by
φ = φ
I
dx
I
,
where φ
I
is a C
∞
function on U, I = (i
1
, ..., i
r
) is an rtuple of indices
1 ≤, i
1
< . . . < i
r
≤ m = dim
R
X, and dx
I
= dx
1
i
∧ . . . ∧ dx
1
r
.
Let g
i j
be the metric tensor in U, and let g
i j
(i, j = 1, ..., m) be the
elements of the inverse matrix g
−1
= (g
i j
)
−1
. Then τ
−1
φ is deﬁned in U
by its components (with respect to the basis
∂
∂x
1
, ...,
∂
∂x
m
)
ϕ
i
1
...i
r
= g
i
1
k
1
. . . g
i
r
k
r
ϕ
k
1
...k
r
.
If
(∗ϕ)
J
dx
J
is the local representation in U of the (m−r)form ∗ϕ, J being the multi
index J = ( j
1
, . . . , j
m−r
)(1 ≤ j
1
< . . . < j
m−r
≤ m), then
(∗ϕ)
J
= ℓ
IJ
ϕ
I
.
where ℓ = ℓ
1...
m
dx
′
∧ ... ∧ dx
m
is the representation of the volume ele
ment in U(ℓ
1
...
m
=
_
det(g
i j
)) and ℓ
IJ
= εℓ
′
1...m
ε being the sign of the
summation (1, ...m) (I J).
12 1. Vanishing theorems for hermitian manifolds
The proof is straightforward. It follows that for ϕ, ψ ∈ C
r
,
ϕ ∧ ∗ψ =
1
r!
_
¸
¸
_
¸
¸
_
I
ϕ
I
ψ
I
_
¸
¸
_
¸
¸
_
ℓ (1, II)
the summation being over all multiindices I = (i
1
, ..., i
r
) or
ϕ ∧ ∗ψ =
ϕ
I
ψ
I
if we extend the summation only to the multiindices I = (i
1
, ..., i
r
),
l ≤ i
1
< . . . < i
r
≤ m clearly,
clearly, ϕ ∧ ∗ψ = ψ ∧ ∗ϕ
and ∗ϕ∧ ∗ ∗ψ = ϕ ∧ ∗ψ
_
¸
¸
_
¸
¸
_
. . . (1, III)
We now revert to the situation when the manifold X is a complex 12
manifold and the Riemannian structure is the one canonically associated
to a hermitian structure on X . Let C
pq
(X) be the complex vector space
of C
∞
forms of type (p, q) on X .
Lemma 1.3. The ∗operator maps C
pq
(X) into C
n−q,n−p
(X).
Proof. The isomorphism τ of the (real) tangent space to X at xǫX onto
space of diﬀerentials extends to a complex linear isomorphism of the
“complexiﬁed” tangent space onto the space of complex valued diﬀer
entials. Denoting the extension again by τ and noting the fact that every
complex valued diﬀerential form may be regarded as a multilinear form
on the complexiﬁed tangent space, we see that the formula
(∗ϕ)(t
1
, . . . , t
m−r
)ℓ = ϕ ∧ τ(t
1
) ∧ ∧ τ(t
m−r
)
holds for complex tangent vectors (t
1
, . . . , t
m−r
) as well. Moreover from
the deﬁnitions of the associated Riemannian structure, we see easily that
for every α, τ
_
∂
∂z
α
i
_
is a linear combination of the dz
β
i
while τ
_
∂
∂¯ z
α
α
_
is
a linear combination of the dz
β
i
. Our assertion now follows from the
deﬁnitions of forms of type (p, q).
1. The space [
p,q
13
Now, let E be a holomorphic vector bundle on X deﬁned by (G =
¦U
i
¦
i∈I
; e
i j
: U
i
∩ U
j
→ GL(n, C)).
Let h = (h
i
)
i∈I
be a hermitian metric along the ﬁbres. Let C
p,q
(X, E)
denote the space of C
∞
 forms of type (p, q) on X with values in E
. The operator * of the associated Riemannian structure deﬁnes also a
canonical isomorphism
∗ : C
p,q
(X, E) → C
n−p,n−p
(X, E)
as follows. If a form ϕ in C
p,q
(X, E) is represented in U
i
by a column 13
_
¸
¸
¸
¸
¸
_
.
.
.
ϕ
1
i
ϕ
m
i
_
¸
¸
¸
¸
¸
_
where each ϕ
k
i
is a scalar (p, q) form, then ∗ϕ is represented by
_
¸
¸
¸
¸
¸
_
.
.
.
∗ϕ
1
i
∗ϕ
m
i
_
¸
¸
¸
¸
¸
_
in U
i
. Next, the hermitian metric h enables us to deﬁne an operator
# : C
p,q
(X, E) →C
q, p
(X, E
∗
),
where E
∗
is the dualbundle to E.(E
∗
is deﬁned by the transition func
tions ¦x t
−1
e
i
(x)
¦ with respect to the covering G = ¦U
i
¦
i∈I
). In fact if
we represent a (p, q)form ϕ in U
i
by a column
_
¸
¸
¸
¸
¸
_
.
.
.
ϕ
1
i
ϕ
m
i
_
¸
¸
¸
¸
¸
_
, where each ϕ
k
i
is a scalar form of type (p, q), then #ϕ in U
i
is given by the column of
(p, q)forms
(#ϕ)
i
=
_
¸
¸
¸
¸
¸
¸
¸
¸
_
h
i
ϕ
1
i
h
i
ϕ
2
i
h
i
ϕ
m
i
_
¸
¸
¸
¸
¸
¸
¸
¸
_
That ¦(#ϕ)
i
¦
i∈I
deﬁne forms with values in E
∗
follows from the fact
h
i
ϕ
i
= (t
e
ji
h
j
e
ji
e
i j
ϕ
j
) = (t
e
ji
h
j
ϕ
j
)
−
= e
t−1
i
j
h
j
ϕ
j
Remark . C
p,q
(X, E) has a structure of an Fmodule where F is the 14
ring of global complexvalued C
∞
functions. For this structure of F
modules on the C
p,q
(X, E)
∗ : C
p,q
(X, E) → C
n−q,n−p
(X, E)
14 1. Vanishing theorems for hermitian manifolds
is an Flinear isomorphism, while the operator
# : C
p,q
(X, E) → C
q, p
(X, E
∗
)
satisﬁes the condition
#( f .ϕ) = f (#ϕ) where f ∈ F, ϕ ∈ C
p,q
(X, E).
It follows that the operators * and # deﬁne actually homomorphisms
∗ : A
p,q
(E) → A
n−q,n−p
(E) (resp. A
p,q
(E)
♯
→ A
q, p
(E
∗
)
where A
rs
(E)(resp. A
rs
(E
∗
))
is the vectorbundle of exterior diﬀerential forms of type (r, s) with val
ues in E (resp. E
∗
); here * is Clinear while # is anti linear.
In the special case when E is trivial bundle, the map # : C
◦◦
(X, E) →
C
◦◦
(X, E
∗
) is simply the map
f −f .
Let α ∈ C
p,q
(X, E) and β ∈ C
rs
(E
∗
). Then (α ∧ β) is deﬁned as a
scalar form of type (p + r, q + s). For ϕ, ψ ∈ C
pq
(X, E) we set
A
E
(ϕ, ψ)l = ϕ∧(# ∗ ψ). (We note that #∗ = ∗#.) If (supp ϕ∩supp ψ)
is compact, then
_
¸
¸
¸
¸
¸
_
X
¦ϕ ∧ ∗#ψ¦ =
_
¸
¸
¸
¸
¸
_
X
¦A
E
(ϕ, ψ)¦ < ∞
Let U be an open coordinate set in X and let ϕ, ψ be represented in
U by
ϕ =
_
1
p!q!
ϕ
a
A
¯
B
dz
A
∧ dz
B
_
, ψ =
_
1
p!q!
ψ
a
AB
dz
A
∧ dz
B
_
,
where: a = 1, . . . , m = rankE, A = (α
1
, . . . , α
p
), B = (β
1
, . . . , β
q
), 15
1 ≤ α
i
, β
i
≤ n, dz
A
= dz
α
1
∧ . . . ∧ dz
α
p
, dz
β
= dz
β
1
∧ . . . ∧ dz
β
q
.
Then it follows from (1 II) that
A(ϕ, ψ) =
1
p! q!
h
ba
ϕ
a
AB
φ
ψ
bAB
,
2. The space W
p,q
15
(h
ba
)
a,b=1,...,m
being the local representation in U of the metric along the
ﬁbres of E. We shall call A(ϕ, ϕ)
1
2
the “length of the form ϕ.
Sometimes it will be denoted also by ¦ϕ¦.
Let ϑ
p,q
(X, E), be the space of C
∞
− (p − q)− forms with compact
support. Then for ϕ, ψǫD
pq
(X, E) we can deﬁne a scalar product:
(ϕ, ψ) =
_
X
ϕ ∧ (∗#ψ) =
_
X
A
E
(ϕ, ψ).
Also, we denote (ϕ, ϕ) by jϕj
2
; j j is in fact a norm: the hermitian
bilinear form (ϕ, ψ) deﬁnes on D
p,q
(X, E) the structure of a prehilbert
space over C. The completion of 1
p,q
(X, E) is denoted [
p,q
(X, E).
This latter space is referred to in the sequel also as the space of square
summable forms. Because of the RieszeFisher theorem [
pq
(X, E) can
be identiﬁed with the space of square forms of type (p, q)i.e. the space
of ϕ such that
_
X
A(ϕ, ϕ) < ∞.
2 The space W
p,q
The operator ∂ : A
p,q
(E) → A
p,q+1
(E) deﬁnes homomorphisms (again
denoted by the same symbol)
∂ : C
p,q
(X, E) → C
p,q+1
(X, E)
∂ : D
p,q
(X, E) →D
p,q+1
(X, E).
We can now deﬁne the adjoint of ∂ with respect to hermitian metric 16
on X and a hermitian metric along the ﬁbres of the vector bundle π :
E → X, i.e. an operator
ϑ : C
p,q
(X, E) → C
p,q−1
(X, E)
satisfying
(∂ϕ, ψ) = (ϕ, ϑψ) (1.1)
for ϕǫD
p,q
(X, E), ψǫD
p,q+1
(X, E). Such an operator, if it exists is
immediately seen to be unique. For the existence we have the
16 1. Vanishing theorems for hermitian manifolds
Lemma 1.4. If ϕǫD
p,q
(X, E), ψǫD
p,q+1
(X, E) are Lipschitz continuous
forms with compact support, then (∂ϕ, ψ) = (ϕ, − ∗ #
−1
∂ ∗ #ψ).
Proof. We note ﬁrst that the scalar products on both sides are deﬁned.
Firstly, the ∗ and # being Clinear (resp. Canti linear) bundle homomor
phisms, these can also be deﬁned on arbitrary forms  that is arbitrary
sections of the vector bundle A
p,q
(E). Secondly, since ϕ and ψ are Lips
chitz continuous ∂ϕ and −
∗
#
−1
∂ ∗ # are deﬁned almost every where and
are bounded measurable forms with compact support. We have more
over,
d(ϕ ∧ ∗#ψ) = ∂(ϕ ∧ ∗#ψ)
=
_
∂ϕ ∧ ∗#ψ
_
+ ¦(−1)
p+q
ϕ ∧ ∂ ∗ #ψ¦
=
_
∂ϕ ∧ ∗#ψ
_
− ¦ϕ ∧ ∗#ϑψ¦
since ∗ ∗ = (−1)
(p+q)(2n−p−q)
= (−1)
p+q
(here n is the complex dimension of X). Applying Stoke’s formula, we
obtain (since ϕ and ψ are compact support),
0 =
_
X
(∂ϕ ∧ ∗#ψ) −
_
X
ϕ ∧ ∗#ϑψ
i.e. (Jϕ, ψ) = (ϕ, ϑψ), which proves the lemma.
The operator ϑ depends on the hermitian metric on X and on the 17
hermitian metric along the ﬁbres of E. To emphasise this fact, we may
write sometimes ϑ
E
for ϑ.
We have already introduced the norm j j on D
p,q
(X, E) and denoted
the corresponding completion by [
p,q
(X, E). We will now introduce
another norm on D
p,q
(X, E).
Deﬁnition 1.2. For ϕ, ψǫD
p,q
(X, E).
a(ϕ, ψ) = (∂ϕ, ∂ψ) + (ϑψ, ϑψ) + (ϕ, ψ).
and N(ϕ)
2
= a(ϕ, ϕ).
As before we see that the hermitian scalar product a(ϕ, ψ) deﬁnes a
complex prehilbertspace structure on D
p,q
(X, E). We denote by W
p,q
2. The space W
p,q
17
(X, E) the corresponding completion. Clearly, we have for ϕ, ψǫD
p,q
(X, E) the inequality jϕj ≤ N(ϕ), so that the identity
D
p,q
(X, E) →D
p,q
(X, E)
extends uniquely to a continuous linear map,
i : W
p,q
(X, E) → [
p,q
(X, E).
Proposition 1.1. i is injective.
Proof. Let ϕ
ν
ǫD
p,q
(X, E) be a Cauchy sequence in N and assume that
jϕ
ν
j → 0. Since ϕ
ν
is a Cauchysequence in N, ∂ϕ
ν
and ϑϕ
ν
are Cauchy
sequences in the norm j j. Hence ∂ϕ
ν
and ϑϕ
ν
tend to a limit in
[
p,q+1
(X, E) and [
p,q−1
(X, E) respectively.
We denote these limits by u and v. By our identiﬁcation of [
p,q
(X, E) as the space of square summable Evalued forms, u and v may be
regarded as forms on X with values on E.
Now for any ψ ∈ D
p,q+1
(X, E) 18
(u, ψ) = lim
ν→∞
(∂ϕ
ν
, ψ) = lim
ν→∞
(ϕ
ν
, ϑψ) = 0
since jϕ
ν
j → 0. Since ψ is arbitrary in D
p,q+1
(X, E), U = 0. Similarly
v = 0. That is, ∂ϕ
ν
→ 0 and ϑϕ
ν
→ 0 in [
p,q+1
(X, E) and [
p,q−1
(X, E)
respectively. Thus N(ϕ
ν
) −→ 0 as ν −→ ∞, hence the proposition.
By continuity we obtain from (1.1) that, if ϕ ∈ W
p,q
(X, E), ψ ∈
W
p,q+1
(X, E), then
(∂ϕ, ψ) = (ϕ, ϑψ)
Clearly, we may now regard W
p,q
(X, E) also as a space of measur
able Evalued forms on X. From the deﬁnitions, it is moreover clear that
if f ∈ W
p,q
(X, E), ∂ f and ϑf in the distribution sense are currents repre
sentable by square summable (p, q+1) and (p, q−1) forms respectively.
In general W
p,q
(X, E) is not the space of all square summable forms ωof
type (p, q) whose ∂ and ϑ (in the sense of distributions) are again square
summable. We have however, the
18 1. Vanishing theorems for hermitian manifolds
Theorem 1.1. If the Riemannian metric (associated to the hermitian
metric) on X is complete, then
W
p,q
(X, E) =
_
ϕ ¦ ϕ ∈ [
p,
∧
q
(X, E); ∂ϕ ∈ [
p,q+1
(X, E); ϑϕ ∈ [
p,q−1
(X, E)
_
Proof. We will ﬁrst establish three lemmas which are needed for the
proof.
Lemma A. There exists C
◦
> 0 depending only on the dimension of X
such that for any scalar form u and any v ∈ C
p,q
(X, E),
A
E
(u ∧ v, u ∧ v)(x) ≤ ¦C
◦
¦ u ¦
2
A
E
(v, v)¦(x)
where ¦ u ¦
2
denotes the length of the scalar form a). (We recall that 19
A
E
(ϕ, ψ)ℓ = ϕ ∧ ∗#ψ). This is simply a lemma on ﬁnite dimensional
vector space with scalar products. We omit the proof.
Lemma B. Let p
◦
∈ X be any point. The function ρ(x) = d(p
◦
, x) =
distance from 0 of x, is locally Lipschitz continuous and wherever ρ has
partial derivatives, ¦ dρ ¦
2
≤ 2n.
Proof. We have by the triangle inequality
¦ ρ(x) − ρ(y) ¦≤ d(x, y).
Now if U is a coordinate open set with coordinates (X
1
, . . . , X
m
) (X
considered as a diﬀerentiable manifold of dimension m = 2n) and the
Riemannian metric is given on U by
_
g
i j
dx
i
x
j
then for any V ⊂ 0 ⊂ U,
there exists λ and µ such that
λ
dx
i
2
≤
g
i j
dx
i
dx
j
≤ µ
dx
i
2
so that if V is a ball about the origin in U, there exist constants C
1
C
2
> 0
such that
C
1
¦ x − y ¦≤ d(x, y) ≤ C
2
¦ x − y ¦ for x, y ∈ V.
Thus ¦ ρ(x) − ρ(y) ¦< C
2
¦ x − y ¦ for x, y ∈ V. Hence the ﬁrst
assertion.
2. The space W
p,q
19
For the second assertion consider about a point p a neighbourhood U
in which we can introduce geodesic polarcoordinates. Let (x
1
, . . . , x
m
)
be a coordinate system such that x
i
(p) = 0 for 1 ≤ i ≤ m.
For ε > 0 we can choose U so small that ¦ g
i j
¦< δ
i j
+ ε, i.e.
d(x, p)
2
≤ (1 + ε)
_
x
2
i
. In such a coordinate system, we have setting
e
i
= (0, . . . , 1 . . . , 0) (1 at the i
th
place)
∂̺
∂x
i
(ρ) = Lt
h→0
ρ(he
i
) − ̺(0)
h
On the other hand we have 20
¦ ρ(he
i
) − ρ(0) ¦≤ d(0, h, e
i
) ≤ (1 + ε)h
Since we are working with a coordinate system such that d(x, 0) ≤
(1+ε
_
_
x
2
i
_ 1
2
. Hence ρ is Lipschitz continuous and, if the derivatives ex
ist,
∂ρ
∂x
i
(p) ≤ 1. Now, by deﬁnition ¦ dρ ¦
2
=
_
g
i j
∂ρ
∂X
i
∂ρ
∂X
j
and in the coor
dinate system introduced above, g
i j
(0) = δ
i j
so that ¦ ρ ¦
2
=
m
_
1=1
_
_
∂ρ
∂x
i
_
_
2
≤
m. When X is a complex manifold of complex dimension n, we have,
¦ dρ ¦
2
≤ 2n.
Lemma C. Let
˙
D
p,q
∂
(X, E) =
_
ϕ ¦ ϕ ∈ [
p,q
(X, E), ∂ϕ ∈ [
p,q+1
(X, E), Supp ϕ ⊂⊂ X
_
˙
D
p,q
ϑ
(X, E) =
_
ϕ ¦ ϕ ∈ [
p,q
(X, E),
ϑϕ ∈ [
p,q
(X, E), ϑϕ ∈ [
p,q−1
(X, E), supp ϕ ⊂ X
_
(where ∂ and ϑ are in the sense of distributions) and ﬁnally
˙
D
p,q
(X, E) =
˙
D
p,q
∂
(X, E) ∩
˙
D
p,q
ϑ
(X, E).
Then
20 1. Vanishing theorems for hermitian manifolds
(i) D
p,q
(X, E) is dense in
˙
D
p,q
∂
(X, E) w.r.t. the norm
P(ϕ) : P(ϕ)
2
= jϕj
2
+ j∂ϕj
2
and
(ii) D
p,q
(X, E) is dense in
˙
D
p,q
ϑ
(X, E) with the norm
Q(ϕ) : Q(ϕ)
2
= jϕj
2
+ jϑϕj
2
(iii) D
p,q
(X, E) is dense in
˙
D
pq
(X, E) with the norm
N : N(ϕ)
2
= jϕj
2
+ j∂ϕj
2
+ ¦ϑϕj
2
.
Proof. We will prove (i) ; the proofs for the other two cases are similar. 21
Let G = ¦U
i
¦
i∈I
be a locally ﬁnite covering of X such that E ¦
U
i
is trivial.
Let m be the rank of E. Assume further that each U
i
is a coordinate open
set. Let ¦V
i
¦ V
i
⊂⊂ U
i
¦
i∈I
be a shrinking of U
i
. Let (ρ
i
)
i∈I
be a partition
unity subordinate to ¦V
i
¦
i∈I
. Let ϕ ∈
˙
mathscrD
p,q(X,E)
σ
. E ¦
U
i
being
trivial, we may take ρ
i
ϕ
i
to be a C
m
valued from on C
n
with compact
support in V
i
. For any ε > 0, we can ﬁnd a C
m
valued C
∞
from ψ
i
with compact support in V
i
⊂ C
n
, such that ¦¦ ∂ρ
i
ϕ
i
− ∂ψ
i
¦¦< ε and
jρ
i
ϕ
i
−ψ
i
j < ε. This can be secured by the usual regularisation methods.
Since the support of ϕ is compact, we may take ψ
i
= 0 except for a
ﬁnite number of i. Since E ¦
U
i
is trivial and ρ
i
ϕ
i
were regarded as scalar
forms through suitable trivialisations, we may now revert the process
and consider ψ
i
as Evalued C
∞
forms with support in V
i
. It follows
that ψ =
_
ψ
i
is a C
∞
form with compact support and
jϕ − ψj = j
ρ
i
ϕ
i
−
ψ
i
j ≤
i
jρ
i
ϕ
i
− ψ
i
j < Mε
and similarly j∂ϕ − ∂ψj < Mε where M is the number of indices of the
ﬁnite set
¦i ¦ i ∈ I, U
i
∩ support ϕ φ¦.
Since M is ﬁxed for a given ϕ and ε is at our choice, the lemma is
proved.
2. The space W
p,q
21
Lemma C enables us to prove the following more general statement.
Theorem 1.1. If the Riemannian metric (associated to the hermitian 22
metric) of X complete, then
i) D
p,q
(X, E) is dense in the space
_
ϕ ¦ ϕε[
p,q
(X, E), ∂ϕ ∈ [
p,q+1
(X, E)
_
,
with respect to the norm P(ϕ);
ii) D
p,q
(X, E) is dense in the space
_
ϕ ¦ ϕ ∈ [
p,q
(X, E), ϑϕ ∈ [
p,q−1
(X, E)
_
,
with respect to the norm Q(ϕ);
iii) D
p,q
(X, E) is dense in the space
_
ϕ ¦ ϕ ∈ [
p,q
(X, E), ∂ϕ ∈ [
p,q+1
(X, E), ϑϕ ∈ [
p,q−1
(X, E)
_
,
with respect to the norm N(ϕ).
Proof. We will prove i). The proofs of ii) and iii) are similar.
In view of Lemma C it is suﬃcient to prove that every distributive
form ϕ in [
p,q
(X, E) such that ∂ϕ ∈ [
p,q+1
(X, E) can be approximated
as closely as we want by forms ψ such that ψ, ∂ψ are square summable
and supp ψ, ⊂⊂ X.
Let µ :¦ R
1
→ [0, 1] be a C
∞
function such that
(i) µ(t) = 1 if t < 1 and (ii) µ(t) = 0 if t > 2.
Let M = Supp
t
¦
dµ
dt
¦. Let d(x, y) be the distance function deﬁned by
the complete Riemannian metric of X. Then we ﬁx a point p
◦
∈ X and 23
set ρ(x) = d(x, p
◦
) for x ∈ X, the function
ω
ν
(x) = µ
_
ρ(x)
ν
_
ν > 0
is locally Lipschitz, and where the derivatives
∂ρ
∂x
i
exist, then
¦ dω
ν
¦
2
≤¦
1
γ
∂µ
∂t
_
ρ(x)
ν
_
dρ ¦
2
≤
2nM
2
ν
2
22 1. Vanishing theorems for hermitian manifolds
in view of Lemma B. Since the metric is complete the ball of centre p
and radius c,
B
c
= ¦p ¦ d(p, p
◦
) < c¦
is relatively compact in X for all c > 0.
Consider now the form ω
ν
.ϕ = ϕ
ν
; ϕ
ν
has compact support. It is
easily seen that ω
ν
.ϕ = ϕ
ν
is in
˙
D
p,q
(X, E). We will now prove that
ϕ
ν
→ ϕ in the norm N as ν → ∞. In fact, we have ﬁrst of all
jϕ − ϕ
ν
j = j(1 − ω
ν
)ϕj
X−B
ν
≤ jϕj
X−B
ν
→ 0 as ν → ∞.
Secondly
j∂ϕ − ∂ϕ
ν
j = j∂ϕ − ω
ν
∂ϕ − ∂ω
ν
∧ ϕj
≤ j∂ϕj
X−B
ν
+ j∂ω
ν
∧ ϕj
Now, j∂ω
ν
∧ ϕj = A(∂ω
ν
∧ ϕ, ∂ω
ν
∧ ϕ)
≤¦ ∂ω
ν
¦
2
A(ϕ, ϕ) by Lemma A.
On the other hand, one checks easily that almost everywhere
¦ ∂ω
ν
¦
2
≤¦ dω
ν
¦
2
≤
2nM
2
ν
2
,
so that, we obtain
j∂ϕ
ν
− ∂ϕj ≤ j∂ϕj
X−B
ν
+
C
′
ν
jϕj.
Hence ¦¦ ∂ϕ
ν
−∂ϕ ¦¦→ 0 as ν → ∞. This completes the proof of Theorem 24
1.1 (i).
Deﬁnition 1.3. : C
p,q
(X, E) → C
p,q
(X, E) is the operator ∂ϑ + ϑ∂.
The following result is easily checked.
Lemma 1.5. The operator deﬁned above is strongly elliptic.
The operator depends on the metric on X and on the metric along
the ﬁbres of E. To emphasize this fact we may write
E
for .
For the operator we have the following result.
2. The space W
p,q
23
Theorem 1.2 (Stampacchia Inequality). We assume the Riemannian
metric on X (associated to the hermitian structure) to be complete. Let
p
o
∈ X and for ν > 0, let B
ν
= ¦x ¦ d(x, p
o
) < ν¦. Then there exists a
constant A > 0 such that for every σ > 0, for every choice r < R of a
pair of positive reals r, R and every ϕ ∈ C
p,q
(X, E),
¦¦ ∂ϕ ¦¦
2
B
r
+ ¦¦ ∂ϕ ¦¦
2
B
r
≤ σ ¦¦ ϕj
2
B
r
+
_
1
σ
+
A
(R − r)
2
_
jϕj
2
B
R
Proof. Let us choose as before a C
∞
function µ : R
1
→ [0, 1] such that
µ(t) = 1 for t < 1 and µ(t) = 0 for t > 2.
Let M = sup¦
dµ
dt
¦. consider the function deﬁned by
W(x) = µ
_
ρ(x) + R − 2r
R − r
_
where σ(x) = d(p
o
, x). Evidently then, ω has support in B
R
and w ≡ 1
on B
r
(B
R
⊂⊂ X).
Moreover, ¦dω¦
2
= ¦
1
R−r
∂µ
∂t
_
ρ(x)+R−2r
R−r
_
d
ρ
¦
2
. It follows that ¦dω¦
2
≤ 25
2nM
2
(R−r)
2
.
Suppose now that ϕ ∈ C
pq
(X, E) and ψ is any Lipschitz continuous
form with compact support in B
R
, then
(∂ϕ, ∂ψ)
B
R
+ (ϑϕ, ϑψ)
B
R
= (ϕ, ψ)
B
R
.
Set ψ = ω
2
ϕ. We have then
∂ψ = ω
2
∂ϕ + 2ω∂ωΛϕ
ϑψ = ω
2
ϑϕ − ∗(2ω∂ωΛ ∗ ϕ).
This leads to
(ω∂ϕ, ω∂ϕ)
B
R
+ (ωϑϕ, ωϑψ)
B
R
= (ϕ, ω
2
ϕ) − (ω∂ϕ, 2∂ωΛϕ) + (ωϑϕ, ∗(2∂ωΛ ∗ ϕ)).
Now, by Schuarz inequality,
¦ (ϕ, ω
2
ϕ)
B
R
¦≤
1
2
σjϕj
2
B
R
+
1
2σ
jω
2
ϕj
B
R
σ
2
jϕj
2
B
R
+
1
2σ
jϕj
2
B
R
24 1. Vanishing theorems for hermitian manifolds
¦ (ω∂ϕ, 2∂ωΛϕ)
B
R
¦
1
2
jω∂ϕj
2
B
R
+
4c
o
nM
2
(R − r)
2
jϕj
2
B
R
and ¦ (ωϑϕ, ∗(2∂ωΛ ∗ ϕ)) ¦
1
2
jωϑϕj
2
B
R
+
4c
o
nM
2
(R − r)
2
jϕj
2
B
R
(where c
o
is the positive constant which has been introduced in lemma
A)
If follows that 26
jω∂ϕj
2
B
R
+ jωϑϕj
2
B
R
≤ σjϕj
2
B
R
+ σjϕj
2
B
R
+
8c
o
nM
2
(R − r)
2
jϕj
2
B
R
.
The inequality follows now from
jω∂ϕj
2
B
R
j∂ϕj
2
B
and jωϑϕj
2
B
R
jϑϕj
2
B
R
This completes the proof of theorem 1.2.
Corollary 1. For ϕ
2
∈ C
p,q
(X, E) and for any σ > 0,
j∂ϕj
2
+ jϑϕj
2
≤ σjϕj
2
+
1
σ
jϕj
2
.
Proof. Set R = 2r in stampachia inequality and let r → ∞.
Corollary 2. If ϕ ∈ C
p,q
(X, E), jϕj < ∞and jϕj < ∞, then j∂ϕj < ∞,
jϑϕj < ∞. If squareϕ = 0, then ∂ϕ = ϑϕ = 0.
Proof. The ﬁrst assertion follows from corollary 1. The second again
from corollary 1 since σ is arbitrary.
Remark . On any (paracompact) complex manifold X there exists a
complete hermitian metric. More exactly we shall prove that, given any
hermitian metric ds
2
on X, there exists a C
∞
function F : X → R such
that F ds
2
is a complete metric.
Proof. Let ¦B
ν
¦
ν∈N
be a sequence of compact sets such that
B
ν
⊂
o
B
ν+1
, ∪B
ν
= X.
3. Wellipticity and a weak vanishing theorem 25
Let f
ν
: X → R be a C
∞
function on X, satisfying the following 27
conditions:
0 f
ν
1,
f
ν
= 1 on B
ν+1
− B
ν
,
Supp f
ν
⊂ B
ν+2
− B
ν−1
Let d(x, y) (x, y ∈ X) be the distance function determined by the
hermitian ds
2
, and let
ε
ν
= inf
x∈B
ν
y∈∂B
ν+1
d(x, y).
Then ε
ν
> 0. Consider the positive C
∞
function
F(x) =
+∞
ν=0
f
ν
(x)
ε
ν
and the hermitian metric
¯
ds
2
= F(x)ds
2
. Let
˜
d(x, y) be the distance
function determined by ds
2
. We have
∽
d(B
ν
, B
ν+1
)
x ∈
inf
B
ν+1
− B
ν
F(x)d(B
ν
, B
ν+1
)
1
ε
ν
ε
ν
= 1.
This implies that every Cauchy sequence for the distance
˜
d(x, y) con
verges. Hence
¯
ds
2
is a complete hermitian metric.
3 Wellipticity and a weak vanishing theorem
28
We now introduce a seminorm on W
p,q
(X, E) as follows: for ϕ, ψǫW
p,q
(X, E), let b(ϕ, ψ) = (∂ϕ, ∂ψ)+(ϑϕ, ϑψ); then b is a positive semi deﬁnite
form and b(ϕϕ) deﬁnes a seminorm on W
p,q
(X, E). the map
j : W
p,q
(X, E) → W
p,q
b
(X, E)
26 1. Vanishing theorems for hermitian manifolds
where the latter space is W
p,q
(X, E) provided with the (not necessarily
Hausdorﬀ) topology deﬁned by the seminorm b(ϕ, ϕ) and j is the iden
tity is clearly continuous: in fact, b(ϕ, ϕ) N(ϕ).
Deﬁnition 1.4. We say that E is W
p,q
elliptic with respect to the hermi
tian metric on X and the and the hermitian metric along the ﬁbres of E,
if j admits a continuous inverse.
In particular, the W
p,q
ellipticity of E implies that b(ϕ, ϕ)
1
2
is actu
ally a norm. In fact, there is a K > 0 such that N(ϕ)
2
Kb(ϕ, ϕ). Since
by deﬁnition, N(ϕ) = b(ϕ, ϕ) + jϕj
2
, W
pq
ellipticity is equivalent to fol
lowing: there is a constant C > 0(= K − 1 when K is as above) such
that
jϕj
2
Cb(ϕ, ϕ) = C(j∂ϕj
2
+ jϑϕj
2
). (1.2)
We shall call C a W
p,q
ellipticity constant.
Proposition 1.2. Assume given a hermitian metric on X and a hermitian
metric along the ﬁbres of π : E → X. Suppose further that E is W
pq
elliptic with reference to these hermitian metrics. Then there is a linear
map G : [
p,q
(X, E) → W
p,q
(X, E) such that, for f ∈ [
p,q
(X, E), ϕ ∈
W
p,q
(X, E), we have
( f , ϕ) = (∂ Gf , ∂ϕ) + (ϑGf , ϕ).
The linear map G is continuous more exactly: 29
b(Gf , Gf ) Cj f j
2
(1.3)
C being the constant which appears in (1.2). Moreover Gf is uniquely
determined by the above formula.
Proof. ϕ (ϕ, f ) deﬁnes a linear form on W
p,q
(X, E) which is evi
dently continuous. By the Riesz representation theorem (since E is W
p,q
elliptic, b(ϕ, ψ) deﬁnes a Hilbert space structure on W
p,q
(X, E) equiva
lent to that deﬁned by N), there is a unique element Gf ∈ W
p,q
(X, E)
such that
( f , ϕ) = b(Gf , ϕ) = (∂Gf , ∂ϕ) + (ϑGf , ϑϕ).
3. Wellipticity and a weak vanishing theorem 27
We have, now
b(Gf , Gf ) = (∂Gf , ∂Gf ) + (ϑGf , ϑGf ).
= ( f , Gf ) by the above equation.
Hence b(Gf , Gf )
2
j f j
2
.jGf j
2
Cj f j
2
b(Gf , Gf ) in view of W
pq

ellipticity. That is,
b(Gf , Gf ) Cj f j
2
.
We obtain
N(Gf )
2
(C + 1)j f j
2
.
This completes the proof of Proposition 1.2.
Corollary. Gf = f (in the sense of distributions).
Proof. For u ∈ D
p,q
(X, E), we have ( f , u) = (∂Gf , ∂u) = (ϑGf , u).
This proves the lemma.
Proposition 1.3. Let E be W
p,q
elliptic with respect to a given metric 30
along the ﬁbers and to a complete hermitian metric on X. In the notation
of Proposition 1.2 we have the following.
(i) If f ∈ [
p,q
(X, E) and ∂ f ∈ [
p,q+1
(X, E), then
∂Gf ∈ W
p,q+1
(X, E); ∂Gf = ∂f ;
jϑ∂Gf j
2
1
σ
j∂ f j
2
+ σj∂Gf j
2
for σ > 0.
(ii) If f ∈ [
p,q−1
(X, E) and ϑf ∈ [
p,q+1
(X, E), then
ϑGf ∈ W
p,q−1
(X, E); ϑGf = ϑf ;
¦¦
¯
∂ϑGf ¦¦
2
σ¦¦ϑ f ¦¦
2
+ σ¦¦ϑ Gf ¦¦
2
f orσ > 0.
Proof. Since the metric is complete, by Theorem 1.1’ 1
p,q
(X, E) is
dense in the space
_
ϕ ¦ ϕ ∈ [
p,q
(X, E), ∂ϕ ∈ [
p,q+1
(X, E)
_
provided with
the norm P(ϕ)
2
= jϕj
2
+ j∂ϕj
2
. Let f ∈ [
p,q
(X, E) and ∂f ∈ [
p,q+1
(X, E). Then there is a sequence f
n
∈ 1
p,q
(X, E) such that j f
n
− f j →
28 1. Vanishing theorems for hermitian manifolds
0, j∂ f
n
− ∂ f j → 0 as n → ∞. We have moreover, from Proposition 1.2,
that
N(Gϕ)
2
Kjϕj
2
for ϕ ∈ L
p,q
(X, E). Hence
jGf
n
− Gf
m
j Kj f
n
− f
m
j
and j∂Gf
n
− ∂Gf
n
j + jϑGf
n
− ϑGf
n
j Kj f
n
− f
m
j.
Applying now Stampachia inequality (Corollary 2, Theorem 1.2) to
∂Gf
n
, and taking into account the fact that and ∂ commute, we obtain,
for σ > 0,
jϑ∂(Gf
n
− Gf
m
)j
2
1
σ
¯
∂( f
ν
− f
µ
)j
2
+ σj(∂G( f
ν
− f
µ
)j
2
Since the right hand side tends to 0 as n, m → ∞, it follows that
_
ϑ∂ f
n
_
is a Cauchysequence in [
p,q
(X, E). It is immediate then that 31
ϑ∂Gf = lim
n→∞
ϑ∂Gf
n
∈ [
p,q
(X, E). On the other hand ∂(∂Gf ) = 0.
Hence ∂Gf ∈ W
p,q+1
(X, E). The equation ∂Gf = ∂ f follows from
the fact that ∂ and commute and the fact that Gf = f . The last
inequality is the Stampachia inequality applied to ∂Gf . The proof of
Part (ii) of the proposition is entirely analogous.
Theorem 1.3. Let π : E → X be a holomorphic vectorbundle which
is W
p,q
elliptic with respect to a complete hermitian metric on X and
a hermitian metric along the ﬁbres of E. Then if q > 0, given f ∈
[
pq
(X, E) with ∂ f = 0, there is a unique x ∈ W
pq
(X, E) such that f =
∂ϑx and ∂x = 0. Moreover, we have jϑxj
2
Cj f j
2
, C being a W
p,q

ellipticity constant.
Proof. We set x = Gf . We have then Gf = f . (Corollary to Propo
sition 1.2). Clearly ∂ f = 0 ∈ [
p,q+1
(X, E). Hence by (i) of proposition
1.3, we have ∂Gf ∈ W
p,q+1
(X, E) and further jϑ∂Gf j
2
σj∂Gf j
2
+
1
σ
j∂Gf j
2
. On the other hand since and ∂ commute, it follows again
from the Corollary to Proposition 1.2, that ∂Gf = 0. Since σ is ar
bitrary, ϑ∂x = 0. It follows that (∂x, ∂x) = (x, ϑ∂x) = 0 (note that
∂x ∈ W
p,q+1
(X, E): Proposition 1.3). Hence ∂x = 0. Now = ∂ϑ + ϑ∂
so that Gf = f leads to (∂ϑ + ϑ∂)Gf = f that is, ∂ϑGf = f .
3. Wellipticity and a weak vanishing theorem 29
Finally (1.3) yields 32
jϑxj
2
Cj f j
2
.
The uniqueness of x satisfying ∂ϑx = f and ∂x = 0 is easily
checked.
This completes the proof of the theorem.
It follows from the regularity theorem for elliptic systems that , if f
∈ C
p,q
(X, E) ∩ [
p,q
then Gf (can be modiﬁed on a null set so that it)
will be of class C
∞
on X.
In particular we have the
Corollary. If f ∈ 1
p,q
(X, E) and ∂ f = 0 then there exists ψ ∈ C
p,q−1
(X, E) such that ∂ψ = f
Remark. The corollary above clearly implies the following.
If E is W
pq
elliptic with respect to a complete hermitian metric, then
the natural map
H
q
k
(X, Ω
p
(E)) → H
q
(X, Ω
p
(E))
where the leftside stands for the q
th
cohomology with compact supports
of X with values in Ω
p
(E), is the trivial map α 0 for every α ∈
H
q
k
(X, Ω
p
(E)).
Remark. Let G =
_
U
i
_
be a covering of X such that π : E → X is
deﬁned with respect to G by holomorphic transitions e
i
j
: U
i
∩ U
j
→
GL(m, C)(m = rank of E).
Let
_
h
i
_
i∈I
be a hermitian metric along the ﬁbers of E; then h
i
is a
C
∞
function on U
i
whose values are positive deﬁnite hermitian matrices.
We have on U
i
∩ U
j
33
h
i
=
t
e
ji
h
j
e
ji
,
and therefore t
h
−1
i
=
t
(
te
−1
ji
)
t
h
−1
j
t
e
−1
ji
30 1. Vanishing theorems for hermitian manifolds
That means that
_
t
h
−1
i
_
deﬁnes a matric along the ﬁbres of the dual
bundle E
∗
.
We note here for future reference the following identities, holding
with respect tot he metrics ¦h
i
¦ on E and ¦t
h
−1
i
¦ on E
∗
. They follow
immediately from the deﬁnitions of the operators involved.
A
E
∗ (∗#ϕ, ∗#ψ) = A
E
(ϕ, ψ) (1.4)
∂ ∗ #ϕ = (−1)
p+q
∗ #ϑ
E
ϕ, (ϕ, ψ ∈ C
pq
(X, E)) (1.5)
ϑ
E
∗ ∗ #ϕ = (−1)
p+q+1
∗ #∂ψ, (1.6)
E
∗ ∗ #ϕ = ∗#
E
ϕ. (1.7)
As a corollary we have that ∗# deﬁnes an isometry of [
pq
(X, E) onto
[
n−p n−q
(X, E) which maps W
pq
isometrically onto W
n−p n−q
(X, E
∗
).
Furthermore, if E is W
pq
elliptic with respect to the metric ¦h
i
¦ on
E, then E
∗
is W
n−p n−q
 elliptic with respect to the metric ¦t
h
−1
i
¦ on E
∗
the Wellipticity constants being the same.
4 Carleman inequalities
We will now formulate certain further conditions on the vector bundles 34
and show how these more stringent conditions lead to stronger vanishing
theorems than the one above.
We assume always that the hermitian metric denoted ds
2
on X is
complete. Let the given metric on E be denoted by h.
C
1
. There is given a C
∞
function φ : X → R
+
.
C
2
. For every non decreasing convex C
∞
function λ : R → R.
E is W
pq
elliptic with respect to (ds
2
, e
λφ
h
).
C
3
. The W
pq
ellipticity constant is independent of.
That is, there is C > 0 independent of λ such that
j f j
2
λ
C
_
j ∂ f j
2
λ
+ j ϑ
λ
f j
2
λ
_
for f ∈ D
pq
(X, E) where j j
λ
stands
for the norm with respect to (ds
2
, e
λ(φ)
. h), and ϑ
λ
denotes the ϑoperator
with respect to these metrics.
4. Carleman inequalities 31
Conditions C
1
, C
2
, C
3
imply C
′
1
, C
′
2
, C
′
3
below. This weaker set of
conditions are suﬃcient for the “vanishing theorems” we will now prove.
C
′
1
. There is given a C
∞
function φ : X → R
+
(this is the same as C
1
).
C
′
2
. Given C
o
> 0, there is a nondecreasing C
∞
function λ : R → R
that λ(t) = 0 for t C
o
and λ(t) > 0 for t > C
o
, such that, E is
W
pq
elliptic with respect to (ds
2
, e
νλ(φ)
h) for every positive integer
ν.
C
′
3
. The W
pq
ellipticity constant is independent of ν that is, there is a 35
C > 0 such that for f ∈ D
pq
(X, E)
j f j
2
ν
≤ C
_
j ∂f j
2
ν
+ j ϑ
ν
f j
2
ν
_
,
where j j
ν
stands for j j
νλ
and ϑ
ν
for ϑ
νλ
.
Lemma 1.6. Assume given a constant C
o
> 0 and that the condition
C
′
1
, C
′
2
, C
′
3
above are satisﬁed (C
o
in condition C
′
2
is taken as the above
constant). For λ as in condition C
′
2
, let [
p,q
ν
(X, E) denote the space
of Evalued forms on X which are squaresummable with respect to
(ds
2
, e
νλ(φ)
.h). Then for f∈ ∩
ν
[
p,q
ν
(X, E)(q > 0) such that ∂f = 0, there
exist Ψ
ν
for every integer ν ≥ 0 such that Ψν ∈ [
p,q−1
ν
(X, E), ∂Ψ
ν
= f
and j Ψ
ν
j
ν
≤ C j f j
ν
(We assume that ds
2
is complete).
Proof. This follows from Theorem 1.3.
Theorem 1.4. Assume that conditions C
′
1
, C
′
2
, C
′
3
are satisﬁed. Then
for every squaresummable form f of type (p, q)(q > 0) having compact
support such that ∂ f = 0, there exists Ψ ∈ [
p,q−1
(X, E) such that, f =
∂Ψ, j Ψ j
2
≤ C j f j
2
and (support of Ψ)
⊂ ¦xjφ(x) < (sup φ(y)) y ∈ support f )¦.
Proof. Let C
o
= sup φ(x), x ∈ support f . Then we have
j f j
2
ν
=
_
X
e
νλ(φ)
A( f , f )dx =
_
Supp . f
e
νλ(φ)
A( f , f )dX
32 1. Vanishing theorems for hermitian manifolds
=
_
Supp f
A( f , f )dx =
_
X
A( f , f )dX =j f j
2
Since νλ(ϕ(x)) = 0 on the support of f . It follows that f ǫ ∩
ν
36
[
p,q
ν
(X, E) for every integer ν > 0. Hence by Lemma 1.6, we can ﬁnd
ψ
ν
ǫ [
p,q−1
ν
2
(X, E) for every νǫZ
+
such that
−
∂ψ
ν
= f and ¦¦ ψ ¦¦
2
ν
≤ C ¦¦ f ¦¦
2
.
On the other hand, we have ¦¦ ψ
ν
¦¦
ν
≥¦¦ ψ
ν
¦¦ so that, ¦¦ ψ
ν
¦¦
2
≤¦¦ ψ
ν
¦¦
2
ν
≤
C ¦¦ f ¦¦
2
= C
1
say. It follows that we can assume (by passing to a subse
quence if necessary) that ψ
ν
converges weakly in [
p,q−1
(X, E) to a limit
ψ. We have ¦¦ ψ ¦¦< C ¦¦ f ¦¦. On the other hand, for every ε > 0,
_
φ(x)>c
0
+ε
e
νλ(φ)
A(ψ
ν
, ψ
ν
)dX ≤ C
1
and since λ is nondecreasing, we have,
e
νλ(C
0
+ε)
_
ϕ≥C
0
+ε
A(ψ
ν
, ψ
ν
)dX ≤ C
1
.
It follows that
_
ϕ≥C
0
+ε
A(ψ
ν
, ψ
ν
)dX tends to zero and hence ψ
ν
→ 0
almost everywhere in ϕ ≥ C
0
+ ε. (The integrand is positive for all
ν.) Hence ψ = 0 on ¦x ¦ ϕ(x) ≥ C
0
+ ε¦ for every ε. Hence support
ψ ⊂ ¦x ¦ φ
′
(x) ≤ C
0
¦. Finally for
u ∈ D
n−p,n−q
(X, E
∗
), (−1)
p+q
(ψ,
−
∂u) = (−1)
p+q
_
ψ ∧
−
∂u
= Lt(−1)
p+q
_
ψ
ν
∧
−
∂u
=
_
f , u
_
.
Hence
¯
∂ψ = f in the sense of distributions. This completes the proof
of the theorem.
Remark. The proof contains the following lemma (which has no con 37
nection with W
p,q
ellipticity.
4. Carleman inequalities 33
Lemma 1.7. Suppose given a C
∞
nondecreasing function λ : R → R
such that λ(t) = 0 for t ≤ C
0
, λ(t) > 0 for t > C
0
, and a C
∞
function
ϕ : X → R. Suppose further that there is given an f ∈ [
pq
(X, E) such
that sup ϕ(x) on the support of f is C
0
. If for every integer ν > 0 there
exists ψ
ν
such that
¯
∂ψ
ν
= f and ¦¦ ψ
ν
¦¦
2
ν
≤ C ¦¦ f ¦¦
2
for a constant C
independent of ν, then there exists ψ ∈ [
p,q−1
(X, E) such that
¯
∂ψ = f , ¦¦
ψ ¦¦
2
≤ C ¦¦ f ¦¦
2
and support ψ ⊂ ¦x ¦ ϕ(x) ≤ C
0
¦. (Here by) ¦¦ ψ
ν
¦¦
2
ν
we
mean
_
e
νλ(φ)
A(ψ
ν
, ψ
ν
)dX).
Suppose now that the function φ in C
′
1
satisﬁes the following addi
tional condition.
C
′
4
. For every C > 0, the set ¦x ¦ φ(x) < C¦ ⊂⊂ X.
Remark. If C
′
4
is satisﬁed in addition to C
′
1
, C
′
2
and C
′
3
then in Theorem
1.4, we can assert that the support of ψ is compact.
Here we can state, as a corollary to Theorem 1.4 the following The
orem 1.4
′
. If the hypothesis of Theorem 1.4 are fulﬁlled and if C
′
4
holds,
then
H
q
k
(X, Ω
p
(E)) = 0.
Lemma 1.8. Let V ⊂ [
p,q
(X, E) be the set V= V
p,q
(X, E) = ¦ϕ ∈
[
pq
(X, E)
¸
¸
¸
¸
¸
there exists ψ ∈ [
p,q−1
(X, E) such that
−
∂ψ = ϕ¦ and N = 38
N
p,q
(X, E) = ¦ϕ ∈ [
p,q
(X, E) ¦ ϑϕ = 0¦. Then N is the orthogonal
complement of V provided that the metric is complete.
Proof. Let ρ ∈ [
p,q
(X, E). Then ρ ∈ ¦ orthogonal complement of
V¦ if and only if (ρ,
−
∂ψ) = 0 for every ψ ∈ [
p,q−1
(X, E) with
−
∂ψ ∈
[
p,q−1
(X, E). Since the metric is complete, this is equivalent to (ρ,
−
∂ψ) =
0 for every ψ ∈ D
p,q−1
(X, E). Hence the lemma.
Theorem 1.5. Assume conditions C
′
1
, C
′
2
, C
′
3
to hold for π : E → X. Let
f ∈ [
p,q+1
(X, E) be such that sup ϕ(x) on the support of f is C
0
. Suppose
further that ( f , g) = 0 for all g which belong to some [
p,q+1
ν
, and are
such that ϑ
ν
g = 0(ν = 1, 2, ....)(
1
). Then there exists ψ ∈ [
p,q
(X, E) such
that
−
∂ψ = f and support ψ ⊂ ¦X ¦ ϕ(X) ≤ C
0
¦.
1
In view of the choice of λ, we have ( f , g) = ( f , g)
ν
.
34 1. Vanishing theorems for hermitian manifolds
Let V
p q+1
ν
be the space V
p q+1
constructed with respect to the met
rics
(e
νλ(φ)
h, ds
2
). Since λ = 0 on supp f , then f ǫ[
pq
ν
(X, E) for all ν > 0.
By the previous lemma
f ∈ V
p,q+1
ν
for ν = 1, 2, ...
First assume that
f ∈ V
p q+1
ν
for a particular ν.
In this case, there exists ϕ
ν
∈ [
p,q
ν
(X, E) such that f =
−
∂ϕ
ν
Now by
Proposition 1.3, there is an X
ν
= G
ν
ϕ
ν
∈ W
p,q
ν
(X, E).
Such that
ν
x
ν
= ϕ
ν
. We now set ψ
ν
= ϑ
ν
¯
∂. By Proposition 1.3 ψ
ν
∈ 39
[
pq
ν
(X, E). Further we have
¯
∂ψ
ν
=
¯
∂ϑ
ν
¯
∂x
ν
=
¯
∂(
¯
∂ϑ
ν
+ ϑ
¯
∂)x
ν
=
¯
∂ϕ
ν
= f .
Since ϑ
ν
ψ
ν
= 0 , then by Theorem 1.1 ψ
ν
∈ W
pq
ν
(X, E). On the other
hand, in view of W
p,q
ν
ellipticity,
jψ
ν
j
2
ν
≤ C(j
¯
∂ψ
ν
j
2
ν
+ jϑ
ν
ψ
ν
j
2
ν
) = Cj f j
2
ν
= C
1
say C
1
is independent of ν since λ(ϕ(x)) = 0 for x ∈ support f .
Suppose now f ∈ V
pq+1
ν
, then there exists a sequence f
i
ν
∈ V
pq+1
ν
such that jf
i
ν
 f j
ν
→ 0. Then choosing ψ
i
ν
∈ W
pq
ν
(X, E) is above for
each f
i
ν
(∈ V
pq+1
ν
) we have
¯
∂ψ
i
ν
= f
i
ν
jψ
i
ν
ψ
2
ν
Cj f
i
ν
j
2
ν
and jψ
i
ν
− ψ
j
ν
j
2
ν
Cj f
i
ν
− f
l
ν
j
2
ν
.
Hence ψ
i
ν
converge to a limit ψ
ν
in [
p,q
ν
(X, E). Clearly
¯
∂ψ
ν
= F.
Further, from jψ
i
ν
j
2
ν
Cj f
i
ν
j
2
ν
, we deduce that jψ
ν
j
2
ν
Cj f j
2
. Now the
proof follows from Lemma 1.7.
The conditions C
∗
1
, C
∗
2
, C
∗
3
below are dual to C
′
1
, C
′
2
, C
′
3
: that is
if C
∗
1
, C
∗
2
, C
∗
3
hold for Evalued (p, q)forms, then C
′
1
, C
′
2
, C
′
3
hold for
E
∗
valued (n − p, n − q) forms.
C
∗
1
. (= C
′
1
= C
1
) There is given a C
∞
function φ : X → R
+
.
4. Carleman inequalities 35
C
∗
2
. For every C
o
0, there exists a nondecreasing C
∞
function, λ :
R → R such that λ(t) = 0 if t C
0
and λ(t) > 0 for t > C
0
and E
is W
pq
elliptic with respect to (ds
2
, e
νλ(φ)
h)
C
∗
3
. The W
p,q
ellipticity constant is independent of ν = 1, 2, . . .
Then we have analogous to Theorem 1.5, the following 40
Theorem 1.6. Assume C
∗
1
, C
∗
2
, C
∗
3
. Then if f ∈ [
p,q−1
(X, E), is such
that sup φ(X) on support of f = C
0
, and if further ( f , u) = 0 for every u
∈ [
p,q−1
−ν
(X, E), such that
¯
∂u = 0, then there exists χ ∈ [
p,q
(X, E) such
that f = ϑχ and support χ ⊂ ¦X¦φ(X) ≤ C¦.
Theorem 1.7. Let T = T
p,q−1
be a distribution valued form type (p, q −
1) (that is a current of type (p, q − 1)). Suppose further that
¯
∂T ∈
C
p,q
(X, E) and let K = support of T. Then for any neighbourhood U
of K, there is a form η ∈ C
p,q−1
(X, E) such that
¯
∂T =
¯
∂η and support
η ⊂ V. In particular if K is compact, we can ﬁnd an η with the above
property.
Proof. We recall (Prerequisites, 4) that we have ﬁne resolutions
0 →Ω
p
→ A
p,o
¯
∂
−→ A
p,1
→ . . .
¯
∂
−→ A
p,n
→ 0
and 0 →Ω
p
→ K
p,o
¯
∂
−→ K
p,1
→ . . .
¯
∂
−→ K
p,n
→ 0.
Further there is canonical injection A
p,q
→ K
p,q
which is compatible
with the operator
¯
∂. Hence by a standard result on cohomology of
sheaves the induced map of complexes
q0
Γ
Φ
(X, A
p,q
) →
Γ
Φ
(X, K
p,q
)
is a homotopy equivalence of complexes for any para compactifying
family Φ of closed sets on X. We consider in particular the family of
all closed sets of X which are contained in U. This is evidently a para
compactifying family and theorem follows from the general result stated
above (for a more detailed proof, see [2], 9799).
Theorems 1.5 and 1.7 together enable us to prove 41
36 1. Vanishing theorems for hermitian manifolds
Theorem 1.8. Let π : E → K be a holomorphic vector bundle. Assume
that for a suitable complete hermitian metric ds
2
on X and a suitable
hermitian metric h along the ﬁbres of E, conditions C
′
1
, C
′
2
, C
′
3
and
C
′
4
hold. Then the image
¯
∂D
pq
(X, E) of
¯
∂ in D
p,q+1
(X, E) is a closed
subspace for the usual topology on D
pq
(X, E). Hence H
q+1
k
(X, Ω
p
(E))
has a structure of a separated topological vector space.
Proof. The “usual” topology on D
pq
(H, E) may be described as fol
lows. Let ¦K
ν
¦ be an increasing sequence of compact sets such that
¯
K
ν
⊂
o
K
ν
+ 1 and UK
ν
= K. Further, let G = ¦U
i
¦
i∈I
be a locally ﬁnite
covering of X by open sets such that each U
i
is a relatively compact open
subset of a coordinate open set V
i
on X. Let G
ν
= ¦U
i
¦U
i
∩ K
ν
φ¦. We
topologize each D
pq
(K
ν
, E) = ¦ϕ¦ϕ ∈ D
pq
(X, E), supp ϕ ⊂
ν
¦ as follows:
for U
i
∈ G
ν
, ϕ¦U
i
may be regarded as a C
∞
vectorvalued function ϕ
i
on
U
i
; a fundamental system of neighbourhoods of zero in D
pq
(K
ν
, E) is
given by,
_
ϕ ∈ D
pq
(K
ν
, E)
¸
¸
¸¦
¯ α
ϕ
i
¸
¸
¸ < ε
α
, for every U
i
∈ G
ν
_
where ¦ε
α
¦ is an arbitrary family of positive reals.
D
pq
(K
ν
, E) is a Frechet space. There is a natural injection
D
pq
(K
ν
, E) →D
pq
(K
ν+1
, E).
The image of D
pq
(K
ν
, E) is a closed subset of D
pq
(K
ν+1
, E). The in 42
duced topology on the image coincides with the topology of D
pq
(K
ν
, E).
This shows that D
pq
(X, E) is a strict inductive limit of the Frechet spaces
D
pq
(K
ν
, E) [8] 6667; [9] (225227). The “usual” topology of D
pq
(X, E)
is the inductive limit topology.
A subset of D
pq+1
(X, E) is closed if, and only if, it is sequentially
closed ([19], 228). Therefore it will be suﬃcient to prove that
¯
∂D
pq
(X, E) is sequentially closed in D
pq+1
(X, E).
Let ¦ϕ
i
¦
i∈N
be a sequence of (p, q) forms of D
pq
(X, E). Such that
the sequence ¦∂ϕ
i
¦
i∈N
⊂ ∂D
pq
(X, E) converges to an element ϕ of D
pq+1
(X, E). The sequence ¦ξϕ
i
¦ is a bounded set in D
pq+1
(X, E).
4. Carleman inequalities 37
Hence there exists a compact K
s
such that ∂ϕ
i
∈ D
pq+1
(K
s
, E) for
i = 1, 2, . . ., and ϕ ∈ D
pq+1
(K
s
, E) ([8], 70; [19], 226).
Let c
o
= sup φ(x) on K
s
. Let λ : R → R be a non decreasing C
∞
function such that
λ(t) = 0 for t c
o
,
λ(t) > 0 for t > c
o
,
and let us consider the spaces [
pq+1
ν
(X, E) = [
pq+1
νλ(ϕ)
(X, E) for ν =
1, 2, . . ..
Let g ∈ [
pq+1
ν
(X, E) be such that
ϑ
ν
g = 0.
Then we have
(ϕ, g) = (ϕ, g)
ν
= lim(∂ϕ
i
, g)
ν
= lim(ϕ
i
, ϑ
ν
g) = 0.
Thus by theorem 1.5 there exists a ψ ∈ [
pq
(X, E) with compact
support (since we have assumed c
′
4
), such that 43
ϕ = ∂ψ.
Now, since in addition ϕ is C
∞
, ψ may be assumed to be a C
∞
form,
i.e. ψ ∈ D
pq
(X, E). Hence ∈ ∂D
pq
(X, E). This completes the proof of
theorem 1.8.
Chapter 2
Wellipticity on Riemannian
manifolds
5 Wellipticity on Riemannian manifolds
We carry over the results proved for ∂ on a complex manifold to the 44
operator dexterior diﬀerentiation of ordinary forms. In the sequel until
further notice we consider only real valued forms. Through all this chap
ter X will be a connected orientable Riemannian manifold. We choose
an orientation for X, and we denote by C
q
= C
q
(X) (resp. D
q
(X) = D
q
)
the space of C
∞
qforms (resp. C
∞
forms with compact support). We
deﬁned in Chapter 1 the operator
∗ : C
q
→ C
n−q
and the scalar product,
(ϕ, ψ) =
_
A(ϕ, ψ)dX =
_
X
ϕ ∧ ∗ψ
of two forms ϕ, ψ in D
q
(X).
We deﬁne the operator
δ : C
q
−→ C
q−1
39
40 2. Wellipticity on Riemannian manifolds
as the linear diﬀerential operator
ϕ (−1)
q
∗
−1
d(∗ϕ)
(ϕ ∈ C
q
(X)). If ϕ ∈ C
q
(X), ψ ∈ C
q+1
(X), and if supp ϕ ∩ supp ψ is
compact, then (dϕ, ψ) = (ϕ, δψ). We also introduce the Laplace operator
△ = dδ + δd;
△ is a strongly elliptic operator.
We have then 45
(△ϕ, ψ) = (dϕ, dψ) + (δϕ, δψ)
for ϕ, ψ ∈ D
q
(X).
On the space D
q
(X), we deﬁne another scalar product: for ϕ, ψ ∈
D
q
(X),
a(ϕ, ψ) = (ϕ, ψ) + (dϕ, dψ) + (δϕ, δψ).
Then N(ϕ) = a(ϕ, ϕ)
1
2
deﬁnes a norm on D
q
(X). We denote the comple
tion of D
q
(X) under N by W
q
. If [
q
denotes the completion of D
q
(X)
with respect to the norm, j ϕ j
2
= (ϕ, ϕ), then the identity extends canon
ically to a continuous map
i : W
q
−→ [
q
.
We have then the following result.
Proposition 2.1. i is an injection.
The proof is analogous to that of proposition 1.1. As in theorem
1.1 (Chapter 1), we have the following characterisation of W
q
(rather
i(W
q
)) when the Riemannian metric on X is complete.
Theorem 2.1. If the Riemannian metric on X is complete, then
W
q
(X) = ¦ϕ ¦ ϕ ∈ [
q
(X), dϕ ∈ [
q+1
(X), δϕ ∈ [
q−1
(X)¦.
Moreover D
q
(X) is dense in the space ¦ϕ ¦ ϕ ∈ [
q
, dϕ ∈ [
q−1
¦,
with respect to the norm (j ϕ j
2
+ j dϕ j
2
)
1
2
, and in the space ¦ϕ ¦ ϕ ∈
[
q
, δϕ ∈ [
q+1
¦ with respect to the norm (j ϕ j
2
+ j δϕ j
2
)
1
2
.
Once again, we omit the proof which is analogous to that of Theorem
1.1 (Chapter 1). We state also, without proof, the analogue of Corollary
1 to Theorem 1.2 of Chapter 1.
5. Wellipticity on Riemannian manifolds 41
Proposition 2.2. For ϕ ∈ C
q
and for any σ > 0 46
j dϕ j
2
+ j δϕ j
2
σ j ∆ϕ j
2
+
1
σ
j ϕ j
2
.
Deﬁnition 2.1. The Riemannian manifold X is W
q
elliptic if there is a
constant C > 0 such that
j ϕ j
2
C
_
j dϕ j
2
+ j δϕ j
2
_
.
We shall call such a constant C a W
q
ellipticity constant.
In further analogy to Chapter 1, we have ﬁnally the following
Theorem 2.2. If X is W
q
elliptic, then for f ∈ [
q
, there is a unique
x ∈ W
q
such tht f = △x (in the sense of distributions). Further,
j x j c j f j
j dx j
2
+ j δx j
2
C j f j
2
,
C being a W
q
− ellipticity constant. If f is C
∞
, then x ∈ C
q
∩ W
q
. If the
metric on X is complete and j d f j< ∞, then, for σ > 0,
j δdx j
2
≤
1
σ
j d f j
2
+σ j dx j
2
.
In particular, if df=0 , f=dδx and dx=0.
Except for obvious modiﬁcations, the proof is contained in Proposi
tions 1.2 and 1.3 and Theorem 1.3 (Chapter 1).
Before we give criteria for W
pq
ellipticity, on a complex manifold,
we give suﬃcient conditions for W
q
ellipticity, on an orientable Rie
mannian manifold X. In order to do this we ﬁrst write down explicitly in 47
a coordinate open set the eﬀect of the Laplacian ∆ on a C
∞
− p−form.
Let (x, . . . x
n
) be a coordinate system on an open set U in X. Let the
Riemannian metric be
g
i j
dx
i
dx
j
in this open set. We denote ▽ the covariant derivation with respect to
the Riemannian metric. For a tensor ϕ, ▽
α
ϕ denotes the tensor
(▽ϕ)(
∂
∂x
α
).
42 2. Wellipticity on Riemannian manifolds
(▽ϕ is regarded as a 1form with values in the tensor bundle). In the
localcoordinate system, a form ϕ ∈ C
q
(X) may be written in the form.
ϕ =
1
1
q!
ϕ
I
dx
I
where I runs over all qtuples (i
1
, . . . , i
q
) of integers with 0 < i
j
n and
if I = (i
1
, . . . , i
q
) is a permutation (σ( j
i
), . . . , σ( j
q
)) of I
′
= ( j
1
, . . . , j
q
),
then
ϕ
I
= ε
σ
ϕ
I
′ .
(In particular, if I has repeated indices, ϕ
I
= 0). We have
A(ϕ, ψ) =
1
q!
ϕ
I
ψ
I
(ϕ, ψ ∈ C
q
)
For ϕ ∈ C
q
and a qtuple I = (i
1
, . . . , i
q
),
dϕ
iI
=
∂ϕ
I
∂x
i
+
r
(−1)
r
∂ϕ
i
∂x
i
r
i
1
, . . . , i
r
, . . . i
q
= ▽
i
ϕ
I
+
r
(−1)
r
▽
ir
ϕ
ii
1
, . . . i
r
. . . , i
q
(2.1)
as is seen by a direct computation. Similarly we have 48
δϕ
J
= −Σ▽
i
ϕ
i
J
(2.2)
where J is a (q − 1)tuple and ϕ
i
is the (q − 1)form deﬁned by
ϕ
i
J
=
g
i j
ϕ
iJ
for any (q − 1)tuple J = ( j
1
, ... j
q−l
).
From (2.1) and (2.2), it follows that
δdϕ
I
= −
n
i=1
▽
i
(dϕ)
i
I
= −
n
i=1
▽
i
▽
i
ϕ
I
−
q
i=1
(−1)
r
▽
i
(▽
ir
ϕ
i
)i
i
...
ˆ
i
r
...i
q
(2.3)
5. Wellipticity on Riemannian manifolds 43
(where ▽
i
=
_
j
g
i j
▽
j
) and
dδϕ
I
=
q
r=I
n
i=I
(−1)
r
▽
i
r
▽
i
ϕ
i
i
1
...
ˆ
i
r
...i
q
(2.4)
Hence (▽δ)
I
= −
▽
i
▽
i
ϕ
I
+ (κϕ)
I
(2.5)
In the equation (2.5)
(κϕ)
I
=
i
r
(−1)
r
(▽
i
r
▽
i
− ▽
i
▽
i
r
)ϕ
i
i
1
...
ˆ
i
r
...i
q
=
q
r=1
j
(−1)
r−1
R
j
i
r
ϕ
ji
1
...
ˆ
i
r
...i
q
+
q
r,s=1
(−1)
r+s
R
i
j
i
r
i
s
ϕ
i. j.i
.
.
ˆ
i
r
.
ˆ
i
s
In the special case when ϕ is a function,
▽ϕ = −▽
i
▽
i
ϕ (2.6)
Lemma 2.1. For every ϕ ∈ D
q
we have 49
j ▽ϕ j
2
+(Kϕ, ϕ) =j dϕ j
2
+ j δϕ j
2
(2.7)
Proof. From (2.6) above, we have setting ¦ϕ¦
2
= A(ϕ, ϕ) =
1
ϕ
ϕ
I
ϕ
I
,
∆(¦ϕ¦
2
) = −
▽
i
▽
i
¦ϕ¦
2
= −
1
q!
▽
i
▽
i
ϕ
I
ϕ
I
= −
2
q!
▽
i
(▽
i
ϕ
I
.ϕ
I
)
= −
−2
q!
▽
i
▽
i
ϕ
I
.ϕ
I
− 2¦▽ϕ¦
2
=
I
_
2
q!
▽ϕ
I
ϕ
I
−
2
q!
(Kϕ)
I
ϕ
I
_
− 2¦▽ϕ¦
2
That is
▽¦ϕ¦
2
= 2A(▽ϕ, ϕ) − 2A(Kϕ, ϕ) − 2A(▽ϕ, ▽ϕ)...
If ϕ has compact support,
_
▽¦ϕ¦
2
dx =
_
dδ¦ϕ¦
2
dX = 0
44 2. Wellipticity on Riemannian manifolds
by Stokes formula.
On the other hand,
_
A(▽ϕ, ϕ)dX =j dϕ j
2
+ j δϕ j
2
.
Hence
(Kϕ, ϕ)+ j ▽ϕ j
2
=j dϕ j
2
+ j δϕ j
2
: (ϕ ∈ D
q
(X)).
If there exists a positive constant C such that
A(Kϕ, ϕ) ≥ C¦ϕ¦
2
for every ϕ ∈ C
q
and at each point of X, then
¦¦ϕ¦¦
2
≤ C
′
(Kϕ, ϕ) ≤ C
′
_
¦¦dϕ¦¦
2
+ ¦¦δϕ¦¦
2
_
where C
′
=
1
C
. Hence the following.
Lemma 2.2. If there is a positive constant C such that 50
A(Kϕ , ϕ) ≥ C A(ϕ, ϕ)
for every ϕ ∈ C
q
and at each point of X, then the Riemannian manifold
X is W
q
elliptic.
6 A maximum principle
In the sequel we consider weaker condition on the expression A(κϕ , ϕ).
We shall deal with the case where the quadratic from A(κϕ , ϕ) is positive
semi deﬁnite outside of a compact.
From now on we shall always assume X to be oriented and con
nected, and the Riemannian metric of X to be complete.
Lemma 2.3. Assume A(κϕ , ϕ) ≥ 0 (i.e. A(κϕ, ϕ) positive semideﬁnite),
outside a compact set K in X. Then for ϕ ∈ W
q
, j ▽ϕ j
2
< ∞ and (2.7)
holds.
6. A maximum principle 45
Proof. By Lemma 2.1 we have for ϕ ∈ 1
q
(X),
j ▽ϕ j
2
+(κϕ, ϕ) = (dϕ , dϕ) + (δϕ, δϕ).
Since 7 is compact, there is C ≥ 0 such that (7ϕ, ϕ) > −CC¦ϕ¦
−2
for every ϕǫC
q
and each point of 7. Hence,
j ▽ϕ j
2
+
_
X
A(kϕ, ϕ) ≤ C j ϕ j
2
k
+ j dϕ j
2
+ j δϕ j
2
.
Since A(κϕ, ϕ) ≥ 0 on X − 7 we have for ϕ ∈ 1
q
,
j ▽ϕ j
2
≤ C
_
j ϕ j
2
+ j dϕ j
2
+ j δϕ j
2
_
The lemma follows now from the fact that 1
q
is dense in W
q
. (Theorem
2.1.)
Remark. We have proved more: there is a C > 0 such that 51
j▽ϕj
2
C
_
jϕj
2
K
+ jdϕj
2
+ jδϕj
2
_
Let X be a manifold as in the above lemma.
Let ϕ ∈ C
q
. Identity (2.8) shows that △¦ϕ¦
2
0 at each point of the
set Y where △ϕ = 0 and A(Kϕ, ϕ) 0
Applying a classical lemma of E. Hopf ([36], 2630) we see that
¦ϕ¦
2
cannot have a relative maximum at any point of Y. Thus, if X is
compact, ¦ϕ¦
2
takes its, maximum in the set Supp f ∪ K.
If X is not compact we cannot draw the same conclusion. However
the following proposition will provide an estimate of ¦ϕ¦
2
on X in terms
of Sup ¦ϕ¦
2
on Supp f ∪ 7.
Proposition 2.3. Let X be a connected oriented and complete Rieman
nian manifold. Assume give a compact set 7 on X such that for X
7, A(7ϕ, ϕ)(x) ≥ 0 for every ϕ ∈ C
q
. Suppose that ϕ ∈ C
q
∩ [
q
is such
that △ϕ = f ∈ [
q
. Then ϕ ∈ W
q
and sup
x∈X
¦ϕ(x)¦
2
C
0
where
C
0
=
Sup
x∈supp f ∪7
¦ϕ(x)¦
2
.
46 2. Wellipticity on Riemannian manifolds
For the proof of this proposition we utilise the following result
Lemma 2.4 (Gaﬀney [12]). Let 7 be an oriented complete Riemannian
manifold and ω a C
∞
1form. Then if
_
X
¦ω¦dX < ∞and
_
X
¦δω¦dX < ∞,
we have
_
X
δωdX = 0.
Proof. Let λ be a real value C
∞
function on such that λ(t) = 1 for 52
t 0 and λ(t) = 0 for t 1. Let p
0
∈ X be a ﬁxed point and ρ = ρ(x)
denote the distance function (from p
0
). Then ρ(x) is locally Lipchitz as
was remarked earlier (Lemma B of Chapter 1). Let 0 < r < F and let us
consider the form
λ
_
ρ(x) − r
F − r
_
ω
This form is locally Lipchitz and has compact support in the ball
¦x¦ρ(x) R¦.
Hence by Stoke’s formula (which is applicable to Lipchitz continu
ous form),
_
X
d ∗ λ
_
ρ(x) − r
R − r
_
ω = 0;
We have,
_
X
1
R − r
λ
_
ρ(x) − r
−r
_
dρΛ ∗ ω +
_
X
λ
_
ρ(x) − r
R − r
_
d ∗ ω = 0.
Now set R = 2r. Then
_
X
1
r
λ
′
_
ρ(x) − r
r
_
dρΛ ∗ ω +
_
X
λ
_
ρ(x) − r
r
_
d ∗ ω = 0
Now ¦dρ¦
2
(lemma B of Chapter 1).
Hence, since ∗ω is integrable, we have 53
_
X
¦dρΛ ∗ ω¦dX < ∞.
6. A maximum principle 47
Hence, using the fact that λ
′
is bounded, we see that
1
r
_
X
λ
′
_
ρ(x) − r
r
_
dρΛ ∗ ω → 0 as → ∞.
It follows that
Lt
r → ∞
_
X
λ
_
ρ(x) − r
r
_
d ∗ ω = 0
On the other hand, since
_
¦d ∗ ω¦dX < ∞ and λ
_
ρ(x)−r
r
_
is bounded
and tends to 1 as r → ∞,
r
Lt
→ ∞
_
λ
_
ρ(X) − r
r
_
d ∗ ω =
_
d ∗ ω.
This proves the lemma.
Proof (of Proposition 2.3) Let C
0
and C
1
be two positive constants,
0 < C
0
< C
1
, and let λ : R → R be a C
∞
function such that
λ(t) 0, λ
′
(t) 0, λ
′′
(t) 0,
λ(t) = 0, for t C
0
λ
′
(t) > 0, for t > C
0
λ
′′
(t) = 0, for t < C
0
and for t > C
1
,
and such that λ
′
(t) be bounded. Since λ
′′
(t) has compact support, it is
bounded. We have then for ϕ as in the proposition,
∆λ(¦ ϕ ¦
2
) = δd(λ(¦ ϕ ¦
2
))
= δ
_
λ
′
(¦ ϕ ¦
2
).d ¦ ϕ ¦
2
_
= − ∗ d ∗
_
λ
′
(¦ ϕ ¦
2
)d ¦ ϕ ¦
2
_
= − ∗ d
_
λ
′
(¦ ϕ ¦
2
) ∗ d ¦ ϕ ¦
2
_
= − ∗
_
dλ
′
(¦ ϕ ¦
2
) ∧ ∗d ¦ ϕ ¦
2
+λ
′
(¦ ϕ ¦
2
)d ∗ d ¦ ϕ ¦
2
_
= − ∗
_
λ
′′
(¦ ϕ ¦
2
)d ¦ ϕ ¦
2
∧ ∗ d ¦ ϕ ¦
2
+λ
′
¦ ϕ ¦
2
d ∗ d ¦ ϕ ¦
2
_
= −λ
′′
¦ ϕ ¦
2
¦ d ¦ ϕ ¦
2
¦
2
+ λ
′
¦ ϕ ¦
2
.∆(¦ ϕ ¦
2
).
48 2. Wellipticity on Riemannian manifolds
In the view of (2.8), 54
△(λ¦ϕ¦
2
) = −λ
′′
(¦ϕ¦
2
)¦d¦ϕ¦
2
¦
2
+ 2λ
′
¦ϕ¦
2
¦A(△ϕ, ϕ) − A(ϕ, ϕ) − A(▽ϕ, ▽ϕ)¦
On the other hand by proposition 2.2 we have for ϕ ∈ C
q
,
¦¦∂ϕ¦¦
2
+ ¦¦δϕ¦¦
2
¦¦∆ϕ¦¦
2
+ ¦¦ϕ¦¦
2
In particular, if ∆ϕ ∈ [
q
and ϕ ∈ C
q
∩ [
q
then ¦¦dϕ¦¦
2
< ∞ as also
¦¦δϕ¦¦
2
< ∞. In the view of theorem 2.1 this proves that ϕ ∈ W
q
.
Now
_
x
∆λ(¦ϕ¦
2
)dX =
_
X
∗d ∗ dλ(¦ϕ¦
2
)
We shall show that these integrals vanish.
Using Schwartz inequality, one checks that there exists a positive
constant C
2
such that
¦d¦ϕ¦
2
¦ C
2
¦ϕ¦¦▽ϕ¦ at each point of X. (2.10)
Hence, since ¦λ
′
(t)¦ is bounded it follows that
_
X
¦dλ(¦ϕ¦
2
)¦dX < ∞.
We will prove now that
_
X
¦∆λ(¦ϕ¦
2
)¦dX < ∞.
Letting C
3
= Sup λ
′′
(t), we have, by (2.10),
_
X
λ
′′
(¦ϕ¦
2
)¦d¦ϕ¦
2
¦
2
dX =
_
λ:(¦ϕ¦
2
2
)¦d¦ϕ¦
2
¦
2
C
◦
<¦ϕ¦
2
<C
1
dX C
3
_
¦d¦ϕ¦
2
¦
2
C
◦
<¦ϕ¦
2
<C
1
¦d¦ϕ¦
2
¦
2
dX
C
2
Cc
3
_
¦ϕ¦
2
¦▽ϕ¦
2
C
◦
<¦ϕ¦
2
<C
1
dX C
1
C
2
C
3
_
X
¦▽¦
2
dX = C
1
C
2
C
3
¦¦▽ϕ¦¦
2
Since ϕ ∈ W
q
, then by Lemma 2.3 ¦¦▽ϕ¦¦
2
< ∞, and therefore 55
_
X
λ
′′
(¦ϕ¦
2
)¦d¦ϕ¦
2
¦
2
dX < ∞
6. A maximum principle 49
Since K is compact, there is a constant C
4
> 0 such that
¦A(7u, u)¦ C
4
A(u, u)
for any u ∈ C
q
and at each point of 7. Then we have
_
X
¦A(Kϕ, ϕ)¦dX =
_
X−7
A(7ϕ, ϕ)dX +
_
7
¦A(7ϕ, ϕ)¦dX
_
X
A(7ϕ, ϕ)dX + 2
_
7
¦A(7ϕ, ϕ)¦dX
(7ϕ, ϕ) + 2C¦¦ϕ¦¦
2
2C¦¦ϕ¦¦
2
+ ¦¦dϕ¦¦
2
+ ¦¦δϕ¦¦
2
< ∞.
Finally we have
_
X
¦A(△ϕ, ϕ)¦dX ≤
1
2
_
X
A(△ϕ, △ϕ)dX +
1
2
_
X
A(ϕ, ϕ)dX < ∞.
Then, it follows from (2.9) and from the fact that ¦λ
′
¦ and ¦λ
′′
¦ are
bounded, that
_
X
¦△λ(¦ϕ¦
2
)¦dX < ∞.
Hence, by Lemma 2.4,
_
X
△λ(¦ϕ¦
2
)dX = 0, i.e. by (2.9)
_
X
λ
′′
(¦ϕ¦
2
)
2
(d¦ϕ¦
2
)
2
dX + 2
_
X
λ
′
(¦ϕ¦
2
).A(7ϕ, ϕ)dX + 2
_
X
λ
′
(¦ϕ¦
2
)¦▽ϕ¦
2
dX =
_
X
2λ
′
(¦ϕ¦
2
)A(△ϕ, ϕ)dX. (2.11)
Let us consider now sup ¦ϕ¦ on Supp f ∪ k. If it is not ﬁnite there is 56
nothing to prove. If it is ﬁnite, we set C
0
= Sup ¦ϕ¦
2
on K
_
Support f .
In view of our choice of the function λ all the terms on the left hand side
of (2.11) are non negative. We obtain
0 2
_
X
λ
′
(¦ϕ¦
2
)¦▽ϕ¦
2
dX
_
X
2λ
′
(¦ϕ¦
2
)A(△ϕ, ϕ)
50 2. Wellipticity on Riemannian manifolds
= 2
_
X
λ
′
(¦ϕ¦
2
).A( f , ϕ).
Once again since ¦ϕ¦
2
C
0
on support f , the integral on the right is
zero. Hence
_
X
λ
′
(¦ϕ¦
2
)¦▽ϕ¦
2
dX = 0
By consequences we have also d¦ϕ¦
2
= 0, i.e. ¦ϕ¦ is a constant, on the
open set ¦¦ϕ¦
2
> C
o
¦. Hence ¦ϕ¦
2
≤ C
o
on X. This concludes the proof.
Corollary. If ϕ ∈ [
q
∩ C
q
and △ϕ ∈ D
q
(in particular, if △ϕ = 0), then
¦ϕ¦ is bounded on X.
7 Finite dimensionality of spaces of harmonic forms
Let γ be a real number and let M
q
γ
be the vector space
M
q
γ
= ¦ϕ¦ϕ ∈ [
q
, △ϕ = γϕ¦.
we set M
q
= H
q
Lemma 2.5. M
γ
⊂ W
q
and γ 0.
Proof. If ϕ ∈ M
q
γ
, then △ϕ ∈ [
q
. Thus, by the ﬁrst part of Proposition
2.3, ϕ ∈ W
q
. Since ϕ is a solution of the elliptic equation ∆−γ, we may 57
assume ϕ ∈ C
q
. We have moreover (since the metric is complete)
jδdϕj
2
j∆dϕj
2
+ jdϕj
2
(γ
2
+ 1)jdϕj
2
,
jdδϕj
2
j∆δϕj
2
+ jδϕj
2
(γ
2
+ 1)jδϕj
2
.
Hence dϕ ∈ W
q+1
, δϕ ∈ W
q−1
and therefore
γjϕj
2
= (∆ϕ, ϕ) = jdϕj
2
+ jδϕj
2
(2.12)
This proves that γ 0
It follows from (2.12) that, if γ = 0, i.e. if ϕ ∈ H
q
, then dϕ = 0,
δϕ = 0.
7. Finite dimensionality of spaces of harmonic forms 51
Theorem 2.3. Let X be a connected, oriented and complete Riemannian
manifold and assume that A(7ϕ, ϕ) C¦ϕ¦
2
for all ϕ ∈ C
q
for some
C > 0, outside a compact set K ⊂ X. Then for γ < C, M
q
γ
is ﬁnite
dimensional. In particular dim H
q
< ∞.
Proof. Let ϕ ∈ M
q
γ
. Since ϕ ∈ W
q
, then by Lemma 2.3
j▽ϕj
2
+ (7ϕ, ϕ) = jdϕj
2
+ jδϕj
2
and by (2.12)
j▽ϕj
2
+ (7ϕ, ϕ) = jdϕj
2
+ j¦¦δϕj
2
= γjϕj
2
Now 7 being compact, there is a constant C
2
> 0 such that (7u, u)
−C
2
¦u¦
2
for every u ∈ C
q
and at each point of 7. Since (7ϕ, ϕ) C
2
¦ϕ¦
2
on the complement X − 7 we have
j▽ϕj
2
+ Cjϕj
2
X−7
C
2
jϕj
2
7
+ γjϕj
2
or again,
j▽ϕj
2
+ Cjϕj
2
(C + C
2
)jϕj
2
7
+ γjϕj
2
Hence, 58
(C − γ)jϕj
2
(C + C
2
)jϕj
2
K
(2.13)
Suppose now that M
q
γ
is inﬁnite dimensional for γ < C, then the evalu
ation map
ω ω(x) (x ∈ K)
(which associates to a form ω, the element of the q
th
exterior power of
the tangent space at x which is deﬁned by ω) is a linear map into a ﬁnite
dimensional space. Hence the kernel of this map is of ﬁnite codimen
sion. Since the ﬁnite intersection of subspaces of ﬁnite codimension
is nonzero (over an inﬁnite ﬁeld !) it follows that given any sequence
x
1
, . . . , x
ν
, . . . of points of K, we can ﬁnd forms ϕ
∈
∈ M
q
γ
such that
ϕ
ν
(x
i
) = 0 for i ν. We may further assume that jϕ
ν
j = 1. It follows
then that we can choose a subsequence ψ
k
= ϕ
ν
k
of ¦ϕ
ν
¦ such that ψ
k
converges weakly to a limit ψ in [
q
. Now this implies that ψ ∈ M
q
γ
,
since for any ϕ ∈ D
q
,
(ψ, ∆ϕ) = Lt(ψ
k
, ∆ϕ) = Lt(∆ψ
k
, ϕ) = γ(ψ, ϕ).
52 2. Wellipticity on Riemannian manifolds
Hence we can assume ψ ∈ C
q
. Now the ψ
k
are solutions of the elliptic
equation ∆ − γ = 0. Hence the weak convergence of ψ
k
implies that ψ
k
converge uniformly together with all partial derivatives on every com
pact set. Clearly we have Lt ψ
k
(x
ν
) = 0 for every ν. If we choose x
ν
to
be dense in K then ψ ≡ on K since ψ ∈ C
q
. Hence it follows that
¦ψ
k
¦ < ε on K.
for any given ε > 0 and all large k. We conclude that 59
jψ
k
j
2
K
·−→ 0
as k → ∞. On the other hand since ψ
k
∈ M
q
γ
, we have by (2.13),
(C − γ)jψ
k
j
2
(C + C
2
)jψ
K
j
2
K
.
Since C − γ > 0 and jψ
k
j
2
= 1, we arrive at contradiction. Hence
M
q
γ
is ﬁnite dimensional for every γ < C. This proves Theorem 2.1
completely.
Theorem 2.4. Under the hypothesis of Theorem 2.3 we have the fol
lowing. Let S
q
denote the orthogonal complement of H
q
in the Hilbert
space W
q
. Then there is a C > 0 such that
jϕj
2
C¦jdϕj
2
+ jδϕj
2
¦ for ∀ϕ ∈ W
q
For the proof of the theorem we need the following result due to
Rellich.
Lemma 2.6. Let X be a Riemannian manifold and Ω ⊂⊂ X Let ϕ
ν
∈
[
q
(Ω) be a sequence such that
jϕ
ν
j
2
< ∞ and j▽ϕ
ν
j
2
Ω
< ∞
(We denote
_
Ω
¦▽ϕ
ν
¦
2
dX by j▽ϕ
ν
j
2
Ω
). Suppose that ¦¦ϕ
ν
¦¦
2
Ω
+¦¦▽ϕ
ν
¦¦
2
Ω
M for a ﬁxed constant M > 0. Then if ∂Ω is smooth, we can ﬁnd a
subsequence ψ
k
= ϕ
ν
k
of ϕ
ν
such that ¦¦ψ
k
− ψ
k
′ ¦¦
Ω
→ 0 as k, k
′
→ ∞.
If ϕ
ν
ǫ[
q
2
(X) is a sequence such that j ϕ
ν
j
2
+ j ▽ϕ
ν
j
2
≤ M for a 60
ﬁxed constant M > 0, then we can ﬁnd a subsequence ψ
k
= ϕ
ν
k
such
that for each compact K
′
⊂ X,
j ψ
k
− ψ
k
′ j
K
→ 0 as k, k
′
→ ∞.
7. Finite dimensionality of spaces of harmonic forms 53
Proof. The set Ω being relatively compact, it is suﬃcient to prove the
lemma in the case q = 0, i.e. in the case of L
2
functions. For this proof
see e.g. [7], 339.
Proof of Theorem 2.4. Suppose that the theorem is false. Then we can
ﬁnd a sequence ϕ
ν
ǫW
q
such that
j dϕ
ν
j
2
+ j δϕ
ν
j
2
+jϕ
ν
j
2
= 1
while jdϕ
ν
j + jδϕ
ν
j
2
≤
1
ν
jϕ
ν
j
2
. In view of Remark under Lemma 2.3
this implies that
j▽ϕ
ν
j
2
≤ C
1
for some C
1
> 0. By Rellich’s Lemma (Lemma 2.6) we can ﬁnd a ϕǫ[
q
(by passing of a subsequence if necessary) such that
jϕ
ν
− ϕj
Ω
→ 0
for every relatively compact Ω ⊂ X.
In particular, we have
jϕ
ν
− ϕj
2
K
→ 0.
We assert that ϕ is harmonic, that is △ϕ = 0. In fact for u ∈ 1
q
, we 61
have
(ϕ, △u) = Lt(ϕ
ν
, △u)
= Lt(dϕ
ν
, du) + (δϕ
ν
, δu)
≤ Lt(jdϕ
ν
j jduj + jδϕ
ν
j.jδuj)
≤ lim
1
√
2
jϕ
ν
j(jduj + +jδuj) = 0
since jϕ
ν
j
2
is bounded. Hence ϕǫW
q
∩ H
q
.
Now we have for f ǫW
q
,
j▽f j
2
+ (7 f , f ) = jdfj
2
+ jδ f j
2
.
54 2. Wellipticity on Riemannian manifolds
We have (7 f , f ) ≥ C¦ f ¦
2
on X − K . On the other hand there is a
C
2
≥ 0 such that (7 f , f ) > −C
2
¦ f ¦
2
on K. Since j▽f j
2
≥ o, we obtain,
Cj f j
2
≤ (C + C
2
)j f j
2
k
+ jd f j
2
+ jδ f j
2
.
Setting f = ϕ − ϕ
ν
, we obtain,
Cjϕ − ϕ
ν
j
2
≤ (C + C
2
)jϕ − ϕ
ν
j
2
k
+ jd(ϕ − ϕ
ν
)j
2
+ jδ(ϕ − ϕ
ν
)j
2
.
Now ϕǫH
q
∩C
q
so that
(dϕ, dϕ) + (δϕ, δϕ) = (△ ϕ, ϕ) = 0.
Hence
Cjϕ − ϕ
ν
j
2
≤ (C + C
2
)jϕ − ϕ
ν
¦¦
2
k
+ ¦¦dϕ
ν
¦¦
2
+ ¦¦δϕ
ν
¦¦
2
≤ (C + C
2
)¦¦ϕ − ϕ
ν
¦¦
2
k
+
1
ν
jϕ
ν
j
2
.
It follows that ϕ
ν
→ ϕ in [
q
on the whole of X. On the other hand 62
since
jdϕ
ν
j
2
+ ¦¦δϕ
ν
¦¦
2
≤
1
ν
jϕ
ν
¦¦
2
dϕ
ν
and δϕ
ν
converge to zero. Hence ϕ
ν
converges in W
q
to a limit ϕ
satisfying darphi = 0, δϕ = 0, i.e. ϕǫH
q
. But ϕ
ν
ǫS
q
and S
q
is a closed
subspace. Hence ϕǫS
q
and therefore ϕ = 0. But then ϕ
ν
→ 0 in W
q
, and
this is absurd, since ¦¦ϕ
ν
¦¦
2
+ ¦¦dϕ
ν
¦¦
2
+ ¦¦δϕ
ν
¦¦
2
= 1. It follows that there
exists C
′
> 0 such that for ϕǫS
q
C
′
¦¦¦dϕ¦¦
2
+ ¦¦δϕ¦¦
2
¦ ≥ ¦¦ϕ¦¦
2
Remark. Since for ϕǫH
q
, dϕ = 0, δϕ = 0, S
q
is also the orthogonal
complement of H
q
with respect to the scalar product on W
q
induced by
[
q
.
The hypothesis of Theorem 2.3 can be weakened, further the space
H
q
.
7. Finite dimensionality of spaces of harmonic forms 55
Theorem 2.5. Let X be an oriented, connected and complete Rieman
nian manifold and let A(7ϕ, ϕ) ≥ 0 outside a compact set 7 for every
ϕ ∈ C
q
. Then dimH
q
< ∞.
Proof. If dimH
q
= ∞, then given any dense sequence ¦x
ν
¦ of points
x
ν
ǫX, we can ﬁnd a sequence ¦ϕ
ν
¦ of forms ϕ
ν
ǫH
q
such that
ϕ
ν
(x
i
) = 0 for i ≤ ν, ¦¦ϕ
ν
¦¦
2
= 1.
We assert now that ¦¦ϕ
ν
¦¦
Ω
> 0 for any relatively compact open subset
Ω ⊃ K. In fact, we have
¦¦▽ϕ
ν
¦¦
2
+ (7ϕ
ν
ϕ
ν
) = ¦¦dϕ
ν
¦¦
2
+ ¦¦δϕ
ν
¦¦
2
= o since ϕ
ν
ǫH
q
.
Hence from the Remark under Lemma 2.3 it follows that there is a 63
constant C > 0 such that
j ▽ϕ
ν
j
2
< C j ϕ
ν
j
2
K
(2.14)
Hence if j ϕ
ν
j
Ω
= 0, then j ϕ
ν
j
2
K
= 0 so that ▽ϕ
ν
= 0. But in the
case, ¦ ϕ
ν
¦
2
is a constant. Hence j ϕ
ν
j
Ω
> 0, a contradiction.
We may therefore assume that for a ﬁxed Ω ⊃ K j ϕ
ν
j
Ω
= 1. From
(2.14),j ϕ
ν
j
2
Ω
+ j ▽ϕ
ν
j
2
Ω
< C
′
for some C
′
> 0. Hence by Rellich’s
Lemma (Lemma 2.6), we can, by passing to a subsequence, if necessary,
assume that j ϕ
ν
− ϕ
µ
j
2
Ω
→ 0 as ν, µ → ∞ provided that ∂Ω is smooth.
Hence ¦ϕ
ν
¦ converges to a limit ϕ in [
q
(Ω).
But since ¦ϕ
ν
¦ is a sequence of solutions of an elliptic operator,
ϕ
ν
converges to ϕ uniformly on every compact subset of Ω. Since
ϕ
ν
(x
i
) = 0 for i ν, we see that ϕ ≡ 0 on Ω. Hence jϕ
ν
j
Ω
→ 0 as
ν → ∞, a contradiction. It follows that H
q
is ﬁnite dimensional. Hence
the theorem. An estimate for the dimension of H
q
is provided by the
following.
Proposition 2.3. Assume that, under the same hypotheses for X, 7 ≡ 0
on X. Then for ϕǫH
q
, ▽ϕ = 0. Hence, dimH
q
_
n
q
_
. Moreover, if
dimH
q
> 0, then vol X < ∞.
56 2. Wellipticity on Riemannian manifolds
Proof. Since now 64
j ▽ϕ j
2
= jdϕj
2
+ jδϕj
2
= 0 for ϕǫH
q
we have ▽ϕ = 0 for ϕǫH
q
. A form invariant under parallel translation
is determined by its value at one point. Hence dim H
q
_
n
q
_
. If ϕǫH
q
is nonzero, then jϕj
2
< ∞; on the other hand since ▽ϕ = 0, ¦ ϕ ¦
2
is a
constant and the last assertion follows.
Lemma 2.7. Under the hypotheses of Theorems 2.3 and 2.4, for every
sǫS
q
there exists a unique xǫS
q
such that
s = △x
in the sense of distributions. Moreover
j x j C
′
j s j,
jdx¦
2
+ ¦¦δxj
2
≤ C
′
j s j
2
C
′
being the constant which appears in Theorem 2.4. If s ǫC
q
∩S
q
, then
(x can be modiﬁed on a null set in sch a way that) xǫC
q
∩ S
q
and the
equation △x = s holds in the ordinary sense.
Proof. S
q
as a (closed) subspace of the Hilbert space W
q
is a Hilbert
space. By theorem 2.4 the norm N is equivalent on S
q
to the norm
(¦¦d¦¦
2
+¦¦δ¦¦
2
)
1
2
. f (s, f ) is a continuous linear form on S
q
. Thus there
is a unique x ∈ S
q
such that
(s, f ) = (dx, d f ) + (δx, δ f )for all f ǫS
q
.
Let now ϕ be any element of W
q
and let h and ψ be its orthogonal
projections into H
q
and S
q
. They are uniquely deﬁned, and furthermore
ϕ = h + ψ
Since h is orthogonal to s and dh = 0, δh = 0, then 65
(s, ϕ) = (s, ψ) = (dx, dψ) + (δx, δψ) = (dx, dϕ) + (δx, δϕ).
This proves the lemma.
8. Orthogonal decomposition in [
q
57
Since
W
q
= H
q
⊕ S
q
we can state the following.
Proposition 2.4. Under the hypotheses of Theorem 2.3, for every ϕǫW
q
there exists a unique hǫH
q
and a unique xǫS
q
such that
ϕ = h + △x
(in the sense of distributions). Moreover
¦¦x¦¦ C
′
¦¦ϕ¦¦
¦¦dx¦¦
2
+ ¦¦δx¦¦
2
C
′
¦¦ϕ¦¦
2
8 Orthogonal decomposition in [
q
Let X be an oriented, complete and connected Riemannian manifold.
Assume that there exists a compact set K ⊂ X and a positive constant C
such that
A(7ϕ, ϕ) CA(ϕ, ϕ)
outside K, for each ϕǫC
q
.
The space [
q
can be decomposed as direct sum of orthogonal sub
spaces
[
q
= H
q
⊕ dD
q−1
⊕ δD
q+1
,
where dD
q−1
and δD
q+1
are the closure of d D
q−1
and [1
q+1
with 66
respect to the norm j j.
Let ϕ ∈ [
q
;  if dϕ = 0, then the orthogonal projection of ϕ into
δ1
q+1
is zero; if δϕ = 0, then the orthogonal projection into d1
q−1
is
zero (see [17] 602605; [29], 165).
Lemma 2.8. For every φ ∈ d1
q−1
there is a unique form x ∈ S
q
such
that
φ = dδx, dx = 0.
58 2. Wellipticity on Riemannian manifolds
Moreover
j x j≤ C
′
j φ j,
j δx j
2
≤ C
′
j φ j
2
(C
′
being the constant introduced in Theorem 2.4)
Proof.
s (φ, s)
is a continuous form on S
q
. Then there exists a unique x ∈ S
q
such that
(φ, s) = (dx, ds) + (δx, δs) for all s ∈ S
q
.
Moreover, by Theorem 2.4,
j x j
2
≤ C
′
(j dx j
2
+ j δx j
2
) = C
′
(φ, x) ≤ C
′
j φ j . j x j,
j dx j
2
+ j δx j
2
= (φ, x) ≤j φ j . j x j≤ C
′
j φ j
2
.
Let now u ∈ 1
q
u can be written
u = h + s, h ∈ H
q
, s ∈ S
q
.
Since h⊥ d1
q−1
in [
q
and dh = 0, δh = 0, then (ϕ, u) = (ϕ, s) = 67
(dx, ds) + (δx, δs) = (dx, du) + (δx, δu) = (x, △u) i.e.
ϕ = △x
in the sense of distributions.
Let φ ∈ C
q
∩d1
q−1
. By the regularity theorem, we can modify x on
a null set in such a way that x ∈ C
q
∩ S
q
, and that the latter equation
holds in the ordinary sense. Thus, by proposition 2.2
j δdx j
2
≤
1
σ
j dφ j
2
+σ j dx j
2
= σ j dx j
2
for all σ > 0.
Hence
δdx = 0, ϕ = dδx.
8. Orthogonal decomposition in [
q
59
But then, by Theorem 2.1, δx ∈ W
q−1
, and therefore
j dx j
2
= (x, δdx) = 0 i.e. dx = 0.
If ϕ is not C
∞
, there exists a sequence ¦u
ν
¦ of forms u
ν
∈ 1
q−1
such
that j du
ν
− ϕ j−→ 0. Clearly du
ν
⊥ H
q
; hence du
ν
∈ S
q
. Setting
ϕ
ν
= du
ν
and applying to ϕ
ν
the above argument, we can ﬁnd, for each
ϕ
ν
, a unique x
ν
∈ S
q
∩ C
q
such that
ϕ
ν
= dδx
ν
, dx
ν
= 0
j x
ν
j≤ C
′
j ϕ
ν
j,
j δx
ν
j
2
≤ C
′
j ϕ
ν
j
2
Hence ¦x
ν
¦ is a Cauchy sequence in S
q
. Let 68
x = lim x
ν
.
Then
dx = 0,
j x j ≤ C
′
j ϕ j,
j δx j
2
≤ C
′
j ϕ j
2
,
and ﬁnally
ϕ = dδx,
in the sense of distributions. This proves the lemma.
An analogous argument yields
Lemma 2.9. For any ψ ∈ δ1
q+1
there is a unique form y ∈ S
q
such that
ψ = δ dy δ y = 0
Moreover
j y j ≤ C
′
j ψ j
j dy j
2
≤ C
′
j ψ j
2
60 2. Wellipticity on Riemannian manifolds
Theorem 2.6. Every current f ∈ [
q
can be decomposed as the sum of
three currents
f = h + dϕ
1
+ δψ
1
with h ∈ H
q
, ϕ
1
∈ W
q−1
, ψ
1
∈ W
q+1
. Moreover
N(h)
2
=j h j
2
≤j f j
2
, N(ϕ
1
)
2
≤ (C
′
+1) j ϕ j
2
, N(ψ
1
)
2
≤ (C
′
+1) j ψ j
2
Proof. First of all any element f ∈ [
q
can be expressed in a unique way
as the sum
f = h + ϕ + ψ
of three forms h ∈ H
q
, ϕǫ dD
q−1
, ψǫδD
q+1
69
¦¦ f ¦¦
2
= ¦¦h¦¦
2
+ ¦¦ϕ¦¦
2
+ ¦¦ψ¦¦
2
.
Next we apply to ϕ and ψ Lemma 2.8 and 2.9 setting then
ϕ = δϕ
1
, ψ = dψ
1
;
ϕ
1
∈ W
q−1
ψ
1
∈ W
q+1
.
Moreover
N(ϕ
1
)
2
= ¦¦ϕ
1
¦¦
2
+ ¦¦ϕ¦¦
2
≤ (C
′
+ 1)¦¦ϕ¦¦
2
≤ (C
′
+ 1)¦¦ f ¦¦
2
N(ψ
1
)
2
= ¦¦ψ
1
¦¦
2
+ ¦¦ψ¦¦
2
≤ (C
′
+ 1)¦¦ψ¦¦
2
≤ (C
′
+ 1)¦¦ f ¦¦
2
Remark. If the hypotheses of Theorem 2.3 are satisﬁed, not only for the
forms of degree q, but also for those degree q − 1 and q + 1 then spaces
H
q−1
, H
q+1
, S
q−1
and S
q+1
can be introduced and one checks that
ϕ
1
⊥ H
q−1
, i.e. ϕ
1
∈ S
q−1
ϕ
1
⊥ H
q+1
, i.e. ψ
1
∈ S
q+1
.
Chapter 3
Local expressions for and
the main inequality
9 Metrics and connections
We now go back to the case of holomorphic vector bundles on a complex 70
manifold.
Let x be a complex manifold and π : E → X a holomorphic vector
bundle. Let 1 = ¦U
i
¦
iǫI
be a covering of X such that π : E → X is
deﬁned with respect to 1 by transition functions e
i j
: U
i
∩ U
j
→ GL
(m, C) (E is of rank m).
Let (h
i
)
iǫI
be a hermitian metric along the ﬁbres of E: then h
i
are C
∞
functions U
i
whose values are positive deﬁnite hermitian matrix such
that.
h
i
=
t−
e
ji
h
j
e
ji
=
t
e
−1
i j
h
j
e
−1
i j
on U
i
∩ U
j
.
Consider the 1form l
i
= h
−1
i
∂h
i
where ∂ is the exterior diﬀerentia
tion with respect holomorphic coordinates. We have for this family of
forms the following.
Lemma 3.1. If l
i
= h
−1
i
∂h
i
, then l
i
are (1, 0) forms with values in M
m
(C)
and on U
i
∩ U
j
we, have
l
i
= e
−1
ji
l
j
e
ji
+ e
−1
ji
∂e
ji
.
61
62 3. Local expressions for and the main inequality
In other words l
i
are the local representations of a connection of type 71
(1, 0) on the principal bundle associated to E.
Proof. We have
e
−1
ji
l
j
e
ji
+ e
−1
ji
∂e
ji
= e
−1
ji
∂h
j
e
ji
+ e
−1
ji
∂e
ji
Now h
i
=
t
e
ji
h
j
e
ji
, so that
h
−1
i
∂h
i
= e
−1
ji
h
−1
j
t
¯ e
−1
ji
_
∂
t
e
ji
h
j
e
ji
+
t
e
ji
∂h
j
e
ji
+
t
e
ji
h
j
∂e
ji
_
.
Since the e
i j
are holomorphic functions,
∂
t
¯ e
ji
= 0.
Hence h
−1
i
∂h
i
= e
−1
ji
h
−1
j
∂h
j
e
ji
+ e
−1
ji
∂e
i j
.
The principal bundles
⌣
ω : P −→ X associated to E is deﬁned, with
respect to the covering 1 = ¦U
i
¦, by the transition function ¦e
i j
¦ acting
on the ﬁbre GL(m, C) as follows:
Z
i
= e
i j
Z
j
(Z
i
, Z
j
∈ GL(m, C)).
The computations above shows that
Z
−1
i
(l
i
Z
i
+ dZ
i
) = Z
−1
j
(l
j
Z
j
+ dZ
j
) on
⌣
ω
−1
(U
i
∩
⌣
ω(U
j
)
Hence ω
i
= Z
−1
i
(l
i
Z
i
+dZ
i
) is the local representation on
⌣
ω
−1
(U
i
) of
a global 1form ω with values in the Lie algebra of GL(m, C).
Let P
ξ
be the tangent space to P at a point ξ ∈ P. Let Q
ξ
⊂ P
ξ
be 72
the subspace annihilated by ω. The family of these subspaces deﬁnes a
connection in P [26]. This proves the lemma.
The covariant derivation associated to the connection deﬁned above
is a map
▽ : Γ(X, 7(E)) →Γ(X, 7(E ⊗ Θ
∗
)
where for a vectorbundle F, 7(F) denotes the sheaf of germs of diﬀer
entiable sections of F, and Θ
∗
is the bundle of 1forms with value in C.
9. Metrics and connections 63
(In the above we regard E as a diﬀerentiable vector bundle). ▽ is deﬁned
in an open set ¦U
i
¦ of the covering 1 as follows.
Let z
1
, . . . , z
n
be a system of local coordinates on U
i
. Since we are
given a local trivialisation over U
i
of E, a section σ over U
i
of E may
be regarded as function on X with values in C
m
(rank E = m). We then
have
▽
α
(σ) = ▽(σ)
_
∂
∂z
α
_
= ∂σ
_
∂
∂z
α
_
+ l
i
_
∂
∂z
α
_
(σ)
and ▽
α
(σ) = ▽(σ)
_
∂
∂z
α
_
= ∂σ
_
∂
∂z
α
_
.
It is easy to check that these local representation deﬁned a global
map ▽ as above.
We consider next the curvature form of the connection.
Let Ω be the curvature form of the connection form ω in the princi 73
pal bundle P associated to E. The values of Ωon a pair, u
1
, u
2
, of tangent
vector ﬁelds to P is given by the structure formula ([26], 3435):
Ω(u
1
, u
2
) = dω(u
1
, u
2
) +
1
2
[ω(u
1
), ω(u
2
)].
Hence the component of type (1,1) of Ω is given on ¯ ω
−1
(U
i
) by
∂ω
i
= Z
−1
i
s
i
Z
i
,
where s
i
= ∂l
i
.
We shall call s the curvature form of the (connection ¦l
i
¦ deﬁned by the)
metric ¦h
i
¦ along the ﬁbers of E. On U
i
∩ U
j
s
i
= e
i j
s
j
e
ji
.
Hence we have
Lemma 3.2. The curvature form of the connection deﬁned above is
given in U
i
by the matrix valued 2form
(s
a
t
)
i
= s
i
= ∂l
i
Hence it is a form of type (1,1) with values in the “adjoint” bundle End
(E).
64 3. Local expressions for and the main inequality
The bundle Θ
∗
decomposes canonically as a direct sum
Θ
∗
⋍ Θ
∗
◦
⊕
¯
Θ
∗
◦
where Θ
◦
(resp. Θ
◦
) is the holomorphic tangent bundle (resp. anti holo 74
morphic tangent bundle) of X. (The two bundles in the above decompo
sition are regarded as diﬀerentiable bundles. As a diﬀerentiable bundle
Θ
∗
◦
(resp. (Θ
∗
◦
), is simply the bundle of C
∞
− forms on X of type (1,0)
(resp. type (0,1)). The decomposition above gives a directsum repre
sentations
7(Θ
∗
) ⋍ 7(Θ
∗
◦
) ⊕ 7(Θ
∗
◦
),
and for any holomorphic vector bundle E on X,
Γ(X, 7(E ⊗ Θ
∗
)) ⋍ Γ(X, 7(E ⊗ Θ
∗
◦
)) ⊕ Γ(X, 7(E ⊗ Θ
∗
◦
)).
Now if we are given a connection on E, it deﬁnes as we have re
marked a covariant derivation
▽ : Γ(X, 7(E)) →Γ(X, 7(E ⊗ Θ
∗
));
composing with the natural projections deﬁned by the direct sum de
composition above we obtain maps
▽
′
: Γ(X, 7(E)) →Γ(X, 7(E ⊗ Θ
∗
◦
))
and ▽
′′
: Γ(X, 7(E)) →Γ(X, 7(E ⊗ Θ
∗
◦
)).
Remarks. (1) As it has been remarked at the end of §3, the metric ¦h
i
¦
on E induces a metric on the dual E
∗
: this is simply given by the
family of positive deﬁnite matrixvalued functions
_
t
h
−1
i
_
i∈I
. The
corresponding connection is given locally on U
i
by the form −
t
l
i
. 75
It is also easy to see that the curvature form of this connection is
given by ¦−
t
s
i
¦
i∈I
(2) Let
¯
E denote the “anti holomorphic” bundle associated to E.
¯
E
is the bundle with transition functions ¯ e
i j
. Then
t
h
i
=
¯
h
i
deﬁne a
metric on
¯
E as well. In this case, the local forms
¯
l
i
=
¯
h
−1
i
¯
∂
¯
h
i
9. Metrics and connections 65
deﬁne a connection of type (0,1). When we speak of a connection
on
¯
E without further comment, it will always be of this connection
(of course, it is necessary to assume given a metric or at least a
connection on E.)
(3) Suppose that E and F are holomorphic vector bundles on X and
that we given hermitian metrics ¦h
1
i
¦
i∈I
and ¦h
2
i
¦
i∈I
on E and F re
spectively. (We assume here, as we may, that E and F are deﬁned
by means of transition function ¦e
i j
¦¦ f
i j
¦ with respect to the same
covering 1 = ¦U
i
¦
i∈I
). Then the family of matrix valued functions
¦h
1
i
⊗ h
2
i
¦ deﬁnes a metric ¦h
i
¦ along the ﬁbres of E ⊗ F. We have
then
∂(h
i
) = ∂(h
1
i
⊗ h
2
i
) = ∂h
1
i
⊗ h
2
i
+ h
1
i
⊗ ∂h
2
i
so that l
i
= h
−1
i
∂h
i
= h
−1
i
∂h
1
i
⊗ I
r
2
+ I
r
1
⊗ h
2
−1
i
∂h
2
i
where r
1
(resp. r
2
) is the rank of E (resp. rank F) and for an 76
integer m, I
m
denotes the (m × m) identity matrix. One sees im
mediately then that the curvature form s
i
is given by the formula
s
i
= ∂l
i
= s
1
i
⊗ I
r
2
+ I
r
1
⊗ s
2
i
.
It is clear that the above considerations can be carried over to the
case when one or both of E and F are anti holomorphic.
In the sequel, we call a connection on a (diﬀerential) vector bundle
(with complex ﬁbres) a ∂−connection (resp. ∂ connection) if the associ
ated connection form is of type (1,0) (resp. type (0,1)). The connection
l
i
on a holomorphic bundle E deﬁned with respect to a metric h
i
is then a
∂− connection while that on the conjugate bundle E is a ∂− connection.
Now if ¦h
i
¦
i∈I
is a hermitian metric along the ﬁbres of E, then it
can be regarded as a section h of the bundle E
∗
⊗
¯
E
∗
. We have also a
canonical connection on E
∗
⊗
¯
E
∗
. We denote the covariant derivative on
section of E
∗
⊗
¯
E
∗
again by ▽:
▽ : Γ(X, 7(E
∗
⊗
¯
E
∗
)) →Γ(X, 7(E
∗
⊗
¯
E
∗
⊗ Θ
∗
)).
66 3. Local expressions for and the main inequality
Proposition 3.1. ▽h = 0
Proof. Let z
1
, .., z
n
be a coordinate system on U
1
. In this proof we write
l for l
i
and h for h
i
.
We have 77
▽
α
h = (▽h)
_
∂
∂z
α
_
= ∂
α
h
ab
−
c
l
c
bα
h
ac
= ∂
x
h
ab
−
(h
−1
)
cd
∂
α
h
db
h
ac
= ∂
α
h
ab
−
δ
d
a
∂
α
h
db
= ∂
α
h
ab
− ∂
α
h
ab
= 0.
Similarly one shows that
▽
α
h = ▽h(
∂
∂z
α
) = 0.
Hence the proposition.
(Here h is regarded as a function from U
i
into
_
C
m
⊗C
m
_
∗
using the
given trivialisation of E ¦ U
i
. A vector t ∈ C
m
is an mtuple denoted
(t
a
)
1≤a≤m
and tǫC
n
is again an mtuple but is denoted (t
a
)
1≤a≤m
. Further
conventions are as follows: a vector in the dual of C
m
is an mtuple but
with subscripts instead of super scripts. In other words we write
t =
t
a
e
a
for t ∈ C
m
t =
t
a
e
a
for t ∈
¯
C
m
t =
t
a
e
∗
a
for t ∈ C
m
∗
t =
t
a
e
∗
a
for t ∈ C
m
∗
where ¦e
a
¦ (resp. ¦e
a
¦, ¦e
∗
a
¦, ¦e
∗
a
¦ is the canonical (resp. conjugate of the
canonical, dual of the canonical, conjugate dual of the canonical) basis
of C
m
).
Proposition 3.2. Let z
1
, . . . , z
n
be a coordinate system on U
i
. Then in 78
U
i
, the curvature form can be written as
a
i
= (s
a
b
¯
βα
)
ab
d¯ z
β
∧ dz
α,
9. Metrics and connections 67
where
(s
a
b
¯
βα
)
ab
= ▽¯
β
▽
α
− ▽
α
▽¯
β
(3.1)
Proof. Let t = (t
a
) be a section of the bundle over U
i
.
Then
▽¯
β
t = ∂¯
β
t = (∂¯
β
t
a
)
Hence
(▽
α
▽¯
β
t)
a
= ∂
α
∂t
a
β
+
b
l
a
bα
∂¯
β
t
b
where l =
_
l
a
bβ
dz
α
is the connection form.
▽
α
t = ∂
α
t + (l
a
bα
)(t) = ∂
α
t
a
+
b
l
a
bα
t
b
so that
▽¯
β
▽
α
t = ∂¯
β
∂
α
t
a
+
b
∂¯
β
l
a
bα
t
b
+
b
l
a
bα
∂¯
β
t
b
= ∂¯
β
∂
α
t
a
+ s
a
b
¯
βα
t
b
+
b
l
a
bα
∂¯
β
t
b
.
Hence −(▽
α
▽¯
β
t − ▽¯
β
▽
α
t)
a
=
_
b
s
a
b
¯
βα
t
b
.
This proves the proposition.
The equation (3.1) is known as the Bianchi Identity.
(We remark that s is a form of type (1,1).).
The following lemma is easy to prove. 79
Lemma 3.3. ▽# = #▽.
Next we specialize the preceding considerations to the case E = Θ
o
the homomorphic tangent bundle. Let z
1
, . . . , z
n
be a local coordinate
system in an open set U ⊂ X. Let
g =
g
¯ αβ
dz
−α
dz
β
be the expression for the hermitian metric in this coordinate system. The
associated ∂connection is then given in U by the (1,0) form
C
α
βγ
dz
γ
68 3. Local expressions for and the main inequality
where C
α
βγ
=
g
¯ τα
∂g
¯ τβ
∂z
γ
.
(here, as is usual, g
¯ αβ
is deﬁned by (g
¯ αβ
)(g
¯ αβ
) = Identity).
That means that
_
¸
¸
_
¸
¸
_
▽
β
_
∂
∂z
α
_
=
_
C
α
βγ
∂
∂z
γ
▽¯
β
∂
∂z
α
= 0
(3.2)
where
▽ : Γ(X, 7(Θ)) →Γ(X, 7(Θ
o
⊗ Θ
∗
))
is the covariant derivation.
The ∂ connection deﬁned above is not in general symmetric: C
γ
αβ
80
C
γ
βα
; in fact this connection is symmetric if and only if the hermitian
metric g is Kahler. We set then
S =
1
2
(C
γ
βα
− C
γ
αβ
)dz
β
∧ dz
α
.
Clearly S is an alternating (2,0) form with values in the tangent bun
dle Θ
0
. It is called the torsion form of the ∂ connection. Its vanishing
characterises the Kahler metrics.
A hermitian metric on the holomorphic tangent bundle deﬁnes a Rie
mannian metric as well on X. We have corresponding to this metric a
Riemannian connection on X. The associated covariant derivative is a
map
D : Γ(X7(Θ)) →Γ(X, 7(Θ ⊗ Θ
∗
)).
In particular, we have
D : Γ(X7(Θ
0
)) →Γ(X, 7(Θ
0
⊗ Θ
∗
)).
Denoting, as is usual, the RiemannChristofel symbols by Γ
γ
αβ
, Γ
γ
αβ
,..
etc, one can prove easily that these are related to the C
γ
αβ
as follows:
Γ
α
βγ
=
1
2
(C
α
βγ
+ C
α
γβ
)
9. Metrics and connections 69
γ
Γ
¯ γ
β¯ γ
= −
r
S
γ
βγ
Since Θ ⋍ Θ
◦
⊕Θ
◦
as a diﬀerentiable vector bundle, we have a direct
sum decomposition
Γ(X, 7(Θ ⊗ Θ
∗
)) ⋍ Γ(X, 7(Θ ⊗ Θ
∗
◦
)) ⊗ Γ(X, 7(Θ ⊗ Θ
∗
◦
))
and hence a natural projection 81
Γ(X, 7(Θ ⊗ Θ
∗
)) →Γ(X, 7(Θ ⊗ Θ
∗
));
composing this projection with D and ▽, we obtain two linear maps
▽
′′
: Γ(X, 7(Θ
o
)) →Γ(X, 7(Θ
o
⊗ Θ
∗
o
))
and D
′′
: Γ(X, 7(Θ
o
)) →Γ(X, 7(Θ
o
⊗ Θ
∗
o
)).
If the metric is Kahler, these two maps coincide. In a similar way,
composing the natural projection
Γ(X, 7(Θ ⊗ Θ
∗
)) →Γ(X, 7(Θ ⊗ Θ
∗
o
))
with ▽ and D we deﬁne two linear maps,
▽
′
: Γ(X, 7Θ
o
)) →Γ(X, 7(Θ
o
⊗ Θ
o
∗
))
D
′
: Γ(X, 7Θ
o
)) →Γ(X, 7(Θ
o
⊗ Θ
∗
o
))
which coincide if the hermitian metric of X is a Kahler metric.
Suppose now that we are given a vector bundle E on X and hermitian
metrics on X and along the ﬁbres of E. We then have canonical ∂ 
connections on the bundle E and Θ
o
and a ∂ connection on Θ
o
. These
connections extend canonically to connections on each of the bundles
p
ΛΘ
∗
o
⊗
q
ΛΘ
∗
o
⊗ E
We denote by ▽ the covariant derivation in any of these bundles: 82
▽ : Γ(X, 7(
p
∧Θ
∗
o
⊗
q
∧Θ
∗
o
⊗ E)) →Γ(X, 7(
p
∧Θ
∗
o
⊗
q
∧Θ
∗
⊗ E ⊗ Θ
∗
))
70 3. Local expressions for and the main inequality
Once again, composing with the natural projections, we can deﬁne
△
′′
: Γ(X, 7(
p
∧Θ
∗
o
⊗
q
∧Θ
o
∗
⊗ E)) →Γ(X, 7(
p
∧Θ
∗
o
⊗
q
∧Θ
∗
o
⊗ E ⊗ Θ
∗
o
)),
△
′
: Γ(X, 7(
p
∧Θ
o
q
∧ Θ
o
⊗ E)) →Γ(X, 7(
p
∧Θ
o
q
∧ Θ
o
⊗ E ⊗ E ⊗ Θ
∗
o
),
Let s =
_
α,β
s
a
b¯ αβ
d¯ z
α
∧ dz
β
be the curvature form of the ∂ connection
on E in a coordinate open set Y over which the bundle E is trivial, the
complex analytic coordinates in U being (z
1
, . . . , z
n
) : if (rank E)= m,
a, b run through 1 to m and for ﬁxed α, β, (s
a
bαβ
) is an (m × m)matrix.
In the same coordinate neighborhood, let
L =
σ,τ
L
α
β ¯ στ
d¯ z
σ
∧ dz
τ
be the curvature form of the ∂connection on the holomorphic tangent
bundle Θ
o
. For ﬁxed σ, τ, (L
α
β ¯ στ
) is an (n × n)matrix.
10 Local expressions for ∂, ϑ and
We will now obtain a local expression for the operator
= ∂ϑ + ϑ∂
in terms of ▽
α
, ¯ ▽
α
and the forms L and s above.
We adopt the following notation : for a p tuple A = (α
1
, . . . α
p
), we 83
set dz
A
= dz
α
1
∧ . . . ∧ dz
α
p
(resp. dz
−A
= d¯ z
α
1
∧ . . . ∧ d¯ z
α
p
). Then if for
φǫC
pq(X,E)
, we set
ϕ
_
∂
∂z
α1
, . . . ,
∂
∂z
αp
,
∂
∂¯ z
β1
, . . . , . . .
∂
∂¯ z
βq
_
= ϕ
a
A
¯
B
we have ϕ =
A,B
1
p!q!
ϕ
a
A
¯
B
dz
A
∧ d¯ z
B
where A = (α
1
, . . . , α
p
) and B = (β
1
, . . . , β
q
) for a ptuple A and a (q+1)
tuple (β
1
, . . . , β
q+1
) we have
(
¯
∂ϕ)
a
A
¯
β
1
. . .
¯
β
q+1
= (−1)
p
q+1
r=1
(−1)
r−1
¯
∂¯
β
r
ϕ
a
A
¯
β
1
...β
r
...β
q+1
.
10. Local expressions for ∂, ϑ and 71
On the other hand
∇¯
β
r
ϕ
a
¯
β
1
...
ˆ
¯
βr...β
q+1
= ∂¯
β
r
ϕ
a
A
¯
β
1
...
ˆ
¯
β
r
...
¯
β
q+1
−
ir
C
¯ α
β
i
β
r
ϕ
a
A
¯
β
1
...
¯
β
i−1
¯ α
¯
β
i+1
...
ˆ
¯
β
r
...
¯
β
q+1
= ∂¯
βr
ϕ
a
A
¯
β
1
...
ˆ
¯
β
r
...
¯
β
q+1
−
ir
B
α
β
i
β
r
+ S
α
β
i
β
r
ϕ
a
A
¯
β
1
...
¯
β
i−1
¯ αβ
i+1
...
ˆ
¯
β
r
...β
q+1
where C
γ
αβ
has been deﬁned before and 84
B
γ
αβ
=
1
2
(C
γ
αβ
+ C
γ
βα
)
while S
γ
αβ
=
1
2
(C
γ
αβ
− C
γ
βα
)
so that S =
S
γ
αβ
dz
α
∧ dz
β
is the torsion of the connection on Θ
◦
.
We have therefore
(
¯
∂ϕ
a
)
A
¯
β
1
...
¯
β
q+1
= (−1)
p
(−1)
r−1
∇
¯
β
r
ϕ
a
A
¯
β
1
...
ˆ
¯
β
r
...
¯
β
q+1
+ (−1)
p
ir
(−1)
r−1
¯
S
α
β
i
β
r
ϕ
a
A
¯
β
1
...(α)
1
...
ˆ
¯
β
r
...
¯
β
q+1
.
Let 85
S : C
pq
(X, E) → C
p,q+1
(X, E)
be the operator deﬁned by
(S ϕ)
a
A
¯
β
1
...
¯
β
q+1
= (−1)
p
ir
(−1)
r−1
¯
S
α
β
i
β
r
ϕ
a
Aβ
1
...β
i−1
αβ
i+1
β
γ
¯
β
−−q−1
(3.4)
and set
¯
∂ =
˜
∂ + S, (3.5)
so that
(
˜
∂ϕ)
a
A
¯
β
1
...
¯
β
q+1
= (−1)
p
(−1)
r−1
∇ ¯
β
r
ϕ
α
A
¯
β
1
...
ˆ
¯
β
r
...β
q−1
(3.6)
Let
˜
ϑ = − ∗ #
−1
˜
∂ ∗ # : C
pq
(X, E) → C
p,q−1
(X, E) (3.7)
72 3. Local expressions for and the main inequality
and
T = − ∗ #
−1
S ∗ # : C
pq
(X, E) →C
p,q−1
(X, E) (3.8)
so that
ϑ =
˜
θ + T. (3.9)
Finally let
˜ =
˜
∂
˜
ϑ +
˜
ϑ
˜
∂
If the hermitian metric on X is a Kahler metric, then 86
¯
∂ =
∼
∂, ϑ =
∼
ϑ, =
∼
.
For ϕ ∈ C
pq
(X, E)
_
˜
ϑϕ
_
a
AB
′
= (−1)
p−1
∇
α
ϕ
a
A
α
B
′
, (3.10)
so that, exactly as in the case of the Laplacian △ in Chapter 2, we have
_
∼
ϕ
_
a
AB
= −∇
α
∇
α
ϕ
a
AB
+
q
r=1
(−1)
r−1
(∇
α
∇
β
r
− ∇
β
r
∇
α
)ϕ
a
A
α
B
′
r
(3.11)
where
∇
α
= g
αβ
∇
β
,
and A = (α
1
, ........α
p
), B = (β
1
, ......β
q
), B
′
r
= (β
1
, .......,
ˆ
β
r
, ...., β
q
).
In view of the Ricci identity, the summand of (3.11) can be ex
pressed by
q
r=1
(−1)
r−1
(∇
α
∇
β
r
− ∇
β
r
∇
α
)ϕ
a
A
α
B
′
r
= (
∼
7ϕ)
a
AB (3.12)
where
∼
7 is a mapping
∼
7 : C
pq
(X, E) → C
pq
(X, E),
which is linear over C
∞
functions, whose local expression involves lin
early (with integral coeﬃcients) only the coeﬃcients of the curvature 87
10. Local expressions for ∂, ϑ and 73
forms, s and L, of E and Θ
◦
.
By Remark (3) after Lemma 3.2 we have
_
∼
7ϕ
_
a
AB
=
q
r=1
(−1)
r
s
a
bβ
r
α
ϕ
b
A
α
B
′
r
+
_
∼
7
◦
ϕ
_
a
AB
, (3.13)
where
∼
7
◦
involves only the curvature tensor of Θ
◦
, and is completely
independent of E.
Formula (3.11) can be also written as
(
∼
ϕ)
a
AB
= −∇
α
∇
α
ϕ
a
AB
+ (
∼
7ϕ)
a
AB
. (3.14)
Now
= (
¯
∂ϑ + ϑ
¯
∂) = (
¯
∂ + S )(
∼
ϑ + T) + (
∼
ϑ + T)(
∼
∂ + S )
=
∼
+
∼
∂T + T
∼
∂ +
∼
ϑS + S
∼
ϑ + S T + TS.
It follows that
Lemma 3.4. For any ϕǫC
pq
(X, E)
(ϕ)
a
AB
= (
∼
ϕ)
a
AB
+ (F
1
ϕ)
a
AB
+ (F
2
∇
′
ϕ)
a
AB
+ (F
3
∇
′′
ϕ)
a
AB
where
F
1
: C
pq
(X, E) → C
pq
(X, E),
F
2
: C
pq
(X, E ⊗ Θ
∗
o
) → C
pq
(X, E),
F
3
: C
pq
(X, E ⊗ Θ
∗
o
) → C
pq
(X, E),
are linear over C
∞
functions. Their local expression involves the tensor 88
tensor and its ﬁrst derivatives.
If the metric on X is a Kahler metric, then F
1
≡ 0, F
2
≡ 0, F
3
≡ 0.
Applying proposition 3.1 to the hermitian metric on X we have that
∇∗ = ∗∇.
74 3. Local expressions for and the main inequality
Hence, by (3.14),
∼
− ∗
−1
∼
∗ =
∼
7− ∗
−1
∼
7 ∗ . (3.15)
If the metric on X is Kahler, this identity becomes
− ∗
−1
∗ =
∼
7 − ∗
−1
∼
7 ∗ .
A direct computation shows that, in this case,
∼
7 − ∗
−1
∼
7∗ does not
depend on the curvature form of the Kahler metric. We have in fact (see
[2], 103)
((
∼
7 − ∗
−1
∼
7∗))
a
AB
=
q
i=1
(−1)
i
s
a
bβ
i
β
ϕ
bβ
AB
′
i
+
p
j=1
(−1)
j
s
a
bαα
j
ϕ
bα
A
′
j
¯
B
+ s
a
bβ
βϕ
b
AB
where A
′
j
= (α
1
, . . . , ˆ α
j
, ...., α
p
).
Starting from this formula, it is easy to check that
∼
7 − ∗
−1
∼
7∗ = e(s)Λ − Λe(s),
where 89
e(s) : C
pq
(X, E) → C
p+1,q+1
(X, E)
is the linear mapping locally deﬁned by
(e(s)ϕ)
i
=
√
−1s
a
ib
∧ ϕ
b
,
and Λ is the classical operator of the Kahler geometry (see e.g. [35],
42). Thus we have, in the Kahler case,
− ∗
−1
∗ = e(s)Λ − Λe(s)
which, by (1.7), can also be written
E
− #
−1
E
∗ # = e(s)Λ − Λe(s).
This formula, which was ﬁrst obtained in [4], 483, yields Wellipti
city conditions on K¨ ahler manifolds (see [2], ).
11. The main inequality 75
If the metric on X is not Kahler, then it follows from Lemma 3.4 and
from (3.15) that
− ∗
−1
∗ =
∼
K − ∗
−1
∼
K ∗ +G
1
ϕ + G
2
▽
′
ϕ + G
3
∇
′′
ϕ
where
G
1
: C
pq
(X, E) →C
pq
(X, E),
G
2
: C
pq
(X, E ⊗ Θ
∗
◦
) →C
pq
(X, E),
G
3
: C
pq
(X, E ⊗ Θ
∗
◦
) →C
pq
(X, E),
are linear over C
∞
functions. Their local expressions involve the torsion
tensor and its ﬁrst covariant derivatives. If the hermitian metric is a 90
K¨ ahler metric, then G
1
≡ 0, G
2
≡ 0, G
3
≡ 0.
11 The main inequality
We shall now establish an integral inequality which, under convenient
hypotheses, yields a suﬃcient condition for the W
pq
 ellipticity of E.
Let ϕ ∈ C
pq
(X, E)(q > 0). Let ξ and η be two tangent vector ﬁelds
to X, deﬁned by
ξ = (ξ
β
= h ¯
ba
▽
¯ γ
ϕ
a
A
β ¯
B
′ ϕ
b
¯
AγB
′
, ξ
¯
β
= 0)
η = (η
γ
= 0, η
¯ γ
= h ¯
ba
▽
β
ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
′
)
We have (η being the complex dimension of X)
divξ =
2n
i=1
D
i
ξ
i
=
n
β=1
▽
β
ξ
β
−
α,β
S
β
α
βξ
α
,
where
▽
β
ξ
β
= h¯
ba
▽
β
▽
¯ γ
ϕ
a
A
β ¯
B
′ .ϕ
b¯ aγB
′
+ h¯
ba
▽
¯ γ
ϕ
a
A
β ¯
B
′ .▽¯
B
ϕ
b
¯
AγB
′
= h¯
ba
▽
¯ γ
▽
β
ϕ
a
A
β ¯
B
′ .ϕ
b¯ aγB
′
+ h¯
ba
▽
¯ γ
ϕ
a
A
β ¯
B
′ .▽¯
B
ϕ
b
¯
AγB
′
+ h¯
ba
(▽
β
▽
¯ γ
− ▽
¯ γ
▽
β
)ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
′
.
76 3. Local expressions for and the main inequality
The last summand can be evaluated using the Ricci identity. We 91
have, by (3.12).
h¯
ba
(▽
β
▽
¯ γ
−▽
¯ γ
▽
β
)ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
=
1
q
h
ba
(
˜
Kϕ)
a
A
¯
B
ϕ
b
¯
AB
= p!(q−1)!A(
˜
Kϕ, ϕ).
Furthermore a direct computation, starting from (3.6), shows that
A(
˜
∂ϕ,
˜
∂ϕ) = A(▽
′′
ϕ, ▽
′′
ϕ) −
1
p!(q − 1)!
h¯
ba
▽
¯ γ
ϕ
a
A
β ¯
B
′ .▽¯
β
ϕ
b
¯
AγB
′
.
Hence we have
▽
β
ξ
β
= h¯
ba
▽
¯ γ
▽
β
ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
′
+ p!(q − 1)!
_
A(
˜
Kϕ, ϕ) + A(▽
′′
ϕ, ▽
′′
ϕ) − A(
˜
∂ϕ,
˜
∂ϕ)
_
.
Now
div η =
2n
i=1
D
i
η
i
=
▽
¯ γ
η
¯ γ
− 2
γ,β
S
β
γβ
η
¯ γ
,
▽¯
∂
η
¯ γ
= h¯
ba
▽
¯ γ
▽
β
ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
′
+ h¯
ba
▽
β
ϕ
a
A
β ¯
B
′ .▽
γ
ϕ
b
¯
AγB
′
= h¯
ba
△
¯ γ
△
β
ϕ
a
A
β ¯
B
′ .ϕ
b
¯
AγB
′
+ p!(q − 1)!A(
˜
ϑϕ,
˜
ϑϕ), by (3.10).
Thus 92
div ξ − div η = p!(q − 1)!
_
A(
∼
Kϕ, ϕ) + A(▽
′′
ϕ, ▽
′′
ϕ)
−A(
∼
∂ϕ,
∼
∂ϕ) − A(
∼
ϑϕ,
∼
ϑϕ)
_
− 2
_
S
α
β
β
ξ
α
− S
β
γβ
η
γ
_
.
Let ϕǫD
pq
(X, E). Then by Stokes’ theorem we have
j▽
′′
ϕj
2
+(
∼
Kϕ, ϕ) = j
∼
∂ϕj
2
+j
∼
ϑϕj
2
+
2
p!(q − 1)!
_
X
_
S
β
αβ
ξ
α
− S
β
γβ
η
γ
_
dX.
(3.16)
Let ¦ S ¦ be the length of the torsion form. Applying lemma A of
Chapter 1 and the Schwartz inequality we get from (3.4), (3.5) and from
(3.8), (3.9) the following estimates
j
∼
∂ϕj
2
≤ 2(∂ϕj
2
+ jS ϕj
2
) ≤ 2j∂ϕj
2
+ c
_
X
¦ S ¦
2
A(ϕ, ϕ)dX
11. The main inequality 77
= 2j∂ϕj
2
+ +c(¦ S ¦
2
ϕ, ϕ),
j
∼
ϑϕj
2
≤ 2(jϑϕj
2
+ jTϕj
2
) ≤ 2jϑϕj
2
+ c
_
X
¦ S ¦
2
A(ϕ, ϕ)dX
= 2jϑϕj
2
+ c(¦ S ¦
2
ϕ, ϕ),
c being a universal positive constant (depending only on the dimension
of X).
Furthermore, again by the Schwartz inequality, we have 93
1
p!(q − 1)!
¦
S
α
β
β
ξ
β
¦ ≤
1
p!(q − 1)!
¦ S jξ ¦≤ q ¦ S ¦ A(▽
′′
ϕ, ▽
′′
ϕ)
1
2
A(ϕ, ϕ)
1
2
≤
qε
2
¦ S ¦
2
A(ϕ, ϕ) +
q
2ε
A(▽
′′
ϕ, ▽
′′
ϕ),
1
p!(q − 1)!
¦
S β
γ
β
η
¯ γ
¦ ≤
1
p!(q − 1)!
¦ S jη ¦≤ q ¦ S ¦ A(
∼
ϑϕ,
∼
ϑϕ)
1
2
A(ϕ, ϕ)
1
2
≤ q
ε
2
¦ S ¦
2
A(ϕ, ϕ) +
q
2ε
A(
∼
ϑϕ,
∼
ϑϕ),
for any ε > 0.
Substituting these estimates in (3.16) we obtain for any ε > 0
_
1 −
q
ε
_
j ▽
′′
ϕj
2
+
_
∼
Kϕ −
_
2c + 2qε +
qc
ε
_
¦ S ¦
2
ϕ, ϕ
_
≤ 2j∂ϕj
2
+ +2
_
1 +
q
ε
_
jϑϕj
2
.
Setting, for instance ε = 2q , and
7 =
∼
7 −
_
5
2
c + 4n
2
_
¦ S ¦
2
.Id : C
pq
(X, E) →C
pq
(X, E) (3.17)
we obtain the following
Proposition 3.3. For any ϕǫD
pq
(X, E) (q > 0) we have
1
2
j▽
′′
ϕj
2
+ (7ϕ, ϕ) ≤ 3
_
j∂ϕj
2
+ jϑϕj
2
_
.
78 3. Local expressions for and the main inequality
If the hermitian metric on X is a K¨ ahler metric, the above proposition
can be considerably sharpened. In fact, in this case, S ≡ 0, ∂ = 0, 94
∼
ϑ = ϑ. It follows from (3.16) that for every ϕǫD
pq
(X, E) (q > 0) satisﬁes
the identity
j▽
′′
ϕj
2
+ (7ϕ, ϕ) = j∂ϕj
2
+ jϑϕj
2
(K¨ ahler) (3.18)
where we have set 7 =
∼
7. Also in the hermitian case the choice of the
numerical constants can be improved.
As a corollary to proposition 3.1 we have the following
Theorem 3.1. If there exists a positive constant C
2
such that
A(7ϕ, ϕ) ≥ C
2
A(ϕ, ϕ)
for every ϕǫC
pq
(X, E) (q > 0) and at each point of X, then E is W
p,q

elliptic.
Remark. Eﬀect on 7 of a change of the hermitian metric on E.
Let ϕ : XR be any C
∞
 function. We wish to examine how the
operator 7 above changes when we replace the hermitian metric h on E
by h
′
= e
ϕ
h. (The hermitian metric on the base X remains undisturbed).
First, the curvature form s
′
of h
′
is given by
s
′
= s +
α,β
∂
2
ϕ
∂z
α
∂z
β
dz
α
∧ dz
β
.I
m
,
where s, as above, is the curvature form of h; (z
α
)
1≤α≤n
is a local coordi
nate system in an open set U ⊂ X on which a local trivialization of E is
assumed given; the symbol I
m
stands for the (m × m) identity matrix. It
is immediate from (3.13) that
(7
′
ϕ)
a
AB
= (7ϕ
a
)
AB
+
i
(−1)
i
∂
2
ϕ
∂z
α
∂z
β
i
ϕ
aα
AB
′
i
.
where 7
′
is the analogue of 7, deﬁned now with respect to the new 95
hermitian metric h
′
.
Chapter 4
Vanishing Theorems
12 qcomplete manifolds
Let X be a complex manifold and φ : X → R a C
∞
function. The Levi 96
formL(φ) of φ is the hermitian quadratic diﬀerential form on X deﬁned
as follows: let z
1
, ..., z
n
be a coordinate system on an open set U ⊂ X;
then
L(φ)
_
∂
∂z
α
,
∂
∂z
β
_
=
∂
2
ϕ
∂z
α
∂z
β
.
(L(ϕ) is thus a hermitian form on the holomorphic tangent space).
Deﬁnition 4.1. A C
∞
function φ : X → R is strongly qpseudoconvex
if the Levi form L(φ) has at least (n − q) positive eigenvalues at every
point of X.
Deﬁnition 4.2. A complex manifold X is qcomplete if there exists a C
∞
function φ : X → R such that
(i) φ is strongly qpseudoconvex
(ii) for c ∈ R, the set
_
x ¦ x ∈ X, φ(x) < c
_
is relatively compact in X.
Lemma 4.1 (E. Calabi). Let H be a hermitian quadratic diﬀerential
form on X and G a hermitian metric on X. Assume that H has at least p
79
80 4. Vanishing Theorems
positive eigen values. Let ε
1
(x), ..., ε
n
(x) be the eigen values of H (w.r.t.
G) at x in decreasing order: ε
r
(x) ≥ ε
r+1
(x)
Then given c
1
, c
2
> 0, G can be so chosen that 97
l
H
(x) = c
1
ε
p
(x) + c
2
Inf(0, ε
n
(x)) > 0 for all x ∈ X.
Proof. Let G be any complete hermitian metric whatever. Let σ
1
(x), . . . ,
σ
n
(x) be the eigenvalues of H with reference to G arranged in de
creasing order. We construct now a metric G on X whose eigen val
ues are functions of
_
σ
i
(x)
_
1≤i≤n
as follows: let λ : X → R be a C
∞
−
function (we will impose conditions on λ letter); let U be a coordi
nate open set in X with holomorphic coordinates (z
1
, ..., z
n
); in U, we
have G = G
Uαβ
dz
α
dz
β
so that (G
Uαβ
)
αβ
is a function whose values
are positive deﬁnite hermitian matrices; then the matrix valued function
¯
G
U
= (
¯
G
Uαβ
)
αβ
where
¯
G =
_
¯
G
Uαβ
dz
α
dz
β
in U is deﬁned by
¯
G
−1
U
= G
−1
U
∞
=
r=0
λ(x)
r
(r + 1)!
(H
U
G
−1
U
)
r
where H
U
is the matrix valued function (H
Uαβ
) deﬁned by
H =
H
Uαβ
dz
α
d¯ z
β
in U.
We now assert that
¯
G
U
deﬁne a global hermitian deﬀerential form on
X and under a suitable choice of λ, it is positive deﬁnite. To see that
¯
G
U
deﬁnes a global hermitian diﬀerential form on X, we need only prove
the following. Let V be another coordinate open set with coordinates
complex (w
1
, ..., w
n
). Let J =
∂(z
1
, ..., z
n
)
∂(w
1
, ..., w
n
)
be the Jacobian matrix. As 98
before let G
V
= (G
Vαβ
) be deﬁned by G =
_
G
Vαβ
dw
α
d ¯ w
β
in V. Then
if
¯
G
V
is deﬁned starting from G
V
as
¯
G
U
from G
U
, we have
J
¯
G
t
U
J =
¯
G
V
.
12. qcomplete manifolds 81
We have in fact, writing J
∗
for
t
¯
J
−1
,
J
∗
¯
G
−1
U
J
−1
= G
−1
V
J
_
¸
¸
¸
¸
¸
¸
_
∞
r=0
λ(x)
1
(r + 1)!
(H
U
G
−1
U
)
r
_
¸
¸
¸
¸
¸
¸
_
J
−1
, since JG
U
t
¯
J = G
V
.
It follows from the above that
J
∗
¯
G
−1
U
J
−1
= G
−1
V
∞
r=0
λ(x)
r
(r + 1)!
(JH
U
G
−1
U
J
−1
)
r
= G
−1
V
∞
r=0
λ(x)
r
(r = 1)!
(H
V
J
∗
G
−1
U
J
−1
)
r
since JH
t
U
JH
V
. Hence we obtain
J
∗
¯
G
−1
U
J
−1
= G
−1
V
r=0
(λ(x))
r
(r + 1)!
(H
V
G
−1
V
)
r
=
¯
G
−1
V
This proves that
¯
G
U
deﬁnes on X a global hermitian diﬀerential. We
next show that
¯
G is positive deﬁnite. For this we look for the eigen
values of
¯
G with reference to G. TO compute these, we may assume, in
the above formula for
¯
G
U
, that G
U
is the identity matrix
Then we have 99
¯
G
U
=
+∞
r=0
λ(x)
r
(r + 1)!
H
r
U
It follows that the eigen values of
¯
G
U
are
_
¸
¸
_
¸
¸
_
+∞
r=0
λ(x)
r
(r + 1)!
σ
q
(x)
r
_
¸
¸
_
¸
¸
_
1≤q≤n
.
It is easily seen that these are all strictly greater than zero: this as
sertion simply means this: f (t) =
e
t
−1
t
=
∞
_
r=0
t
r
(r+1)!
for t 0, f (0) = 1
(which is continuous in t) is everywhere greater than 0.
82 4. Vanishing Theorems
We will now look for conditions on λ such that
¯
G satisﬁes our re
quirements. From the formula for
¯
G, we have
H
U
¯
G
−1
U
=
+∞
r=0
λ(x)
r
(r + 1)
(H
U
G
−1
U
)
r+1
.
Now the eigen values of H with respect to
¯
G (resp. G) are simply
those of the matrix H
U
¯
G
−1
U
(resp. H
U
G
−1
U
), Hence these eigenvalues
ε
q
(X) of H with reference to
¯
G are
f (λ(x), σ
q
(x))
where f (s, t) is the function on R
2
deﬁned by
f (s, t) =
t∞
r=0
S
r
(r + 1)
t
r+1
Since
∂f (s,r)
∂t
= e
st
> 0 for any, f (s, t) is monotone increasing in t.
Hence we have 100
ε
r
(x) ε
r+1
(x) for 1 r n − 1.
Moreover f (s, t) t for s 0. Thus , if we choose λ(x) 0 for
every xǫX, then ε
q
(x) σ
q
(x) > 0.
The choice of λ(x) is now made as follows . Let , for every integer
ν > 0. B
ν
=
_
x ¦ d(x, x
0
) γ
_
for some x
0
ǫX, the distance being i the
metric G. The B
ν
are then compact. Let b
ν
= Inf
x∈B
γ
(σ
p
(x)).
Then b
1
b
2
. . . b
ν+1
. . ..
Let b(x) be a C
∞
function on X such that b(x) > 0 for x ∈ X and
b(x) < b
ν
in B
ν
− B
ν−1
. Then clearly b(x) ≤ σ
p
(x).
Finally let ρ(x) be a C
∞
function on X such that ρ(x) d(x, x
0
), and
k >
_
C
2
C
1
b
1
be a real constant. Set λ(x) =
2ke
ρ(x)
b
2
(x)
. We have then
ε
q
(x) = f (λ(x), σ
p
(x)) = σ
p
(x) +
λ(x)
2!
σ
p
(x)
2
+
12. qcomplete manifolds 83
so that ε
p
(x)
ke
ρ(x)
b
2
(x)
σ
p
(x)
2
ke
ρ(x)
k.
On the other hand,
ε
n
(x) = f (λ(x), σ
n
(x)) =
1
λ(x)
_
e
λ(x)σ
n
(x)
−1
_
−1
λ(x)
= −
b
2
(x)
2ke
ρ(x)
−
b
2
1
k
,
C
1
ε
p
(x) + C
2
Inf(0, ε
n
(x)) C
1
k − C
2
b
2
1
k
1
k
(C
1
k
2
− C
2
b
2
1
) > 0.
This proves the lemma.
Combining this with the Remark at the end of §2, we can assert 101
moreover that G can be chosen to be complete and satisfy the condition
of Lemma 4.1.
Let φ : X → R be a strongly qpseudoconvex function on X. As
suming as a form H in the above lemma the Levi form L(φ), and taking
p = n − q, c
1
= 1, c
2
= n, we obtain
Lemma 4.2. Let φ : X → R be a strongly qpseudoconvex function on
X. Then there exists a complete hermitian metric on X such that at each
point of X.
l
L(φ)
(x) = ε
n−q
(x) + n Inf (0, ε
n
(x)) > 0.
Let u ∈ C
rs
(X, E) and let
L(φ)¦u, u¦ =
1
p!q!
h¯
ba
∂
2
φ
∂z
α
∂z
β
u
a
A
α
B
′
u
b
¯
AβB
′
Lemma 4.3. If φ : X → R is strongly qpseudoconvex then for any
u ∈ C
rs
(X, E), with s ≥ q +1 the following inequality holds at any point
of X
L(φ)¦u, u¦ ≥ l
A(u,u)
L(φ)
.
Proof. At any point x ∈ X, we have
L(φ)¦u, u¦ = h¯
ba
β
ε
β
(x)u
AβB
′
u
¯
AβB
′
84 4. Vanishing Theorems
Since 102
s ≥ q + 1, i.e s + n − q ≥ n + 1,
then every stuple of indices contains at least one the indices 1, 2 . . . , n−
q of the positive eigenvalues ε
1
(x), . . . , ε
n−q
(x). Hence
1
p!q!
h
ba
n−q
β=1
u
AβB
′
u
¯
AβB
′
≥ A(u, u),
and therefore
L(φ)¦u, u¦ ≥ ε
n−q
(x)A(u, u) +
1
p!q!
_
ε
n−q+1
(x)u
An−q+1B
′
u
An−q+1B
1
¦ + ε
n
(x)u
AnB
′ u
¯
AnB
′
_
≥ ε
n−q
(x)A(u, u) + Inf(o, ε
n
(x))A(u, u).
This proves our lemma.
Lemma 4.4. φ be a strongly qpseudo convex function on X and µ a
positive real valued function on R such that µ
′
(t) ≥ 0, µ
′′
(t) > 0.
Then
L(µ(φ))(u, u) = µ
′
(φ)L(φ)(u, u) + (µ
′′
)(φ)
¸
¸
¸
¸
¸
¸
∂φ
∂z
α
u
α
¸
¸
¸
¸
¸
¸
2
Hence
L(µ(φ))(u, u) ≥ µ
′
(φ)L(ϕ)(u, u),
and l
L(µ(φ))
≥ µ
′
(φ)l
L(φ).
The Lemma follows from a direct computation of the Levi form
L(µ(φ)) and from the fact since µ
′′
(φ)
¸
¸
¸
¸
¸
¸
∂φ
∂z
α
u
α
¸
¸
¸
¸
¸
¸
2
is positive semi deﬁnite,
the eigenvalues of L(µ(φ)) are not than the corresponding eigenvalues 103
of µ
′
(ϕ)L(ϕ).
13. Holomorphic bundles over qcomplete manifolds 85
Lemma 4.5. Let X be a hermitian manifold and φ a strongly qpseudo
convex proper function such that at each x ∈ X
l
L(φ)
(x) = ε
n−q
(X) + n Inf(0, ε
n
(x)) > 0,
where ¦ε
p
¦ are the eigen values of L(ϕ) with respect to the metric on
X arranged in decreasing order. Then given any continuous function
g : X → R there exists a sequence ¦a
ν
¦
1≤ν<∞
of real constants such that
for any C
∞
function µ : R → R satisfying (i) µ
′
(t) > 0 (ii) µ
′′
(t) ≥ 0 and
(iii) µ
′
(t) ≥ a
ν
for ν ≤ t < ν + 1, we have
l
L(µ(φ))
(x) > g(x).
Proof. Since the sets K
ν
= ¦x¦ν ≤ φ(x) ≤ ν + 1¦ are compact and
l
L(φ)
(x) > 0 for all x ∈ X, we can ﬁnd a
ν
such that
a
ν
l
L(φ)
(x) > g(x) for x ∈ K
ν
the lemma now follows from Lemma 4.4
13 Holomorphic bundles over qcomplete
manifolds
Lemma 4.6. Let E be a holomorphic vector bundle on a qcomplete
complex manifold. Let ϕ : X → R be a strongly qpseudoconvex C
∞
function on X such that ¦x¦φ(x) < c¦ ⊂⊂ X for every c ∈ R. Let ds
2
be a
hermitian metric on X such that l
L(φ)
(x) > 0 with respect to this metric.
Let h be any hermitian metric on X. Then there is a sequence (a
ν
)
1≤γ<∞
of real constants such that the following property holds.
Let µ be any C
∞
functions on R such that µ
′
(t) ≥ a
ν
for ν ≤ t ≤ ν+1 104
and µ
′
(t) > 0, µ
′′
(t) ≥ 0 for all t. Let 7 and (resp. 7
−µ
′
)
be the opera
tors (linear over C
∞
functions) from C
rs
(X, E) −→ C
rs
(X, E) deﬁned in
chapter 3. (See (3.12) and (3.17)) with respect to the hermitian metrics
ds
2
and h (resp. h = e
−µ(φ)h
. Then for ψ ∈ C
r,s
(X, E) with s > q, we
have
A
−µ
(7
−µ
ψ, ψ) ≥ e
−µ(φ)
A(ψ, ψ).
86 4. Vanishing Theorems
Proof. From the Remark at the end of Chapter 3,
7
−µ
ψ = 7ψ −
(−1)
i
∂
2
µ(φ)
∂
β
z
∂z
β
i
ϕ
a
A
β
B
′
i
.
It follows that
A
−µ
(7
−µ
ψ, ψ) = e
−µ(φ)
¦L(µ(φ))(ψ, ψ)¦ + e
−µ(φ)
A(7ψ, ψ)
≥ e
−µ(φ)
¦l
L(µ(φ)
+ f ¦A(ψ, ψ),
Where f : X −→ is a continuous function depending on 7. By
Lemma 4.4. (a
ν
)
1≤ν<∞
can be so chosen that for any convex µ satisfying
(i) µ
′
(t) > 0, (ii) µ
′′
(t) ≥ 0 (iii) µ
′
(t) > a
ν
for ν ≤ t < ν+1,
we have
c
1
l
L(µ(ϕ)
+ f ≥ 1.
Hence
A
−µ(ϕ)
(7
−µ
ψ, ψ) ≥ e
−µ(ϕ)
A(ψ, ψ).
Theorem 4.1. Let X be a qcomplete manifold so that there exists ϕ : 105
X −→ satisfying : (i) ϕ is strongly qpseudoconvex and (ii), for c ∈
, ¦x ¦ ϕ(x) ≤ c¦ is compact X. Let E be a holomorphic vector bundle
on X. Then there exists a complete hermitian metric ds
2
on X and a
hermitian metric h along the ﬁbres of E and real constants ¦a
ν
¦
1≤ν<∞
and c > 0 such that for every C
∞
function µ : −→ satisfying
(i) µ
′
(t) > 0, (ii) µ
′′
(t) ≥ 0 and (iii) µ
′
(t) ≥ a
ν
for v ≤ t < ν+1
we have
3(j∂ψj
−µ
+ jϑ
−µ
ψj
2
−µ
) ≥ jψj
2
−µ
for every ψ ∈ D
r,s
(X, E), s > q.
Hence E is W
rs
elliptic for s > q, with respect to the metric e
−µ(ϕ)
h.
Proof. From proposition 3.1., we have
(7
−µ
ψ, ψ) ≤ 3
_
j∂ψj
2
−
µ + jϑ
−µ
ψj
2
_
.
(The subscript −µ means that the operators scalar product etc. are
deﬁned with reference to the metric ds
2
on the base and e
−µ(ϕ)
h along
the ﬁbres of E). The theorems then follow from Lemma 4.2, Lemma 4.6
and the above inequality.
13. Holomorphic bundles over qcomplete manifolds 87
Remark . If we choose a µ satisfying the conditions of Theorem 4.1,
then for any function λ : −→ such that λ
′
(t) ≥ 0, λ
′′
(t) ≥ 0 for
t ∈ , µ + λ again satisﬁes the conditions of the theorems.
Consequently if we denote the metric on the base by ds
2
and the
hermitian metric e
−µ(ϕ)
h by h
′
, with respect to (ds
2
, h
′
) we have the fol 106
lowing properties for s > q:
1) There is a C
∞
function φ : X → R such that φ(x) ≥ 0
2) For every C
∞
non decreasing convex function λ : R → RE is W
rs
−
elliptic with respect to (ds
2
, e
−λ(φ)
h
′
)
3) The W
rs
− ellipticity constant is independent of ν.
4) The set B
c
= ¦x ¦ x ∈ X, ϕ(x) ≤ c¦ is compact in X for every c ∈ R.
These conditions imply conditions C
∗
1
, C
∗
2
, C
∗
3
, C
∗
4
of §4.
From Theorem 4.1 and from Theorem 1.3 we have the following
corollary.
Theorem 4.2. Let X be a qcomplete complex manifold and E a holo
morphic vector bundle on X. Then for
s ≥ q + 1, r ≥ 0, H
s
(X, Ω
r
(E)) = 0.
Proof. In fact, for ϕ ∈ C
rs
(X, E)(s > q), there is a suitable µ as in
Theorem 4.1 above (i.e µ : −→ is a C
∞
function such that µ
′
(t) >
0, µ
′′
(t) ≥ 0 and µ
′
(t) > a
ν
for ν ≤ t < ν + 1, ¦a
ν
¦
1≤ν<∞
being chosen so
that Theorem 4.1 holds for this sequence of real constants) such that
jϕj
2
−µ
=
_
X
e
−µ(φ)
A(ϕ, ϕ)dX < ∞.
On the other hand, Theorem 4.1. ensures us that with respect to the
metric e
−µ(ϕ)
h (h as in Theorem 4.1.) and the complete hermitian metric
ds
2
(as in Theorem 4.1.) on X, we have W
rs
− ellipticity for s > q. We
deduce Theorem 4.2. from Theorem 1.3 of Chapter 1.
88 4. Vanishing Theorems
Theorem 4.3. The cohomology groups H
s
k
(X, Ω
r
(E)) (cohomology with 107
compact supports) vanish for s ≤ n − q − 1 for any vector bundle E on
the qcomplete complex manifold X. Moreover, H
p−q
k
(X, Ω
r
(E)) has a
structure of a separated topological vector space.
Proof. The theorem can be proved by applying Serre’s duality to Theo
rem 4.2. It can also be given the following direct proof, which is based
on the results of §4.
Let us choose a metric h
′
along the ﬁbres of the dual bundle E
∗
in
such a way that conditions 1), 2), 3), 4) of the Remark after Theorem
4.1. be satisﬁed (by E
∗
). Corresponding to the metric e
−λ(φ)
h
′
on E
∗
,
we choose the “dual” metric
t
(e
−λ(φ)
h
′
)
−1
= e
λ(φ)
t
h
′−1
on E. Let ϕ ∈
C
rs
(X, E), with s ≤ n − q − 1. Then ∗#ϕ ∈ C
n−rn−s
(X, E
∗
). Since
n − s ≥ q + 1, ∗#ϕ satisﬁes the inequality
A
E
∗
,−λ
(7
E
∗
,−λ
∗ #ϕ, ∗#ϕ) ≥ A
E
∗
,−λ
(∗#ϕ, ∗#ϕ),
i.e. A
E
∗
,−λ
(7
E
∗
,−λ
∗ #ϕ, ∗#ϕ) ≥ A
E,λ
(ϕ, ϕ).
Hence, for any ϕ ∈ D
rs
(X, E), with s ≤ n − q − 1 we have
jϕj
2
E,λ
≤ 3(j∂ ∗ #ϕj
2
E
∗
,−λ
+ jϑ
E
∗
,−λ
∗ #ϕj
2
E
∗
,−λ
)
i.e. by (1.4), (1.5), (1.6),
jϕj
2
E,λ
≤ 3(j∂ϕj
2
E,λ
+ jϑ
E,λ
ϕj
2
)
This inequality holds for any ϕ ∈ D
rs
(X, E), with s ≤ n − q − 1, and 108
for any C
∞
, nondecreasing, convex function λ : →.
Our theorem follows from Theorem 1.4. of §4 and from Theorem
1.8.
Let now X be a qcomplete manifold and φ : X → R
+
a strongly
qpseudo convex function on X such that for
c ∈ , B
c
= ¦x¦x ∈ X, φ(x) < c¦ ⊂⊂ X. We set B
c
= Y for some c ∈ .
Lemma 4.7. B
c
is qcomplete.
13. Holomorphic bundles over qcomplete manifolds 89
Proof. Let µ be a C
∞
, nondecreasing convex function on (−∞, c) such
that ¦t ¦ t ∈ , µ(t) < t
◦
¦ is relatively compact in (−∞, c) for every
t
◦
∈ . Then the two conditions of Deﬁnitions 4.2 are satisﬁed by the
function µ(φ). This proves the lemma
We adopt the following notation
F
rs
(X, E) =
_
ψ ∈ L
rs
loc
(X, E); ∂ψ = 0
_
.
Q
rs
(X, E) =
_
ψ ∈ L
rs
−νφ
(X, E) for positive integer ν, ∂ψ = 0
_
.
Finally ρ : F
rs
(X, E) → F
rs
(Y, E) is the restriction map. Clearly
Q
rs
(X, E) ⊂ CF
rs
(X, E) and we have
Theorem 4.4. If s ≥ q, ρ(Q
rs
(X, E)) is dense in F
rs
(Y, E). (The metric
on X and that along the ﬁbres are chosen as ds
2
and h respectively in
the Remark after Theorem 4.1.)
Proof. We denote Q
rs
(X, E) simply by Q. Let µ be a continuous linear
functional on F
rs
(Y, E) which vanishes on r(Q). It is suﬃcient to prove
that µ vanishes on all of F
rs
(Y, E).
First, we extend µ to a continuous linear form on L
rs
loc
(Y, E) (Hahn 109
Banach extension theorem). By the representation theorem it follows
that there is a ψ ∈ L
rs
(Y, E) with compact support in Y such that µ(u) =
(ψ, u) for every u ∈ L
rs
loc
(Y, E).
Let c
◦
= sup¦ϕ(x) ¦ x ∈. Support ψ¦. Since Support ψ is compact in
Y = B
c
, then
c
◦
< c.
Now, let λ : → be a real C
∞
nondecreasing, convex function
such that
λ(t) = 0 for t ≤< c
◦
λ(t) = 0 for t < c
◦
,
0 ≤ λ
′
(t) ≤ 1.
Then, for t > c
◦
,
λ(t) ≤ t − c
◦
≤ t.
90 4. Vanishing Theorems
Let
u ∈ L
rs
−νλ(φ)
(X, E) for some ν > 0, ∂u = 0.
Then u ∈ Q, because
juj
2
−νφ
=
_
X
e
−νφ
A(u, u)dX ≤
_
e
−νλ(φ)
A(u, u)dX = juj
2
−νλ(φ)
Thus we have
(ψ, u) = µ(ρ(u)) = 0
Now under these conditions, it follows from Theorem 1.6, taking 110
into account the remark after Theorem 4.1 that there exists X ∈ L
r,s+1
such that
(i) ψ = ϑX and (ii) support X ⊂
_
x¦ϕ(x) c
◦
_
.
Let c
◦
< c
1
< c and introduce a complete hermitian metric on Y
which coincides with the given one B
1
. With respect to the new metric,
ψ ∈ L
rs
(Y, E), X ∈ L
r,s+1
(Y, E)
Now let u ∈ F
rs
(Y, E) be an arbitrary form such that ∂u = 0. Then
there is a C
∞
, nondecreasing function σ : (−∞, c) →such that σ(t) =
0 for t c
1
, and
u ∈ L
rs
−σ(φ)
(Y, E).
We have then
(ψ, u) = (ϑX, u) = (ϑ
−σ(φ)
X, u)
−σ(φ)
= (X, ∂u)
−σ(φ)
= 0.
Hence the linear form µ vanishes on F
rs
(Y, E). This proves the the
orem.
Corollary 1. Q
rs
(X, E) is dense in F
rs
(X, E) (in the topology of L
2
−
convergence on compact sets).
Proof. This is simply the case X = Y (note that in the proof of Theorem
4.4, the case c = ∞is covered).
Corollary 2. ρ(F
rs
(X, E)) is dense in (F
rs
(Y, E). 111
13. Holomorphic bundles over qcomplete manifolds 91
Proof. In fact Ω ⊂ F
rs
(X, E).
Suppose now that X is a stein manifold. Then X is is 0complete.
Hence we can ﬁnd a C
∞
function φ : X → R such that L(φ) is positive
deﬁnite and such that for c ∈ R,
_
x¦x ∈ X, φX < c
_
⊂ αX. If then E
is a holomorphic vector bundle on X, we obtain from Corollary 1, the
following result.
Corollary 3. Let Q(X, E) denote the space of holomorphic sections of
E over the Stein manifold X. There exists a complete hermitian metric
on X and a hermitian metric along the ﬁbres of E such that the set
¦ f ¦ f ∈ Q(X, E);
_
X
e
−νφ
¦ f ¦
2
< ∞ for some integer ν¦ is dense in Q(X, E)
in the topology of uniform convergence on compact sets.
In view of Corollary 1, to prove Corollary 3, we need only prove the
following
Lemma 4.8. Let D be a domain in C
n
and ¦ f
m
¦ a sequence of holomor
phic functions on D such that for every compact K ⊂ D, ¦ f
m
¦ converges
in L
2
(K) (= space of square summable functions on K). Then ¦ f
m
¦ con
verges uniformly on every compact set K.
Proof. Let K be a compact set contained in D. Then there is a constant
c = c(K) > 0 such that for every point z
◦
∈ K. The polydisc P
z
◦
= ¦z
¸
¸
¸¦z
i
− z
i
o
¦≤ c¦ of radius c is contained in D and
_
z
◦
∈K
P
z
o
⊂⊂ D. We
have then from the Cauchyintegral formula,
f (z
◦
) =
1
(πc
2
)
n
_
P
zo
f (z)dv. (dv = Lebesque measure in C
n
)
It follows on applying this to f
2
, that 112
¦ f (z
◦
) ¦
2
1
volP
z
◦
_
P
z
o
¦ f ¦
2
dv
1
volP
z
◦
_
P
z
o
j f ¦¦
2
K
′
where K
′
is the (compact) closure of
_
z
◦
∈K
P
z
o
. The lemma is immediate
from the above inequality.
In the special case q = 0, Corollary 2 together with the lemma above
yields the following result.
92 4. Vanishing Theorems
Theorem 4.5. Let X be a Stein manifold and ϕ : X → R a strongly
0pseudo convex C
∞
function on X such that for
c ∈ R,
_
X ¦ φ(X) < c
_
⊂⊂ X. Then the pair (X, B
c
), c ∈ R is
a Runge pair. In other words, the set of holomorphic functions on B
c
which are restrictions of global holomorphic functions on X is dense in
the space of all holomorphic functions on B
c
in the topology of uniform
convergence on compact subsets (of B
c
).
14 Examples of qcomplete manifolds
Let X be an mcomplete manifold and f
1
, . . . , f
q+1
be any q+1 holomor
phic functions on V. Let Z =
_
x¦ f
i
(x) = 0 for 1 i q + 1
_
.
Assertion Y = V − Z is (m + q)complete.
Proof. Let ϕ : V → R be a strongly mpseudoconvex function V such
that for every c ∈ R,
B
c
=
_
x ¦ x ∈ V, ϕ(x) < c
_
⊂⊂ X.
Because of the second condition, we can choose a C
∞
function 113
λ : R → R such that λ
′
(t) > 0, λ
′′
(t) ≥ 0,
λ(t) > 0 for t ∈ R, λ(t) → ∞as t → ∞and further
λ(ϕ) >
q+1
i=1
f
i
f
i
on Y.
Set ψ = λ(ϕ) − log
q+1
_
i=1
f
i
¯
f
i
. Since L(log
q+1
_
i=1
f
i
¯
f
i
) is positive semi
deﬁnite with at most q positive eigenvalues, ψ strongly (m+ q)pseudo
convex. It remains to prove that
B
′
c
= ¦X ¦ x ∈ Y, ψ(x) < c¦ ⊂⊂ y
14. Examples of qcomplete manifolds 93
for every c ∈ R. Now ψ(x) ≤ c implies that
λ(ϕ) −
q+1
i=1
log f
i
¯
f
i
≤ c
or again that
e
λ(φ)
q+1
_
i=1
f
i
f
i
≤ e
c
,
i,e.,
q+1
i=1
f
i
f
i
≥ e
λ(ϕ)−c
> δ > 0.
Hence B
c
⊂ X − U, where U is a neighborhood of Z. On the other
hand; since λ(ϕ) >
q+1
_
i=1
f
i
f
i
, e
λ(ϕ)
/λ(ϕ) < e
c
, hence λ(ϕ) < e
c
= C, on B
′
c
. 114
Hence B
′
c
⊂ B
c
. On the other hand since B
′
c
is closed in X − Z in
X − U, it is closed in X. Hence B
′
c
is compact. This concludes the proof
of our assertion.
In particular, if X is a Stein manifold, then Y is qcomplete.
If Z is a complete intersection, then q+1 is the complex codimension
of the submanifold Z of X. Let now Z be any analytic subset of the Stein
manifold X, of complex codimension q + 1 at each point. To Y = X − Z
qcomplete?
If q = 0, and if Z is a divisor, then the answer to this question is
positive i.e. Y = X − Z is a Stein manifold as it was shown by Serre
([6] p. 50) R.R. Simha [31] has proved that, if q = 0, if X is a complex
codimension 2 and if Z is an analytic set in X of complex codimension
1 at each point, then Y = X − Z is a Stein space. Examples ([23] Satz
12, 17; [14]) show that this result is false when dim
C
X > Z.
An example, due to G. Sorani and V. Villani [33] , shows that if
q > 0 then Y is not necessarily qcomplete. Before discerning that
example, we shall establish the following
94 4. Vanishing Theorems
Theorem 4.6 ([32]). Let X be qcomplete of complex dimension n. Let
Ω
n
denote the sheaf of germs of holomorphic nforms on X. The natural
map
H
q
(X, Ω
n
) → H
n+q
(X, C)
is surjective.
Proof. For any complex manifold, we have spectral sequence (E
st
r
)
o
≤ 115
r ≤ ∞ such that E
∞
is the associated graded group of H
∗
(X, C), and
E
st
1
= H
t
(X, Ω
s
) [11]. In the case when X is qcomplete, H
t
(X, Ω
s
) = 0
for t > q. In particular for t +s = n+q, E
st
t
= 0 for all (s, t) diﬀerent from
(n, q). Hence E
s,t
r
= 0 for (s, t) (n, q), s + t = n + q. Moreover, all of
E
n,q
1
are cycles for d
1
since Ω
n+1
= 0. Hence E
n,q
1
→ E
n,q
2
is surjective.
For r ≥ 2, E
n−r,q+r−1
r
= 0 since q + r − 1 > q and E
n+r,q+r+1
= 0. Hence
E
n,q
2
∽E
n,q
∞
. Hence H
n+q
(X, C) ≃ E
n,q
∞
= E
n,q
2
. (We note that E
s,t
∞
= 0 for
s + t = n + q, (s, t) (n, q)). This proves the theorem.
We consider now the subsets of C
2n
Z
1
=
_
z ∈ C
2n
¸
¸
¸z
i
= 0 for i > n
_
, X
1
= C
2n
− Z
1
,
Z
2
=
_
z ∈ C
2n
¸
¸
¸z
i
= 0 for i ≤ n
_
, X
2
= C
2n
− Z
2
,
Z = Z
1
∪ Z
2
, Y = C
2n
− Z = X
1
∩ X
2
.
The analytic set Z ⊂ C
2n
has complex codimension n at each point.
We shall prove that Y is not (n − 1)complete, when n ≥ 2.
Since the point ¦0¦ is a complete intersection of codimension 2n in
C
2n
, then U = C
2n
− ¦0¦ is (2n − 1)complete. Let Ω
2n
be the sheaf of
group of holomorphic 2nforms. Then by Theorem 4.6.
H
2n−1
(U, Ω
2n
) → H
2n−1
(U, C)
is surjective. On the other hand, since U is contractible on the unit 116
sphere of C
2n
, then H
2n−1
(U, C) ≈ C. Hence
H
2n−1
(U, Ω
2n
) 0.
The exact cohomology sequence of MayerVictoris yields
15. A theorem on the supports of analytic functionals 95
→ H
r
(U, Ω
2n
) → H
r
(X
1
, Ω
2n
) ⊕ H
r
(X
z
, Ω
2n
)
→ H
r
(X
1
∩X
2
, Ω
2n
) → H
r+1
(U, Ω
2n
) → (4.1)
Since Z
i
(i = 1, 2) is a complete intersection of codimension n, X
i
is
(n − 1)complete.
Hence
H
r
(X
i
, Ω
2n
) = 0 f or r ≥ n(i = 1, 2).
It follows from (4.1) that
H
r
(X
1
∩ X
2
, Ω
2n
) ≈ H
r+1
(U, Ω
2n
) for r ≥ n.
Then
H
2n−2
(X
1
∩ X
2
, Ω
2n
) 0.
If n ≥ 2, then 2n − 2 ≥ n. This shows that X
1
∩ X
2
is not (n − 1)
complete.
15 A theorem on the supports of analytic function
als
We shall apply the methods developed above to obtainfollowing the
ideas of [16], §2.5 with minor technical changes  a generalization, due
to Martinean [23] of a theorem originating with P´ olya [27]. We ﬁrst
prove two propositions which are of independent interest.
Proposition 4.1. Let φ be a C
∞
plurisubharmonic function on C
n
. Let 117
jωj
−ϕ
denote the norm of a C
∞
form ω of type (p, q) with respect to
the Euclidean metric on the base and the metric e
−φ
on the trivial line
bundle. Let ψ = φ + 2 log(1 + ¦z¦
2
). Then there is a constant C > 0 such
that for any C
∞
form ω of type (p, q), q ≥ 1, with
¯
∂ω = 0, jωj
φ
< ∞,
there is a C
∞
form ω
′
of type (p, q−1) with
¯
∂ω
′
= ω, jω
′
j
−ψ
≤ Cjωj
−φ
.
Proof. Consider the operator 7
−ψ
with respect to the metric e
−ψ
. We
have, since the Euclidean metric is ﬂat,
A
ψ
(7
−ψ
u, u) = e
−ψ
L(ψ)¦u, u¦
96 4. Vanishing Theorems
≥ 2e
−ψ
(1 +
¯
¦z¦
2
)
−2
A(u, u) = 2(1 + ¦z¦
2
)
−2
AA
−ψ
(u, u),
since L(ψ)¦u, u¦ ≥ 2L(log(1 +¦z¦
2
))¦u, u¦ ≥ 2(q +¦z¦
2
)
−2
A(u, u). Hence
7
−ψ
is positive deﬁnite with eigenvalues ≥ 2(1 + ¦z¦
2
)
−2
at the point z;
7
−1
−ψ
is also positive deﬁnite, and its eigenvalues are ≤
1
2
(1 + ¦z¦
2
)
2
. By
consequence we have
¦A
−ψ
(u, v)¦ ≤ A
−ψ
(7
−1
−ψ
u, u)A
−ψ
(7
−ψ
v, v)
≤
1
2
A
−ψ
((1 + ¦z¦
2
)
2
u, u)A
−ψ
(7
−ψ
v, v).
Hence, for forms u, v of type (p, q), we have
¦(u, v)
−ψ
¦
2
≥
1
2
j(1 + ¦z¦
2
)uj
2
−ψ
(7
−ψ
v, v)
−ψ
, (4.2)
whenever the forms u, v are such that the norms on the right are ﬁnite.
On D
p,q
(C
n
), we introduce the norm M by 118
M(u)
2
= (7
−ψ
u, u)
−ψ
+ j
¯
∂uj
2
−ψ
+ jϑ
−ψ
uj
2
−ψ
,
and let V
p,q
be the completing of D
p,q
with respect to this norm. Obvi
ously V
pq
⊂ L
pq
loc
. For u ∈ D
p,q
, we have, by (3.18),
(7
−ψ
u, u) ≥ j
¯
∂uj
2
−ψ
+ jϑ
−ψ
uj
2
−ψ
.
Thus the norm M on D
p,q
is equivalent to that deﬁned by
M
′
(u)
2
= j
¯
∂uj
2
−ψ
+ jϑ
−ψ
uj
2
−ψ
.
Let ω ∈ L
p,q
loc
, and let jωj
−ϕ
< ∞. Then, for u ∈ D
p,q
, we have, by
(4.2) (dropping the factor
1
2
as we may)
¦(u, ω)
−ψ
¦
2
≤ j(1 + ¦z¦
2
)ωj
2
−ψ
M
′
(u)
2
= jωj
2
−φ
M
′
(u)
2
(4.3)
15. A theorem on the supports of analytic functionals 97
Thus, the linear form u (u, ω)
−ψ
is continuous on V
p,q
so that
there exists a unique x ∈ V
p,q
with
(u, ω)
−ψ
= (
¯
∂u,
¯
∂x)
−ψ
+ (ϑ
−ψ
u, ϑ
−ψ
x)
−ψ
= (
−ψ
u, x)
−ψ
, u ∈ D
p,q
.
Thus we have
ω =
−ψ
x
and, moreover, by (4.3) 119
M
′
(x)
2
= (x, ω)
−ψ
j ω j
−φ
M
′
(x),
so that j ∂x j
2
−ψ
+ j ϑ
−ψ
x j
2
−ψ
j ω j
2
−φ
.
Let us assume now that ω is C
∞
and that ∂ω = o. Then by Corol
lary 1 to Theorem 1.2, we have, for any σ > 0,
j ϑ
−ψ
∂x j
ψ
1
σ
j ∂ω j
2
ψ
+σ j ∂x j
2
−ψ
= σ j ∂x j
−ψ
,
and letting σ → 0, we obtain
ϑ
−ψ
∂x = 0,
so that
−ψ
x = ∂ϑ
−ψ
x;
since, as proved above, j ϑ
−ψ
x j
−ψ
≤j ω j
−φ
, we obtain the proposition
on setting ω
′
= ϑ
−ψ
x.
We point out that the above proposition depends on a weaker condi
tion than W
p,q
ellipticity, in fact, proposition 4.1 is a particular case of
the following general statement, which can be established by a similar
argument.
Proposition 4.2. Let X be a complex manifold endowed with a complete 120
hermitian metric. Let π : E → X be a holomorphic vector bundle on X
and let h be a hermitian metric along the ﬁbres of E. Assume that
A(7ϕ, ϕ) > 0
98 4. Vanishing Theorems
at each point of X and for every ϕ ∈ C
pq
(X, E). Let ε(x) be the least
eigenvalue of the (positive deﬁnite) hermitian from A(7ϕ, ϕ)
x
(x ∈ X)
acting on the space of the (p, q)form with values in E. Then ε(x) > 0.
If ϕ ∈ C
pq
(X, E), ∂ϕ = 0 is such that
_
X
1
ε(x)
A(ϕ, ϕ)dX < ∞,
there exists a from ψ ∈ C
p,q−1
(X, E) such that
∂ψ = ϕ,
_
X
A(ψ, ψ)dX ≤ 3
_
X
1
ε(x)
A(ϕ, ϕ)dX.
Proposition 4.3. Let φ be a plurisubharmonic function on C
n
such that
there exists a constant C > 0 for which ¦φ(z) − φ(z
′
)¦ ≤ C whenever
¦z − z
′
¦ ≤ 1. Let V be a (complex) subspace of C
n
of codimension k,
and f a holomorphic function on V such that
_
¦ f ¦
2
e
−ϕ
d
V
v < ∞ (d
V
v =
Lebesgue measure on V). Then, there exits a holomorphic function F on
C
n
with F¦V = f and
_
C
n
¦F¦
2
e
−ϕ
(1 + ¦z¦
2
)
3k
dv M
_
V
¦ f ¦
2
e
−ϕ
d
V
v,
where M is a constant independent of f ; here dv = d
C
n v is the Lebesgue 121
measure in C
n
Proof. We may clearly suppose that V has codimension 1 in C
n
. We
assume that V is the subspace z
n
= 0.
Then f is a holomorphic function in C
n
, depending only on z
1
, . . . ,
z
n−1
. Let λ : R → R be a C
∞
function on R, such that
0 λ(t) 1,
λ(t) = 1 for t
1
4
λ(t) = 0 for t 1.
Let C
1
= Sup
¸
¸
¸
¸
¸
dλ
dt
¸
¸
¸
¸
¸
.
15. A theorem on the supports of analytic functionals 99
We will construct a function µ in such a way that
F(z
1
, . . . , z
n
) = λ(¦z
n
¦
2
) f (z
1
, . . . , z
n−1
) − z
n
µ(z
1
, . . . , z
n
)
satisﬁes all the requirements of the lemma.
Let
σ = ∂λ(¦z
n
¦
2
) f ,
σ is a C
∞
, ∂closed form, such that ¦z
n
¦ ≤
1
2
. Thus the form ω =
1
z
n
σ
is a (0, 1) from which is C
∞
and ∂closed on C
n
, i.e. ω ∈ C
01
(C
n
, C).
Moreover
Supp ω ⊂
_
z = (z
1
, . . . , z
n
)¦ ¦z
n
¦
1
2
1
_
.
Since 122
¦ φ(z, ..., z
n−1
, z
n
) − φ(z, ..., z
n−1
, 0) ¦≤ C for ¦ z
n
¦ 1,
then
_
¦z
n
¦<1
e
−ϕ
′
¦ f ¦
2
dv πe
C
_
V
e
−ρ
¦ f ¦
2
dv
V
On the other hand
jωj
2
−φ
=
_
1
2
<¦z
n
¦<1
e
−φ
A(ω, ω)dv (2)
2
_
1
2
<¦z
n
¦<1
e
−φ
¦ f ¦
2
d C
2
_
V
e
−φ
¦ f ¦
2
dv
V
(C
2
= 4C
2
1
πe
C
). In particular, jωj
2
−φ
< ∞
Assume now that φ is C
∞
. By Proposition 4.1, there exists a C
∞
function µ : C
n
→ C such that
= ∂µ
_
C
n
e
−φ
(1+ ¦ z ¦
2
)
n
¦ µ ¦
2
dv jωj
2
−φ
c
2
_
V
e
−φ
¦ f ¦
2
dv.
If φ is not C
∞
, we introduce a regularising function, i.e. a C
∞
func
tion α(z) = α(¦ z ¦) 0, on C
n
, with support ⊂ ¦¦ z ¦ 1¦, and such that
_
C
n
αdv = 1. Given 0 < ε < 1, the function
ϕ
ε
(z) =
_
C
n
φ(z − εz
′
)α(z
′
)dv(z
′
)
100 4. Vanishing Theorems
is a C
∞
plurisubharmonic function on C
n
, such that
φ
ε
(z) → φ(z) as ε → 0.
Moreover 123
¦ φ
ε
(z) − φ(z) ¦
_
C
n
α(z
′
) ¦ φ(z − εz
′
) − φ(z) ¦ d(z
′
) C,
and therefore ¦ φ
ε
(z) − φ
ε
(z
′
) ¦ 3C for ¦ z − z
′
¦ .
Hence there exists a C
∞
function µ
ε
: C
n
→ C, such that
∂µ
ε
= ω
_
ε
n
e
−φ
(1+ ¦ z ¦
2
)
2
¦ µ
ε
¦
2
dv e
c
_
C
n
e
−φ
ε
(1+ ¦ z ¦
2
)
2
¦ µ
ε
¦
2
dv C
3
_
V
e
−φ
¦ f ¦
2
dv
V
.(C
3
= C
2
1
πe
5c
)
Letting ε → 0, we can ﬁnd a distribution µ such that
∂µ = ω
1
_
C
n
e
−φ
(1+ ¦ z ¦
2
)
2
¦ µ ¦
2
dv C
3
_
V
e
−φ
¦ f ¦
2
dv
V
.
Consider now the distribution
F = λ(¦ z
n
¦
2
) f − z
n
µ.
We have
∂F = ∂λ f − z
n
∂µ = 0.
Hence F is a holomorphic function C
n
.
Furthermore 124
_
C
n
e
−φ
(1+ ¦ z ¦
2
)
3
− ¦ F ¦
2
dv
15. A theorem on the supports of analytic functionals 101
2
_
¦z
n
¦<¦
e
−φ
(1+ ¦ z ¦
2
)
3
− ¦ f ¦
2
dv + 2
_
C
¦ z
n
¦
2
e
−φ
(1+ ¦ z ¦
2
)
3
¦ µ ¦
2
dv
2πe
C
_
V
e
−φ
¦ f ¦
2
dv
V
+ 2
_
C
n
e
−φ
(1+ ¦ z ¦
2
)
2
¦ µ ¦
2
dv
C
4
_
V
e
−φ
¦ f ¦
2
dv
V
(C
4
= 2πe
C
+ 2C
3
).
This proves our proposition.
Lemma 4.9. Let φ be as in Proposition 4.2. Let F : C
n
→ C be a
holomorphic function such that
_
C
e
−2φ
(1+ ¦ z ¦
2
)
N
¦ F ¦
2
dv = M
2
< ∞,
for some N 0. Then there exists a positive constant C
5
depending only
on C and on N such that
¦ F(z) ¦ C
5
Me
φ(z)
(1+ ¦ z ¦)
N
e
φ(z)
.
Proof. Let z
0
∈ C
n
and let P(z
0
) be the polydisc with centre z
0
and
radius 1. Then by Cauchy’s formula we have
F(z
0
)
2
=
1
π
n
_
P(z
0
)
F(z)
2
dv,
whence 125
¦ F(z
o
)
2
¦
1
π
n
_
P(z
o
)
¦ F ¦
2
d.
Furthermore
e
−φ(z
o
)
e
ε
e
−φ
(z) for z ∈ P
1
(z
o
)
and
1
(1+ ¦ z
o
¦)
2
_
1 + 2n
1+ ¦ z ¦
_
2
(1 + 2n)
2
1+ ¦ z ¦
2
f or z ∈ P
1
(z
o
)
102 4. Vanishing Theorems
In conclusion
e
−2φ(z
o
)
(1+ ¦ z
o
)
2N
¦ F(z
o
) ¦
2
e
2ε
(1 + 2n)
2N
π
n
_
P(z
o
)
e
−2φ
(1+ ¦ z ¦
2
)
¦ F
2
¦ dv
e
2ε
(1 + 2n)
2N
π
n
M
2
.
This concludes the proof of our lemma.
Before going on to state the theorem of Martineanp´ olya, we need
some deﬁnitions.
Let H = H (C
n
) denote the vector space of holomorphic function
on C
n
with the topology of compact convergence. An analytic func
tional µ is a continuous linear functional on H µ is said to be supported
by a compact set K if for any open set U ⊃ K there is a constant M
U
such that sort any f ∈ H we have
¦ µ( f ) ¦≤ M
U
sup
z∈U
¦ f (z) ¦ .
For z = (z
1
, . . . , z
n
), ζ = (ζ
1
, . . . , ζ
n
) ∈ C
n
, let (z, ζ) =
_
z
i
ζ
i
. For 126
ζ ∈ C
n
, let e
ζ
∈ H be deﬁned by e
ζ
(z) = e
(z,ζ)
. The Laplace transform
of µ is the holomorphic function
∽
µ on C
n
deﬁned by
∽
µ(ζ) = µ(e
ζ
).
Let K be compact and H the function deﬁned by H
K
(ζ) = sup
z∈K
Re
(z, ζ). Clearly H
K
is continuous, and hence, being the supremum of
plurisubharmonic functions, is itself plurisubharmonic. If K
ε
denotes
the set ¦z ∈ C
n
¦ d(z, K) ≤ ε¦, we have H
K
ε
(ζ) = H
K
(ζ) + ε ¦ ζ ¦
Theorem 4.7 (MartineaoPolya). In order that the analytic functional µ
be supported by the convex compact set K, it is necessary and suﬃcient
that for every ε > 0, there exist a constant C
ε
> 0 such that
¦
∽
µ(ζ) ¦ C
ε
e
H
K
(ζ)+∈¦ζ¦
.
15. A theorem on the supports of analytic functionals 103
Proof. Suppose that µ is supported by K. Then, by deﬁnition
¦ ˜ µ(ζ)¦ = ¦µ(e
ζ
)¦ ≤ C
ε
sup
z∈K
ε
¦e
ζ
(z)¦ = C
ε
e
¦sup
Kε
(z,ζ)¦
≤ C
ε
e
H
K
(ζ)+ε¦ζ¦
To prove the converse, we proceed as follows. H is a closed sub
space of the space C of continuous function on C
n
; hence (Hahn  Ba
nach) µ extends to a linear functional µ : C → C, hence deﬁnes a
measure. It is clearly suﬃcient to prove that any ε > 0, there exists
distribution ν with support in K
ε
such that
µ( f ) = ν for all f ∈ H .
Let ξ
′
be the space of distributions with compact support in C
n
. For 127
ν ∈ ξ
′
, let ˆ ν denote the Fourier transform of ν considered as a function
of 2n real variables u
1
, . . . , u
2n
:
ˆ ν(u
1
, . . . , u
2n
) = ν(e
′
u
),
where e
′
u
(z) = exp (−i(u
1
Re z
1
+ u
2
Imz
1
+ . . . + u
2n
Imz
n
)).
Clearly¯ν has an extension to a holomorphic function on C
2n
; further
since linear combination of the function e
ζ
are dense in H , we have
ν(g) = µ(g) for all g ∈ H if and only if µ(e
ζ
) = ν(e
ζ
) for all ζ, i.e. if
and only if
∽
µ(ζ) =¯ν(iζ
1
, −ζ
i
, ..., iζ
n
, −ζ
n
). (4.4)
Therefore, because of the PaleyWiener theorem, it is suﬃcient to
construct an entire function ¯ν on C
2n
satisfying (4.4), for which, we
have further,
¦ ¯ν(u) ¦ C
ε
(1+ ¦ u ¦ ¦)
Nφ(u)
. (4.5)
for some N > 0; here φ(u) = sup
x∈K
ε
(x
1
Imu
1
+ . . . + x
2n
Imu
2n
) and z
j
=
x
2j−1
+ ix
2j
.
Consider the subspace V of C
2n
consisting of points (iζ
i
, −ζ
1
, ..., i
ζ
n
, −ζ
n
), where ζ = (ζ
1
, ..., ζ
n
) ∈ C
n
. If u = (iζ
1
, ..., −ζ
n
), we have
φ(u) = H
K
ε
(ζ) = H
K
(ζ) + ε¦ζ¦.
104 4. Vanishing Theorems
Further, φ(u) is plurisubharmonic in C
2n
and there exists a constant 128
C > 0 such that ¦φ(u) − φ(u
′
)¦ C for ¦u − u
′
¦ 1. Clearly
∽
µ deﬁnes
a holomorphic function f on V if we set f (iζ
1
, ..., −ζ
n
) =
∽
µ(ζ
1
, ..., ζ
n
).
Since for any δ > 0, ¦
∽
µ(ζ)¦ C
δ
e
H
K
(ζ)+δ¦ζ¦
by hypothesis and φ(u) =
H
K
(ζ) + ε¦ζ¦ on V, we have
¦ f (u)¦ C
δ
e
φ(u)+(δ−ε)¦u¦
on V.
If δ < ε, we clearly have therefore
_
V
¦ f (u)¦
2
e
−2φ(u)
d
V
v < ∞.
By Proposition 4.2 there exists F, holomorphic in C
2n
such that
F¦V = f , and
_
C
2n
¦F(u)¦
2
e
−2φ(u)
(1 + ¦u¦
2
)
−3n
dv < ∞.
By Lemma 4.9 this implies that there exists a constant C > 0 such
that
¦F(u)¦ M(1 + ¦u¦)
3n
e
φ(u)
,
so that if we set ¯ν(u) = F(u), the conditions (4.4) and (4.5) are satisﬁed.
This completes the proof of the theorem.
Bibliography
[1] A. Andreotti and H. Grauert: Th´ eor` ems de ﬁnitude pour la coho 129
mogipdes espaces complexes, Bull. Soc. Math. Frances 90(1962),
193259.
[2] A. Andreotti and E. Vesentini: Carleman estimates for the Laplace
Beltrami equation on comples manifoldes, instit. Heautes
´
Etudes
Sci., Publ. Math. 25.81130: Erratum ibd, to appearss
[3] A. Andreotti and E. Vesentini Les th´ eor` emes fondaientaux de la
th´ eorie des espaces holomorhiquement complets, S´ eminaire de
G´ eometrie Diﬀ´ erentielle Vol IV (196263) 131; Paris, Secr´ etariat
Math´ emateque.
[4] S. Bochner: Tensor ﬁelds with ﬁnite bases, Ann. of Math.,
53(1951), 400411.
[5] E. Calabi and E. Vesentini: On compact, locally symmetric Kahler
manifolds, Ann. of Math., 71(1960), 472507.
[6] H. Cartan: S´ eminaire E.N.S., 195354; Paris, Secr´ etariat
Math´ ematique.
[7] H. Cartan: Vari´ et´ es analytiques complexes et cohomologie, Col
loque sur les fonctiones de plusieurs Variables, Bruxelles, 1953;
4155.
[8] J. Deny and J.L. Lions: Les espaces de B. Levi, Ann. Inst. Fourier.
5(195354), 305370.
105
106 BIBLIOGRAPHY
[9] J. Dieudonn´ e and L. schwartz: La dualit´ e dans les espaces (F) et
(LF), Annales de I’ Institut Fourier, 1(1949), 61101.
[10] K.O. Friedrichs: On the diﬀerentiability of the solutions of linear
elliptic diﬀerential equations, Comm. Pure Appl. Math., 6(1953),
299325.
[11] K.O. Friedrichs: Diﬀerential forms on Riemannian manifolds,
Comm. Pure. Appl. Math., 8(1955) 551590.
[12] A. Frolicher: Relations between the cohomology groups of Dol
beault and topological invariants, Proc. Nat. Acad. Sci. U.S.A.,
41(1955), 641644.
[13] M.P. Gaﬀney: A special Stokes’s theorem for complete riemannian 130
manifolds, Ann. of Math., 60(1954), 140145.
[14] R. Godement: Topologie alg´ ebrioue et th´ eorie et theorie des fais
ceaux, Hermann, Paris, 1958.
[15] H. Grauert and R. Remmert: Singularitaten komplexer Man
nigﬂtigkeiten and Riemannscher Gebiete, Math. Zeitschr.,
67(1957), 103128.
[16] L. Hormander: L
2
estimates and existence theorems for the ∂
operator, Diﬀenrential Analysis, OXford University Press, 1964,
6579.
[17] L. Hormander: Existence theorems for the ∂operator by L
2
meth
ods, Acta mathematica, 113 (1965), 89152.
[18] K. Kodaira: Harmonic ﬁelds in Riemannian manifolds (General
ized potential theory), Ann. of math., 50 (1959), 587665.
[19] J.L. Koszul: Vari´ et´ es Kahl´ eriennes, S˜ ao Paulo, 1957;
mimeographed notes.
[20] G. K¨ othe: Topologische lineare R¨ aume, SpringerVerlag, 1960.
BIBLIOGRAPHY 107
[21] A. Lichn´ erowicz: Th´ eorie globale des connexion et des groupes
d’holonmie, Cremonese, Roma, 1955.
[22] A. Lichn´ erowicz: G´ eom´ etrie des groupes de transformationes,
Dunod, Paris, 1958.
[23] B. Malgrange: Lectures on the theory of functions of several com
plex variables, Tata Institute of Fundamental Research, Bombay,
1958.
[24] A. Martineau: Sur les fonctionnelles anaytiques et la trans
formation de FourierBorel, Journal d’Analyse Math´ ematique,
XI(1963), 1164.
[25] T. Meis: Die Minimale Blmtterzhl der Konkretisierungen einer
kompakten Riemannschen Flmche Schriftenreihe des Mathema
tischen Instituts der Universitmt munster, Heft 16, 1960.
[26] M.S. Narasimhan: A remark on curvature and the Dirichlet prob 131
lem, Bull. Amer. Math. Soc., 65(1959), 363364.
[27] K. Nomizu: Lie groups and diﬀerential geometry, Publications of
the Mathematical Society of Japan, 1956.
[28] G. Polya: Untersuchungen ¨ uber L¨ ucken and Singularit¨ aten von
Potenzreihen, Math. Zeitschr., 29 (1929) 549640.
[29] F. Rellich: Ein Satz ¨ uber mittlere Konvergenz, Nachr. Ges.
G¨ ottingen (math.phys. Kl.), 1930, 3035.
[30] G. de Rham: Vari´ et´ es diﬀ´ erentiables, Hermann, Paris, 1955.
[31] L. Schwartz: Th´ eories des distributioes, Hermann, Paris; Tome I,
1957; Tome II, 1959.
[32] R.R. Simha: On the complement of a curve on a Stein space of
dimension two, Math. Zeitschr., 82(1963), 6366.
[33] G. Sorani: Homologie des qpaires de Runge, Annali Scuola Nor
male Superiore, Pisa, (3), 17 (1963), 319322.
108 BIBLIOGRAPHY
[34] G. Soraniand V. Vesentini: Spazi qcompleti e coomologia,
Mimeographed notes, 1964.
[35] E. Vesentini: Sulla coomologia delle variet´ a diﬀerenziaoili, Rend.
Seminario Matematico e ﬁsico di Milano, 34 (1964), 1930.
[36] A. Weil: Introduction a l’´ etude des vari´ et´ es k¨ ahl´ eri ennes, Her
mann, Paris, 1958.
[37] K. Yano and S. Bochner: Curvature and Betti numbers, Ann. of
Math. Studies, 32, Princeton, 1953.